Application of low-cost sensors and deep autoencoders for monitoring water pumps in particle accelerators

In particle accelerator facilities, cooling-water pumps play a critical role in removing substantial amounts (in megawatts) of waste heat from numerous high-power accelerator components (e.g., magnets, radio frequency structures, power supplies) and beamline components. Despite their role in daily operations, inspecting hundreds of water pumps is labor-intensive and performed only occasionally. Their unexpected failures can potentially lead to degradation of beam quality, hardware damage, and costly unplanned downtime. This study introduces an innovative method for real-time monitoring of water pump vibrations to identify anomalies that signal potential mechanical failures. Our approach integrates (i) low-cost vibration sensors, which will consistently sample pump vibration data and transmit it to a (ii) Deep Autoencoder model for detecting anomalies. The autoencoder model recognizes each pump's normal pump vibration patterns and identifies subtle deviations. This monitoring framework can facilitate proactive maintenance by enabling early detection of anomalies, enhancing pump reliability, lowering maintenance expenses, and minimizing costly downtimes.