Handbook of Dynamic Data Driven Applications Systems
Volume 1
2nd ed. 2022
Editors
Erik P. Blasch
Air Force Of ice of Scienti ic Research, Arlington, VA, USA
Frederica Darema
InfoSymbiotics Systems Society, Boston, MA, USA
Sai Ravela
Earth, Atmospheric and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, MA, USA
Alex J. Aved
Air Force Research Lab, Rome, NY, USA
ISBN 978-3-030-74567-7 e-ISBN 978-3-030-74568-4
https://doi.org/10.1007/978-3-030-74568-4
© This is a U.S. government work and not under copyright protection in the U.S.; foreign copyright protection may apply 2022
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, speci ically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on micro ilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a speci ic statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional af iliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Contents
1 Introduction to the Dynamic Data Driven Applications Systems (DDDAS) Paradigm
Erik P. Blasch, Frederica Darema and Dennis Bernstein
Part I Measurement-Aware: Data Assimilation, Uncertainty Quanti ication
2 Tractable Non-Gaussian Representations in Dynamic Data Driven Coherent Fluid Mapping
Sai Ravela
3 Dynamic Data-Driven Adaptive Observations in Data Assimilation for Multi-scale Systems
Hoong C. Yeong, Ryne Beeson, N. Sri Namachchivaya, Nicolas Perkowski and Peter W. Sauer
4 Dynamic Data-Driven Uncertainty Quanti ication via Polynomial Chaos for Space Situational Awareness
Richard Linares, Vivek Vittaldev and Humberto C. Godinez
Part II Signals-Aware: Process Monitoring
5 Towards Learning Spatio-Temporal Data Stream Relationships for Failure Detection in Avionics
Sida Chen, Shigeru Imai, Wennan Zhu and Carlos A. Varela
6 Markov Modeling via Spectral Analysis: Application to Detecting Combustion Instabilities
Devesh K. Jha, Nurali Virani and Asok Ray
7 Dynamic Space-Time Model for Syndromic Surveillance with Particle Filters and Dirichlet Process
Hong Yan, Zhongqiang Zhang and Jian Zou
Part III Structures-Aware: Health Modeling
8 A Computational Steering Framework for Large-Scale Composite Structures: Part I—Parametric-Based Design and Analysis
A. Korobenko, M. -C. Hsu and Y. Bazilevs
9 Development of Intelligent and Predictive Self-Healing
Composite Structures Using Dynamic Data-Driven Applications Systems
Mishal Thapa, Bodiuzzaman Jony, Sameer B. Mulani and Samit Roy
10 Dynamic Data-Driven Approach for Unmanned Aircraft Systems
Aero-elastic Response Analysis
R. Kania, A. Kebbie-Anthony, X. Zhao, S. Azarm and B. Balachandran
Part IV Environment-Aware: Earth, Biological, and Space Systems
11 Transforming Wild ire Detection and Prediction Using New and Underused Sensor and Data Sources Integrated with Modeling
Janice L. Coen, Wilfrid Schroeder and Scott D. Rudlosky
12 Dynamic Data Driven Application Systems for Identi ication of Biomarkers in DNA Methylation
Haluk Damgacioglu, Emrah Celik, Chongli Yuan and Nurcin Celik
13 Photometric Stereopsis for 3D Reconstruction of Space Objects
Xue Iuan Wong, Manoranjan Majji and Puneet Singla
Part V Situation Aware: Tracking Methods
14 Aided Optimal Search: Data-Driven Target Pursuit from OnDemand Delayed Binary Observations
Luca Carlone, Allan Axelrod, Sertac Karaman and Girish Chowdhary
15 Optimization of Multi-target Tracking Within a Sensor Network Via Information Guided Clustering
Alexander A. Soderlund and Mrinal Kumar
16 Data-Driven Prediction of Con idence for EVAR in Time-Varying Datasets
Allan Axelrod, Luca Carlone, Girish Chowdhary and Sertac Karaman
Part VI Context-Aware: Coordinated Control
17 DDDAS for Attack Detection and Isolation of Control Systems
Luis Francisco Combita, Jairo Alonso Giraldo, Alvaro A. Cardenas and Nicanor Quijano
18 Approximate Local Utility Design for Potential Game Approach to Cooperative Sensor Network Planning
Su-Jin Lee and Han-Lim Choi
19 Dynamic Sensor-Actor Interactions for Path-Planning in a Threat Field
Benjamin S. Cooper and Raghvendra V. Cowlagi
Part VII Energy-Aware: Power Systems
20 Energy-Aware Dynamic Data-Driven Distributed Traf ic Simulation for Energy and Emissions Reduction
Michael Hunter, Aradhya Biswas, Bhargava Chilukuri, Angshuman Guin, Richard Fujimoto, Randall Guensler, Jorge Laval, Haobing Liu, SaBra Neal, Philip Pecher and Michael Rodgers
21 A Dynamic Data-Driven Optimization Framework for Demand Side Management in Microgrids
Haluk Damgacioglu, Mehrad Bastani and Nurcin Celik
22 Dynamic Data Driven Partitioning of Smart Grid for Improving Power Ef iciency by Combinining K-Means and Fuzzy Methods
Antonia Nasiakou, Miltiadis Alamaniotis, Lefteri H. Tsoukalas and Manolis Vavalis
Part VIII Process-Aware: Image and Video Coding
23 Design of a Dynamic Data-Driven System for Multispectral Video Processing
Honglei Li, Yanzhou Liu, Kishan Sudusinghe, Jinsung Yoon, Erik P. Blasch, Mihaela van der Schaar and Shuvra S. Bhattacharyya
