CUAD-Mo: Continuos Unsupervised Anomaly Detection on Machining Operations

Computer Numerical Control (CNC) machine tools have become essential elements in manufacturing industries because they allow optimizing time and effort during production. Monitoring and diagnostic of such machines is a particularly important task to limit the number of defective products and to meet the required quality goals. In the context of Industry 4.0, Machine Learning approaches for Anomaly Detection have proven to be successfully used to detect abnormal events in industrial environments. When consumed by human operators in Decision Support Systems, it is desirable for AD technologies to indicate what particular feature triggered the abnormal prediction, thus, the usage of explainable methods is of fundamental importance to enable Root Cause Analysis and to allow timely and adequate interventions. In this work, we propose an approach, named CUAD-Mo (Continuos Unsupervised Anomaly Detection on Machining Operations), an Anomaly Detection pipeline that, from a set of features extracted from the time series obtained during the execution of a CNC machine, obtains an anomaly score using the Isolation Forest model and provide post-hoc explanations about the score obtained. CUAD-Mo was evaluated using real time data from a CNC machine used in production and achieved satisfactory both qualitative and quantitative results.