Neural network-based adaptive fault-tolerant control for load following of a MHTGR with prescribed performance and CRDM faults

Abstract

A neural network-based adaptive fault-tolerant control scheme (NN-AFTCS) for load following of a modular high-temperature gas-cooled reactor (MHTGR) system possessing parameter uncertainties, exogenous disturbances, and control rod drive mechanism (CRDM) faults, capable of guaranteeing prescribed performance is developed in this paper. From practical point of view, the mathematical model of the MHTGR system is first described with consideration of total CRDM fault effects together with lumped uncertainties that consist of parameter uncertainties, exogenous disturbances, and unmeasured states, where the unknown values of both time-varying total CRDM fault effects and lumped uncertainties are approximated by utilizing the radial basis function neural network (RBFNN) owing to its online learning capability. After that, with the help of the RBFNN, a fault-tolerant controller is proposed to compensate the adverse effects of the CRDM faults and lumped uncertainties, while at the same time achieving prescribed performance guarantees by invoking error transformation technique. Moreover, the dynamic surface control technique is also employed such that the problem of “explosion of complexity” existing in the conventional backstepping is averted. With Lyapunov stability theorem, it is proved that the load following error in the closed-loop MHTGR system would converge to a small residual set within a finite time. At last, simulation and comparison results confirm that the proposed NN-AFTCS exhibits higher load following accuracy, better CRDM fault-tolerant capability, and stronger robustness against lumped uncertainties than a disturbance observer-sliding mode controller and an active disturbance rejection controller.