24 Light Field and Plenoptic Point Cloud Compression
Li Li and Zhu Li
25 On Compression of Machine-Derived Context Sets for Fusion of Multi-modal Sensor Data
Nurali Virani, Shashi Phoha and Asok Ray
Part IX Cyber-Aware: Security and Computing
26 Simulation-Based Optimization as a Service for Dynamic DataDriven Applications Systems
Yi Li, Shashank Shekhar, Yevgeniy Vorobeychik, Xenofon Koutsoukos and Aniruddha Gokhale
27 Privacy and Security Issues in DDDAS Systems
Li Xiong, Vaidy Sunderam, Liyue Fan, Slawomir Goryczka and and Layla Pournajaf
28 Multimedia Content Analysis with Dynamic Data Driven Applications Systems (DDDAS)
Erik P. Blasch, Alex J. Aved and Shuvra S. Bhattacharyya
Part X Systems-Aware: Design Methods
29 Parzen Windows: Simplest Regularization Algorithm
Jing Peng and Peng Zhang
30 Multiscale DDDAS Framework for Damage Prediction in Aerospace Composite Structures
A. Korobenko, M. Pigazzini, X. Deng and Y. Bazilevs
31 A Dynamic Data-driven Stochastic State-Awareness Framework for the Next Generation of Bio-inspired Fly-by-feel Aerospace Vehicles
Fotis Kopsaftopoulos and Fu-Kuo Chang
32 The Future of DDDAS
Erik P. Blasch, Frederica Darema, Sai Ravela and Alex J. Aved
Index
About the Editors
Erik P. Blasch is a program of icer with the Air Force Of ice of Scienti ic Research. His focus areas are in multi-domain (space, air, ground) data fusion, target tracking, pattern recognition, and robotics. He has authored 750+ scienti ic papers, 22 patents, 30 tutorials, and 5 books. Recognitions include the Military Sensing Society Mignogna leadership in data fusion award, IEEE Aerospace and Electronics Systems Society Mimno best magazine paper award, IEEE Russ bioengineering award, and founding member of the International Society of Information Fusion (ISIF). Previous appointments include adjunct associate professor at Wright State University, exchange scientist at Defense Research and Development Canada, and of icer in the Air Force Research Laboratory. Dr. Blasch is an associate fellow of AIAA, fellow of SPIE, and fellow of IEEE.
Frederica Darema
retired as Senior Executive Service (SES) member and director of the Air Force Of ice of Scienti ic Research, Arlington, Virginia, where she led the entire basic research investment for the AF and served as research director in the Air Force’s Chief Data Of ice, and as associate deputy assistant secretary at the Air Force Of ice for Science, Technology and Engineering. Prior career history includes: research staff positions at the University of Pittsburgh, Brookhaven National Laboratory, and Schlumberger-Doll; management and executive-level positions at the T. J. Watson IBM Research Center and the IBM Corporate Strategy Group, the National Science Foundation, and the Defense Advanced Research Projects Agency; and director of the AFOSR Directorate for Information, Math, and Life Sciences. Dr. Darema, PhD in nuclear physics, is a fellow of the Institute of Electrical and Electronics Engineers (IEEE), among other professional recognitions. She pioneered the DDDAS paradigm, and since 2000, she has organized and led research initiatives, programs, workshops, conferences, and other forums to foster and promote DDDAS-based science and technology advances.
Sai Ravela, PhD
, directs the Earth Signals and Systems Group (ESSG) in the Earth Atmospheric and Planetary Sciences (EAPS) Department at the Massachusetts Institute of Technology. His primary interests are in statistical pattern recognition, stochastic nonlinear systems, and computational intelligence with application to earth, planets, climate, and life. Dr. Ravela has pioneered dynamic data driven observing systems for wildlife and luids, the latter with application from the laboratory to localized atmospheric phenomena. He has advanced several DDDAS topics with new methods for application to coherent luid dynamical regimes. Dr. Ravela proposed and co-organized the Dynamic Data Driven Environmental Systems Science Conference (DyDESS 2014, Cambridge), and then co-organized the irst, second, and third general DDDAS conferences (2016 Hartford, 2017 Cambridge, 2020 MIT/Online). Dr. Ravela also teaches machine learning with system dynamics and optimization, which introduces the informative approach, a key DDDAS concept, to design learning and hybrid stochastic systems and solve inverse problems and inference.
Alex J. Aved
is a senior researcher with the Air Force Research Laboratory, Information Directorate, Rome, NY, USA. His research interests include multimedia databases, stream processing (via CPU, GPU, or coprocessor), and dynamically executing models with feedback loops incorporating measurement and error data to improve the accuracy of the model. He has published over 50 papers and given numerous invited lectures. Previously, he was a programmer at the University of Central Florida and database administrator and programmer at Anderson University.
