Turning insight into action Trust before autonomy Building AI-ready digital twins AI-native operating system
ISSUE 70 | OCTOBER 2026
Decision boundaries for AI at scale The future of industrial AI
Enablon
Building a safer, more responsible and sustainable world using the power of innovative technology. Award-winning software for optimizing environmental performance, improving worker safety, streamlining reporting and achieving regulatory compliance Recognized as a global market leader by third party analysts including Verdantix and Gartner
25+ years of experience of delivering innovative software that adapts to evolving regulations
Part of Wolters Kluwer, a global leader in professional information, software and services with operations in over 40 countries
Trusted by the world’s largest companies
90%
of the largest oil & gas companies
70%
of the top 10 pharmaceutical companies
70%
of the world’s top 10 technology giants
70%
of the world’s largest aerospace & defence companies
65%
of the world’s largest food & beverage companies
40%
of the top 10 chemical producers Scan the QR and contact us
CEO Adam Soroka
From digital representation to operational intelligence Digital twins have reached an important point in their evolution. For much of the past decade, the focus has been on creating richer digital representations of physical assets, bringing together engineering models, operational data, maintenance information and visualisation tools. That work has created substantial value, but the conversation is now moving on. The question is no longer simply how accurately we can represent an asset, but how effectively that digital environment can improve the decisions made around it.
Editor Mark Venables
That shift is particularly important for oil and gas. Complex assets, long operating lifecycles and demanding safety and integrity requirements mean there is little tolerance for digital systems that are impressive in demonstration but difficult to trust in day-to-day operations. As artificial intelligence becomes increasingly integrated with digital twins, the quality of the underlying engineering context will matter more than ever. AI may be able to interrogate vast volumes of information, identify patterns and accelerate analysis, but its recommendations are only as reliable as the data, models and relationships on which they are based.
Turning insight into action Trust before autonomy Building AI-ready digital twins
Decision boundaries for AI at scale The future of industrial AI
AI-native operating system
ISSUE 70 | OCTOBER 2026
Throughout this magazine, many of those themes come to the fore. Contributors explore interoperability, engineering data quality, industrial knowledge graphs, modelling fidelity, explainability and the growing role of AI in supporting operational decision-making. Collectively, they point towards a more mature vision of the digital twin: not simply a visual representation of an asset, but a trusted information environment connecting engineering, operations and increasingly intelligent software. There is also a growing recognition that the next stage will require more than technology. Organisations need to decide where automation can genuinely add value, where human judgement must remain decisive and how digital twins can stay aligned with physical assets as those assets change.
Governance, provenance, model validation and management of change may not generate the same excitement as generative AI, but they will determine how confidently that AI can eventually be deployed. That is what makes this year’s Future Digital Twin & AI conference particularly timely. The industry is moving beyond experimentation towards the harder questions of scale, trust and operational value. The opportunities are significant, but so too is the responsibility to ensure that increasingly powerful digital capabilities remain grounded in engineering reality. I hope this magazine adds another layer to those conversations, offering ideas, experiences and perspectives that continue well beyond the conference itself.
Future Digital Twin & AI
Cavendish Group
Mark Venables Editor
Second Floor Front 116-118 Chancery Lane London WC1A 1PP Tel: +44 (0)203 675 9530
www.futureoilgas.com | Future Digital Twin & AI 01
TRANSFORMING AI INSIGHT INTO RELIABLE PERFORMANCE BUILDING TRUST TOGETHER IN AI
www.dnv.com
06 The next digital twin challenge is not visibility, but decision quality As oil and gas operators connect digital twins with industrial AI, the focus is shifting from representing assets to improving the decisions made around them. The next stage will depend on trusted context, engineering physics and clearly defined boundaries between human judgement and machine autonomy.
10 Digital Twin journey: Turning insight into action Susan Burrell, Global Product Owner Digital Twin at Shell Research Limited, explains how digital twins are moving beyond integrated asset visualisation to become part of how operational work is planned and executed. She explores why trusted digital context is increasingly important to improving decisions and ultimately changing the way work is done.
16 Trust before autonomy: assuring industrial AI in energy As artificial intelligence moves from exploration and decision support towards operational systems, the energy sector is discovering that technical capability is only part of the equation. Building confidence in data, governance and system behaviour will determine how quickly AI can progress into safety-critical applications.
22 Turning the digital twin into a case for action As AI becomes more deeply embedded in industrial operations, digital twins are moving beyond asset visualisation towards a much more active role. The opportunity lies in combining trusted data, engineering knowledge and AI to understand not only what is happening, but what action should follow.
28 Building AI-ready digital twins starts with trusted inspection data Peter Rosiepen, CEO of DIMATE, explains how lessons from medical imaging are helping asset-intensive industries turn NDT inspection data into a trusted foundation for Asset Integrity, RBI and AI-ready Digital Twins.
34 From digital twin to AI-native operating system As digital twins evolve from static asset representations into AI-enabled operating environments, the real value will come from the workflows, decisions and actions they support. For oil and gas operators, that makes ownership of data, operational context and the control plane increasingly strategic. www.futureoilgas.com | Future Digital Twin & AI 03
40 Drawing the line: Decision boundaries for AI at scale
Pedro Alcântara Nunes Neto, Chief Revenue Officer at Falkor, examines one of the defining questions in industrial AI: which decisions should be entrusted to machines and which must remain with people. He argues that getting this boundary right is fundamental to designing AI systems that are both effective and trusted.
44 How AI transforms process safety management
AI has the potential to transform process safety management by moving organisations from fragmented, reactive approaches towards continuous, data-driven risk intelligence. Dan McLean, Content Marketing Manager at Wolters Kluwer Enablon, examines where AI can deliver the greatest value in EHS and PSM, and why strong digital foundations and human judgement remain essential.
48 The future of industrial AI – better and faster decisions
Hermias Hendrikse, Global Industry Development Lead at Siemens, examines how manufacturers can move beyond isolated AI pilots to achieve scalable, enterprisewide impact. He explains why connected data, digital twins and physics-based AI are becoming essential foundations for faster, more trusted industrial decision-making.
52 Separation modelling rigour: the key to unlocking hidden production capacity Tom Ralston, Digital Process Engineering Business Development at MySep Pte Ltd, explains why modelling fidelity is becoming a critical differentiator in digital twin performance. He explores how rigorous separation modelling can expose hidden production constraints, strengthen AIdriven optimisation and unlock significant operational and commercial value.
56 Trustworthy Industrial AI: Solving the black box problem in critical energy environments Dr Gerhard Koch, Chief Software and Product Officer at Siemens Energy Global, explains why explainability, stability and context are becoming essential foundations for scaling AI safely and effectively across industrial operations. 04 Future Digital Twin & AI | www.futureoilgas.com
60 From Interoperability to Intelligence: Building Digital Twins and Industrial AI on Engineering Data That Can Be Trusted Dr. Michael Wiedau, Chairman of DEXPI e.V., explains why trusted, machinereadable engineering data is essential to unlocking the full potential of digital twins and industrial AI. He explores how open standards and semantic interoperability can turn fragmented information into a reliable foundation for smarter, scalable industrial decision-making.
64 AI needs more than data: why standardized digital twins matter AI needs more than access to industrial data. Sandeep Rudra, Technical Project Manager at IDTA, explains how standardised Digital Twins and the Asset Administration Shell can provide the structure, context and interoperability needed to turn AI into reliable, scalable industrial applications.
68 From digital twin to industrial knowledge graph Arun Govind, Co-Founder & CPO of Rive Labs, Inc, explores how digital twins are evolving beyond representation into Industrial Knowledge Graphs that connect assets, systems, documents and operational data through meaningful relationships. He argues that this connected context will provide the foundation AI needs to reason effectively across the industrial enterprise.
72 Final Word: AI will not fix a digital twin you cannot trust Mark Venables argues that oil and gas operators should resist the rush towards autonomous industrial AI until they can trust the engineering context beneath it. The competitive advantage will come from building digital twins that accurately reflect the assets AI is being asked to optimise.
Digital Twins Within AI-Native Operations For years, digital twins have been defined by how faithfully they mirror physical assets. The next era will be defined by something more ambitious: integrating into the environment where operational work actually happens. That means a shared canvas, where humans and AI agents work side by side. The barrier is no longer technology. It’s trust and, critically, an organization’s readiness to change. Trust begins with ownership of your data, your workflows and the control plane that governs them.
See how EPAM builds AI-native operating systems for energy. www.epam.com.energy
FUTURE DIGITAL TWIN & AI
The next digital twin challenge is not visibility, but decision quality As oil and gas operators connect digital twins with industrial AI, the focus is shifting from representing assets to improving the decisions made around them. The next stage will depend on trusted context, engineering physics and clearly defined boundaries between human judgement and machine autonomy. 06 Future Digital Twin & AI | www.futureoilgas.com
FUTURE DIGITAL TWIN & AI
D
igital twins have spent the past decade becoming progressively better at representing complex industrial assets. For oil and gas operators, that has meant combining 3D models, engineering information, real-time process data, maintenance records and analytics into integrated views of the asset. That progress has improved visibility, but it has also exposed the next challenge: how to turn connected information into better operational decisions. A digital twin that lets an engineer locate a pump, open its datasheet and view its vibration trend is useful. A twin that can help determine why that pump is behaving differently, understand the operational consequences and support a decision about what should happen next is considerably more valuable. Maturity is therefore becoming less about how much information has been connected and more about how effectively it improves decisions. Context becomes the critical infrastructure Oil and gas companies do not lack data. The more persistent problem is maintaining consistent context around it. A compressor may exist as an engineering tag, an object in an enterprise asset management system, a collection of historian signals, a maintenance hierarchy, inspection records and a location inside a 3D model. Each system may be accurate while describing the same physical asset differently. Engineers can navigate many inconsistencies through plant knowledge, but AI cannot safely make the same assumptions. This makes the semantic layer of the digital twin increasingly important, with equipment, measurements, documents, work orders, failure modes and operating constraints connected through explicit relationships. Industrial knowledge graphs are one way of extending this capability. For an AI assistant asked why a compressor is approaching an operating limit, context determines whether its answer is useful. It may need to combine current process conditions with performance curves, maintenance history, upstream restrictions, anti-surge behaviour and the configuration installed today. Retrieval gives AI access to information; context allows it to understand how that information relates to the asset.
Physics still matters
large language models can make complex information easier to interrogate. Neither automatically understands why a separator carries liquid into a gas stream, how compressor performance changes across its map or whether an attractive operating point increases risk elsewhere. The strongest industrial AI architectures are therefore likely to combine different approaches. Data-driven models can identify anomalies and predict behaviour, while first-principles models, reduced-order models and engineering rules provide the physical boundaries within which potential interventions can be evaluated. The digital twin then moves beyond describing what is happening to testing the consequences of different decisions.
From insight to execution Many digital programmes still stop one step too early. An analytics platform identifies a deteriorating asset, a dashboard highlights the problem and an engineer receives an alert. At that point, value is often assumed to have been created, even though the economic outcome depends on what happens afterwards. Someone must determine whether intervention is necessary, assess production consequences, check spares, identify the correct maintenance procedure and find an appropriate window for action. That chain is where digital twins and AI can have their greatest operational impact because it connects insight with execution. Consider a predicted failure on an offshore injection pump. A mature digital environment should connect that prediction with asset configuration, criticality, redundancy, maintenance history, available spares, planned shutdowns and production schedule. AI can then help engineers evaluate credible options rather than simply generating another notification. One system tells the organisation that something may be wrong; the other helps determine what should be done about it and why. This is also why workflow integration will become more important than adding another visual interface. Digital twin context needs to reach the systems where work is actually executed, including maintenance management, inspection, planning and engineering change. Trusted digital context needs to be available within the workflows engineers already use.
The arrival of generative AI does not reduce the importance of engineering models. In many respects, it increases it because the most valuable decisions in oil and gas operations remain constrained by physical reality. Production cannot simply be optimised against a commercial target. It must remain within pressure, temperature, flow, integrity, emissions and equipment operating envelopes, with changes in one part of the process often creating consequences elsewhere.
The twin must follow the asset
Machine learning can identify difficult-to-detect patterns, while
For AI, this is more than a data-quality issue. A confident answer
Industrial assets do not stand still. Components are replaced, temporary modifications are introduced, control logic changes, equipment degrades, documents are revised and inspection findings alter the understanding of remaining life. A twin that accurately represented a facility at commissioning can therefore gradually become a sophisticated description of a plant that no longer exists.
www.futureoilgas.com | Future Digital Twin & AI 07
FUTURE DIGITAL TWIN & AI
based on an obsolete configuration may be actively misleading, particularly when used to support maintenance, integrity or operational decisions. Digital twin governance therefore needs to become closely connected to management of change, with the digital representation changing as the physical asset changes. Provenance, versioning and validation status become part of operational assurance. The question is whether that information is correct for this asset, in its current configuration, at the moment the decision is made.
Defining the boundary of autonomy As agentic AI enters industrial environments, the boundary between recommendation and action will become one of the most important design decisions. An agent might assemble evidence for an equipment investigation, compare similar assets across a fleet, prepare a maintenance recommendation or identify affected documents following an engineering change.
08 Future Digital Twin & AI | www.futureoilgas.com
Other decisions will continue to demand explicit human authority, particularly where safety, integrity, production or environmental consequences are significant. The boundary should not be determined purely by what an AI system is technically capable of doing. It should reflect consequence, reversibility, confidence and the quality of the evidence available. This creates a natural role for the digital twin as an assurance layer between AI and the physical operation. An AI agent should be able to show which asset configuration it used, which information informed the conclusion, which engineering constraints were applied and where uncertainty remains. Explainability means giving engineers sufficient evidence to judge whether a recommendation deserves to be trusted and acted upon.
