优化数据生成算法的往复压缩机故障诊断研究  

Compressor Fault Diagnosis Method Based on Optimized Data Generation Algorithm

作　　者：王鹏[1] 李颖 王金东[2] 巴鹏[1] WANG Peng;LI Ying;WANG Jindong;BA Peng(School of Mechanical Engineering,Shenyang Ligong University,Shenyang 110159,China;School of Mechanical Science and Engineering,Northeast Petroleum University,Daqing 163318,Heilongjiang,China)

机构地区：[1]沈阳理工大学机械工程学院,沈阳110159 [2]东北石油大学机械科学与工程学院,黑龙江大庆163318

出　　处：《噪声与振动控制》2025年第1期133-138,共6页Noise and Vibration Control

基　　金：国家自然科学基金资助项目(51934002);辽宁省教育厅科学研究经费资助项目(青年科技人才“育苗”项目LG202031);沈阳理工大学引进高层次人才科研支持计划资助项目(101014700081)。

摘　　要：为解决实际生产中存在的往复压缩机故障信息样本缺失、样本不平衡等引起的长尾分布所造成故障诊断不准确的问题,提出一种基于经优化的对抗生成网络(Generative Adversarial Network,GAN)的方法,该方法在既保证样本质量又增强样本数量的情况下,应用改进的卷积神经网络(Convolutional Neural Networks,CNN)进行故障诊断分类。首先对往复压缩机一维故障数据进行整理并通过小波变换生成故障时频图;然后构建适应于样本的LS-SAGAN框架模型并利用原始故障时频图训练模型,生成满足实验数量的时频图;最后通过经天鹰算法优化CNN进行快速准确的故障诊断。将实验方法与其他方法进行效果对比验证,结果表明,所提方法在故障诊断中的平均准确率达到99.6%,相较其他分类方法分类效果明显提高。To solve the problem of inaccurate fault diagnosis caused by the long tail distribution of missing and imbalanced samples in the fault information of reciprocating compressors in actual production,a diagnosis method based on optimized generative adversarial network(GAN)was proposed.In this method,under the promise of ensuring sample quality and enhancing sample size,the improved convolutional neural network(CNN)was applied for fault diagnosis classification.Firstly,the one-dimensional fault data of reciprocating compressors was organized and the fault time-frequency maps were generated through wavelet transform.Then,the LS-SAGAN framework model suitable for the samples was built and trained by using the original fault time-frequency maps to generated the time-frequency maps that meet the experimental quantity requirement.Finally,the Tianying algorithm was used to the optimized CNN for fast and accurate fault diagnosis.The effectiveness of this method was verified by comparing the results of this method with those of other methods.The experimental results indicate that the average accuracy of the proposed method in the fault diagnosis process reaches 99.6%,which significantly improves the classification performance compared to other classification methods.

关 键 词：故障诊断 往复压缩机 LS-SAGAN 卷积神经网络 

分 类 号：TH457[机械工程—机械制造及自动化] TP206.3[自动化与计算机技术—检测技术与自动化装置]