Open-Source Framework for Fault Detection in District Heating Systems Released

A new open-source framework aimed at predictive maintenance in district heating substations has been launched, filling a significant void in the sector. This innovative system integrates a publicly accessible dataset, validated by service reports, with an evaluation technique that measures the accuracy, reliability, and timeliness of fault detection. The dataset comprises time series operational data from 93 substations belonging to two manufacturers, featuring annotations for disturbances caused by faults and maintenance actions, alongside examples of normal events and comprehensive fault metadata. The baseline results utilized EnergyFaultDetector, an open-source Python tool for automated anomaly detection in energy system data. Timely fault detection is crucial for lowering return temperatures and enhancing overall efficiency. Previously, advancements in this domain were hindered by a lack of public, labeled datasets. EnergyFaultDetector's evaluation employs three metrics: accuracy for identifying normal behavior, an eventwise F-score for dependable fault detection, and a measure of earliness. Both the framework and dataset are freely accessible to researchers and industry professionals.