Thermoplastic Recipe Generation via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing process for manufacturing thermoplastic compounds is time-consuming and costly, often requiring multiple iterations to meet target specifications for multiple material properties, as changes to optimize one property can negatively impact others, such as color and mechanical properties.
Innovation Solution
A computer-implemented method using a machine learnable model to predict material properties from input recipes, generating candidate recipes, and selecting the best one based on a scoring function to approximate target values, with the option to retrain the model if deviations are significant.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If empirical methods are used to determine recipes by iterative testing, then material properties can be optimized, but the development time and cost increase significantly
Solution Approach 1:
The system performs preliminary actions by training a predictive model on historical compound data before actual recipe development. This pre-established knowledge base enables rapid prediction of material properties for candidate recipes, eliminating the need for iterative physical testing and significantly reducing development time while maintaining specification accuracy.
Solution Approach 2:
The patent replaces the mechanical/physical iterative testing system with a computational prediction system. Instead of physically manufacturing and testing multiple recipe iterations, the system uses a trained machine learning model to predict material properties, substituting empirical mechanical testing with intelligent computational analysis.
2Manufacturing precision
If recipes are optimized for one material property, then that property improves, but other material properties may deteriorate
Solution Approach 1:
The predictive model serves multiple functions simultaneously - it can predict various material properties (color, mechanical properties, UV stability, etc.) for any given recipe. This multi-functional capability allows the system to evaluate and optimize multiple material properties together, ensuring that improving one property does not negatively impact others, as all properties are considered in the recipe generation process.
Solution Approach 2:
The system changes the approach from single-property optimization to multi-property simultaneous optimization by using a predictive model that evaluates multiple material properties. The recipe generation process adjusts ingredient proportions and combinations to achieve target values for multiple material properties concurrently, rather than sequentially optimizing one property at a time.
3Manufacturing precision
If multiple iterations of testing and adjustment are performed, then target specifications can be met, but the number of samples to be manufactured increases
Solution Approach 1:
The system creates virtual copies of physical testing through computational prediction. Instead of manufacturing multiple physical test samples to evaluate different recipes, the trained predictive model generates virtual predictions of material properties for numerous candidate recipes. This digital copying approach eliminates the need for multiple physical iterations, reducing sample manufacturing while maintaining specification compliance through accurate predictive evaluation.
Data Source
AI summary
A processor system and method (100) are provided for generating a recipe for a thermoplastic compound, wherein the recipe defines a set of ingredients and a relative contribution of the ingredients for manufacturing the thermoplastic compound. The ingredients may comprise additives to be added to a base polymer. The recipe may be generated by training (110) a machine learnable model on compound data (20) of existing (historical) compounds to predict values of compound material properties from an input recipe, providing (120) candidate recipes, selecting 1 (40) a best recipe based on a scoring function, outputting (150) the selected recipe, e.g., via a display, to enable a sample of the compound to be manufactured (200) and measured (210), receiving (160) measurement data of the sample and determining a deviation to a target specification, and determining (170) if the recipe is acceptable. If the recipe is not acceptable, the machine learned model may be retrained or updated based on the measurement data and the recipe of the sample and the above-identified steps may be repeated until a recipe meets the target specification.