In the context of intelligent manufacturing, product quality control increasingly relies on deep insights into massive, high-dimensional, and temporal manufacturing process data. Traditional quality control methods have limitations in dealing with complex nonlinear relationships and spatiotemporal correlations. This article proposes a quality prediction and control framework that integrates improved spatiotemporal graph convolutional networks with Bayesian optimization. Firstly, by constructing a manufacturing unit relationship diagram and utilizing a spatiotemporal graph convolutional network to capture the spatiotemporal dependencies between multiple processes and multi-sensor data, accurate prediction of key quality attributes can be achieved. Furthermore, by combining the quality prediction model with Bayesian optimization, an active control strategy is constructed to dynamically optimize the process parameter set points and achieve feedforward quality control. Experiments on publicly available datasets and simulation cases have shown that the proposed prediction model outperforms traditional time series models (ARIMA, SVR) and baseline deep learning models (LSTM, CNN-LSTM) in terms of root mean square error, mean absolute error, and coefficient of determination. The proactive control strategy based on prediction is significantly superior to traditional statistical process control and rule-based feedback control in terms of control accuracy, stability, and cost-effectiveness. This study provides effective theoretical methods and technical paths for constructing an adaptive and forward-looking intelligent quality control system.
Dai et al. (Thu,) studied this question.