Abstract
This study presents an Artificial Neural Network (ANN)-based adaptive switching frequency control strategy for series Direct current (DC) motor drives used in battery-powered mining locomotives, aiming to extend the lifetime of critical power-electronic components such as Insulated Gate Bipolar Transistors (IGBTs) and DC bus capacitors. In embedded systems for electric traction, two dominant degradation factors, motor current ripple and IGBT temperature fluctuation, significantly shorten component lifetimes. Conventional fixed switching frequencies impose a trade off: higher frequencies reduce current ripple but increase IGBT losses and temperature, while lower frequencies yield the opposite effect. Consequently, an adaptive variable switching frequency control algorithm is proposed to perform real-time decision making by predicting the optimal switching frequency that minimizes both motor current ripple and IGBT thermal fluctuations. The proposed algorithm was trained with a dataset acquired from current sensors, NTC temperature sensors, and a potentiometer defining the target current (PWM duty). Performance comparisons with a fixed frequency demonstrate that the ANN-driven approach maintains an average current ripple of less than 5% (average) and 10% (maximum), while the lifetime of the IGBT and capacitors improves. A fairness index was defined to quantify the relative lifetime improvement of the IGBT and capacitor, revealing that the proposed variable frequency switching model enhances the overall system performance by up to 13 times compared to fixed-frequency operation. These results confirm that the integration of embedded machine learning and adaptive control algorithms can substantially enhance the durability and efficiency of power-electronic systems in real-time industrial applications.