Industrial process fault detection method and system based on weighted key principal elements

Abstract

The invention relates to an industrial process fault detection method and system based on weighted key principal elements, and the method comprises the following steps: S1, obtaining equipment sample data under a normal working condition, constructing a score matrix, calculating the local outlier factor value of the score matrix, and obtaining the control limit of equipment; and S2, collecting sample data of the equipment on line, obtaining a Hailinger distance change rate of the equipment, constructing a weighted score matrix based on the Hailinger distance change rate, calculating a local outlier factor value of the weighted score matrix, judging whether the local outlier factor value of the weighted score matrix is greater than a control limit of the equipment or not, if so, judging that the equipment has a fault, and if not, judging that the equipment is normal and returning to the step S2. Compared with the prior art, the fault can be well detected, the fault missing report rate is low, and the false alarm rate is low.

Description

technical field [0001] The invention relates to the field of fault detection, in particular to an industrial process fault detection method and system based on a weighted key pivot. Background technique [0002] With the rapid development of modern industry, the safety and reliability of industrial production has become particularly important. Process monitoring methods and technologies dominated by fault detection have become a hot spot in the field of industrial safety. At present, there are three main types of fault detection methods: fault detection methods based on mathematical models, fault detection methods based on qualitative knowledge, and fault detection methods based on data-driven; among them, fault detection methods based on data-driven methods do not require accurate system models. With the rapid development of sensor technology, the acquisition of massive data has become easier, so this method has attracted more and more attention. Principal component analy...

AI Technical Summary

Problems solved by technology

Jiang et al. counted the rate of change of single pivot statistics, and detected faults by selecting and monitoring sensitive pivots. However, the traditional CPV method was still used in the stage of selecting pivots, which would lead to the loss of useful information and affect the detection effect.

Cang et al. expressed the degree of variation of the principal component by constructing the rate of change of cumulative statistics, but traditional statistics require data variables to obey a normal distribution, and it is obviously difficult for the data collected in actual industrial processes to meet this restriction

At the same time, from a geometric point of view, the statistic is essentially an elliptical control boundary, and the error is large

Song proposed a pivot selection method for full variable expression (FVE), selecting the pivot with the greatest explanatory power for each variable as the key pivot, and retaining all variable information, but this method is still based on the form of statistics Construct corresponding statistics for detection. Secondly, all the above methods treat the selected pivots equally. In fact, when a fault occurs, only some pivots contain important information related to the fault.

Images