Now we may ask what we have learned behind such characterization of
stall precursors. One of the questions is under what condition which type of
stall inception may happen. The first attempt to study this question was given
by Day [16] linking the type of stall inception with rotor blade tip clearances.
Although no parametric relation could be built, the tip clearance did influence
the type of stall inception presumably by altering the blade passage flow in the
tip region. In further research along the same line, based on the experimental
results, a simple model of critical incidence was proposed by Camp and Day [22].
The model states that in the throttling process of compressor, if the peak of the
overall characteristic is reached before the critical value of rotor incidence is
exceeded anywhere in the compressor, then modal oscillations will occur. If, on
the other hand, the critical incidence is exceeded before the peak of compressor
characteristic, then spikes will occur and lead the compressor to stall before
any chance of modal activity. This model, by showing the interplay of system
mode and localized event in blade passages, could successfully link the physical
difference of stall precursors with the flow parameter in compressor.
But this is not the end of the answer. A question remains of whether any
events preceding the stall precursors exist, and if they do, what is the influential
mechanism for their existence. In fact, Tryfonidis et al. had tried this question
in their early study on pre-stall behavior in compressors [23]. They distinguished
the whole stall events into three stages — the fully developed rotating stall, stall
inception and so called pre-stall. While the stall inception, as described above, was
attributed to the interrelated system and blade flow effects, the pre-stall, according
to the authors of Ref. [23], was susceptible to excitation by geometric non-
uniformities in the compressor. However this idea was not clarified in Ref. [23]
and the evidences shown in pre-stall stage were limited to the type of travelling
wave, in spite that the wave can be detect at least 100-200 rotor revolutions prior
to the inception of rotating stall.
It was perhaps because of the general lack of more effective technique then
to treat the measured data, and thus prevented from the further study of the pre-
stall behavior. Indeed, to identify the pre-stall disturbances out of the weak and
unsteady signal is difficult because they are blended within the blade passage
flow and measurement noise. Furthermore, we want to not only detect them,
but also to track their development with time. This is especially complicate for
the unsteady and nonlinear spiky type signals. Several authors tried different
techniques to this problem. Bright et al. [24] and Lin [25] applied nonlinear
time series analysis and Liao et al. [26] introduced the wavelet transform to the
stall inception analysis. The initial motivation of using wavelet transform was
mainly due to its auto-focusing capability to adapt the large frequency range
for the signal under consideration, and thus the time-frequency analysis can be
performed. As this work went on, more advantages of using wavelet transform
were explored [27–30] and more applications were obtained in stall inception
analysis [31, 32]. Early flow disturbances can be detected over thousand rotor
revolutions prior to stall, the behavior of disturbances in different scales can be
distilled, and the whole evolution process to stall can be tracked in both time and
space.
In one example taken from the test on a low-speed three-stage axial
compressor [30], the time and frequency plots of wavelet power spectrum for
three different time intervals along the pre-stall process are shown in Figs. 1, (
a
)–
(
c
). In these figures, the signals processed are taken from seven pressure sensors
116 ISSN 0236-3941. Вестник МГТУ им. Н.Э. Баумана. Сер. “Машиностроение”. 2006. № 2
