Alloy Property Prediction Model for Conditions Beyond Past Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting material properties of alloy materials are limited to ranges within past manufacturing track records, making it difficult to select manufacturing conditions that meet target design ranges, and are prone to errors due to biased data and interpolation/extrapolation issues in machine learning models.
Innovation Solution
A manufacturing support system that uses a combination of theoretical expressions and machine learning to predict alloy material properties and select manufacturing conditions, allowing for predictions and condition selection beyond past data ranges by training models with additional parameters like weather data and optimizing manufacturing processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are trained using only past manufacturing track record data, then the model can predict material properties within the ranges of past data, but the model cannot provide accurate predictions for manufacturing conditions outside the past data ranges
Solution Approach 1:
The system performs preliminary actions by collecting and integrating multiple data sources (past manufacturing track records, material composition data, process condition data) before training the machine learning model. This preliminary data preparation and integration enables the model to learn from diverse data patterns, improving both prediction accuracy within past ranges and adaptability to new conditions outside past ranges.
Solution Approach 2:
The system changes the parameters of the training data by transforming raw manufacturing data into structured formats with multiple features (material composition parameters, process condition parameters, property measurement parameters). This parameter transformation enables the machine learning model to capture complex relationships between manufacturing conditions and material properties, improving prediction accuracy and extrapolation capability.
2Measurement precision
If machine learning models rely solely on interpolation within past data ranges, then predictions are relatively accurate within known ranges, but the models cannot identify optimal manufacturing conditions that fall outside historical data ranges
Solution Approach 1:
The system performs preliminary reverse prediction operations by using the trained machine learning model to predict material properties from process conditions, then using these predictions to identify optimal manufacturing conditions that achieve target material properties. This preliminary prediction and iteration process enables the system to suggest innovative manufacturing conditions outside historical data ranges while maintaining prediction accuracy through the model's learned relationships.
Solution Approach 2:
The system applies reverse prediction by inverting the typical prediction direction - instead of only predicting properties from known process conditions, the system uses the model to work backwards from desired material properties to identify optimal process conditions. This inversion enables design flexibility and the discovery of novel manufacturing conditions that achieve target properties, extending beyond historical data ranges.
3Device complexity
If the system uses traditional prediction methods based on human experience and track records, then the approach is simple and interpretable, but the system cannot efficiently handle complex multi-parameter manufacturing conditions and select optimal conditions meeting target design ranges
Solution Approach 1:
The system introduces a machine learning model as an intermediary between raw manufacturing data and optimal condition selection. This intermediary model learns complex relationships between multiple manufacturing parameters and material properties from training data, then efficiently predicts outcomes and identifies optimal conditions. The model acts as a mediator that handles the complexity of multi-parameter optimization, improving productivity while maintaining system interpretability through the learned relationships.
Solution Approach 2:
The system transforms the complexity of multi-parameter manufacturing optimization into a manageable form by using the machine learning model to process multiple parameters simultaneously. The model converts complex multi-dimensional relationships into predictive outputs, enabling efficient condition selection that meets target design ranges. This parameter transformation approach handles complexity internally while providing clear predictive results.
Data Source
AI summary
In accordance with a program, a processor obtains a plurality of manufacturing parameters and a measured value of an at least one property of an alloy material, calculates a pre-predicted value based on a first manufacturing parameter included in the plurality of manufacturing parameters using a prediction expression describing a relationship between the first manufacturing parameter, and a pre-predicted value of the property representing a roughly calculated value of a target predicted value that is a target value of the property, calculates a difference between the pre-predicted value and a measured value of the property, and trains a model using a training data set including a second manufacturing parameter and the difference, to generate a trained model that is used to predict the at least one property.