Neural Network CAD Model Retrieval for Mating Compatibility
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current CAD model retrieval systems are inefficient in determining mating compatibility between mechanical parts, leading to suboptimal design and assembly outcomes due to reliance on geometrical methods that consume significant computing resources and memory.
Innovation Solution
A computer-implemented method using a neural network trained on pairs of Boundary Representations (B-Reps) to output mating scores, enabling efficient retrieval of compatible CAD models by converting input pairs into single embeddings and utilizing a Siamese graph neural network encoder and multi-layer perceptron neural network for accurate compatibility assessment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If geometrical methods are used for determining mating compatibility between mechanical parts, then the retrieval process can be performed, but the computing resources and memory consumption increase significantly
Solution Approach 1:
The patent replaces traditional geometrical methods with a machine learning-based neural network system. The neural network is trained on pairs of Boundary Representations (B-Reps) to predict mating compatibility scores, substituting computationally intensive geometrical calculations with a trained model that processes embeddings more efficiently.
Solution Approach 2:
The patent transforms the mating compatibility determination from direct geometrical comparison to a parameter-based approach using neural network embeddings. By converting B-Reps into embedded representations and using learned parameters (weights and biases) to predict compatibility, the system reduces computational complexity while maintaining accuracy.
2Productivity
If traditional CAD model retrieval systems are used, then the system structure is simple, but the retrieval efficiency and accuracy are insufficient
Solution Approach 1:
The patent implements preliminary action by pre-training the neural network on datasets of B-Rep pairs with mating compatibility labels before deployment. This offline training phase prepares the model to make rapid predictions during actual retrieval operations, improving productivity while the complexity is managed through batch processing.
Solution Approach 2:
The patent uses Siamese graph neural network encoders that process pairs of B-Reps through identical network architectures. This copying approach ensures consistent feature extraction while enabling efficient comparison through embedding similarity, balancing complexity with retrieval efficiency.
Data Source
AI summary
A computer-implemented method of machine-learning for CAD model retrieval based on a mating score. The method includes obtaining a dataset of pairs of Boundary Representations (B-Reps) representing mechanical parts, each pair being labeled with mating compatibility data, the mating compatibility data representing an extent of mating compatibility between the mechanical parts represented by the pair. The method also includes training a neural network based on the dataset, the neural network being configured for taking as input a pair of B-reps representing mechanical parts, and outputting a mating score of a pair of single embeddings, each single embedding corresponding to a B-Rep of the pair, the mating score representing a score of mating compatibility between the mechanical parts represented by the pair.