Problem Statement:
Shipping warehouses face the "box problem": identifying the most cost-effective set of boxes to minimize packaging material usage for a vast array of orders, each containing items with varying dimensions and orientations. This project tackles this challenge by harnessing the power of deep learning.
Solution:
We present a novel deep learning-based approach that streamlines the box selection process for warehouses, leading to significant cost savings in packaging materials. This is a very rare and unique problem. sOur solution leverages:
- Synthetic Data Generation: We create a comprehensive dataset of 100,000 simulated historical orders, accurately reflecting real-world order patterns. This dataset includes:
- Order ID's
- Number of items per order
- Item descriptions
- Item dimensions (x, y, z)
- **Optimization: Taking advantage of existing ILP, LP solvers by identifying how you should frame the problem:
- we use PuLP module in python which can help us in using the ILP and LP solvers.
We also tried it using other approaches
- Deep Learning Model: We employ a state-of-the-art 3D Convolutional Neural Network (CNN) to effectively capture the spatial relationships between items in different orientations within potential boxes. This model is meticulously trained using the synthetic data to:
- Analyze order details (number of items, item dimensions)
- Predict the optimal box configuration, including:
- Predicted box dimensions (x, y, z)
- Predicted number of boxes required (if multiple boxes are necessary)
Winning at SusHacks'24 Wild Card Entry (Aegion Dynamics):
We are thrilled to announce that our innovative solution, "Team DEDSEC - Optimal Box Packing with Deep Learning," has emerged victorious as the Wild Card Entry winner at SysHacks'24, the prestigious nationwide hackathon, sponsored by Aeon Dynamics! This recognition is a testament to the effectiveness and ingenuity of our approach.
Key Features and Benefits:
- Data-Driven Optimization: Our deep learning model learns from a massive dataset, enabling it to make increasingly accurate predictions over time.
- Cost Reduction: By recommending the most space-efficient box configurations, our solution helps warehouses minimize packaging material usage, leading to substantial cost savings.
- Scalability and Adaptability: This approach is designed to handle large datasets of orders efficiently and can be readily adapted to accommodate changing order patterns.
Getting Started:
- Prerequisites: Ensure you have Python (3.x) and the necessary deep learning libraries (TensorFlow, PyTorch, etc.) installed on your system.
- Clone the Repository: Use
git clone https://github.com/your-username/optimal-box-packing.gitto clone this repository. - Set Up the Environment: Create a virtual environment (recommended) and install the required dependencies using
pip install -r requirements.txt. - **Run the final phase notebook file in the guthub 'Final_Phase.ipynb'
Further Enhancements:
- Explore alternative deep learning architectures (e.g., Graph Neural Networks) to potentially improve the model's performance.
- Integrate the solution with a warehouse management system for seamless box selection during order processing.
- Implement a continuous learning mechanism to allow the model to adapt to evolving order patterns over time.
Team DEDSEC:
We are a team of passionate developers and problem-solvers who are dedicated to creating innovative solutions. Our team members include:
- Sasi - Core ML and DL Developer
- Vinod - ML Developer
- Rahaman - UI/UX Designer and Data Analyst
- Siddu - Full Stack Developer.
License:
This project is licensed under the MIT License (https://opensource.org/license/mit).
We are confident that this comprehensive and well-structured README will effectively showcase your project's merits, celebrate your SUS HACKS'24 victory, and attract the interest of the developer community.