Anomaly detection in injection molding process data based on unsupervised learning
Plastic processing companies in high-wage countries are facing continuously increasing cost and quality pressures. In many applications, a 100 % quality control leads to unreasonable efforts. Hence, quality forecasting or control based on process data would be desirable. Neural Networks have been applied. However, their success depends on the appropriate labeling of the process data. Since during the process, it is usually unknown whether a good or bad part has been produced in one cycle, supervised machine learning is not applicable. Here, we present approaches to anomaly detection in injection molding process data by means of unsupervised machine learning.
READ FULL TEXT