Measure the decision, not the twin The next phase of digital twin investment will require different measures of success. Connected data sources and model completeness reveal
little about operational value. More meaningful measures include faster diagnosis, reduced engineering search time, avoided production loss, fewer unnecessary interventions and improved maintenance planning. Those measures shift attention from the technology itself to the decisions and workflows it improves. The twin provides the connected representation, engineering context and operational state, while AI helps interrogate that environment, detect patterns and evaluate options. For oil and gas operators, the opportunity is not simply to build ever larger digital representations of assets. It is to create a trusted decision environment in which engineering models, operational data, human expertise and AI can work together. The digital twin began as a way to represent the physical world. Its next role is helping the organisation make better decisions about what should happen in it.
DIGITAL TWIN
Digital Twin journey: Turning insight into action Susan Burrell, Global Product Owner Digital Twin at Shell Research Limited, explains how digital twins are moving beyond integrated asset visualisation to become part of how operational work is planned and executed. She explores why trusted digital context is increasingly important to improving decisions and ultimately changing the way work is done.
D
rawing on her experience working with digital twin technologies, data foundations and ERP, Susan shares her personal observations on how digital twins are evolving from integrated asset visualisation tools into capabilities that can support operational decision-making and execution. The views expressed in this article are her own and are based on her professional experience. Digital twins have long promised to transform how complex industrial assets are understood, operated and improved. Yet visibility alone does not create value. Value is created when insight changes decisions, decisions change workflows and workflows change outcomes. For many years, the industry has focused on integrating data and creating increasingly sophisticated virtual representations of physical assets. These capabilities remain
10 Future Digital Twin & AI | www.futureoilgas.com
important, but they are not the end goal. The next era of Digital Twin will be defined less by how accurately an asset can be visualised and more by how effectively that representation changes the way work is planned, risk is managed and decisions are executed. For Susan , this shift is becoming increasingly tangible. Digital Twin is evolving from an integrated visualisation environment into a workflow-enabled decision platform whose purpose is not simply to make information available, but to make action easier. Achieving
DIGITAL TWIN
this at scale means balancing trusted data and semantic foundations, which take time to build, with outcomes that individual assets can value today.
From integrated information to operational value Susan has seen Digital Twin capability develop from a platform for integrating and visualising asset information into an increasingly mature environment for operational awareness, engineering context and workflow support. At its core, it provides a virtual representation of physical
asset elements and their dynamic behaviour across the asset lifecycle, bringing together engineering information, operational data and asset context to help users investigate issues and make betterinformed choices. The early focus was necessarily on establishing trust. Engineers, operators, planners and maintenance teams need confidence that information is accurate, accessible, relevant and contextualised before they will use it in operational decisions. Much of the initial value therefore came from making trusted information easier to find
and creating a common view across previously fragmented systems. The experience reinforced a basic principle: a sophisticated tool does not create value simply because it has been deployed. Value emerges when people adopt that tool as part of the way they perform their work and break the journey into smaller, practical steps rather than attempting one large leap from integrated data to a completely new operating model. Asset readiness, acceptance and perception of
www.futureoilgas.com | Future Digital Twin & AI 11
DIGITAL TWIN
value, can vary significantly. Some assets prioritise risk visualisation, while others see more immediate value in isolation planning or work visualisation. The long-term product vision remains important, but the route towards it has to reflect local operational priorities, data maturity and willingness to change.
Solving problems that assets recognise The strongest and fastest value has generally emerged when Digital Twin capabilities address a “problem worth solving”: a priority recognised and defined by the asset that will use the resulting workflow. Examples include Cumulative Risk Visualisation, isolation planning and Asset Copilot. Cumulative Risk Visualisation helps teams understand the combined risk picture across an asset. Isolation planning connects engineering and operational information to support the safe preparation of equipment for work, while Asset Copilot allows users to ask contextual questions about asset information through a natural-language interface. Familiarity with copilots can lower the barrier to adoption, provided responses remain grounded in trusted and appropriately governed data. The deployment of a cockpit for Operations Cockpit has also shown how strongly asset context influences adoption. At a greenfield site, people, processes and tools were new, allowing integrated work processes to be designed into the operating model from the outset. At a brownfield site, established practices had been shaped by the asset’s history and existing systems, so the product had to be adapted to lower the threshold for
12 Future Digital Twin & AI | www.futureoilgas.com
acceptance. The lesson is clear: transformation has to begin from the reality of how people work today. Leadership support also matters. Clear sponsorship and direction make adoption easier because the capability is understood as an operational priority rather than an optional digital initiative. Data discipline also matters, because a Digital Twin depends on current, relevant and integrated information and is more readily trusted where ownership and governance are already established.
The next frontier: Digital Twin as a system of action The next phase of Digital Twin development is not primarily about richer visualisation or more integrations. The dominant shift is from systems of record to systems of insight, and increasingly from systems of insight to systems of action. For Digital Twin, that means moving from representing an asset to helping orchestrate the workflows through which people plan, coordinate and execute work. Susan describes this as the Workflow Digital Twin. Capabilities including cumulative and integrated risk management, isolation planning, dynamic bow ties, Operations Cockpit concepts and work-package support bring engineering information, operational data, risk context and work processes into a common environment. The aim is not to transfer an existing process into a new interface, but to redesign the workflow around trusted context and the decisions users need to make.
Trusted NDT Data for AI-Ready Digital Twins DIMATE helps asset-intensive industries transform fragmented NDT inspection data into structured, traceable and decision-ready information. Our digital inspection workflows connect NDT results, inspection assessments and integrity systems, creating a structured and traceable foundation for Asset Integrity, Risk-Based Inspection, Digital Twins and AI-assisted decision support.
www.futuredigitaltwin.com | Future Digital Twin & AI 13
www.dimate.de
DIGITAL TWIN
AI will be important to this transition, not because it replaces the Digital Twin, but because it can make the Digital Twin easier to use at the point of work. Natural-language interfaces, copilots and, over time, agentic workflows can reduce the friction between users and complex industrial information. Their value, however, depends on the quality of the foundations beneath them: trusted data, semantic context, appropriate governance and clear human accountability. The Digital Twin is also becoming more spatial and predictive. By connecting engineering drawings, process information, three-dimensional models, photographs and video, users can move more naturally between equipment relationships and physical reality. As simulation capabilities mature, teams will increasingly be able to explore future states through operational rehearsal, risk-scenario testing and what-if analysis before actions are taken on the physical asset.
That requires a composable architecture built around reusable data foundations, modular applications, open integration patterns and interoperability across platforms. No single technology will solve every problem, so capabilities must be assembled rapidly without losing consistency in the underlying information model. Semantics are central to this approach because they are what allow workflow components to travel. A shared language for assets, equipment, risks, processes and work allows workflows, AI assistants and decisionsupport tools to interpret information consistently across sites without forcing every asset to operate in exactly the same way. In that sense, semantics are not primarily a technical concern; they are a business scaling mechanism.
Towards an asset-centred operating layer Looking ahead, these developments are beginning to converge. Workflow Digital Twins, AI-enabled interaction, semantic interoperability, composable deployment and scenario-based operations are not separate destinations. Together, they point towards a Digital Twin that acts as the contextual operating layer around the asset, helping teams understand current conditions, anticipate possible outcomes, coordinate work and execute decisions with greater confidence.
Scaling through reuse, not uniformity One of the early assumptions Susan had to reconsider was that a successful workflow could be replicated through a largely standard deployment model. Often, In practice, there is neither a o single way of working across every asset, nor is there likely to be one in future. Scale will not necessarily come from imposing uniformity. It will come from designing for controlled variation: common foundations, reusable components and enough local flexibility for assets to adopt confidently. In one sense, this makes replication more difficult than transferring a successful product from one site to another. A workflow component developed for a pressing problem at its originating asset may arrive elsewhere where the need is perceived differently or an alternative solution is already in place. The challenge is to accommodate legitimate variation to facilitate and accelerate adoption, without allowing every deployment to become entirely bespoke.
14 Future Digital Twin & AI | www.futureoilgas.com
Digital Twin should therefore not be viewed as a standalone application. It is becoming part of a broader operating layer that connects enterprise data, operational systems, semantic context, AI capabilities and workflow execution around the asset. One lesson Susan would approach differently today is the need to secure early agreement from participating assets on adopting features being developed elsewhere. These early “handshakes” help align priorities, identify where components can be reused and maximise the value of development investment. Success will continue to depend on balancing two realities. Assets need value today, while integrated data foundations and semantic models take time to mature. The most effective approach is therefore to deliver high-value workflows incrementally while continuously strengthening the reusable foundations beneath them. The next chapter of Digital Twin is not about building a more sophisticated representation of the asset. It is about building a more effective operating model around it, connecting data, semantics, workflows, technology and people in a way that improves decisions and accelerates execution. The organisations that succeed will be those that most effectively convert insight into action and action into measurable operational outcomes.
SCALE EXPERTISE. NOT JUST TOOLS. Turn expert-led work from months and weeks into hours. UptimeAI AI Reasoning Agents bring together operational data, engineering knowledge and context to accelerate complex analysis and decisions across your operations.
200X
STOP MONITORING. START DECIDING.
FASTER ANALYSIS
2–5%
EBITDA OPPORTUNITY
uptimeai.com
www.futureoilgas.com | Future Digital Twin & AI 15
AI TRUST
Trust before autonomy: assuring industrial AI in energy As artificial intelligence moves from exploration and decision support towards operational systems, the energy sector is discovering that technical capability is only part of the equation. Building confidence in data, governance and system behaviour will determine how quickly AI can progress into safety-critical applications.
F
or energy operators, the attraction of industrial AI is easy to understand. Better reservoir interpretation, faster analysis, more effective drilling decisions and increasingly sophisticated optimisation all promise improvements in performance. But as AI moves closer to operational decision-making, the question changes from what the technology can do to whether an operator can trust it enough to act on its output. That shift is already visible in DNV’s work with the Norwegian energy sector. Lars Meloe, Principal Consultant, DNV points to recent work with power transmission system operator Statnett, where DNV has supported AI governance, including how solutions can be used safely while meeting relevant rules and regulation. “Governance, risk and trust in the solutions have risen to the top of the checklist,” he says. “AI can add value to operations and to the business, but companies need to know that they can trust it before they use it.” The growing caution does not mean enthusiasm for AI has disappeared. Rather, the industry is beginning to distinguish between successful experiments and systems that can be relied upon in real operations. “There has been a lot of focus on the added value, but not many have
16 Future Digital Twin & AI | www.futureoilgas.com
AI TRUST
succeeded in the operational phase because using AI can become high risk,” Meloe says. “I think the energy sector has taken one or two steps back and become more conservative, making sure the risk mitigation is in place before these solutions are used operationally.”
From insight to operational consequence Oil and gas already provides examples of where AI can create value without placing the technology in direct control of critical equipment. Meloe highlights reservoir analysis as one area where AI is accelerating interpretation and improving understanding of potential resources. “Equinor has used AI for reservoir work, and there are good results around finding reservoirs more quickly and analysing them more precisely,” he says. “We also see AI being used in applications and technologies within drilling operations.”
The risk profile changes significantly when an AI system moves from analysing information to influencing what happens next. DNV’s wider work argues that risk does not sit within the algorithm alone. It can emerge from interactions between the AI component, the wider system, the operational environment and the governance around it. Data quality, uncertainty, changing operating conditions, human oversight and accountability therefore become part of the assurance question.
The data problem sits at the centre of that calculation. Industrial organisations have large volumes of operational history, but quantity is not the same as suitability for AI. “Everyone struggles with the data and the quality,” Meloe adds. “You need full control of the data you are using and the data on which these AI solutions are running. The industry is spending a lot of time cleaning databases and making sure the solutions are working with data that is good enough.”
Meloe sees the cost of reaching that level of confidence as one reason operators are becoming more selective about deployment. “Someone said at a conference that AI is everywhere except on the bottom line,” he says. “Energy companies are being more careful about spending money on AI because making a solution trustworthy enough for operational use can involve a very high cost. You must look at whether the value really justifies the work needed to make it safe.”
Simulation before action Digital twins can help close the gap between AI output and operational confidence by providing a controlled environment in which decisions can be explored before they affect the physical asset. Meloe stresses that the term covers a wide spectrum. “A digital twin can be anything from a 3D model to a full simulation of the actual operation,” he says. “With the more advanced twins, you can run simulations faster than the real operation and start predicting what may happen in the future.” That capability can give operators a way to test AIsupported decisions against credible representations of asset behaviour. “Digital twins are increasingly being used to see the risk before it arrives and, where possible, avoid it by running simulation tests in advance,” Meloe says. “That can help the industry use AI solutions in a much safer way.” The lifecycle issue is equally important because an AI system does not necessarily remain the same after deployment. Models can be retrained, data can drift and operating conditions can change, so a one-off qualification provides only a snapshot. “There is no value in entering a customer one day and assuring an AI technology if it can change the day after we have been there,” Meloe continues. “We need continuous assurance solutions that enable the end user to trust the solution day by day.” DNV’s position is that assurance therefore needs to become continuous, context-aware and evidencebased, with claims about safety or performance linked to supporting evidence and revisited as the system evolves. For oil and gas companies already familiar with lifecycle risk management, the principle is recognisable. What changes is the pace at which the evidence may need to be reassessed. That wider view is important because an AI model cannot be judged solely on whether it performs
www.futureoilgas.com | Future Digital Twin & AI 17
AI TRUST
accurately in isolation. DNV’s assurance approach considers the AI component, the wider system in which it operates, its operational context and the governance surrounding its use. A model may perform strongly in testing but still introduce unacceptable risk if the data changes, if it interacts unexpectedly with another system or if operators misunderstand the confidence they should place in its output. The level of assurance also needs to reflect the consequence of failure. An AI tool providing an engineer with additional insight does not require the same level of scrutiny as a system capable of directly influencing production equipment or safety-critical controls. As greater autonomy is introduced, the evidence needed to justify that autonomy must increase with it, creating a staged path from advisory applications towards more automated operation rather than treating autonomy as a single technological leap.
