Fault-tolerant neural-network iterative learning control for a cascade of interconnected continuous stirred tank reactors with output constraints

The continuous stirred tank reactor (CSTR) is a cornerstone of chemical production processes, where maintaining precise control of key variables is critical for product quality and operational safety. This paper addresses the challenge of controlling a Cascade-interconnected CSTR (CCSTR) system subject to actuator faults and stringent output constraints. We develop a novel Neural Network-based Iterative Learning Control (NNILC) algorithm to achieve high-precision tracking over repeated operational batches. The complex CCSTR dynamics are first transformed into a pure-feedback nonlinear structure via the Mean Value Theorem, facilitating controller design. Given the fundamental requirement to constrain reactant concentration and mixture temperature within specific bounds, a command-filtered backstepping technique is integrated with universal Barrier Lyapunov Functions (UBLFs). This integration guarantees the satisfaction of output constraints, ensures controller feasibility, and delivers prescribed transient performance. Through a composite energy function-based analysis, we rigorously prove that under the proposed NNILC scheme, all closed-loop signals remain bounded and the tracking errors uniformly converge to an arbitrarily small residual set over the iteration domain. The efficacy and robustness of the proposed approach are further demonstrated through numerical simulations.