Acoustic signal based tool wear monitoring using a computationally efficient FFT-CNN-BiLSTM architecture

Abstract This paper presents a new approach to tool wear monitoring (TWM) in milling processes using acoustic signals. A publicly available dataset of Acoustic signals collected from various cutting tools operating at different wear levels and rotational speeds, was utilized to evaluate the performance of the proposed predictive models. A new FFT-CNN-BiLSTM architecture is proposed, integrating Fast Fourier Transform for spectral feature extraction, Convolutional Neural Networks for spatial pattern recognition, and Bidirectional Long Short-Term Memory networks for temporal sequence modeling. Comparative analyses against both conventional machine learning techniques (SVM) and other deep learning approaches (CNN, LSTM) demonstrate that the FFT-CNN-BiLSTM model achieves competitive performance metrics while maintaining computational efficiency. The findings demonstrate the effectiveness of audio signal analysis and deep learning models in optimizing tool condition assessment, contributing to advancement in smart manufacturing practices.