How gas sensing gets intelligent with BME688 and BME AI-Studio (long version)

  Переглядів 13,844

Bosch Sensortec

Bosch Sensortec

День тому

The new BME AI-Studio tool enables customers to train the BME688 gas scanner on their specific application, like in home appliances, IoT products or smart home.
▬▬▬▬ Chapters ▬▬▬▬▬▬
0:00 Introduction
0:25 The BME688 Gas Sensor
1:03 Example: Coffee vs. Normal Air
1:26 A Few Things To Keep In Mind
2:08 Step 1: Record Normal Air
2:48 Step 2: Record Espresso Coffee
3:19 Step 3: Record Normal Air Again
3:36 Step 4: Record Filter Coffee
4:04 Step 5: Import & Label The Data
5:37 Step 6: Create New Algorithm And Classes
6:38 Step 7: Train And Evaluate The Algorithm
8:55 Step 8: Export The Algorithm
9:09 Conclusion
It is all around us, but unless there is a problem, we usually do not concern ourselves with what is in the air that we breathe. In both indoor and outdoor environments, poor air quality can greatly impact our health and well-being.
The BME688 is the first gas sensor with AI and integrated high-linearity and high-accuracy
pressure, humidity and temperature sensors. It is housed in a robust yet compact 3.0 x 3.0
x 0.9 mm³ package and especially developed for mobile & connected applications where size and low power consumption are critical requirements.
The gas sensor can detect Volatile Organic Compounds (VOCs) and other gases such
as carbon monoxide and hydrogen, in the part per billion (ppb) range.
You can find more information here:
BME688 product website:
www.bosch-sensortec.com/produ...
www.bosch-sensortec.com/softw...

