Data-driven predictive adaptive iterative learning fault-tolerant control for networked batch processes

Abstract

This article studies the fault-tolerant control (FTC) problem for a class of networked nonlinear batch processes. Firstly, the controlled batch process is converted to an adaptive data-driven model equivalent to the original system by using the iterative dynamic linearization technique, with actuator faults and fading communication phenomena considered in the control input and output channel, respectively. Among them, the fading communication phenomenon is modeled as an independent identically distributed over the iteration and time domains with known mathematical expectation and variance. Then, by fully combining the idea of predictive control and the output fading compensation algorithm, the data-driven predictive adaptive iterative learning FTC (DDPAILFTC) scheme is designed based on the dual-domain (iteration and time domains) compensation mechanism, which can avoid a short-sighted control decision and suppress the adverse effect brought by fading communication. Next, the strict convergence analysis of the presented DDPAILFTC approach is carried out by using the contraction mapping principle. The design and analysis process of the control scheme is completely data-driven and does not require any explicit model information. Ultimately, the effectiveness of the developed control method is demonstrated with a temperature tracking control example of a nonlinear batch reactor. The results show that the proposed DDPAILFTC strategy reduces the average MAE, average MSE, and calculation time by 20%, 21 %, and 31%, respectively, compared with ILFTC, and 18%, 15%, and 52%, respectively, compared with PILFTC.