End milling tool condition monitoring with vibration signals and machine learning classifiers: A comparative study

Contemporary industrial businesses are encountering issues related to tool wear. To address this issue, the implementation of tool condition monitoring (TCM) is imperative, necessitating the development of such solutions. These technologies will help improve the tool's product quality and life and ultimately reduce the tool wear. The present article focuses on multipoint milling tool supervision (Health Monitoring of End Milling Tool) using a Machine Learning approach. In this experimentation, vibration signature data has collected for the multipoint milling tool, extracted vibration signatures (signals) have further used for statistical feature extraction. The J48 algorithm's decision tree have taken to select the best features. The fault classification has carried with a different classifier such as Logit Boost, linear regression, Simple Logistic, and Bagged trees classifier. Comparison of these classifiers is carried out based on vibration signatures, and out of these classifiers, the Logit Boost classifier gives better classification and accuracy than other classifiers. In the Logit Boost, linear regression, Simple Logistic, and Bagged trees classifier, maximum classification accuracy was 95.83%, 95%, 94.38%, and 94.54%, respectively, with a selection of the best statistical features. Similarly, classification accuracy was 96.25%, 95.2083%, 93.75%, and 92.7083% without selecting the best features. The wavelet analysis has been used to compare maximum classification accuracy with statistical feature extraction.In this Daubechies, Symlets, Haar and Discrete Meyer wavelet, Coiflet wavelet family classification accuracy have obtained. In the Daubechies wavelet family, utmost classification accuracy is 99.38% with db5 wavelet, and in the symlets wavelet family, utmost classification accuracy has received with a simple logistic classifier, i.e., 99.17% with sym5 wavelet.