Using operational data for decision making: a feasibility study in rail maintenance

Abstract

In many organisations, large databases are created as part of the business operation: the promise of ‘big data’ is to extract information from these data- bases to make smarter decisions. We explore the feasibility of this approach for better decision-making for maintenance, specifically for rail infrastructure. We argue that the data should be used within a Bayesian framework with the aim of inferring the underlying state of the system so we can predict future failures and improve decision-making. Within this framework, some data is diagnostic of this underlying state and other data have a causal influence. The framework can be realised as a Bayesian network and the probabilistic rela- tionships in this network can be learnt from data. However, the network can- not be created just from data; instead experts’ knowledge is vital for the model’s structure as some variables representing the underlying state of the system may not be present in the data. We outline an architecture for a smart decision tool and show that the GB railway industry has the data needed. The challenges of developing such a tool are also discussed. For example, the required data are distributed across multiple databases and both within and between these databases important relationships, such as physical proximity, may not be represented explicitly.