Prediction of toolwear from tool noises in turning using deep learning methods

The prediction of progressive tool wear and tool breakage, as well as the real-time monitoring of machining processes, is crucial for optimizing cutting parameters and establishing databases. CNC lathe systems optimized for high precision and accuracy are preferred in this context. Nowadays, an increasing number of sensors are being used in machine tools, leading to an exponential growth in input data. Appropriate systems for tool condition monitoring are selected based on the nature of the signals. Some of these systems include DL (deep learning) and ML (machine learning) models. In this study, sounds recorded during turning were analyzed using DL and ML models. The workpiece materials used were 316L stainless steel and 1050 steel. The features of the sound data were extracted using amplitude-time, mel-spectrogram, MFCCs (Mel Frequency Cepstral Coefficients), ZCRs (Zero Crossing Rates), and RMS (Root Mean Square Energy). DL models, including 1D CNN (Convolutional Neural Networks) and 2D CNN, as well as ML models such as KNN (K-Nearest Neighbors), SVM (Support Vector Machines), RF (Random Forest), and ensemble learning models, were trained using the extracted features. During the machining of 316L stainless steel, high prediction accuracies were achieved for tool wear with the 1D CNN model at 98.08%, the 2D CNN model at 96.72%, the KNN model at 94.26%, the SVM model at 90.43%, and the ensemble learning model at 96.99%. For the machining of 1050 steel, prediction accuracies for tool wear were achieved at 89.32% with the 1D CNN model, 91.24% with the 2D CNN model, 89.6% with the KNN model, 84.27% with the SVM model, 92.28% with the RF model, and 90.35% with the ensemble learning model. The DL and ML models demonstrated favorable results, indicating their applicability in the manufacturing industry. This study is noteworthy for its applicability in both industrial and academic fields, contributing to the literature and laying a foundation for future research.