© This is a U.S. government work and not under copyright protection in the U.S.; foreign copyright protection may apply 2022
E. P. Blasch et al. (eds.), Handbook of Dynamic Data Driven Applications Systems https://doi.org/10.1007/978-3-030-74568-4 1
1. Introduction to the Dynamic Data Driven Applications Systems (DDDAS) Paradigm
Erik P. Blasch1 , Frederica Darema2 and Dennis Bernstein3
(1) (2) (3)
Air Force Of ice of Scienti ic Research, Arlington, VA, USA
InfoSymbiotics Systems Society, Boston, MA, USA
Department of Aerospace Engineering, University of Michigan, Ann Arbor, MI, USA
Erik P. Blasch (Corresponding author)
Email: erik.blasch.1@us.af.mil
Dennis Bernstein
Email: dsbaero@umich.edu
Abstract
Dynamic Data Driven Applications Systems (DDDAS) is a paradigm for systems analysis and design, and a framework that dynamically couples high-dimensional physical and other analysis models and methods, runtime measurements, and computational architectures. Some of the foremost early applications of DDDAS successes range from environmental assessment of adverse weather and natural disasters such as tornadic activity, hurricane formation and trajectory, wild ire monitoring and volcanic plume detection and tracking, to real-time structural health monitoring in aerospace systems and electrical power grids operation, and to medical and societal applications. Monitoring, understanding and predicting behaviors of complex and dynamic systems with DDDAS principles has expanded over the years to demonstrate new
and advanced capabilities in other applications that span space situational awareness, unmanned aerial vehicle (UAV) design and operation, and complex systems adaptive management and security applications. Recent efforts re lect the digital age of information management such as multimedia analysis, electrical power grid control, other civilian infrastructures, and biohealth concerns. Underlying DDDAS developments are advances in sensor design, signal processing and iltering, as well as computational architectures and communications. The book highlights for the reader DDDAS-based advances, with more information available in the DDDAS society’s website: www.1dddas.org.
Keywords Dynamic Data Driven Application Systems – Hurricane – Data Assimilation loop – Sensor recon iguration loop – Feedback control –
High-dimensional modeling – Weather forecasting – Volcanic ash detection – Wild ire monitoring – Orbital awareness – Structural health monitoring – Self-aware – Estimation – Context – Cyber networks –Sensing-learning-adaptation – Autonomy – Smart sensing – Autonomy in use (AIU) – Machine learning
1.1 Introduction
The methods presented in the book capture the DDDAS paradigm and the essence of DDDAS-based systems’ analysis and design. Invariably, the DDDAS paradigm and ensuing frameworks proposed by one of the authors (Dr. Frederica Darema) inspires many researchers for engineering and science advances.
As articulated by Dr. Darema who pioneered the DDDAS paradigm1 [1–5]:
“in DDDAS, instrumentation data and executing application models of these systems become a dynamic feedback control loop, whereby measurement data are dynamically incorporated into an executing model of the system in order to improve the accuracy of the model (or simulation), or to speed-up the simulation, and in reverse the executing application model controls the instrumentation process to guide the measurement process. DDDAS presents opportunities to create new capabilities through more accurate understanding, analysis, and prediction of the behavior of complex systems, be they
natural, engineered, or societal, and to create decision support methods which can have the accuracy of full-scale simulations, as well as to create more ef icient and effective instrumentation methods, such as intelligent management of Big Data, and dynamic and adaptive management of networked collections of heterogeneous sensors and controllers. DDDAS is a unifying paradigm, bringing together computational and instrumentation aspects of an application system, which extends the notion of Big Computing to span from the high-end to the real-time data acquisition and control, and it’s a key methodology in managing and intelligently exploiting Big Data. ”
DDDAS (Dynamic Data Driven Applications Systems), beginning in 1998–1999, is a paradigm in which computation and instrumentation aspects of an application system are dynamically integrated in a feedback control loop, in the sense that instrumentation data can be dynamically incorporated into the executing model of the application, and in reverse the executing model can control the instrumentation [6]. Such approaches have shown to enable more accurate and faster modeling and analysis of the characteristics and behaviors of a system. Methods based on the DDDAS paradigm can exploit data in intelligent ways to provide new capabilities, including: (1) analysis and understanding of the behaviors of complex systems and dynamic conditions; (2) decision support systems with the accuracy of full-scale modeling; (3) ef icient data collection and data mining (including heterogeneous and distributed data); and (4) complex systems resource management, such as heterogeneous collections of sensors and controllers, and optimized operations of systems-of-systems.
