Stamping defect detection has been one of the last remaining frontiers in automotive machine vision for years. It’s a difficult problem to solve at every step: big parts, tiny defects, shiny parts, complex part shape/depth, enormous variability between defects, and a high number of part styles, just to name a few. To find tiny defects you need a lot of pixels, and the larger the parts get, that number increases exponentially . This massive amount of data then needs to be analyzed and done so before the next part finds its way down the line to
restart the process. Technology has finally gotten to a place with the advancement of GPUs where we can crunch that information into useable form. By combining modern cameras capable of acquiring images with incredible resolution, and the improvement of deep learning, artificial intelligence models, a fully in-line stamping defect detection system is now possible.
At USS, we’ve spent the last three years working R&D on building out a system that picks up splits exclusively. The first challenge, mechanically designing a system with the ability to see every nook and cranny of a freshly stamped part and casting even illumination across the entirety to produce a good, usable image. The second, taking those images and building a robust AI model to define exactly what is and what’s not to be considered a defect in the stamping. After compiling millions of images of raw data and sorting through to provide a working model of hundreds of thousands of different splits, we’ve finally cracked the code in this development. Our system deploys a fleet of strategically placed lights and a
vast array of cameras capturing features from all angles to inspect every part as it exits the press. With calculated redundancy and pixel overlay, our model then analyzes the signal to noise ratio and flags even the smallest cracks in metal surfaces. This is a massive step-forward for automotive production as this process has always been done by manual operators. In developing this solution, our systems will bring more consistency to end users while saving the stamping facility huge amounts of both time and money.