I am pleased to announce that I will be presenting at the 2018 Global Congress on Process Safety (AIChE) in sunny Orlando, FL. I will be presenting in the Instrumented Protective Systems track on the subject of A Hierarchical Bayesian Approach to IEC 61511 Prior Use.
This paper builds on my prior work showing the advantages of a Bayesian approach to the analysis of failure rate data for Safety Instrumented Systems. The Bayesian approach takes into account prior knowledge about failure rates and uncertainties and then uses a Bayesian updating process to update our knowledge as new failure data is available.
The intent is to go beyond just reliability modeling and actually close the loop on SIS performance monitoring
This latest paper looks at the question of how to handle failure data that comes from similar (but not identical) populations. We demonstrate how using a Hierarchical Bayesian approach , the data from all of the different non-homogenous populations can be combined and used to draw inferences.
We step through the mathematical basis at a high level. However, the integral equations to solve a hierarchical model are generally intractable, so Markov Chain Monte Carlo simulation is used for the analysis.
We give a practical example of the analysis using the JAGS software running inside RStudio. Finally, we demonstrate how this approach fits nicely into an enterprise data management framework.
The significance of this approach is that it allows the failure rate distribution for each population (i.e. service type) to “borrow strength” from other populations where appropriate. In addition to making use of prior knowledge, the hierarchical approach makes the maximum use of the available data.
For more about Bayesian statistics, please check out courses and resources from our affiliate partners, including: