Determination of Risk-Reducing Intervention Programs for Railway Lines and the Significance of Simplifications

Because failures on railway lines have nonzero probabilities of occurrence and can result in significant costs if they occur, railway infrastructure managers are interested in determining intervention programs that best reduce this risk, taking into consideration their budget constraints. In this paper, a base model is proposed to determine optimal risk-reducing intervention programs for railway lines based on the states of its objects and budget availability. The base model is an integer nonlinear model with an objective function that maximizes net benefit without exceeding budget constraints. The net benefit is the difference between the amount of risk reduction and the costs of executing interventions in terms of both direct costs (i.e., materials and labor forces) and indirect costs (i.e., the travel time costs). Three variations of the base model, i.e., the object model, the block model, and the line model, are used to investigate the trade-offs between increasingly simplified models and decreasing ability to determine the optimal intervention program due to decreasing ability to accurately estimate costs and benefits. In the object model, objects are considered in isolation, i.e., while each object fails and is restored all other objects in the line are fully functional. In the block model, blocks are observed in isolation, i.e., while each block fails and is restored all objects in the line are fully functional, whereas the multiple objects within a block may fail and be restored simultaneously. In the line model, all objects in the line may fail and be restored simultaneously. Fault tree analysis is used in the block and line models to estimate the costs of the combined failures of multiple objects. The three models are demonstrated by using them to determine the optimal intervention program for a fictive railway line between two stations that consisted of eight track sections, a bridge, two switches, and two signals. The intervention programs determined using the three models in terms of the interventions included and the net benefit obtained are compared. It is shown that all three variations produce useful results but that there are significant differences in the estimation of the net benefits using the three different models, and that these differences lead to different interventions being included in the determined intervention programs, and consequently in the net benefit that will be achieved through their implementation. It was also shown that each improvement in the estimation of accuracy comes with an increase in modeling complexity.