Keeping the human in the loop For oil and gas, the question of autonomy makes the human role particularly important. Full autonomy may be technically achievable in some applications over time, but Meloe believes the current risk picture still makes human oversight essential in most safety-critical energy operations. “The human factor and the human in the loop are still crucial to safe operations,” he explains. “That is clearly the case in oil and gas, but it also applies in power grids and eventually in areas such as offshore wind.” The challenge is not simply to place a person somewhere in the process and call that oversight. The operator must understand what the AI is doing, where its limits are and when intervention is genuinely possible. DNV’s assurance work also warns against treating humans as an automatic fallback if they do not have the information, time or authority needed to respond effectively. Much of the foundation for managing these risks is already familiar to the energy sector. Operators routinely use technology
18 Future Digital Twin & AI | www.futureoilgas.com
www.futureoilgas.com | Future Digital Twin & AI 19
AI TRUST
qualification, safety cases, testing regimes and recognised standards to build confidence in new systems. Meloe argues that industrial AI requires an extension of that discipline rather than an entirely new philosophy. “It is more of the same in many ways,” he says. “Technology providers and operators want a second opinion on the trustworthiness of the technology and on whether the organisation itself is able to develop, deliver and stand behind its AI solutions.” DNV has developed recommended practices covering areas including digital twins, data quality, simulation models,
AI and machine learning, including DNV-RP-0671 for assurance of AI-enabled systems. The objective is to apply evidence-led industrial assurance to systems that are more adaptive, data-dependent and difficult to explain. DNV’s offshore AI safety work similarly argues that AI should be assured with the same rigour that has long underpinned major-hazard management, while adapting safety cases to capture new and less visible risks. For operators considering the next stage of industrial AI, Meloe’s message is therefore less about slowing adoption than about defining what must be true before a system is trusted. Governance, data quality, system boundaries, standards, human responsibility and continuous evidence all need to be considered alongside model performance. “There are a lot of similarities between power grids, oil and gas and other parts of the energy sector, even if the domain-specific standards and challenges are different,” he says. “The governance is quite similar, and the technology solutions need to be assured in much the same way.” AI may ultimately enable more autonomous and efficient energy operations, but the route to that future will be defined by confidence rather than capability alone. As Meloe puts it: “The industry sees a lot of added value in AI. But the question is: how can I trust my industrial AI solution? That is where risk and assurance must become part of the technology from the beginning.”
20 Future Digital Twin & AI | www.futureoilgas.com
www.futuredigitaltwin.com | Future Digital Twin & AI 21
DIGITAL TWIN
Turning the digital twin into a case for action As AI becomes more deeply embedded in industrial operations, digital twins are moving beyond asset visualisation towards a much more active role. The opportunity lies in combining trusted data, engineering knowledge and AI to understand not only what is happening, but what action should follow. 22 Future Digital Twin & AI | www.futureoilgas.com
DIGITAL TWIN
D
igital twins and AI have spent years near the top of the energy sector’s technology agenda, yet their real test is increasingly operational. A sophisticated model or powerful algorithm may demonstrate what is technically possible, but value comes when those capabilities improve the way an asset is maintained, optimised and operated. That is pushing the conversation away from technology in isolation and towards the industrial foundations required to make it useful. John de Koning, President EMEA region at Radix, points to maintenance, reliability and upstream optimisation as areas where that shift is already visible. “AI is a brilliant toolkit to help with the full contextualisation of data from all the sources you have available,” he says. “But when you look at value generation, particularly in asset maintenance and efficiency, you see a lot of use cases moving from reactive to predictive ways of operating. Large volumes of data need to be processed, and AI is an excellent toolkit for that.” The same capability can support drilling optimisation and reservoir modelling, where enormous datasets need to be interpreted quickly enough to influence decisions. “Speed of delivery is important for value generation, but so are the reliability and resilience of operating facilities,” de Koning adds.
those decisions. Radix combines newer models with firstprinciples analytics, giving engineers a reference against which an AI-generated recommendation can be tested. That becomes particularly important with generative AI, where convincing language should not be confused with engineering validity. “The good news is that AI always gives you an answer. The bad news is that AI always gives you an answer,” de Koning continues. “As an operator or engineer, how do you know that the answer or instruction set you are receiving is valid? You need reference points, and that is why we still use a lot of traditional firstprinciples engineering.” Once the underlying industrial information has been integrated and contextualised, conversational AI can also change how users interact with it. Instead of searching through documentation and multiple applications, an engineer can ask questions in natural language about the current performance of a compressor or gas turbine, its operating window and potential optimisation opportunities. “You can pinpoint from a very wide range of information to the specific area where you need to focus,” he explains. “It helps drive you towards the information and the actions you need rather than having to look through all the documentation yourself.”
From representation to operational tool The digital twin itself has undergone a similar evolution. Early implementations frequently centred on engineering information and 3D visualisation, providing users with a clearer representation of how a facility had been built and how its equipment fitted together. “The value really started once we added the day-to-day data,” de Koning says. “When you bring in operational data, maintenance records and other information, you can start building relationships, analysing those relationships, identifying trends and making things more predictable. That is where AI comes in.” For Radix, the next step is what de Koning describes as the “case for action”. Better access to information is valuable, but an operational system should also help users determine whether a changing condition matters, understand its likely cause and decide what should happen next. Rotating equipment provides a practical example. “You want to understand what the anomalies are, what is changing in the operational behaviour, what is good and what is bad, but most importantly, what actions need to be taken,” he adds. “We work a lot with rotating equipment to identify anomalies, whether they are technical or process-related, understand the failure modes and identify the actions required to bring the equipment back to normal.” AI does not replace the established engineering science behind
This combination of engineering, operational technology, data and software expertise also shapes how Radix approaches technology selection. Existing systems are retained where they continue to provide useful capabilities rather than replaced simply to accommodate a new digital programme. “We first look at how you are operating, what your business challenges are, where you want to be and what capabilities you need to get there,” de Koning says. “Then we look at what technologies can provide those capabilities, what you already have in-house and where the gaps are.”
Building trust in the data That approach is particularly relevant to brownfield oil and gas assets. Historians, maintenance systems, engineering databases, control platforms and shift reports can contain decades of valuable information, but the data was rarely created with an enterprise digital twin or AI programme in mind. Radix is frequently brought into projects after a proof of concept has demonstrated technical potential, but the operator has encountered difficulties taking it further. “The data sources need to be accessible, the data has to be of good enough quality, and then you need to integrate and aggregate it,” de Koning explains. “Once you have that integrated data, you can contextualise it and create the links between the different sources. That gives you the core data layer for a digital twin.”
www.futureoilgas.com | Future Digital Twin & AI 23
DIGITAL TWIN
The work does not finish when those connections have been established. Source systems continue to change; new information is created and relationships between datasets have to remain current. A technically impressive implementation can quickly lose operational credibility if users begin finding information they know to be wrong. “We often see end users reach a point where they no longer trust the data because failures start appearing,” he says. “That usually comes back to the maintenance processes around the source data, the integrations and the contextualisation. The change-management workflow around that is one of the most essential parts of the implementation.” For facilities with different generations and brands of technology, Radix often introduces a middle layer between plant systems and the applications above them. This allows the enterprise architecture to accommodate the realities of the installed base without requiring a wholesale technology replacement. “You can standardise the data model and the accessibility for the use cases built on top,” de Koning says. “When you scale to another facility, you may connect to a different brand of control system or historian, but from the usecase perspective the connectivity remains the same. That is an essential part of scaling quickly.”
Scaling beyond the pilot A successful pilot can expose a different challenge when an operator attempts to reproduce it across assets with different systems, configurations and data structures. The objective is not to make every facility identical, but to minimise how much of the solution has to be rebuilt each time. “The more you can standardise the data model, the more it becomes reusable across facilities and use cases,” de Koning says. “The second part is interoperability: using industry standards to connect to different data sources. That makes the underlying systems easier to access while allowing the use cases above them to remain consistent.” Radix is working with industry partners on this interoperability challenge, including efforts to make source systems and engineering information more accessible through standard interfaces. In the meantime, platforms such as AVEVA Connect and Cognite can provide the middle-layer technology needed to separate enterprise use cases from individual plant-level systems. The harder part of scaling may ultimately have less to
24 Future Digital Twin & AI | www.futureoilgas.com
Structural Integrity Management Breathing new life into ageing offshore facilities At Rosenxt, we look far beyond tomorrow. We are a niche, next-generation offshore structural integrity management (SIM) specialist service provider offering pragmatic engineering & consultancy services worldwide. As existing offshore facilities approach or exceed their original design life, traditional assessment methods built on outdated standards often struggle to cost-effectively demonstrate continued safety and integrity. We adopt a future ready, risk-informed approach towards fitness-for-service assessment, life extension and facilities reuse/repurposing projects. We offer cutting edge SIM services that are powered by an advanced assessment toolkit, powerful Risk-based Inspection (RBI) methodology and digital twin solutions. We at Rosenxt, deliver enhanced asset safety, tangible OPEX savings and the highest levels of technical assurance borne out of ex-operator experience. Backed by a lean team of integrity specialists, we eliminate waste without compromising rigor and deliver value to our customers.
www.rosen-nxt.com
DIGITAL TWIN
do with software. Oil and gas operators are understandably reluctant to act on recommendations produced by a black box, particularly where equipment reliability, production, safety or environmental performance could be affected. “The technology is not the challenge anymore. Most of it is available off the shelf,” de Koning says. “The essential part is people change and business change: helping operators and engineers understand how to use the output of a digital twin or AI model, how it fits into
their work processes and why an action is being recommended.” More automation can eventually take that logic further. Better sensor coverage, accessible operational data and engineering models allow anomalies to be identified earlier and responses to become more standardised and predictable. Newer facilities can embed more autonomous capabilities from the outset, while brownfield sites will follow a more gradual route as existing systems are connected and modernised. “The more stable you operate, the more you reduce risk, increase safety and reduce environmental impact,” de Koning says. “Automation also gives you more standardised and predictable actions, and that is what moves you towards autonomous operations.” For Radix, that journey reflects its broader “vision to value” proposition. Digital twins, industrial data platforms and AI are components of the solution rather than the starting point, with technology selected according to the operational problem it needs to solve. “We are not technology pushers; we are problem solvers,” de Koning concludes. “The key is to understand where the challenges are, how they can be addressed, and then leverage technology to implement the solution.”
26 Future Digital Twin & AI | www.futureoilgas.com
Securing the future of your floating assets.
Calum MacLean Projects Director
John Walden CTO
Visit our website
Meet MTL to discuss PYXIS and our end‐to‐end digital inspection tools, digital twin, scope creation, offshore capture and fast reporting, plus VR, visualisation and engineering‐led www.futureoilgas.com | FutureAI. Digital Twin & AI 27
AI-READY DIGITAL TWINS
Building AI-ready digital twins starts with trusted inspection data Peter Rosiepen, CEO of DIMATE, explains how lessons from medical imaging are helping asset-intensive industries turn NDT inspection data into a trusted foundation for Asset Integrity, RBI and AI-ready Digital Twins.
T
his interview with Peter Rosiepen, CEO of DIMATE, explores how lessons from medical imaging are helping asset-intensive industries build trusted inspection workflows for the age of Digital Twins and AI. The conversation addresses some of the industry’s biggest challenges: fragmented inspection information, manual corrosion monitoring workflows, and the lack of continuity between NDT results, engineering documentation and integrity systems. Rather than focusing on AI alone, it explains why structured inspection workflows provide the foundation for reliable Asset Integrity, Risk-Based Inspection (RBI), Digital Twins and future AI applications. Using practical examples - including AI-assisted RT wall thickness evaluation, Corrosion & CML Automation, digital Component Lifecycle Files and structured collaboration between owner-operators, NDT service providers and EPC companies - the interview illustrates how DIMATE helps organizations transform inspection information into validated, decision-ready data that can be integrated with IDMS, AIMS, RBI and Digital Twin environments.
28 Future Digital Twin & AI | www.futureoilgas.com
AI-READY DIGITAL TWINS
Peter, DIMATE has a rather unusual origin story. How did a company focused on industrial NDT software emerge from the healthcare imaging world? Peter Rosiepen: The origin of DIMATE is indeed closely connected to healthcare imaging. My business partner Jens and I came from the healthcare sector and from our work with VISUS, where Picture Archiving and Communication Systems (PACS) played a central role in digitalizing diagnostic imaging workflows. In healthcare, PACS fundamentally changed the way medical images, findings and related information are managed. Instead of working with film, paper and disconnected systems, hospitals were able to archive, communicate, compare and evaluate imaging data within a structured digital workflow. We sometimes describe this with a smile as the “non-destructive testing of humans.” In medicine, imaging technologies are used to understand the condition of a patient without damaging the body. In industry, NDT serves a very similar purpose by assessing the condition of pipelines, vessels, welds and other critical assets without interrupting operation.
