Overview

This project implements a robust Adaptive Neuro-Fuzzy Inference System (ANFIS) to control magnetic levitation systems. MagLev systems are inherently unstable and nonlinear, making them difficult to control with traditional linear PID methods over a wide operating range.

System Features

• Adaptive Learning: The controller uses a hybrid learning algorithm (Backpropagation + Least Squares) to tune fuzzy membership functions based on input-output data.
• Fuzzy Logic: Operates on a Sugeno-type inference system with 14 fuzzy rules, providing smooth transitions between control states.
• Inputs:
  • Error: Position deviation from the setpoint.
  • Error Rate: The velocity of the error change.

Methodology

1. Data Collection: Training data was gathered from a PID-controlled simulation to capture the system’s fundamental dynamics.
2. Preprocessing: Data was cleaned, normalized, and partitioned to ensure robust training.
3. Training: The ANFIS structure was trained for 50 epochs to minimize the RMSE between the controller output and the target control signal.