How AI is changing simulation and modelling

The old mantra ‘check twice, cut once’ is a vital reminder when making anything – and engineers know more than most about the risks of costly mistakes. While physical prototypes are always useful, the rise of technology enables engineers to simulate and model ideas before putting hard cash into building things.

AI is poised to play a “huge role” in the future of simulation platforms, particularly in creating high-fidelity digital twins that mirror the real world, says Saber Fallah, professor of safe AI and autonomy at the University of Surrey. These digital twins – physical infrastructure or products recreated in silico, where they can be tested with a number of variables – can help optimise and validate assumptions, improving designs in the long run.

“The value of these AI systems is the generation of high-fidelity simulation platforms,” says Fallah. “These models have access to the different types of data – for example, images, LiDAR, canvases, all these things – so they can produce more realistic environments based on data and generate different sorts of environments.”

Seeing double

It’s not just in the ability to produce these simulations that AI has proven transformational. It also makes it easier to spin up simulations without coding expertise. Traditionally, harnessing the power of those data streams and producing high-fidelity simulations would require high-level knowledge of how to develop such models and code them by hand. 

With “classical models or physics-based models, it was really hard to generate a multi-purpose or general-purpose simulation platform,” says Fallah. However, large language model-based platforms can turn plain text queries into complicated simulations in an instant – and troubleshoot any errors that arise.

Digital twins have become a buzzphrase within engineering, perceived as a solution to many issues faced by a sector with high overheads and constrained investment. A 2024 survey of industry found that around two-thirds of businesses say the main reason for using digital twins is to optimise processes to be more efficient, while six in 10 say they help support decision making.

One area for more growth is in design and development, with only 40% of respondents saying they use twins for that. The team behind the survey, representing several universities worldwide, say digital twins “are gaining popularity in both academic and industrial contexts”.

Beyond catalysing new ideas, the supercharging of simulation also reinforces the ability to fine-tune pre-existing designs. Take the design and testing of vehicle safety features, such as crumple zones, roll bars or other areas. “Using reinforcement learning, we can train simulations or agents of the simulation to create cases that rarely happen in the real world,” says Fallah. That allows manufacturers to test the impact of one-off crashes in silico, rather than in real life. 

It’s a principle Fallah and his lab tested themselves with Jaguar Land Rover, designing specific agents using reinforcement learning to maximise the chance of accidents with autonomous vehicles to check the safety of the technology. 

Looking ahead 

But it’s not just worst-case scenarios modelling benefits. For transport infrastructure on roads, rail and in the air, simulation can understand population-level flows. “We can use AI-based simulations to optimise traffic for managing fleets,” he says. “We can also look at public transport network systems and logistics, on-demand and mobility models.”

Fallah can envisage a future where governments and testing bodies could adopt simulation and modelling tech to develop policy and approve infrastructure – though issues inherent within AI will have to be ironed out before that occurs, particularly in sensitive areas such as healthcare. “I believe AI and large language models will play a significant role in the next generation of simulation platforms, which maximise their productivity in the real world,” he says, “and provide near-real world performance.”

IMechE's Simulation and Modelling event will take place in Birmingham from 30 September to 1 October, covering the latest applications of simulation technologies, AI and machine learning, digital twins, multiphysics modelling and much more. Register now.