1. Introduction
Axial waterjets represent a widely adopted propulsion solution for high-speed marine vessels, offering a combination of efficiency, maneuverability, and reduced appendage drag compared to conventional propellers. They are particularly advantageous in applications where shallow draft, high-speed operation, and rapid maneuverability are essential, such as military vessels, high-speed ferries, and rescue boats. In recent years, much attention has been paid to this type of unconventional propulsors and to pumpjets, as they use the very similar concept of combining a rotor with one or more stator stages to recover the energy losses of the rotor wake. The most obvious difference is only the presence of a nozzle (accelerating or decelerating, depending on the type of application) in the case of pumpjets (which makes them more similar to ducted propellers), whereas axial waterjets are mainly designed to operate inside a converging pipe.
Experimental and numerical studies of the very complex phenomena occurring for this type of device evolved hand in hand. Beyond general cavitation evolution, research has primarily focused on blade Tip Leakage Vortex (TLV, originating in the gap between the rotor blade tip and the internal surface of the pipe) and the resulting TLV cavitation. Zhang et al. [
1] conducted experimental investigations into TLV cavitation and the periodic collapse and interactions of the TLV-induced suction-side-perpendicular cavitating vortex with the cloud cavitation at the trailing edge of the sheet cavity at the blade tip also observed in [
2,
3,
4,
5,
6,
7]. This interaction, which causes the re-orientation of the cloud cavitation in a direction perpendicular to the suction side of the rotor and the downstream convection of the resulting vapor structures through the blade passage (the so-called “perpendicular cavitating vortices” of [
7]) is a very peculiar cavitating feature of axial waterjets. Turbulent flows within an axial waterjet pump have been analyzed using stereoscopic particle image velocimetry measurements on both meridional planes and on the leakage vortex region [
6,
8,
9,
10] while Tan et al. [
7] also experimentally investigated the pump performance breakdown in relation to large-scale cavitating vortical structures, concluding that interactions between TLV and cloud cavitation on the blade suction side of the neighboring rotor blades are the primary cause of performance degradation.
Numerical analyses were used to replicate and complement experimental observations. Lindau et al. [
11] and Schroeder et al. [
12] carried out RANSE calculations to investigate the performance reduction, observing that choking of axial waterjets was related to cavity bubbles extended up to the trailing edge of the rotor blades. From the same authors [
13], further studies were carried out to model the cavitating flow by employing Transport Equation Models (TEMs) and the homogeneous mixture approach, combined iteratively with axisymmetric body force distributions (the “powering iteration”), to account for the blockage effect of blades not directly included in the calculations. Avanzi et al. [
14] tuned the Zwart mass transfer model for cavitation and studied its interactions with several turbulence models (1-, 2-, and 4-equations models) within RANSE to identify an optimal setup to predict thrust breakdown. Huang et al. [
15] similarly adopted the well-established k-omega Shear Stress Transport (STT) turbulence model and the Zwart mass transfer model to predict the unsteady cavitation when the axial waterjet of their analyses was subjected to the non-uniform inflow represented by the boundary layer of the hull after its interaction with the pipe at inlet. Guo et at. [
16] modified the Zwart cavitation model and performed investigations on the tip leakage vortex evolution and its influence on induced pressure pulses. However, since experiments of Wu et al. [
5,
6,
9,
10] have shown the anisotropic nature of the TVL, which is controlled by several interacting shear layers, most of the numerical investigations aimed at accurately characterizing the dynamics of tip vortices inside axial waterjets addressed the problem by using Large Eddy Simulations (LES). Arabnejad et al. [
17] proposed LES analyses of the AxWJ-2 axial waterjet pump [
18] using subgrid-scale models and transport equations for the cavitation modeling from the OpenFOAM library. Han et al. [
19] studied the cavitation of the tip vortex and the mutual influence with the tip vortex dynamic, while Liu et al. [
20] investigated the spatio–temporal evolution of cavitating vortical structures at the tip of axial pumps employing LES with the wall-adapting local eddy (WALE) subgrid model and the Burger vortex cavitation model (BVCM), based on the Burgers vortex, to simulate the tip leakage vortex cavity [
21]. Hybrid models (i.e., DES, Detached Eddy Simulations, and its variants) were employed as well, with the specific aim of analyzing the wake and tip vortex dynamics, their destabilization, and their interactions with the stator blades. It was mainly RIM-driven thrusters (with their peculiarities attributed to the completely sealed gap at the blade tip) and pumpjets, in Rotor/Stator and Stator/Rotor configurations further than conventional propellers [
22,
23,
24,
25,
26,
27,
28,
29,
30], that underwent these systematic analyses showing promising results at a fraction of the computational cost of LES calculations.
