Why Pipe Age Is a Poor Predictor of Pipe Failure

XGBoost predicted 30% of pipe failures across Barcelona's 4,700 km network, challenging Europe's reliance on age-based renewal

10 Apr 2026

European water utilities may have a more reliable way to manage aging infrastructure, after researchers showed that machine learning can predict nearly a third of pipe failures before they occur.

A study published in February 2026 in Applied Water Science deployed an XGBoost model across Barcelona's 4,700-kilometre underground water network, drawing on more than two decades of failure records. Researchers from Aigues de Barcelona and the Barcelona Supercomputing Centre tested more than 28,800 algorithm configurations, optimising for operational utility targets rather than standard statistical benchmarks.

The model predicted 30.2% of expected failures at a 5% annual renewal rate, and more than 10% at the 1% rate common among most utilities. Both figures represent a measurable advance on conventional methods.

The most consequential finding may be what the model revealed about pipe age. Most European utilities use age as the primary basis for scheduling replacements. The research found it is not the dominant failure predictor. Hydraulic pressure conditions and the geographic clustering of failures in neighbouring pipe sections proved far more informative. That result alone could prompt utilities across the continent to rethink how they allocate capital.

Non-revenue water losses average around 25% across EU member states. Identifying which pipes are most likely to fail, before they do, reduces emergency repair costs, limits service disruption, and supports more efficient long-term planning.

The research also resolved a methodological problem that has long complicated pipe failure analysis. By developing a new method to trace pipe section histories across decades of network changes, the team reconnected 22% of historical failure records that would otherwise have been misattributed. That contribution may prove as durable as the predictive results themselves.

The European Commission is preparing an Action Plan on Digitalisation in the Water Sector for 2026. Whether utilities can act on findings of this kind at scale, given the cost and complexity of replacing legacy data systems, remains an open question.