New methodology for identification and control of plants with static input or output nonlinearities

In this study, it is assumed that a plant consists of input or output static nonlinearities and a linear process. A novel hybrid identification method is developed which consists of a feedforward multilayer neural network (MFNN) and an auto-regressive moving average (ARMA) linear model. The MFNN is used to identify the static nonlinearities and the ARMA model captures the dynamics of the process. A new recursive algorithm which estimates the weights of the MFNN and the parameters of the ARMA model in a combined procedure is developed. Closed loop control is achieved by inserting the inverse of the identified MFNN in the appropriate loop location and designing a linear controller for the linear part of the plant. Invertibility of the MFNN is addressed and the robustness of the method is guaranteed by exhaustive simulation studies. Two case studies for a heat exchanger and a fluid flow process are reported here.