Quality Prediction using Multiscale Convolutional VAEs for Thin Plate Parts

The machining quality of thin-walled parts is critical to the performance and reliability of high-value equipment. This study proposes a Multi-SPP-VAE model to improve the accuracy and robustness of dimensional error prediction in thin-plate machining. The model incorporates a multiscale convolutional architecture to extract both local and global features from cutting force signals, an attention mechanism to refine latent-space representations, and the fusion of static machining parameters to enhance contextual awareness.Key innovations include a novel multi-scale spatial pyramid pooling structure for improved noise suppression and temporal pattern representation, and an enhanced Grey Wolf Optimization (EGWO) algorithm with nonlinear convergence control and distance-weighted update mechanisms for automated hyperparameter tuning.Experimental evaluations demonstrate that with 108 convolutional channels and a 32-dimensional latent space, the Multi-SPP-VAE significantly outperforms conventional CNN, RNN, and LSTM-based baselines in MSE, RMSE, and MAE across multiple datasets, confirming its strong generalization and predictive performance.This work provides new insights into feature-level error prediction in thin-plate machining and offers a scalable, high-fidelity solution for real-time quality monitoring in intelligent manufacturing environments.