Page 21 - EE Times Europe Magazine | February 2020

EE|Times EUROPE 19

ARTIFICIAL INTELLIGENCE & INDUSTRY 4.0 for engineers to create and train reinforce-
In 2020, What Will Be the Key ment-learning policies and the development
of simulation data for training purposes.
Additional enablers for RL comprise
Trends in AI and Industry 4.0? straightforward incorporation of reinforce-
ment-learning agents into system simulation
tools and code generation for embedded
By Jos Martin, Senior Engineering Manager, MathWorks hardware. Taking autonomous driving sys-
2020 is set to be an important year for Industry 4.0 and tems as a real-world example, by including
an RL agent into the system, it is possible to
arti cia inte i ence A as t e tec no o ies are set to refine and enhance driver performance, lower
fuel consumption, shrink response time, and
contin a rede ne t e i its o at is ac ie a e or ultimately increase driving speeds.
en ineers and scientists. o o er t e ne t ear at are THE MARCH OF MODEL-BASED DESIGN
t e e trends t at e can e pect to see e er e TOOLS
AI-driven systems that are increasingly
design-complex are becoming more prevalent
COBOTS AND AI FACILITATE FLEXIBLE edge-computing devices. As data is sourced in industry. However, these systems require
PRODUCTION LINES beyond an individual machine alone, from significantly more testing processes as a
Collaborative robots, otherwise known as across multiple sites and different vendor result of the significant impact of AI model
cobots, that augment human capabilities and equipment, predictive maintenance will behavior on the performance of the entire
that are parameterized and refined by AI will dramatically improve. Additionally, AI-based system. Consequently, this year, it’s likely that
be the technology that finally unlocks manu- algorithms will boost productivity in factories there will be greater adoption of model-based
facturers’ ability to realize flexible production because these tools can dynamically optimize design tools that deliver simulation, integra-
lines in 2020. Visions of the future factory the entire production line throughput as well tion, and testing on an ongoing basis.
floor have predominantly focused on auto- as lower any energy costs that are imposed. The benefits of these are that simulation
mation, with production lines creating single permits designers to analyze how artificial
items, thereby aiding reduced inefficiencies REINFORCEMENT LEARNING IN AN intelligence interacts with a system, integra-
and long changeover times. However, for this INDUSTRIAL SETTING tion allows them to trial designs within the
type of production to be realized, and for When a computer learns to perform a task context of the complete system, and contin-
Industry .0 to reach the next level, produc- via repeated trial-and-error interactions uous testing makes it conceivable to easily
tion lines need to become more flexible. With with a dynamic environment, this is known identify limitations in AI training datasets and
numerous mechatronic modules that can as reinforcement learning (RL). This year, other design flaws in the system’s components.
be reorganized on an ad hoc basis, and with we will witness this technology progressing Engineers and scientists are set to experi-
additional cobots on standby — which can be from winning games like chess and Go against ence a plethora of benefits as a result of new
tuned by AI according to the following item human competitors to developing into a vital technologies. However, they must be careful
rolling down the manufacturing line — we will support for engineers. The technology will to make the most of the tools available to
be closer to the goal of full autonomy. lend itself to implementing controllers and them as well as encourage their teams to learn
decision-making algorithms for complex new skills and adapt to working with bigger
ACCESSIBLE AI BECOMES A REALITY systems, including autonomous systems datasets. They will also have to contend with
This year, AI project work will become more and robots. Key drivers for the deployment building new models and testing AI-driven
widely available to engineers and scientists of RL as a way to improve large industrial systems, which will be critical to realizing the
as access to existing deep-learning models systems include implementing simpler tools full potential of Industry 4.0. ■
and research improves dramatically. Con-
ventionally, AI models tend to be mainly
image-based, but in 2020, they will integrate
a much broader selection of data, from time
series to text to radar, sensor data, and more.
Engineers and scientists are best placed to
succeed with their AI plan due to their exten-
sive domain knowledge. However, to truly
excel, the use of tools, including automatic
labeling to quickly curate vast and high-qual-
ity datasets, is crucial. Teams with access to
a greater quantity of high-quality data have
a far better chance of producing accurate AI
models that generate the required outcomes.

EDGE COMPUTING IMPROVES PREDICTIVE
MAINTENANCE
New functionality of software on production IMAGE: SHUTTERSTOCK
systems is becoming a reality as a result of the
use of cloud systems as well as the superior
calculation power of industrial controllers and

NOVEMBER 2019 | www.eetimes.eu www.eetimes.eu | NOVEMBER 2019www.eetimes.eu | FEBRUARY 2020