In most cases, the object of experimental and numerical investigations was the AxWJ-2 propulsor [
18]. In the literature, indeed, there is a lack of usable geometries of axial waterjets and that of the AxWJ-2, thanks to the availability of several experiments including cavitation observations and flow visualizations using non-intrusive laser-based techniques, that could serve as the preferred candidate for the validation of CFD methods. The scarcity of available geometries for testing and validation depends on the common applications, which are of military nature, this type of device had over the years. Only recently, indeed, were axial waterjets exploited for commercial activities, before being extensively studied for this scope. For the same reasons, design methods are not fully established, and sometimes still rely on empirical correlations, simplified potential flow models, and low-fidelity simulations. The AxWJ-2 waterjet, for instance, was designed itself thanks to a combination of different tools. As described in [
18], the waterjet was the result of several steps: a preliminary parametric study, using concentrate parameters models and empirical correlation to assess the efficiency, iterations between a vortex-lattice blade design code and an Euler solver to define the shape of the rotor and stator blades, plus some manual iterations, based on full RANSE analyses of intermediate geometries, to adjust the input values (i.e., advance coefficient) of the design process based on vortex-lattice plus Euler solver to match the absorbed torques (RANSE and vortex-lattice) and ensure shock-free entry of the blades. This design strategy is mainly based on the procedure developed by Taylor et al. [
31] and Kerwin et al. [
32], who employed a lifting-surface design code combined with RANSE or Euler flow predictions of the flow in the passage realized by the hub and the casing of the pipe [
33]. Other potential-based approaches (i.e., Boundary Element Methods, BEM), like PROPCAV [
34,
35] were adapted as well to deal with axial waterjets thanks to the axial flow predictions provided by finite-volume methodologies and provide slightly more accurate predictions of pressure distributions over the blades, which are potentially usable in place of vortex-lattice methods (VLMs) in the design process. However, while these methods have been successful in providing reliable designs, they often fail to fully exploit the potential performance gains achievable through more advanced numerical optimization techniques.
One of the main limitations of conventional design methodologies is their reliance on pre-defined geometric parameters and semi-empirical performance predictions, which may not fully capture the complex flow physics governing axial waterjet functioning. Additionally, the use of low-fidelity CFD models often results in suboptimal performance predictions, particularly when dealing with complex flow structures such as tip leakage vortices, secondary flow interactions, and cavitation inception. These limitations hinder the ability to explore a broader design space, thereby limiting opportunities for efficiency improvements. Moreover, conventional approaches may not adequately account for variations in operating conditions [
36], leading to designs that perform well under nominal design conditions but suffer from reduced efficiency and increased cavitation risk in off-design scenarios.
