Milling Cutter Wear Prediction Based on Bidirectional Long Short-Term Memory Neural Networks

Abstract:
The wear of milling cutters during high-speed milling processes affects the quality of the work-piece, productivity and manufacturing costs. Accurate tool wear prediction can therefore optimize production decisions and avoid losses due to tool wear. To improve the prediction precision, this paper proposes a method for predicting the wear of milling cutters based on bidirectional long short-term memory neural networks and attention mechanism. By extracting the time domain features, frequency domain features and time-frequency domain features based on time series data generated during industrial processes and using bidirectional long short-term memory neural networks to learn the complex mapping relationship between the extracted features and the wear amount. Experimental results show that this method reduces mean square error by 70.81% and 47.97% compared to the SVR model and the BPNN model respectively, and reduces mean square error by 79.69% and mean absolute percentage error by 5.977% compared to the CNN, enables a more accurate and reliable milling cutter wear prediction.