Background

Many complex and structured fluids exhibit a wide range of rheological responses to different flow characteristics. The ability to predict this complex rheological behavior is essential in better understanding and designing these complex fluids and their processing conditions. Despite the increasing trend in using artificial intelligence and machine learning algorithms in many avenues of science, technical issues have precluded their effective implementation in rheology, fluid mechanics and material science and engineering.

Over the past few decades, software packages have been developed to perform fluid mechanical and rheological simulation of a given condition, however, none of these find the same success in industrial settings as they do in academic environments. This is due to lack of accuracy, adaptability, and ease of use. There is a need for a new methodology that utilizes machine learning algorithms in prediction and design of rheological behavior of complex fluid with higher accuracy and adaptability.

Technology Overview

Northeastern spinout, Rheopoint, leverages advances in artificial intelligence and machine learning in solving complex problems in materials science and complex fluids. This technology, uses two interconnected neural networks, each consisting of hidden layers and neurons. In one of the two neural networks, synthetic data generated from conventional models are used as input, and the other one uses actual experimental data on the problem under investigation. This new methodology significantly reduces the number of data points required to perform meaningful machine-learning predictions. This new algorithm also incorporates physical intuition into the neural network to capture the experiment conditions, including temperature, salinity, and aging, which significantly enhances the prediction reliability. In contrast to usual deep learning platforms, using physical models enables this technology to predict behaviors outside the training window. This technology can be used to accelerate material design and discovery, and for predicting the rheological behavior of complex systems across a wide range of conditions.

Benefits

Significantly less data required to provide reliable predictions
Agile and adaptable model for new materials/processes
Enables accelerated material design and discovery by predicting behavior
Fast and accurate modeling of complex fluids in very complex processing/operating conditions conventionally not accessible by other models

Applications

Adaptive modeling of processing/design/behavior of complex fluids in:

Consumer products
Polymers/plastics
Oil/gas
Pharmaceuticals

Opportunity

Building business plan
Applying for grant funding

Rheology-Informed Neural Networks for Complex Fluids

IP Status

Seeking