论文部分内容阅读
提出一种新的针对KPCA模型的故障识别方法——贡献率图法。该方法是在微分贡献率图和核函数导数的基础上提出来的,它采用统计量T2和SPE对每个变量的偏导数来度量每个变量对统计量T2和SPE的贡献率。和基于数据重构法的KPCA故障识别方法相比,该方法不需要任何迭代近似计算和数据的重构,计算量小且可避免重构产生的误差对识别结果的影响。通过在某型涡扇发动机故障检测与诊断中的应用表明,该方法比基于数据重构法的故障变量识别准确率更高,再结合发动机故障机理分析,便可准确地确诊故障,从而大为缩短故障定位及排故的时间,预防重大事故的发生。
A new fault identification method based on KPCA model is proposed. This method is based on the differential contribution rate graph and kernel function derivatives. It uses the statistic T2 and SPE to measure the contribution of each variable to the statistic T2 and SPE using the partial derivative of each variable. Compared with the KPCA fault identification method based on data reconstruction method, this method does not need any iterative approximation calculation and data reconstruction, and the calculation is small and the influence of the reconstruction error on the recognition result can be avoided. The application of fault diagnosis and diagnosis in a turbofan engine shows that this method is more accurate than the fault variable identification based on data reconstruction method. Combined with the analysis of the mechanism of engine failure, the method can accurately diagnose the fault and make it greatly Shorten the time to locate and troubleshoot the accident and prevent the occurrence of major accidents.