AI in automotive design simulations: breaking the barriers to adoption
Key challenges slowing AI adoption — and how to overcome them
Imagine an automotive engineer optimizing a new vehicle design by running high-fidelity (detailed and realistic) simulations. Typically, running such simulations (also referred to as virtual-prototyping) for crash tests, aerodynamics, and structural integrity, takes days. Now, imagine an AI-enhanced system that cuts that time by half, while improving usability and providing better insights. Sounds like a game-changer, right? Yet, despite the potential benefits, AI adoption in engineering simulations for automotive design remains surprisingly low. Why is this so?
This article explores the key factors behind the low adoption of AI in engineering simulations, identifies the most critical of these factors, and suggests potential ways to address them. The information presented is based on recently conducted industry research (Kini, 2024) which derives information from existing literature, from real-world examples and case studies, and from interviews with experts in the field of simulations and AI. A survey of automotive industry professionals was also conducted as part of the research, to validate the gathered information and insights. Although this article is written with AI/simulation vendors, automotive engineers, and decision-makers in mind, it is relevant to anyone working in AI, product design or new technology adoption.
The promise of AI in engineering simulations
As organizations lock horns in a fiercely competitive market, the need to deliver new products with higher frequency is expected to escalate. A McKinsey/NAFEMS joint article (Ragani, Stein, Keen, & Symington, 2023) states that Automotive, Aerospace other major industries expect one-third of their revenues to come from the sale of new products — amounting to over $30 trillion in revenues over the next 5 years — highlighting the importance of an agile and fast product development process.
Virtual prototyping has shortened design cycles and sped up product development — vehicle design testing durations have reduced from months, when using only physical prototypes, to days when also employing virtual prototyping simulations (HBR, 2021). Such high-fidelity simulations though are computationally demanding and still considered time-consuming. The simulation industry has thus continued its relentless search for ways to reduce the number of simulations needed to be run, and the simulation and analysis time, to optimize a design. The advent of Artificial Intelligence (AI), and it’s rapid progress in the last decade, has provided the simulation industry with a potentially powerful lever to achieve those objectives.
AI offers solutions that can dramatically accelerate simulations, reduce computational costs, and enhance design flexibility. This includes use of Traditional AI applications for faster prediction — by using existing data, intelligent parameter selection, and smarter algorithms. Traditional AI applications also help improve outcomes by covering a broader design space. And the use of Generative AI applications, for generative design (generating entirely new concepts that designers would otherwise not have thought of), synthetic data generation and creation of virtual engineering assistants.
An overwhelming majority of industry professionals, working on engineering simulations and/or AI, feel that AI can bring great value to simulations.
But if AI is so promising and capable, most automotive companies would have already adopted it to enhance productivity and streamline workflows. Yet, this is not the case and AI adoption remains low.
Breaking the AI barrier: Why adoption stalls
Despite all the buzz, the adoption of AI for engineering simulations remains low (~20%) in production. This is despite the fact that over 90% of companies have explored or piloted AI projects. The following visual shows the percentage of companies in the various stages of AI adoption, highlighting a gap between exploration, piloting and full deployment/adoption in production.
What is keeping these companies from progressing from the exploring/piloting stage to adopting AI for production? An expansive set of possible reasons/factors were collected from existing research and expert interviews — these have been categorized into Technological, Organizational and Environmental factors, as shown below.
An industry survey was carried out to ascertain which of these factors were of primary or highest concern. The most critical factors were found to be:
- AI Integration: Concerns and complexities related to integrating AI into existing workflows — even when companies are successful in exploratory projects, practical roadblocks arise during full-scale integration. The root causes for these could range from workflow, architecture, and data incompatibilities, to a general resistance to change. Integration challenges are real — 58% of surveyed industry professionals said that bringing AI into existing simulation workflows requires major effort.
- AI Explainability: Concerns about the lack of explanation given by AI algorithms to support their predictions and outputs (the black-box problem) — engineers like to understand why a model gives a certain result, but with AI, that transparency isn’t always there. No wonder only 13% of the respondents were satisfied with the current state of AI explainability.
- AI Maturity: Concerns that AI may not be sufficiently field-proven and thus not reliable — AI is still seen by some as a tool that’s in its infancy, and not quite ready for prime time. In fact, only 17% of surveyed professionals felt that AI tools for simulations were sufficiently mature.
- Budget/Investment: Realization that sizeable investment may be needed to achieve formal and full-scale AI adoption — Budget constraints remain an issue for 43% of companies.
Factors of secondary concern (less critical, but important nonetheless) were:
- Lack of AI expertise: Concern that the skill and knowledge to understand and use AI may be lacking — a split verdict — half of the survey respondents felt that sufficient AI expertise is typically available in-house (or can easily be acquired — via training, hiring, or outsourcing), while the other half felt AI expertise is limited and difficult to acquire.
- Forthcoming government regulations: Concerns regarding possible impact of anticipated regulations for the responsible use of AI — here too almost 50% expressed concerns that forthcoming regulations may impose constraints adversely impacting their use of AI.
- Availability of usable data: Concern that although data exists, it may not be in a readily usable form — almost 33% of respondents expressed this concern — and felt that extensive data management, storage and processing systems need to be in place to support AI use.
- AI Scalability: Concern that scaling an AI application, even for the same use case (to multiple components/vehicles/sites, etc.), is difficult, and may result in limited or localized adoption — although only 26% of respondents had this concern, a high majority agreed that scalability should be an important consideration from the exploration stage itself.
