The Smith Institute conducted an independent review of Network Rail’s methodology for estimating the rates of degradation of their earthwork cohorts, providing Network Rail with a benchmark against which to compare different statistical approaches. This project highlighted the versatility of mathematics and the value of cross-domain learning.
Maintaining railway assets to ensure their longevity for future generations is a key responsibility of Network Rail. This includes the management of almost 200,000 earthwork assets, composed of embankments, soil cuttings and rock cuttings up to 100 metres long. Understanding the rates at which earthworks are degrading is essential to enabling Network Rail to accurately assess the effectiveness of their interventions. Furthermore, since inaccurate degradation rates lead to sub-optimal maintenance and renewal planning, by improving their accuracy Network Rail can reduce asset spends and the risks to railway safety.
In our review we provided a critique of Network Rail’s existing statistical methodology based on data collected from historical examinations. We then developed a refined algorithmic approach using a Markovian framework and demonstrated its implementation on the historical inspection data. In addition, we proposed a method for measuring the effectiveness of interventions using the historical data.
A key step in the development of our alternative algorithmic approach was the identification of an analogous problem in biostatistics, in analysing rates of disease progression. From a modelling perspective the problems share many key features – disease progression and degradation both have random aspects and for each there is a final state (death for a patient or failure for an earthwork). Perhaps the most important of their common features is that earthwork degradation is only ever in one direction, from better to worse, just as for the disease progression in the absence of medical treatment. In both cases, improvement in condition occurs only as the result of some intervention.
The disease progression problem that we identified as analogous had well-established methodologies for calculating the rates of ‘degradation’ and even associated packages in the R programming environment. By making use of this existing mathematics and code, the Smith Institute were able to rapidly deliver recommendations for improvement to the earthworks degradation methodology to Network Rail.