Product Disassembly Sequence Planning

Overview

Product disassembly is a critical area of research promoting sustainable development by facilitating effective end-of-life (EOL) strategies such as reuse, remanufacturing, and recycling. In this project, we developed an AI model that leverages 3D data from CAD assembly models to generate feasible disassembly sequences, enhancing the efficiency of EOL processing.

Annotated DMU-Net Dataset

My role focused on creating a structured dataset from raw 3D CAD models. Starting with the Electric Motor dataset, I reverse-engineered the assembly sequences using Fusion 360 and manually annotated the correct disassembly sequences, including the directions for component removal. However, the Electric Motor dataset presented challenges:

1. Complexity and Diversity: The dataset encompassed multiple types of electric motors, each with distinct components, leading to over 20 classes, complicating the model's training process.

2. Non-uniformity: Some 3D models were highly detailed, while others lacked sufficient detail, resulting in an inconsistent dataset.

3. After a team discussion, we discontinued using the Electric Motor dataset and shifted focus to more manageable 3D models. I examined all 29 remaining parts and selected three—Coupling, Shock Absorber, and Rod Piston Assembly—for dataset creation. These selections provided a balanced mix of complexity and uniformity.

I also maintained an Excel sheet detailing component occurrences in each file, generating bar graphs for deeper insights into the dataset's characteristics. This helped refine our understanding and ensure the dataset's utility for our AI model.

Project Outcomes

1. Paper Acceptance: Our research paper, "3D-DSPNET: Product Disassembly Sequence Planning," was accepted at the ICME Workshop 2022. The paper introduced our custom dataset, 3D-DSP, which enhanced the DMU-Net dataset with ground truth annotations for disassembly order and direction. We also proposed 3D-DSPNet, a data-driven approach that significantly outperformed existing baselines in disassembly sequence planning.
   
   

2. Proof of Concept (PoC): We developed a PoC for 3D-DSPNet by integrating a plug-in into Autodesk Fusion 360, allowing users to visualize the disassembly plans generated by our AI model. This PoC validated our system's efficacy and showcased the potential impact of our research in practical applications.
   
   

Through this project, I contributed to advancing sustainable product lifecycle management by enhancing disassembly sequence planning with AI, demonstrating the value of integrating advanced data-driven methods into traditional manufacturing processes.