Nonlinear bidirectional dynamic latent modeling for closed-loop process monitoring

Closed-loop control brings new challenges for process monitoring. Nonlinearities in the correlation structure and the dynamic relations in closed-loop process data further complicate the task. To tackle these issues, a new dynamic latent variable scheme is proposed in this paper for nonlinear closed-loop processes. The nonlinearities of the processes are investigated in detail first, followed by a dynamic latent variable object for latent structure extraction of nonlinear closed-loop processes. An algorithm is then proposed to solve the constructed optimization object which has guaranteed convergency. The relations of the extracted dynamic latent model are revealed in detail afterwards, and the process monitoring strategy is built with comprehensive analysis of variations in closed-loop process data. A numerical simulation and a study on the thruster system of the “Jiaolong” deep-sea submarine are carried on to show the performance of the proposed scheme.