基于重构的多工况过程无监督故障幅值估计

Unsupervised fault magnitude estimation of multi-mode industrial process based on reconstruction

针对先验知识不足的多工况工业过程的故障幅值估计问题,提出了一种基于重构算法的故障幅值估计方法。通过保局投影(Locality Preserving Projection,LPP)对历史数据进行降维,然后利用模糊c均值(Fuzzy C-means,FCM)实现对各个工况的划分,并确定检测指标和控制限、建立故障检测模型,最后利用重构算法来提取故障方向矩阵,并估计出故障的幅值。以田纳西-伊斯曼(Tennessee Eastman,TE)过程为研究对象进行仿真,结果表明提出的算法可以准确地估计出多工况工业过程的故障幅值、描述出故障幅值曲线。

To solve the problem of fault magnitude estimation during the muhi-mode industrial process lack of prior knowledge, a fault magnitude estimation method based on reconstruction is proposed. The dimensionality reduction of historical data is realized by LPP （locality preserving projection）, and the division of each operational mode is realized by FCM（fuzzy C means clustering）. Then the weighted monitoring index and control limit are defined. A fault detection model is established. At last, the fault direction matrix is extracted by reconstruction, and then the amplitude of the fault is estimated, the Tennessee Eastman benchmark process is selected to carry out simulation. The results show that the proposed algorithm can accurately estimate the fault magnitude of the multi-mode industrial process and describe the fault amplitude curve.