The DDDAS paradigm, and opportunities and challenges in exploiting the DDDAS paradigm have been discussed in a series of workshops, starting with the National Science Foundation (NSF) workshop in March of 2000 [7], and subsequent ones. Notably, shortly after the March 2000 workshop, Kelvin Droegemeier and his team applied the DDDAS paradigm for analysis of a March 2000 tornadic event in Dallas, Texas; the event had been missed by the then forecasting models, and Droegemeier showed that using DDDAS they would have correctly predicted the onset and path of the tornado [8]. The reports from this and subsequent forums, identi ied new science and technology capabilities, inspired by and enabled through the DDDAS paradigm. New capabilities include
modeling approaches, algorithm developments, systems software, and instrumentation methods, as well as the need for synergistic multidisciplinary research among these areas [9]. DDDAS brings together practitioners of application domains, researchers in mathematics, statistics, electrical engineering, and computer sciences, and designers involved in the development of instrumentation systems and methods. Through a series of forums and sponsored funding programs, research efforts commenced to address the challenges and create new frontiers in complex and dynamic systems’ capabilities. As shown through the increasing body of work, DDDAS is applicable to many areas, such as: (1) engineering: aerospace-, civil-, electrical- and mechanical-engineering, and nanotechnologies; (2) systems: manufacturing, transportation, and energy systems; (3) science: materials, environmental-, weather-, climate-sciences, and biomedical sciences; as well as (4) decision support for: operations in civilian and industrial infrastructures, medical diagnosis and treatment, multimedia analysis, and cyber security evaluation. This book presents examples of advances through DDDAS to inform and motivate scientists, engineers and developers who may use the DDDAS paradigm and exploit it in these in such and many other science and technology areas.
The remainder of this chapter seeks to provide the reader a better understanding of the DDDAS paradigm and its applicability. Section 1.2 discusses the features of the DDDAS paradigm. Section 1.3 highlights, as examples, the methods of estimation and assimilation for processing data. Section 1.4 presents key components in enabling DDDAS methods. Section 1.5 provides a review of major areas where DDDAS has been applied in the last 20 years. Section 1.6 provides an overview of the book contents. Section 1.7 discusses DDDAS future directions, and Sect. 1.8 presents a summary on the book scope.
1.2 What Is DDDAS?
Consider an approaching hurricane (e.g., the 2005 “Katrina” event [10, 11]). A meteorological model of the storm can be constructed, constituted from the atmospheric aspects of the storm system, and its interface with ground, litoral, or aqueous (especially oceanic) aspects. However, the model has limited predictive value without knowledge of initial and varying conditions, boundary conditions, inputs, parameters and states,
especially dynamic ones, such as velocities and accelerations - all these data collected with various sensor measurements, spanning many spatial positions, time-scales, and modalities. In order to make predictions, data are needed to estimate unknown quantities. Although the storm can be imaged at low resolution by satellite, measurements by aircraft with high resolution are expensive and limited in range, and therefore the size of the storm makes it impossible to obtain detailed measurements over a large area.
In a scenario such as the dynamic environmental example above, in the DDDAS-based approach, the (physical) model of the system can be used to select adaptively (by the executing model) measurements (needed for example to improve the predictive accuracy of the model), and in return the executing model guiding and recon iguring the sensors and their modalities, so that data useful in improving the model accuracy are collected. Thus, through the selected measurements, the information content of the data is enhanced, for the ultimate objective of accurately predicting the path and intensity of the storm. At the same time, the data collected by the sensors enhances the accuracy of the model by providing estimates of initial and varying conditions, boundary conditions, and dynamic inputs, parameters, and states. The integration of on-line and off-line (archival) data with the executing model creates a positive feedback loop, where the model judiciously guides the sensor selection and data collection, from which (“in return”) the sensor data improves the accuracy of the model; the other aspect of the DDDAS paradigm is to use the on-line (or archival) data to replace parts of the computation to speedup the modeling, and in return the executing model can request additional data to be collected or from archival storage, and use these data to further speedup the modeling; these approaches can be combined to speed-up and make the modeling more accurate.
The hurricane example illustrates the essence of Dynamic Data Driven Applications Systems (DDDAS). DDDAS is a paradigm and conceptual framework that synergistically and dynamically integrates models and data in order to facilitate and improve the analysis and prediction of the characteristics and behaviors of physical phenomena, engineered systems, societal systems. The term “physical” is used here in the broader sense of any model that represents the reality of a system (natural, engineered, or other system). In a broader context, DDDAS encompasses adaptive state estimation that uses an instrumentation recon iguration
loop (IRL) as shown in Fig. 1.1, in [12]. The IRL loop seeks to recon igure the sensors in order to enhance the information content of the measurements. The sensor recon iguration is guided adaptively by the simulation of the physical process. Consequently, the sensor recon iguration is dynamic, and the overall process is (dynamic) data driven.
Fig. 1.1 A Depiction of Dynamic Data Driven Applications Systems (DDDAS) feedback loop
The core of DDDAS is the data augmentation loop (DAL), which integrates instrumentation data to drive the physical system simulation modeling (of an actual system) so that the trajectory of the simulation more closely follows the trajectory of the physical system. The dynamic data assimilation (which combines theory with observations) in DAL, uses input data if input sensors are available. An innovative feature of DDDAS, which goes beyond the traditional “data assimilation” methods, is the additional instrumentation recon iguration loop (IRL), shown in Fig. 1.1, which controls the instrumentation system (such as the set of sensors, actuators, sensor executing models and sensors signal analysis), to collect additional data in targeted ways, where data are needed to improve or speedup the model, or apply coordinated control of sensors or actuators. For example if the instrumentation consists of a set of physical sensors, in DDDAS the executing model (or simulation)
adaptively guides the physical sensors, in a coordinated way, in order to enhance the information content of the collected data. The dynamic data augmentation and the instrumentation recon iguration feedback loops are computational and physical feedback loops. The simulation guides the instrumentation recon iguration and what data to be collected, and in turn, uses this additional, selected data, to improve the accuracy of the physical system simulation. This “meta” (positive) feedback loop is the essence of DDDAS.
