Designing low-emissions vehicles with modeFRONTIER®, Sculptor® and AVL FIRE®: external shape aerodynamic optimization

Fig. 1 The original simplified Volvo car model with streamlines. Hot colors represent higher velocities of the flow.

External shape aerodynamic optimization plays a major role in achieving low-emissions in designing the next generation of vehicles, no matter the underlying propulsion and powertrain technologies. To tackle such a challenge, parametric fluid-dynamics simulations, driven by efficient numerical optimization algorithms, represent a key know-how. In this context, the modeFRONTIER multi-objective optimization tool brings a user-friendly and powerful solution to designers. It can connect in the optimization loop Computational Fluid Dynamic (CFD) models, as well as mesh-deformation software such as Sculptor (“mesh” is the computational grid that is built around the geometry to perform the fluid flow simulation). Together, modeFRONTIER ® and Sculptor represent a flexible solution to optimize shapes with only few key parameters but great freedom, without involving “expensive” parametric Computer Aided Design (CAD) and mesh-generator software in the optimization loop.
This approach is applied at Chalmers University of Technology on a simplified Volvo car model in a project supported by AVL List GmbH, which provide the CFD simulation software AVL FIRE, involving pure fluid-dynamics optimization objectives (Fig. 1).
It should be also underlined that such concept is fully and easily expansible to cover the real multi-disciplinary nature of the design challenge. In fact, it could be easily completed simply by plugging into the modeFRONTIER “optimization workflow” suitable comfort simulation (i.e. handling, cross-wind stability, aero-acoustics, …) and aesthetic design and cost models.

As mentioned, aerodynamic shape optimization is an important element in improving cars of the future, where there is a demand for more energy-efficient and comfortable vehicles.
The project described here aims at creating an automated shape optimization process, able to optimize any geometry with respect to aerodynamic properties. Such an optimization process is always multi-objective, and often such objectives are connected in a way that improvement in one objective leads to deterioration in another. Here, two conflicting goals are considered: the vehicle lift coefficient (Cl), to be decreased for better handling performances; the drag coefficient (Cd), to be decreased for lower consumption and hence to achieve the low-emission concept.

Fig. 2 - The modeFRONTIER ® workflow integrating Sculptor and AVL FIRE®

Since the rear end of any personal car is responsible for most of the aerodynamic drag, the choice was to focus the shape modifications on such a region, while keeping the rest at the previously defined frozen-design stage. In this case, the optimization is performed on the rear end of a simplified full size car model from Volvo Cars Corporation.
To tackle such a challenge within a timeframe compatible with the ever accelerating development pace of the automotive industrial standards and environmental requirements, all the phases of this process should take advantage of the best-in-class technologies. Hence, the software used is modeFRONTIER for the “process integration” and “design optimization” part, Sculptor for mesh morphing and AVL FIRE for initial mesh creation and CFD calculations.
One need is to limit the number of considered independent parameters controlling the shape. They should be as few as possible, to speed up the optimization search. On the other hand, they should be able to generate the widest set of shapes to be explored. Sculptor’s mesh deformation technology makes the difference compared to a traditional parametric CAD approach, allowing to control the key shape-features of the vehicle’s rear end with only two parameters.
Another key factor is the efficiency of multi-objective numerical optimization algorithms, that should be able to find an optimal shape configuration out of billions of possible ones, by evaluating only a few variants. Here modeFRONTIER plays again a major role in achieving the expected improvements with its sophisticated “Evolutionary Strategy” optimization algorithm with multi-objective capability.
Last but not least, the CFD model needs an accurate tuning. When performing CFD computer simulations, huge computational power is required and proper pre-processing is necessary to keep the simulation time within reasonable limits. The simulation time ranges from days to months, and therefore a compromise must be done between the resolution of the simulation and simulation time, where the goal is to get enough information to improve the aerodynamic properties of the vehicle in as short simulation time as possible. This optimization process ran twice, initially using a ”k-ε” turbulence model, and it has been repeated with a higher-accuracy “large-eddy simulation” (LES) approach using the Smagorinsky model. The optimization with the lower-accuracy k-ε model took around 18400 CPU hours to be completed, while the LES took 46100 CPU hours (the sum of computational time over all the CPU‘s used, here around 40).

