Estimating tool wear using multi-sensor data fusion and machine learning techniques

Abstract

Modern manufacturing industries are being transformed by the integration of sensor technology, data science, and machine learning, leading to smarter, more efficient operations. Advancements in equipment health monitoring are crucial for improving productivity, extending equipment lifespan, and ensuring consistent product quality. In computer numerical control (CNC) machining, worn tools contribute to increased forces and vibrations, negatively impacting both machine performance and part quality. Traditional tool condition monitoring methods, which rely on manual offline inspections, result in machine downtime and decreased productivity. Modern tool condition monitoring methods involve monitoring tools based on single-sensor analysis. While a single sensor can detect tool wear within a machine, it fails to capture the full range of system behavior, potentially overlooking critical anomalies indictive of tool wear. To address these challenges, automated monitoring systems utilizing multisensory data and machine learning techniques have been developed, enabling real-time monitoring and prediction of tool wear. This research introduces a novel three-level data fusion framework for predicting tool flank wear in CNC machining. Force, vibration, and sound data was collected using various sensors during a CNC milling operation. The raw sensor data was processed and transformed into distinct statistical features to train machine learning models. A stacking ensemble method combining a random forest, artificial neural network, and extreme gradient boosting algorithm was employed to enhance predictive accuracy, achieving an R 2 value of 0.982, and root mean squared error of 37.146 micrometers. The proposed three-level fusion framework proved to be highly effective in predicting tool flank wear and shows great potential for monitoring the health of engineering equipment across a variety of industries.