Unsupervised Detection of Tool Breakage: A Novel Approach Based on Time and Sensor Domain Data Analysis

During metal machining processes, cutting tool chipping or breakage can deteriorate finished surfaces, leading to significant quality problems and economic loss. Tool breakage detection (TBD) can detect potential cutting tool failures, facilitate timely machine tool shutting down, and prevent product quality degradation and high manufacturing costs. Current TBD methods include both supervised and unsupervised techniques and generally consist of signal feature extraction and decision-making model building. The problems with existing supervised TBD are mainly associated with limited availability of training data while, for unsupervised TBD, the difficult issue is how to determine an appropriate TBD threshold. In the present study, a novel unsupervised TBD approach is proposed based on an innovative time and sensor domain data decomposition and analysis framework. Within this framework, multisensor signals are compressed into a significantly smaller set of time and sensor domain data. Then the time series analysis and sequential clustering are performed on the time domain and sensor domain data, respectively. The time series analysis is used to generate the features for TBD while the sensor domain data-based sequential clustering is exploited to determine the baseline passes and adaptively update the TBD threshold. The application to the multisensor data collected from two milling experiments demonstrates the effectiveness of the novel approach and its potential to be applied for real-time cutting TBD.