To overcome these challenges, the application of optimization-based design methodologies has gained increasing attention in recent years. By leveraging more accurate flow solvers, hydrodynamic shape optimization enables the systematic exploration of a wide range of geometrical configurations, leading to performance enhancements that would be difficult to achieve through traditional approaches. This is the case for conventional propeller design [
36,
37,
38], which was addressed in recent years using Simulation-Based Design Optimization methodologies based on BEM calculations in place of traditional lifting-line and lifting-surface codes. The same can be said for unconventional propulsors and devices [
24,
39,
40] which, for the complexity of the involved hydrodynamic phenomena, were designed for exploiting RANS-based optimization techniques. Simulation-Based Design Optimization allows for the identification of non-intuitive design improvements by coupling advanced optimization algorithms with detailed hydrodynamic analyses. The higher fidelity of the flow solvers employed for the characterization of the key performance indicators of the design allows for a widening of the design space beyond limits usually forced by the inherent limitations of low-fidelity and simplified design methods. This usually is the most effective action towards better performance, offering new opportunities to successfully exploit improved propulsive performance, flow features, and interactions not seen or roughly approximated by low-fidelity approaches. Furthermore, optimization techniques can be employed to balance trade-offs between multiple performance criteria, such as maximizing efficiency while minimizing cavitation risk and structural loads. The integration of surrogate modeling, machine learning-based design space exploration, multi-objective optimization strategies, and space dimensionality reduction methodologies in the design process, made possible by the optimization paradigm, represents a step further towards improved and affordable designs. Especially for unconventional propulsors, for which the computational cost of RANSE-based design-by-optimization strategies is prohibitive, several surrogate-aided frameworks were developed. RIM-driven thrusters, in this respect, were extensively analyzed and designed by combining high-fidelity RANSE analyses and response surface methods [
41,
42], albeit applied to simplified parametric descriptions (through the expanded area ratio and average pitch) of the blade geometry. Adjoint optimization [
43] was used as well to try reducing the computational cost of achieving convergence when using pure metaheuristic, genetic algorithms. Machine learning approaches—including online interactions with the designer [
44]—and surrogate models at different levels of fidelity [
45] for propeller and energy-saving device designs [
46] provided, once included in the optimization process, interesting design trends and guidelines for further- and fully resolved investigations of the optimal geometries. Space dimensionality reduction tools, like those developed in [
47,
48], provided further assistance towards the reduction of the computational cost to achieve convergence in a multi-objective design problem by identifying a priori the most relevant geometrical features to be included in the exploration and exploitation of the design space of the optimization.
In this context, the paper presents an optimization-driven approach to the design of axial waterjets for marine applications, demonstrating how high-fidelity RANSE CFD can be integrated into a rigorous optimization framework to enhance the performance of the propulsor. To this aim, a geometry, derived from the AxWJ-2 waterjet, is used as a reference for the identification of the design space and the baseline performances. The optimization-based design, using some simplifications of the hydrodynamic problem (involving the use of cavitation inception criteria in place of truly cavitating analyses, mixing plane model in place of fully unsteady rotor/stator interactions, needed for the efficient use of RANSE analyses in a design process that would require thousands of CFD calculations) is carried out with the aim of reducing cavitation and simultaneously increasing (or not excessively worsening) the propulsive efficiency. Detailed CFD analyses, using the cavitation model of Schnerr and Sauer [
49], are finally employed to better characterize the performances of a set of optimal geometries identified, by different balancing of the design objectives, on the Pareto front of the optimization. The results highlight the advantages and the flexibility of this methodology over conventional design techniques, which has the potential for delivering more efficient and high-performance waterjet propulsion systems, emphasizing its feasibility for real-world applications.
This discussion is organized as follows. In
Section 2, the test case is described and the approximated geometry of the AxWJ-2 waterjet is presented.
Section 3 introduces the numerical models adopted for the analyses and the grid sensitivity studies needed to identify the most computationally efficient setup for the systematic calculations of the optimization process presented in
Section 4. In addition, the results of the initial geometry (pressure distributions and cavity extensions) are introduced to provide a reference for the design. The optimization problem (parametrization and objectives/constraints of the design) is introduced in
Section 4 while results and detailed analyses of the most promising optimized configurations are provided in
Section 5.
2. Test Case
Figure 1 shows the geometry of the axial waterjet used as a reference for current design activity. The geometry is a variant of the AxWJ-2 waterjet [
18] since the original documents reported only the main geometrical data (chord, pitch, sectional hydrofoil shape at few radial positions) but not all the details regarding the conical/streamline adaptive surfaces used for the 3D-shaping of the rotor and stator blades.
