NANOEDGE AI STUDIO

Version 3

Your fast track to smart

products
 160 billion machines
just “want” to do a better job

Add quality services Make home security

to your offer - systems more
Monitor usage of intelligent without
equipment and plan impacting user
after-sales privacy
intervention Optimize building
Detect equipment drift occupancy and
before failure happen HOME management power
and optimize
SECURITY consumption - make
MAINTENANCE
maintenance cost – computer vision
save up to 20% without expensive
equipment downtime cameras
INDUSTRIAL PEOPLE
MAINTENANCE COUNTING

3
 Product development new paradigm

From rule-based engineering to data-driven engineering

Standard programming Machine learning

Handcrafted rules based on experience Rules learned from real-world data

▪ Requires digital signal ▪ Generate code from real-

processing skills world observations
▪ Manual feature ▪ Automated feature
extraction? extraction?
▪ Need to rewrite if ▪ Re-learn from data if
environment evolves environment evolves

4
 Distributed artificial intelligence approach

Leverage billions of devices at the edge!

Data center cloud

Thousands Analytics, storage, compute

Edge nodes
Millions IoT gateways, micro datacenters

Ultra-low-power Edge things

100 billions devices and sensors Real time, local processing

5
 Artificial intelligence at the edge

Why move AI closer to the edge?

Ultra-low latency Privacy by design

Real-time applications GDPR compliant

Sustainable on energy
More reliability
Low-power consumption

Security of data
Better user experience
No sharing in the cloud

6
 For most companies, creating an edge AI device is
a long journey with extraordinary challenges
Investment, complexity and development time are often barriers to AI
adoption

Important upfront
investment
RAM & energy
Challenge
Cloud
dependency
Lack of data
science resources
No qualified
data sets

8
 NanoEdge AI Studio

NanoEdge AI Studio, an automated ML design solution

Generate ultra optimized ML library for

STM32
ML model benchmark to speed up your
development time
State-of-the-art ML implemented
continuously: no specific AI skills needed
 Create a state-of-the-art AI solution
in a simple, fast, and affordable way
The power to create Edge AI solutions in a simple, quick, & affordable
way

Small and
efficient dataset
Your developers
can use it now

Zero Cloud
Lack of data dependency
science resources
No qualified 1 to 16Kb of RAM
< 10Kb Flash
data sets
3 x savings in $
and 2 times faster

12
 NanoEdge AI Studio in the AI development workflow

1 Data preparation 2 Model development 3 Model implementation

Model
Data Data Model Model library Model
selection and
acquisition processing testing creation inference
training

Automated Edge AI Software

Data logging
tools
All the tools needed for data preparation

I want the tools to easily create my dataset”

14
All the tools needed for data preparation

I want to quickly and efficiently clean my data”

15
State-of-the-art machine learning for
ever smarter products

I want them to autonomously adapt to their target environment

and detect anomalies by themselves”

I want to detect any outliers”

I want to know by name what problems are occurring”

I want to anticipate the vibration level so that I have time for

corrective actions”
16
 NanoEdge AI Studio V3.2
Same ease of use, more performance

NanoEdge AI Studio
.. And then optimizes it to
improves the performance
reduce footprint and latency
of the model..
17
 NanoEdge AI Studio
enabling on-device learning
ON THE PC ON THE MCU

1 Create the library, ONCE. 2 Use the library, MANY TIMES.

Create and embed

a self learning engine

Standalone PC (Win/Linux) solution For anomaly detection, the model is

self-trained at the Edge.
 Learn in-situ what normality is
ON THE PC ON THE MCU

1 Creation of an ANOMALY DETECTION 2 On-target learning for a fine-tuned model

Machine Learning library

Model A Model B

Learn Infer Learn Infer

ML library OK
NOK
OK
NOK
Contextual
Signals

Embed to Embed to
MCU project MCU project
equipment A equipment B

19
 Anomaly detection use case
ON THE PC

1 Creation of an ANOMALY DETECTION

Machine Learning library
 Quick & effective n Class classification
ON THE PC ON THE MCU

