When other experimental approaches are too limited to effectively predict a fluid production process across a variety of operating conditions, engineers can create a CFD digital twin. Here are four steps for the industrial practitioner.
A digital twin is a digital replica of a physical system. Designed to run real-time numerical experiments based on relevant mathematical models, the predictions from a digital twin should be identical to real experimental data—hence, twin.
Thanks to modern algorithms and advancements in graphics processing unit (GPU) hardware, digital twins are becoming an optimal method for engineers who are trying to optimize processes with complex fluid mechanics and design limitations that make other approaches inapplicable or too limited.
For a successful digital twin, four things need to be true:
Literature correlations and numerical modeling are not applicable and/or physical experiments are too limited.
The digital twin produces repeatable results that are indistinguishable to experimental data.
The digital twin takes less time and resources to develop compared to experimental measurement, at a much lower cost.
The digital twin generates process design correlations.
For certain applications, developing a digital twin makes sense. Let’s take a look at an example to see why.
Use Case for a CFD Digital Twin Model
In the biopharmaceutical manufacturing process, it’s common to mix two miscible liquids with differences between fluid density and viscosity. But these differences make predicting mixing processes within agitated tanks complicated, causing significant processing and scale-up challenges.
To optimize pharmaceutical blending and mixing, manufacturers can hybridize the three traditional approaches—literature correlations, numerical modeling, and experiments—with a digital twin. This is because too few literature correlations are applicable to multi-fluid systems with large variations in density and viscosity, experiments are limited by costs and access to equipment, and the transport physics involved limit the applicability of numerical modeling approaches.
In this example, a digital twin that pairs Lattice Boltzmann–based transport algorithms with GPU resources allowed the pharmaceutical manufacturer to simulate minutes/hours of fluid mechanics within hours/days of computer wall time. The transient processing insights that the twins generated rivaled experimental data—but at a cost orders-of-magnitude lower.
These three equations together can be solved via the Lattice Boltzmann method.
3. Build the Digital Twin Model
To build the digital twin model—and solve the above equations—you need a tool that supports the Lattice Boltzmann method.
The Lattice Boltzmann method is high-resolution, which means direct numerical simulation and/or large-eddy simulation can be used to build a digital twin that can handle laminar, transitional, and turbulent flow regimes with no reconfiguration.
Compared to traditional finite element and finite difference approaches, the Lattice Boltzmann approach can model multiphase and multiphysics transport processes in fluid mechanical systems at much faster computational speeds. This speed is only further amplified when run in a GPU-based computing environment.
Modern CFD software solves Lattice Boltzmann algorithms on GPUs, which can be used to build digital twins and quickly produce detailed, accurate process simulations.
4. Validate the Model
But before you can apply the model, you have to validate it.
In our pharmaceutical blending example, the manufacturer validated the twin by first comparing the single fluid blend times and power numbers predicted from the twin to experimental data across a wide range of Reynolds numbers. From there, they used the twin to explore blending in two-fluid, density stratified systems. Then, they verified output against experimental data taken at multiple impeller speeds.
The takeaway: Before you can use the twin for process optimization and design on unmeasured systems, you must first measure and supply relevant fluid properties to the twin. Appropriate experimental data is crucial for validating output.
The initial development of a digital twin does not eliminate the need for experiment. However, once developed, it can be used to generate processing insights with a fidelity that rivals experimental measurement at a much lower cost.
When you’re dealing with complex fluid mechanics, the initial set up of a digital twin is worth it—especially when backed by modern CFD software.
This week, guest Aaron Sarafinas, Principal of Sarafinas Process & Mixing Consulting LLC, joins John to discuss one loaded question: “Will CFD or some CFD parallels ever replace experiment, and in an increasingly digitized world, what’s the role of experiment?”
This week, guest Piero Armenante, Distinguished Professor of Chemical Engineering at NJIT, joins John to talk about his background in mixing, what it takes to join his research group at NJIT, and the hybridization of CFD and experiment in his research.