Based on available data and comparison with 3D views, the geometry was reconstructed as similar as possible to that of the AxWJ-2 waterjet. In any case, since it serves only for the shape parametrization and the definition of the design space of the optimization process, such minor variations with respect to the original geometry seem irrelevant to the scope of the work. The comparison with the experimental measurements of the real AxWJ-2 [
50,
51], shown in
Section 3.4, reveals some differences in predicted head and absorbed power (both slightly overestimated for the reconstructed waterjet) but very similar values of efficiency that, in the end, confirm the overall reliability of the reconstructed geometry for the scope of the design activity. As the original AxWJ-2, the reference waterjet consists of a six-blade rotor (diameter
DR = 0.3048 m) housed inside a cylindrical casing with a tip clearance of 0.5 mm (internal pipe diameter
D1 = 0.3058 m). The nozzle, which has the shape of a revolved spline curve, houses a rectifier stator made of eight blades and has a throat diameter (
D2) of 0.2134 m. The internal diameter of the outlet pipe (
D3) is equal to 0.315 m. A sketched view of the pipe/nozzle arrangement [
7] is given in
Figure 2, which also shows the locations of the pressure transducers used to measure the head of the pump.
Since waterjets are closer to turbomachinery than to propellers, during experiments, performances have been collected [
50,
51] using turbomachinery conventions and quantities. The functioning condition is controlled by the non-dimensional flow coefficient
Q* which depends on the volumetric flow rate
QJ. Key performance indicators of the waterjets are the head coefficient
H*, the power coefficient
P*, and the hydraulic efficiency
η:
where
n is the rate or revolution (equal to 900 rpm, as in the experiments of [
7]),
Q is the absorbed torque, ρ is the fluid density, and
pt3 and
pt6 are the total pressures measured at the inlet and outlet reference planes.
5. Results
Results of the design are given in
Figure 16. The design space was sampled with 600 configurations using Uniform Latin Hypercube. They evolved 20 times using a genetic process (MOGA-II [
56]) to deal with the complex non-convex and non-linear nature of objectives and constraints. Of the 12,000 configurations analyzed, only those satisfying the design constraint and not subjected to rotor or stator face cavitation (precautionally assuming a cavitation index equal to 3, i.e., −
CPNface > 3) were considered for the final selection. To ease the visualization of the Pareto front, suction side inceptions were merged, and the diagram shows the worst between tip and root.
The identification of the best trade-off configurations in the multi-dimensional space of solutions is not obvious and several criteria, also based on previous experiences or expected cavitating behaviors given by peculiar pressure distributions, could be employed as usual in the case of conventional propellers. Since the focus is on the reliability and the flexibility of this design-by-optimization approach, for well-consolidated conventional propellers, when applied to waterjet propulsors rather than (or not only) the identification of the best overall geometry, different balances between objectives were accepted for the selection of the optimal candidates.
This allows for a critical comparison among different designs and, through detailed cavitating analyses, for a discussion on the reliability of the simplified criteria employed throughout the optimization process. To this end, four geometries were considered. All complied differently with the design goals (i.e., minimum cavitation inception, maximum efficiency, or intermediate conditions) and provided a general overview of what was achievable in the given design space. The selection criteria were mainly based on cavitation avoidance, looking for geometries with the minimum overall inception; at fixed inception, those with maximum efficiency are preferred. Selected waterjets, shown in
Figure 17 and
Figure 18, are:
WJ-7801, which provides the lowest possible inception (σN = 5.3) among all the feasible configurations. Without further observations, it can be seen as the overall best geometry. Inception at σN = 5.3 occurs on the suction side, at the tip of the rotor blade, in the midchord region. Suction side, leading-edge cavitation at rotor blade root is the second cavitation phenomenon in order of appearance (σN = 4.4), followed by midchord cavitation at the root (σN = 4.3) and leading-edge, suction side cavitation at the blade tip (σN = 3.9). On the stator blade, the first occurrence is midchord cavitation on the suction side (σN = 4.7), while on the pressure side, the risk of cavitation inception at the leading edge is postponed to σN < 2.5. At the trailing edge of the blade (pressure side) there is a zone of low pressure due to the contraction of the pipe. Efficiency is about 0.87, then slightly lower than the reference.
