Development of cavitation detection methodology for hydraulic turbomachines using experimental and numerical tools

Abstract

Centrifugal pumps constitute one of the most common and essential component in most industrial procedures and other technical works and applications, while hydraulic turbines operate in numerous hydroelectric power plants worldwide, producing a substantial percentage of the consumed electricity. Consequently, the reliable and efficient operation of these machines is highly required. An important hydraulic excitation mechanism that may affect the steady and dynamic operation of a hydraulic turbomachinery is cavitation, which may be created in the low static pressure zone of the rotor. The re-liquification of the vapour bubbles at downstream positions is a very violent implosion procedure that generates shock pressure waves of extreme local intensity, which can cause extended wear of the solid surfaces of the machine. Long-term operation of a machine under cavitating conditions results in significant material removal, accompanied by strong vibrations and drop of performance. The present Thesis deals with the analysis of the flow conditions in centrifugal pumps that suffer from cavitation, and aims at the development of detection tools capable of identifying the presence of vapour formations promptly. The present work attempts to approach the problem both experimentally and computationally. Two testing configurations that allow the flow visualisation of the phenomenon are developed; one at the Engineering Department of Lancaster University (LU) and one at the Laboratory of Hydraulic Turbomachines of the NTUA. In both test rigs, the regulation of the pump suction conditions (static pressure) is achieved by throttling the suction valve. The pump installed at the LU has its volute made from transparent material and uses three semi-open, radial impellers; i) one with six backward-curved blades and six splitter blades, ii) one with twelve radial blades, and iii) one with twelve forward-curved blades. On the other hand, the test configuration of the NTUA incorporates an industrial centrifugal pump with a closed impeller that has five backward-curved blades. Significant modifications are made to this pump in order to allow the visualizaton of the flow in its impeller. More specifically, the impeller is re-manufactured from scratch, with the shroud made of transparent material, while a transparent window is opened in the front casing of the pump. Performance experiments are conducted for all four impellers and they are accompanied by noise and vibration measurements, as well as flow visualizaition observations. Two accelerometers and one acoustic emission sensor are used in each test rig, and their time-series measurements are processed so as to extract cavitation characteristics and develop a detection tool of general application. The signals obtained are processed in the time and frequency domain, with the application of statistical, Root Mean Square and Power Spectral Density tools. Moreover, the Spectral Kurtosis methodology for the construction of band pass filters that unmask cavitation characteristics in the measured data is implemented and tested in the framework of the Thesis. In parallel, the fluid flow in two of the impellers is simulated under normal and two-phase flow conditions with the use of a commercial CFD algorithm. The modelling approach solves the Reynolds Averaged Navier Stokes Equations (RANS), along with a transport equation that regulates the mass transfer between the vapour and the liquid phase. The turbulence is modelled with the k-ω SST model, while the equations of the flow that correspond to the computational domain of the impeller are solved in the rotating reference frame (‘frozen’ rotor approach). The characteristic operation curves derived for each impeller of the two test configurations, exhibit the expected trend of low specific speed centrifugal pumps. During cavitating conditions, the total head value remains unchanged until the point where the vapour cavity starts to block the flowpath and finally, makes the impeller unable to provide power to the working medium (head break down). The flow visualisation data in the LU impellers depict the inception of the cavitation phenomenon and the progressive increase of its extent, as the available suction head level of the test rig decreases. Moreover, the photos obtained from the impellers’ flow field, feature the existence of different large-scale cavitating formations, such as travelling bubble, cloud, tip clearance, and attached cavitation. The flow visualization data and the laboratory measurements confirm the ability of the numerical model to reproduce the total head drop curve due to vapour phase development. The model faces difficulties to identify the vapour travelling bubbles close to their visual onset, due to their strong dynamic nature. However, its ability to predict the actual size and shape of the cavities improves significantly when the attached cavitation mechanism becomes stronger, at the intermediate stages of cavitation development, and well before the total head drop. In the NTUA pump, the computational results are also used to provide an improved prediction for the visual inception point of cavitation. Consequently, the proposed modelling approach can be used for detection of cavitation, provided that both the geometrical and operational data of a pump are given. Moreover, in case of the unshrouded impeller, the numerical results reveal the complex flow pattern at the blades tip clearance region, which includes secondary flows that recirculate part of the tip flow, back to the suction side of the blade, forming the backflow phenomenon. The latter may turn to backflow tip cavitation, when the pump operates under low static pressure conditions. The noise and vibration measurements are obtained using the AE and accelerometer sensors, which are located on the body of the centrifugal pumps in both configurations. The raw noise and vibration data measured under severe cavitating conditions include peaks that are related to bubble implosion and affect their Gaussian distribution. The statistical analysis of the data associates these differentiations with the fourth statistical moment, the kurtosis parameter, which value deviates significantly from that of the normal distribution. The examination of the frequency spectrum of the noise and vibration signals illustrates the excitation of the wide ranges of the frequency spectrum. The use of RMS and powerband tools is qualified so as to integrate the power of the excited areas and plot it as function of suction conditions. The results demonstrate the successful use of these tools for the prompt detection of cavitation in the vast majority of the examined conditions. However, their difficulty to unmask the phenomenon in low loading conditions of the closed impeller, along with the risk to detect a different kind of impulsive fault (like faults related with rolling-element bearings) instead of cavitation, makes necessary the further analysis of the obtained signals. For that reason, the Spectral Kurtosis signal processing method is tested to efficiently provide the band pass filter’s characteristics that aim to unmask cavitation behaviour. In addition, the filtered signals are demodulated with the use of Hilbert Transformation, so as to confirm the possible periodicity of the characteristics depicted in the filtered signals. The results prove the successful implementation of this methodology that manages to distinguish the appearance of impulses in the filtered signals, which are modulated by the blade passing frequency (BPF) component. The proposed approach specifies the analogy between the impulsive behaviour of the filtered signal and the shock pressure mechanism, and between the BPF modulation of the impulses with the shock pressure waves that occur in the rotating impeller. The latter relate the detection characteristics with the physical mechanism of cavitation wear, and can constitute a robust and reliable detection tool of general application in hydraulic turbomachinery.