travelling microbubbles alone could account for much of the observed be- 

 havior. We reproduce here in Figure 39 his summary results using the 

 water-tunnel derived estimates of microbubble populations. Plainly al- 

 though much remains to be learned about the cavitating environment of these 

 flows, bubbly cavitation is evidently an important factor in them. From 

 all of this work, it was clear that new experimental methods had to be 

 devised to measure the transfer function of a pump. 



Experiments in Unsteady Pump Dynamics 



There has been growing concern by many workers with the proper 

 representation of unsteady pump characteristics. We may mention, for 

 example, the work of Kolesnikov and' Kinelev (1973), Fanelli (1972), and 

 Black and Santon (1975) , who are all concerned in various respects with 

 this problem. Experimental work is comparatively rare, however. Among the 

 first of these is that of Ohashi (1968) and later Anderson et al. (1971), 

 who carried out experiments on oscillating flow in fully wetted pumps. 

 There, in principle, only the term z needs to be measured; this, of 

 course, greatly simplified measurement demands. Rather similar problems 

 have encroached into related fields of naval hydrodynamics, namely the re- 

 sponse of lift fans of surface effect vehicles to periodic disturbances. 

 There too, a transfer function approach is taken (Durkin and Luehr 1978). 

 Durkin and i_,uehr did not carry out measurements of the transfer function 

 itself, but based their conclusions, as did earlier workers, on lumped 

 parameter system model tests. It would appear that the first direct 

 attempt to measure the transfer function of a cavitating pump system is 

 due to Ng (1976). The problem, as described there, is to determine the 

 eight unknown coefficients of the transfer matrix [z] of Equation (39) from 

 measurements of fluctuating inlet and outlet pressures and volumetric flow 

 rates. Any given hydraulic system with a fixed method of excitation will 

 produce a unique relation between input and output quantities, thereby 

 providing insufficient data to determine [z]. The solution devised by Ng 

 was to provide two sources of excitation to a test loop, these sources 

 being locked in phase, but separated by an isolation section. Then, with 



83 



