基于学习矢量量化的空调压缩机声纹诊断方法  

Voiceprint diagnosis method of air conditioning compressor based on learning vector quantization

作　　者：欧阳城添[1] 袁瑾 OUYANG Cheng-tian;YUAN Jin(School of Information Engineering,Jiangxi University of Science and Technology,Ganzhou 341000,China)

机构地区：[1]江西理工大学信息工程学院,江西赣州341000

出　　处：《计算机工程与设计》2021年第9期2634-2641,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61561024、61462034、61563019);江西省自然科学基金项目(20151BAB207035);江西省研究生创新专项基金项目(YC2018-S332)。

摘　　要：针对传统空调压缩机故障诊断工况信号采集困难的缺点,提出一种基于学习矢量量化(learning vector quantization,LVQ)神经网络的空调压缩机声纹识别模型用于空调压缩机故障诊断,将声纹识别技术引入压缩机故障诊断。对压缩机的声音数据进行预处理,包括预加重、分帧、加窗,在分帧步骤中针对压缩机的声音特性进行改进,通过计算声音信号的梅尔倒谱系数(Mel frequency cestrum coefficient,MFCC)得到压缩机声音的特征向量。在模型训练阶段,重点分析原始的LVQ算法和改进的LVQ算法的优缺点,对3种LVQ算法进行对比实验。实验结果表明,使用LVQ3算法学习的压缩机声纹识别模型在测试集上可以达到90%的召回率,研究结果为压缩机故障诊断提供了一种依据。To overcome the shortcomings of difficult signal acquisition of traditional air conditioning compressor fault diagnosis operating conditions,a voiceprint recognition model of air conditioning compressor based on learning vector quantization(LVQ)neural network was proposed for air-conditioning compressor fault diagnosis.Voiceprint recognition technology was introduced into compressor fault diagnosis.The sound data of the compressor were pre-processed,including pre-emphasis,framing,and windowing,and the sound characteristics of the compressor were improved in the framing step,and the Mel cestrum coefficient of the sound signal was calculated by Mel frequency cestrum coefficient(MFCC)to get the feature vector of compressor sound.In the model training stage,the advantages and disadvantages of the original LVQ algorithm and the improved LVQ algorithm were analyzed,and three LVQ algorithms were compared and tested.The results verify that the compressor voiceprint recognition model learned using the LVQ3 algorithm can reach a recall rate of 90%on the test set.Therefore,the research results provide a basis for compressor fault diagnosis.

关 键 词：空调压缩机 故障诊断 声纹识别 学习矢量量化 梅尔倒谱系数 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]