7 minute read
Inspection: Applied Vision
Brian Ensinger* discusses how Applied Vision designs multispectral illumination solutions for glass container sidewall inspection to help manufacturers operate with speed and efficiency.
Applied Vision Corporation has engineered multispectral illumination solutions for all types of containers since 1997. As one of the first companies to utilise programmable, multispectral light-emitting diode (LED) light sources and colour cameras to benefit container inspection, Applied Vision has developed an approach that allows vision systems to acquire and process transparent, opaque, and other categories of defects from a single illumination pattern where each defect has its own optimal filtered light, thereby creating the best defect signature for detection.
The use of colour illumination for glass container inspection can provide manufacturing facilities with a customisable solution when faced with process-induced defects and the ever-growing complexity of containers, particularly for foods and beverages. This led Applied Vision to engineer the Volcano inspection station that is fully programmable in terms of intensity, colour selection, and pattern geometry. It also uses the highest-quality colour LEDs available, which improves defect detection and recognition along with improved defect classification for both top-down and sidewall inspection regardless of heavy embossing, the colour of the glass, or even the thickness of the container. In tandem with highresolution color cameras, lensing and optics, powerful software, and selflearning algorithms, plant engineers and QA managers gain the ability to enhance or ‘tune’ images prior to applying inspections, thus supporting greater operational efficiency and consistency.
Defects require different approaches
All glass container defects require a different lighting geometry and optical
� Fig 1. Multispectral illumination puts colour to work so that glass container manufacturers can optimise the accuracy, speed, and efficiency of sidewall inspections.
configuration to optimise detection accuracy and precision while minimising false rejects. That is, the illumination source and the illumination geometry must be developed to allow as many defects as possible to be detectable on the surface of the container, within the container, and below the surface (with some defects being highly detectable and others remaining less so). This is also dependent upon where the problem is occurring (Fig 1).
In the past, using conventional lighting techniques, cameras, and geometries, it has often been necessary to compromise one type of defect detection for another. Essentially, the user must optimise lighting and settings for a particular defect of interest while working to minimise negative impacts to other inspections being performed. As a consequence, machine setup becomes more complicated and timeconsuming while detection capabilities are diminished for certain defects, which can lead to more false rejects or defective containers escaping notice.
Helping companies avoid compromises and implement effective sidewall inspection on their production lines has been an area of focus for Applied Vision for decades. Today, sidewall inspection is a priority for glass manufacturers intent on ensuring quality and reducing cost amid rising global demand for food and beverage containers. By continuing to develop proprietary tools and approaches, Applied Vision aims to deliver the next generation of ‘no-touch’ solutions for multispectral illumination of glass sidewall surfaces.
Ahead of the curve
Combining many sidewall defect detections into a single solution that helps to reduce false reject rates is achieved with Applied Vision’s patented approach to machine vision colour inspection. This methodology improves on incumbent approaches whereby defects compete (often unsuccessfully) for optimal lighting geometry, lensing and optics, and camera resolution within the inspection station.
By leveraging high-resolution colour cameras and a configurable geometry of multispectral, software-controlled LEDs, the contours of the glass container being imaged cause different colours of light to be reflected back to the camera. Applied Vision designed a system that incorporates 16 equally spaced colour cameras and fully calibrated imaging geometry. This optical configuration also utilises a lensing principle to correct for perspective distortion in photographs.
Within the inspection station, optics are positioned to place a horizontal ray of lower cameras at the base of the containers (viewing parallel across the
� Fig 2 & Fig 3. Applied Vision’s multispectral illumination module is offered in the Volcano sidewall inspection system that makes light work of heavily decorated glass containers without contact or rotation.
conveyor view) and a horizontal ray of upper cameras at the finish (viewing directly across the top of the container), which helps to ensure the entire image will be sharp and in focus with capabilities in the heel and push-up areas of the container (Figs 2&3). Positioning each of the illuminators orthogonal to the central camera ray also reduces shadowing while allowing for complete coverage of the container across all views.
Imaging geometry, which does not rely on sensitive folding optics, is designed to minimise the standoff distance between cameras and the surface being inspected.
Furthermore, the image angle provided by the upper cameras facilitates improved inspection of the shoulder regions of the container.
As a result, a near infinite number of filtered colour images can be produced from a single image then utilised for specific defect inspection.
Within this scope falls opaque defects that are typically inclusions (stones and bits of refractory, for instance) as well as transparent defects ranging from blisters and bubbles to seeds and tears. Dedicated cameras for internal stresses using circular polarisers (as used in lab equipment) are included in the Volcano sidewall inspection solution.
Dimensional inspection, meanwhile, is concerned with lean, filler lean, height, diameter, and cap and neck finish (‘E’ and ‘T’ standard dimensional measurements) as the machine ‘sees’ the ideal shape and size of the container outlined by its design. Optimal edge detection lighting is important for dimensional inspections so as not to compromise other inspections. All the defect detections noted here are achievable using multispectral illumination within a glassmaking facility on lines operating at typical production speeds.
This is made possible in part by other built-in inspection station technologies developed by Applied Vision that bring novel software architecture and learning algorithms to bear on the problems that cost manufacturers time and money. A brief look at these capabilities includes: � A ‘sentinel’ software tool that highlights regions of the color image that deviate from the established statistical model for a particular glass container. These anomalous regions are candidate defects. The tool learns what is normal and what is abnormal by training on a large number of containers. It learns the appearance of the container as well as how much variation in appearance to expect. Notable is that the training set does not require perfect containers. Rather, it may include a representative distribution of containers, even those exhibiting defects. � Defect classification that utilises an algorithm termed the ‘blob classifier’. Plant engineers and QA managers want to know exactly what defects are being produced and at what frequency. This tool locates and categorises anomalies, grading each by severity. It then applies user adjustable criteria to identify which (if any) anomalies are defects. If defects are identified, the container is rejected. � A Bottle Geometry Tool that essentially learns, locates, and registers the sidewall in the image automatically, taking the burden of location setup away from the machine operator. Once an inspection sequence has been established, it can be saved and recalled for future use when running the container again or as a starting point for similar products. As images are registered, a large library of algorithms can be utilised for inspecting for defects in these processed images. The difference compared to many other machine vision technologies is that each targeted inspection is assigned to a specific colour filter that provides the best contrast and signal for detection.
Applied Vision inspection algorithms are some of the fastest in the industry enabling a tremendous amount of image processing and decision-making to be accomplished in a short amount of time. At the same time, harnessing the full colour spectrum to benefit sidewall inspection greatly improves the ability to reject nonconforming containers and provide feedback to the hot end.
Multispectral illumination solutions from Applied Vision meet the company’s mission to provide contactless inspection for glass containers throughout the manufacturing process. Handling and rotating containers create material losses while also limiting line speeds and capabilities in certain cases.
A no-touch inspection can provide substantially increased line speeds and layout efficiencies, improved yields by reducing loss associated with container breakage, energy consumption decreases, and dramatically improved job change times. Manufacturing space is also freed up and maintenance costs come down when containers do not have to be grasped or held to be rotated.
Today, machine vision inspection technologies powered by full spectrum colour capabilities, feature-rich software, and deep-learning algorithms are helping manufacturers of glass containers deploy effective solutions for sidewall inspection at line speeds. For the food and beverage industry, there are many benefits to explore. �
*Product Line Manager, Glass Inspection, Applied Vision Corporation, Ohio, USA www.appliedvision.com
