An Enhanced Unsupervised Extreme Learning Machine Based Method for the Nonlinear Fault Detection

Although the unsupervised extreme learning machine (UELM) based methods have been widely used to diagnosis nonlinear process faults recently, UELM algorithm is only designed preserve local adjacency similarity of input dataset instead mining intra-class variations. Besides, determination optimal hidden nodes number a tough issue. In order deal with these two problems, novel enhanced (EUELM) scheme developed effectively detect in our work. proposed EUELM approach, first improved by naturally incorporating diversity analysis technique into original objective function both variation and dataset. Then, settle difficult issue selecting nodes, kernel trick further employed approach mine data strong nonlinearity. Based on extracted low dimensional feature information, k-nearest neighbor (KNN) principle applied derive monitoring statistic for fault detection. At last, experiments comparisons effectiveness suggested are made numerical system benchmark Tennessee Eastman (TE) process. The obtained results illustrate that significant improvements can be achieved detection compared other popular related approaches.