Imperial College London

Sensing and IoT Coursework Project

Reducing water waste by automatically dispensing lower bound water amounts, calculated using a ML Model using present weather data

About

This project aims to integrate hardware IoT sensors (some RPi Zero's with I2C probes) and Machine Learning (Deep-NN) to predict how much water a outdoor garden will need every day.

The ML Model uses the last 24hrs of collected weather data to make a prediction around mow much water needs to be dispensed to keep the plants alive.

This approach hopes to reduce the overall water use by not watering plants in excess.

Project diagram

TODO

File Structure

This part is long, click to expand

• data_collector/ Main data collector and uploader module
  • secrets/ Private API keys
    • icl-iot-weather-firebase-adminsdk.json Firebase key for database access
    • weather_api_key.txt OpenWeather API key
  • data_colelctor.py Python script for data collection
  • Dockerfile Containerizing the application
  • requirements.txt Python requirements for running the script
• data_processor/ ML on demand data processing module
  • firebase_python_image/ Since building a tf image takes ages, we build it once and use it for testing the ml script later
    • Dockerfile Building a tensorflow image for armv7l
  • secrets/ Private API keys
    • icl-iot-weather-firebase-adminsdk.json Firebase key for database access
  • watering_model.model/ Saved ML modelfor water predicitons
  • Dockerfile Containerizing the application
  • water_predictor.py On demand, real time watering predictor script
• Diagrams/ Process and block diagrams
  • block_and_process_diagrams.drawio TODO
• lora_nodes/ Scripts running on the Pi Zero nodes
  • master/ Scripts running on the primary, internet connected Pi
    • irrigator/ Daily watering module
      • Dockerfile Containerizing the application
      • irrigator.py Daily watering script
      • requirements.txt Python requirements for running the script
    • Dockerfile Containerizing the application
    • main.py Main data server, collects data from satellite over LoRa and returns over http
    • requirements.txt Python requirements for running the script
  • satellite/ Data reading and sending module for the sensor Pi
    • Dockerfile Containerizing the application
    • main.py LoRa commend listener, temperature reader and pump controller
    • requirements.txt Python requirements for running the script
• manual_data_processing/ iPython notebooks used for data processing and model training
  • datasets/* Various datasets used for processing and training
  • TODO
• site/* Monitoring website, HTML+CSS+JS, hosted on Firebase
• .gitignore Gitignore file preventing all my API keys from showing up online...
• README.md See README.md

Installation and running

1. Clone this repo
2. Create the secrets folders
3. Generate a set of firebase and OpenWeather API credentials and place them in the secrets folders, rename accordingly
4. Move the master folder to the main Pi
5. Move the satellite folder to the sensor Pi
6. Install Docker on all Devices
7. Build the docker images on the relevant architectures (for me everythong was build on armv7)
   • data_colelctor: docker build -t siot-weather-collector .
   • data_processor: docker build -t siot-data-processor .
   • master/irrigator: docker build -t master-node .
   • master: docker build -t irrigator .
   • satellite: docker build -t satellite .
8. NOTE: The ML Dockerfile is very long and complicated because tensorflow does not play well with a 32 bit arm architeture, if you are building for x86 or arm64, you may need to change the file
9. Run the docker containers:

• The data_collector on the cloud device: docker run -d --restart always --name siot_weather_collector siot-weather-collector
• The data_processor on the cloud device: docker run -dp 3535:3535 --restart always --name siot_watering_predictor siot-data-processor
• The master on the master Pi: docker run -dp 3333:3333 --privileged --restart always master-node
• The master/irrigator on the master Pi: docker run -d --privileged --restart always irrigator
• The satellite on the sensor Pi: docker run -d --privileged --restart always satellite

10. While you are free to use my ML model included in this repo, I suggest you explore the manual_data_processing folder and create your own.
11. You will also need at least 24hrs of data before the model can make predictions, so the siot-weather-collector image must be started at least 24hrs before the others

Maintainers and Contributors

Max Hunt: me@maxhunt.design