Generic Bayesian network models for making maintenance decisions from available data and expert knowledge

Abstract

To maximise asset reliability cost-effectively, maintenance should be scheduled based on the likely deterioration of an asset. Various statistical models have been proposed for predicting this, but they have important practical limitations. We present a Bayesian network model that can be used for maintenance decision support to overcome these limitations. The model extends an existing statistical model of asset deterioration, but shows how (1) data on the condition of assets available from their periodic inspection can be used, (2) failure data from related groups of asset can be combined using judgement from experts and (3) expert knowledge of the deterioration's causes can be combined with statistical data to adjust predictions. A case study of bridges on the rail network in Great Britain (GB) is presented, showing how the model could be used for the maintenance decision problem, given typical data likely to be available in practice.

Overview on performance predictive models – Application to Bridge Management Systems

“…A Bayesian network (BN) represents the joint probability of a set of random variables [24], which are linked under certain dependencies between them. Those relationships can be represented by a directed acyclic graph [25] where the variables are nodes, and the arrows directions designate the parents and the children.…”

Towards a model based asset deterioration framework represented by probabilistic relational models

“…We illustrate the use of the framework with multiple variants of deterioration models. can be adapted when there is insufficient data, both with expert knowledge (Frangopol et al, 2004, Zhang andMarsh, 2018), or by learning from similar groups (Memarzadeh et al, 2016, Zhang andMarsh, 2018). In the work of Zhang and Marsh (2018), six generic BN models for asset deterioration were developed, which both provides us the possibility of adopt different deterioration models, but also enables us to include alternative data and unused expert knowledge.…”

“…can be adapted when there is insufficient data, both with expert knowledge (Frangopol et al, 2004, Zhang andMarsh, 2018), or by learning from similar groups (Memarzadeh et al, 2016, Zhang andMarsh, 2018). In the work of Zhang and Marsh (2018), six generic BN models for asset deterioration were developed, which both provides us the possibility of adopt different deterioration models, but also enables us to include alternative data and unused expert knowledge. These model variants cannot yet be presented to an asset deterioration domain expert in a unified framework: adapting the underlying concepts to a particular context requires a deep understanding of their implementation as BNs.…”

“…The generic models in the model library are represented as abstract probabilistic relational models (PRMs). Figure 4 shows an example developed from Zhang and Marsh (2018). In Figure 4, a square (called a class) defines a group of identical objects that share the same set of variables or probabilistic models.…”

“…Customised instantiation of this framework for practical uses have been developed, for example, general maintenance problems in Zhang and Marsh (2016), and rail bridge deterioration in Zhang and Marsh (2018). Since the focus of this paper is to show how the model-based asset deterioration framework to deal with different modelling assumptions that may happen in practice, we present a series of instantiation variants of a basic deterioration model, which is sufficient to convey the idea.…”