关键词: 相对测地距离, 子流形, 核ridge回归, 故障检测, 非线性过程, 多模态过程

Abstract: Industrial processes are often operating under different modes, while there are nonlinear correlations between data monitored.Aiming at these problems, a fault detection method based on relative isometric mapping (RISOMAP) was proposed.Relative geodesic distance was used to establish distance matrix in the high dimensional space, and multi dimensional scaling (MDS) was used to calculate output in the low dimensional embedded space.Information of sub-manifold and error could be obtained, and then monitoring statistics were built for fault detection.Meanwhile, kernel ridge regression was used to obtain the lower dimensional output of test data.Besides,kernel matrix was updated through integrated similarity.The simulations of visualization case and TE process illustrated that in contrast to fault detection methods based on kernel principal component analysis (KPCA) and ISOMAP, the proposed method could detect process fault more effectively and quickly.It also provided an idea to implement fault detection without prior knowledge in the multimode process.

Key words: relative geodesic distance, sub-manifold, kernel ridge regression, fault detection, nonlinear process, multimode process