Hybrid GA-ANFIS and PSO-ANFIS techniques for nonlinear DC motor system modeling

This research focuses on nonlinear modeling techniques for direct current (DC) motor, with permanent magnets in the stator, using optimized adaptive neuro-fuzzy inference systems (ANFIS). The traditional linear model fails to accurately represent the dynamics of a DC motor due to nonlinear friction effects. To address this limitation, a nonlinear model incorporating Tustin’s friction model is proposed and evaluated against experimental data. Despite improvements over the linear model, challenges remain due to the discontinuity introduced by the signum function in friction representation, necessitating smoother approximations like hyperbolic tangent for control applications. The nonlinear modeling approach also does not fully capture the dynamics of the real-world behavior of the object. To achieve a robust and accurate model across all operational conditions without approximations, three ANFIS variants are developed. These models employ diverse approaches to generate fuzzy rules, such as grid partitioning and fuzzy C-Means clustering. The second and third models undergo optimization using two different nature inspired optimization algorithms. Comparative analysis reveals that all ANFIS models yield superior performance, with GA-ANFIS on top, accurately predicting DC motor velocity under varying input conditions such as step, sinusoidal, and chirp signals. Experimental validation demonstrates that the optimized ANFIS model closely tracks the real-world behavior of the DC motor, offering promising prospects in every type of control, especially in direct inverse control in which the system model is inverted and used as a controller. This approach enables precise control actions based solely on observed system dynamics, avoiding the pitfalls of approximation-based methods.