Dealing With Epistemic Uncertainties in Reliability-Based Pipeline Integrity Management

Abstract A key advantage of reliability-based integrity management is that it provides a consistent framework for explicit consideration of various uncertainties in the modelling and decision-making process. Uncertainties involved in reliability-based analyses are often classified into two categories, aleatory uncertainties, and epistemic uncertainties. Aleatory uncertainties are uncertainties considered “intrinsic” to system and thus are often challenging to further reduce, whereas epistemic uncertainties are subjective uncertainties that arise due to incomplete analyst knowledge. As such, epistemic uncertainties are often referred to as lack of knowledge uncertainties and they can be further reduced as new information comes to light. There are two major implications stemming from the differentiation between aleatory and epistemic uncertainties: (1) the estimated reliability performance and risk level calculated from reliability-based models are the results from a combination of both aleatory and epistemic uncertainties, and thus differing from the actual reliability performance of a system due to additional uncertainties from the lack of complete knowledge; (2) the modelling and decision-making process involves understanding and characterizing aleatory and epistemic uncertainties, including evaluating options for further reduction of epistemic uncertainties for more refined risk and reliability results. This paper discusses two commonly used approaches for dealing with epistemic uncertainties in reliability-based integrity management. The first approach is to have an explicit separation of epistemic and aleatory uncertainties involved in the analysis and produce a set of reliability and risk results corresponding to different levels of epistemic uncertainties. The advantage of this approach is that it can provide a simplified overview of the sensitivity of the results to epistemic uncertainties, which eases the evaluation of various options related to further reducing the epistemic uncertainties. However, for certain cases, the separation of epistemic and aleatory uncertainties can be challenging due to the subjectivity that is involved and potential correlations between epistemic and aleatory uncertainties. The second approach is to perform the full reliability-based analyses initially without considering the types of uncertainties. The differentiation of aleatory and epistemic uncertainties is performed in follow-up sensitivity analyses that evaluate the cost and benefits associated with various options to reduce epistemic uncertainty. This approach provides a more comprehensive analysis and decision-making process for dealing with epistemic uncertainties, but it could require significantly more modelling and analysis effort. Examples are presented in the paper to illustrate the two approaches and their advantages and disadvantages are demonstrated and discussed.