AI and looking into the crystal ball

Using artificial intelligence in production for error prediction and identification of error source

Errors occur in production processes. Even in carefully set up and maintained equipment, every now and then a defective piece is produced resulting from faulty interaction of the machines involved. In terms of quality management, it is crucial for companies to identify the source of such errors. After all, avoiding them reduces manufacturing costs and increases competitiveness. Sometimes even a small change in temperature can make the production process more reliable - but that first has to be found out.

Formula for quality

Gathering and evaluating quality data to enable troubleshooting is generally time-consuming and laborious. Many companies evaluate their production data manually, meaning employees are examining data with simple tables and statistics. Using Data Science to analyse this data also requires a great deal of effort and preparatory work. This is where the "PREFERML" research project, which has been headed by Professor Dr. Holger Ziekow at Furtwangen University for the past three and a half years, gets involved. "Very briefly, the idea of the project can be explained with this formula," says Ziekow, "AI + human = production quality."

Detailed analysis of processes

In the PREFERML project, Ziekow and his HFU team have succeeded in using artificial intelligence to identify correlations in production processes, as well as to predict and identify the causes of errors. The company partner in the PREFERML research project was SICK AG, who allowed their sensor manufacturing production processes to be examined in minute detail. "This was no easy task, as there was already a very, very low error rate there anyway," reports Prof. Ziekow. "That made it all the more important for us to use our software to accurately evaluate the huge amounts of data available for these minimal errors." The PREFERML project was funded by the German Ministry of Education and Research.

Transparency and learning effect

For optimised machine learning, Ziekow's team "fed" the AI with the domain knowledge that only skilled personnel familiar with production can have. With this knowledge and data from the production lines, the AI learned the conditions which must exist for the error to occur. Using technologies known as automatic learning (AutoML), relevant data was identified, and different types of errors were analysed virtually at the push of a button. "Of course, the point was to be able to proactively prevent errors from occurring in the first place," Ziekow said. "Technically, we also integrated Big Data solutions for this," he explains, "and so that users can understand what the AI is doing, we also integrated so-called Explainable Artificial Intelligence as a 'trust aspect' – like an MRI, so to speak, with which you can look into the AI brain to understand the factors involved. After the machine has learned, the human can then learn from the machine where something can be done better!"