WJ-10128 is the second overall best geometry. At the cost of a slightly anticipated cavitation inception, this geometry occurs at the leading edge of the tip zone on the suction side of the rotor (σN = 5.5), with the hydraulic efficiency increasing to 0.89. With respect to WJ-7801, WJ-10128 seems to have better-balanced behavior. After inception at the tip, midchord cavitation at the tip (σN = 5.2), followed by leading edge and midchord cavitation at the root (σN = 4.4), characterize this geometry. Cavitation on the stator blades is substantially postponed with respect to WJ-7801, and the first occurrence is midchord cavitation for a cavitation index of 3.95. There is a slightly higher risk of pressure side cavitation on the stator blades since the inception cavitation index is equal to 3 at the leading edge.
WJ-9400 is the geometry with the highest efficiency (0.9) when accepting a cavitation index equal to 6. The first occurrence of cavitation is midchord at the tip of the rotor (σN = 5.95), after which the cavitation inception occurs at the root of the blade (σN = 4.9 both at leading edge and midchord) and at the tip (leading edge, σN = 4.4). For this geometry, there is a slightly higher risk of pressure side cavitation of the rotor (σN about equal to 5) corresponding to a possible local inversion of the angle of attack at the very tip of the blade. Stator cavitation is a bit better than WJ-7801 since inception at the midchord of the blade is predicted at a cavitation index of 4.4.
WJ-4837 is the last geometry that underwent extensive analyses. The selection criteria include the acceptance of a very anticipated inception at the leading edge of the blade tip (σN = 7) to favor the delay of other cavitating phenomena. Cavitation at the blade root is postponed at σN = 4; at the tip, midchord inception starts at a cavitation index of 4.8 (the lowest inception, among the selected geometries), pressure side is always avoided. On the stator blades, suction side midchord cavitation occurs quite early since the inception is for σN = 5.2.
To have a broader view of the optimization results, these four geometries underwent detailed analyses, including not only open-water performance (
Q* from 0.65 to 1.05) but also cavitation predictions using the cavitation model for a direct comparison with the reference geometry results of
Figure 13 and
Figure 14.
The open-water performances of
Figure 19 highlight how the relatively small modifications (given the constraint on the head coefficient) of the geometry at the design condition (
Q* = 0.85) can have heavy influences in the off-design functioning conditions. WJ-7801 and WJ-10128 are the geometries with the head coefficient at a design closer to that of the reference geometry. The trend of the head, except for very high flow rates, is very similar to the original one and not that of the absorbed power and, consequently, the hydraulic efficiency: WJ-7801 is extremely close to the reference geometry in the entire functioning range; WJ-10128 provides the same (or slightly higher) head at a much higher efficiency at any functioning condition. Combined with the better balance of cavitation inceptions and anticipated for this waterjet only at the leading edge of the tip (
Figure 17), it is clear how this geometry is a valid and overall more efficient alternative to WJ-7801. WJ-9400 and WJ-4837 are, instead, the geometries which provide, within the assigned tolerance, the most different values of head coefficient at the design point: WJ-9400 increases the delivered head to 2.48; the delivered head of WJ-4837 is 2.2. For WJ-9400, the increase of the head is consistent over the entire range of functioning conditions and much higher than the increase in absorbed power. This leads to a tremendous increase in hydraulic efficiency, which is higher than 0.8 at any functioning. The increase of the head is coherent with the overall higher suction distributed quite uniformly over the back side of the blade of this waterjet (
Figure 17c), which determines cavitation inception at the highest midchord cavitation index (
σN = 6).
Results of the cavitation model (
Figure 20 and
Figure 21) obviously closely resemble the behavior discussed on the basis of pressure distributions, adding additional information of the extension, and not only the inception, of the cavitation bubbles. Two flow rates (the design condition and
Q* = 0.75) and several cavitation indexes (4, 5, and 6 for the design condition, 4 and 5 only for the loaded functioning) are considered. At the design functioning, rotor blades are almost free of cavitation for cavitation indexes higher or equal to 5, which is a substantial improvement when compared to the well-developed leading-edge bubble observed for the original waterjet. The only appreciable phenomenon (except WJ-4837) is the presence of vapor in the gap between the rotor tip and the casing. This occurrence, associated with a cavitating leakage vortex that is underpredicted in these analyses due to the insufficient grid resolution at tip [
17,
19], is consistent with the pressure distributions of the modified geometries, which are very close to shock-free entry with most of the sectional load uniformly distributed at midchord. WJ-9400 has the thickest bubble in that zone, in accordance with the earlier inception at midchord (
σN = 5.95) predicted by non-cavitating analyses.
