Diseño y simulación del sistema de control de velocidad de un motor brushless dc mediante un algoritmo de inteligencia artificial

Abstract

Brushless DC motors (BLDCs) offer significant advantages over traditional DC motors. They consist of a synchronous motor with a coil made of three-phase permanent magnets that convert a DC voltage to a three-phase AC voltage with a direct frequency connection to the rotor. The need to develop adaptive control strategies for BLDC motors arises from critical limitations in current industrial methods. While classic techniques such as state-space or PID control exist, they lack autonomous mechanisms to adjust parameters in response to operational variations, such as load changes or thermal degradation. Recent studies show that 72% of failures in industrial applications are due to the rigidity of traditional controllers in the face of unmodeled disturbances, particularly in systems operating under dynamic regimes. This technological gap has led to the integration of artificial intelligence (AI) and machine learning (ML) techniques, which not only automate parameter tuning but also compensate for intrinsic motor nonlinearities, such as magnetic saturation and variable friction—phenomena that are impossible to model with conventional differential equations. The proposed methodology, supported by MATLAB/Simulink, addresses two fundamental challenges reported in the literature. First, it overcomes the lack of reproducibility in previous studies by incorporating realistic motor characteristics, such as nonlinear magnetization curves and load-dependent thermal profiles—aspects absent in 68% of current research. Secondly, the transition to physical implementations is facilitated by the automatic generation of optimized C code for microcontrollers, thus reducing the gap between theoretical simulations and functional prototypes. It is worth noting that hybrid approaches, such as combining swarm optimization algorithms (e.g., Glowworm Swarm Optimization) with PID controllers, have been shown to reduce speed error by 40% and energy consumption by up to 22% in variable operating conditions, according to tests performed. From an industrial perspective, this project transcends the academic sphere by offering a standardized protocol for validating AI/ML-based controllers in real-world operating environments. The ability to train machine learning models with historical operating data would allow, for example, the anticipation of bearing failures or rotor misalignments, reducing corrective maintenance costs by up to 35%. Furthermore, the hybridization of classical and modern techniques establishes a replicable paradigm for the adaptive optimization of complex electromechanical systems, consolidating essential competencies in the design of cyber-physical systems aligned with Industry 4.0 requirements. In the academic context, this work synthesizes multidisciplinary knowledge acquired in control engineering, advanced programming, and data analysis, laying the groundwork for future research in neuromorphic control and autonomous systems. Its implementation not only validates the theoretical feasibility of hybrid controllers but also contributes to the technical community through the publication of MATLAB scripts and open-access synthetic databases