КОМЕНТАРІ: 31
@soovui
@soovui 2 роки тому
Wow great to have such helpful algorithm software and where I am looking for it all this while... Such a brilliant move, well done Bosch. Keep it up. Like to see more videos especially people counting or crowded area detection. Maybe some motion Ai sensor capability too.
@ullithepilot4995
@ullithepilot4995 2 роки тому
Hey guys. I like the video very much. Very well structured and good explanations. What is miss is the last step, the verification of the learned data. Another question in this context is, if the sampelt data can be programmed to the sensor, or do they need to be in the algorithm in the controller. In other words the step from the development board to the real application would be interesting (if I look to the comments, I think I am not alone). In my case I would especially be interested to identify CO. At the moment I try out in my project the Chinese MEMS Sensor GM702-B. The BME688 could be a very interesting alternative.
@alchenist
@alchenist 2 роки тому
Wonderful tech. Thank you. Do I need the board with 8 sensors? It's a tad too large for my use case. Can I just buy single sensors attached to the breakout boards already available, bundle them with iot devices (arduino nano iot) and use the raw data to train with your AI studio ar my own python code??
@jensschroder8214
@jensschroder8214 2 роки тому
The BME280 and BME680 are already very popular among hobbyists with the ATMega, ESP8266 and ESP32. The BME280 is cheap to try out. The BME680 is for more serious applications.
@testme2026
@testme2026 Рік тому
that's great, but how do you load the inference back into the board ? and show how it identifies the two types of coffee ? you didn't show that
@alejandrolopez-wp2yb
@alejandrolopez-wp2yb 2 роки тому
hello, how many gases can 1 single sensor measure at the same time? Can I measure the smell of coffee and the quality of the air at the same time?
@agenthandy
@agenthandy 2 дні тому
Can this sensor be used with an arduino?
@frankfahrenheit9537
@frankfahrenheit9537 Рік тому
Is the training data applicable for all Boards now? Or is it valid for this particular board only?
@steffen_kuellmer
@steffen_kuellmer 2 роки тому
Mega!
@inputvolume6361
@inputvolume6361 Рік тому
Hi! I like the explaination so far. What I'm missing though is the next step. How to use the config file to measure data. Is it just copying the file onto the SD card? Or are there more steps involved? The documentation is very confusing here
@BoschSensortec
@BoschSensortec Рік тому
Hello Input Volume, Thank you for your message. To give an overview: there are mainly two times there is an exchange of configuration file between the BME688 development kit SD card and the BME AI Studio Desktop: 1.) First, while configuring the development kit board to record data - In this the configuration file is exported from the BME AI Studio Desktop and copied directly to the SD Card of the development kit board. This file can be identified by an extension “.bmeconfig” file. And once the board is powered on, it automatically starts recording data based on the configuration file. 2.) Second, to test the trained algorithm after the training is complete - In this case also the configuration file is exported from the BME AI Studio Desktop and copied directly to the SD card of the development kit board. To do the live test with BME688 Development kit App both the files with extension “.config” and “.aiconfig” has to be copied to the SD card. We hope that this information is helpful for you. Many greetings Your Bosch Sensortec Team
@santoshkumar-yh2bc
@santoshkumar-yh2bc 2 роки тому
I have a few queries : 1. Does the ESP32 board has application built on phython ? is there any link for application built on Arduino IDE? 2. i'm not able to get the sensor data on my SD card, Possibly because i overwrite the pre-Programmed application, Does anyone has link to the application code that comes with the module ? 3. Is the algorithm generated by the AI stdio portable to Arduino IDE ? Thanks in advance.
@chrisanderson1513
@chrisanderson1513 3 роки тому
Can you export training/test data for use with your own algorithms?
@riscv
@riscv 2 роки тому
most likely you could use the data generated on the microSD card in a separate environment (example: jupyter notebook (or google colab), numpy, tensorflow and tensorboard). This will allow you to train more advanced models with deeper neural architectures.
@gherdi
@gherdi Рік тому
Gibt es eine Möglichkeit den Sensor wieder zurückzusetzen? Bei mir blinkt das rote Licht mittlerweile mehrfach pro Sekunde und es werden keine Daten mehr gesammelt. LG
@TheOzarkWizard
@TheOzarkWizard 2 роки тому
Does this work with breakout, single chip boards?
@riscv
@riscv 2 роки тому
Hey buddy, this demo uses a simple pcb breakout for the BME688 sensor (3.3v) connected to a popular ESP32-WROOM-32, all data was saved to a microSD card, the sampling rate was very low (500 samples in several minutes ); So, YES, any microcontroller could interface with that sensor and get the data easily.
