Distributed dynamic process fault detection method based on mutual information

Abstract

The invention relates to a distributed dynamic process fault detection method based on mutual information. The distributed dynamic process fault detection method based on mutual information comprises the steps: introducing a time-delay measured value for each measured variable of the process; by means of the correlation index defined by mutual information, distinguishing the auto-correlation and the cross-correlation displayed at different sampling time for each measured variable of the process; respectively establishing a corresponding principal component analysis fault detection model for a data set subblock corresponding to each variable; and at last, during the process of implementing on-line monitoring, utilizing Bayesian reasoning to integrate the results of different fault detection models into one probability type monitoring index so as to make a final fault decision conveniently. Compared with the prior method, the distributed dynamic process fault detection method based on mutual information fully considers the auto-correlation and the cross-correlation among different measured variables at different sampling time, and avoids losing the useful information which may be hidden in the complex dynamic characteristic of the process data.

Description

technical field [0001] The invention relates to an industrial process fault detection method, in particular to a distributed dynamic process fault detection method based on mutual information. Background technique [0002] In recent years, data-driven fault detection methods have been paid attention to by researchers as an important technical means to ensure production safety and product quality stability because of their simplicity and versatility. Research on multivariate statistical process monitoring methods represented by Principal Component Analysis (PCA) has attracted extensive attention from industry and academia. potential information. This type of method can avoid establishing an accurate process mechanism model, so it is very suitable for monitoring modern large-scale and complex industrial processes. [0003] Generally speaking, the data collected in the production process is inevitably dynamic (or called autocorrelation) due to the short sampling interval. Am...

AI Technical Summary

Problems solved by technology

However, traditional methods such as DPCA and DLV generally assume that the autocorrelation and cross-correlation between measurement variables are consistent in sampling time, ignoring the useful information that may be hidden by the complex dynamic characteristics of process data.

Images