Machine Learning for Barocaloric Refrigerant Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The selection of suitable Barocaloric materials for refrigeration systems is challenging due to the need for materials that exhibit efficient solid-to-solid phase transitions with minimal pressure application, while also being available and cost-effective.
Innovation Solution
A machine learning model is trained to predict properties such as temperature of fusion (Tfusion) and entropy of fusion (ΔSfusion) for a large set of molecules, enabling the identification of candidate molecules with improved Barocaloric effect characteristics. These candidates are then ranked based on their predicted properties to determine a target molecule for use in refrigeration systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional refrigeration systems use liquid or gas refrigerants, then cooling effect is achieved, but leakage problems occur
Solution Approach 1:
The patent changes the physical state parameter of the refrigerant from liquid/gas to solid, utilizing solid-to-solid phase transitions under pressure to achieve cooling while preventing leakage inherent in liquid/gas systems
Solution Approach 2:
The patent employs solid-to-solid phase transitions of Barocaloric materials under applied pressure to generate cooling effects, replacing the liquid/gas phase transitions used in conventional refrigeration systems to eliminate leakage issues
2Reliability
If Barocaloric materials are selected for refrigeration systems, then leakage is avoided, but material selection becomes challenging due to requirements for efficient solid-to-solid phase transitions with minimal pressure
Solution Approach 1:
The patent replaces complex manual material evaluation processes with machine learning algorithms that automatically predict Barocaloric properties, transforming the material selection from a complex mechanical/experimental process to an efficient computational process
Solution Approach 2:
The machine learning model enables the system to self-evaluate and select optimal Barocaloric materials by predicting their properties, eliminating the need for extensive manual testing and expert analysis in material selection
3Productivity
If machine learning model predicts molecular properties, then candidate molecules can be identified efficiently, but training data requirements increase
Solution Approach 1:
The patent performs preliminary filtering of molecules based on basic criteria (molecular weight, elemental composition) before training the machine learning model, reducing the dataset size and preparing pre-processed data that requires less training volume
Solution Approach 2:
The patent extracts and utilizes only the essential molecular properties and Barocaloric characteristics needed for prediction, removing unnecessary data elements to reduce training data requirements while maintaining prediction accuracy
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach allows for the selection of Barocaloric materials that can achieve a larger temperature change with lower pressure application, while being suitable for use as solids at room temperature, thereby enhancing the efficiency and practicality of refrigeration systems.
Implementation Method 1
Barocaloric materials are substances that exhibit thermal change in response to pressure. In a phenomena known as the Barocaloric effect, a Barocaloric material may undergo a solid-to-solid phase transition (e.g., from one crystal structure to another), and release heat as a result of the transition, when a mechanical force or pressure is applied to the material.
Implementation Method 2
a fan configured to conduct an airflow relative to the container. The airflow may move heated air and/or cooled air generated by applying the pressure to the solid refrigerant.
Data Source
AI summary
A system can train a machine learning model to predict one or more properties of a molecule. The one or more properties may include a temperature of fusion and/or an entropy of fusion. The machine learning model can be trained based on a sample of molecules from a plurality of molecules. The system can apply the machine learning model to the plurality of molecules to predict the one or more properties for molecules of the plurality of molecules. The system can determine a plurality of candidate molecules from the plurality of molecules. The plurality of candidate molecules may be determined based on the one or more properties predicted for molecules of the plurality of molecules. The system can determine a target molecule of the plurality of candidate molecules to implement in a refrigeration system.