Deep Learning for Automated Quality Control
It is here. Fully automated, intelligent systems that can aid the new-age factory in driving up productivity, crashing down waste and instilling confidence in the manufactured & shipped outputs. Real time display of results in easy to understand charts, highlighting key metrics that are crucial to every manufacturing process. Integrated into existing systems to upgrade rather than replace.
The past few years have seen immense growth in Artificial Intelligence (AI) and its implementations in Industry 4.0. With Internet of Things becoming mainstream in industry, the smart factories of the future are now here in the present. There are more computers monitoring manufacturing processes than ever before, and with these computers, comes the power of intelligent automation.
Traditional Vision Systems
Vision systems have been part of manufacturing for a better part of the last three decades. Non-contact measurement has been a significant application of vision systems so far. They work to automate inspection, with a high degree of restriction (e.g. position, orientation, size, part, to name a few), and to provide a highly accurate and reliable result. With the improvement in camera technology, such systems can provide measurements accurate to the nearest 10nm.
Traditional systems work by identifying 'features' - These features can include edges, corners, colours etc. Essentially, regions of interest, containing these features are identified on images for evaluation towards a decision on quality. Therefore, in order to identify these regions of interest in every image that is to be inspected, a particular orientation and strict standardization of part needs to be maintained.
Traditional vision systems work wonders in high accuracy measurement requirements, image stitching and 3D image reconstructions. However, they are limited in applications where consistency of positioning, orientation and lighting cannot be maintained.
AI Powered Vision Systems
AI, and specifically Deep Learning, has been making significant inroads into the world of machine vision. Deep Learning in Machine Vision involves the development of Neural Networks to correctly identify and classify features. Most commonly, AI systems work on the principle of 'It identifies what you teach it to identify'.
The development of AI powered vision systems starts with a set of images that roughly represent the variations of the feature to be identified. For the case of quality control, these features are mostly a set of defects. The variations generally include differences in size, location, color, illumination etc. This set of images is used to train an AI model, wherein it becomes acquainted with the defects. Then the trained AI model can then identify defects similar to those in the training dataset. The nature of models, choice of models and their evaluation is a very interesting study, but beyond the scope of this article.
Now, what makes these AI powered vision systems stand out from their traditional computer vision counterparts? In a nutshell, it is the flexibility and reliability.
AI vision systems do not rely on defining 'features' to detect. The neural network itself defines these features in an automated manner during training.
With a good training dataset, AI vision systems remain extremely tolerant towards change in ambient light, change in position and orientation of the object, and also, sometimes, a change in object itself.
AI vision systems can become highly scalable. With minimal, and low-complexity additional effort, new parts, defects, variations can be taught to the model to detect.
AI has been in the market for a long time now. However, with the rise of computational capabilities, much more powerful computers are now available at a reasonable price point. With this advent of available hardware, the applications of AI have skyrocketed. This is being increasingly pronounced in the machine vision arena.
Technology and a drive to push the limits of production efficiency have formed an excellent bond to develop innovative solutions for automation of quality inspection processes. The 'new normal' that we are in is increasingly leading towards strong remote monitoring & remote action capabilities and AI powered computer vision is beckoning to become a main stalwart in this need.
EaglAI Detect has been developed on a core of AI in Deep learning machine vision. Drop us a message for a detailed discussion and a demo on automation of your quality control process.