Estudio comparativo de un controlador por red neuronal artificial y un controlador por red neuronal evolutiva para un motor de corriente continua.

Abstract

This study presents an analysis of two intelligent control methods used in a DC motor: one based on artificial neural networks and the other on evolutionary neural networks. The goal was to analyze how both approaches perform in contrast to the drawbacks of conventional controllers, such as PID, in nonlinear, noisy, or changing situations. A mathematical model was created for the DC motor, using differential equations that illustrate both its electrical and mechanical components. The simulation of motor performance and the development of the controllers were carried out in Python, usingspecialized libraries such as SciPy, NumPy, Matplotlib, Scikit-learn, and DEAP. The ANN (Artificial Neural Network) regulator was developed using a multi-layered neural network trained with simulated data, applying the MLPRegressor algorithm. This method acquires the ability to create control signals that help the motor stay within a pre-established reference. In contrast, the ENN (Evolutionary Artificial Neural Network) regulator used genetic algorithm techniques to refine variables such as proportional gain and offset, increasing its tuning capacity without requiring supervised training. Both controllers were evaluated using the metrics of mean square error (MSE), response time, stability and robustness against disturbances. The results showed that the ENN obtained a better overall performance, with a lower MSE (0.0121 vs 0.0152) and a smoother response, consequently it was more stable to changes in the input signal. While both controllers proved to be effective, the ENN method stood out more for its accuracy, flexibility and consistency, this research underlines the capacity of artificial intelligence methods, especially evolutionary ones, to create more effective and robust control systems in the industrial and mechatronic fields.