基于多尺度特征提取的风机音频信号故障诊断方法研究

Fault diagnosis method of fan audio signal based on multi-scale feature extraction

针对风机传动链部件音频信号成分复杂,依靠单一特征提取方法难以实现故障识别的问题,提出了一种基于多尺度特征提取的风机主轴音频信号故障诊断方法。首先,采集风机传动链部件的音频文件,将其转换为数字信号后对数据进行了预处理;然后,对音频信号进行了多尺度特征提取,从时域、频域和倒谱域3个维度提取了5大特征,组成了多维复合特征矩阵,并对提取的特征进行了分析及降维;最后,利用支持向量机(SVM)分类预估器对多维复合特征矩阵进行了有监督学习,并使用粒子群算法(PSO)对SVM参数选择过程进行了优化,并通过多组对比实验,对PSO-SVM分类预估器在音频信号模式分类上的性能进行了验证。研究结果表明:所提取的多尺度特征能很好地表征音频信号的信息,具备一定的鲁棒性;使用经PSO优化后的SVM分类模型,对风机传动链部件音频信号模式识别的准确率可达98%以上,并具备良好的泛化能力。

Aiming at the problem that the audio signal of fan drive chain components was complicated and fault identification was difficult to realize with a single feature extraction method,a fault diagnosis method for audio signals based on multi-scale feature extraction was proposed.Firstly,the collected audio files of fan drive chain components were used to convert to digital signals and preprocess the data.Secondly,multi-scale feature extraction was performed on the audio signal.Five major features were extracted from the three dimensions of time domain,frequency domain and cepstral domain to form a multi-dimensional composite feature matrix.Then,the features were analyzed and the dimension of the feature matrix was reduced.Finally,the support vector machine(SVM)classification predictor was used to perform supervised learning on the multi-dimensional composite feature matrix,and the particle swarm optimization(PSO)was used to optimize the SVM parameter selection process.Through multiple sets of comparative experiments,the performance of PSO-SVM classification predictor on audio signal pattern classification was verified.The experimental results indicate that the extracted multi-scale features can well represent the information of the audio signal,and have certain robustness.The SVM classification model optimized by PSO eliminates the blindness of parameter selection,which can achieve more than 98%accuracy rate of audio signal pattern recognition for fan drive chain components,and has good generalization.