Polypropylene Composition Prediction via ML Fingerprinting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The development of new polypropylene compositions is challenging due to the need to balance mechanical, rheological, and aesthetic properties, requiring significant human expertise and resources, and existing methods for predicting material properties are not accurate or efficient for polypropylene compositions.
Innovation Solution
A computer-implemented method that trains a fingerprinting model by mapping polypropylene composition ingredients to analytical features, allowing for the prediction of attributes such as amorphous and crystalline fractions, chemical, and physical properties, using machine learning algorithms and analytical techniques like CRYSTEX QC, to optimize polypropylene composition recipes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a trial-and-error approach is used to develop new polypropylene compositions, then the development process can identify appropriate recipes through iterative testing, but the process becomes time-consuming, resource-intensive, and costly
Solution Approach 1:
The patent applies preliminary action by training a machine learning model in advance using historical composition data and analytical features. This pre-trained model can then rapidly predict properties of new compositions without requiring iterative trial-and-error testing, significantly reducing development time while maintaining prediction accuracy.
Solution Approach 2:
The patent uses copying by creating a virtual model (machine learning fingerprinting model) that replicates the relationship between composition ingredients and analytical features. This digital twin allows prediction of composition properties without physical trial-and-error experimentation, reducing both time and resource consumption.
2Measurement precision
If analytical features including amorphous and crystalline fractions are used in machine learning prediction, then prediction accuracy of polypropylene composition attributes is improved, but the complexity of the prediction model increases
Solution Approach 1:
The patent applies segmentation by dividing the complex prediction task into distinct analytical features: amorphous fraction, crystalline fraction, and other composition properties. Each feature is predicted separately by the machine learning model, allowing complex composition analysis to be broken down into manageable prediction components that improve overall accuracy without overwhelming model complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention relates to a computer-implemented method for predicting attributes of a polypropylene composition. The method includes training a fingerprinting model with polypropylene training compositions by mapping contents of ingredients of the polypropylene training compositions to analytical features of the polypropylene training compositions, feeding the trained fingerprinting model, as an input, with contents of ingredients of a polypropylene test composition, and receiving, as an output, analytical features of the polypropylene test composition. In this context, the analytical features comprise at least one feature characteristic of the amorphous fraction or the crystalline fraction of the polypropylene composition. Furthermore, the present invention provides a computer readable storage medium for storing computer program instructions defining the steps of the inventive method. Besides that, the present invention encompasses the use of analytical features comprising at least one feature characteristic of the amorphous fraction or the crystalline fraction of a polypropylene composition and contents of ingredients of polypropylene compositions for training a computer-implemented model.