Please use this identifier to cite or link to this item:
|Title:||Dempster-Shafer fusion for collaborative fault detection and identification (FDI) with application to a distillation column case study||Authors:||Ghosh, K.
|Issue Date:||2008||Citation:||Ghosh, K.,Ng, Y.S.,Srinivasan, R. (2008). Dempster-Shafer fusion for collaborative fault detection and identification (FDI) with application to a distillation column case study. AIChE 100 - 2008 AIChE Annual Meeting, Conference Proceedings : -. ScholarBank@NUS Repository.||Abstract:||Most of the literature on fault detection and identification (FDI) for chemical processes depend on a single method such as principal component analysis (PCA), artificial neural-networks (ANN), qualitative trend analysis (QTA), signal processing methods or first principles model. Each of these methods inherits certain strengths and shortcomings. A method that works well under one circumstance might not work well under another when different features of the underlying process come to the fore. Any single FDI method alone is not sufficient enough for monitoring and fault diagnosis of a chemical process under all circumstances. But the shortcomings of each individual FDI method can be overcome by means of meaningful collaboration among them. Ng and Srinivasan proposed an agent-based approach for collaborative FDI, where each FDI method was considered as a monitoring and / or diagnostic agent. The results produced by the agents were fused by a consolidator agent. The advantages of the collaborative FDI approach - significant improvement in recognition rate, and reduction in both fault detection and diagnostic time - were demonstrated through two case studies. These advantages accrue when good collaboration can be achieved. This paper focuses on effective collaboration strategy. When multiple FDI methods are used, the results produced by different FDI methods might be in conflict. Ng and Srinivasan used Voting and Bayesian-based fusion strategies to combine the results of the FDI agents and to resolve conflicts among them. The voting-based fusion technique relies solely on the predictions produced by all the FDI methods. Each FDI method is usually treated equally, although the predictions from some FDI methods outperform those from others in some situations and should hence be given more weight. Such information is incorporated in Bayesian-based fusion. The Bayesian technique is a popular method for conflict resolution among multiple classifiers. It estimates the posteriori probability of an event from the a-priori knowledge of class-specific performance of each individual classifier. The class-specific performance of a classifier is captured in a confusion matrix that is constructed by testing the classifier on training datasets. While the confusion matrix is adequate for resolving conflicts when all the classifiers make some prediction, it cannot effectively handle situations when some of the classifiers are unable to classify an event into any known class. This is an important issue since this precludes specialization of the classifiers. We seek to develop this issue in the current work The Dempster-Shafer based fusion method is a generalization of the Bayesian theory and provides a systematic way to account for the classification inability of classifiers. In Dempster-Shafer technique the performance of each classifier is measured in terms of recognition rate, substitution rate, and rejection rate. The first two measures reflect the correct and incorrect classification performance respectively. The last measure - rejection rate - accounts for the inability of classifiers to provide an answer (class) in a given situation. The Dempster-Shafer fusion strategy estimates the likelihood of the fault based on the available evidence without requiring every FDI method to contribute an answer in every situation. In this paper, a collaborative Fault Detection and Identification (FDI) method using Dempster-Shafer fusion approach is presented. The proposed collaborative approach is illustrated through the monitoring and fault diagnosis of a continuous lab-scale distillation column where different FDI methods, principal component analysis, self-organizing maps, model-based filters/observers and neural-networks are employed. The results obtained from these FDI methods are fused through Dempster-Shafer based fusion. The results show that Dempster-Shafer based collaboration performs better than any single FDI method, as well as other fusion schemes.||Source Title:||AIChE 100 - 2008 AIChE Annual Meeting, Conference Proceedings||URI:||http://scholarbank.nus.edu.sg/handle/10635/74541||ISBN:||9780816910502|
|Appears in Collections:||Staff Publications|
Show full item record
Files in This Item:
There are no files associated with this item.
checked on Nov 24, 2022
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.