基于独立分量分析特征提取的故障诊断系统

Fault diagnosis system based on ICA feature

针对矿山破碎机的声音故障诊断受复杂现场环境制约、确诊率低的难题,结合独立分量分析(ICA)在自然图像和连续语音信号中特征提取的方法,采用两层ICA分别用于从混杂声音中提取各采集通道(部位)的统计独立声音信号和进一步提取该信号的特征基.训练阶段生成的特征基系数序列用来生成矢量量化(VQ)的码书,设计出ICA-VQ破碎机故障诊断系统.现场采集数据的实验中系统的故障诊断准确率达到96.8%,表明系统的高效性.

To overcome the difficulty of complex background in mining machine fault diagnosis, a fault diagnosis system based on independent component analysis （ICA） and vector quantization （VQ） was developed. A fault sound ICA model was presented to get the fault sound feature bases with ICA algorithms in extracting nature images and continuous speech features. One ICA separated the sounds from different parts of the machine and the other extracted the feature basis of fault sound. The coefficients of the basis were used in designing codebooks. The diagnosis accuracy of this system is 96.8 % in the experiment with the realistic mine machine fault data, so the ICA-VQ is a high efficient fault diagnosis system.