Polymer Property Prediction via Substructure Density Regression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for predicting polymer physical properties require human intervention to determine structural patterns, leading to low accuracy and limited application to polymer structures.
Innovation Solution
A polymer physical property prediction device that automatically calculates substructure numbers, atom counts, and densities, constructs regression models using experimental data, and predicts properties based on input polymer structures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human-determined structure descriptors are used to predict polymer physical properties, then the prediction process can be performed, but the prediction accuracy is low
Solution Approach 1:
The system performs self-service by automatically extracting substructure patterns from polymer structures without requiring human intervention. The substructure pattern extraction unit autonomously identifies and quantifies substructures, enabling the system to improve prediction accuracy while maintaining high automation throughout the prediction process
Solution Approach 2:
The patent replaces the manual mechanical process of human experts determining structure descriptors with an automated computational system. The substructure pattern extraction unit uses computer-based algorithms to automatically identify and quantify substructures, substituting human cognitive work with automated information processing to achieve both high accuracy and full automation
2Extent of automation
If automated substructure extraction methods for low-molecular weight organic molecules are applied to polymers, then automation is improved, but the prediction accuracy remains low
Solution Approach 1:
The system applies local quality by focusing on specific substructure patterns within the polymer rather than treating the entire structure uniformly. The substructure pattern extraction unit identifies and quantifies particular local structural features that are most relevant to the physical properties being predicted, enabling accurate predictions while maintaining automation
Solution Approach 2:
The patent changes the parameters used for structure description from conventional human-determined descriptors to automatically extracted substructure pattern quantities. By transforming the structural information into quantified substructure patterns and using their number densities as predictive parameters, the system achieves both automation and high prediction accuracy for polymer physical properties
3Reliability
If conventional regression models are used with human-determined structure descriptors, then the model construction can be performed, but the prediction reliability is limited
Solution Approach 1:
The system extracts the essential predictive information by taking out and quantifying specific substructure patterns from the complex polymer structures. The substructure pattern extraction unit isolates and counts key structural features, and the regression model construction unit uses these extracted quantities as inputs, simplifying the relationship between structure and properties while improving prediction reliability
Solution Approach 2:
The patent segments the polymer structure into identifiable substructure patterns that can be independently quantified. By dividing the complex molecular structure into discrete substructure units and counting their occurrences, the system creates a simplified representation that maintains predictive reliability while reducing the complexity of the regression model
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A polymer physical property prediction device including a substructure number calculating unit configured to read a structural unit from a storage unit and use the structural unit to calculate numbers each indicating how many substructures are in a polymer, the polymer being formed of repetition of the structural unit, an atom calculating unit configured to calculate a number indicating how many atoms are in the structural unit, a substructure number density calculating unit configured to calculate number densities of the substructures from the numbers of substructures and the number of atoms in the structural unit, a regression model construction unit configured to construct a regression model that predicts a physical property value by using an experimental value of a physical property of the polymer and the number densities of the substructures, the experimental value being obtained from the storage unit, a polymer structure input unit configured to input a polymer structure of which the physical property value is to be predicted, and a polymer physical property prediction unit configured to predict the physical property value corresponding to the input polymer structure by using the regression model, is provided.