A Fault Detection Method Based on Autoencoder and Bayesian Network

Abstract

The invention proposes a fault detection method based on an automatic encoder and a Bayesian network, which belongs to the field of chemical process fault diagnosis. In the off-line stage of the method, variable data is selected from any continuous chemical production process to construct a chemical process data set and a sample data set; the sample data set is used to train the autoencoder model and the statistics T2 and SPE detection threshold are calculated; the chemical process data set is used to construct Bayesian network and estimate the conditional probability; at this stage, obtain real-time data and input the autoencoder model to obtain the corresponding estimated value, calculate the T2 and SPE values ​​corresponding to the input data and compare them with the detection threshold: if the conditions are met, the chemical production The process is normal; if it is not satisfied, calculate the contribution of each variable, and find the root cause of the fault through the Bayesian network. The invention automatically extracts features from process data of chemical production, is effectively applied to nonlinear dynamic chemical process, and realizes detection and rapid diagnosis of chemical process faults.

Description

technical field [0001] The invention relates to the field of chemical process fault diagnosis, in particular to a fault detection method based on an automatic encoder and a Bayesian network. Background technique [0002] The chemical production process is a complex process. In order to maximize economic benefits, the chemical production process often requires "safe, stable, long, full, and excellent", and requires the production equipment to be able to run smoothly for a long period of time. With the continuous development of automatic control technology, the control of the device is tightly coupled with the state variables. When an abnormality occurs in a major variable in the chemical production process, these abnormalities will propagate through the device through mass transfer, heat transfer, etc., thereby As a result, the entire device fluctuates, causing a flood of alarms in the device, making it difficult for the operator to diagnose the fault correctly. If the oper...

AI Technical Summary

Problems solved by technology

The encoder and decoder of traditional autoencoders are generally composed of multi-layer feedforward neural networks, which do not consider the expansion of dynamic time

Images