Logo image
Probabilistic modeling of process systems with application to risk assessment and fault detection
Dissertation   Open access

Probabilistic modeling of process systems with application to risk assessment and fault detection

Taha Mohseni Ahooyi
Doctor of Philosophy (Ph.D.), Drexel University
Dec 2015
DOI:
https://doi.org/10.17918/etd-6598
pdf
Mohseni-Ahooyi_Taha_20156.41 MBDownloadView

Abstract

Data Interpretation, Statistical Medical statistics Chemical Engineering
Three new methods of joint probability estimation (modeling), a maximum-likelihood maximum-entropy method, a constrained maximum-entropy method, and a copula-based method called the rolling pin (RP) method, were developed. Compared to many existing probabilistic modeling methods such as Bayesian networks and copulas, the developed methods yield models that have better performance in terms of flexibility, interpretability and computational tractability. These methods can be used readily to model process systems and perform risk analysis and fault detection at steady state conditions, and can be coupled with appropriate mathematical tools to develop dynamic probabilistic models. Also, a method of performing probabilistic inference using RP-estimated joint probability distributions was introduced; this method is superior to Bayesian networks in several aspects. The RP method was also applied successfully to identify regression models that have high level of flexibility and are appealing in terms of computational costs.

Metrics

31 File views/ downloads
20 Record Views

Details

Logo image