Remaining Useful Life Prediction of Milling Tool Based on Pyramid CNN

Remaining useful life prediction of a milling tool is one the determinants in making scientific maintenance decision for CNC machine tool. Predicting RUL accurately can improve machining efficiency and quality product. Deep learning methods have strong capability are extensively used. Multiscale CNN, typical deep model prediction, has large number parameters because its parallel convolutional pathways, resulting high computing cost. Besides, MSCNN ignores various influences different scales degradation features on accuracy. To address issue, pyramid CNN (PCNN) proposed this paper. Group convolution used to replace pathways extract multiscale without additional parameters. And channel attention with soft assignment select key features, considering sensors scales. The wear experiments show that score value method achieved 51.248 ± 1.712 RMSE 19.051 0.804, confirming better performance compared traditional other methods. reduced by 62.6% 54.8% self-attention methods, lower