Bidirectional Dynamic Latent Variable Analysis for Closed-Loop Process Monitoring

Closed-loop data are widely encountered in modern industrial systems, which require special data analytics to gain insight for system monitoring. A closed-loop dynamic latent analysis scheme named closed-loop DiCCA (CL-DiCCA) is proposed in this article. Bidirectional dynamic latent variable relationships are proposed with a new objective to extract the closed-loop dynamic latent structure. An iterative algorithm is proposed to solve the constructed optimization problem for closed-loop processes. Four statistically independent residuals are generated, which monitor the dynamic and static variations of the process data. A process monitoring logic with the CL-DiCCA model is established, which offers further separation of faults into output-relevant and output-irrelevant ones. A numerical simulation and a case study on the thruster system of the Jiaolong deep-sea submersible are provided to illustrate the effectiveness of the proposed method.