A multi-information dynamic fitting weighted strategy based on process knowledge for quality-driven process monitoring

The quality-oriented process monitoring method has become a prominent research focus in modern industrial applications. However, product quality is generally determined by multiple process variables, whereas quality measurements are typically sparse and subject to significant analytical delays. Therefore, to capture the intricate nonlinear dependencies between process variables and quality variables, a novel multivariate fitting weighted strategy based on graph convolutional networks (MFW-GCN) is proposed for quality indicator monitoring. Specifically, the framework incorporates a time-lagged weighting mechanism to address information propagation delays, coupled with graph convolution to model variable interactions. Crucially, a multivariate regression weighting mechanism is designed to quantify the specific contributions of process variables to quality indicators, enabling the extraction of quality-related latent features. Finally, the effectiveness of the proposed method has been validated on two industry benchmarks, demonstrating that it yields improved results compared to existing methods.