Development of a Demonstator Plant for Hot Stamping of Metal Sheets with a Machine Learning Assisted Anomaly Detection Control System

Due to the demographic developments of an increasingly aging society, the number of employees retiring is also rising. Likewise, the fluctuation within the individual companies has increased drastically within the last few years. As a result, experience gained over many years is being lost. On the other hand, there is a strong trend towards an ever-increasing number of very complex manufacturing processes with a large number of process influencing variables, where the human monitoring capabilities are limited. In this work we study the complex process of hot stamping of metal blanks for car body parts with a demonstrator plant. The demonstrator plant consists of five main components, comprising a magazine for material feed, an industrial robot for transferring the raw material and the finished part, an annealing furnace, and a hydraulic press with the temperature controlled hot forming tool inside. All machines have open communication interfaces with which all sensor data can be accessed. The demonstrator plant is additionally outfitted with different sensors measuring crucial process parameters. We use temperature sensors in the punch and die and make use of capturing the structure-borne sound and mechanical vibrations during the process to closely monitor the process. This sensor data is analyzed and methods of machine learning are used to detect anomalies in the process data. The goal of this work is to develop a methodology for the systematic detection of anomalies in the hot forming process using machine learning methods within a demonstrator plant that are suitable for real time anomaly detection for a high level control system. We use unsupervised machine learning techniques and neural networks to detect anomalies with great accuracy on our test data.