That parallel was the starting point for DIMATE. What made you believe that the industrial market needed a PACS-like solution? Peter Rosiepen: The key moment came when we discovered that the industrial NDT community had adopted concepts from the medical DICOM standard and established them as DICONDE (Digital Imaging and Communication in Non-Destructive Evaluation). For us, this was a strong indication that the industry recognized the need for structured, interoperable and long-term accessible inspection information. When we analyzed the market, we found inspection equipment, reporting tools, document repositories, inspection management systems and asset integrity platforms. What was missing was an industrial PACS that could bring together inspection images, measurements, reports and related documentation in a structured workflow. That was the opportunity we saw. In 2018, DIMATE was established as a spin-off from VISUS. We acquired the usage rights to the PACS software solution from VISUS and adapted the technology to the specific requirements of industrial NDT workflows. What was one of your first industrial applications? Peter Rosiepen: One of our first customers was Applus RTD in Germany. Together, we digitalized RT inspection workflows at their customer site, Shell Rheinland in Cologne. The project showed that the real industrial challenge was not image storage but workflow continuity: connecting inspection orders, images, reports and review processes so inspection data remains available for later integrity decisions. Building on those lessons, we continued to develop DIMATE PACS for industrial NDT. Today, we support customers such as bp in digitizing inspection workflows for refineries and other asset-intensive environments. Why is NDT inspection data becoming so important for Digital Twin and AI initiatives? Peter Rosiepen: Building AI-Ready Digital Twins starts with trusted inspection data. Many Digital Twin initiatives focus on engineering models, process information and sensor data. These provide valuable operational context, but inspection information provides the physical evidence of an asset’s actual condition. Wall thickness measurements, corrosion findings, repairs and inspection history are therefore essential for integrity decisions. If this information is fragmented across reports, spreadsheets, local archives or contractor handovers, it becomes difficult to use consistently within Asset Integrity, RBI and Digital Twin environments. AI and Digital Twins do not simply require more data - they require inspection information that is structured, accessible and understood in its engineering context. Interviewer: Where do asset owners typically struggle most with inspection data? Peter Rosiepen: One of the most common challenges is corrosion monitoring. Operators routinely collect large volumes of wall thickness measurements from UT and RT inspections at Corrosion Monitoring Locations (CMLs) or Thickness Monitoring Locations (TMLs). These measurements form the basis for integrity assessments and maintenance planning. The challenge is not collecting the data but preserving its context. As inspection results move between reports, spreadsheets and integrity systems, the connection to the original inspection, component history and engineering context is often weakened, creating manual effort and increasing the risk of inconsistencies. For integrity engineers, a wall thickness value alone is rarely sufficient. They also need access to the underlying inspection image, inspection report, previous measurements and component history before making an informed decision. How does DIMATE address this challenge?
AI-READY DIGITAL TWINS
Peter Rosiepen: One common use case we support is the automation of Corrosion & CML workflows. In many organizations, corrosion monitoring data is still transferred manually from inspection reports into IDMS, AIMS, RBI or Digital Twin systems. This process is time-consuming, increases the risk of transcription errors and often delays the availability of validated inspection data for integrity decisions. DIMATE supports this workflow by keeping inspection results connected to their images, reports, component information and inspection history throughout the entire process. Instead of extracting wall thickness measurements from reports, validated CML/TML data can be transferred directly into downstream integrity systems. The result is a more efficient inspection workflow with less administrative effort, improved traceability and greater confidence in the data used for corrosion monitoring and maintenance planning. Where do DIMATE AI solutions fit into this workflow? Peter Rosiepen: We see AI as an extension of trusted inspection workflows rather than a replacement for engineering expertise. One practical use case is the automated evaluation of wall thickness from RT inspection images. Reviewing large numbers of radiographs is timeconsuming, particularly for extensive inspection programmes. AI can support inspectors by identifying measurement locations and generating structured wall thickness results that engineers subsequently review and validate. For us, transparency is essential. AI-generated results remain linked to the underlying inspection images, reports and component information, ensuring that every result can be reviewed and understood. In safety-critical industries, that level of traceability is just as important as automation itself. You also developed DIMATE CCM. What role does it play? Peter Rosiepen: DIMATE CCM (Component Content Management) extends the concept of structured inspection information beyond NDT data. In most industrial organizations, information about a component is distributed across multiple systems. Inspection data, drawings, certificates,
30 Future Digital Twin & AI | www.futureoilgas.com
Smarter decisions start with connected data Empower your teams with AI-driven digital twin and integrated information management.
As a global software company, we make complex engineering and operational data simple so you can focus on results.
www.cadmatic.com
AI-READY DIGITAL TWINS
reports and historical documentation are stored separately, although integrity engineers need this information together to make reliable decisions. The Component Lifecycle File brings these sources together in a structured digital component record. By combining inspection information with engineering documentation, it provides the context needed for Digital Twins and helps engineers understand not only a component’s current condition, but also how it has evolved over time. How can NDT service providers and EPC companies benefit from this approach? Peter Rosiepen: Digitalization is not only changing the way asset owners manage inspection information. It is also changing expectations towards inspection contractors and EPC companies. Many operators no longer want to receive inspection results solely as PDF reports or document handover packages. They increasingly expect structured information that can be integrated directly into their Asset Integrity, maintenance or Digital Twin environments. This requires closer collaboration between operators and their service providers.
32 Future Digital Twin & AI | www.futureoilgas.com
One important use case is the structured exchange of inspection information between contractors and asset owners. Instead of delivering inspection results as isolated documents, validated inspection information and the associated documentation can be provided in a structured, reusable format that supports downstream integrity and maintenance processes. For service providers, this creates an opportunity to offer additional value beyond the inspection itself. For operators, it reduces manual processing and improves the quality and availability of inspection information throughout the organization. What will DIMATE present at Future Digital Twin & AI Amsterdam? Peter Rosiepen: Our focus will be on demonstrating how inspection information becomes the foundation for Digital Twins and AIsupported integrity management. Visitors will see how DIMATE connects the complete inspection workflow - from managing UT and RT inspection data, through AI-supported RT wall thickness evaluation, to transferring validated inspection information and engineering
documentation into IDMS, AIMS, RBI and Digital Twin environments. Rather than presenting individual software products, we want to demonstrate how these connected workflows support more efficient integrity management throughout the asset lifecycle. What is the one question visitors should ask themselves when they think about AI and Digital Twins? Peter Rosiepen: The first question should be a simple one: Do we really trust the inspection information behind our Digital Twin? If the answer is uncertain, organizations should first ensure that inspection information is structured, connected and accessible across images, measurements, reports and engineering documentation. Only then can engineers - and AI use it with confidence. That has been DIMATE’s guiding principle from the beginning: better integrity decisions start with trusted inspection information. With that foundation in place, Digital Twins and AI can deliver their full potential.
Site context for enterprises. Visual Operations brings together industrial assets, enterprise systems, and spatial reality data to create actionable operational intelligence.
Turning Site Reality into Operational Intelligence Unified 3D visualization, asset data integration, and spatial context
to enable AI-assisted planning, predictive insights, and automated recommendations. Operational intelligence: The critical spatial context that will enable you to make better decisisons faster.
ASSET INFORMATION
+
ENTERPRISE DATA
+
SPATIAL REALITY
→
VEERUM OPERATIONAL CONTEXT
Connect systems. Understand reality. Operate intelligently. Learn more at veerum.com
DIGITAL TWIN
From digital twin to AI-native operating system As digital twins evolve from static asset representations into AI-enabled operating environments, the real value will come from the workflows, decisions and actions they support. For oil and gas operators, that makes ownership of data, operational context and the control plane increasingly strategic. 34 Future Digital Twin & AI | www.futureoilgas.com
DIGITAL TWIN
D
igital twins have been sold for years as increasingly sophisticated representations of industrial assets. The next phase will be defined less by how accurately they reproduce the physical world than by whether they become the environment through which operational work is carried out. That distinction matters because the term itself has become so broad that almost anything can now be described as a digital twin. Berdiyar Saurbayev, Managing Principal, Head of Operational Transformation at EPAM[JH1.1], has seen interpretations range from fully contextualised asset environments to little more than dashboards. “Some companies still have Excel-based tools with simple graphs and call that a digital twin,” he says. “Traditionally, the understanding has been that you digitise the data, contextualise it, place that data layer on top of a 3D model and visualise the asset, and this is where most operators stop.” The problem is not that 3D models or visualisation lack value. It is that they are often treated as the destination rather than an enabling layer. Oil and gas operators already have established systems for finding engineering documents, executing work orders, planning maintenance and managing turnarounds. Unless the twin improves those workflows, users have little reason to change the way they work. “A company can buy a digital twin as a platform and still not use it,” Saurbayev says. “It becomes a shiny 3D object sitting there, while operators continue using the systems they have worked with for years. What is missing are the workflows on top of the twin with the control layer and the management of change around how people operate.”
From visualisation to workflow Rather than measuring a twin by the sophistication of its interface, the value case is better tested against the operational decisions it changes. Turnaround planning provides one example. Operators may use Primavera for scheduling, ERP systems for maintenance and materials, engineering platforms for asset information and separate 3D environments for visualisation. Each may perform its own task well, but the overall workflow remains fragmented. “Imagine an AI-native turnaround where you can see the asset through the digital twin, bring in the schedules, connect information from Primavera and SAP, identify simultaneous work on equipment and create the work packs in the same environment,” Saurbayev adds. “An AI agent could then help optimise the schedule, show different scenarios and support the operator in making the decision. That is very different from simply having a 3D model.” Turnarounds, maintenance and inspection may look similar
DIGITAL TWIN
across the sector, but individual operators have different processes, applications and operating philosophies. The enterprise twin, therefore, needs enough common architecture to support reusable capability while remaining flexible enough to reflect how each organisation works. This is also why EPAM views the digital twin as a technology stack rather than a single product, with each layer providing the foundation for the next.
Build the stack in the right order The starting point is data. Decades-old assets can contain fragmented engineering records, inconsistent tags, unstructured documents and information spread across multiple systems. Before advanced use cases are added, operators must first establish a clear picture of the data they hold, put governance in place and prepare it for digital use. “If you put garbage in, garbage comes out,” Saurbayev says. “You need the data foundation first. That means inventorying the data, making sure the governance is there, having the catalogues and engineering documents in order and understanding what data you are going to use.” Digitisation comes next, but simply converting documents into machine-readable form is not enough. The depth of digitisation has to match the intended use case. A use case built for document search through tags, for example, may require less detail than one intended to support an AI-enabled HAZOP orchestration or another more complex engineering workflow. The next layer is contextualisation: connecting equipment, documents, sensor information, maintenance history and other operational data so digital systems understand the relationships between them. The practical difference becomes clear when AI enters the picture. “If you ask an agent how to repair a pump without context, it can only give you a generic answer,” Saurbayev explains. “With the context layer, it understands exactly which pump you mean. It can find the correct vendor manual, check SAP to analyse maintenance history or to see whether the spare parts are available to execute the work. It
36 Future Digital Twin & AI | www.futureoilgas.com
could even initiate a work order, subject to human approval.” Once that layer is established, 3D visualisation can provide a useful interface for navigating the asset and its associated information. Saurbayev regards visualisation as a mature capability that operators can generally buy rather than something they need to build themselves. Above it sit AI-native workflows supporting maintenance, inspection, turnaround management, production planning and optimisation, followed by an orchestration layer capable of bringing those different activities into the same operating environment. “This is where the digital twin goes from a legacy system to an AInative operating system,” Saurbayev says. “The use cases are where the value sits because they help people solve day-to-day operational problems. Then you put the agentic orchestration on top so it becomes one control plane through which different users can manage the asset.”
Own the control plane The shift towards agentic AI makes the build-versus-buy decision much more strategic. Commodity capabilities can come from the market, while ownership needs to be retained over the areas that define how the organisation operates. “The operator has to own the data, the workflows and the control plane,”
Engineering Excellence Through Knowledge Driven by AI innovation, powered by data
Quality. Safety. Compliance. Delivered. Experience the Integrity platform, empowering teamwork, efficiency, and safer operations with validated engineering insights.
www.hazid.com
DIGITAL TWIN
Saurbayev continues. “You do not want to outsource your own ways of working and lose interoperability, because once the business is tied to a specific platform, the switching cost can become very high.” At the same time, the market supplying models and agentic capabilities is developing far more rapidly than the traditional lifecycle of industrial technology. “You want the ability to plug and play,” he adds. “If something better, faster or cheaper becomes available, you should be able to use it. You do not want to put your whole business and your future flexibility in the hands of one vendor.” The challenge is that operators’ adoption and procurement cycles can run from months into years, while the technologies they are evaluating may change within weeks. Keeping pace will require those processes to become more flexible.
38 Future Digital Twin & AI | www.futureoilgas.com
AI is also changing the economics of software development. Workflows that once required substantial manual effort can now be built much more quickly, making it possible for operators to recreate previously licensed applications at a fraction of the cost and bring them into the digital twin’s control plane as native workflows. Lowercomplexity processes such as shift handovers and electronic permit-to-work provide a natural starting point, where the risks are more contained and the benefits can be realised more quickly.
also goes a long way towards reducing adoption friction later on.”
“The key step is to identify and build the capabilities that can be reused across use cases,” Saurbayev says. “Things like generating and optimising schedules, producing reports, running approval workflows. These are the building blocks. Identifying those use cases has to begin with the business, and specifically with the end users, which
The same changes are influencing how EPAM approaches delivery. Saurbayev points to its Dark Factory, which brings together product engineering, domain expertise, data and AI within a single delivery model, as a way of accelerating the development of use cases and reducing the time required to put them into operation.
