Neighborhood Preserving Embedding Based on Temporal Extension and Its Application in Fault Detection

In order to handle the feature extraction and dimension reduction of dynamic autocorrelation data, this paper presents a temporal extension method of neighborhood preserving embedding(TNPE) for fault detection. Taking the limitation of the existing NPE into account, a new optimizing target is constructed by incorporating both the spatial feature and the temporal relation among the process data. Comparing to original high dimensional variable space, the obtained low dimensional projection space has similar local spatial structure and the temporal dynamic structure, such that more feature information can be extracted. By means of the TNPE algorithm, the original variable space is divided into the feature space and residual space. Moreover, Hotelling’s T2 and squared predication error are constructed upon the TNPE model to monitor the variations among the two spaces. A case study on the Tennessee Eastman process demonstrates the feasibility and efficacy of the proposed method in this paper, which also shows the superiority in fault detection.