Machine learning based defect detection for OEMs
AI First and Fabrimex Systems launch new software product
Fabrimex Systems relies on AI First to expand their product offering: the joint venture by 4Quant and Netcetera offering AI-driven software services is developing the computer vision product fs-vision that can detect defects on production lines. Fabrimex Systems, specialist for industrial cameras, embedded computing and machine vision in Switzerland, is an early adopter of this machine learning based technology from AI First. The software can be trained to identify defective work pieces during the manufacturing process, speeding up the supply chain and lowering production costs.
The development of technical machinery and equipment has become increasingly complex in recent years. At the same time, the product life cycles have become shorter. Detection of defects has to keep up with this development. Traditional computer models were in many cases unsuccessful in detecting the proper object shapes and the defects. Leveraging machine learning (ML) leads to successful and automated detection without human interaction. For Fabrimex Systems, AI First is currently developing the product fs-vision that can be deployed on production lines equipped with an overhead camera. Based on a small training set of images, the product detects similar and previously unseen defects on its own. Original equipment manufacturers (OEMs) applying the technology can therefore speed up their entire supply chain while lowering costs.
Thomas Graf, Head of Image Processing at Fabrimex: “We are certain that AI First is the right partner to build this visionary product with us. Together, we are a diverse team with longstanding technological experience as well as industry knowhow – just the right combination that enables us to serve better our customers in the manufacturing industries.”
Leverage ML to detect the smallest defects in real-time
fs-vision from AI First is based on deep learning techniques and consists of two modules: a trainer and a detector. The trainer requires a set of labeled (“OK” and “not OK”) image data each to produce a machine learning model. This model is then used by the detector to identify defects in real-time. When new objects need to be processed or when environmental conditions change, a retraining of the model requires almost no efforts or expert knowledge. This allows OEMs to rapidly adapt their production to market requests.
As the work pieces can be rather small with dimensions of 100mm x 80mm or even much smaller, the solution has to be very precise. As a result, defects can be very small, like e.g. an eyelash, and hard to detect. Defects might include foreign particles, impurities like grease or dirt, damages (holes, tears, rips), bad or unclear printing (no or partial printing, double printing).
Nithin Mathews, Business Developer at AI First: “Most OEMs have existing standards and business know-how when it comes to measuring the quality of their products. Machine learning leverages this knowledge and automates many of the manual steps still part of the quality assurance process in the manufacturing industry.”