Machine Learning Sensitivity Landscape for Shape Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The calculation of gradient-based sensitivity maps in manufacturing physical objects is computationally intensive, leading to high costs and long design phases, hindering interactive design processes.
Innovation Solution
A method using machine learning to create a calculation device that assigns a sensitivity landscape to shape data records, allowing for the determination of physical shapes with predefined target properties without computationally intensive simulations, by capturing shape data and target properties, and using machine learning to generate sensitivity values for optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If gradient-based sensitivity maps are calculated using computational fluid dynamics (CFD), then sensitivity information for shape optimization is obtained, but computational effort and time costs increase significantly
Solution Approach 1:
The patent pre-calculates sensitivity maps using CFD for a set of representative shapes and stores them in a database. During actual design optimization, the system retrieves pre-computed sensitivity information rather than performing new CFD calculations, dramatically reducing computational time while maintaining accuracy for the queried shape configurations
Solution Approach 2:
The patent creates simplified representations of complex CFD simulations by storing sensitivity map results as reusable data structures. These copied sensitivity results can be quickly retrieved and applied to shape optimization without re-running the full computational fluid dynamics simulation, reducing computational effort while preserving the essential sensitivity information
2Measurement precision
If gradient-based sensitivity maps are calculated using computational fluid dynamics (CFD), then sensitivity information for shape optimization is obtained, but manufacturing costs increase due to high computational resources required
Solution Approach 1:
The patent performs expensive CFD-based sensitivity calculations in advance and stores the results in a database. When designing or optimizing shapes, the system retrieves pre-computed sensitivity data rather than performing new CFD simulations, significantly reducing computational resource requirements and associated costs while maintaining the accuracy of sensitivity information
Solution Approach 2:
The patent creates reusable copies of sensitivity map results from CFD simulations and stores them in a database. These copied sensitivity results can be quickly retrieved and applied to multiple shape optimization problems without re-running the full computational fluid dynamics simulation, reducing both computational effort and manufacturing costs while preserving accuracy
3Measurement precision
If traditional computational methods are used for shape determination, then accurate sensitivity analysis is achieved, but interactive design capability is hindered due to long calculation times
Solution Approach 1:
The patent pre-computes sensitivity maps for a comprehensive set of representative shapes and stores them in a database organized by shape characteristics. During interactive design sessions, the system quickly retrieves pre-computed sensitivity information matching the current shape configuration, enabling real-time feedback and iterative design exploration without the delays of traditional CFD-based sensitivity analysis
Solution Approach 2:
The patent creates reusable sensitivity map data structures from CFD simulations and organizes them in a searchable database. During interactive design, the system copies and retrieves appropriate sensitivity results based on shape similarity, providing immediate feedback to designers while maintaining the accuracy of gradient-based sensitivity analysis, thus enabling true interactive design capability
Data Source
AI summary
Provided is a method for determining a physical shape having a predefined physical target property that includes calculating a sensitivity landscape on the basis of a shape data record for the physical shape with the aid of a calculation device. The calculation device is a machine-taught artificial intelligence device. The shape data record identifies locations at or on the physical shape. For a plurality of these locations, the sensitivity landscape respectively indicates how the target property of the physical shape changes if the physical shape changes in the region of the location. Furthermore, the shape data record for the physical shape to be determined is changed on the basis of the sensitivity landscape in such a manner that the predefined physical target property is improved.