Learn how one platform helped generate numerous design variations automatically at approximately one minute per case.
The automotive industry is rapidly changing. New competitors are emerging with the innovation and the agility needed to deliver on these trends, from software-defined vehicles, to AI-driven advanced driver-assistance systems, to hybrid technologies against significantly compressed timelines.
What does this mean for legacy automakers? For Honda, the challenge isn’t just about building vehicles, but building them faster while fulfilling requirements for quality, performance, and appeal the brand is known for.
That effort has led to a growing focus on simulation, computational fluid dynamics (CFD), and artificial intelligence. By applying AI-powered geometry generation through the Ansys GeomAI platform, the automaker is accelerating performance evaluation workflows and exploring vehicle designs more efficiently.
The Challenge of Balancing Performance and Design
“Our department is responsible for vehicle attribute development as it relates to customer experience,” says Kazuaki Hiwatashi, assistant chief engineer at Honda. “This includes wind noise mitigation caused by frontal grill or A-pillar designs, and mirror visibility under bad weather conditions.”
One such challenge involved A-pillar complexity. According to Honda's technical research, airflow separation around the A-pillar geometry located on either side of the windshield could potentially strengthen vortices and increase wind noise. In this instance, accurate analysis of this region quickly became an important aspect of vehicle development for Honda engineers.
Simulation Workflows Uncover New Benefits and Hidden Obstacles
At one time, exterior performance evaluations were driven primarily by physical testing involving full-scale models and wind tunnel testing. While these methods remain valuable today, they can be cumbersome and costly for OEMs pressed for time.
Honda recognized that accelerating development would require moving more performance evaluation activities into digital workflows. To do this, the engineering team decided to replace portions of traditional wind tunnel testing with CFD analysis using Ansys Fluent fluid simulation software and Ansys Discovery 3D product simulation software. Yet as simulation adoption increased, a new bottleneck emerged: computational model preparation.
Typically, engineers at Honda must evaluate many possible geometries to improve performance. Successful analysis of each geometry requires the creation of a corresponding computational model.
“We do see a significant time savings using Fluent and Discovery versus physical testing alone,” says Hiwatashi. “However, the model generation needed for every design iteration slows the process considerably. We needed to find a way to generate computational models that were suitable for engineering analysis faster.”
Accelerating Model Creation with the Ansys GeomAI Platform
In this case, the GeomAI platform was the ideal solution for generating large numbers of computational models rapidly to deliver dramatic efficiency enhancement for geometry-specific model creation at Honda.
Using the GeomAI platform empowers engineers and designers to generate and explore innovative geometrical design ideas with AI-trained models. It does so by observing prior designs to produce an AI model that can then generate new designs inspired by the concepts included in the training dataset.
The study used only four training geometries representing different combinations of A-pillar width and gap dimensions. From this limited dataset, the GeomAI platform could then generate numerous new design variations automatically without model inconsistency.
For the engineers involved in the study, the benefits of using the GeomAI platform were clear.
“Using a GeomAI-enabled workflow made it possible to quickly populate a large design space with AI-generated geometry variants,” says Ryuta Yonemitsu, staff engineer at Honda. “More opportunities for broader design exploration emerged than would be practical through manual modeling alone.”
Looking beyond specific training examples, it quickly became obvious during proof-of-concept testing that the GeomAI platform could generate shapes beyond the specific examples used in training.
“During our evaluation of the platform, the successful generation of designs in extrapolation regions revealed something else,” says Yonemitsu. “And that was the ability to create useful geometry variations outside the original training set.”
Measuring the Impact
In one A-pillar study, engineers previously required approximately half a day of manual effort to prepare a single model for analysis. Although the GeomAI platform requires an initial training dataset, once the model is trained, new geometry variants can be generated in approximately one minute.
“For a CFD study involving 10 different design alternatives, the traditional workflow required approximately five days of manual work,” says Yonemitsu. “Using GeomAI, we reduced the effort to approximately two days, even when accounting for model training time.”
A potential for platform use post-training was also discovered.
“Once training is complete, additional model generation requires virtually no incremental effort,” says Yonemitsu. “We can continue to explore optimized designs and perform additional validation studies without being constrained by the workload associated with model preparation.”
As for the future, Honda plans to expand its use of the GeomAI platform.
“Currently, we are in the proof of concept (POC) stage,” says Hiwatashi. “By streamlining model creation, we have been able to shift engineers’ time from simple repeatable tasks to detailed flow field analysis and design optimization. We believe this will allow us to make rational proposals based on a deeper and more accurate understanding of the phenomenon, ultimately leading to our goal of achieving a high level of balance between vehicle design and performance.”
Learn more about the GeomAI platform.
