基于小波分析的三角钢琴踏板机械故障检测方法  

Method of mechanical fault detection based on wavelet analysis

作　　者：陈莎莎 CHEN Shasha(Shaanxi Railway Institute,Weinan Shaanxi 714099,China)

机构地区：[1]陕西铁路工程职业技术学院,陕西渭南714099

出　　处：《自动化与仪器仪表》2022年第12期25-30,共6页Automation & Instrumentation

基　　金：新时代高校美育背景下的原创经典文化品牌培育推广机制研究(2020FKT67)。

摘　　要：针对采集的三角钢琴踏板机械信号故障检测准确率低的问题,提出基于小波分析的三角钢琴踏板机械信号故障检测方法。采用频谱感知算法构建一个三角钢琴音频信号的多源信息采集模型,获取三角钢琴踏板机械音频信号;为提高钢琴踏板机械信号获取的准确率,通过频谱特征分离方法对采集信号进行降噪处理,采用小波包变换中的Mel倒谱系数MFCC对处理后的信号进行频谱特征分解;最后基于分解频谱特征进行三角钢琴音频信号的特征识别优化。实验表明,在相同的数据量下,对比于12维的MFCC频谱特征分解方法,提出的小波包分解的MFCC方法可分解出更多的钢琴频谱特征峰值,为后续三角钢琴踏板机械信号故障检测提供了有效的数据。实验发现,对比于传统卷积神经网络的音频特征识别方法,本方法的识别精度最高可达95.6%,且均保持在90及以上。综合分析可知,本方法可实现三角钢琴踏板机械信号故障准确识别。The new method of fault detection method of the mechanical signal based on wavelet analysis is proposed. A multi-source information acquisition model for the grand piano audio signal is used to improve the accuracy of the acquisition of the piano pedal mechanical signal, and the Mel inversion coefficient MFCC analyzes the processed audio signal based on the decomposed spectrum features. Experiments show that, compared with the 12-dimensional MFCC spectrum feature decomposition method, the proposed MFCC method of wavelet packet decomposition can decompose more piano spectral feature peaks under the same amount of data, which provides effective data for the subsequent fault detection of mechanical signals of the grand piano pedal. It is found that compared with the traditional convolutional neural network audio feature recognition method, the recognition accuracy of this method can reach 95.6%, and all remain at 90 and above. Comprehensive analysis shows that this method can accurately identify the mechanical signal fault of the grand piano pedal.

关 键 词：小波分析 钢琴踏板 机械故障检测 音频信号 特征分解 

分 类 号：TP392[自动化与计算机技术—计算机应用技术]