Key aspects of DDDAS include the algorithmic and statistical methods that can incorporate dynamically the measurement data with that of the high- idelity modeling and simulation, and as needed invoke models of higher or lower levels of idelity (or resolution) depending on the dynamic data inputs.
1.3 State Estimation and Data Assimilation
The goal of state estimation is to combine models with data in order to estimate model states that are not directly measured. State estimation is a foundational area of research in systems and control. Relevant techniques date from the 1960’s in the form of the Kalman ilter and the Luenberger observer. An observer is a model that emulates the dynamics of a physical system and is driven by instrumentation data in order to approximate unmeasured states. The Kalman ilter is a stochastically optimal observer that estimates unmeasured states. In large-scale physics applications, such as applications involving structures or luids, state estimation is called data assimilation.
The Kalman ilter was developed for linear systems. However, most real applications involve nonlinear dynamics, and the development of observers and ilters for nonlinear systems is a challenging problem that remains largely unsolved. Numerous techniques, which can be described as suboptimal, ad hoc, application-based, or approximate, have been developed, and many of these methods are widely used. These techniques include the extended Kalman ilter (KF), ensemble Kalman ilter (EKF), ensemble adjustment Kalman ilter (EnAKF), unscented Kalman ilter (UKF), stochastic integration ilter (SIF), and particle ilters (PF) [13, 14]. As discussed in the next Section, DDDAS is an advanced adaptive state estimation methodology.
1.3.1 DDDAS and Adaptive State Estimation
State estimation algorithms are based on prior information about the (physical) system [15]. The information typically includes a model of the (physical) system as well as knowledge of the initial state, inputs (e.g., disturbances), and instrumentation (e.g., sensor data) noise. Likewise, stochastic representation, for example, as a statistical description of the disturbances and instrumentation-data noise, is one method to process the information. An adaptive state estimation algorithm may attempt to learn and update online the information, states, and parameters.
DDDAS uses adaptation in a different sense. In particular, DDDAS seeks to recon igure the instrumentation during operation, in real-time. Instrumentation recon iguration, driven by the model of the system, enhances the information content of the measurements. The information recon iguration loop is shown in Fig. 1.1. Together, the integration of the data into the executing model (dynamic data augmentation) loop and the sensor recon iguration loop are central to methods using the DDDAS paradigm.
1.3.2 Does DDDAS Use Feedback Control?
DDDAS uses computational feedback, in addition to physical feedback. As Fig. 1.1 shows, state estimation is a feedback process, where the instrumentation data corrects the simulation of the physical system. The data augmentation feedback loop is implemented in computation, and thus has no effect on the physical system.
DDDAS employs an additional feedback loop by recon iguring the instrumentation based on the sensor data, much like the KF method. The instrumentation recon iguration feedback loop is also computational, and thus does not affect the response of the physical system. In contrast, feedback control uses physical inputs (such as forces and moments) in order to affect the behavior of a physical system, such as an aircraft autopilot that drives the control surfaces and modi ies the aircraft trajectory. Consequently, DDDAS employs two computational feedback loops, in addition to the physical feedback control. The power of DDDAS is to also use simulated data from a high-dimensional model to augment measurement systems, for systems analysis and design, and to leverage statistical methods, simulation, and computation architectures.
1.4 DDDAS Methods
The DDDAS paradigm and ensuing framework, as the term “Applications Systems” in the name implies, has been applied to many applications where modeling and data collection are utilized in engineering and scienti ic analysis, but also for other application areas such as inancial and societal systems. To enable such capabilities through DDDAS, synergistic and multidisciplinary advances are needed along four science and technology axes: (1) modeling and simulation methods of real-world applications, (2) their associated mathematical (numeric and nonnumeric) and statistical algorithms, (3) instrumentation and measurement methods, and (4) computer systems software, as shown in Fig. 1.2.
Fig. 1.2 Characteristic attributes, or characteristic or inherent elements of DDDAS
Instrumentation methods include multidomain components in realworld situations such as for example, terrestrial and space sensors monitoring the atmosphere; avionics sensors detecting air movements, computer vision detecting vehicles on a terrain road network, or sonar sensing of water currents and turbulence in the ocean. Representing the application are comprehensive, high- idelity simulation models as well as model abstractions and models of lower idelities, such as for example those in the case-study areas referenced above, the space Global ionosphere–thermosphere model (GITM) model, the National Climate Atmospheric Reference (NCAR) model, ground-based vehicle traf ic models, and oceanic radar scatter models. The dynamic integration of modeling and the instrumentation data requires software systems to process the dynamically changing computational and communication requirements of the application models and algorithms, which change depending on the dynamic data inputs and model parameters; such include: models with good convergence properties under perturbation from dynamic data inputs, and invocation of other models (representing other modalities of the system and/or different levels of abstraction) depending on the dynamic data inputs. The seamlessly integrated coordination of high-end with real-time computing requires new hardware and software approaches in the ields of optimization, data low, and architectures, together with modeling and instrumentation methods for real world applications.
