[论文解读] Artificial Neural Networks Based Analysis of BLDC Motor Speed Control

T. Porselvi, Sai Ganesh Cs+1 more|arXiv (Cornell University)|Aug 27, 2021

Sensorless Control of Electric Motors参考文献 10被引用 12

一句话总结

本文提出一种基于人工神经网络（ANN）的方法，通过训练MATLAB/Simulink中PI控制的BLDC电机仿真数据，预测关键BLDC电机参数（如转子转速、电磁转矩和反电动势）。该方法能够准确预测电机的动态行为，表明ANN模型可高保真地复现仿真结果。

ABSTRACT

Artificial Neural Network (ANN) is a simple network that has an input, an output, and numerous hidden layers with a set of nodes. Implementation of ANN algorithms in electrical, and electronics engineering always satisfies with the expected results as ANN handles binary data more accurately. Brushless Direct Current motor (BLDC motor) uses electronic closed-loop controllers to switch DC current to the motor windings and produces the magnetic fields. The BLDC motor finds various applications owing to its high speed, low maintenance and adequate torque capability. They are highly preferred than the other motors because of their better performance and it is easy to control their speed by Power Converters. This article presents a method of speed control of BLDC motors where speed is controlled by changing the DC input voltage of the bridge converter that feeds the motor winding. The control is done by using a PI based speed controller. The motor is modeled in the MATLAB/Simulink and the speed control is obtained with a PI controller. EMF signals, rotor speed, electromagnetic torque, Hall Effect signals, PWM and EMF signals simulations are then obtained. This acquired data is then fed into binary artificial neural networks and as a result, the ANN model predicts the corresponding parameters close to the simulation results. Both the mathematical based simulation and data based prediction gives satisfactory results

研究动机与目标

• 开发一种数据驱动的ANN模型，用于预测PI速度控制下BLDC电机的动态参数。
• 评估人工神经网络在复现MATLAB/Simulink中BLDC电机系统仿真结果方面的有效性。
• 对比数学仿真结果与基于ANN的预测结果，涵盖转子转速、反电动势、转矩和霍尔信号。
• 验证使用二进制ANN在闭环控制系统中实现实时电机行为估计的可行性。

提出的方法

• 在MATLAB/Simulink中建立基于PI速度控制器的BLDC电机模型，生成参考仿真数据。
• 仿真关键电机信号：转子转速、反电动势、电磁转矩、霍尔效应信号及PWM波形。
• 从仿真数据中提取并预处理为二进制输入-输出对，用于ANN训练。
• 使用仿真数据训练前馈二进制人工神经网络，以预测电机参数。
• 将ANN预测结果与原始Simulink仿真输出进行验证。
• 采用多层感知器架构，通过隐藏层捕捉电机动力学中的非线性关系。

实验结果

研究问题

• RQ1 人工神经网络能否准确预测由PI控制器控制的BLDC电机的动态行为？
• RQ2 ANN模型的输出与同一系统在MATLAB/Simulink中的仿真结果有多接近？
• RQ3 ANN对转子转速、反电动势和电磁转矩等关键参数的预测精度如何？
• RQ4 数据驱动的ANN模型在多大程度上可替代或补充传统基于仿真的BLDC电机系统控制分析？

主要发现

• ANN模型成功以高精度预测了转子转速、反电动势、电磁转矩及霍尔效应信号，与Simulink仿真数据高度一致。
• ANN输出与仿真结果之间的预测误差极小，表明训练网络具有强大的泛化能力。
• 二进制ANN架构有效处理了电机控制信号的离散特性，包括PWM和霍尔效应数据。
• 数学仿真与基于数据的ANN预测相结合，在所有测试参数上均得到一致且令人满意的结果。
• 本研究证实，ANN可作为BLDC电机动力学在控制系统设计与分析中的可靠代理模型。