That also places greater emphasis on understanding where emerging technologies are best applied. Building a prototype may be relatively straightforward, but the total cost of ownership can be less apparent at the outset. Operators therefore need to consider not only the capabilities and limitations of frontier technologies, but also their economics and suitability for individual use cases.
www.futureoilgas.com | Future Digital Twin & AI 39
AI AT SCALE
Drawing the line: Decision boundaries for AI at scale Pedro Alcântara Nunes Neto, Chief Revenue Officer at Falkor, examines one of the defining questions in industrial AI: which decisions should be entrusted to machines and which must remain with people. He argues that getting this boundary right is fundamental to designing AI systems that are both effective and trusted.
40 Future Digital Twin & AI | www.futureoilgas.com
AI AT SCALE
T
ake a compressor trending toward failure, flagged by an anomaly model 200 hours before the compressor fails. The detection is agent-led work: pattern recognition across thousands of sensor readings, done faster and more consistently than any person could manage across a full shift. But the decision to isolate that compressor or keep it running through a planned maintenance window needs a trained engineer’s judgement, built over years and applied to a situation the agent probably has not seen before. That is the challenge with industrial intelligence: deciding where agent authority should start and stop.
Agent-led or human-led? Agent-led work makes sense where tasks are high-volume, time-critical and bounded by clearly defined rules or operating constraints. In our industry, that could include: • Anomaly detection across hundreds of sensor readings • Predictive maintenance signals ranked by failure probability • Correlating operational data to identify emerging issues • Calculating and optimizing operating parameters • Scheduling and prioritizing work within defined constraints Human-led decisions are different. These include: • Safety-critical calls involving shutdown, isolation or personnel • Decisions that materially change the operating envelope or accept operational risk
• Situations with no reliable precedent or pattern to draw on • Exceptions where context, uncertainty or consequence demands experienced judgement Between the humans and the agents lies the decision boundary that we aim for: a person supervising the system, accountable for outcomes, intervening by exception rather than by default. After all, approving every single agent output would defeat the purpose of having the agent in the first place. Instead, humans define the operating envelope and agents operate within it. Exceptions, uncertainty and decisions outside that envelope escalate to accountable people.
From in the loop to on the loop Human-in-the-loop means that a person must approve every operational decision before it is acted on. It is the right model for occasional, highstakes recommendations, and it is the model most industrial organizations are still running today (for good reason). But at scale, it becomes a bottleneck. On an asset generating thousands of signals an hour, requiring sign-off on every decision slows the system down before it can deliver value. Human-on-the-loop is the alternative: not approving everything, but watching for exceptions and staying accountable for outcomes on both sides of the boundary. As confidence, evidence and capability improve, that boundary can move. More decisions can become agent-led but autonomy should expand only where the operating envelope, escalation criteria, traceability, and accountability are clear. This is a standard we build into our own agentic AI tools - every response traceable, every
www.futureoilgas.com | Future Digital Twin & AI 41
AI AT SCALE
recommendation accountable to a named user, models that never retain a customer’s data. Important to note: getting to human-on-the-loop depends on having the basics for industrial intelligence in place today. The path to scale requires a solid foundation, then decision boundaries, and finally more AIled autonomy where evidence supports it.
Proof in the field Norske Shell’s Ormen Lange gas field and its onshore plant at Nyhamna, independently assessed by S&P Global Energy* earlier this year, shows what that groundwork looks like in practice. What they have today is a single, contextualized, real-time picture of the asset delivered through a digital twin built on Falkor’s platform: a basis that lets systems calculate and optimize while giving operators the information they need to make operational decisions. *Abbink, O. (2026). A most advanced digital twin: Optimizing gas production from reservoir to dispatch point. S&P Global Energy, Strategic Report, July 2026. The MEG problem is a good example. Monoethylene glycol prevents hydrates forming in deepwater pipelines. One cubic metre injected displaces roughly 2,000 cubic metres of gas. Overinjecting pours away production, underinjection risks a blockage. Limited flow control at each wellhead made hitting the precise rate genuinely hard. Falkor and Norske Shell developed an optimization workflow inside a digital twin built on the Falkor platform, processing live model inputs and returning injection parameters in seconds. The system calculates, and the control room operator decides. Even without agentic AI, the results are specific enough to matter: • A 1.3% increase in gas output from optimizing chemical injection across the wellheads • A 1.8% reduction in electricity consumption from tuning compressor operation against real-time data • Maintenance tool time, the actual proportion of a shift spent doing the job rather than finding the information needed to do it, up from around 45% to over 80%. Every one of these results still came from a person deciding, but with better information than before. Ormen Lange has not moved the decision boundary yet – but when they are ready to do so, it is made safer by the foundation already built.
42 Future Digital Twin & AI | www.futureoilgas.com
What happens when the boundary moves too quickly Ford offers a useful warning from another assetintensive industry. The company replaced veteran quality inspectors and engine specialists with AI systems, expecting the technology to uphold product quality standards on its own. What happened? Quality dropped to the point where Ford ended up rehiring around 350 of those same specialists, this time as consultants, at rates well above what they had been paid as employees (“Ford Has Been Rehiring Quality Inspectors After AI Fell Short”, Bloomberg, June 25, 2026). Is this AI failing? No. The technology did what it was built to do. The lesson here is that AI can extend expertise across far more data and far more decisions, but only if the underlying knowledge, context and accountability workflows are correct first. Ford collapsed the boundary before the foundation was built – they are still paying the cost of rebuilding it. Ormen Lange laid the foundation first, and is already seeing the return.
Operator or agent, the foundation comes first Today, the system calculates and the operator decides. Somewhere in the future, agents will make more operational decisions themselves. Getting there starts with the foundation long before the agent.
Industrial Digital Twin Industrial Digit Association Association
make Digital Twins We make Digit ASSETWe ADMINISTRATION SHELLDigital (AAS) – Industrial Industrial Digital Twin Twin Association Association THE STANDARD FOR THE DIGITAL We We make make Digital Digital Twins Twins TWIN We shape the future of the Digital Twin together.
We implement industry requirements for the Digital Twin.
We shape the future of the Digital Twin together.
W re fo
WeWe shape shape thethe future future of the of the Digital Digital Twin Twin together. together.
WeWe connect industry We implement implement industry industry know-how for requirements requirements a common solution. for for thethe Digital Digital Twin. Twin.
We establish an international standard.
We connect industry know-how for a common solution.
W an st
WeWe connect connect industry industry know-how know-how for for a common a common solution. solution.
We establish establish WeWe demonstrate an an international international valuable use standard. standard. cases.
We demonstrate valuable use cases. www.futureoilgas.com | Future Digital Twin & AI 43 www.idtwin.org
FUTURE DIGITAL TWIN & AI
How AI transforms process safety management AI has the potential to transform process safety management by moving organisations from fragmented, reactive approaches towards continuous, data-driven risk intelligence. Dan McLean, Content Marketing Manager at Wolters Kluwer Enablon, examines where AI can deliver the greatest value in EHS and PSM, and why strong digital foundations and human judgement remain essential.
A
rtificial intelligence (AI) is gaining traction in environment, health, and safety (EHS) and process safety management (PSM), but AI is only as strong as the quality, completeness, and accessibility of data consumed. Many organisations still rely on paperbased records, siloed systems, and unstructured data. This impacts AI’s ability to generate meaningful insights for EHS and PSM. According to research conducted by Wolters Kluwer Enablon and the National Safety Council that surveyed 1,053 safety and operations professionals in the US, only 11% say their safety processes are fully digitalised and integrated. A considerable 55% still rely heavily on manual and paperbased safety systems. While AI can still be applied in partially digital EHS/PSM environments, use cases tend to focus on productivity improvements. For example, natural language processing (NLP) can support incident reporting, audit analysis, document classification and retrieval, trend identification in historical lagging
indicators, and for reporting or regulatory submission drafting. While valuable, these applications primarily reduce administrative burden rather than fundamentally transform risk management.
A necessary digital foundation Unleashing the full power of AI requires EHS and PSM processes to become digitally integrated and data driven. This allows AI to support drastic organisational improvements in understanding, managing, and mitigating workplace risk. A digital foundation offers a more dynamic platform for providing continuous, real-time views of operational risk across people, processes, and assets. When PSM and EHS systems are fully digital, both structured and unstructured data become accessible and useful. AI can then continuously analyse this data to identify patterns, detect emerging risks, and correlate signals across operations to see early signals such as outdated or degraded safeguards, and quickly recognise recurring near misses, permit deviations, training gaps, or changes introduced through management of change. With this foundation, organisations can move
44 Future Digital Twin & AI | www.futureoilgas.com
FUTURE DIGITAL TWIN & AI
beyond periodic reviews and fragmented reporting towards ongoing risk awareness. Leaders gain confidence that workplace safety is being managed systematically, operators receive clearer guidance at the point of work, and safety professionals can prioritise interventions based on forward-looking indicators instead of lagging outcomes.
Improved and proactive management In traditional PSM programs, much effort is spent on periodic reviews of procedures, hazard analyses, inspections, and incident records. While essential, these processes are often conducted in isolation, allowing risks to accumulate and go unnoticed between review cycles. AI transforms PSM by driving continuous analysis and potentially highlighting early signs of weakness across safeguards, human performance, and
operating conditions. AI models can ingest data from control systems, maintenance records, safe work permits, training history, management of change requests, and incident reporting systems, then detect patterns that show rising risk. Operational leaders can then be alerted to intervene earlier, particularly when corrective actions might be less disruptive to the business, and far less costly to implement. In PSM, the integrity of alarms, interlocks, relief systems, procedures, and human actions is critical. AI can analyse test results, bypass frequency, alarm behaviour, and maintenance history to identify degrading barriers or those being relied upon too heavily. AI can support human factors management, which is a frequent contributor to incidents. By analysing patterns in permits, training compliance, shift handovers, and incident narratives, AI could identify conditions where human error is more likely to happen, such as during high workload periods, night shifts, complex temporary changes, or inconsistent procedural adherence. This insight allows leaders to adjust staffing, training, or work planning proactively rather than reacting after an error occurs. AI-enabled PSM potentially puts ongoing risk awareness into routine operations. Operations managers gain clearer foresight into where hazards are increasing, why controls may be under strain, and what actions will most effectively prevent escalation. The likelihood of catastrophic incidents could be significantly reduced and there’s potential for significant improvement in factors including operational stability, regulatory confidence, and trust between leadership and the workforce.
What the research says Research from our 2026 survey of US safety professionals supports the idea that AI has exciting potential as a
www.futureoilgas.com | Future Digital Twin & AI 45
FUTURE DIGITAL TWIN & AI
risk tool for core PSM activities. We asked the 1,053 survey participants how AI could benefit EHS processes in general. A total of 84% rank risk assessment and hazard identification as a “very high” potential benefit, 83% said AI benefits could be “very high” when applied to predictive maintenance and asset safety, and 77% said likewise for incident reporting and investigation. Confidence in AI for what might be considered more human-centric domains such as training and workforce safety engagement (76%), health surveillance and ergonomics (74%) and emergency response planning (76%) received fewer “very high” potential endorsements. Our research reveals respondents are cautious about AI’s role in human behaviour, emergency decision-making, and experiential learning, all of which are critical to PSM, but harder to formalise and validate. Additionally, 51% of respondents said they worry about overreliance on AI instead of human judgment. A total of 38% said they have concerns over the lack of transparency in how AI makes decisions. Our research concludes that organisations expect AI to support rather than replace engineering and operational judgment, and that AI trust is highest where it enhances rigour, consistency, and signal detection, rather than human decision authority. Among AI’s clearest values in PSM appears to be as a risk intelligence layer across processes using highly hazardous chemicals, helping teams see hazard accumulation and barrier erosion sooner.
46 Future Digital Twin & AI | www.futureoilgas.com
DEXPI
® Data Exchange in
the Process Industry
Seamless Data Exchange Across the Plant Lifecycle
OUR MISSION: We develop and promote a common data exchange standard for the process industry, simplifying data exchange from concept to operation.
Boost Engineering and Maintenance
Unify Your Data Ecosystem
Empower Collaboration
DEXPI.ORG OPEN • VENDOR-NEUTRAL • INDUSTRY-DRIVEN
DEXPI e.V. Heinrich-Hoffmann-Straße 3 60528 Frankfurt am Main
www.futureoilgas.com | Future Digital Twin & AI 47
FUTURE DIGITAL TWIN & AI
The future of industrial AI – better and faster decisions Hermias Hendrikse, Global Industry Development Lead at Siemens, examines how manufacturers can move beyond isolated AI pilots to achieve scalable, enterprise-wide impact. He explains why connected data, digital twins and physics-based AI are becoming essential foundations for faster, more trusted industrial decision-making.
I
t’s no secret that we are speeding towards a future that includes industrial AI. Most companies have accepted this and are at least already experimenting with AI in isolated pilot projects. The real challenge is transforming these fragmented AI initiatives into scalable, enterprise-wide impact while protecting the company’s data, security and reputation. McKinsey & Company research shows that while 88% of organizations use AI in at least one business function, true enterprise-wide success is rare. Just 6% qualify as “high performers” achieving significant bottom-line value. How do you transform fragmented AI initiatives into measurable operational impact? The answer is a connected foundation where: • Data is contextualized across the enterprise • Engineering and operations are connected • Digital Twins become the system of understanding • Physics-based AI makes insights trustworthy • Open ecosystems accelerate adoption and democratization
• AI becomes an embedded decision-making capability, not a standalone technology Companies have access to a wide range of industrial software, automation technology and R&D facilities that generate a gold mine of data. But a nice-looking dashboard is not the end goal. Who is making sense of that data – and how long does it take before specific decisions surface? Industrial AI can speed up contextualization and decision-making.
