Green-Guard

FPGA & LSTM-Driven Smart Irrigation System

EECS 3216 Digital Systems Engineering - Final Project

This project implements an intelligent home automation system for plant care, integrating environmental sensing, hardware control, and a machine learning model for optimal watering recommendations, displayed via a custom VGA output on an FPGA.

Table of Contents

1. Project Overview
2. Demo
3. Goals and Rationale
4. System Architecture
5. Hardware Components
6. Software and AI Model
7. Experimentation and Results
8. Challenges Encountered
9. Getting Started
10. Group Members
11. License

Project Overview

Green-Guard is a real-time embedded system designed to automate plant care by monitoring environmental conditions (soil moisture, humidity), controlling actuators (water pump, fan), and providing visual feedback. A key innovation is the integration of a machine learning model to analyze historical data and recommend optimal environmental conditions to improve plant health. The system leverages a hybrid architecture combining a Raspberry Pi Pico microcontroller for sensing and control with a DE10-Lite FPGA for real-time VGA display of sensor data.

Demo

Check out a demonstration of the system in action: https://www.youtube.com/watch?v=Yn1OlFF0FPE

Goals and Rationale

The primary goal is to design and implement an intelligent system that provides personalized care for plants, enhancing health and reducing risks of over/under-watering.

The rationale is rooted in addressing challenges in:

• Agricultural Applications: Managing environmental variability and stress on crops.
• Conservation: Providing tailored care for sensitive and exotic plant species (like the ghost orchid).
• Home Gardening: Empowering urban dwellers and hobbyists with automated, consistent, and water-efficient plant care.

Engineering goals include contributing to Biodiversity Preservation by supporting diverse plant types and promoting Environmental Sustainability through water conservation.

System Architecture

The system employs a dual-controller architecture for modularity and leveraging the strengths of different platforms:

1. Raspberry Pi Pico: Acts as the sensing and control hub. It interfaces with sensors (DHT11, soil moisture), processes readings, implements threshold-based control logic for actuators (water pump, fan), and encodes sensor data into a binary format for the FPGA.
2. DE10-Lite FPGA: Handles the visual feedback. It receives binary sensor data from the Pico via GPIO, decodes it, converts it to BCD format, and renders real-time humidity and moisture values as 7-segment digits on a standard VGA monitor.

Communication between the Pico and DE10-Lite is achieved using parallel binary GPIO, providing a fast and lightweight data path.

Hardware Components

• Raspberry Pi Pico: Microcontroller for sensor reading and actuator control.
• Intel DE10-Lite: FPGA for VGA display output.
• DHT11 Sensor: Measures ambient humidity and temperature.
• Analog Soil Moisture Sensor: Measures water content in soil (outputs analog voltage).
• 5v Relay Module: Controls the water pump.
• MOSFET Switch: Optionally controls a cooling fan for humidity regulation.
• Small Water Pump: Delivers water to the plant.
• Breadboard, Jumpers, Power Supplies: For system assembly and power distribution.

Software and AI Model

• Raspberry Pi Pico Firmware: Written in Arduino-style C++ using the Arduino IDE. Manages sensor readings, implements control logic based on predefined thresholds, and encodes data for FPGA transmission.
• DE10-Lite FPGA Logic: Developed in Verilog using Intel Quartus Prime. Decodes incoming binary GPIO data, performs BCD conversion, and generates VGA signals to display sensor values. Includes custom logic for rendering 7-segment digits.
• AI Model (Python): A time series forecasting model built using TensorFlow/Keras.
  • Architecture: Long Short-Term Memory (LSTM) network.
  • Input Features: Moisture, Humidity, Watering levels from historical data.
  • Target: Predicted Plant Health score.
  • Training Data: Concatenated data from five distinct environmental scenarios (collected over 14-day cycles on an orchid).
  • Preprocessing: MinMaxScaler for normalization, sliding window approach for sequence creation.
  • Optimization Strategy: Randomized search to find environmental conditions (Moisture, Humidity, Watering amount) that maximize the predicted plant health for the next day.

The AI model is separate from the real-time embedded system but is intended to provide recommendations for optimizing system thresholds or future iterations.

Experimentation and Results

The system was tested under various conditions (different soil moisture levels, simulated humidity).

Key findings:

• The soil moisture threshold ( < 20000 ADC) and humidity threshold ( > 60% ) effectively triggered the water pump and fan, respectively.
• Real-time environmental feedback was successfully displayed on the VGA monitor via the DE10-Lite, with minimal perceived latency despite the Pico's 2-second update rate.
• The parallel binary GPIO communication scheme proved functional for the data transfer, enabling real-time display.
• The system demonstrated reliable closed-loop sensing and actuation based on the defined thresholds.

Challenges Encountered

During development, several challenges were faced:

• Unstable GPIO Data Transfer: Inconsistent connections via breadboard jumpers sometimes caused brief flickering on the VGA display.
• Timing Mismatch: The difference between the Pico's sensor update rate (2s) and the VGA refresh rate (60Hz) occasionally caused temporary visual glitches during data transitions.
• Voltage and Power Issues: Powering high-current actuators from the same rail as logic components caused voltage dips, requiring mitigation with capacitors, flyback diodes, and a separate power source for the pump.
• Moisture Sensor Inconsistency: Sensor readings were sensitive to soil type and environmental noise, necessitating the use of a buffered threshold.

Most of these challenges were addressed through hardware adjustments and minor logic refinements.

Getting Started

To replicate this project, you will need the hardware listed above and the corresponding software environments:

• Arduino IDE (for Pico code)
• Intel Quartus Prime (for DE10-Lite Verilog code)
• Python with necessary libraries (TensorFlow, Keras, scikit-learn, pandas, numpy, matplotlib) for the AI model.

The final code for the Pico (Arduino C++), DE10-Lite (Verilog), and the AI Model (Python) are available in this repository, in dedicated folders (e.g., Verilog, Arduino, ML). Refer to the code comments and the project documentation for detailed setup and usage instructions.

Group Members

• Svastik Sharma (217997081)
• Ahmed Abdulahi (218407635)
• Sathira Williams (218131938)
• Yaroslava Samokhina (217316217)

License

This project is licensed under the Apache-2.0 license.