Create an Edge AI solution

for STM32 without any AI

knowledge using
NanoEdge AI Studio
 Webinar agenda

1 The challenges of implementing Edge AI solutions

2 What is NanoEdge AI Studio and how it works

Predictive maintenance and condition monitoring use cases

3
using NanoEdge AI Studio

4 Demo: the design process with NanoEdge AI Studio

5 Q&A session

2
The challenges
of implementing Edge AI solutions
 AI momentum: buzz versus business value

2 to 5 years Confusion around AI

Edge AI

You need to start

now to meet the
market in-time
Companies struggle to assign a
realistic value and business outcome

AI products will be a standard on the

market in 2 to 5 years (Gartner)

Source: Gartner 2021

4
 Artificial intelligence at the deep edge
Moving part of Artificial Intelligence closer to the data acquisition
brings several benefits

Better user Optimized Cloud

experience usage

No latency Privacy by design

(real-time) (GDPR compliant)

Sustainable on
More reliability Add new functions and energy
services with Embedded AI

5
 A new way to add context-awareness
to your products
Create more robust software using Machine Learning on STM32

Standard programming Machine Learning

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

Procedural Desired output

Input Data Input Data
algorithm from the system

General ML model trained

Desired output from the system
for the specific problem

▪ Requires domain expertise to code ▪ Generate code from real-world observations

▪ Need to rewrite if environment evolves ▪ Re-learn from data if environment evolves

6
 160 billions machines
just “want” to do a better job

The washer isn't An unusual noise has

draining properly been detected
because and
a belt is showing recognized as a
signs of wear window break Enterprise restaurant
The pump is about to is full
break down HOME and
SECURITY
due to MAINTENANCE your waiting time is
a failure on a ball currently estimated
bearing to be 15min

INDUSTRIAL PEOPLE
MAINTENANCE COUNTING

7
 The over-promise & under-deliver trap

• Business leaders tend to

overestimate the impact of AI and Vision from business
But R&D means...
leaders
underestimate its complexity and
associated costs
• As a result, business leaders in
data and analytics must manage
their expectations or run the risk of
costly projects failing due to the
problems that occur during the move
from POC to production.

Source : Gartner report (2021) 8

 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

9
10
 AI/ML solutions for STM32

USE CASES
COMPANY’S PROFILE
Anomaly detection Classification Deep Learning

Engineering
Services
Embedded developers
▪ No dataset available
▪ No dedicated AI Team

Team with AI expertise

▪ Dataset available
▪ AI Team

11
What is NanoEdge AI studio
and how it works?
 NanoEdge AI studio: create a state-of-the-art AI solution
in a simple, fast, and affordable way
The power to create Edge AI solution, simply, quickly and affordably.

No data set Your developers Zero Cloud 1 to 16 Kb of RAM 3 x savings in $

required can use it now dependency < 10 Kb Flash and 2 times faster

13
 For embedded developers

NanoEdge AI Studio, an automated ML design solution

A solution A solution A solution

that creates that allows your that detects
custom STM32 to learn anomalies,
Edge AI Library on device classifies them
for you and much more…

14
 NanoEdge AI studio V3

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.

15
NanoEdge AI studio
Key features
 Anomaly detection use-case
ON THE PC ON THE MCU

1 Creation of an ANOMALY DETECTION 2 Use of an ANOMALY DETECTION

Machine Learning library Machine Learning library

Infer Infer

Model A Model B

ML library
Contextual
Signals Learn Learn

17
 One class classification use case
ON THE PC ON THE MCU

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

Machine Learning library Machine Learning library

Infer

Embedded
Static
Model

Normal ML library
Condition
Signal

18
 N-class classification use-case
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

19
 Extrapolation use-case
ON THE PC ON THE MCU

1 Creation of an EXTRAPOLATION 2 Use of an EXTRAPOLATION

Machine Learning library Machine Learning library

SPEED 10%
Signals
Vibration level 80% Vibration level 87%
Vibration level 65%
SPEED 25%
Vibration level 25%
Signals Vibration level 10%
ML library

SPEED 65% Classification

Signals

SPEED 80%
Signals

20
 From idea to datalogging
in a matter of minutes
• Streamlined data logging process
• No code
• All settings done using a graphic interface

The STWIN SensorTile

wireless industrial node
(STEVAL-STWINKT1B)

21
Predictive maintenance
and condition monitoring solutions
using NanoEdge AI studio
 Why do we need to monitor equipment state?

