基于分块特征提取的近场声全息故障诊断方法在相干故障下的应用

The application of NAH-based fault diagnosis method based on blocking feature extraction in coherent fault conditions

通过引入近场声全息和分块特征提取技术,改进了基于声像的故障诊断方法,发展了基于近场声全息模式识别的故障诊断技术。针对多个机械部件对应相同故障频率,并产生相干声场的故障情形进行了加肋板激振的模拟实验,使用传声器阵列扫描技术测取各种状态下声信号,在利用近场声全息技术得到声像进行噪声源识别与定位的基础上,对声像进行整体和分块相结合的奇异值特征,提取方式构造识别向量,然后采用多分类支持向量机进行训练分类,进而用于机械工作状态的诊断。实验结果表明,根据声像的物理特征使用整体和分块相结合的特征提取技巧能够较好改善诊断效果,同时进一步验证了声成像方法在故障诊断领域应用的可行性,并与常规的基于单点或几个孤立测点测试的声学故障诊断方法相比具有优越性,拓展了声学故障诊断技术的应用范围。

To further study the fault diagnosis method based on acoustic images, one near-field acoustical holography （NAH）-based fault diagnosis method is developed which firstly introduce the NAH technology and blocking feature extraction method into fault diagnosis. In allusion to the coherent fault conditions in which different machine components correspond to one and the same feature frequency and the coherent sound field is generated, one rib plate multi-excitation experiment is implement- ed. The scanning measurement technique is employed to sample the sound signals, and then the NAH algorithm is utilized to reconstruct the sound pressure distribution of sound sources for source recognition. Considering the physical meaning of the acoustic images, blocking feature extraction technique is applied to compose the eigenvectors. At last, the multiclass-support vector machine （SVM） is employed to train the feature vectors and diagnose the machine conditions. The experiments in labo- ratory demonstrate that the new diagnosis technology based on NAH technology is feasible and is one appropriate method comparing to acoustic-based diagnosis technique based on isolated point test in coherent fault conditions, and simultaneously widens the applications of NAH technique in the area of fault diagnosis.