Materials Database With ML Property Prediction and QM Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Materials discovery is hindered by the vastness of unexplored chemical space, incomplete experimental data, and the high computational cost of quantum mechanics calculations, limiting the effectiveness of machine learning models due to a lack of training data.
Innovation Solution
An interactive materials discovery platform using machine learning models and a database that incorporates quantum mechanics calculations, allowing user-directed queries, generative models to create candidate structures, and a feedback loop for improving machine learning models with QM-calculated data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If quantum mechanics calculations are used to obtain accurate material property data, then measurement precision is improved, but use of energy and computational cost increase
Solution Approach 1:
The system performs quantum mechanics calculations in advance to generate a training dataset, storing the results in a database. This preliminary action allows the machine learning model to be trained on accurate QM data without requiring real-time QM calculations, thus achieving high measurement precision while reducing subsequent computational costs.
Solution Approach 2:
The system creates a copy of the accurate quantum mechanics calculation results by training a machine learning model on QM-calculated data. The trained model then serves as a surrogate that can predict material properties with high accuracy without requiring additional expensive QM calculations, effectively copying the benefits of QM accuracy at lower computational cost.
2Productivity
If machine learning models are trained on existing experimental data, then productivity is improved, but measurement precision deteriorates due to incomplete data
Solution Approach 1:
The system merges experimental data with quantum mechanics calculation data to create a comprehensive training dataset. This combination allows the machine learning model to be trained on both real experimental measurements and accurate theoretical predictions, improving both the completeness of the training data and the precision of predictions while maintaining high productivity.
Solution Approach 2:
The system uses quantum mechanics calculations as an intermediary to generate missing or incomplete material property data. When experimental data is incomplete, the QM calculations provide the missing information, which then serves as additional training data for the machine learning model, improving overall data completeness without reducing discovery speed.
3Reliability
If the database includes only experimentally characterized materials, then reliability is improved, but extent of automation deteriorates due to limited coverage
Solution Approach 1:
The system enables automated generation and validation of material data through the machine learning pipeline. The model can predict properties for new materials, and these predictions can be automatically added to the database, allowing the system to expand its coverage adaptively while maintaining quality through the rigorous ML training process on validated QM and experimental data.
Solution Approach 2:
The system changes the criteria for database inclusion from requiring complete experimental characterization to accepting machine learning predicted properties based on the trained model's confidence and performance metrics. This parameter change in data acceptance criteria expands database coverage while maintaining reliability through the robust ML training framework.
Data Source
AI summary
Examples are disclosed that relate to materials discovery using machine learning models. One example provides a method enacted on a computing system. The method comprises receiving a query comprising one or more of element information and material property information, and, based on the query, retrieving material data from a materials information database. The material data comprises structural information for each material within a set of materials matching the query, the set comprising one or more materials, and for one or more materials in the set of materials, one or more predicted material properties determined using one or more trained machine learning models. The method further comprises outputting the material data.