|
چکیده
|
Background and Problem Statement Proportional–Integral–Derivative (PID) controllers are widely used in industrial control systems due to their simple structure and ease of implementation. However, in nonlinear systems with time-varying parameters, fixed tuning of PID gains often leads to performance degradation, increased settling time, and undesirable phenomena such as overshoot. In recent years, artificial neural networks have attracted considerable attention as an effective approach for adaptive tuning of PID parameters. On the other hand, implementing such algorithms on microcontrollers faces serious challenges due to limited computational resources and real-time constraints. In this context, Field-Programmable Gate Arrays (FPGAs) have emerged as a suitable platform for the real-time implementation of intelligent controllers. Research Objective The objective of this research is to design and simulate an adaptive PID controller based on a multilayer perceptron neural network (MLP-PID) with a VHDL-aware approach, and to evaluate its performance in motion control systems, while simultaneously ensuring efficient, reliable, and practical implementation on FPGA hardware. Research Methodology In this study, a three-layer feedforward neural network with a 5–8–3 architecture is designed for online tuning of the PID gains KpK_pKp, KiK_iKi, and KdK_dKd. The network inputs include the current control error, past errors, error variations, and system output. To facilitate hardware implementation, a VHDL-compatible piecewise hyperbolic tangent activation function is employed in the hidden layer. Moreover, to guarantee initial system stability, an intelligent initialization method for the output layer biases is adopted based on stable initial PID gains. The online learning algorithm is developed using the backpropagation method with gradient descent, and the controller is implemented in an incremental PID structure. Results Simulation results obtained in the MATLAB/Simulink environment demonstrate that the proposed controller provides a fast and stable step response without overshoot. Furthermore, the PID gains converge smoothly and remain stable, and the overall system performance under nonlinear conditio improvement compared to methods reported in the existing literature. Conclusion Based on the obtained results, integrating a multilayer neural network with a PID controller while incorporating VHDLa ware considerations at the design provides an effective solution for developing intelligent controllers with real stage time capability and efficient implementation on FPGA platforms. The proposed approach enhances control performance while ensuring hardware feasibility and reliability. Th e hardware implementation of the proposed controller on FPGA and its experimental evaluation are currently in progress, and the corresponding practical results will be reported in future extensions of this research.
|