We are surrounded by wide range of machines, which eventually

break down if they are not maintained properly

23
 Why do we need to monitor equipment state?
Different maintenance strategies increase working reliability and
reduce operational cost
Reliability: OEE and uptime

Predictive
Predict future Prescriptive
Condition Based issues,
maintenance Controlled usage to
Reactive Preventive Continuous sensing scheduled
Planned to identify defects maximize lifetime
Repair on failure based on
maintenance but and optimize
leads to cost and lifetime to
loss of usable life performance
loss of production reduce
until repair productivity loss

Level I Level II Level III Level IV Level V

24
 Steps to a predictive maintenance system

Anomaly
Data Condition Predictive
detection &
Acquisition Monitoring maintenance
classification

▪ Acquisition sensor setup ▪ Data cleaning / denoising ▪ Machine learning of the ▪ Model deployment
▪ Retrieve data over wired/ ▪ Data visualization system behavior ▪ Remaining Life
wireless connectivity ▪ Preprocessing and ▪ Unsupervised learning at prediction models
▪ Label data Feature Extraction the edge for anomaly ▪ Overall efficiency
▪ Store data ▪ Feature Engineering detection optimization
▪ Supervised learning to ▪ Operational systems
classify anomalies integration

Edge - Factory Level (processed sensors data) Company Level (ERP, etc.)

25
 Filter clogging
CHALLENGE
• In an air conditioning system, it is very difficult to detect when a
filter is clogged.
• The engineers had imagined installing cameras to film the
colorimetry of the filters and compare it to a pre-learned model
to detect when a filter was obstructed. They did not succeed.
SOLUTION
• At the time of the first start-up of an air conditioner, NanoEdge
AI learns the “shape” of the high-frequency directly inside the
motor control card. No additional sensor is needed.
• When the filter is slightly obstructed, the shape of the high
frequency current is alternated and detected by the NanoEdge
Implementation of NanoEdge AI Studio AI library.

Microcontroller STM32 BENEFITS

• Ability to learn any filter automatically
Library Type Anomaly Detection • Works even for compatible replacement filters with different
density
• Software only upgrade to existing air condition system to bring
Signals used Current (motor control)
clogging detection capability. No additional hardware cost
involved.
RAM / FLASH 6kB / 11kB
 Pump maintenance
CHALLENGE
• Every pump has its unique signature according to pipe size,
shape and mounting
SOLUTION
• Learn the pattern of every pump in operation and detect
anomalies as they occur using vibration or current signal
• When anomaly is detected the second library is activated to
recognize the fault (classification)
BENEFITS
• Close to 100% accuracy due to local learning
• Extreme adaptability of model to wide range of pumps
Implementation of NanoEdge AI Studio • Ability to add seasonal learning phases

Microcontroller STM32 M33

Library Type Anomaly detection & classification

Signals used Vibrations or current

2kB / 5kB (lib 1)

RAM / FLASH 4kB/ 14kB (lib 2)
 In practice
Explore different use cases of NEAI

Visit https://data.cartesiam.ai/
28
 The ultimate no brainer AI solution stack
A unique one-stop shop solution

✓ Lead with true innovation

✓ Improved time to market
Your Industry Expertise ✓ Optimize cost
✓ Minimize risks

✓ Proven methodology to accelerate ML innovation process

AI Design Services ✓ Delivered direct or through certified partner ecosystem
✓ Direct R&D assistance for all sprint projects
Stay focused on
your expertise, ✓ Best AI offering portfolio on the market
AI Software and ecosystem
we bring you ✓ Most comprehensive AI stack ranging from deep
learning computer vision to self-learning anomaly
everything else detection.
✓ Most comprehensive HW portfolio to address all
Hardware projects and communication environment

29
Demo: the design process
with NanoEdge AI studio
 Summary and next steps

• NanoEdge AI Studio is a powerful solution to bring machine

learning to the edge.
• Optimal data security and privacy: data stays local, no cloud
connection required.
• Helps developers bring smart solutions to the edge quickly without AI
skills.
• Works on all STM32 microcontrollers with ARM Cortex-M cores.

To get started,
contact us at edge.ai@st.com
 Find out more at www.st.com/stm32nanoedgeai

© 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.