Statistical Monitoring of Machining Operations: Building Reference Models to Detect Process Abnormalities

Abstract

Numerous methods have been developed to detect process anomalies in machining, with statistical approaches for semi-supervised anomaly detection being among the most effective. These methods establish decision boundaries using data from normaloperations to evaluate process performance. This thesis presents a method for statistically building a reference model defined by envelopes calculated from successful similar operations, comparing operations of the same type to evaluate disparities from the reference model, summarizing the disparities into Key Performance Indicators (KPIs) and defining a logic to detect process anomalies based on the KPIs. Additionally, the thesis provides a detailed description of the developed service that implements this method for quality assessment within the gemineers GmbH product. Following implementation, a case study was conducted using data from various manufacturing processes for the same product to evaluate the method’s anomaly detection capabilities. As a result, the gemineers software now features a flexible and powerful tool for identifying machined parts or tools requiring inspection, without incurring long-term costs such as external library licenses.