The key developments of the DDDAS-based dynamic integration of executing application models with the application instrumentation, and the support runtime software (computer “systems-software”) include: theory, algorithms, and computation, which the book seeks to highlight. The theory includes mathematical advances including numeric and non-numeric (e.g., retrospective cost modeling, graph-, agent-based, etc); while the algorithms support new methods (e.g., ensemble Kalman ilter, Particle ilter, optimization techniques). The computational considerations align with developments in the computer networking and communications areas for cases such as non-convex optimization, data low architectures, and systems design.
The challenges to enable DDDAS include data modeling, context processing, and content application. To bring together data, context and content requires addressing issues in model idelity, dimensions, and
usability such as how many parameters are needed for system control. When data are collected, they need to be preprocessed to determine whether the inherent information matches the context. An example is: clutter reduction, sensor registration, and confuser analysis in vehicle tracking. Finally, another key challenge is that of sampling, as shown in Fig. 1.3. Sampling is the multi-resolution needed to monitor the situation, environment and network context to explain the content desired.
Fig. 1.3 DDDAS challenges and processes
Three examples are presented in Fig. 1.4 which demonstrate DDDAS methods applied to enhance awareness. The examples are from the areas of air, space, and cyberspace, where modeling, instrumentation, and systems software have been designed for realistic platforms and actual environments. On the left is weather modeling with nonlinear tracking methods for unmanned aerial vehicle (UAV) light routing. The middle includes multi-domain robotics of space and ground vehicles with iltering methods for distributed autonomous coordinated control. Finally, the cyber example comes from power grids performance that integrates cyber physical systems (CPS) [16] with the internet of things (IoT). It should be noted that DDDAS is a more powerful paradigm than CPS (which was coined by the embedded systems community), as DDDAS involves comprehensive models of the system under study, and where
said models interact with the instrumentation aspects of the system and cognizantly control the instrumentation, as for example demonstrated for power grids in [17] for optimized power grids operation.
1.5 DDDAS Research Areas of Historical Development
The impact of the DDDAS concept has been manifested for almost two decades, starting with the initial NSF workshop in March 2000 that brought together researchers, engineers, scientists, and developers [7]. The March 2000 workshop introduced the DDDAS paradigm to the research community, and articulated the need for multidisciplinary research harnessing the power of theory, modeling, instrumentation, and (computer) software and hardware advances to instantiate applicationssystems-level opportunities. As articulated in that workshop and in the subsequent ones in 2006 and 2010, and in the 2005 NSF Program Solicitation, the DDDAS environments spanned and integrated the highend and the real-time computing, and computing at the sensors and controllers side – what was called later the IoT and Edge Computing. The proliferation of DDDAS impact is demonstrated in the literature, as shown in Fig. 1.5. The statistics from Fig. 1.5 only capture those papers that explicitly call-out DDDAS as the underlying paradigm; while many other papers that have brie ly acknowledged DDDAS are not included in Fig. 1.5. There is a growing trend in approaches using DDDAS, which is established through the website (www.1dddas.org).
Fig. 1.4 DDDAS Awareness Examples
Fig. 1.5 DDDAS Trends of papers per year (dashed line 2002–2016; solid line 2002–2020)
Many forums have provided opportunities for showcasing science and technology advances in DDDAS and through DDDAS. Some of the primary meetings that highlighted DDDAS-based advances include:
IEEE International Parallel and Distributed Processing Symposium (IPDPS) [18];
International Conference on Computational Science (ICCS) [19]; and
IEEE Winter Simulation Conference (WSC) [20, 21].
The opportunities have expanded into engineering conferences:
IEEE American Controls Conference (ACC) [22];
ISIF International Conference on Information Fusion (Fusion) [23];
AIAA Aviation [24]; and
ASME Embedded Systems and Applications [25, 26].
These and other science forums have hosted DDDAS workshops, panels, and presentations; such include: Data Stream (STREAM), American Geophysical Union (AGU), American Society of Mechanical Engineers (ASME), International Conference in Computational Sciences (ICCS), and Society for Industrial and Applied Mathematics (SIAM).
Along the way, numerous meetings and workshops were convened, including the Dynamic Data-driven Environmental Systems Science Conference (DyDESS) (2014). DyDESS focused on scienti ic methods in environmental sciences, such as: (1) Perspectives from Ocean State Estimation; (2) Imaging Earth’s interior with active and passive source
seismic data; (3) Objective Detection of Lagrangian Vortices in Unsteady Velocity Data; and (4) Data Assimilation and Controls for atmospheric mutiscale dimensional processing. The DDDAS/InfoSymbiotics conference (2016) spurred the genesis of this book, which includes research work presented at that conference, as well as other work done over the years, starting in 2000.
