A hybrid neuro-inverse control approach with knowledge-based nonlinear separation for industrial nonlinear system

This brief presents a general methodology of controller design by the hybrid neuro-inverse control with knowledge-based nonlinear separation for industrial nonlinear systems. In industrial nonlinear systems, various kinds of uncertainties may cause serious deterioration of system performances. Unfortunately, these uncertainties are usually difficult to identify and compensate from the entire system point of view. With using the knowledge-based nonlinear separation, nonlinear dynamics of a nonlinear system is possibly separated into the input-output nonlinear static part and the nonlinear dynamic part to form a nonlinear separation structure. Hence, uncertainties in the nonlinear system are bounded in the nonlinear dynamic part. In the proposed hybrid neuro-inverse control method, the input-output nonlinear static part is controlled by an inverse controller. A neurocontroller with a rigidly defined and trained neural network using available prior knowledge of the nonlinear system is constructed for the control of the nonlinear dynamic part. Owing to using the knowledge-based nonlinear separation, the neurocontroller is only needed to control a part of the original nonlinear dynamics of industrial nonlinear systems contaminated by uncertainties. The structure of the neural network employed by the neurocontroller becomes simpler and the consumption of time in training is reduced. Meanwhile, system performances of the nonlinear system can be improved. In this brief, with two typical control problems in industries, i.e., high-precision contour control of industrial articulated robot arm (IARA) and outlet working fluid heat rate control of evaporator in energy conversion plant, the proposed method was explained clearly. The experimental and simulation results demonstrated the generality, practicality and significant potential of this method for realizing the high-performance control of industrial nonlinear systems.