Machine Learning Material Property Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning methods for discovering new materials and optimizing material properties face challenges in finding global minima or maxima, as they often require excessive computation time and lead to unacceptable errors due to the lack of analytical methods and inefficiencies in numerical solution strategies.
Innovation Solution
A machine learning system that maps a training dataset of material compositions and properties from an input space to an output space to create a convex function, learns this function using gradient descent, and maps the minimum back to the input space to predict optimal material properties and compositions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If numerical solution strategies are used for global optimization, then material property optimization can be achieved, but computation time becomes excessive and error becomes unacceptable
Solution Approach 1:
The patent replaces traditional numerical optimization methods with a machine learning-based approach. A neural network model is trained to predict material properties and guide the optimization process, substituting the mechanical/numerical computation system with an intelligent system that can find global minima more efficiently without excessive computation time or error.
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between the material composition space and the property optimization goal. This intermediary model learns the complex relationships from training data and provides guidance for finding optimal compositions, acting as a mediator that reduces the direct computational burden and improves accuracy.
2Loss of time
If analytical methods are used for global optimization, then computation time is reduced, but they are typically not available for material property optimization
Solution Approach 1:
The patent creates a computational model that copies the complex relationships between material compositions and properties from training data. Instead of deriving analytical formulas (which are not available), the model learns and stores the relationships through training, providing a versatile approach that works for various material systems without requiring explicit analytical expressions.
Solution Approach 2:
The patent changes the approach from seeking analytical solutions to using data-driven parameter learning. The machine learning model learns optimal parameters from training data, allowing the system to adapt to different material systems and property optimizations without requiring analytical methods, thus achieving both speed and versatility.
Data Source
AI summary
A machine learning system includes a processor and a memory communicably coupled to the processor. The memory stores an acquisition module, a mapping module, a machine learning module, a fitting module, and a minimization module that include instructions that when executed by the processor cause the processor to: select a training dataset, map the training dataset from an input space to an output space such that the mapped training dataset is convex; train a machine learning model to learn a convex function that approximates the mapped training dataset in the output space; learn a minimum of the convex function; map the minimum of the convex function to the input space; and predict, based at least in part on the minimum of the convex function mapped to the input space, an optimum material property value and a corresponding material composition.