Yield Criteria Estimation via Evolutionary Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for estimating anisotropic yield criteria of materials are inflexible and computationally expensive, often relying on iterative processes like Newton-Raphson optimization that require initial guesses and gradient calculations, leading to unreliable results and excessive material usage due to potential convergence on local optima.
Innovation Solution
A computer-implemented method using an evolutionary search process to iteratively identify parameters of a function descriptive of yield criteria, which does not require calculating gradients or initial guesses, allowing for more accurate and globally optimal estimates of material properties.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If iterative Newton-Raphson optimization is used to fit yield criteria parameters, then the data fitting process can converge to a solution, but it may converge to a local optimum rather than global optimum, leading to unreliable results
Solution Approach 1:
The patent replaces the traditional Newton-Raphson iterative optimization method with a neural network-based system. The neural network is trained on a database of material test data and yield criteria parameters, enabling it to directly predict accurate yield criteria parameters without iterative optimization. This substitution eliminates the risk of converging to local optima while maintaining computational efficiency.
Solution Approach 2:
The patent performs preliminary work by collecting and storing extensive material test data and corresponding yield criteria parameters in a database before actual use. This pre-prepared database allows the neural network to make direct predictions without requiring real-time iterative optimization, thus avoiding local optimum convergence issues while providing reliable results.
2Manufacturing precision
If iterative optimization methods are used to determine gradient or Hessian matrix, then parameter fitting can be performed, but the computational expense increases significantly
Solution Approach 1:
The patent replaces computationally expensive iterative optimization methods that require gradient or Hessian matrix calculations with a trained neural network system. The neural network, once trained on comprehensive material data, provides direct predictions of yield criteria parameters without requiring real-time iterative computations, thus dramatically reducing computational expense while maintaining modeling precision.
Solution Approach 2:
The patent creates a computational model (neural network) that copies and learns from extensive pre-collected material test data and yield criteria parameters. This trained model can then predict yield criteria for new materials or conditions without requiring expensive iterative optimization, effectively copying the knowledge from the training database to new applications.
3Ease of manufacture
If initial guesses are used as seed values in data fitting processes, then the optimization can start, but poor initial guesses lead to convergence on poor quality results
Solution Approach 1:
The patent replaces the traditional data fitting process that requires manual or automated selection of initial guess values with a neural network system. The neural network, trained on comprehensive material data, directly predicts accurate yield criteria parameters without requiring initial guesses, thereby eliminating the reliability issues associated with poor seed values while maintaining ease of use.
Solution Approach 2:
The neural network system is self-sufficient and does not require external provision of initial guess values or seed data. It autonomously predicts accurate yield criteria parameters based on material test data, eliminating the need for user intervention in selecting initial parameters and ensuring consistent reliable results regardless of user expertise.
Data Source
AI summary
Yield criteria of a material are estimated by obtaining test data representing anisotropic material properties of the material and performing an iterative evolutionary search to identify parameters of a function descriptive of the yield criteria of the material. The evolutionary search includes determining an error value based on a first data point of a first population, where the first data point representing potential values of the parameters. The evolutionary search also includes performing an evolutionary process to generate a second data point as a candidate for replacing the first data point in a second population and determining a second error value based on the second data point. Either the first data point or the second data point is selected for inclusion in the second population. Output data is generated based on estimated values of the parameters that are identified by the evolutionary search.