5.2.1. Analysis of the Influence of the Rotating Speed
In this section, the cavitation number
and oblique angle
are used to analyze the influence of the PJP rotating speed. The bearing force comprises the transverse force
, vertical force
, and axial force
.
Figure 12 shows the changing trend of the bearing force with different rotating speeds under the cavitation condition (N) and non-cavitation condition (Y). Under the no-cavitation conditions,
Figure 12a shows that for the different rotating speeds, the transverse force
only differs in the peak value and there is no phase difference. In contrast, under the cavitation condition, there is not only a difference in the peak values of the transverse force but also a certain phase difference. Moreover, the fluctuation range of the transverse force increases with the rotating speed, and the phase simultaneously shifts to the left. As shown in
Figure 12b, the change law of the vertical force
with the rotating speed is similar to that of the transverse force. The difference between the transverse and vertical forces is only a certain peak phase difference under the no-cavitation and cavitation conditions. The axial force
still exhibits good periodicity under the two conditions (N and Y) regardless of the rotation speed, which is verified by the pressure cloud in
Figure 13. The fluctuation range of the axial force under the cavitation condition is relatively smaller than that under the non-cavitation condition. With increasing rotating speeds, the axial force first increases and then decreases after the rotating speed reaches 3200 rpm. The conclusion is consistent with
Figure 10a.
Figure 13 shows the water vapor phase volume fraction cloud of the rotor and stator blades at different rotating speeds. The figure clearly shows that the cavitation area of the rotor and stator blades increases with increasing rotating speeds, and the cavitation phenomenon basically disappears for low rotating speeds
R = 2000 and 2600 rpm. For a low rotating speed (such as 3200, 2600, and 2000 rpm), the rotation of the rotor blades affords discontinuities when the fluid passes through the leading edges of the rotor blades. The upper end of the guide edge yields a tip vortex, which causes the local low pressure and cavitation phenomenon (as shown in
Figure 13c–e). The fluid flowing through the rotor blades significantly accelerates at high rotating speeds (such as 3600 and 4000 rpm), where the pressure significantly drops, and it can afford a large cavitation area (as shown in
Figure 13a,b), especially at the rotor blade tip. Furthermore, the cavitation phenomenon occurs at the suction surface of the stator blades, which occurs due to the influence of the rotor blades. The cavitation area and intensity both increase with increasing rotating speeds, which can further decrease the total thrust of the PJP.
Figure 14 shows the pressure cloud of the rotor domain at different rotating speeds when the oblique angle
. The pressure distributions on the rotor disc surface at different rotating speeds exhibit obvious differences. The overall pressure of the fluid environment on the rotor disc surface continuously decreases with increasing rotating speed. For any rotating speed, a low-pressure area is present at the suction surface of the rotor disc surface. The low-pressure area continues to expand with increasing rotating speed. The pressure surface of the rotor is not prone to cavitation, which is in a relatively high-pressure area. Moreover, the area of the high-pressure zone decreases with increasing rotating speed. The pressure difference between the pressure and suction surface increases with the rotating speed. This verifies the changing trend of
Figure 13: the cavitation area is larger at higher rotating speeds (
R = 3600 and 4000 rpm).
The above analysis shows that the transient analysis of the PJP bearing force is an important way to analyze the PJP flow field. A PJP has a complex structure and requires numerous grids; thus, considerable calculation time and resources are required for the numerical simulation of the bearing force using the CFD method until it converges and yields a stable flow field. Thus, a fast and accurate method for predicting the transient bearing forces is extremely important. Herein, the deep neural network model CNN-Bi-LSTM was adopted to obtain the prediction results, which were compared with the CFD results.
Figure 15 shows the difference in the bearing force between the predicted value (pred) and CFD value at different rotating speeds.
Figure 15a clearly shows that the CNN-Bi-LSTM model successfully predicts the phase difference of different rotating speeds and the change trends of the transverse force
. Other than the position of the transverse force crest value, the predicted values of other positions well agree with the CFD values. This may be due to the relatively large fluctuation of the crest value at five rotating speeds, and the predicted values are within the acceptable error range. For the different rotating speeds, the differences between the predicted peak data and CFD data under lower speed conditions are relatively small compared to those at higher rotating speed conditions (3600 and 4000 rpm), which is because the fluctuations in the same time interval conditions are relatively larger than the crest value of the low rotating speeds. For the vertical force
, the change trend of the difference between the predicted value and CFD value at different rotating speeds is similar to that for the transverse force. Additionally, the number of cycles of the axial force
is equal to the number of stator blades (Qiu et al., 2020), which has a stronger periodicity than the transverse and vertical forces.
Figure 15c shows that the predicted value of the axial force affords a better accuracy than that of the transverse and vertical forces at different rotating speeds. A large error is only present at the trough at the rotating speed R = 3200 rpm. As shown in
Figure 15d, the losses of the training and testing datasets converge very quickly and the fluctuation is relatively small, which may be due to the good periodicity of the time series dataset of PJP, and the deep learning method can more easily learn time and the nonlinear relation compared to other methods. The convergence speed and training stability show that the bearing force prediction of PJP requires very little time when using a data-driven method compared to that when using the CFD method. Deep learning methods afford a certain feasibility in the prediction of the PJP bearing force considering the training time and prediction accuracy.
