Method for detecting industrial process faults basing on multi-sampling rate factor analysis model

Abstract

The invention discloses a method for detecting industrial process faults basing on a multi-sampling rate factor analysis model. The method comprises the following steps of: collecting data of different sampling rates in a normal working state; establishing a multi-sampling rate factor analysis model; constructing two detection statistic values T<2> and SPE and the corresponding statistical limitsthereof (T<lim><2>) and SPE<lim> as shown in the specification using latent variables and prediction errors of the model; detecting an online industrial process to obtain a test sample; and estimatingthe latent variables and the prediction errors of the test sample using a existing model structure, calculating corresponding statistic values, and obtaining a final fault detection result. Accordingto the industrial process fault detection method based on multi-sampling rate factor analysis model, the multi-sampling rate information processing is achieved, data information can be fully utilized, and the variance differences of data noise matrices can be fully considered, so that a few latent variables after dimension reduction are better explained and described for the original variables, thereby improving the fault detection precision.

Description

technical field [0001] The invention relates to the technical field of chemical process detection, in particular to an industrial process fault detection method based on a multi-sampling rate factor analysis model. Background technique [0002] In modern industrial processes, with the widespread application of distributed control systems (DCS) in industrial fields, industrial sites can achieve high sampling rate acquisition and storage of process variables such as flow, temperature, and liquid level, but they are not related to product quality or process processing. The quality variables related to the results are difficult to achieve high sampling rate acquisition due to the limitation of assay costs and other aspects, resulting in the multi-sampling rate characteristics of information acquisition, which is a challenge for rational and effective use of these data information. [0003] At the same time, with the continuous advancement of multivariate statistical analysis-bas...

AI Technical Summary

Problems solved by technology

Among them, the process detection technology based on principal component analysis (PCA) can realize the dimensionality reduction of the data through the selection of the original variables, but cannot realize the utilization of complete data information

The process detection technology based on partial least square estimation (PLS) uses quality variables to guide the decomposition of the process variable sample space, and the obtained projection space only reflects the changes related to the quality variables in the process variables, and it is also impossible to realize the massive data information. full explanation

However, although the process detection technology based on Probabilistic Principal Component Analysis (PPCA) considers the influence of noise when building the model, it is not detailed enough to describe and deal with the variance of noise variance, which affects the accuracy of fault detection.

Images