FUTURE MANUFACTURING
Processes should not be held up by quality control . Rather , it has to be integrated into the production line and capable of reliably coping with fast cycle times of less than one second .
Visual Inspection in the automotive sector : Trained for different types of welds , AI can detect even the smallest anomalies .
bottles no longer have to be rejected and reheated later , but can be melted down and reshaped while still warm . As a result , the company not only saves a great deal of energy , but also the costs for all subsequent production steps , up to and including the removal of rejects .
Automated control directly in the production process
AI-supported quality control is integrated directly into the manufacturing process and offers numerous benefits compared to older solutions . Image recognition is by no means new in the field of quality assurance . Computer Vision has long been used to capture images and check them against predetermined criteria . Until now , this involved counting pixels , checking colour values and finding out whether a part was good or defective . The major disadvantage here is the time it takes to develop the corresponding algorithms – at least several weeks – and the fact that such bespoke solutions cannot be transferred to other areas .
Modern AI solutions are a game changer in this respect : Visual Inspection enables companies to implement initial use cases in a matter of days . There are two machine learning methods involved : firstly , the system is fed with a minimum number of labelled photos , i . e . images clearly marked as good or bad . By means of regular AI training , this knowledge base is constantly refined . The second option is to work with a reference image that is as perfect as possible , which the system compares against each of the relevant components . Pre-trained algorithms are at work in the background here , too .
Pre-analysis reveals functionality of application
GFT Technologies cooperates closely with its clients during the development of solutions . “ We either first consider together where Visual Inspection could be used or look at the process the client has already selected for the application on site ,” reports Denis Häussler , GFT ’ s Lead Consultant . “ We then check what infrastructure may already be in place and optimise it .” During the preliminary analysis stage , the company also uses its Spanish AI Lab to inspect faulty components . “ During this phase , we check whether we can actually solve the problem and with what degree of certainty the application can deliver correct results ,” explains Häussler .
This initial analysis phase is crucial for the progress of the project . “ There were some cases – minor paint defects , for example – where we were unable to deliver 100 percent defect detection . This is something we are quite open about ,” says Häussler . Since AI use cases can easily break down , the implementers work in small steps so they can react swiftly and successfully implement as many use cases as possible within a short space of time . Meanwhile , the users are trained in a kind of co-development process , enabling them to apply the knowledge increasingly themselves . “ We have customers with 60 or 70 such use cases and naturally they don ’ t want to be too dependent . They would rather be in charge themselves without having to set up their own team of AI experts ,” explains Häussler .
21