基于PSO-ChOA优化的轴流风机故障诊断模型  

Axial flow fan fault diagnosis model based on PSO-ChOA optimization

作　　者：吕亚楠 赵康 马草原[3] 郑璐[3] LV Yanan;ZHAO Kang;MA Caoyuan;ZHENG Lu(School of Aeronautical Engineering,Jiangsu Aviation Technical College,Zhenjiang 212134,China;Zhenjiang Branch of State Grid Jiangsu Electric Power Co.,Ltd.,Zhenjiang 212134,China;School of Electrical Engineering,China University of Mining and Technology,Xuzhou 221116,China)

机构地区：[1]江苏航空职业技术学院航空工程学院,江苏镇江212134 [2]国网江苏省电力有限公司镇江分公司,江苏镇江212134 [3]中国矿业大学电气工程学院,江苏徐州221116

出　　处：《机电工程》2025年第2期373-386,共14页Journal of Mechanical & Electrical Engineering

基　　金：江苏航空职业技术学院院级课题项目(JATC22010105)。

摘　　要：传统的风机故障诊断技术依赖大量的历史数据,在参数优化和算法选择上存在早熟收敛问题,且在风机故障诊断过程中需要精确采集信号,但实际应用中受限于传感器安装条件,影响了数据的准确性和诊断的有效性。针对这些问题,提出了一种融合改进粒子群优化算法(PSO)与黑猩猩优化算法(ChOA)混合优化策略(PSO-ChOA)的VMD-CNN-Transformer模型,应用于轴流风机故障诊断。首先,通过仿真和实验获取了七种风机典型电气故障信号和三种离心风机轴承故障信号,并进行了预处理以满足算法训练要求;然后,使用PSO对ChOA的狩猎搜索阶段进行了优化,减少了人为设定参数对模型训练的影响,通过构建23个标准测试函数,分析了PSO-ChOA算法在收敛速度和全局优化上的优势;最后,利用变分模态分解(VMD)提取了故障特征,并利用卷积神经网络-Transformer(CNN-Transformer)模型进行了分类,采用实例分析了该模型在处理非线性和高维数据时的强大能力。研究结果表明:相较于传统算法,PSO-ChOA算法在收敛速度上的优势显著,能够更快地跳出局部最优,避免早熟收敛,同时保持较高的搜索精度,最终找到更接近全局最优的解;采用PSO-ChOA优化的VMD-CNN-Transformer模型在风机故障诊断任务中达到了97.76%的准确率,较VMD-CNN-Transformer方法,准确率提升了6.64%;PSO-ChOA在参数优化领域的应用潜力,为工业设备故障诊断研究提供了新的视角。The traditional fan fault diagnosis technology relies on a large amount of historical data,precocious convergence in parameter optimization and algorithm selection,and accurate signal acquisition is required in the process of fan fault diagnosis.However,the actual application is limited by the installation conditions of sensors,affecting the accuracy of data and the effectiveness of diagnosis.To solve these problems,a VMD-CNN-Transformer model was proposed,which combines the hybrid optimization strategy(PSO-ChOA) of improved particle swarm optimization(PSO) and chimpanzee optimization(ChOA),so as to be applied to the fault diagnosis of axial flow fans.Firstly,seven typical fan electrical fault signals and three centrifugal fan bearing fault signals were obtained by simulation and experiment,and pre-processing was carried out to meet the training requirements of the algorithm.Then,PSO was used to optimize the hunting and searching phase of ChOA to reduce the influence of artificial parameters on model training.By constructing 23 standard test functions,the advantages of PSO-ChOA algorithm in convergence speed and global optimization were analyzed.Finally,the fault characteristics were extracted by variational mode decomposition(VMD) and classified by convolutional neural network Transformer(CNN-Transformer) model.The powerful ability of the model in processing nonlinear and high-dimensional data was analyzed by examples.The research results show that comparing with the traditional algorithm,PSO-ChOA algorithm has obvious advantages in convergence speed,can jump out of the local optimal faster,avoid premature convergence,while maintaining high search accuracy,and finally find the solution closer to the global optimal.The VMD-CNN-Transformer model optimized by PSO-ChOA achieves 97.76% accuracy in fan fault diagnosis,and it is 6.64% higher than the VMD-CNN-Transformer method.The application potential of PSO-ChOA in the field of parameter optimization provides a new perspective for the research of industrial equipment

关 键 词：离心式风机 复杂非线性信号 粒子群优化 黑猩猩优化算法 卷积神经网络-Transformer模型 变分模态分解 

分 类 号：TH432.1[机械工程—机械制造及自动化]