CAD-IT, the Korean partner of Neural Concept, ran a multi-objective optimisation of a car's side view mirror for an automotive customer. The platform minimised both lift and drag while holding the mirror's surface unchanged, training on an existing dataset of Star-CCM+ CFD simulations to predict the surface pressure field at a given speed.
Key takeaways
Two objectives were minimised at once, lift and drag, under a fixed surface constraint.
The training data already existed as Star-CCM+ CFD simulations of different mirror designs.
The model predicts the surface pressure field at a given speed rather than a single scalar value.
The work was delivered through CAD-IT, a Korean partner and reseller, on behalf of an automotive customer.
Neural Concept's predictive models and powerful multi-objective optimization features can be used to find highly innovative and performant designs. Our Korean partners and resellers from CAD-IT used Neural Concept to run a multi-objective optimization of a car's side view mirror design, on behalf of an automotive customer. The platform minimized both lift and drag forces while ensuring that the mirror's surface stays unchanged.
At the beginning of the project we had at our disposal a pre-existing dataset of Star CCM + CFD simulations of different mirror designs, that we used to train a model to predict the surface pressure field at a given speed.
Difference between Fixed Volume and Optimized Volume
This dataset was created in a previous phase of the project, and it was made with an old parametrization that was inevitably changing the mirror's surface. However, thanks to the uniqueness of Neural Concept's predictive models, the free morphing used by Neural Concept's optimization algorithm can be totally unrelated to the parametrizations used to generate the training dataset of the predictive model. This made it possible for us to explore totally new, disruptive design spaces, in a matter of a few minutes, without the need of launching any extra simulation.
Once the predictive model was trained, it was used by a genetic algorithm for the design's optimization to obtain the pressure fields of each geometry, which was then integrated to obtain the final objectives: the drag and lift forces. The results are shown in the following figures:
Pareto FrontDifference between Base Geometry and Morphed Geometry
