University of St. Gallen

← Theses

3D-Printed Mini Hydropower Plant that Senses Its Own Faults

Open topic · MA · Advisors: Bruno Rodrigues, Jinyuan Zhang

Motivation

Hydropower turbines emit acoustic signatures long before they fail. Cavitation bubble collapse radiates broadband energy between 6 and 25 kHz, and seal leakage produces a characteristic hiss with little accompanying structural vibration. Condition monitoring in hydropower today rests almost entirely on contact vibration sensing standardised in ISO 20816-5, which measures only the structure it is bolted to and, at a usable bandwidth of 3.7 kHz, cannot reach the higher-frequency signatures at all. Microphones capture them, and our group has shown that machine learning on turbine audio detects developing faults reliably (Khamaisi et al., 2025).

Problem statement

Detecting a fault, however, does not locate it, and an operator who knows only that a machine is degrading must still open it to find where, while unplanned downtime at a 295 MW plant costs upward of 28,000 EUR per hour. Locating a sound needs nothing more than a few microphones and an accurate clock, since sound travels at roughly 343 m/s, so a noise that starts closer to one microphone arrives there a fraction of a millisecond before it reaches the others, and each pair turns that small difference into a geometric constraint on where the noise began. Enough pairs narrow the constraints to a point, and this is what time difference of arrival (TDoA) localisation does. Inside a real turbine it is harder than the geometry suggests, because sound reflects off the housing and arrives many times over, flow noise is broadband and loud, structure-borne paths arrive ahead of the airborne signal, and any offset between node clocks is indistinguishable from a genuine delay. It remains unclear how much localisation accuracy survives them in a working machine.

The project

We will build a scaled hydropower plant, 3D printed and instrumented with a synchronised microphone array, using an open Francis turbine design with a spiral casing and an eight-blade runner (Martlul, 2020). Printing the runner is what makes the experiment tractable, because a damaged or imbalanced blade is printed rather than machined, so fault severity becomes a controlled variable and every fault sits at a position known to within a millimetre. Parts are already printed and our localisation algorithms exist as a starting point, so the project extends a running line of work that scales down our nine-microphone array at Rodundwerk II, a 295 MW pumped-storage plant operated with illwerke vkw. No background in hydropower or acoustics is required, and the hardware is provided. The specific goal of this project is to measure how accurately TDoA locates an induced fault on the rig, to determine how many microphones and which array geometry that accuracy requires, and to identify the conditions under which the approach fails.

Objectives

  • Understand TDoA localisation, GCC-PHAT delay estimation, and hyperbolic solvers.
  • Build and instrument the 3D-printed rig with a synchronised microphone array, and quantify the residual timing error and the spatial resolution it bounds.
  • Extend our existing localisation pipeline in Python (or programming language of your choice).
  • Induce controlled faults at known positions and record a labelled dataset across fault types, severities, and sessions.
  • Measure localisation error against ground truth across sessions, and derive a design guideline for array geometry.

Previous theses in this line

References

[1] Khamaisi, K., Keller, N., Krummenacher, S., Huber, V., Fässler, B., & Rodrigues, B. (2025). From Noise to Knowledge: Acoustic Anomaly Detection in Pumped-storage Hydropower Plants. arXiv:2509.22881.

[2] Zhang, J., Fässler, B., & Rodrigues, B. (2026). Multi-Modal Acoustic-Vibration Array Monitoring for Anomaly Detection and Source Localization in a Pumped-Storage Turbine. 23rd International Seminar on Hydropower Plants (ViennaHydro 2026), Vienna.

[3] Omarar, Z. (2026). Multimodal Sensing for Acoustic–Vibration Based Fault Localization in Hydropower. Master’s thesis, University of St.Gallen. https://sensing-group.com/files/theses/ma-zakaria-omarar.pdf

[4] Klindworth, L. F., & Berendt, C. W. (2026). BATradar: Acoustic Drone Localization Engine. Bachelor project, University of St.Gallen. https://sensing-group.com/files/theses/bp-loris-colin-batradar.pdf

[5] Huwyler, M. (2026). Distributed Acoustic-Vibration System for Industrial Predictive Maintenance. Master’s thesis, University of St.Gallen. https://sensing-group.com/files/theses/ma-maximilian-huwyler.pdf

[6] Knapp, C., & Carter, G. (1976). The generalized correlation method for estimation of time delay. IEEE Transactions on Acoustics, Speech, and Signal Processing, 24(4), 320-327.

[7] Martlul (2020). Francis turbine complete: Spiral hydro turbine. Printables model 234219, free STL. https://www.printables.com/model/234219-francis-turbine-complete-spiral-hydro-turbine-comp

Interested? Email the advisors: Bruno Rodrigues, Jinyuan Zhang.