Engineering AI Siemens is already helping companies do what they are already doing faster and more efficiently. By introducing AI capabilities like ROM AI, Physics AI and embedded AI capabilities in the Siemens Xcelerator Digital Twin solutions – spanning Teamcenter, Simcenter and Opcenter and others – users are empowered to work with faster engines and improve overall productivity for smarter decision-making. In turn it becomes a learning accelerator. For example, with Physics AI, simulations are accelerated by one thousand, allowing engineers to combine the virtual and physical world to evaluate and optimize multi-scale, multi-domain innovations at an unprecedented speed. In addition, embedding AI and co-pilots seamlessly in Siemens Xcelerator portfolio is unleashing new productivity gains to organizations. For instance, let’s imagine a global enterprise with operations in one part of the globe, where an operator has discovered a code issue in a piece of equipment. The operator can use a Teamcenter app on his phone to send a voice message describing the problem and upload a photo. Within seconds, AI extracts the most pertinent details and maps it to an EBOM/MBOM view, sent to an engineer working in Frankfurt, Germany. Working together, the engineer and his AI copilot simulate the problem in Simcenter and send a new code to the operator to get the equipment back online within minutes, not hours or days. The simulation and resolution are used to help predict and resolve the same issue if it crops up anywhere else. Silos have disappeared and engineering and operations are connected in a continuous improvement loop. Engineering informs operations, and operations improve engineering.
Expanding pilots to the agentic enterprise The next critical aspect of Siemens industrial AI is to unleash enterprise-wide agentic AI with Intelligence Center X, which was unveiled at Realize LIVE Americas in June 2026. Scaling industrial AI across the enterprise requires three things to work together. First, a way to connect and contextualize operational data so AI agents reason from facts rather than guesswork. Second, a way to build and govern the precision ML models that power real industrial decisions. And finally, a way to deploy those agents into
48 Future Digital Twin & AI | www.futureoilgas.com
FUTURE DIGITAL TWIN & AI
the hands of people, quickly, safely and at scale. Intelligence Center X brings all three together. An Enterprise Knowledge Graph (Graph Studio) connects data across every source and domain, grounding every agent response in traceable, structured enterprise knowledge. Purpose-built ML models (in AI Studio) are trained on that context. They deliver the predictive and prescriptive intelligence that language models alone cannot. And a low-code development application platform (Mendix) puts governed, agentic applications in the hands of the people who need them, without losing control or ownership of the underlying data. With this contextualized data, industryspecific intelligence extraction and governed actions, energy companies can more confidently turn isolated experiments into scalable, real-world business impact, where people and AI agents work together with shared context, workflows and lifecycle intelligence. Since AI models are empowered to understand complete operational data and relations; allow AI models to work with current data; allow validation of actions through the digital twin; and ensure that implemented actions through agents are compliant with company guardrails, companies can unleash their process-specific improved actions without the risk of hallucinations.
How Siemens AI solutions are unique Every company worries about how to protect IP, keep control over their data and ensure governance while deploying AI at enterprise level. Siemens’ unique value to customers pivots on the following: • Industry-leading low-code accelerators: Intelligence Center X combines the Mendix low-code platform with software from the Rapidminer portfolio, allowing anyone to build applications to collaborate more closely, outsource basic tasks, and focus on the tougher challenges to move faster and have a bigger impact. In this way, Intelligence Center X complements a
company’s in-house AI experts in scaling industrial AI deployment. • Keep control of data: Build rich, interactive, AI-driven enterprise applications that put AI in the hands of your people without losing control or ownership of your data. • Open and composable: Companies can future-proof industrial AI projects by building on a foundation that doesn’t lock them into one vendor, one architecture or one strategy. Siemens’ open technology, open architecture and ecosystem approach allows customers, EPCs, OEMs, SIs and partners to innovate together. Siemens is uniquely situated to deliver the power of industrial AI for companies to accelerate innovation, improve production quality, reduce OPEX and meet regulatory obligations. First, Siemens Xcelerator software portfolio is seamless on cloud, desktop and edge devices with no loss of data fidelity between internal teams or locations. Second, products are configurable and evolve as your business evolves. And finally, adaptive systems apply AI, agents and executable digital twins to sense change, respond instantly, and optimize decisions in real time. When you apply Siemens Industrial AI to this foundation, you get the following: • Faster engines – By embedding AI agents into some of the best physics, geometry and data engines, you get insights faster and more accurately. This also accelerates the engine for simulation, verification and optimization. Native AI reduces compute times, and allows more design exploration, converging on answers better and faster.
www.futureoilgas.com | Future Digital Twin & AI 49
FUTURE DIGITAL TWIN & AI
• Smarter execution – Intelligence embedded directly into workflows across simulation and operations helps define and deliver insights instantly while even automating some of the repeatable routines. • Trusted outcomes – Outcomes require context, not just data. Most data lakes give you only disconnected snapshots. You need an understanding of operations and how changes affect them. Turn lifecycle data into decisionready intelligence.
Concepts in action Leading companies are creating enterprise-wide decision intelligence by connecting trusted data, digital twins, Physics AI and operational context. The result is faster decisions, higher efficiency, improved reliability, and a stronger foundation for the industrial enterprise of the future. Use case #1: Siemens Energy manages a broad fleet of global turbines, which presents a major logistics challenge. Managing spare parts catalogs and specific customer configurations for each turbine requires manual processing and labor. Knowledge Graphs – built by subject matter experts and augmented by intelligent machine learning algorithms and application development – are the ideal solution to organize, manage and query its machine data structures across the fleet. Read more. Use case #2: Aker Solutions embraced engineering simulation and predictive engineering analytics to ensure their equipment is optimized through the complete lifecycle. The company delivered its digital twin “Thermal Insight” to manage thermal performance even in locations that are largely inaccessible. Read more. Use case #3: Assystem operates fusion power plants and wanted to combine as-designed plant models with real-world sensor data to create a Digital Twin that helped engineers understand the plant’s structural integrity and optimize inspection and maintenance schedules. Rapidminer AI Studio software formed the solution’s backbone and analytic front end, creating a seamless data flow between the plant and the physics-based models to create a true Digital Twin. Read more. The future of industrial AI is not better models. It’s better decisions. Learn more about how enterprise AI software improves oil and gas operational efficiency, reduces costs and minimizes environmental risk.
50 Future Digital Twin & AI | www.futureoilgas.com
MODELLING
Separation modelling rigour: the key to unlocking hidden production capacity Tom Ralston, Digital Process Engineering Business Development at MySep Pte Ltd, explains why modelling fidelity is becoming a critical differentiator in digital twin performance. He explores how rigorous separation modelling can expose hidden production constraints, strengthen AI-driven optimisation and unlock significant operational and commercial value.
D
igital Twin Adoption Is Widespread. Fidelity Is Not. Digital twins are now embedded across production assets. AI-driven optimisation, real-time dashboards, and predictive analytics dominate the conversation. Yet one critical question remains:
Does the model reflect physical reality with sufficient depth? In many facilities, production is not limited by compression or equipment nameplate capacity. It is limited by separation performance. performance And separator behaviour is frequently simplified inside digital twins. That modelling gap can quietly suppress significant revenue. The Hidden Constraint: Separation Physics Real separator performance depends on droplet size distributions, internal geometry, entrainment, coalescence, and separation efficiency. These directly influence hydrocarbon dew point compliance, gas export specifications, and allowable throughput. Yet many digital twin environments treat separators as idealised flash stages. The mass balance may converge; the physics may not. Whilst it is typical to model many process operations with rigour, including heat exchangers, pipelines, valves, pumps, compressors, controls etc with fully rigorous representations: simulator default separator models either, neglect carry-over, or require the user to specify arbitrary linear factors. The consequence is subtle but material: operators may accept conservative production ceilings without identifying the true constraint.
52 Future Digital Twin & AI | www.futureoilgas.com
Modern intelligence for asset integrity The system prepares. The engineer decides.
Deeplify connects inspection findings, asset history, maintenance and engineering data into one intelligence layer. Integrity teams work from quality-checked, decision-ready context instead of rebuilding it from scattered systems and reports.
One connected record per asset, from inspection through to decision Quality-checked findings, with the source always traceable Less time reconstructing information, more time on engineering judgment
Future Digital Twin & AI · Amsterdam · 18 September 2026
deeplify.de
MODELLING
uplift without major capital expansion or export uplift, specification breach. The value came from modelling fidelity.
Turning Modelling Rigour into Revenue MySep embeds rigorous separator physics into commercial simulation platforms used to construct HYSYS®, AVEVA™ digital twins, such as Aspen HYSYS® PRO/II™ and AVEVA™ DYNSIM® DYNSIM®, Honeywell UniSim® Design, Design KBC Petro-SIM® Petro-SIM®, Kongsberg K-Spice® and SLB Symmetry. Instead of approximating equilibrium separation, the model predicts real droplet behaviour and carry-over risk — exposing bottlenecks invisible to simplified representations. This enables operators to identify hidden separation limits, quantify throughput sensitivity, evaluate retrofit scenarios virtually, and optimise production within specification. Rigour becomes decision clarity. Case Study: US$325 Million Through Fidelity On a producing FPSO, oil production was capped at ~74,000 BOPD and gas at ~79 MMSCFD due to the reason that the export gas system was limited by hydrocarbon dewpoint specification. specification From an engineering perspective: the processing plant appeared mechanically sound, yet its capacity was capped well below its design potential. Conventional simulation models that represented separators as ideal devices showed no root cause. A process digital twin integrating MySep’s rigorous separator models revealed liquid carry-over from the first-stage separator as the true constraint. For the first time, the model could replicate how small variations in gas and liquid flow affected entrainment and downstream
54 Future Digital Twin & AI | www.futureoilgas.com
dewpoint. Utilising the MySep Studio software, retro-fit internals upgrades were explored for each separator vessel, and the overall process operational envelope could be explored, with these revamp configurations simulated in the digital twin. The upgraded process provided improved droplet capture efficiency throughout, significantly reducing the amount of mist carried into the export gas stream even during high throughput operation. In practice, this meant the facility could safely raise production rates without exceeding the dewpoint limit. The improvement unlocked an increase in production to ~82,000 BOPD and ~88 MMSCFD — delivering an estimated US$325 million annual
AI Needs Engineering Truth AI enhances optimisation. But AI operating on simplified physics cannot uncover physics-driven constraints. Digital twins are only as good as the models they contain. By ensuring that separation — often the hidden bottleneck — is represented with sufficient physical depth, operators can uncover significant efficiency and revenue potential while maintaining compliance and integrity. When digital twins are grounded in rigorous engineering models, AI becomes materially more powerful — because its optimisation operates on physical credibility. Modelling rigour is not refinement. It is competitive advantage. Digital twin adoption is no longer the differentiator. Model fidelity is. The future is not AI alone. It is AI built on engineering rigour.
From Data Chaos to Operational Intelligence Platform that turns years of industrial data into Operational Intelligence in weeks.
Unify
Decide
Act
Connect data, systems & context
Understand, reason & determine the next best action
Execute across systems, agents & 0eo0/e
Weeks, not months. Accelerate asset readiness. Drive measrale impact.
Book a personalized demo
rive.ai/demo rive.ai
contact@rive.ai
INDUSTRIAL AI
Trustworthy Industrial AI: Solving the black box problem in critical energy environments Dr Gerhard Koch, Chief Software and Product Officer at Siemens Energy Global, explains why explainability, stability and context are becoming essential foundations for scaling AI safely and effectively across industrial operations.
A
rtificial intelligence is no longer a futuristic promise for the energy sector. It is already reshaping how energy companies improve speed, efficiency, and decision quality across engineering and operational processes. But for energy and heavy industry, the defining question is no longer whether AI can create value. It is whether AI can be trusted in environments where safety, reliability, and asset protection are non-negotiable. That is where Industrial AI must move beyond generic automation and become explainable, stable, and grounded in physical reality. At Siemens Energy, we see this as one of the most important shifts in the next phase of digital transformation: AI will only scale in critical industries when trust is engineered into the system from the start. For energy companies, that means combining advanced analytics with domain expertise, physics-based understanding, and clear accountability for how recommendations are generated and used. This is the fundamental dilemma of Industrial AI: the same technology that can
56 Future Digital Twin & AI | www.futureoilgas.com
Mastering digitalization is like a rally. With a proactive partner, you take every turn at full throttle. Digital transformation in the process industry isn’t just a challenge, it’s a chance to grow. It calls on even the smallest teams to rethink what’s possible, to turn complexity into clarity, and to lead with purpose. With Endress+Hauser as your partner, digitalization becomes a shared journey. Together, we turn data into direction, infrastructure into insight, and daily work into real progress. You’re not alone on this path. With the right partner, transformation is achievable – and momentum becomes your advantage. #TeamUpToImprove!
