Mechanical Assembly Generation With Hybrid Search and AI Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional automated design techniques struggle to efficiently identify functional solutions for complex mechanical systems due to exponential growth in the solution space, requiring extensive manual tuning and lacking interactivity, which limits designer input and domain-specific insights.
Innovation Solution
A hybrid approach combining search algorithms with transformer models to iteratively generate mechanical assemblies, allowing for interactive design workflows and incorporating user feedback, while leveraging generative AI to rapidly explore design spaces and satisfy structural feasibility and design intent.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional automated design techniques are used to explore the solution space for complex mechanical systems, then the completeness of solution exploration improves, but the computational time and complexity increase exponentially
Solution Approach 1:
The patent segments the design space exploration into two distinct phases: (1) a limited sampling phase using conventional search algorithms to generate initial candidate assemblies, and (2) a generative phase using a transformer model to expand and refine these candidates. This segmentation allows the system to balance thoroughness with computational efficiency, avoiding exhaustive exploration while still identifying high-quality solutions.
Solution Approach 2:
The patent performs preliminary action by using search algorithms to generate an initial set of candidate mechanical assemblies before applying the generative transformer model. This preliminary sampling provides a foundation of feasible configurations that the transformer can then refine and expand, reducing the overall computational burden compared to starting from scratch.
2Ease of operation
If conventional automated design techniques are used without manual tuning, then the ease of operation improves, but the quality of functional solutions deteriorates
Solution Approach 1:
The patent implements self-service by enabling the system to automatically tune and optimize its own parameters through the generative transformer model. The transformer learns optimal design configurations and parameters autonomously from the sampled data, eliminating the need for manual expert intervention while maintaining or improving solution quality.
Solution Approach 2:
The system incorporates feedback mechanisms where the results from physics-based simulations and constraint satisfaction evaluations are fed back into the generative transformer model. This feedback loop allows the model to learn from successful configurations and improve its generation of high-quality solutions in subsequent iterations, automatically adapting without manual tuning.
3Extent of automation
If conventional closed-loop design systems are used, then the extent of automation improves, but the interactivity and designer control deteriorate
Solution Approach 1:
The patent introduces dynamics by creating an interactive loop between the automated generative system and the human designer. The system can generate multiple candidate assemblies, present them to the designer for evaluation, and incorporate designer feedback to refine subsequent generations. This dynamic interaction maintains high automation while restoring designer control and interactivity.
Solution Approach 2:
The system establishes a feedback mechanism where designer preferences, constraints, and evaluations are continuously incorporated into the generative process. The transformer model can be retrained or fine-tuned based on designer feedback, allowing the automated system to adapt to human expertise and preferences while maintaining computational efficiency.
Data Source
AI summary
A computer-implemented method is disclosed for generating mechanical assemblies using iterative optimization and generative artificial intelligence (AI). The method includes receiving a mechanical parts catalog and assembly requirements, and executing an iterative generation process. The process comprises generating, via limited sampling, at least one combined mechanical assembly that may satisfy the requirements; generating, via a generative AI model, at least one complete mechanical assembly based on the combined assembly and the requirements; and generating assembly metrics by applying at least one physics simulation to the complete assembly. A reward score is generated based on the assembly metrics, and the iterative generation process is repeated based on the reward score until a convergence threshold is satisfied. The method further includes performing at least one operation associated with the complete mechanical assembly.