Learning-based fault tolerant control in dynamic chemical processes under incipient faults

This paper proposes a learning-based fault-tolerant control (FTC) framework for dynamic chemical processes subject to incipient faults. The proposed method employs an ANFIS-based estimator trained under healthy operating conditions to generate state estimates for residual-based fault detection. Autocorrelation-enhanced residuals are then used to improve sensitivity to slowly developing faults. After fault detection, a residual-driven reference reconfiguration law updates the control objective of a constrained sliding-mode controller, thereby reducing undesirable feedback compensation while preserving bounded closed-loop behavior under the stated assumptions. The approach is evaluated on the DTS200 three-tank benchmark under simultaneous leakage and outlet-jamming faults. The simulation results indicate earlier detection and lower tracking-error indices than a detection-only baseline, while maintaining bounded operation after reconfiguration.