Disassembly Sequence Planning Using Graph Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Disassembly sequence planning remains challenging due to combinatorial explosion and complexity in determining feasible disassembly sequences, especially with discrete parameters and actions, limiting the efficiency of existing optimization-based and search methods.
Innovation Solution
The use of 3D model-based graph representations and an encoder/decoder system to generate disassembly data, including order and direction for component removal, through graph embeddings and message passing techniques, facilitating a data-driven approach for disassembly planning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional optimization-based methods or search methods with heuristic rules are used to generate disassembly sequences, then feasible disassembly sequences can be found more robustly in lesser time, but the complexity increases significantly due to combinatorial explosion from discrete parameters and actions
Solution Approach 1:
The patent replaces traditional mechanical optimization and search methods with a deep learning-based system. An encoder processes 3D model data and component relationships to generate embeddings, which a decoder then uses to predict disassembly sequences. This substitution of mechanical algorithms with neural network-based processing reduces computational complexity while maintaining feasibility assessment capabilities.
Solution Approach 2:
The patent transforms the discrete parameter space of traditional methods into continuous vector representations through embeddings. By converting discrete component relationships and spatial configurations into continuous vector spaces, the system enables smoother optimization and reduces the combinatorial explosion inherent in discrete search spaces.
2Reliability
If global search methods are used to explore all possible disassembly sequences, then comprehensive solutions can be found, but the time required increases significantly compared to optimization-based methods
Solution Approach 1:
The patent performs preliminary encoding of 3D model data, component geometries, and relationship information into embeddings before the actual sequence prediction. This preprocessing step organizes and compresses the search space information, enabling the decoder to efficiently predict sequences without exhaustive exploration, thus reducing time while maintaining solution quality.
Solution Approach 2:
The patent creates vector representations (embeddings) as copies of the actual component data and relationships. These embeddings capture essential features without requiring processing of the full discrete search space, allowing efficient prediction of disassembly sequences that would otherwise require comprehensive global search.
Data Source
AI summary
Systems and methods that support functionality for performing disassembly sequence planning are disclosed. The systems and methods may obtain 3D model data for a product to be disassembled and may convert the 3D model to one or more graph-based representations. An encoder is provided to generate a set of embeddings based on the one or more graph-based representations. A decoder is provided to generate disassembly data based on the set of embeddings. In an aspect, the disassembly data may indicate an order in which each component is to be removed during disassembly of the product and a direction in which each component should be removed.