Multi-Fidelity Neural Network for Complex Fluid Rheology
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current engineering and scientific software packages for fluid mechanics and rheology lack accuracy, adaptability, and ease of use in industrial settings, limiting the application of machine learning algorithms due to technical issues related to complex materials and fluids.
Innovation Solution
A physics-informed neural network framework, specifically a Multi-Fidelity Neural Network (MFNN) architecture, is developed to reduce data requirements and improve predictive accuracy by incorporating physical intuition, enabling reliable predictions of complex fluid behavior across various conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional software packages are used for fluid mechanics and rheology simulation, then industrial applications can be supported, but accuracy and adaptability are insufficient
Solution Approach 1:
The patent combines physics-based constitutive models with machine learning algorithms to create a hybrid framework. The physics model provides theoretical foundation and constraints, while the ML component learns from experimental data to improve predictions. This composite approach resolves the contradiction by integrating the reliability of physics models with the adaptability of data-driven methods.
2Measurement precision
If machine learning algorithms are applied to complex fluids, then predictive capability can be enhanced, but technical issues related to complex materials and fluids limit their use
Solution Approach 1:
The patent introduces physics-based constitutive models as intermediaries between the complex fluid system and the machine learning algorithm. These models serve as a bridge that translates complex fluid behavior into a form that ML can process, reducing the technical complexity barrier while maintaining predictive accuracy.
Solution Approach 2:
The framework transforms complex material properties and processing conditions into standardized input parameters for the neural network. By changing the representation of complex fluid characteristics into suitable parameter formats, the system reduces technical complexity while preserving predictive capability.
3Adaptability or versatility
If traditional phenomenological and empirical modeling methods are used, then existing materials can be modeled, but a lot of time and money are needed to perform experiments and create large data sets
Solution Approach 1:
The patent performs preliminary action by incorporating physics-based constitutive models that encode prior knowledge about fluid behavior. This pre-encoded knowledge reduces the need for extensive experimental data collection, as the model starts with theoretical understanding rather than requiring large datasets to be gathered first.
Solution Approach 2:
The framework uses physics models as simplified copies or representations of complex fluid behavior. Instead of performing numerous physical experiments to understand fluid behavior, the system creates a theoretical copy through constitutive models that captures essential physics, reducing experimental requirements.
4Device complexity
If reduced-order modeling with correlated parameters is used, then model complexity is reduced, but the ability to reflect actual parameters in life is compromised
Solution Approach 1:
The patent creates a universal framework that can handle both simplified and detailed parameter representations. The neural network is designed to accept various input formats and automatically determine the appropriate level of detail needed, making the model multi-functional in terms of parameter representation while maintaining accuracy for the specific application.
Data Source
AI summary
A comprehensive machine-learning algorithm, namely a Multi-Fidelity Neural Network (MFNN) architecture, is disclosed for data-driven constitutive meta-modelling of complex fluids. The physics-based neural networks are informed by underlying rheological constitutive models through synthetic generation of low-fidelity model-based data points.