Just two years ago, a new technology called Generative AI (ChatGPT) made its debut and in just one year, it spread rapidly across various industries. This was followed by competitors such as Microsoft and Google releasing more advanced and intelligent AI tools like Gemini 2.0, Copilot, and others.
Traditional jobs such as digital marketing, content writing, and even coding began to feel the impact, with major companies like Accenture, Dell, Meta laying off employees in these roles.
While AI tools have undoubtedly boosted productivity and accelerated research, their growing influence on the mechanical engineering field will be significant in the coming years. Given this situation, it’s natural for CAE engineers to wonder: Will AI eventually replace our jobs? Do we need to learn AI tools and shift to new roles?
For now, the answer is ‘No.’ But what about the near future? Could AI-powered CAE become the new standard? To understand this, we need to see the direction Artificial Intelligence is taking in simulations and identify the key skills that CAE engineers need to develop today.
In an article by Simscale, Samir Jaber points out that while AI and simulation have been driving progress especially in the automotive sector—most manufacturers are still in the early stages of integrating AI into simulations. This doesn’t mean the technology isn’t compatible; it’s simply still evolving.
Finite Element Analysis (FEA) has been the base of simulations for decades. By discretizing a structure into elements and solving governing equations, FEA provides reliable results for various engineering problems. Let’s look at AI approaches which are already being utilized by companies (like Monolith).
Data-Driven Simulation:
In a purely data-driven simulation, models rely entirely on data instead of incorporating physical laws. This can work well for scenarios where abundant experimental data is available, but the physics is complex or poorly understood.
Consider predicting crash behavior in vehicles. By using a large dataset of previous crash tests, a machine learning model can predict outcomes for new designs (this approach is currently used by a company called Monolith for Crash Simulations). However, without integrating physical laws, these models might struggle to generalize beyond the dataset.
Understanding PINNs - The Next Step in Simulation:
Physics-Informed Neural Networks (PINNs) are a modern approach to solving complex engineering problems. They combine the power of neural networks with physical laws like equations of motion or heat transfer. Unlike traditional simulation methods, PINNs directly embed these equations into the network’s training process, making it possible to solve problems faster and with less computational effort.
Imagine predicting how heat spreads through an engine block during operation. Traditional methods require meshing and significant computational resources. A PINN could solve this by learning the heat equation and predicting temperature distribution much faster, potentially speeding up design iterations.
Companies such as NVIDIA, DeepMind, and MathWorks are actively developing these advancements, which could transform simulation processes in the years to come.
To effectively use PINNs, engineers need to learn programming (e.g., Python), understand neural networks, and have a firm grasp of numerical methods. These skills are currently outside the expertise of many CAE engineers, whose area focuses on FEA tools and Engineering principles.
Collaboration is the Key:
Given the growing interest in AI-driven simulations, industries are likely to facilitate collaboration between CAE professionals and data scientists. Joint projects, shared tools, and cross-disciplinary training programs could become common as companies explore the potential of PINNs.
Each brings unique strengths—CAE engineers contribute domain knowledge and physics expertise, while neural network experts handle model development and optimization. (CAE engineers with programming skills in Python or Fortran may have an edge over others, but the full scope of this skillset is still not entirely clear).
CAE engineers should therefore focus on strengthening their domain expertise and understanding of physics. This foundation will make them valuable partners in collaborations, ensuring accurate and meaningful applications of AI-driven techniques like PINNs.
What will the future of simulations look like?
According to experts, the future of simulation will likely combine the strengths of FEA and PINNs. Traditional methods will remain crucial for detailed and well-understood scenarios, while PINNs will complement them in areas requiring faster, multi-scale solutions. Collaboration between domain experts and data scientists will drive this evolution, creating smarter and efficient simulations.
In summary, this is the direction the simulation industry is moving towards:
PINNs will accelerate simulations by integrating physics and AI for faster results.
Teams of CAE experts and AI specialists will drive effective PINNs implementation.
CAE engineers will contribute with domain expertise and physics knowledge.
Simulation will evolve into a mix of FEA and PINNs.
It's likely that your friends or colleagues working in CAE have similar questions. Instead of debating whether AI will replace your jobs, share this article with them to spark a meaningful conversation.
You can find useful resources on this topic below:
1) Library of examples - PINNs validation with physical data
2) Working of PINNs (with example) - NVIDIA
3) Data-driven simulation - Case Studies
4) Deep Learning for CAD/CAE - Topology Optimization
5) FEA and PINNs - Basic Difference
If you found this article useful, feel free to share your feedback in the comments.
Do you think instead of creating models , assigning physics and meshing for a simulation will move towards giving results just by giving a prompt?