Over the years, many researchers have embraced the DDDAS concept with a variety of applications, as shown in Fig. 1.6, which is an illustration from the AFOSR-NSF 2010 Workshop Report [7]. Areas of interest shown in the illustration include adverse weather events (e.g., tornadoes, hurricanes, etc), environmental (e.g., atmospheric contaminant transport, wild ires, etc), aerospace (e.g., UAV swarms, decision support modeling, etc), and medical imaging, among many others. The DDDAS community is dedicated to showcasing scienti ic and technological advances in complex systems modeling and instrumentation methods. The next section cites a set of publications over the last 20 years, organizing them into the areas of theory, methods, and systems analysis and design.
Fig. 1.6 Examples of DDDAS impact (From: Report of the August 2010 Multi-Agency Workshop on Info/Symbiotics/DDDAS: The power of Dynamic Data Driven Applications Systems, AFOSR-NSF (Air Force Of ice of Scienti ic Research-National Science Foundation), 2010 [7])
The history of DDDAS extends over four decades, from the inception of the concept in 1980 to the last two decades of developments by the broader scienti ic communities, academic, industry, and federal labs in the US and internationally. To organize the diverse set of applications, three areas are highlighted here: (1) theory, (2) methods, and (3) designs. Key areas for theory are based in the scienti ic areas with large data collections and complex models. Methods include various engineering designs for a broad set of domains – space, air, and ground, where DDDAS supports dynamic response and control. Finally, examples are presented that include elements needed to support applications that require systems design and computational architectures. Given the large size of the DDDAS literature, various taxonomies could be highlighted; however, the organization here is an effort to provide the reader with perspectives of the wide-ranging in luence the DDDAS paradigm has had on the scienti ic, development, and design communities.
1.5.1 Theory: Modeling and Analysis
The DDDAS paradigm began with enhancing the phenomenology of science and engineering models such that, measurement information would (dynamically and adaptively) improve the resulting model. In 2003, several projects were launched that demonstrated key attributes in DDDAS, including faster and more accurate modeling, measurement information, dynamic data assimilation and adaptive sampling (going beyond the traditional data assimilation), incorporated into multiphysics [27], atmospheric modeling [28], and ocean forecasting [29]; as well as pandemics [30] and enterprise resource planning [17], discussed later (Sect. 5.3) in the context of application systems design. An application that bene ited from the DDDAS principles using science models was oil well placement [31].
As the DDDAS methods showed promise in science applications, a key area was in weather forecasting [8]. Researchers assessed tornado prediction [32], weather and climate analysis [33], and chemical transport models [34]. Simultaneously, DDDAS began addressing theoretical uncertainty and quantifying error minimization (Uncertainty
Quanti ication - UQ) [35]. Some years later, Ravela et al. [36] and others began to use the information from weather forecasting (e.g., coherent luid analysis) for advances in applications controls for UAVs and aircraft routing.
Along with weather forecasting, another set of environmental systems-related applications on wild ire monitoring were developed, such as agent-based simulations for ire propagation modeling [37], which remains of interest to the present. A set of researchers, led by Mandel and Coen with their initial work [38], continued to use the DDDAS paradigm for inclusion of advanced physical models of wild ire prediction with that of real-time sensing. Such work includes the CAWFE® (Coupled Atmosphere-Wildland Fire Environment) modeling system, which together with various sensors such as the Visible Infrared Imaging Radiometer Suite (VIIRS), provided analysis of smoke plume detection [39] in the United States. The wild ire assessment method was extended to other geographic locations such as in Europe [40]. Furthermore, ire detection and mitigation sought to understand the management of water distribution [41].
Another example in the environmental systems area, is that of volcanic ash detection by Bursik and Singla, et al. [42]. Atmospheric analysis can have impacts on commercial air transport, such as the 2010 eruption of the Eyja jallajokull volcano in Iceland. The particulates in the air from an eruption could have disastrous effects, for example on combustion engines of an aircraft lying through the sky region affected by volcanic ash spread. Likewise, with the detection of changes in the weather content, environmental wind context, and navigational data could be used to alter the air traf ic management of the networked skies. Advances in uncertainty quanti ication were incorporated into the ash movement modeling so as to prepare aviation for future events and provide passenger safety [43]. Uncertainty quanti ication helps in estimating error reduction in complex modeling and estimation methods [44].
Additional science and engineering applications bene itting from DDDAS-based methods include areas for bio-sensing and analysis for medical applications. One example is using image recognition for tracking human responses to stress and expressions. Metaxas et al. [45] developed DDDAS-based methods using image recognition and face tracking [46]. Other examples include using the sensing to update models
of humans in support of neurosurgery [47]. As a third set of examples, Oden et al. [48, 49] utilized DDDAS principles for laser treatment of cancer. In each of these cases, DDDAS methods have supported enhancements in medical treatment through advanced modeling. Other DDDAS-based developments presented in this book include diagnostics, chemical treatment, and pandemics.