@cloudstergaming5730
@cloudstergaming5730 22 дні тому
what if you dont have a brandnew board?
@BoschSensortec
@BoschSensortec 19 днів тому
Hi cloudstergaming5730, If a brand new board is not used (i.e. the board was already in use/powered on for more than 24hrs) then the first step of running the sensor to stabilize can be skipped. All the best, your Bosch Sensortec team
@ressamendy
@ressamendy 5 місяців тому
Hello. What is the esp32 board model to which the demo card is connected?
@BoschSensortec
@BoschSensortec 5 місяців тому
Hi ressamendy, It is Adafruit Huzzah32 - ESP32 board. All the best, your Bosch Sensortec team
@ressamendy
@ressamendy 4 місяці тому
What exactly will be done when we first receive the BME688 Development Kit? I purchased 2 separate products from Mouser. Will my "Adafruit Huzzah32 - ESP32 board" be programmed first? If so, isn't there a manual or video explaining this? I could not find any information or detail in your videos or explanations about these first things to do.@@BoschSensortec
@agenthandy
@agenthandy 2 дні тому
@@BoschSensortecis it possible to us an arduino board?
@flisboac
@flisboac 6 місяців тому
Just curious: why are 8 BME688 sensors needed in this board?
@BoschSensortec
@BoschSensortec 6 місяців тому
Hi flisboac, During the data collection phase, the 8 sensors in the BME688 development kit enables to collect more dataset, additionally different Heater Profiles (possible to choose up to 4 Heater Profiles) can be enabled for each sensor to test for the best Heater Profile for the use case within the same time. Also, in the testing phase performance of one sensor can be compared to other. Your Bosch Sensortec team
@flisboac
@flisboac 6 місяців тому
@@BoschSensortec Thank you for the answer! But I still have some questions. 1. How different are deployment scenarios when compared to training? Do I need to have all 8 sensors to make the detection I trained the sensor for? (and if not, is there any chance that I could deploy less sensors, say 2, or 4?) 2. Considering that the sensor can't distinguish specific gases (ie. it's just a total count/estimate, in a way, AFAIK), and thinking in an Air Quality monitoring scenario, how feasible would it be to use only one BME688 (or an array of them, if not possible), instead of a sensor per gas type (e.g. HCHO, VOC, NH2, CO, etc.)? The way I see it, the BME688 is much more than just a VOC sensor, but I couldn't grasp quite yet if it can substitute some more traditional, distinct gas sensors.
@BoschSensortec
@BoschSensortec 5 місяців тому
@@flisboac Thank you for reaching out again. We would like to discuss your questions and refer you to our community in this regard. The Bosch Sensortec Community is for discussion, exchange and learning about sensor solutions. Registration and participation is free of charge: bit.ly/3KmLeQT
@erfindertreffhmbtec7747
@erfindertreffhmbtec7747 3 роки тому
ich habe noch nicht verstanden, wieso 8 gleiche Sensoren benötigt werden und wie der "footprint" ermittelt wird.
@thomasblock3117
@thomasblock3117 3 роки тому
8 Sensoren lernen Zusammen besser als einer und das mit besserer Statistik. Heute würde man das auch als Schwarmintelligenz bezeichnen.
@remy-
@remy- 2 роки тому
I expect a difference between the sensors by the warming plate temperature. And therefor a different ‘fingerprint’
@riscv
@riscv 2 роки тому
According to the BME680(great sensor by the way) datasheet, there is a sensor to sensor offset of 15%, so making a array of sensors will result in better probabilities. To be honest, the challenge when training a model from data coming from metal oxide-based sensors is to keep data acquisition fidelity between training/eval and prediction/test, for example if you train your model using data collected from sensors MOS in a perfect environment (clean, no contaminants, temperature, humidity) and then you try to predict data using real world (different temp / humidity / contaminants), your model will fail miserably, so it is necessary to prepare the environment, perhaps using small fans, heaters and filters.
Надеюсь никто не включит свет😰
00:59
Mikhail Boldurev
Переглядів 2,1 млн
Your home's air could be making you sick. Fight back!
13:24
Jeff Geerling
Переглядів 318 тис.
How Gas Sensing gets Intelligent with BME688 and BME AI-Studio with Bosch
9:47
Future Electronics
Переглядів 14 тис.
Simulating Tastes and Smells Using a Digital Lollipop - with Danielle George
3:51
The Royal Institution
Переглядів 28 тис.
How gas sensing gets intelligent with BME688 and BME AI-Studio
1:37
Bosch Sensortec
Переглядів 11 тис.
BME AI-Studio: Record and Live-Test with the AI-Studio Mobile App
5:08
Bosch Sensortec
Переглядів 1,1 тис.
DIY ESPHome Multisensor - Temp, Humidity, RGB LED, Motion and LUX
27:16
digiblur DIY
Переглядів 184 тис.
Redmi Turbo 3 - НЕ покупайте Apple, Samsung и Xiaomi в 2024 году…
14:33
Thebox - о технике и гаджетах
Переглядів 45 тис.
Обзор на мой iPhone👉 ПЕРЕПИСКИ + ИГРЫ + ПРИЛОЖЕНИЯ
21:02
Ксения Куприкова
Переглядів 149 тис.
Iphone yoki samsung
0:13
rishtan_vines
Переглядів 4,2 млн