Do you want to learn more? www.endress.com
INDUSTRIAL AI
dramatically improve performance can also introduce new risks if it behaves like a black box. In industrial environments, blackbox behavior, hallucinations, and weak accountability are not abstract technical concerns. They directly affect whether AI can be deployed in systems where safety, reliability, and asset protection are essential. If the logic behind an AI recommendation cannot be understood, verified, or challenged, it becomes difficult to use that recommendation with confidence in environments where mistakes carry operational, financial, or safety consequences. The stakes are especially high in energy and heavy industry. In critical environments, operational reliability is not optional. Unverified control signals can put valuable machinery such as turbines or compressors at risk. A wrong AI-driven decision can also lead to unplanned shutdowns and significant financial loss. These examples show why trust is not a soft factor in Industrial AI. It is a prerequisite for adoption, scale, and value creation. To move beyond the black box, energy companies need an “AI glass box” strategy. This means developing AI systems that are not only powerful, but also explainable, stable, and context aware. Explainable
AI helps identify the true “why” behind a decision by providing easily understandable logic. Stability tracking AI ensures smooth and stable decision paths, helping AI follow physical logic without erratic conclusions. Contextual AI adds the ability to understand industrial relationships — from individual components to machines and entire systems. Together, these approaches shift AI from a mysterious decision engine to a more transparent and reliable partner for industrial experts. Trustworthy Industrial AI also requires the right architecture. A robust trust architecture starts with data governance, including unified data lineage and data quality. It then needs a model trust layer, built around approaches such as physics-informed machine learning and uncertainty quantification. On top of that, an intelligence layer can provide explainable AI and root cause attribution, while a guardrail layer uses deterministic safety enforcers to make systems safe by design. In this layered view, trust is not added at the end; it is built into the system from data to decision. This matters because trust is what turns AI potential into industrial value. Trusted AI can support higher uptime through more reliable predictive maintenance, lower operating costs through better optimization, faster rootcause analysis, and fewer false alarms through contextual validation. AI creates the greatest value when people can understand its recommendations, challenge its assumptions, and apply it responsibly in critical workflows. For energy leaders, the message is clear. The next phase of AI adoption will not be won by the organizations that deploy the most algorithms, but by those that can prove that AI is understandable, reliable, and aligned with the physical reality of industrial operations. Trustworthy AI is therefore not just a technology topic. It is a business enabler, a safety requirement, and a condition for scaling AI where it matters most.
58 Future Digital Twin & AI | www.futureoilgas.com
2026 EVENTS PROGRAMME
FUTURE DIGITAL TWIN & AI
USA 2026
13th May 2026 | Houston
FUTURE DIGITAL TWIN & AI
AMSTERDAM 2026 September 2026 | Amsterdam
FUTURE DIGITAL TWIN & AI
MIDDLE EAST 2026
29th October 2026 | Middle East
FUTURE AI
USA 2026
17th November 2026 | Houston
futuredigitaltwin.com
www.futureoilgas.com | Future Digital Twin & AI 59
DATA
From Interoperability to Intelligence: Building Digital Twins and Industrial AI on Engineering Data That Can Be Trusted Dr. Michael Wiedau, Chairman of DEXPI e.V., explains why trusted, machine-readable engineering data is essential to unlocking the full potential of digital twins and industrial AI. He explores how open standards and semantic interoperability can turn fragmented information into a reliable foundation for smarter, scalable industrial decision-making.
I
ndustrial companies are being asked to extract more value from their information than ever before. Digital twins should support decisions across the asset lifecycle. Knowledge graphs should connect facts that were previously locked in separate systems. Artificial intelligence should help engineers find, compare and interpret information at speed. Yet these ambitions encounter the same practical obstacle: the underlying engineering data is often fragmented, inconsistent and difficult for machines to understand.
60 Future Digital Twin & AI | www.futureoilgas.com
A plant may be described by piping and instrumentation diagrams, equipment lists, simulation models, vendor documents, databases and maintenance systems. Each source is useful, but names, classifications and relationships can differ. A person can often reconstruct the intended meaning by reading a drawing, recognising a symbol and drawing on experience. Software cannot safely make those assumptions. If an AI system cannot distinguish a pump from its tag, or cannot identify which line is connected to which
nozzle, fluent output is not the same as a reliable answer. This is why interoperability is not a peripheral integration topic. It is the foundation for trustworthy digitalisation. DEXPI e.V. develops an open, vendor-neutral information model for the standardised exchange of engineering data in the process industries. The aim is simple but farreaching: information should retain its meaning when it moves between tools, organisations and lifecycle phases.
DATA
Figure 1 — Intelligence becomes dependable when fragmented documents and application data are transformed into a shared, machine-readable information layer.
The missing layer is meaning Many digital initiatives begin with access: collect files, connect databases and move content into a common platform. Access matters, but it does not automatically create understanding. Two systems may exchange a field called “type” while assigning different meanings to it. A tag may identify an asset in one application and a functional location in another. A line drawn between symbols may be visually obvious to an engineer but have no explicit, computable relationship behind it. Semantic interoperability addresses this problem. It describes not only values but what those values represent and how objects relate. For process engineering, that means identifying equipment, instruments, piping, properties and connections in a consistent structure. The information can then be validated, queried and reused instead of repeatedly interpreted or manually remodelled. DEXPI brings owner-operators, engineering companies, software vendors, equipment suppliers
and research organisations together to turn these requirements into implementable specifications. Openness is essential. A plant outlives individual software releases and often the systems in which it was designed. Its information must remain usable without binding the owner to one vendor, one exchange project or one moment in the lifecycle. The immediate benefit is less friction during engineering exchange and handover. The strategic benefit is much larger: a shared model provides stable ground for services that depend on context. A digital twin can link its representation to the engineering definition of the asset. A knowledge graph can traverse explicit relationships. Analytics can compare like with like. AI can retrieve a smaller, better-defined body of evidence and return answers with traceable sources.
Complementary building blocks for the industrial digital twin No single standard needs to describe every aspect
of an industrial asset. A more scalable approach is to combine standards with clear responsibilities. DEXPI focuses on rich engineering information for process plants: the objects, properties and connectivity needed to exchange and interpret plant design data. The Industrial Digital Twin Association advances the Asset Administration Shell (AAS), a standardised digital representation that organises information about an asset into interoperable submodels and exposes it across the lifecycle. These approaches are complementary. DEXPI can provide precise engineering semantics and relationships; the AAS can package and deliver relevant information in a consistent digital envelope. Together they support a twin that is more than a three-dimensional image or a collection of documents. It becomes an interoperable information resource connected to the physical asset and usable by engineering, operations, maintenance and supply-chain processes.
www.futureoilgas.com | Future Digital Twin & AI 61
DATA
Figure 2 — Engineering semantics and a modular digital asset representation serve different purposes, but together connect the physical asset with lifecycle applications
This combination also supports federation. Not
Machine-readable engineering information
were applied. Access controls must reflect
every fact must be copied into a monolithic
makes several valuable patterns possible.
commercial sensitivity and operational risk.
repository. Information can remain under the
Engineers can search by meaning instead
Most importantly, AI output must be treated
authority of the system best able to maintain
of guessing filenames. A knowledge graph
according to its purpose. A drafting aid and a
it, while common identifiers, semantics and
can reveal dependencies around an item
safety-critical recommendation cannot share the
interfaces make it discoverable and usable. That
of equipment. Validation rules can identify
same assurance threshold.
reduces duplication and clarifies responsibility.
incomplete or inconsistent handover data.
The twin becomes a coordinated view across
Retrieval-based AI assistants can ground
trusted sources rather than another isolated
answers in structured plant information and
Treat information as a lifecycle asset
system that gradually diverges from reality.
associated documents. Agents can prepare
The highest return comes when interoperability
comparisons or workflows while keeping human
is designed into the lifecycle rather than
From connected data to trustworthy AI
approval at consequential decision points.
added during a late data-cleaning exercise.
None of this removes the need for governance.
Requirements should define what information
Industrial AI is frequently discussed as a model-
Standards make quality visible; they do
must be delivered, in which semantic form and
selection problem. In practice, much of the
not manufacture quality automatically.
against which quality rules. Engineering tools
work lies upstream. The model needs relevant
Organisations still need ownership, validation
should create and exchange that structure.
context, controlled terminology and evidence
rules, version control and change processes.
Handover should validate both content and
it can cite. It needs to know whether a value is
Provenance must accompany data so users
relationships. Operations should preserve
current, which object it describes, where it came
can understand who supplied it, when it
context as the plant changes, and operational
from and whether it passed appropriate checks.
was updated and which transformations
learning should flow back into future engineering.
62 Future Digital Twin & AI | www.futureoilgas.com
DATA
Figure 3 — Shared semantics, validation, operational stewardship and feedback form a governed cycle around a trustworthy data core.
A practical starting point is a focused use case with measurable value: reducing manual checks at handover, finding the latest approved information for an asset, or accelerating impact analysis for a change. Map the information required for that outcome, identify authoritative sources and expose the semantic gaps. Use open standards to close those gaps, test exchange with real software and measure the reduction in rework, search time or uncertainty. Then extend the pattern. Progress also depends on treating conformance as a product capability, not a one-time project test. Implementations should be checked against examples, validation services and exchange scenarios. When an interpretation differs, the community needs a transparent route to resolve it and improve the specification. This feedback loop turns a standard into an interoperable ecosystem. It also gives asset owners confidence that information delivered today can be read by another compliant tool tomorrow, without sacrificing the applications that engineers rely on. Competition can then focus on workflows and
insight rather than proprietary control of meaning. Collaboration is crucial because interoperability cannot be achieved by one participant acting alone. Owner-operators must state durable information requirements. Engineering contractors need workable delivery processes. Software vendors must implement standards consistently. Equipment suppliers should provide structured product information. Associations such as DEXPI and IDTA create the neutral environment in which those perspectives can converge.
An open foundation for the next generation of industry Digital twins and industrial AI will not succeed through compelling demonstrations alone. They must survive changes of tool, project partner, organisational boundary and lifecycle phase. They must be able to explain what an answer is based on. They must operate on information whose meaning is explicit enough for machines and stable enough for industry. That foundation is achievable. Open, vendor-neutral
standards convert engineering knowledge from application-specific content into a shared industrial resource. DEXPI’s work on process-plant information, combined with complementary building blocks such as the Asset Administration Shell, offers a route from document exchange to connected lifecycle information—and from connected information to intelligence that engineers can use with confidence. The decisive question is therefore not whether a company has a digital twin or an AI pilot. It is whether the information beneath it can move, retain its meaning and remain trustworthy. Solve that, and innovation can scale beyond individual demonstrations. Interoperability is where industrial intelligence begins. About DEXPI e.V. DEXPI is an independent, memberdriven association advancing data interoperability and digitalisation across the process industries. Its international community develops and maintains an open, vendor-neutral information model for the standardised exchange of engineering data. Learn more at www.dexpi.org.
www.futureoilgas.com | Future Digital Twin & AI 63
FUTURE DIGITAL TWIN & AI
AI needs more than data: why standardized digital twins matter AI needs more than access to industrial data. Sandeep Rudra, Technical Project Manager at IDTA, explains how standardised Digital Twins and the Asset Administration Shell can provide the structure, context and interoperability needed to turn AI into reliable, scalable industrial applications.
A
I can only deliver reliable results when the data behind it is meaningful, structured and accessible. Across manufacturing, energy and oil and gas companies are dealing with vast amounts of data generated by machines, equipment, products and infrastructure. Yet this data is often fragmented across systems, vendors and stages of the asset lifecycle. The challenge is therefore making existing information understandable and usable across organisational and technological boundaries. The Asset Administration Shell (AAS) provides a standardised way of describing assets and their associated information. By giving industrial data a common structure and semantic context, it can support the integration of AI applications across different systems and environments. Digital Twins based on the AAS can therefore serve as digital representations of physical assets as well as a foundation for AI-driven applications and new digital services. Industrial Digital Twin Association (IDTA) is working with its community to make this approach practical. Alongside the AAS specifications and implementation guidance, the association provides standardised Submodel Templates and use cases that address concrete industrial requirements, including AI-related applications.
64 Future Digital Twin & AI | www.futureoilgas.com
FUTURE DIGITAL TWIN & AI
How AAS and AI Complement Each Other The relationship between AAS and AI is best understood as a functional complement. The AAS provides structure, semantics and context for industrial information, while AI analyses this information and derives predictions, classifications or recommendations. The two technologies therefore address different parts of the problem with the AAS not replacing analytics and AI not replacing semantic and interoperable information models. For manufacturing, energy and other assetintensive industries, this distinction is important. AI methods require technically meaningful inputs instead of only raw values. An AI system needs to know what a signal represents, to which asset it belongs, in what unit it is measured, and under which conditions it is valid. The AAS can provide this structured context and thereby create a common information basis for interoperable digital applications, including AI. Once industrial information is structured and available in a standard form, AI can use it for classification, pattern recognition, anomaly detection, prediction and decision support. This creates a layered architecture in which the AAS
provides the information foundation and AI acts as the analytical layer built on top of it.
What IDTA Already Provides IDTA has already established several elements that make this interaction more concrete. The AAS is supported by specifications and implementation guidance, while the IDTA Submodel repository provides standardised templates that can be used and reused across applications. Industrial use cases demonstrate how AAS-based information exchange can support concrete applications, including AI. Three Submodel Templates currently address AIrelated information: Artificial Intelligence Dataset, Artificial Intelligence Deployment and Artificial Intelligence Model Nameplate. Together, they cover different stages of the AI lifecycle and demonstrate how AI-related information can be represented in a structured and machine-readable form. The Artificial Intelligence Dataset Submodel addresses the data basis of AI models, including schema, labels, provenance, usage constraints and quality indicators. The Artificial Intelligence Model Nameplate is designed to document information related to AI models and their training process. The Artificial Intelligence Deployment Submodel supports the operational phase and includes
information such as storage location, model access, input and output, hardware requirements and runtime-related parameters. By standardising this information, the templates can help reduce the information loss that occurs when AI moves from data collection and model development into industrial operation. They also make AI-related information accessible not only to people but to machine-readable processes and applications.