1 Creation of a n CLASS CLASSIFICATION 2 Use of an n CLASS CLASSIFICATION

Machine Learning library Machine Learning library

Bearing
Problem
Signals Bearing 90% Bearing problem
Misalignment 0%
Misalignment Cavitation 3%
Problem Shaft Imbalance 7%
Signals ML library

Cavitation Classification
Problem
Signals

Shaft imbalance
Problem signals

23
 Predict states with extrapolation features
ON THE PC ON THE MCU

1 Creation of an EXTRAPOLATION 2 Use of an EXTRAPOLATION

Machine Learning library Machine Learning library

Vibration level 80%

Vibration level 65% Vibration level 87%
SPEED 10%
Signals
Vibration level 25%
Vibration level 10%

SPEED 25%
Signals
ML library Speed 82 %

SPEED 65% Regression

Signals

SPEED 80%
Signals

24
 NanoEdge AI Studio
Always more added value, and simple to use
Step 1 Step 2 Step 3

The pump is about to The pump is about to The pump is about to Anomaly
break down break down break down detection

due to due to
a failure on a ball bearing a failure on a ball bearing Classification

and will reach 35% of

anomaly next Tuesday at Extrapolation
& 1 Class
5:32 PM

25
Always more added value, always simple to use

➔ Click here to watch the YouTube video 26

NKE WATTECO

CHALLENGE
Easily and quickly deploy predictive maintenance 4.0 solutions
without the usual constraints of deploying an AI-based solution,
which are to:
• Collect huge amounts of data to train static models
• Work with data scientists to develop complex algorithms
SOLUTION
The “Shazam of vibrations” right out of the box.
With embedded AI based on NanoEdge AI Studio, snap Bob
assistant to the machines in your factory and Bob will carefully
learn and analyze the vibrations and warn you before a problem
impacts the production line. Application of NanoEdge AI Studio
BENEFITS
Microcontroller M3 @ 80Mhz
With just a few contextual inputs, anyone can generate optimized
and effective machine learning algorithms with NanoEdge AI
Studio. The ability to learn algorithms on the device allows the Library Type Anomaly detection
user to tailor anomaly detection to a specific equipment or
environment to reduce false positives and false negatives. Signals used Vibrations

RAM / FLASH 6K / 28K

IRMA
Intelligent Resource for
Maintenance Application
CHALLENGE
Being able to predict issues in machinery is a considerable
advantage in many respects. Yet, it is often complicated to
understand precisely the nominal operating state of an equipment
to deduce a malfunction.
SOLUTION
IRMA can predict failures thanks to its multiple sensors and its
embedded AI at the edge. First, IRMA collects data (vibration,
temperature, pressure, humidity, acoustics and gas) then compares LoRaWAN® + BLE + Zigbee, IP67, ATEX Zone 0
them via the embedded AI algorithms with a database that is
constantly enriched. IRMA warns you at the slightest anomaly.
Application of NanoEdge AI Studio
BENEFITS
"Without data science experts on the team and within a very short Microcontroller STM32L4
development time, we managed to create a unique solution on the
market combining embedded intelligence and 6 sensors. IRMA has
Library Type Anomaly detection and classification
already been recognized as a major innovation by several industrial
awards and is already installed in production by prestigious Humidity, acoustic, vibration (1 and 3
customers” Signals used
axes),temperature, gas, pressure

RAM / FLASH 9Kb / 17Kb

Predictive maintenance for
electrical solutions
CHALLENGE
The mechanical ageing of circuit breakers is virtually impossible to
anticipate, leading to untimely power outages and costly production
downtimes.
SOLUTION
NanoEdge AI solution has been successfully implemented inside
circuit breakers to learn the normal switching patterns and help plan
predictive maintenance activities.
BENEFITS
“Using Machine Learning libraries developed with NanoEdge AI
Studio, we were able to anticipate behaviors that were previously Application of NanoEdge AI Studio
difficult to detect. An AI that is simple to implement and directly
integrated into the heart of our equipment is an important added Microcontroller STM32G4
value in future developments.”

