聚类分析作为一种先进的无监督模式识别技术, 能在无先验知识的情况下对数据进行分类并揭示数据内部结构。通过优化K均值聚类算法, 并应用浮动门槛值重新计算声发射检测波形数据的常用特征参量作为聚类算法的输入向量, 采用浮动门槛计算得到的特征向量更能反映声发射波形特征, 取得良好的聚类效果。通过现场储罐声发射检测数据实例的聚类分析, 结果表明K均值聚类能有效地区别不同声源和传播途径的声发射信号, 具有很好的去噪效果, 能有效提高罐底声发射检测评价准确度。

Cluster analysis is an advanced unsupervised pattern recognition technology, which can classify data and reveal the data internal structure without prior knowledge. In this paper, the application of floating threshold value to calculate the characteristic parameters of acoustic emission test common waveform data is used as input vector clustering algorithm by optimizing K means clustering calculation. The featuring vector by floating threshold calculation has more acoustic emission waveform characters, obtaining good clustering effect. By analysis of acoustic emission test data of storage tank, the results show that K means clustering can determine different sound signals from different sound sources and propagation paths, affording excellent de-noise effect and effectively improving acoustic emission test accuracy of tank bottom.