关键词: 故障检测, 数据深度, 核空间深度, 马氏深度, 秩统计量, TE过程

Abstract: In order to monitor the quality of production process of the process industry, a fault detection method based on data depth is proposed. The common mahalanobis depth and spatial depth are selected, and using Gaussian kernel function to generalize spatial depth in order to improve the sensitivity of spatial depth to position deviation. This method maps high-dimensional data to one-dimensional depth value by means of depth function （mahalanobis depth and kernelized spatial depth）, and then constructs asymptotic distribution by combining non-parametric rank statistics to make fault judgment. The effectiveness of the proposed method is verified through the Tennessee Eastman（TE） simulation experiment by referring to the two indicators of false alarm rate and detection efficiency and comparing with other methods.

Key words: fault detection, data depth, kernelized spatial depth, mahalanobis depth, rank statistic, TE process