基于高斯混合模型和变量重构组合法的故障诊断与分离

Fault Diagnosis and Isolation Based on Combination Gaussian Mixture Models and Variable Reconstruction

提出了一种将变量重构与高斯混合模型结合的故障诊断与分离的方法。首先建立过程数据的高斯混合模型,解决了监控过程的测量数据不服从单峰的高斯分布所带来的问题,然后进行故障数据变量重构,估计未知参数并采用最大期望算法来估测均值与协方差矩阵。在此基础上建立统计模型进行故障的诊断与分离。与传统的贡献图分离故障的方法比较,通过田纳西-伊斯曼化工过程进行实验验证,本文提出的高斯混合模型与变量重构相结合对多状态过程进行故障的诊断与分离收到较好效果。

A fault diagnosis and isolation approach is presented combining variable reconstruction and Gaussian mixture models （GMMs）. Firstly, GMMs for process data are built to overcome the problem that the operating data cannot follow a unimodal Gaussian distribution. Then the variable reconstruction, the estimation of unknown factors and its mean and covariance are completed, as well as the fault diagnosis and isdation. The combination methods are illustrated for a simulated Tennessee-Eastman chemical process （TE） tested for fault diagnosis and isolation and result is better than contribution plot.