The automotive landscape is at its most vibrant today. From hydrogen fuel cells and high-voltage electrification to radical new structural architectures and interior comfort paradigms, automotive OEMs are exploring vast territories of innovation. Yet, amidst this explosion of options, the core mission of the structural safety engineer remains absolute, ensuring that every occupant and vulnerable road user returns home safely.
Movable Deformable Barrier Frontal Impact
In this high-stakes environment, computer simulation has evolved to be a foundational pillar of vehicle design. Crashworthiness, the measure of a vehicle’s structural ability to plastically deform while maintaining a sufficient survival space, is no longer a trial-and-error physical process. The goal is to engineer an optimized structure that absorbs crash energy through controlled deformation, ensuring a deceleration pulse with an early peak and gradual decay. This allows restraint systems like seatbelts and airbags to manage the residual energy and minimize load transfer to the human occupant.
The Ground Truth: Non-Linear Structural Solvers and the Physics of Survival
Structural considerations in an LSDYNA crash
Crashworthiness analysis presents a complex structural mechanics challenge, characterized by high-impact loads that trigger large-scale deformation, material buckling, and component stacking within multi-material assemblies. Designing for these high-stakes scenarios demands deterministic modelling of the entire crash event to predict outcomes with precision. For decades, automotive OEMs have used Physics based Finite Element Solvers to provide a first-principles perspective by solving the complex, non-linear mathematical equations that govern structural deformation and energy dissipation.
During vehicle development, crash safety engineers perform thousands of simulations, ranging from component-level assessments, sled models and subassemblies to comprehensive full-vehicle studies incorporating crash test dummies, airbags, seatbelts, and barriers. Because automotive OEMs operate in diverse global markets, vehicles must satisfy a rigorous array of international regulations covering load cases such as frontal, offset-frontal, side, and rear impacts, as well as rollovers and pedestrian protection. Crash simulations have become instrumental in this process, significantly reducing the reliance on costly physical testing while accelerating the overall product development cycle. Yet, the very precision that makes crash simulations indispensable also makes it computationally expensive, often requiring massive high-performance computing (HPC) clusters to run a single full-vehicle simulation.
Beyond the Bottleneck: Synthetic Design Generation and Validation
Geometric AI predicting new structural reinforcement shapes using geometric deep learning
The challenge in modern vehicle development is twofold: the time required for CAE simulation and the time required for a designer to manually iterate on complex geometries. Traditionally, a designer relies on intuition and experience to propose a handful of design candidates, which are then queued for lengthy simulations. This manual process is inherently limited by human bandwidth and the "slow" nature of high-fidelity solvers. As the industry moves toward more iterative and sustainable design, the traditional "simulate-analyze-redesign" loop can become a bottleneck. This is where the integration of data-driven AI becomes a competitive necessity.
If crash simulation is the deep-thinking scientist of the safety world, physics and geometry driven AI is the intuitive strategist. It addresses both the design generation and performance prediction bottlenecks using deep learning technology. By utilizing advanced geometric deep learning, geometric AI modelsdoesn't just store shapes; it understands the underlying "manifold", the mathematical DNA, of automotive parts. By providing the AI with a few "seed" geometries (successful designs from previous programs), it can interpolate between them to algorithmically generate hundreds of new, viable design candidates in seconds. For the engineer, this means the end of "slow design generation." The AI can populate a vast design space with hundreds of structural designs that no human would have the time to manually model. This shifts the designer’s role from a drafter of parts to a curator of possibilities. The job is now to think "outside the box," defining broader, more ambitious design spaces for the AI to explore.
Physics AI Model trained on simulations
AI models canalso learn the behaviour of physics by ingesting vast amountsof data generated by historical crash simulations. Rather than solving thefundamental physics equations from scratch for every minor design change, it identifies the hidden correlations between a structural geometry change, such as a certain curve in a rail or its thickness, and a specific deformation mode.
Coupled together, the geometric design generator and the physics predictor can work together to predict the performance of a new geometry in a fraction of a second. It acts as a high-speed filter, identifying the "top 1%" of designs that show the most promise. Only these "winners" are sent back to physics solver for final, high-fidelity verification, ensuring the speed of AI is always validated by the rigorous truth of physics.
This approach has been proven on several safety critical load-cases such as pedestrian impact, battery integrity, battery cooling, roll-over impact, frontal offset impact to rigid and deformable barriers and many other structural and fluid applications across automotive and other industries. In all these cases, the AI model is up to 100x faster than the physics solver in evaluating the design and thus accelerating the virtual validation process.
Through these dual AI engines, we are no longer just analysing crashes; we are algorithmically engineering a safer future at a pace that keeps up with the speed of innovation. By automating the routine and amplifying the creative, we are ensuring that the next generation of vehicles isn’t just smarter but fundamentally safer for everyone on the road.
Murali Pullela is Country Head (S&A), Synopsys. Views are personal.