Dynamic Latent Variable Analysis for Multirate Closed-Loop Process Monitoring

Closed-loop control is widely adopted in industrial processes. It brings challenges for dynamic latent modeling and monitoring. Multi-rate process measurements in real applications even further complicate these tasks. In this paper, a multi-rate closed-loop dynamic latent variable (M-CLDLV) scheme is proposed for dynamic latent variable analysis and monitoring for closed-loop process with multi-rate measurements. A new dynamic latent variable model is proposed to capture the dynamic latent structure of the multi-rate closed-loop process data. To extract the dynamic latent structure for multi-rate closed-loop process data, an algorithm within the expectation maximization (EM) framework is developed, where a weighted likelihood function is further proposed to handle the asymmetries between the fast and slow sampled variables. A process monitoring strategy is thus proposed based on the extracted multi-rate dynamic latent variable model with comprehensive insight of dynamic and static relations. An experiment study on the deep-sea submersible shows the superiority of M-CLDLV and the introduction of the weighted likelihood function.