FPGA Based Neural Network PID Controller for Line-Scan Camera in Sensorless Environment

This paper presents a neural network PID controller that enables the usage of the line-scan camera in a sensorless environment. The controller uses the BP neural network which has a self-training ability to adjust the parameter of the PID control loop. For the complexity of the neural network when implemented on FPGA, a linear approximation of the activation function is proposed and the maximum use of the sharing resources which is an effective way to save resource area is also discussed. The complete system performance is investigated by the simulation on MATLAB and ModelSim and validated experimentally on a print product line. The results indicate the success of the controller's design.