Note: The level of concern of these factors was analyzed using hypothesis testing of the industry survey responses, as shown below. For each factor, a low level of concern was initially assumed as the null hypothesis (H0, blue-shaded region). This hypothesis was then tested using the survey responses. In simple terms, the null hypothesis was rejected for factors that fell within the left-most 5% of the curve (H1; red-shaded region — most negative) — meaning, it can be concluded that these are the factors of highest concern. Full details of this statistical approach can be found in (Kini, 2024).
These factors present real and daunting challenges, and have slowed AI adoption for many companies. Nevertheless, some industry leaders have taken decisive steps to successfully overcome these challenges — showing that they are not insurmountable. Audi, Volkswagen and Toyota offer valuable examples of how AI adoption can be successfully accelerated — whether through in-house or collaborative development, strategic partnerships, or tailored AI solutions.
Lessons from industry leaders
Audi developed FelGAN, an AI-based software for Rim design (Audi MediaCenter, 2022). The word ‘FelGAN’ is derived from Felge (meaning rim) and GAN, the acronym for Generative Adversarial Network. FelGAN, trained on historical rim design data, generates multiple realistic configurations efficiently, adapting to new design requirements. In the future, this system may be scaled to serve as a platform to be used by designers in other divisions of the company.
Key takeaways: By developing the capability in-house, Audi has had full control over the functionality (addressing maturity and explainability concerns) and integration of the software to meet their specific needs. Such an approach also eliminates concerns about data provenance and integrity.
Volkswagen recently announced the formation of an AI company named ‘AI Labs’ (Volkswagen Group, 2024). This company is expected to serve as a global competence center to incubate new ideas for VW group and enable easier collaboration with tech companies and partners.
Key takeaways: This set up gives VW greater control over AI development — ensuring the scalability and integration factors are scrutinized upfront — whether built in-house or in collaboration with partners.
At Toyota, use of publicly available Generative AI tools by designers had not been very effective. These tools would help generate great designs, but they would not conform to the engineering constraints that are essential for performance and safety of vehicles. Toyota Research Institute (TRI) generated a system in-house that allows these constraints to be included during the generative design process itself (Greenwood, 2023). The example they share is regarding the aerodynamic drag engineering constraint — a critical design factor for energy efficiency, that the engineering team uses simulations to optimize. This allows designers to be creative with ideas, while working within the constraints that the engineering team will need to eventually impose anyway. This leads to fewer iterations between design and engineering teams (and avoids wastage of simulation hours), leading to faster and efficient product design.
Key takeaways: TRI’s success stemmed from its deep understanding of Toyota’s design process and its ability to unify multiple departments under a cohesive AI strategy, ensuring seamless AI Integration. This would not have been achievable using an off-the-shelf solution developed independently by a simulation/AI vendor.
These companies aren’t just throwing AI into their workflows and hoping for the best — they are actively shaping how AI fits into their specific needs. Some build AI in-house for more control, while others collaborate with AI vendors for customized solutions. The key takeaway? There is no one-size-fits-all approach, but automotive companies taking an active role, in both AI strategy and development, is crucial.
How AI/Simulation vendors and automotive companies can bridge the gap
It is understandable that AI/Simulation vendors would like to develop and offer pre-made AI-enabled solutions, as selling the same software solution to multiple customers will optimize costs and generate more revenue. But such off-the-shelf offerings may experience only limited success. AI/Simulation vendors can deliver far greater value by pursuing a more collaborative approach with automotive companies.
Automotive companies should be open for such collaboration (unless they have all the resources to develop and maintain AI solutions themselves). They should not see AI as an experiment, but as a well-planned strategic investment.
The tables below highlight the approach Automotive companies and AI/Simulation vendors could take to work towards addressing the factors that slow down AI adoption. The specific factors that each action/approach helps address are also marked.
*AI and Simulation solutions could be offered by the same vendor or by different vendors.
Conclusion
AI has the potential to revolutionize automotive simulations, but adoption in production remains slow due to concerns related to integration complexity, AI explainability and maturity, budget/funding, scalability, among others. The stakeholders involved (AI/Simulation vendors, Automotive companies, technology/services partners) will need to take decisive steps to address these concerns and shortcomings, for the industry to be able to capitalize on this transformation.
References
Greenwood, M. (2023, September 27). Toyota’s new GenAI Tool is Transforming Vehicle Design. Retrieved from Engineering.com: https://www.engineering.com/story/toyotas-new-genai-tool-is-transforming-vehicle-design
How Simulation Can Accelerate Your Digital Transformation. (2021, September 8). Retrieved from Harvard Business Review: https://hbr.org/sponsored/2021/09/how-simulation-can-accelerate-your-digital-transformation
Kini, S. (2024). An evaluation of the factors impacting adoption of Artificial Intelligence technology in high-fidelity Engineering Simulations for Automotive Design. Conducted as part of the Executive MBA program at S P Jain School of Global Management.
Ragani, A. F., Stein, J. P., Keen, R., & Symington, I. (2023, June 21). Unveiling the next frontier of engineering simulation. Retrieved from McKinsey & Company: https://www.mckinsey.com/capabilities/operations/our-insights/unveiling-the-next-frontier-of-engineering-simulation
Reinventing the wheel? “FelGAN” inspires new rim designs with AI. (2022, December 12). Retrieved from Audi MediaCenter: https://www.audi-mediacenter.com/en/press-releases/reinventing-the-wheel-felgan-inspires-new-rim-designs-with-ai-15097
Tornatzky, L. G., Fleischer, M., & Chakrabarti, A. K. (1990). The processes of technological innovation. Lexington, MA: Lexington Books.
Volkswagen Group establishes artificial intelligence company. (2024, January 31). Retrieved from Volkswagen Group: https://www.volkswagen-group.com/en/press-releases/volkswagen-group-establishes-artificial-intelligence-company-18105
