Cavitation Detection in Hydraulic

Turbines
Application to a Machine Condition Monitoring System
The term cavitation is used to describe the phenomenon of liquid-to-gas and gas-to-liquid
phase changes that occur when the local fluid dynamic pressures in areas of accelerated
flow drop below the vapor pressure of the local fluid. The liquid-to-gas phase change is
akin to the boiling of water, except that it occurs at ambient temperatures. The gas-toliquid phase change produces extremely high local pressures as vapor cavities implode on
themselves. Cavitation commonly occurs in hydroelectric turbines, generally appearing
around guide vanes, wicket gates, the turbine runner, and in the draft tube. Usually,
cavitation within the fluid stream is not damaging to the turbine. However, when
implosions occur near solid boundaries within the machine, flow surfaces can be
damaged and eroded, figure 1. Damage to the runner has to be routinely repaired to
maintain the bucket profiles. If left unrepaired, the erosion damage can lead to drops in
efficiency and ultimately major damage to the rest of the machine.

Detection of the cavitation phenomenon is straight forward. Large increases in noise,

particularly in moderately high frequency ranges (15- to 100-kHz) are characteristic of
cavitation. In addition, vibration levels generally increase. However, in a machine
condition monitoring program, the simple ability to detect cavitation is not too beneficial.
The real need is to learn when the cavitation is damaging parts of the machine, figure 2.

The challenge then is to develop techniques for separating the noise and vibration caused
by cavitation damage from that caused by non-damaging cavitation and other background
noise and vibration sources. A promising approach under development in recent years is
based on the observation that when cavitation occurs in a rotating machine, the periodic
rotational components will amplitude modulate the wide-band high frequency noise
generated by collapsing cavitation bubbles. Modulation may occur due to the periodic
rotation of continuous noise sources relative to a fixed sensor, or because of periodicity in
the hydrodynamics of the flow (e.g., blades passing through wake zones downstream of
wicket gates). This amplitude modulation has spurred development of many techniques
for cavitation detection, including one that is now commercially available.

Amplitude modulated based cavitation detection techniques identify cavitation as a

hidden periodicity within an isolated band of the high frequency noise floor. Typically
the raw signal is bandpass filtered for the desired high frequency band and the discovery
of the hidden periodicity is accomplished through demodulation of the bandpassed signal
using the principal of envelope detection. Most of the cavitation detection techniques use
half- or full-wave rectification spectral analysis or true RMS detection to perform wideband demodulation on the raw measurement signals. These methods can be accomplished
either by hardware or software methods. Generally, the resulting frequencies of interest
are the unit's rotational speed and its harmonics, particularly the blade passing frequency
of the turbine runner.
In a machine condition monitoring system, a fixed sensor is desirable from a practicality
standpoint. Placement of the sensor is very important. Broadband acoustic emission
sensors or high frequency accelerometers (amplified analog output proportional to stress
wave activity or vibration), mounted on the wicket gate assembly or the turbine guide
bearing assembly have successfully detected "damaging" cavitation. Acoustic emission
sensors mounted on the shell of the draft tube have produced similar results. Either of
these techniques has deficiencies, particularly related to universal application on different
machines. Again, the cavitation phenomenon can be detected easily. However
determining the severity of the damage that may be occurring on the runner by either of
these methods is difficult. Calibrating the system with data about known cavitation zones,
duration of exposure, and extent of damage from maintenance data is possible. However,
geometric differences between similar machine types and totally different machine types
(i.e., Francis vs. Kaplan) significantly affect amplitudes produced by these fixed sensor
demodulation techniques and probably would require machine specific calibrations.
One limitation of the half- or full-wave rectification demodulation technique is that it is
affected by discrete components (machine noise) that are present and so bandpass
filtering has to take place prior to demodulation. Otherwise, these discrete frequencies
will generate peaks in the demodulated signal as well, erroneously showing hidden
periodicity.
A new technique that overcomes this limitation has been used successfully to identify
cavitation in high speed turbo pumps. This technique uses a recently discovered unique
coherent phase relationship within the wide-band noise floor of a cavitation-generated
signal. The combination of a Phase-Only filter and an Amplitude-Medium filter removes
discrete components and allows for detection of hidden periodicity generated by the
coherent phase components in the wide-band noise floor and a non-normalized spectral
function detects the strength of the cavitation generated wide-band modulated signals.
Another noticeable advantage is that no high-pass filtering or high frequency analysis is
required since the low frequency noise floor contains the wide-band modulated coherent
phase information.

Detection of cavitation damage directly by monitoring acoustic emissions can be

accomplished but is still in a testing phase. Material damage or the attack of the surface
which may lead to damage is detectable with acoustic emission sensors. The initial attack
is characteristic of an out-of-plane (OOP) source on the runner blade. However as the pit
grows in size and depth, a larger percentage of the acoustic emissions become in-plane
(IP) sources. Most of the energy in the stress waves created by OOP sources occur at
frequencies below 100 kHz. This energy is carried by a low frequency flexure wave in
the plate. The IP sources generate energy in the stress wave which is carried by high
frequency extensional and shear waves. Recent developments in sensor designs have lead
to a transducer which is equally sensitive to both OOP an IP sources. This transducer
utilizes a "false" aperture, consisting of a piezoelectric crystal that is mass loaded over a
small (typically 1/8-inch) area of the center of the crystal. The transmission of acoustic
emissions, especially IP sources, should prove no problem over a lubricated bearing due
to mode conversion at the boundary. Thus, mounting one of these dual IP-OOP sensors
on the turbine guide bearing might prove promising both in detecting the presence of
cavitation and the extent of the damage. This technique has proven successful for
detection of both IP and OOP sources in thick steel plates, however, the complex
geometries characteristic of turbine runners may prove to be too difficult to allow
application of this technique. Monitoring the ratio of IP to OOP levels will provide not
only the onset of cavitation damage but as the ratio becomes more and more dominated
by IP sources, an indication that damage is becoming more severe and may need
attention.
New techniques and sensors have made the detection of damaging cavitation in hydro
turbines a reality. There still exists some questions regarding the absolute comparisons
from unit to unit and among different turbine types. However, we are near a point where
a cavitation detection/damage monitor can be including in a machine condition
monitoring system. The use of a cavitation detection/damage module can provide both
real-time operating information as well as machine condition-based maintenance
information.

Contact: Warren Frizell email: wfrizell@do.usbr.gov.

Last reviewed: 10/25/04
Privacy | Disclaimer | Accessibility | FOIA | Information Quality | FAQ