Knowledge elicitation for fault diagnostics in plastic injection moulding: A case for machine-to-machine communication

Abstract In most manufacturing processes the defect rate is very low. Sometimes, only a few parts per million are defective because of a faulty process. For this reason, fault diagnostics is faced with extremely imbalanced data sets and requires large volumes of data to achieve a reasonable performance. This paper explores whether a machine-to-machine approach can be used, in which several work systems share the process data to improve the accuracy of the fault-detection model. The model is based on machine learning and is applied to industrial data from approximately two million process cycles performed on several injection moulding work systems.