igivis7/ai_sense_by_hazel_n_igor

DRS28_Hazel_n_Igor_portfolio_project

AI sense by Hazel and Igor

The AI_sense_by_Hazel_n_Igor repository is a workspace for portfolio project of Hazel Wat and Igor Isaev. The work on the project was done during the classes of Data Science Retreat in 2021 (Batch 28).

This repository is forked on https://github.com/igivis7/ and https://github.com/hahahazel to reflect the work of all contributors.

The online version of the app is created with Streamlit and can be found here:
https://share.streamlit.io/igivis7/ai_sense_by_hazel_n_igor_4streamlit_online/main/streamlit_AIsense_by_Hazel_n_Igor.py

---

Contents

• About the project
• Live demo
• How to make it running
• Assembling and Connecting to computer
• Features of the project

---

About the project

This project was dedicated to apply our newly gained knowledge of data science on a real-world task.

As the object of study, we decided to take

Grove multichannel Gas Sensor V2 (GGSv2) sensor and implement Machine Learning to solve an odor recognition task.

During the project we have practiced building of Data Science (DS) project from scratch, mastered our Machine Learning (ML) skills and learned ways to overcome problems.

The GGSv2 sensor consists of 4 measurement heads for measuring Nitrogen dioxide (NO2), Carbon monoxide (CO), Ethyl alcohol (C2H5OH) and Volatile organic compounds (VOCs). Each head is an metal oxide semiconductor gas detector made to measure a certain gas molecule. The combination of the 4 readings allows to detect and distinguish different odors. The GGSv2 sensor can be used together with Arduino boards, WIO terminal, Raspberry Pi or any other device that is able to work with I2C communication protocol.

To make classification of odors a fully connected neural network was implemented.

The live demo (with pre-collected samples), code description, project features, hardware list and possible upgrades are listed below.

---

Live demo

There are two options to see run live demo:

1. Online version via Streamlit app. There is possible to see GUI and run with pre-trained model and simulated samples. It does not work with hardware (GGSv2).
   1. go to the app page
   2. press Get random premeasured test data button.
   3. press Predicting button to run prediction using existing model.
2. Local Streamlit app demo.
   1. make sure you have installed all required packages in your environment (see the next section)
   2. go to the folder cd ./soft/streamlit_app/
   3. run streamlit run streamlit_app.py

---

How to make it running.

Here is an instruction to make your copy of the project running.
The procedure is described for a Linux-based OS. It was also was checked with MacOS, but not with Windows.

There a few steps to create a device that is able to detect and distinguish odors:

0. Get an Arduino, Gas Sensor, HDT Sensor, and supplementary parts to combine it.
1. Assemble the device according to the instructions in the section Assembling and Connecting to computer. And connect to a computer.
2. Install Arduino IDE.
3. Upload firmaware to arduino:
   • open file GGSv2_n_DHT11_03_10hz.ino from ./soft/arduino_soft/GGSv2_n_DHT11_03_10hz/ in Arduino IDE
   • select board
   • select port
   • check
   • upload
4. Create Conda environment and install requirements:
   • conda create -n pyAIsense python=3.9.7
   • conda activate pyAIsense
   • pip install ipykernel
   • python -m ipykernel install --user --name pyAIsense --display-name "pyAIsense_env"
   • pip install -r ./info_n_docs/requirements_AIsense.txt
   • jupyter-notebook
5. Collect data with 01_Collect_Data.ipynb
   from ./soft/main_soft/
   Settings:
   • in cell Settings to Change
     • dir_2_save directory to save collected data, located in ./data/readback
     • label name of a sample
     • add_label additional label, not taken into further analysis, but cann't be empty
     • env_condition name of condition for measurement like cold, humid or warm
     • reading_duration time of sample collection in [min] (0.2min=12sec)
   • in cell Additional Settings
     • serial_port the port of Arduino connection. In Linux default is /dev/ttyACM0, but might differ. Use Arduino IDE to find the correct port.
6. Check the collected data with 02_Check_Data.ipynb
   from ./soft/main_soft/
   Settings:
   • cell Change to the directory with data
     • dir_data change to the directory with collected data.
7. Convert multiple files into a single one with 03_Data_to_Single_File.ipynb
   from ./soft/main_soft/
   Settings:
   • cell Change to the directory with data
     • dir_data change to the directory with collected data.
       To remember:
   • cell Save to a new file
     • output of the cell, parameter filename_to_save
8. Train a model with 04_TFmodel_Train.ipynb
   from ./soft/main_soft/
   Settings:
   • cell Settings and Paths
     • filename2load this should be the file from 03_Data_to_Single_File.ipynb cell Save to a new file
     • model_filename give here the model name To remember:
   • cell Lets see the classes
     • output of the cell, parameter classes_values
9. Make live prediction with 05_TFmodel_Predict.ipynb
   from ./soft/main_soft/
   Settings:
   • cell Define the classes to be measured
     • classes_values should be list from 04_TFmodel_Train.ipynb cell Lets see the classes output
   • cell set the paths and load data
     • model_filename the name of the model from 04_TFmodel_Train.ipynb cell Settings and Paths
10. Make Streamlit app running
    • change in file ./soft/streamlit_app/streamlit_app.py
      • line 29 : model_ff set models name from 04_TFmodel_Train.ipynb cell Settings and Paths
      • line 33: classes_values should be list from 04_TFmodel_Train.ipynb cell Lets see the classes output
      • line 63: serial_port set correct port of Arduino, can be found via Arduino IDE
    • run the streamlit:
      • cd ./soft/streamlit_app/
      • streamlit run streamlit_app.py

---

Assembling and Connecting to computer

Parts: Arduino Uno, Grove multichannel Gas Sensor V2 (GGSv2), DHT11

Connections table:

Device Name	Device Pin	Arduino Pin
GGSv2	GND	GND
	VCC	3.3V
	SDA	A4
	SCL	A5
DHT11	GND	GND
	VCC	5V
	Serial Data	Digital Pin 2

List of steps:

1. Connect parts according to the table.
2. Connect to computer via USB.
3. Open Arduino IDE (install it if it is not yet done).
4. Open project "GGSv2_n_DHT11_03_10hz.ino".
5. Select Arduino Uno as current board. IDE: Tools -> Board -> 'Arduino Uno'
6. Check/select the port to which the board is connected. IDE: Tools -> Port -> '/dev/ttyACM0'
7. IDE: Sketch -> Verify/Compile
8. IDE: Sketch -> Upload
9. If appears an error "Error opening serial port '/dev/ttyACM0'"
   then in terminal: sudo chmod 666 /dev/ttyACM0.
10. See the output: IDE: Tools -> Serial Monitor

The data format should be like: 12:40:15.641 -> 258,61,148,86,25,29:

• 12:40:15.641 is a record time
• 6 numbers are: NO2 (GM102B), C2H5OH (GM302B), VOC (GM502B), CO (GM702B), temperature [degC], humidity [%]

Schematics of the connections:

---

Features of the project

See in the separate page: Features and Challenges of work with GGSv2