DATA DRIVEN TOOL WEAR ANALYSIS THROUGH FEATURE SELECTION FROM PROCESS SIGNALS

Abstract

With the increasing adoption of data-driven techniques in manufacturing, industries are leveraging advanced analytics and machine learning to enhance process efficiency and reliability. One critical area where these techniques are making a significant impact is wear monitoring and anomaly detection in machining operations. Traditional machining processes often rely on predefined tool change intervals or operator experience, which can lead to inefficient tool usage, unexpected failures, and increased downtime. In contrast, data-driven manufacturing enables real-time monitoring of machining conditions, al-lowing for predictive maintenance strategies that optimize tool life and process stability. In this study, we explore the role of data-driven approaches in wear monitoring and anomaly detection, focusing on their effectiveness in improving machining performance. By integrating real-time sensor data with machine learning models, we analyzed wear patterns and identified early signs of anomalies that could lead to tool failure. Through extensive experimentation, we investigated the influence of various ma-chining parameters on tool degradation and the effectiveness of different anomaly detection algorithms in predicting failures. The results demonstrated that predictive models based on real-time data sig-nificantly enhanced the ability to detect wear-related anomalies at an early stage, reducing unplanned downtime and improving machining consistency.