Abstract: Aimed at the problem that the vibration sensor is difficult to be installed in engineering application and the feature vector is redundancy, a method of fault diagnosis for rolling bearing based on MFCC and CDET is proposed. The noise signal is used for monitoring the condition of rolling bearing when the machine is running. MFCC features are extracted from the noise signal and CDET is employed to reduce the dimensionality of MFCC features. Finally, the features after feature reduction are used as the inputs of SVM classifier for fault classification and its performance of feature reduction is compared with PCA. The experimental result shows that CDET has better performance of feature reduction in noise diagnosis and the method based on MFCC and CDET can detect the fault category accurately and effectively.

Key words:  Bearing, fault diagnosis; MFCC, CDET, noise signal

摘要： 针对工程应用中振动传感器安装困难、故障特征向量存在冗余等应用问题，提出了一种基于噪声信号美尔倒谱(MFCC)与补偿距离评估(CDET)的滚动轴承故障诊断方法。将机器运行噪声信号作为轴承状态监测信号，提取机器运行噪声信号的MFCC作为诊断特征，采用CDET算法对所提取的MFCC特征进行降维，最后将CDET降维后的MFCC特征向量作为支持向量机(SVM)的输入进行模式分类，并与传统基于PCA的降维算法进行比较研究。实验结果表明：噪声诊断中CDET降维具有更优的降维效果，基于MFCC与CDET的滚动轴承故障诊断能够准确、有效地识别轴承故障类型。

关键词: 轴承；故障诊断；美尔倒谱系数；补偿距离评估；噪声信号