Classification of Production Process Phases with Multivariate Time Series Techniques

This article presents an approach for using Multivariate Time Series Classification (MTSC) to determine the basic production phases from the data of a numerical control (NC) of a manufacturing machine. The milling process in a series production of a contract manufacturer in Germany was selected as the use case. The MTSC techniques LSTM, 1D-CNN, and LSTM-FCN were selected for modeling the classification approach. All modeling phases are explained and finally, these techniques are compared with a standard neural network approach that performs classification time-independent. Overall, using 1D-CNN networks showed a clear superiority over the other models.