5.2.2. Analysis and Prediction of the Influence of the Oblique Angle and Cavitation Number
As shown in
Figure 16, the cavitation numbers
, respectively, are 1.4721, 4.4394, 7.5682, 2.9551, and 5.9225. The oblique flow angles
are
, respectively. For the same cavitation number, no phase difference exists in the transverse force peak of a single-rotor blade. The peak difference of the transverse force is relatively smaller than that of pure axial flow when
. While at a higher oblique angle of
, the peak value of the transverse force is relatively large compared to that of the lower oblique flow angle condition. For different cavitation numbers, the change trends are relatively similar to the pure axial flow condition. The phase and peak difference of the transverse force are relatively small at the lower cavitation numbers
and
. At a higher cavitation number
, the phase difference and peak value of the transverse force significantly change, where the peak value is larger than the smaller cavitation number
and
conditions. The changing trend of the vertical force
is similar to that of the transverse force at different oblique angles and cavitation numbers; only the phase exhibits a certain deviation. For the same oblique angle, the axial force can decrease with increasing cavitation number. The peak value of the axial force reaches the minimum value for the higher cavitation number
. This is similar to the changing trend of the vertical forces: the difference of the peak value is relatively small between
and
, and the peak value of the axial force reaches the maximum when
.
Herein, the rotating speed 3200 rpm is used to study the changing trends of the PJP.
Figure 17 shows the tendencies of the water vapor phase volume fraction cloud. For the same
, the cavitation area of the rotor blade gradually decreases or even disappears with increasing cavitation number, regardless of the oblique flow angle. At the lower cavitation number
, the cavitation area occupies more than half of the rotor blades area, which is extremely detrimental to the propulsion efficiency of the PJP, which is consistent with the situation mentioned in
Section 5.1 where the decline in the propulsion efficiency of PJP is very serious. The difference in the cavitation area is relatively small between
and
, and the efficiency difference is also small, as shown in
Section 5.1.
Figure 18 clearly shows that the rotor of the PJP is in a relatively low-pressure area when
, compared to that with the other two cavitation numbers, which causes the PJP to be more prone to cavitation. As the cavitation number increases, the cavitation phenomenon gradually moves to the rotor blades tip clearance.
Figure 18 shows that the pressure of the rotor blades gradually increases with the cavitation number, and the cavitation area of the rotor blades also continuously reduces with increasing pressure difference of the outlet and saturated vapor pressures. The cavitation area difference is relatively small under different oblique flow angles compared to that under different cavitation numbers. Additionally, the water vapor phase volume fraction is higher at the root of the leaf at a higher oblique angle
, signifying that the rotation of the rotor increases the instability of the PJP flow field under high oblique flow angles. This causes the disc surface of the rotor blades to easily form a low-pressure center, which is conducive to the appearance and development of the cavitation phenomenon. The above phenomenon proves the conclusion that the efficiency (shown in
Section 5.1) decreases under higher oblique flow conditions.
Figure 19 displays the difference between the predicted data and CFD data at different oblique angles and cavitation numbers. The deep neural network model and hyperparameters used here are the same as those in
Section 5.2.1. As shown in
Figure 19a, CNN-Bi-LSTM not only successfully predicts the trend of transverse force but also affords a better prediction accuracy at the three oblique angles (
). The figure clearly shows that the predicted values of the transverse force, except for that at the trough position, are basically consistent with the CFD values. This is because the fluctuation of this value at the other positions is relatively smaller than that at trough position, and the deep learning model can easily learn the relation between different moments. At the trough position, there is a bigger difference between the predicted and CFD values of the transverse force. The prediction accuracy of the oblique flow condition is higher (
) than that of the pure axial flow condition (
). This is because the trough fluctuations under diagonal flow are relatively small. However, the predicted values of the three oblique flow angles are all within the acceptable range.
Figure 19b shows the difference between the predicted and CFD values at different cavitation numbers. In
Figure 19b, the cavitation number
are 1.4721, 2.9551, 4.4394, 5.9225, and 7.5682, respectively. Overall, the deep model CNN-Bi-LSTM successfully predicts the transverse force at different cavitation numbers. In
Figure 19b, the conclusions are similar to those in
Figure 19a under the lower cavitation numbers
. Under the higher cavitation number
, the fluctuation range and change trend of the axial force exhibit a certain difference compared to those under the smaller cavitation numbers. However, the difference between the predicted and CFD values is relatively large compared to the prediction accuracy with the small cavitation number, which is within the acceptable error range.
Figure 19c,d shows the difference of the axial force
at different cavitation numbers and different oblique angles. The axial force has a good periodicity in a certain time interval and the fluctuation amplitude is relatively small compared to the transverse force. For the lower oblique angles, the predicted value well agrees with the CFD value at the peak and trough positions of the axial force. For the higher oblique angle
, the axial force at the trough exhibits a relatively large fluctuation compared to that for the lower oblique flow angles, and the difference between the predicted and CFD values is relatively large compared to that for the other two oblique flow working conditions. As shown in
Figure 19d, the axial force has similar periodicity at the five different cavitation number conditions, and the predicted values of axial force agree well with the CFD values. Based on the above conclusions, CNN-Bi-LSTM has good generalization ability and can accurately predict the bearing force of a PJP at different oblique flow angles and different cavitation numbers.