How does it work?

Predictor system by meticulously preparing and analyzing vast amounts of material data, applying state-of-the-art machine learning models, and providing engineers with material property predictions through a user-friendly interface.

Detailed breakdown of how it functions:

1. Data Collection and Preparation: The system begins by gathering data from an extensive database that includes the physical and mechanical properties of hundreds of thousands of materials. This data includes information on various materials such as steels, non-ferrous alloys, and polymers, along with their properties under different conditions like temperature and heat treatment.

2. Data Curation and Normalization: Given the vast amount of data, the system employs a sophisticated data curation process. This involves cleaning the data by eliminating redundant or inconsistent entries, normalizing the data to ensure uniformity, and handling any missing values. A crucial part of this process is transforming categorical data, such as heat treatment conditions, into a numerical format that the machine learning models can process effectively.

3. Classification and Grouping: To manage the complexity of the data, the system classifies materials into specific groups based on their type, family, and processing conditions. This step is essential because it allows the system to create more focused and accurate machine-learning models for each material group.

4. Machine Learning Model Training: Once the data is curated and classified, the system trains multiple machine learning models, each specialized for a particular material group or property. For this, the system uses powerful algorithms like XGBoost and TabNet, which are known for their effectiveness in handling large, tabular datasets and making accurate predictions.

5. Model Testing and Validation: After training, each model is rigorously tested using separate datasets that were not part of the training process. This testing ensures that the models can generalize well and make accurate predictions on new, unseen data. The system evaluates model performance using metrics such as Mean Absolute Percentage Error (MAPE) and correlation coefficient (R2), aiming for high accuracy and reliability.

6. Prediction Process: When a user inputs specific material parameters (such as chemical composition, processing conditions, etc.) into the system, the relevant machine-learning model is selected. The model then uses the input data to predict the desired material properties, such as tensile strength or hardness, under the specified conditions.

7. User Interface and Integration: The predictions generated by the system are presented to the user through a specialized interface. This interface is designed to be intuitive and user-friendly, allowing engineers to easily select materials, define the properties they need to model, and view the results. The system also provides tools for exporting the data for further use in CAE simulations or other engineering applications.