Concurrent vision technology
Automated surface inspection using camera systems is being widely adopted in factories to save manpower, reduce defect rates, and increase inspection speed. However, traditional vision systems have issues consistently achieving zero false detection rates when light conditions, product surfaces, or defect appearances change. Machine learning provides an easy way to automatically adapt inspection systems to changing conditions without expensive hardware or software adjustments. Concurrent Vision developed intelligent inspection software using a machine learning algorithm and feature detection methods to learn how defects appear in different situations from image data. During training, operators optimize the system's performance until it robustly adapts to changes with long up-times.
Concurrent vision technology
- 1. Quality assurance using vision systems Automated Surface inspection on products at factory floor using camera systems, is increasingly gaining momentum. Flaws and defects are detected at the production line. Automated surface inspection is being widely adopted in factories as a manner of Saving man power Reducing defect escape rates Increasing speed of flaw and defect detection Problem- System needs to adapt Traditional surface inspection methods fall short in producing consistent zero false inspection rates when the following conditions occur in combination or separately. Light conditions vary significantly from the conditions that existed under parameter design and test Product surface shininess and color changes form part-to-part or from batch-to-batch Product surface background varies Defects change appearance and shape Defects appear on different surface backgrounds Normal artifacts on surfaces resemble defects New defect types arise When that happens, the vision inspection system would require adjustments. Such adjustments often involve both SW and HW changes including parameter optimization to adapt the system to the changes and keep low false detection rates. This adaptation process is expensive in terms of system uptime and cost. The process will also be repeated every time a new condition happens. Solution- Artificial intelligence Machine learning provides and easy and cheap method of adaptation to changing conditions. Extremely smart systems adapt themselves automatically. They learn new appearances of defects and build robustness against changes in surface backgrounds and illumination conditions. More realistic systems require intervention from experienced operators to aid learning new defect types and surface backgrounds. The frequency of needed operator intervention declines as function of successful trainings as the system learns in a progressive manner.
- 2. Concurrent Vision Developed an intelligent surface inspection Software. The software is based on a powerful machine learning algorithm and a number of proven efficient feature detection methods. The feature detection methods have all their strong and weak sides with respect to a given problem. In system design phase, the feature detection methods are selected and combined to complement each other in gaining significant features from image patches. These features do carefully characterize different backgrounds and defects in their range of variation. A deep neural network uses these features to learn how different defects or flaws would appear in a wide variety of occurrences. In the training and validation phase, experienced quality inspection operators monitor false detection rates and use facilities in the provided graphical user interface to optimize the performance. The whole method shows to be robust against changes. Systems based on the method are characterized by Long life time High system uptime Fast and easy adaptation