基于MFCC与LPCC的滚动轴承故障音频诊断法  

A Audio-Based Fault Diagnosis Method using MFCC and LPCC features for Rolling Bearings

作　　者：周宴宇[1] 杨胜跃[1] 何正明[1] 龙慧[1] 

机构地区：[1]中南大学信息科学与工程学院

出　　处：《微计算机信息》2009年第31期123-124,74,共3页Control & Automation

摘　　要：本文结合语音识别中运用较广的美尔频率倒谱系数(MFCC)和线形预测倒谱系数(LPCC)作为滚动轴承音频信号的特征参数,并使用具有强学习能力的神经网络方法进行故障诊断,最后依靠Dempster-Shafrg(D-S)证据理论进行分析得出可信度高的判定结果。实验结果表明,该方法在滚动轴承的音频信号故障诊断中有效可行。在积累足够量的样本数后,可望建立相应的专家库,实现轴承故障的快速智能诊断。This article unified the broad Mel frequency cepstrum coefficient （MFCC） and linear forecast cepstrum coefficient （LPCC） as the characteristic parameter of the rolling bearing audio signals,and used neural network method that had the strong learning capability to carry on the fault diagnosis. Finally depended upon the D-S evidence theory, the high confident result can be obtained. The result of experiment indicated that this method was effective and feasible in audio signal of rolling hearing＇s fault diagnosis. After accumulating enough sample data, it can hope to build the corresponding database of expert and realize the fast intelligent diagnosis.

关 键 词：轴承 故障诊断 人工神经网络 美尔频率倒谱系数 线形预测倒谱系数 音频信号 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置]