The use of machine learning and artificial intelligence in engineering is increasing, with the potential to give professionals extensive experience at an accelerated rate.
Siemens Digital Industries Software has submitted this post.
Written by: Stephen Ferguson, Marketing Director, Siemens Digital Industries Software
To become an engineer, you need to put in the hours of studying. Degree courses formed from centuries of engineering knowledge are invaluable as the foundation of a successful career. But the journey doesn’t end there. In fact, it’s only the beginning.
Engineers leave university with a wealth of theoretical knowledge, but we only become truly well-rounded professionals with years of experience in the field. The old saying that “the best way to learn is by making mistakes” is never truer than in engineering. The more you understand what doesn’t work and why, the easier you can find solutions that do work.
It might take decades before you really feel as though you’ve mastered your profession. Engineering isn’t simple, and you must be patient in your ongoing learning process. There’s no way to speed up the acquisition of experience.
Or is there?
The use of machine learning (ML) and artificial intelligence (AI) in engineering is increasing and they have the potential to give professionals that extensive experience at an accelerated rate.
Sounds too good to be true? Read on and I’ll explain how the amount of learning through simulation and test can be increased, and how knowledge can be transferred between projects.
It’s Not What You Do, It’s the Way That You Do It
Simulation and testing produce vast amounts of data that need to be reduced to key defining parameters. Experienced engineers develop the ability to interpret these data fields and identify critical features that directly influence the performance of a design.
But this takes considerable time. So, typically design teams use less experienced (and less expensive) engineers to do much of the legwork—meshing, setting up problems, performing calculations, post-processing. Senior engineers then make decisions based on the results. This works okay, but by isolating the key parameters you are only understanding which designs perform better, not why they perform better.
With software such as Simcenter, engineers use machine learning to scan the results of thousands of previous simulations and train an algorithm to identify the features that are most critical to performance. Not only does this speed up the entire process, but it can also identify critical features that even the most experienced engineer might miss. Machine learning is also applied to previously categorised simulation models to train AI to identify individual CAD components from a combination of their shape and metadata.
All in all, this delivers better results much faster, as well as freeing up engineers to apply their skills to more useful and interesting tasks.
What If… You Could Answer Questions Instantly?
“What if?” questions are fundamental to the engineering design process. They help us explore all the different options on the way to finding the optimum solution. The problem is one person can’t think of every question themselves. So, when someone at a design review meeting asks, “What if we just tried this?” it usually means the engineer must spend more time running simulations to test the new theory.
But what if that question could be answered right there and then?
Machine learning algorithms can accumulate knowledge from previous simulations and instantly predict the outcome of design changes, including full three-dimensional scalar and vector fields. No more rescheduling a follow-up meeting for next week; you can see the probable outcome of any speculations in real-time.
The accuracy does depend on the quality of the training database, so it shouldn’t be taken as complete proof of concept without being backed up by full simulation. However, it can instantly rule out unfeasible solutions, so that engineers can focus their time and resources on simulating designs that have a much higher chance of success. This speeds up the design process significantly as at every stage, large amounts of engineering time and computer processing power aren’t wasted on experimenting with design options that could never work.
Let Everyone Learn From Your Experience
Have you ever been involved in a project where a design parameter has been fixed at an early stage, only for you to realize months later that it might have been better not to fix it? Given that early in the design process you are often working with simple low-fidelity simulations, it’s inevitable that these parameters will look very different further down the line. It can be frustrating, but not if you use it as a learning experience for everyone. “Transfer learning” uses AI to map all the learning from one design process to another, allowing engineers to make better decisions about fixing parameters on the next project.
Making Use of Mistakes
Remember how I said mistakes are the best way of learning? They truly are, but despite the innovation that comes from them it can often feel like a lot of time has been wasted in making them. It takes many failed iterations to create one successfully implemented design improvement. And the data generated from these failures is often archived or even deleted completely.
But now that data is key to training the machine learning algorithms I mentioned earlier. They need as much data as possible to maximize their accuracy, so as far as they’re concerned, the more mistakes the merrier. Now engineers no longer need to worry that a design option they’re exploring will be a waste of time. With ML and AI, there is no such thing as waste.
Bringing Simulation and Test Together
Even in a world of high-fidelity simulation, physical testing is still crucial to the design process. It is only by combining the results from simulation and test that we can find the very best designs. The trouble is this is often a tricky process involving compromises in mapping from one domain to another. With AI and ML, data from multiple simulations and experiments can be combined into a single model, giving the most accurate predictive capability possible.
Delivering a Competitive Advantage
The use of AI and ML in simulation will continue to evolve but it’s clear from my experience that it’s already improving the quality and quantity of simulations for companies that use it. Not only does it facilitate the development of better products, but the time and resources saved in the design process itself are enough to deliver a significant competitive advantage.
Want to find out more about the future of AI in engineering? Check out this podcast, watch this video, and read about Monolith AI, the developers of the leading AI platform for design and engineering.
