Mechanical Assembly Generation Using Parts Grammar and Physics Metrics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional automated mechanical system design methods are computationally inefficient, unable to provide interactive feedback, and often result in locally optimal solutions due to stochastic behavior, leading to unnecessary structural complexity and increased material usage.
Innovation Solution
A method involving the generation of a parts grammar from a mechanical parts catalog, followed by physics simulations to create valid assemblies, and training a generative AI model using transformer-based techniques to optimize mechanical system designs for minimal weight and cost while satisfying constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional optimization algorithms are used to search configuration space, then mechanical system designs can be generated, but computational efficiency deteriorates due to excessive processing time and resource usage
Solution Approach 1:
The patent pre-generates a comprehensive parts grammar database that encodes all valid mechanical part combinations and their compatibility rules before the actual design search. This preliminary structuring of knowledge allows the transformer model to rapidly retrieve valid configurations during design generation, eliminating the need for time-consuming physics simulations and constraint checking during the optimization process itself.
Solution Approach 2:
The patent replaces traditional mechanical optimization algorithms (evolutionary algorithms, simulated annealing, Monte Carlo tree search) with an AI-based transformer model. This substitution transitions from computational search methods to a learned predictive system that generates designs based on patterns learned from training data, dramatically reducing computational time and resource requirements.
2Adaptability or versatility
If conventional non-iterative optimization methods are used, then design configurations can be obtained, but interactive feedback capability is lost, hindering exploration of alternate configurations
Solution Approach 1:
The patent implements a dynamic, iterative design process where the transformer model can generate multiple alternative configurations in response to user feedback and constraint modifications. The system allows designers to interactively adjust parameters, review generated options, and request regenerating with modified constraints, creating a flexible adaptive workflow rather than a fixed one-pass optimization.
Solution Approach 2:
The patent incorporates feedback loops where design evaluations, constraint violations, and performance metrics are fed back to the transformer model to guide subsequent design generations. This enables the system to learn from previous iterations and adjust its output based on designer preferences and performance requirements, enhancing adaptability while maintaining automation.
3Reliability
If stochastic optimization algorithms are used, then design search can proceed, but solution stability deteriorates due to path dependence and local optimality
Solution Approach 1:
The patent transforms the design search from a stochastic process to a deterministic AI-based generation process. The transformer model uses learned parameters from training data to predict optimal configurations, eliminating random variation and path dependence. This results in more stable and reproducible design outcomes while maintaining high solution quality through the model's learned understanding of mechanical design principles.
Data Source
AI summary
Techniques are disclosed for generating training datasets and training generative artificial intelligence (AI) models for mechanical assembly designs. A method includes receiving a catalog of mechanical parts and generating a parts grammar that defines compatibility relationships between the parts. Using the parts grammar, one or more combined mechanical assemblies are generated, each comprising compatible mechanical parts. Assembly metrics are then generated by applying one or more physics simulations to the combined mechanical assemblies. A dataset is created based on the assemblies and corresponding assembly metrics, and used to train a generative AI model. Training includes executing an iterative training process in which assembly metrics are provided as input to the generative AI model to generate predicted assemblies, comparing the predicted assemblies to ground truth assemblies to compute a transformer loss and a complexity loss, and updating model weights based on an aggregated loss metric until a convergence threshold is satisfied.