FLASH-FAULT. Fast learning algorithm for a single sensor based heating system fault detection

Abstract

The potential of machine learning (ML) algorithms for fault detection in residential building heating systems is well established in scientific literature including our recent works. We developed a preliminary ML-pipeline, that utilizes time series forecasting to identify faults in a solar thermal system using minimal initial data unlike existing data-intensive approaches. Although the ML algorithm achieves similar accuracies as the rule-based algorithm integrated into the industrial partner’s single-sensor IoT framework, it lacks theoretical testing and conceptual evaluation.

We aim to advance this ML approach to industrial maturity, necessitating a critical evaluation of our ML pipeline and support for robust deployment.

This project has significant potential for fault detection in other heating systems, such as heat pumps and district heating, projected to play a major role in Switzerland’s future energy mix. Collaboration with SDSC is essential for ensuring the successful translation of our research prototype into a robust, production-ready application, significantly impacting energy efficiency and sustainability in building heating systems.