1.5.2 Methods: Domain Applications
Building upon the DDDAS principles for science and engineering applications, another area of in luence are developments which applied the DDDAS-based dynamic data assimilation analysis to that of control and iltering. As highlighted earlier, an extension of the scienti ic modeling of the air environment was extended to the atmospheric environment for orbital awareness. Bernstein et al. [50] utilized the DDDAS principle for dynamic data assimilation, to use actual data to impart information into a model in aspects (of the system under study) not captured as well by the analytical (physical) model, and applied it in the case of the global ionosphere-thermosphere model (GITM). While it was applied as a scienti ic analysis, it showed how DDDAS can be applied for system-level model-based adaptive control and sensing. Simulations were conducted to determine the effects on planetary motion [51] and movement of atmospheric elements [52]. A third example extends these developments for the Retrospective Cost Model Re inement (RCMR) that includes modeling, sensing and control [53]. The capabilities achieved through this work provide for advances in satellite protection, orbital sensing, and understanding the far-earth environment.
Protection of aerial and space-based platforms, such as satellites, is also a key area for DDDAS-based advanced capabilities, including structural health monitoring (SHM), as exempli ied by several research efforts. Farhat, Michopoulos, Chang et al. [54] utilized the DDDAS principles towards SHM of structural and materials assessments, while Korobenko and Bazilevs et al. followed with aircraft composite structures work which is featured in this book. Creating an accurate model, with DDDAS-based embedded sensing, supports real-time response to a dynamically changing environment. Additionally, developments include reduced-order modeling (ROM) such that the ensemble of models can be re ined over model parameters, uncertainty estimation, and sensing bias [55]. Oden et al. [56] provided additional bene its of SHM for damage
assessment and others highlighted modeling updates that account for materials damage [57]. The book highlights recent advancements in SHM using the DDDAS paradigm such as for aerospace systems.
Expanding DDDAS-based SHM methods to create real-time decision support systems, Willcox and Allaire et al. [58] have utilized online/of line modeling in support of self-aware vehicles which paves the way for autonomous systems. Included in their research is a focus on the model dimensionality for operational performance [59]. As a second example, Mohseni et al. have utilized a wide variety of air and water autonomous systems and applied DDDAS for control and atmospheric sensing [60]. The monitoring of the environment was combined to support the health monitoring of the vehicles in a changing environment. These developments have been incorporated into the control of soaring vehicles [61]. The third example includes on-board avionics to sense fault detection [62]. Varela et al. has led a group to bring together the computational modeling with that of instrumentation (e.g., pitot tubes, electronics) health assessment for safe light [63]. Typically, estimation theory is employed for self-aware vehicles.
To achieve the capabilities for analysis over multiple domains requires the coordination and estimation across the techniques employed. Using the ensemble Kalman ilter, Sandu et al. [64] addressed the computational aspects of dynamic data assimilation for aerosol in the atmosphere while Ravela et al. [65] devised methods for aerial platforms and underwater sensors positioning. Other work examined methods to improve forecasting prediction [66]. If the DDDAS methods are able to forecast the movements, they can use ield alignment to estimate vehicle locations such as with quadrature information [67, 68]. Likewise, the idelity of the parameters affects the estimation of model accuracy [69], which enables a mixture of ensembles [70].
Estimation methods are elements of data fusion techniques. The integration of measurements includes data, sensor, and information fusion. Information fusion aligns well with the DDDAS principles [71], however in DDDAS the data fusion is inherently dynamic. Such an example is an array of sensors whose operation is guided by dynamic data-driven model for target detection and classi ication [72]. DDDAS hence can improve pattern recognition [73] or classi ication especially if data analysis is performed over features [74]. Recent methods have
combined heterogeneous data in support of nonlinear classi ication of moving objects using signal and pixel data [75].
Moving entity analysis includes object estimation. Hoffman et al. [76] used DDDAS-based modeling for adaptive sensor control in analysis of hyperspectral data to gather relevant features of the moving object. Fujimoto et al. [77] used similar DDDAS-based methods for ground vehicle-movement analysis, while others advanced the methods for multidimensional assignment in support of aerial vehicle monitoring [78].
In adaptive context-aware applications, the DDDAS concept leverages models such as scene, roads, or other terrain information. Context aware approaches were investigated [79], along with methods for measurement models which learn through state-augmenting contexts [80]. These methods were furthered by the information fusion community for context-enhanced information fusion which shows how DDDAS techniques can improve tracking over many operating conditions for robust performance [81].
Building on theory and methods, a number of efforts have demonstrated new capabilities for complex systems design and operation.
1.5.3 Analysis and Design: Systems and Architectures
The third section of this overview discusses systems architecture and systems analysis and design, for example for energy networks and cyber network analysis, and other systems design, with recent efforts also exploiting cloud computing. In the early methods of DDDAS, there was a need for scalable applications-systems’ architectures , and agent-based systems where evaluated [82]. DDDAS showed promise for supply chain analysis to improve the logistics and movement of parts [83]. In addition, the advent of web-based methods have provided a use case for distributed simulations for computer data streaming [84]. Web-based methods afford query languages for DDDAS designs [85] and analysis [86].
The distributed aspects of network analysis were adapted to be DDDAS-based and applied for power systems and energy analysis [87, 88]. Power analysis as a function of microgrids is aimed to support not
