Fault detection method based on margin statistics of generalized non-negative matrix factorization

Abstract

A novel fault detection method based on margin statistics of generalized non-negative matrix factorization (GNMF) is proposed. The construction of traditional process monitoring method based on multivariate statistical that neglects the correlation relation and feature distribution of latent variables at different sampling times, and the method also need to assume that latent variables satisfy a particular distribution. Therefore, considering the characteristic of GNMF which has no assumptions on the data distribution, first of all, the GNMF method is applied to extract the latent variables of the process, and the statistics and control upper limits would be obtained in the traditional sense. On this basis, the secondary control limit is constructed on the process data and the control margin is established. Then, the time-varying information of different sampling time is further analyzed. The normal data is modeled to obtain the relevant parameters which are used to calculate the margin of fault data, thus, a new margin statistics is constructed. The fault detection is carried out under the control upper limits. Finally, the proposed method is applied to the Tennessee Eastman process to evaluate the monitoring performance. The experiment results clearly illustrate the feasibility of the proposed method.