Predictive maintenance for water and pipeline infrastructure, phase 2
Built an XGBoost-based pipeline on Stavanger municipality data to predict pipe failure risk across multiple time horizons, combining asset attributes with break history and spatial neighbor features.
In Spring 2026, ReLU moved the Gemini VA collaboration from exploratory analysis into model development and delivery.
Pipe failure risk modeling
Using data from Stavanger municipality, the team built an XGBoost-based pipeline that predicts pipe failure risk across multiple time horizons, ranging from short-term planning to several years ahead. The model combines static pipe attributes with engineered features capturing each pipe’s break history and the failure activity of its spatial neighbors, including a distance-weighted indicator of nearby leakage.
The final deliverable consists of the trained model, the full modeling pipeline, and accompanying documentation.