LLMs and other AI models are proving their real-world value, and they may soon be an essential engineering skill.
The past five years have seen the field of automotive engineering greatly expand its interest in artificial intelligence (AI). The technology’s recent progress has made automakers keen to explore machine learning models, which are increasingly accessible in engineering software.
“[AI] is going to become a tool that you need to know as an engineer to have in your back pocket,” says Seth DeLand, a product marketing manager for MathWorks, the developer behind MATLAB and Simulink.
DeLand told engineering.com that while AI has been in MATLAB since the 1990s, recent AI advances such as large language models (LLMs)—the tech behind generative AI tools like ChatGPT—has elevated the capabilities of AI for engineers. This allows them to automate tasks that could not be automated before.
How automotive engineers can use AI
An automotive engineer can use AI tools in MATLAB to make, for example, a virtual sensor that fuses real sensor data to make additional predictions. Such an AI-based model could pinpoint when driver’s should change their EV’s battery, which is challenging to solve from a physics perspective because degradation mechanisms are hard to model.
In previous years, automotive engineering software allowed an engineer to input data relating to a battery’s state of charge (SoC), state of health, warranty period and average lifespan. Now the engineer can use such AI to understand what interactions occur within a virtual model. Specifically, they can visibly see what problems result when different types of degradation, like loss of lithium, occur.
DeLand says that many automotive MathWorks customers are using AI for novel applications for which there are not pre-trained models. “It’s proprietary information, so they frequently need to retrain or train from scratch these models based on their data,” says DeLand.
MATLAB’s core AI algorithms can be found in its Deep Learning Toolbox and its Statistics and Machine Learning Toolbox. MathWorks then integrates those algorithms in other toolboxes like the Computer Vision Toolbox, System Identification Toolbox or Reinforcement Learning Toolbox.
How Gotion used AI in Simulink to evaluate vehicle charging
In late 2020, Gotion, a Chinese battery company, used AI tools in Simulink to develop a prototype SoC estimation strategy to measure how well its batteries for electric vehicles would charge.
The SoC estimation involved gathering information from a voltage sensor, a current sensor and a thermocouple. Measurements collected from these instruments allowed Gotion to estimate the battery’s SoC over time. This information was used to train an AI model to predict SoC from the group of sensors and provide that information to the vehicle.
Previously, an automotive engineer might have used Kalman filters, algorithms to estimate system parameters. The drawback of these filters is that they are labor-intensive to set up and get working correctly. They also require a physics-based model, which is computationally expensive.
With its AI-based approach to SoC, Gotion was able to continuously improve the models as it worked on the project. This allowed the team to better evaluate the memory footprint of the model. The team could determine whether the model could run on an existing ECU without having to add memory. They could also learn whether the model was performant enough to run in real time.
After evaluating the neural network based SoC estimator on test vehicles, Gotion found there was approximately a 2% Root Square Mean (RMS) SoC error when charging. This low rate of error meant the prototype SoC estimation algorithm could be considered a feasible design strategy.
Simplifying AI upskilling
Another reason MathWorks and other software developers are building AI tools into their software is to help manufacturers retain their automotive engineers.
The demand for automotive engineers is expected to rise over the next three years, according to the U.S. Bureau of Labor Statistics. The auto industry cut jobs before and during the COVID-19 pandemic. Yet there is still a high demand for automotive engineers. The auto industry needs these professionals to solve tech-heavy problems like electrification and connectivity.
Gradually retraining existing engineers saves companies time, money and energy on recruitment and onboarding. Retraining on the job also helps manufacturers keep their engineering workforce on full-time. They do not have to pay for the cost of new degrees or suffer a temporary loss of engineering labor as the professionals earn degrees.
MathWorks’ current training options include “on-ramps,” two-hour tutorials to play with the technology and solve an example problem, instructor-led one or two-day courses and Massive Open Online Courses (MOOCs). The MOOCs typically center on data science and AI and are offered on educational sites like Coursera.
In the next few years, DeLand says MathWorks plans to offer automotive engineers more help as it adds in more AI tools. It also plans to continue assisting engineers with improving design processes.
