Welcome to this repository, where we explore a Reinforcement Learning (RL) approach to optimize electronic waste (e-waste) management. By modeling Printed Circuit Board Assemblies (PCBAs) as graphs and leveraging Graph Neural Networks (GNNs), this project efficiently determines the best inspection sequence (e.g. x-ray, visual inspection, flying probe) and, finally, the most profitable recovery strategies—reuse, repair, recycle, or remanufacture. Additionally, Optuna is used for hyperparameter optimization, and a .env file centralizes important configuration parameters.

In the rapidly growing domain of e-waste, optimizing inspection and recovery strategies for PCBAs is pivotal. This project uses:

• Reinforcement Learning for dynamic decision-making.
• Graph Neural Networks to handle variable PCB structures represented as graphs.
• Optuna for automated hyperparameter tuning.
• A .env file to manage crucial parameters (e.g., learning rate, buffer capacity) without hard-coding them.

With this setup, the system dynamically learns the best inspection steps and recovery methods to maximize value while minimizing costs and environmental impact.

• Graph-Based Modeling: Represents PCBAs as graphs so that each component (node) and connection (edge) is captured.
• Reinforcement Learning Agent: Uses RL to optimize sequential decisions (inspection → strategy).
• Optuna Integration: Automates hyperparameter optimization for efficient experiments.
• .env Configuration: Centralizes adjustable parameters (e.g., batch size, learning rate, gamma).
• Economic & Environmental Focus: Balances cost of inspections with potential profits from reuse, repair, or recycling.
• Scalable & Extensible: Facilitates adding new defect types, measurements, or strategies with minimal code changes.

---

Installation - Prerequisites

• Python 3

• PyTorch, Torch Geometric, SimPy, Optuna, Matplotlib

Run

1. Clone the Repository

   git clone https://github.com/Daviecho/pcb_simulation.git
   cd pcb_simulation

2. Create & Activate a Virtual Environment (Optional)

   python -m venv venv
   source venv/bin/activate  # Windows: venv\Scripts\activate

3. Configure your .env file (e.g., episodes, num PCB per episode).

4. Run the main script:

   python main.py

5. Monitor training progress via console output or TensorBoard logs in ./tensorb.

   tensorboard --logdir=tensorb

Hyperparameter Optimization (Optuna)

1. Set initial guesses in .env (or let the script define search ranges).

2. Run optimization script:

   python opt.py

3. Evaluate the Optuna results, which update the RL agent’s configuration automatically or generate a best trial set of hyperparameters.

4. TensorBoard can be used to track the results:

   tensorboard --logdir=tensorb

5. Parameters are saved in folder opt and PCBA information is saved in an sqlite db in runs. In the folder runs the final trained agent is saved, too.