改进测度下的模糊C均值三元催化器故障诊断方法

针对采用物理建模刻画三元催化器故障演化精确性不足问题,提出一种基于尾气大数据的改进测度模糊C均值(FCM,fuzzy c-means),故障诊断方法。该方法包括分数阶傅里叶变换(FRFT,fractional fourier transform)下的故障特征提取与优化、核熵成分分析(KECA,kernel entropy component analysis)下的分形故障特征降维以及改进相似测度下的FCM故障特征聚类。首先,对不同工况的尾气数据进行FRFT处理,获取三元催化器从时域到频域的精细故障信息,同时利用粒子群算法(PSO,paticle swarm optimization)选取最优的FRFT特征,并由分形算子给出相应精细特征的分形维数;其次,借助KECA对候选的高维分形特征进行维数约简;最后,将获得的故障特征提交给改进测度的FCM故障分类器完成故障诊断。数值实验结果表明,较之采用欧式距离或余弦距离的FCM方法,研究方法的故障诊断精确度更高。

Fault diagnosis of three-way catalytic converter using improved fuzzy C-means clustering

The model precision of three-way catalytic converter is restricted by its complex physical and chemical reaction, which limits the accuracy of fault diagnosis based on its reaction model. To solve this problem, we propose a fault diagnosis method using improved fuzzy C-means (FCM) clustering. The method includes fault feature extraction and optimization using fractional Fourier transform(FRFT), dimensionality reduction of fractal feature using kernel entropy component analysis(KECA) and FCM fault feature clustering based on improved similarity measure. Firstly, we obtain the detailed features of different fault conditions from time domain to frequency domain using FRFT, then select the optimal FRFT order by particle swarm optimization (PSO) algorithm and these high-dimensional FRFT features with optimal order are transformed into fractal feature vectors through the fractal operator. Next, these fractal feature vectors dimensionality is reduced with KECA. At last, the reduced feature vectors are submitted to the improved FCM for fault clustering analysis. Numerical experiment results show that compared with the FCM method of Euclidean distance or cosine distance, the proposed method could obtain better fault identification result.