Stedin geographic AI maintenance models

by Stedin

AI Application

Low Confidence

Stedin's geo data science team built machine-learning models that predict where failures are likely in its gas and electricity networks, from main pipelines to home connections. The models are trained on data from 2013 and produce predictions up to 2042, using about 50 factors such as soil subsidence measured from satellite imagery, tram and rail lines, road works and groundwater levels that affect pipe corrosion. One model estimates the probability of a fault and another the consequences based on the surroundings; together they form a risk model per component. Stedin staff use the yearly risk calculations to rank which maintenance and replacement projects come first and to find measures with structural safety gains. Data are combined and shared as maps in Esri ArcGIS, and results of finished projects are fed back so the models are retrained continuously. The project was nominated in the Digital Transformation category of the Computable Awards 2023.