Library Type Anomaly detection and classification

Signals used Current & Vibrations

RAM / FLASH 12Kb / 20Kb

People presence detection &
counting
CHALLENGE
Being able to detect when people are in a corridor or in a room
even if they are standing still, without implementing camera-based
solutions for smart light and smart heating/air conditioning systems
in smart buildings.
SOLUTION
NanoEdge AI Studio can classify any time series data, such as
waves, particle movements, electrical field changes and more. This
allows users to develop advanced and tailored solutions.
BENEFITS
“Using environmental change to detect presence allows to create Application of NanoEdge AI Studio
computer-vision-like solutions without the disadvantage of having a
camera. It enables low-power solutions, safeguarding privacy and
Microcontroller STM32U5
security while remaining accurate. NanoEdge AI Studio has helped
us develop such a solution in record time thanks to its ease of use
and high performance.” Library Type Classification

Signals used Wave or particle

RAM / FLASH 21Kb / 45Kb

 One step closer to the signal

Possibility to divide the tasks between the sensor and the MCU

Edge

ERROR 1
OK ERROR 2
NOK ERROR 3
ERROR 4
Sensor
Pump MCU Cloud
 Creating highly efficient intelligent sensors with
embedded AI engine
Maximizing efficiency and minimizing energy consumption to make
large-scale sensor proliferation sustainable

Sensors with DSP

Power consumption vs. performance
(Intelligent Sensor Optimization
Processing Unit “ISPU”) Ultra-low power consumption (µW envelope)
for real-time processing
and AI
Optimized algorithms
Efficiency Built-in operations capabilities to create
optimized functions

1000x less Machine Learning & Deep Learning

power consumption
vs. traditional solutions Interoperability Easy programming, interoperable with main
AI frameworks

33
 ISM330ISN – key benefits

Ultra-low power 6-axis IMU with Intelligent Sensor Processing Unit

Embedded ISPU: an ultra-low power

✓ FS: ±2000 dps, ± 16 g
✓ Idd 0.59 mA (Combo mode)
programmable core for AI algos and
✓ Axel noise 70µg/√Hz processing
Highly-specialized DSP
✓ Gyro noise 3.8mdps/√Hz
for ML and processing
(2.5x3.0x0.86mm3 LGA14)

Sensor HUB to connect additional sensors

Higher detection accuracy, always on

monitoring in anomaly detection applications
Home alarms Robotics Condition monitoring

34
 ISPU benefits
Ultra-low power consumption
• Efficiency of the embedded DSP (digital signal processing)
• Very low data exchange with external microcontroller

Ultra Low latency

• Processing / decision taken directly in the sensor

Easily programmable with AI commercial

or open-source models or with C language

Data Privacy & security

• Sensor data are locally processed and not provided outside

Integration / Miniaturization

35
What’s next?

36
 NanoEdge AI Studio Version 3 roadmap
A complete development software for any embedded developers looking to easily make any product smarter.

Additional Families Of Algorithms… … And Tools

We will continue to add families of …and even more useful tools to

algorithms to provide state-of-the-art facilitate prototyping and make
software solution to developers… embedded AI development a breeze.

Increasingly Rich Algorithms

We will continue to deepen the

mathematical richness of each algorithm
family by constantly adding the best of AI
algorithms from our own R&D team.

37
 Find out more at stm32ai.st.com/nanoedge-ai/

© STMicroelectronics - All rights reserved.

ST logo is a trademark or a registered trademark of STMicroelectronics International NV or its affiliates in the EU and/or other countries.
For additional information about ST trademarks, please refer to www.st.com/trademarks.
All other product or service names are the property of their respective owners.