Inverse Material Design via Cell and Basis Image Representation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for materials discovery and design are slow due to reliance on experimentation and are incapable of 'custom material design' or 'inverse design,' where materials are developed for specific target properties or end uses, and suffer from limitations such as restricted generality, permutation variance, and memory-intensive image representations.
Innovation Solution
A system and method using deep learning techniques to represent crystal structures as a combination of 'cell' and 'basis' images, allowing for the generation of new materials with specific properties by sampling a continuous latent space, overcoming limitations of existing methods by enabling inverse design of materials from a wider configurational space and linking property prediction with material generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional experimentation methods are used for materials discovery, then material properties can be verified experimentally, but the material development process becomes extremely slow
Solution Approach 1:
The patent performs preliminary computational screening and property prediction using first principles calculations and machine learning models before experimental verification. This allows the most promising materials to be identified in silico, reducing the number of materials that require slow experimental testing while maintaining reliability through subsequent experimental validation of selected candidates.
Solution Approach 2:
The patent introduces computational models (first principles calculations and machine learning predictors) as intermediaries between material composition and experimental verification. These intermediaries rapidly screen vast numbers of materials and predict properties, acting as a filter that identifies which materials warrant expensive and time-consuming experimental testing.
2Measurement precision
If forward models are used to predict material properties from structure, then property prediction is achievable, but inverse design (custom material discovery for target properties) remains incapable
Solution Approach 1:
The patent inverts the traditional forward modeling approach by implementing inverse design capabilities. Instead of only predicting properties from known structures, the system uses machine learning models trained on property data to work backwards from target properties to identify materials that satisfy those requirements. This is achieved through the ML-predicted property models that can evaluate candidate materials generated from target property specifications.
Solution Approach 2:
The patent transforms the design space by changing parameters from fixed material structures to target property specifications. The system accepts desired property values as inputs and uses machine learning models to identify materials whose predicted properties match the targets. This parameter transformation enables inverse design while maintaining the accuracy benefits of first principles-based training data.
3Loss of information
If image-based representation is used for crystal structures, then structural information is captured, but the approach becomes memory-intensive and lacks generality
Solution Approach 1:
The patent segments the crystal structure representation into discrete compositional features and structural parameters that can be processed efficiently. Instead of using monolithic image representations, the system divides the material description into manageable components (composition, structure, properties) that can be handled by machine learning models with reduced memory requirements while maintaining information completeness.
Solution Approach 2:
The patent changes the representation parameters from high-dimensional images to more efficient data structures such as compositional vectors and structural descriptors. This parameter transformation reduces memory intensity while preserving all necessary crystal structure information for property prediction and inverse design, enabling the handling of larger and more diverse material datasets.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
This disclosure relates to application based design of novel materials. Conventional methods utilize laborious experimentation or costly first principles calculations. Conventional data driven techniques use point cloud-based representation for crystal structures, that suffers from permutation variance which is not inbuilt in a material's representation, the DL model has to learn invariance which may be inaccurate. Other methods use image based representation for crystal structures and separate images for each element type to represent the basis, which is memory and time intensive. Since each element is represented by its own image, it is difficult for model to learn chemical environment and neighborhood pattern of each element. The embodiments used image based representation of materials consistent with physical principles. Also, embodiments utilize elements matrix to obtain atoms and their positions from basis images. Thus, any material, irrespective of lattice geometry, and number and types of elements, is represented by only two images.