An Online Milling Surface Roughness Prediction Method for Thin-Walled Blade Based on Wavelet Transform and Convolutional Neural Network Integrated with a Digital Twin System

The prediction of machining quality is vital for enhancing product performance, reducing costs, and optimizing resource utilization. Current research primarily focuses on offline prediction, limiting adaptability to dynamic machining conditions. To address this, this paper proposes an online prediction method based on Wavelet Transform and Convolutional Neural Network (WTCNN). This approach captures dynamic signals' time-frequency characteristics for deep learning of machining quality. Using five-axis blade milling as an example, cutting force and vibration signals undergo wavelet transform, fused into RGB graphs, and fed into the neural network for predicting surface roughness. A five-axis milling digital twin system triggers WTCNN model execution based on motion information, enabling real-time rendering of predicted results with position information. Experimental results show an 85% accuracy rate and 347ms time delay. The developed digital twin system achieves real-time mapping of machine tool motion, part cutting, and machining quality.