Identifying DC Motor Transfer Function with Few-Shots Learning and a Genetic Algorithm Using Proposed Signal-Signature

The most common actuators in precision or low-power applications are Direct Current (DC) motors. DC motors are in robots, commercial products, automobiles, and appliances for accurately controlling position or speed or in portable devices. Input tracking and system stabilization are the main goals of control theory. There are different approaches for classic control with their proprieties. However, classical controllers and sometimes others with probabilistic methods use mathematical models for designing or generating control laws. Alternatives that represent the model of a system include differential and difference equations, state space models, and transfer functions. Parametrizing those models implies supplying signals to the process and using deterministic or probabilistic algorithms. On the other hand, artificial intelligence and machine learning approaches like genetic algorithms have shown relevant results by tuning those models using the error with the system’s frequency or time response. However, looking for reasonable solutions in search spaces with several dimensions, like in a five-dimensional problem optimizing the parameters of a DC motor, can be timely and computationally demanding. Alternatively, few-shots learning simplifies the data for optimization, resulting in less timely and computational training. In this work, we propose a novel method for transfer function DC motor identification using a few-shots learning approach based on a signature value unique for each motor. We obtained the motor signature by applying a signal that variates length and frequency for each engine and collecting the six numerical values with the exact time sampling. Since we reduce the components of the signals, we reduce the data size. Moreover, the search space becomes significantly simple, allowing us to find the DC motor transfer functions with R-square between 0.99 and 1.0 for at least 95% of the controllers with 6 component signatures.