A New Mechanism Model-Assisted Spatiotemporal Information Fusion Quality-Related Fault Diagnosis Method for Large-Scale Industrial Processes

Modern industrial processes with long production lines and multiple operating units exhibit complex spatiotemporal characteristics under the constraints of information flow, material flow, and energy flow. The quality spatial heritability and dynamic-coupled variables of long-term processes bring serious challenges to conventional quality-related fault diagnosis. Against this background, a new mechanism model-assisted spatiotemporal information fusion quality-related fault diagnosis method is proposed. First, the coupling relationships among variables in the mechanism models are intuitively represented as causal correlation networks (CCNs), and large-scale industrial processes are decomposed into multiple subsystems interpretably by CCNs. Second, a novel 1-D convolutional autoencoder with neighborhood preserving embedding in Riemannian space (1DCAE-RNPE) is designed, where 1DCAE is adopted to capture temporal features stably and RNPE mines spatial features in higher dimensions. On this basis, the correlations among spatiotemporal features and quality variables are explored by distributed canonical correlation analysis (DCCA). Bayesian inference is enforced to establish global monitoring statistics. In addition, after a fault occurs, the contributions of the variables are obtained to identify the fault variables and combined with CCNs to recognize the root cause. Finally, the effectiveness and advantages of the proposed method are demonstrated through a practical large-scale industrial process, the hot strip mill process (HSMP). Superior fault diagnosis performance has been attained compared to other competing methods. Meanwhile, root cause recognition is accurately realized.