Machine Learning Model Suitability for Steel Property Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for evaluating mechanical properties using machine learning models are unreliable when there is no suitable model for the input values, leading to significant deviations in predicted values due to factors like metallic structure and scale on the steel sheet surface, or unlearned substances with different reflectance.
Innovation Solution
A calculation method and apparatus that calculate a feature value using machine learning models and determine the appropriateness of these models by comparing input values against predetermined models, with deviation amount calculations and judgment steps to ensure accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models are used to evaluate mechanical properties from sensor measurements, then non-destructive quality assurance is enabled, but significant deviations occur when no suitable model exists for the input values
Solution Approach 1:
The system performs preliminary actions by constructing multiple different machine learning models in advance, each trained on different subsets of training data. Before evaluating mechanical properties, the system has a ready portfolio of models that can handle various conditions, including those with disturbance effects like metallic structures or scale on steel surfaces.
Solution Approach 2:
The system changes parameters by varying the training data subsets used to construct different machine learning models. Each model is trained with different parameters (different training data combinations), creating a diverse set of models that can adapt to different input conditions. This allows the system to select the most appropriate model for each specific evaluation case.
2Productivity
If a single machine learning model is constructed for evaluation, then the evaluation process is simple and fast, but the model may not be appropriate for all input value conditions
Solution Approach 1:
The system segments the single model evaluation task into multiple parallel model evaluations. Instead of using one monolithic model, the system divides the evaluation into multiple segments, each handled by a different machine learning model constructed from different training data subsets. This segmentation allows parallel processing while improving adaptability.
Solution Approach 2:
The system achieves universality by creating multiple machine learning models that collectively cover a broader range of conditions. Each model serves a specific function based on its training data, and together they provide multi-functional capability to handle various material conditions, including those with disturbance effects that a single model might miss.
3Ease of manufacture
If disturbance effects like metallic structure and surface scale are not considered in model construction, then model construction is simpler, but predicted values deviate significantly from actual mechanical properties
Solution Approach 1:
The system performs preliminary action by constructing multiple models in advance, each trained on different training data that may or may not include disturbance effects. This preliminary construction of diverse models allows the system to later select the most appropriate model for each evaluation case, effectively handling disturbance effects without complicating the real-time evaluation process.
Solution Approach 2:
The system implements feedback by comparing the outputs of multiple machine learning models and selecting the most appropriate prediction. The feedback mechanism evaluates which model's predictions are most reliable for the given input conditions, allowing the system to account for disturbance effects like metallic structures and surface scale by choosing models trained on relevant data.
Data Source
AI summary
A calculation method to be used in production or use of a product includes a step (S2) of calculating a feature value using one or more input values selected from a predetermined input value group and one or more first models, and a step (S3) of calculating a deviation amount, which is an amount of displacement from the first model for a predetermined input value from the input value group, using one or more input values selected from the input value group and one or more second models. The second model and the first model are machine learning models generated using one or more pieces of training data respectively selected from a predetermined training data group.