Retrofitting Water Pumps with Health Monitoring and Machine Learning at the Edge

Water pumps are critical components of modern infrastructure whose maintenance poses significant logistical and financial challenges. Addressing this issue, we propose a solution by retrofitting running pumps with health monitoring, deploying state-of-the-art machine learning (ML) techniques on edge sensors. Our work explores how basic power usage sensors can assess informative features about pump operation cycles. A combination of ML approaches, which can either be run at the edge or as a remote cloud-based analysis service, allows us to then assess the health status of the water pump over time. Our conceptual approach was practically evaluated in a real-world field test on different pump installations, from which we derive \texttt{SYPHM}, a synthetic data generator for pump health monitoring. Tested across various experiments, our approach demonstrated high accuracy in detecting anomalous patterns while balancing privacy, scalability, and cost-efficiency. As such, our research presents novel insights for deploying ML for critical infrastructure, enabling affordable predictive maintenance for sustainable water technology.