Material Embedding for Rubber Recipe Property Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for predicting rubber material properties struggle due to nonlinear characteristics and fail to accurately consider similarity between materials, leading to limitations when using various input materials, especially with categorical data that requires embedding into numerical vectors, where one-hot encoding lacks spatial efficiency and Word2Vec's natural language similarity hypothesis may incorrectly define material similarity.
Innovation Solution
A method and apparatus that utilize material embedding to convert input materials into vector data, considering correlations between materials, and input this data into an artificial neural network (ANN) for predicting recipe properties, employing techniques like K-means clustering and Word2Vec embedding to reconstruct data and improve prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If one-hot encoding is used to convert categorical material data into vectors, then the data can be processed numerically, but the dimensionality increases significantly and similarity between materials cannot be captured
Solution Approach 1:
The patent transforms material data from high-dimensional one-hot encoding to low-dimensional dense vectors by changing the parameter representation from sparse binary vectors to continuous numerical vectors that capture material similarities, thereby reducing dimensionality while improving prediction accuracy
Solution Approach 2:
The patent introduces a new dimensional representation for material data by projecting categorical material features into a continuous vector space where materials with similar properties are positioned closer together, enabling the model to capture material similarities without increasing dimensionality
2Adaptability or versatility
If Word2Vec embedding is used to define material similarity, then the natural language similarity hypothesis can be applied, but completely irrelevant rubber materials may be incorrectly defined as similar
Solution Approach 1:
The patent applies local quality by defining material similarity based on domain-specific chemical and physical properties rather than general patterns, ensuring that similarity definitions are appropriate for the specific rubber material context and avoid incorrect associations between irrelevant materials
Solution Approach 2:
The patent introduces domain experts and domain-specific knowledge as intermediaries to guide the embedding process, using chemical structure, physical properties, and application characteristics as mediating factors to ensure accurate similarity definitions for rubber materials
3Adaptability or versatility
If various input materials are added to improve prediction coverage, then the model can handle more material types, but the rubber recipe complexity exponentially increases
Solution Approach 1:
The patent creates a universal material representation framework that can handle diverse material types through a common embedding approach, allowing the system to process various rubber materials and additives uniformly without requiring separate processing logic for each material type, thereby controlling complexity while maintaining versatility
Data Source
AI summary
A method and apparatus for predicting a recipe property reflecting similarity between chemical materials are provided. The method of predicting a recipe property reflecting similarity between chemical materials, the method includes substituting a plurality of input materials with vector data, respectively, through material embedding and generating recipe data including pieces of vector data selected by considering a correlation between materials, and inputting the recipe data to an artificial neural network (ANN) prediction model and deriving a property prediction result on the recipe data from the ANN prediction model.