A Robust Fault Detection Method Based on Neighborhood Preserving Embedding

Industrial data are in general corrupted by noises and outliers, which do not meet the application assumptions in neighborhood preserving embedding (NPE). Many existing NPE-like algorithms are not robust, overly consider the local features of the data, and cannot capture the key features of the data. To this end, multilevel NPE (MNPE) is incorporated into the self-paced learning (SPL) framework in this study. Compared with single-projection feature extraction algorithms, MNPE consists of two projections: the nonreduced and reduced dimensionality projections. The nonreduced dimensionality projection can remove the redundant features that are unrelated to the key features of the data. The reduced dimensionality projection can reduce the dimensionality of the data and further extract the features of the data. Moreover, the L21 -norm can enhance the robustness of the proposed algorithm, and the SPL framework can avoid the local optimal solution problem caused by nonconvex optimization. Extensive experiments have been conducted on multiple image classification datasets to demonstrate that the proposed method is more effective than other state-of-the-art methods. The numerical simulation and blast furnace fault detection experiment results further validated the effectiveness of the proposed method.