Dynamic Failure Analysis of Process Systems Using Principal Component Analysis and Bayesian Network

Abstract

Modern industrial processes are highly instrumented with more frequent recording of data. This provides abundant data for safety analysis; however, these data resources have not been well used. This paper presents an integrated dynamic failure prediction analysis approach using principal component analysis (PCA) and the Bayesian network (BN). The key process variables that contribute the most to process performance variations are detected with PCA, while the Bayesian network is adopted to model the interactions among these variables to detect faults and predict the time-dependent probability of system failure. The proposed integrated approach uses big data analysis. The structure of BN is learned using past historical data. The developed BN is used to detect faults and estimate system failure risk. The risk is updated subsequently as new process information is collected. The updated risk is used as a decision-making parameter. The proposed approach is validated through a case of a crude oil distillation unit operation.

Knowledge acquisition development in failure diagnosis analysis as an interactive approach

“…Probabilities can be inserted into the BN in order to find the probability of the outcome; same as FT. The flexibility of BN structure and its probabilistic reasoning engine has enough capability for risk analysis in the large and complex systems [51,57,58,[69][70][71][72][73][74]. In a BN model, both forward and backward analysis could be performed.…”

Integrated Diagnostic Framework for Process and Sensor Faults in Chemical Industry

“…Generally, fault diagnosis methods can be divided into three categories: model-based [ 11 ], knowledge-based [ 12 ], and data-based [ 13 ] methods. Among them, the principal component analysis (PCA) in data-based methods is widely used for process and sensor fault diagnosis in chemical processes [ 14 , 15 ]. Ku et al, Lee et al, and Yang et al successively proposed the dynamic principal component analysis (DPCA) [ 16 ], kernel principal component analysis (KPCA) [ 17 ], and dynamic kernel principal component analysis (DKPCA) [ 18 ] for process fault diagnosis.…”

Logic‐based probabilistic network model to detect and track faults in a process system

“…PCA is able to use multivariate data for sensor selection of BN input. 23 Related works used this integration by modeling causal dependencies among the principal components (PCs) for prognostics 24,25 and diagnostics. 26 They used PCA to reduce dimensions of BN into PCA-BN model for inferring the accidental consequences of an abnormality in the works.…”

“…For case studies, predefined models were often used to simulate the abnormal behavior of a process system that fed algorithms for learning and the BN for testing. For instance, the work by Adedigba et al 25 has not been tested PCA-BN applicability on a real plant since data for learning and testing is often generated from a dynamic simulation software. However, the built-in solver of software was unable to find steady state solutions in the whole range of deviations.…”