Paper accepted at Viennahydro 2026

Our paper Multi-Modal Acoustic-Vibration Array Monitoring for Anomaly Detection and Source Localization in a Pumped-Storage Turbine was accepted (Track 1) at Viennahydro 2026, the 23rd International Seminar on Hydropower Plants, organized by TU Wien’s Institute of Energy Systems and Thermodynamics. Now in its 23rd edition, the seminar is one of the long-standing meeting points of the European hydropower community, and takes place on 18 to 20 November 2026 at the Conference Center Laxenburg, on the grounds of the former imperial summer residence just south of Vienna.

The figure summarizes what the system does. Nine microphones and six vibration channels listen to the machine continuously; every second of sound and vibration is reduced to a small set of interpretable features. From these, the system first characterizes the machine and its state: each operating mode has an acoustic fingerprint, and what counts as normal depends on it. It then detects deviations by learning what is normal per operating point, with calibrated alarm thresholds. Finally, the microphone ring localizes the source of a deviation by direction and height, a first demonstration of source localization on this class of machine. Plant data (SCADA: speed, power, guide-vane position) only selects the operating mode and load bin; it is never a feature the score depends on, so the acoustic judgment stands on its own.

Every operating mode has its own acoustic fingerprint. The spectrum below shows the machine’s sound in four operating conditions, turbine, pump, and phase-shifter in two rotation directions, as how much energy it puts into each pitch.

The coloured lines are strikingly different: each mode produces its own pattern of sharp tones and broadband hiss. That is why there is no single “normal”, and why the system learns what healthy sounds like separately for each operating point.

The work presents results from our collaboration with illwerke vkw, Vorarlberg’s hydropower operator: the array ran on a real pumped-storage turbine, with the goal of detecting anomalies early and localizing their source before they become outages. Pumped storage is the backbone of grid flexibility for renewable energy, and condition monitoring that hears problems coming is a direct contribution to keeping it reliable. The full paper appears in the conference proceedings.