摘要
为实现汽车空压机异响的快速诊断,文章对空压机异响机理进行全面分析,针对两大异响特征拍频声和静盘异常声进行详细阐述,并描述了可能发生的异响类型;考虑到空压机运转音频较少的现状,通过在半消声室搭建空压机运转台架,使用数据采集系统进行了音频信号采集,并基于已有的MIMII和IDMT数据集进行筛选对数据量进行扩充。在自编码器模型的基础上根据输入数据的维度特征,通过卷积、正则化以及改变卷积核数量对原有模型进行调整,并结合高斯混合算法,形成高斯混合自编码器模型,使用收集的空压机音频数据集进行训练,其检验的准确性均优于其他基线模型,异响检测的准确率达到93%,检验结果可靠。
In order to realize the rapid diagnosis of abnormal noise of automobile air compressor,the paper comprehen-sively analyzes the mechanism of abnormal noise of air compressor,and elaborates the two characteristics of abnormal noise,beat frequency and static disk abnormal sound,and describes the types of abnormal noise that may occur.Consider-ing the current situation that the air compressor has less running audio,the air compressor running bench is built in the semi-anechoic chamber,the audio signal is collected by using the data acquisition system,and the data volume is expanded based on the existing MIMII and IDMT data sets.Based on the autoencoder model,the original model is adjusted by con-volution,regularization and changing the number of convolutional cores according to the dimensional characteristics of the input data.Combined with the Gaussian mixture algorithm,a Gaussian mixture autoencoder model is formed,and the col-lected air compressor audio data set is used for training.The accuracy of its test is higher than other baseline models.The accuracy of abnormal sound detection reached 93%,and the test results were reliable.
作者
刘志恩
陈磊
陈弯
LIU Zhi-en;CHEN Lei;CHEN Wan(School of Automotive Engineering,Wuhan University of Technology,Wuhan 430o70,China;Hubei Key Laboratory of Advanced Technology for Automotive Components,Wuhan 430070,China)
出处
《武汉理工大学学报》
2025年第7期79-85,共7页
Journal of Wuhan University of Technology
基金
国家自然科学基金(52175111)。
关键词
汽车NVH
无监督学习
自编码器
高斯混合算法
异响检测
automotive NVH
unsupervised learning
autoencoder
gaussian mixture algorithm
abnormal noisedetection
