Sensors selection for tool failure detection during machining processes: A simple accurate classification model

Abstract Tool failure detection is a crucial task for continuous safe machining operations. In this study, a novel approach is proposed to develop an accurate and simple tool condition classification model (TCCM) for early failure detection during machining processes. Signals from current, vibration, and acoustic emission sensors were preprocessed and used for feature extraction in both time and frequency domains, leading to a total of 152 features. Next, a feature reduction was carried out based on relative importance, computed using a fully-grown random forest, which reduced the number of features to 15. To find out the best combination of relevant signal features, a total of 32,767 optimized support vector classifiers were developed. The comparison between different candidate models was based on both accuracy and complexity. The results showed that a classification accuracy up to 0.911 is attainable for a process-independent classification model using only current sensors. Besides, developing an ensemble of material-dependent models showed a good potential for improvement, recording a classification accuracy up to 0.958 while using features extracted only from the current sensors. The novelty in the present study is in its focus on developing a single sensor-based high-accuracy TCCM. This opens the door for wider utilization of such technology, especially that all existing studies focused on increasing the accuracy using multi-sensor TCCMs, which increases the cost of this technology and makes it inaccessible, especially for small and medium enterprises.