Anomaly Detection in Orthogonal Metal Cutting based on Autoencoder Method

Abstract

The choice of appropriate cutting conditions is widely acknowledged as a key performance indicator for efficient machining, since it allows to mitigate tools wear. In precision machinery, tool wear can indeed lead to poor surface quality and even affect the dimensions of the final product. In such a context, the cutting and feed forces, along with other parameters are affected and are not optimal. It is therefore necessary to detect any anomaly and provide the operators in the plant with a decision support, allowing to monitor the machining conditions in order to predict the outputs values and anticipate the wear's damage on the whole cutting process. In this article, cutting speed, cutting angle, cutting width and feed depth are the input parameters. Using the extended-Oxley analytical model of orthogonal metal cutting, a sample of cutting conditions has been simulated in order to analyze the optimality of the cutting force, the feed force and the internal temperature, based on the autoencoder method. Given a threshold for decision, the results allow to identify abnormal parameters and provide significant insights for operators, allowing them to avoid error and make the best choices of the inputs for optimal cutting conditions.