The absence of cavitation in the gap for WJ-4837 is itself consistent with the selection criteria of this geometry. Among the four candidates, inception at midchord is the lowest (about 4) at the cost of a leading-edge risk of cavitation substantially higher (
σN = 7). This is clearly reflected in the vapor bubble, which is indeed very small, at the leading edge, which is the first occurrence of cavitation also based on the local pressure distribution of
Figure 17.
Also, WJ-10128 was selected with a certain risk of leading-edge cavitation (higher than the cavitation index currently discussed for the cavitating calculations) that should correspond to the presence of some vapor at the rotor blade leading edge. This occurrence, however, is not verified. The suction peak, poorly visible in
Figure 17b, is not sufficient (in intensity and local extension over the blade) to sustain a complete phase transition, and the vapor bubble, with the current threshold on the vapor fraction (0.5), is not observable (for completeness, a hint of vapor at the tip is appreciable when the vapor volume fraction is lowered to 0.1). A similar limitation of the computational model has been already discussed for the reference geometry. While calculations at
σN = 6 reveal no cavitation at all for most of the cases, except confirming the risk of WJ-9400 at midchord, the analyses at the lower cavitation index (and those of
Figure 21 at higher loading) magnify the functioning characteristics of the selected waterjets. Partially with the exception of WJ-4837 for the reasons discussed above, rotor blades have a dominant midchord cavity bubble (because of the absence of leading edge back or face suction peaks) which is a clear indicator of the shock-free functioning of the rotor.
WJ-7801, despite having the overall lowest inception criteria on the basis of which it was selected, shows some root and midchord bubbles which, on the other hand, are not observed to the same extent as that of WJ-4837. Tip cavitation from leading edge is confirmed as the most relevant phenomenon of this waterjet, also based on the “thickness” of the vapor bubble observable at Q* = 0.75 despite the poor spatial resolution of the computational grid. The resilience against leading-edge cavitation of WJ-7801 is shown as well, since also in correspondence with the loading condition, leading-edge cavitation at tip is avoided for cavitation index equal to 5. Leading-edge cavitation at the tip for WJ-10128 is only barely observable at σN = 6 (Q* = 0.85), confirming the “weakness” of the suction peak from the non-cavitating analyses adopted as the cavitation indicator in the optimization.
Stator blade cavitation obeys the trends observed from non-cavitating analyses. When the functioning cavitation index is equal to or higher than 5, no vapor bubble is observable on the suction side of the stator. Compared to the reference waterjet, however, the optimized configurations seem more sensitive to this occurrence. This is observable through the presence of root cavitation that characterizes, in increasing order of importance, WJ-10128, WJ-4867, WJ-9400, and WJ-7801 at the design functioning when the cavitation index is reduced to 4. For the latter, stator cavitation is a non-negligible midchord bubble, easily expected from the very intense suction of
Figure 18a. For the loaded condition (
Figure 21), WJ-10128 is again the less cavitating geometry, since for other waterjets, cavitation appears mainly as a large leading-edge bubble.
These results are the consequences of the geometry modifications described in
Figure 22,
Figure 23 and
Figure 24, which correlate, also through
Figure 25 and
Figure 26, with the predicted performances in ways very similar to those of conventional propellers.
