Figure 1: Structured multiblock mesh for a ship hull design variant.
1525 words / 7 minutes read
RANS-based CFD computations are regularly used for studying flow fields for complex marine applications. However, they are used more on a case-by-case basis than as part of a rigorous design cycle. They are used only to simulate the flow field for a few selected final models for confirmation rather than as a day-to-day design tool. On the other hand, simpler potential flow codes are extensively used to do optimization projects. This bias towards potential flow codes for optimization is mainly because, they are easy to set up, computationally cheap and each run takes significantly less time.
Though designers prefer the accurate predictive capabilities of RANS, the higher turnaround time for geometry building, grid-generation and solver runs, often force them to grudgingly settle down for lower-order methods in optimization cycles. Thankfully, this scenario is changing for good, with the development of unique design platforms like CAESES, which enables the engineer to connect and launch all the multiple software needed for shape-optimization from one single platform and make them work in a hands-off automated environment. With this approach, RANS which is already established as the work-horse for design cycles in the aerospace and automotive field can make its assertive mark in marine shape optimization as well.
The following section presents a case study wherein CAESES has been effectively employed in developing a RANS-based fully automated ship hull design cycle.
Case Study – Ship Hull Design Cycle
The study explores the design space for three geometric variables for a twin-screw ROPAX ship using CAE software CAESES, RANS code Neptuno and grid generator GridPro. The key component in this design approach is the CAESES platform which allows easy coupling of all involved software using its integrated Software Connector. All the 3 stages of a CFD-based design cycle, namely – preprocessing (geometry building, grid-generation), solver runs, and post-processing of the simulation results can be efficiently done in the CAESES environment.
CAESES platform has on-board robust CAD capabilities and powerful optimization algorithms. The CAD options in CAESES exports the geometry of the current design variant along with parameters defining the geometry. Once the parametric model is set up, new designs can be quickly generated just by changing the parameter values. Systematic variation of the parameters in the design space can be done using many tools like Sobol, Latin Hypercube Sampling, ensemble design engine, etc, which are rightly available on CAESES platform.
Grid generator and solver plugged into the CAESES platform carry out the grid-generation and simulation as subsequent steps. Once the computations are completed, the results data can be parsed, displayed in plots, tabular forms, flow fields visualized for analyses purpose.
Parametric Modeling
The parameter-based geometry modeling is based on the philosophy that, any product/geometry in all its complexity and details can easily be described by a bunch of parameters. Here, every feature, every contour is defined by a set of parameters, and at any point in time, their values can be modified to bring in a corresponding change in shape. When coupled with optimization algorithms, parametric modeling help in the systematic reduction of the design space to the most important design variables.
The ROPAX ship hull considered in this study is built parametrically using only 23 design variables. The variables define the shape and contours of the hull including bulbous bow and skeg. Out of the 23 variables, only 3 are considered for the optimization study, namely – the length between perpendiculars, the length of parallel midship, and a parameter controlling the position of the parallel midship. The total resistance of the ship was chosen as the objective function.
In total 120 designs were generated using ensemble investigation considering 3 ship lengths, 10 lengths of parallel midship, and 4 positions of the parallel midship.
Why GridPro?
For automatic structured grid generation, GridPro was used. The concept of dynamic boundary conforming -topology adopted in GridPro enables to separate the topology from the geometry and use the same topology for different geometric variants. What it means is that, after constructing a valid topology for one variant, a structured grid can be generated easily for each new design, while maintaining the basic structure and cell distribution of the original grid. This is a much-desired feature, as it helps to maintain consistency in cell count, flow alignment which in turn reduces the dependence of the CFD computations on the mesh resolution and mesh quality.
In order to capture sharp geometric features, GridPro makes use of support surfaces that guide the block faces adjacent to sharp edges. These surfaces aid in avoiding the formation of distorted blocks or skewed cells along the feature edge. In addition, they are also used in high-resolution regions, e.g. near the free surface, to prevent the smoothing algorithm from moving cells out of this region. These support/control surfaces are parametrically generated automatically in CAESES along with the design variant and exported to GridPro.
The grid-generator is allowed to run on each design until a set quality constraint of surface and volume folds are met. However, if the quality criteria are not met even after a sufficiently high number of sweeps, the script stops the grid generator and the design is discarded.
For certain design variants, the topology was not fitting appropriately due to a large variation in design. GridPro’s command-line tools like “transformation ” were called automatically inside CAESES the topology was translated automatically based on the parameters evaluated using the parametric model.
Sufficient resolution of the turbulent boundary layer was ensured by the clustering tool with specific first cell spacing, growth rate, and number of layers.
The entire meshing process was done using a shell script that calls all the essential parameters from a Software Connector. The script contains instructions to perform all the steps from geometry import to meshing to grid export. In addition, certain pre-processing steps like block merging, topology transformation, etc. are also performed to reduce computational effort.
For each configuration with about 1090 blocks, the grid-generator is run for a fixed number of 2500 sweeps, roughly generating a grid with 800,000 cells once in every 6 minutes on a workstation.
RANS Computations
CFD computations are done using the finite volume code Neptuno, a RANS solver that is been validated extensively for many complex marine applications including seakeeping and maneuvering. For the present simulations, a two-phase level set approach is used for the water-free surface. Further, the SIMPLE method is used for pressure-velocity coupling and the standard k-ω turbulence model is chosen for turbulence modeling.
The Neptuno solver is brought into the CAESES framework by 2 software connectors. The first connector facilitates performing the preprocessing steps of GridPro grid conversion to Neptuno format and setting boundary condition values. These steps are executed in the same machine used for grid generation. The second connector enables to do CFD runs and execute postprocessing steps. The tool sshResourceManager in CAESES enables to manage parallel computations on a remote cluster from within CAESES. All runs were completed on the cluster overnight using only one core per design. Once the runs were completed, the solution data are automatically converted for further postprocessing and analysis in CAESES.
Results
In total 120 numerical computations were performed, out of which 20 failed due to the cluster program in GridPro turning itself off (which is being improved) or convergence issues.
Analysis of the results reveals a clear dependence of the non-dimensional resistance of the ship on the position of the parallel midship as can be seen in Figure 5a. The optimal position turns out to be somewhere at the midship position of 0.35. However, no clear trend was seen in the way the parameter length of the parallel midship influences ship resistance. But for the parameter length between perpendiculars, ship resistance shows an obvious reduction in magnitude with the increase in ship length as expected.
Figure 6: Wave patterns for two different designs. The left design has resistance 8% higher than the one on the left.
Figure 6 shows the wave patterns for two different designs. The design variant on the right-hand side has a resistance that is 8% higher than the one on the left-hand side.
Figure 7 shows the response surface model generated from the results. From the successful 100 computations, 16 equally distributed computations were selected to generate the response surface model.
Trailing Thoughts
RANS-based shape optimization under the CAESES umbrella looks very promising. The approach allows for an easy and fully automatic generation of a reliable surrogate model once the parametric model, the grid topology, and the numerical computations are set up. The approach can be further scaled up to include more parametric variables and can be used for more complex computations like optimization of skeg of a ship in oblique inflow conditions, etc.
Author : Sebastian Uharek
Further Reading:
- Shape Optimization for CFD-101
- Piston Bowl Design Optimization and Meshing for CFD
- Did nature or need inspire turbo-machines?
Reference
1.” Fully Automatic Design Space Exploration by RANS Computations “, Sebastian Uharek, et al, 11th International Workshop on Ship and Marine Hydrodynamic, Hamburg, Germany, September 22-25, 2019.
2. https://www.caeses.com
