Bayesian Probabilistic Asset Management Protocol for Infrastructure Systems: Example Application to a Flood Protection System

摘要

A Bayesian probabilistic asset management protocol was developed for risk management of a typical flood protection system. A major advantage of the proposed protocol is its ability to integrate and transform data from field inspections and other information sources into quantitative measures of risk for each system component to help prioritize decisions that reduce risk. While the example application is for a flood protection system subdivided into reaches with varying attributes, its underlying framework is highly versatile and can be applied to a broad spectrum of infrastructure assets and asset systems. The protocol developed for the typical flood protection system uses risk assessment tools to identify reaches with potential risk and the possible measures to reduce that risk. The general approach adopted is to update a background failure probability with information from new observations specific to the reach being evaluated. The background and updated probabilities are termed prior and posterior probabilities, respectively. The prior probability of failure of a reach is determined using historical performance information and subject matter experts. Here, the prior probability of failure does not account for the current condition state or site-specific attributes of the reach being assessed. The posterior probability of failure incorporates new information by using Bayes theorem to combine the prior probability of failure and the likelihoods of observing the condition state for cases of reaches failing and not failing. The proposed protocol consists of three phases: (1) Development of a probability database consisting of prior probabilities and likelihoods; (2) Assessment of the current performance indicators for each reach; and (3) Determination of the updated probability of failure of each reach by combining new observations about the condition state of each reach with information stored in the probability database. The first phase, which is usually performed only once for a given system, starts with assigning potential modes of failure for the reaches. A given mode is then associated with performance indicators that each provide information on the condition state of the reach based on three levels of severity. The likelihoods of observing each level of severity in failure and non-failure conditions are determined using historical failure statistics supplemented with subject matter expert opinion. In some cases, this could be corroborated by available data such as design factors of safety, reliability indices, inspection reports, photographs, and construction records. In the second phase, data from field inspections and field instrumentation are used to assign a current condition assessment for each failure mode for each reach. Subsequently in the third phase, this condition assessment is combined with relevant information in the pre-populated probability database to calculate the posterior probability of failure using Bayes theorem. This revised estimate of probability of failure is then combined with the consequences of failure to provide a quantitative measure of risk for each reach. The posterior probability can be further updated as new information about the condition state becomes available.