Neural network based adaptive pid controller for shell-and-tube heat exchanger

This research presents the design and simulation of nonlinear adaptive control system on the heating process of shell-and-tube heat exchanger model BDT921. Shell-and-tube heat exchanger is a nonlinear process due to the factor of friction, temperature dependent properties, and unmeasured disturbance. As the heating process is nonlinear in nature and conventional PID is a linear controller, change in process dynamics cause instability of the controller parameters i.e proportional gain, integral time and derivative time. Thus, these controller parameters need to be repeatedly retuned. In this circumstance, auto tuning of the controller parameters is incredibly important. In this study, neural network approach was introduced to auto-tune the controller parameters. The dynamic data from the BDT921 plant was collected to formulate the mathematical model of the process using MATLAB System Identification Toolbox. The dynamic behavior of the process is accurately modeled using nonlinear ARX model with 96.17% of validation accuracy and 97.5% of fit to estimation accuracy. Dynamic time series neural network model was used together with Levenberg-Marquardt algorithm as the training method. Single hidden layer feed forward neural networks with 20 neurons in hidden layer was selected. The neural network model consists of 4 input variables and 4 output variables. Simulation and development of the controller was done in the Simulink environment meanwhile the effectiveness of the controller was evaluated based on the set point tracking and disturbance rejection. Simulation result proved that the adaptive PID controller was more effective in tracking the set point with faster settling time and lower or no overshoot respond compared to conventional PID controller. However, there is no significant improvement in controller performance when disturbance is introduced to the controller.