Fig. 3 – A sketch of a single step of the optimization loop, as managed by modeFRONTIER

The solution
The optimization process is fully automated by connecting the software in a closed optimization loop within modeFRONTIER® (so called “process integration”), where the two deformation parameters are controlled by the “Evolution Strategy” optimization algorithm, capable to improve both the two considered conflicting objectives simultaneously (so called “design optimization”). To do so, a symbolic representation of the process is created inside modeFRONTIER, through a block-diagram called “workflow” (see Fig. 2).

Once the “workflow” is ready and the models (Sculptor and AVL FIRE) are plugged-in, together with the specifications of hardware resources to be used, modeFRONTIER takes completely care of managing the whole process. It generates new models with Sculptor, submits the calculations to AVL FIRE, and collects back the results (see Fig. 3). The designer is involved again when the process is completed, in order to focus on the final analysis result and on the decision process for the best trade-off solution between the two
conflicting needs. With only two parameters, and through “Arbitrary Shape Deformation” (ASD), Sculptor controls the mesh and the associated geometry of the whole rear end of the car, one of the main responsible regions for its aerodynamic efficiency (see Figs. 4 and 5). Additionally, this meshdeformation approach allows to keep the CAD and the meshing software out of the optimization loop, with great benefits in terms of such software license usage and on the optimization speed itself. In fact, they’re used only twice, to create the initial computational grid (“mesh”) and to acquire the final optimum. During the whole optimization loop, instead, the mesh is directly accessed, parameterized and manipulated by Sculptor.

Fig. 4 – One-parameter Sculptor’s ASD volume in the XZ plane deforms the whole rear-end geometry and the mesh around it: original shape(up)/mesh(down) in the center, two possible configurations at left and right.

Fig. 5 – The second parameter manipulates the geometry of the rear end of the vehicle and the mesh around it, by compressing/expanding it in the X direction.

The chosen modeFRONTIER “Evolution Strategy” optimization algorithm is very efficient in searching for the global optima: it requires computation of only a few variants over the thousands possible, in order to reach the optimal design. Another crucial benefit is in its capability to generate and handle multiple configurations to be run simultaneously. This way, the available computational power and the solver software licenses are fully exploited, speeding up the overall optimization process by four times respect to any traditional “sequential” optimizer.
In fact, modeFRONTIER managed four design evaluations at the same time, sending CFD computations on a remote cluster where each single design evaluation was itself parallelized.

Fig. 6 - The original car compared to the optimal designs from the k-ε model and LES

Fig. 7. – Results from the optimization viewed as a bubble graph in modeFRONTIER design space.

Conclusions
This work shows that the automatic shape optimization loop is fully functional: modeFRONTIER is able, through its “workflow“, to link and manage the Sculptor mesh-deformation software and the CFD solver AVL FIRE.
Sculptor itself, thanks to its mesh deformation technology, allows to keep CAD and mesh generator software out of the optimization loop, sparing time and resources. In the same moment, it allowed to control the shape of the rear end of the Volvo Cars‘ vehicle model with only two parameters.

Fig. 8 - The flowfield around the original car(left) and the optimal design(right) found using k-ε turbulence model(top) and LES (bottom). The car is colored with pressure and the flow with velocity.

A “twin“ optimization has been run, featuring both a standard k-ε turbulence model and an high-accuracy LES. The last approach shows significant improvements in the CFD evaluations, but also a huge increase in computational time.
Thanks to the efficiency of the modeFRONTIER “Evolutionary Strategy“ algorithm and its parallel nature, it has been possbile to fully exploit the available hardware and software resources, and hence to complete the optimization in an acceptable timeframe even using high accuracy physical models.
This proves that optimization and mesh deformation are key-enabler techniques for the virtual multi-disciplinary design of the next generation of low emissions and high comfort vehicles.
A team at Chalmers University of Technology is pioneering this approach, with their know-how and cutting-edge software technologies.

Prof. Siniša Krajnović, Eysteinn Helgason and Haukur E. Hafsteinsson, Chalmers University of Technology, Sweden
Luca Fuligno, EnginSoft SpA, Italy