Background and objectives

For environmental and safety reasons, our client, a UK gas distribution company, needs to fulfil the regulatory obligation to replace all of its iron gas pipes by PE by 2032. With more than 100k pipes to be replaced with a length of roughly 10,000km, the decision to be optimised is not which pipe to replace but which pipe to replace first. For gas distribution companies, the repair costs amount to roughly 15-20% of the total cost base. Pipe repair is not only a significant cost-driver, it also is an activity with high impact of customer satisfaction and a reflection of asset health. Therefore, increasing the accuracy of this prediction does not only reduce repair cost benefits that go straight to the bottom line, it also increases network reliability and it reduces customer interruptions and therewith increases CSAT for which the gas distributors get an additional allowance.

 

Our approach

Together with our client we developed a model to predict the likelihood of failure of the iron pipes. Because our client does develop an annual replacement program, our model was designed to predict which pipes were most likely to leak in the next regulatory year (some companies work with multiple year programs for which the software can easily be adjusted). The accuracy of this model had to outperform the model that was used by all UK gas distribution companies to determine the risk of a pipe failing. The type of data we used for the training and testing of our model are threefold: 1) asset specific data such as diameter, pressure, volumes, ground type, surface type etc; 2) asset interaction data such as survey data, maintenance data, repair data on surrounding pipes, etc.; 3) fault data, incl. pipe failures specified by connections, erosions, fractures and cost data of repair jobs and replacement jobs. In developing the predictive software, we worked closely together with our client with the aim to transfer the knowledge required to maintain the software and, with the data being enriched year on year, iteratively improve the accuracy of the model.

 

Results

Our predictive model does double the avoided repair costs. In response, our client has decided to embed our software in their annual decision making process for the determination of the optimal replacement program. In addition, to secure the sustainability of these benefits, we have transferred the required knowledge to their internal analytics team.