Machining Accuracy Prediction of Internal Gear Power Honing Process Based on MSACNN-Transformer Under Multisource Error Coupling

The internal gear power honing process is widely used in the gear machining of electric vehicles, and its machining accuracy seriously affects the gear meshing noise and transmission stability. Machining accuracy prediction provides the basis for accuracy modeling and error compensation. At present, most studies utilize single geometric error, thermal error, or cutting force error data to predict machining accuracy. However, gear honing machining accuracy is affected by multiple errors and the prediction accuracy of traditional methods is low. To this end, this article successfully combines multiscale adaptive convolutional neural network (MSACNN) and Transformer and proposes a novel machining accuracy prediction method based on MSACNN-Transformer under multisource error coupling. The influence and coupling mechanism of multisource errors on gear honing machining accuracy are analyzed. The constructed gear honing machining accuracy prediction model using the MSACNN-Transformer method adaptively learns multiscale coupling features and representative machining accuracy information from multisource error data through different convolution scales and the multihead self-attention mechanism. Experimental results demonstrate that the model established using multisource error data provides more abundant machining state information and has superior prediction performance than single error. Moreover, the proposed MSACNN-Transformer model shows higher prediction accuracy than other methods.