Model-Free Neural Fault-Tolerant Control for Discrete-Time Unknown Nonlinear Systems

This paper presents the design of a new fault-tolerant control methodology based on a model-free neural technique. For an unknown nonlinear dynamical system, a recurrent neural network is proposed to design an on-line identifier to develop an adaptive mathematical model that captures the dynamics of the system using only available measurements of its state variables. Employing this neural model, the use of a discrete sliding mode controller is proposed, which is modified to adequately handle the presence of faults in the sensors and actuators, as well as disturbances and uncertainties that occur during the operation of the system. The effectiveness of the proposed control scheme is shown by means of simulation results for a three-phase induction motor. Most of the fault-tolerant controllers start from a nominal model of the system to be controlled, from that neural model methodologies are established that evaluate the behavior of the system with respect to a nominal model, originating a residue between such signals. This residue is analyzed using traditional techniques or artificial intelligence, for which the success of this type of methodology lies in how well the behavior of the system is captured in its mathematical model. Therefore, the main contribution of this methodology is that the proposed controller is model-free, the detection of the fault and the corresponding modification to the control is carried out directly from the measured data, nor does it require exact knowledge of the faults, their duration, or the time they occur, everything is computed implicitly through the neural network.