Automotive Engineering - August 2023 - 9

SPONSORED CONTENT
EXPERT Insight
Augmenting simulation with artificial intelligence
The application of artificial intelligence (AI) to simulation tools has the potential to add new dimensions to the design process. As
AI's capabilities begin to emerge, design and engineering simulation expert MathWorks is exploring the possibilities throughout
the simulation environment. For this Expert Insight interview, Seth DeLand, MathWorks Principal Product Marketing Manager, AI
Products, spoke to SAE's Automotive Engineering about AI's prospects to enhance simulation tools and techniques and how the
company already is incorporating AI into its products and training.
Automotive Engineering: Please describe 'the intersection
of Artificial Intelligence and simulation.'
DeLand: Simulation has become a key tool for developing
advanced algorithms and embedded systems
into modern engineering workflows. We
mean simulation of components, systems, and systems
of systems. Simulation is a tool for different
stages of the design process, including desktop
simulation with models, software in the loop, processor
in the loop and hardware in the loop.
Artificial Intelligence is a megatrend, creating
DeLand: Engineers are upskilling their understanding
of AI technology. MathWorks offers everything
from two-hour on-ramps to instructor-led training
courses to online programs on platforms like
Coursera. Engineers who are not AI experts will
find AI approachable and easier to use.
many new engineering approaches and techniques
to build data-based models that predict outputs
from inputs. AI and simulation intersect when engineers
use simulation tools to build these systems and AI is
used in the research and development phase.
Seth DeLand,
MathWorks Principal
Product Marketing
Manager, AI Products
Automotive Engineering: What challenges arise
from this intersection?
DeLand: One challenge comes from collaborations
with one team doing AI and another simulation. It's
important that the modelling and the simulation
tools work with models from a variety of different
AI frameworks, including some of the big names
like TensorFlow and PyTorch.
Another challenge is the rapid evolution of AI. A discovery
Automotive Engineering: What applications and use cases
result from this intersection?
DeLand: There are four high-level applications. First is Virtual
Sensing, the use of AI as a state estimator, which uses existing
sensors or states to predict something else. Engineers have all
sorts of techniques to do that. AI models are gaining a lot of
attention for this because they can capture non-linear behavior.
Second is Reduced Order Modelling, which approximates a
very accurate physics-based model with a leaner model better
suited for exploring different designs.
AI models can capture a lot of non-linearities from engineered
systems. Such new AI techniques as long short-term memory
(LSTM) models do a good job with time series signals, as well as
neural ordinary differential equations (ODEs) which create hybrid
models combining first-principles physics with data-driven techniques.
AI may be used as a reduced-order model across many
different domains. Two examples are thermal models and fluids
models, which tend to be computationally intensive for applications
like air handling or combustion exhaust.
The third application is use of simulation as a test environment
for that AI model. System simulation is often used to
exercise a system over a variety of different conditions for
things like testing or requirements validation.
Fourth is simulation as a source of AI data. AI models tend
to require a lot of data to train. Often engineers don't have
the right data to train a model for a particular task. Maybe
they want to augment the data they have. Collecting that
data can be expensive or dangerous, so folks use simulation
as a source of data for training their AI models.
Automotive Engineering: What changes are driving more
activity to the AI-simulation intersection?
AUTOMOTIVE ENGINEERING
late in a design process can cause you to rethink how you are
using AI.
Automotive Engineering: How does MathWorks support
these trends?
DeLand: We offer apps and high-level algorithms to support
AI in all steps of a model-based design process, for machine
learning, deep learning, reinforcement learning. We focus on
" examples " to make it easy for a control engineer to find a
relatable workflow. MathWorks products have more than 500
AI examples in our documentation for many domains, including
signals, lidar, radar, wireless and autonomous systems.
AI is more than a modelling step. It's the workflow; data curation,
modelling and simulation testing of the model, through to
deployment. We help build out that workflow and support its
automatic code generation. You can convert AI models developed
by these high-level tools into lower-level C or C++ code to
run on an embedded processor or CUDA code to run on a
Graphics Processing Unit (GPU), enabling engineers to respond
quickly to changes and deploy new models into production.
Automotive Engineering: Where can readers learn more?
DeLand: MathWorks has many learning resources, including a
great e-book on integrating AI into system-level design, and a
bunch of training materials available for upskilling.
Check out mathworks.com for free on-ramps or see what
training courses are available, and for two recent webinars on
AI for model-based design - one on virtual sensing and the
other on reduced-order modelling.
Watch the full interview with Seth.
August 2023 9

Automotive Engineering - August 2023

Table of Contents for the Digital Edition of Automotive Engineering - August 2023