The lowest pitch (P/D) at the tip of WJ-7801 explains the lowest inception at the leading edge; the highest maximum camber (fmax/c) at the intermediate radial positions of WJ-9400 and WJ-10128 (combined, respectively, with the highest and relatively high values of the pitch in the same region) determines the anticipated inception at midchord. The relationship between input (geometrical parameters) and output (performances) variables, studied through the t-Student test, is indeed normal. For efficiency, direct effects (positive values of the normalized effect, i.e., an increase of the input variable causes an increase of the output variable) are ascribable to the pitch, particularly its values at the tip through the value of the last control points (pd_3 to pd_5). On the contrary, efficiency seems negatively influenced (inverse effect) by higher values of maximum sectional camber (fy_1 to fy_3). The expanded area ratio, controlled by the scaling of the chord, has the usual inverse effect on efficiency, also ascribable in the case of waterjets in relation to the increase of the wetted area and the corresponding frictional forces. Higher values of pitch, as expected, are responsible for higher values of pressure at the tip through the inverse effect on CPN in the same figures. Cavitation inception at midchord is driven by sectional camber; as with the case of conventional propellers, highly cambered hydrofoils determine higher suction in the central part of the chord. On the face side as well, when curvature is too high, the risk of suction/pressure side inversion is higher, with the obvious consequences on pressure side cavitation inception. It is worth mentioning that the almost identical shape of the non-dimensional sectional camber lines of the optimized rotor blades (frot./fmax rot. as a function of the non-dimensional chordwise position x/c, from leading to trailing edge) that establish a reference for this type of propulsive device. This is partially ascribable to the very low effect evidenced for the parameters controlling that geometrical feature (fhx and fhy), but it has to be remembered that the test has been performed using the geometries from the optimization and not those from a completely non-correlated sampling of the design space, and this can lower the statistical reliability of the results.
Discussing the influence of casing and hub shapes of the optimized geometries is not obvious. From the t-Student, however, the shape of the casing seems to have a slightly higher influence than the hub with concern to the risk of cavitation at the tip, while for efficiency and other types of cavitation its role, as that of the hub, the results are negligible. This suggests that it is opportune to neglect these design parameters, at least for the preliminary exploration of rotor and stator shapes, thus accelerating the convergence of the process.
6. Conclusions
This study has demonstrated the effectiveness of design-by-optimization methodologies in enhancing the performance of waterjet systems. By leveraging computational fluid dynamics (CFD) simulations and advanced optimization techniques, designs with improved overall performances, including the increase of efficiency and reduction of the risk of cavitation, were identified. This was possible thanks to a simplified, but reliable and verified, computational setup built upon single-blade passage, mixing plane, and moving reference frame models, which are the fundamental assumptions for the computational efficiency needed to evaluate the thousands of different configurations required for any optimization design methodology. With a simulation time between 8 and 10 min per case, using 240 computing cores and parallelization, the 12,000 designs of the optimization were completed in less than 20 days.
Since design-by-optimization methodologies provide sets of (and not single) designs, distinguished by different weighting of the objectives, the accurate analyses of a selection of these optimal geometries allowed discussing the flexibility of the design method and the reliability, and possible limitations, of the selection criteria adopted throughout the process. The reduction of the risk of cavitation (i.e., the minimization of the maximum of the suction, that is the minimization of the inception cavitation index) proved to be, in most cases, a computationally efficient way to monitor cavitation without the burden and the cost of truly cavitating analyses. Detailed cavitating calculations confirmed the overall results of the simplified analyses but pointed out some limitations of the adopted criteria, which are mainly related to the approximated information regarding the real extension of the cavitation bubble. As shown in the paper, indeed, as a consequence of the final development of the cavity bubble (influenced by the suction peak but also by the “extension” of the region under inception conditions, i.e., with a pressure lower than the cavitation index), configurations regarded as the best-performing (WJ-7801, for instance) showed actually delayed inception but not the less-extended bubble, which instead was observed for configurations having a slightly different balance between cavitating phenomena (WJ-10128). In this respect, criteria like the “inception volume” and the “inception area” from non-cavitating analyses [
57] may help to discern among configurations those having similar inception characteristics. In addition, the “cavitating area” was monitored throughout the design process; but without a dedicate experience on this type of propulsors (that can be simply a trustable correlation between cavitation area and final vapor extension), this information was not completely exploited during the selection step.
In any case, the analysis of the set of optimized propellers using both non-cavitating and cavitating models shows the success of the design-by-optimization methodology. Rotor blades are shock-free oriented under the simultaneous influence of casing, and stator blades have an overall delayed cavitation inception and provide not-worsened to significantly improved hydraulic efficiencies, sometimes under off-design conditions as well. Following the strand of propellers and unconventional propulsor designs, future work should explore surrogate-aided design-by-optimization methodologies, machine learning techniques, and design-under-uncertainties which, in the end, may provide configurations less sensitive to perturbations of the functioning conditions at a lower computational cost.