From AI Applications to Digital Product Passports The relevance of standardised digital information extends beyond AI. Another important application area is the Digital Product Passport (DPP), which will introduce new requirements for accessing and sharing product information across value chains and throughout the product lifecycle. The Battery Passport provides a particularly tangible example. From February 2027, batteries covered by the relevant EU regulation will require a digital battery passport. The concept connects information from different actors across the battery value chain and requires data to remain accessible and usable over the product lifecycle. The AAS can provide a modular, standardised structure for representing many of the information elements required in such applications.
www.futureoilgas.com | Future Digital Twin & AI 65
FUTURE DIGITAL TWIN & AI
Standardised Submodels allow information to be modelled according to specific use cases while remaining part of a common digital representation. The IDTA has already published a set of Submodel Templates for the Battery Passport, demonstrating how standardised information models can support concrete regulatory and industrial requirements. AI can add another layer of value to these applications. For example, AI-based interfaces could allow users to interact with product information in natural language without having to understand the underlying data structures. AI can also support companies in creating and maintaining structured product information. The combination of standardised digital information and AI therefore has the potential to make complex product data more accessible and easier to use.
Digital Twins Across the Asset Lifecycle The same principle applies to complex industrial assets. In energy and oil and gas, pipelines, turbines, pumps, compressors and entire
66 Future Digital Twin & AI | www.futureoilgas.com
production facilities can remain in operation for decades. Throughout their lifecycle, they generate and require a wide range of information – from technical specifications and operating data to maintenance records, condition information, upgrades and decommissioning data. When this information is distributed across different systems, vendors and lifecycle phases, maintaining a consistent digital representation of the asset becomes difficult. Standardised digital information models can help connect these sources and make asset information usable across organisational and technological boundaries. For asset-intensive industries, this creates an opportunity to move from isolated Digital Twins to digital representations that can evolve with the asset throughout its lifecycle. The AAS provides a framework for structuring and exchanging this information, while AI can use it to support applications such as predictive maintenance, anomaly detection, optimisation and decision support.
New Business Models The combination of AAS and AI is also relevant from a business perspective. If datasets, models, deployment information and asset-related metadata can be represented in a standard form, AI-supported services become easier to scale across plants, customers and partners. This can support business models such as predictive maintenance services, AI-assisted engineering support, performance benchmarking, model lifecycle services and platform-based data offerings. The underlying principle is straightforward: standardisation can reduce the effort required to connect data, systems and AI applications. Instead of developing individual integrations for every asset, plant or customer, companies can build reusable services on a common information foundation. The AAS creates the basis for digital services that can operate across organisational boundaries and for AI applications that can make use of industrial information in a consistent and scalable way. As Digital Product Passports, Battery Passports and AI-driven applications become increasingly relevant, interoperable Digital Twins can provide an important foundation for the next generation of industrial digital ecosystems.
www.futureoilgas.com | Future Digital Twin & AI 67
INDUSTRIAL KNOWLEDGE GRAPH
From digital twin to industrial knowledge graph Arun Govind, Co-Founder & CPO of Rive Labs, Inc, explores how digital twins are evolving beyond representation into Industrial Knowledge Graphs that connect assets, systems, documents and operational data through meaningful relationships. He argues that this connected context will provide the foundation AI needs to reason effectively across the industrial enterprise. 68 Future Digital Twin & AI | www.futureoilgas.com
INDUSTRIAL KNOWLEDGE GRAPH
F
or more than a decade, the industrial world has been investing in digital twins. The vision has always been compelling: create a digital representation of a physical asset, system or facility, connect it with operational data, and use that representation to understand and improve the physical world. Digital twins have already changed how we design, engineer, commission and operate complex assets. But as industrial organisations enter the age of AI, an important question is emerging: Is representing the physical world enough? Increasingly, the answer is no. The next evolution of the digital twin is not simply a more detailed 3D model or a twin connected to more sensors. It is a connected knowledge layer that understands assets, components, documents, maintenance activities, spare parts, suppliers, operational events and, critically, the relationships between them. This is the Industrial Knowledge Graph. And it has the potential to become the nerve centre for the industrial organisation in the age of AI. The Industrial Data Problem Has Never Been About a Lack of Data Consider something as ordinary as a pump in a large industrial facility. Its engineering information may originate in design systems. Its master record may sit in SAP or another enterprise asset management system. Its maintenance history may be in a CMMS. Sensor readings may reside in a historian. Drawings and manuals may be stored in a document management system. Spare parts information may sit in material masters and bills of materials. Supplier information may be distributed across procurement systems, contracts, emails and technical documentation. There may even be an excellent 3D representation of the facility. Every system contains part of the truth. No system understands the whole. This fragmentation has existed for decades. Industrial organisations have responded by building data warehouses, data lakes, integration platforms and, more recently, cloud data platforms. These technologies are important. They make enormous volumes of information accessible. But making data accessible is not the same as making it understandable. An engineer investigating a recurring pump failure does not simply need more data. The engineer needs to understand that a particular vibration reading belongs to a specific bearing, inside a specific pump, serving a particular process, maintained under a certain work order, using a component supplied by a particular manufacturer. Those relationships are where industrial context lives. From Digital Representation to Connected Industrial Knowledge This is where the digital twin can evolve. Imagine that every important industrial entity becomes part of a connected information model:
an asset and its subassemblies; the sensors monitoring it; the documents describing it; the maintenance plans protecting it; the work orders previously performed on it; the spare parts it requires; the inventory holding those parts; the suppliers providing them; and the other assets that depend upon it. The result is not simply a database containing industrial information. It is a network of industrial knowledge. The knowledge graph understands that a compressor contains a dry gas seal, that the seal is described by a technical datasheet, is maintained through a particular procedure, requires specific spare parts, was affected by previous work orders and is supplied by a particular manufacturer. context. Those relationships create something extremely valuable: context And context is what transforms a digital twin from something we can view into something machines can reason over. Why This Matters Now: AI Needs Context Generative AI has created enormous excitement across industry. Organisations are building copilots, assistants and increasingly autonomous AI agents. But giving an AI agent access to enterprise data does not automatically give it an understanding of the enterprise. A language model can read a maintenance manual. A data platform can provide sensor history. An ERP system can expose material records. But consider asking: “What happens if this component becomes obsolete?” Answering that question properly requires much more than finding a document. The system must understand which equipment uses the component, where those assets are installed, whether alternatives exist, which alternatives are technically compatible, what inventory is available, which maintenance strategies depend on it, which suppliers can provide replacements and what operational risk is created if the component becomes unavailable. That requires reasoning across relationships. Or consider another question: “Why does this compressor keep failing?” The answer may require simultaneously examining failure history, work orders, maintenance procedures, operating conditions, sensor trends, replaced components, engineering specifications and similar equipment elsewhere in the fleet. This is where the Industrial Knowledge Graph becomes critical. It provides AI with a map of the industrial world it is being asked to reason about. The Knowledge Graph Becomes the Nerve Centre The human nervous system does more than collect signals. It connects signals with context and coordinates responses. An Industrial Knowledge Graph can play a similar role. Information continues to live in specialist systems - ERP, EAM, PLM, engineering tools, historians, document repositories and data platforms. Those systems do not need to disappear. Instead, the knowledge graph creates a connected semantic layer
www.futureoilgas.com | Future Digital Twin & AI 69
INDUSTRIAL KNOWLEDGE GRAPH
across them. At the centre is the physical and operational reality of the enterprise: assets, systems, locations, materials, people, documents, processes and events, connected through meaningful relationships. Once that foundation exists, intelligence can operate across traditional application boundaries. • A maintenance optimisation agent can understand equipment criticality, failure history and available resources. • An obsolescence agent can trace a discontinued component across the installed base and identify affected equipment. • A reliability agent can compare behaviour across similar assets. • A technician can move from an asset directly to the relevant drawing, procedure, historical work and spare part without searching across multiple systems. • An AI agent can reason across all of these relationships before recommending an action. The organisation moves from searching across systems to reasoning across knowledge.
70 Future Digital Twin & AI | www.futureoilgas.com
The Digital Twin Becomes a Living Operational Model This also changes what we should expect from a digital twin. A digital twin should not be a static deliverable created during engineering and slowly disconnected from operational reality. It should continuously evolve. A maintenance activity changes the history of an asset. A replacement changes its configuration. A new document changes what is known about it. A supplier notification changes the risk associated with a component. A sensor anomaly changes its operational context. The twin therefore becomes a living operational model of the enterprise. The Industrial Knowledge Graph provides the connective tissue that makes this possible. At Rive, we think of this as moving from fragmented industrial records toward Operational Intelligence: connecting engineering, operational and enterprise information into a living Industrial Knowledge Graph that gives both people and AI the context required to understand the physical
world. The goal is not to replace the systems industrial organisations have spent decades implementing. It is to make those investments work together. The Next Chapter of the Digital Twin The first chapter of the digital twin was about representation. The next is about understanding. As industrial AI moves from answering questions toward making recommendations and ultimately taking actions, the quality of the underlying context will determine the quality of those decisions. The organisations that succeed will not necessarily be those with the most data, the largest language models or the greatest number of AI agents. They will be the organisations that can connect their industrial knowledge in a way that both humans and machines can understand. That is why the Industrial Knowledge Graph is more than another data architecture. It is the intelligence foundation beneath the digital twin - and potentially the nerve centre of the AI-powered industrial enterprise.
TRANSFORMING AI INSIGHT INTO RELIABLE PERFORMANCE BUILDING TRUST TOGETHER IN AI
www.dnv.com
www.futureoilgas.com | Future Digital Twin & AI 71
FINAL WORD
AI will not fix a digital twin you cannot trust Mark Venables argues that oil and gas operators should resist the rush towards autonomous industrial AI until they can trust the engineering context beneath it. The competitive advantage will come from building digital twins that accurately reflect the assets AI is being asked to optimise.
O
il and gas has reached an uncomfortable point in its digital transformation. Artificial intelligence is advancing at remarkable speed, while many of the industrial information environments it is expected to operate on remain fragmented, inconsistent and sometimes out of date. That matters because the industry is no longer talking only about AI summarising documents or helping engineers find information. The ambition is shifting towards systems that diagnose equipment problems, recommend maintenance, optimise production and eventually coordinate operational actions. The more consequential the decision, the less tolerance there is for uncertain information beneath it. This is why I believe the next major challenge for digital twins is not AI integration. It is trust.
that sufficiently capable AI can replace engineering rigour. It cannot. Production facilities are governed by physics. Compressors have operating envelopes, separators have finite performance, pipelines have pressure constraints and equipment deteriorates in ways that cannot always be inferred from statistical correlations alone. AI can identify patterns, anomalies and relationships, but those capabilities become far more valuable when combined with first-principles models and established engineering knowledge. The future digital twin should therefore not become an AI model wrapped in a 3D interface. It should be an environment where physics-based models, operational data, engineering information and AI reinforce one another. That distinction becomes critical as systems move from observation to recommendation. There is a fundamental difference between AI telling an engineer that a compressor appears to be behaving abnormally and recommending that an operating condition is changed in response. The second requires evidence, constraints and explainability.
Bad context produces confident mistakes
Measure decisions, not twins
For years, operators have been encouraged to connect more data to the digital twin. Engineering information, historian data, maintenance records, inspection results, documents and 3D models have progressively been brought together. That has improved accessibility, but connection is not understanding. A compressor can have one identity in an engineering system, another in SAP, hundreds of historian tags and a collection of drawings and maintenance records stored elsewhere. Engineers navigate those relationships because they understand the plant. AI does not possess that institutional knowledge unless the context is made explicit. An AI system recommending action on a pump needs to know which pump it is looking at, its current configuration, what has changed since commissioning, which data belongs to it and which operating constraints apply. It must distinguish between a current drawing and one that has been superseded, and between a design limit and an operating target. Without that context, adding a more powerful model simply creates a more convincing way of being wrong.
Physics cannot become optional There is another temptation the industry should resist: assuming
72 Future Digital Twin & AI | www.futureoilgas.com
Operators also need to rethink how success is measured. Number of connected data sources, percentage of assets modelled and numbers of users are implementation metrics, not business outcomes. The questions that matter are harder. Did an engineer diagnose the problem faster? Was unnecessary maintenance avoided? Was production increased without compromising integrity? Did the organisation identify a developing failure earlier? Did an engineering change take days rather than weeks because everyone could trust the underlying information? The same discipline should apply to AI. Oil and gas does not need the largest possible collection of copilots, agents and pilots. It needs a smaller number of systems that can demonstrably improve decisions within clearly defined operational boundaries. Human oversight will remain essential where safety, integrity and environmental consequences are involved. But keeping a human in the loop is not an excuse for giving that person unreliable machine-generated recommendations. The objective must be better evidence and greater confidence, not simply more information to review. The organisations that lead the next stage of digital transformation will not necessarily be those that deploy AI fastest. They will be those that do the less glamorous work of creating reliable engineering context, maintaining the twin as the physical asset changes and defining exactly where machine intelligence should influence operational decisions. AI is moving quickly. Industrial reality is less forgiving. Before we ask machines to make better decisions about our assets, we need to make certain they understand the assets we have.
Scaling intelligence together Human expertise. Trusted AI. Proven impact. Transform operational knowledge and asset data into measurable results. Discover how trusted, explainable AI can help your teams improve reliability, reduce unplanned downtime, and scale operational excellence across your facilities. Visit us and see trusted AI in action.
Siemens Energy is a trademark licensed by Siemens AG.
You can't visit Ormen Lange, but you can scan this.