CFD Paradigm Shift
Computational Fluid Dynamics (CFD) has evolved into a tool for solving various engineering problems by numerically calculating fluid flows. For a long time, CFD was used as an advanced analysis tool for experts. However, the use of CFD is undergoing a transformation due to advances in open source, automation technologies, the integration of artificial intelligence, and the emergence of user-friendly tools. We summarize this paradigm shift, examining how CFD technology has evolved over time and its current and future directions.
Past and Present
Until the mid-1990s, it was the era of in-house code, developed and used by a small number of experts. It was applied to two-dimensional or very simple three-dimensional problems to elucidate physical phenomena with limited computing power.
Subsequently, with the development of commercial software and high-performance computing, the use of CFD became increasingly widespread. CAD data enabled more precise modeling, and the increased use of clusters enabled large-scale computations, enabling simulations that closely resembled real-world phenomena. Furthermore, CFD was integrated with simulations of diverse physical phenomena, including fluids, heat transfer, structures, chemical reactions, particles, and noise. CFD’s application expanded across diverse industries, spanning healthcare, the environment, safety, and leisure. This expansion not only expanded the scope of its application, but also significantly expanded the scope of its potential to replace experiments, moving beyond its role as a complementary tool.
The need for change
As CFD adoption grows, challenges with current CFD practices are becoming increasingly apparent. Key challenges include high entry costs, expert-centric operations, increasing analysis demands, and the accelerating digital transformation.
Democratization of CFD
CFD democratization is a technological and environmental change that makes CFD analysis, which was previously expensive and dependent on advanced technology, affordable and easy for anyone to use.
The high cost of commercial programs used by most today poses a significant barrier to entry for small and medium-sized businesses, research institutes, and educational institutions. The complex setup and advanced numerical analysis requirements still make them difficult for general designers to access. While demand for applying CFD to rapid performance prediction and optimization from the early stages of product development is increasing, high license fees and the need for expert expertise present significant barriers.
Key factors enabling the democratization of CFD include the proliferation of open source software , automation of CFD workflows, customizable DIY CFD, automation through AI and LLM integration, and the explosion of online tutorials.
The benefits of democratizing CFD include designer-centric analysis (easily usable by product designers who are not CFD experts), reduced development costs (self-analysis without commercial tools for initial analysis), shortened development time (CFD verification before prototype production), and fostered creativity (free experimentation with various designs).
Accelerating Digital Transformation
With the recent acceleration of digital transformation, the demand for simulation-based decision-making is increasing across all industries. However, the long computational times of CFD are a major challenge for simulation-based digital twins. Therefore, a method capable of simulating thermal and fluid changes in real time is required.
The Future of CFD
The major changes in CFD that are currently underway and will accelerate in the future are the proliferation of open source, user-customized DIY CFD, data-driven simulation, and AI CFD .
The spread of open source
Open-source CFD software is a computational fluid dynamics tool whose source code is publicly available, allowing anyone to use, modify, and distribute it freely. Users can develop their own analysis functions based on this software or utilize existing functions to perform analyses. Over the past decade, the number of users has steadily increased, particularly in academia, and it is becoming established as a standard tool that can replace some commercial tools.
The background for the spread of open source includes the high cost of commercial CFD, the need for developer-centric flexibility, the growth of the community, increased use for education/research, and the demand for digital twin and AI integration.
The biggest change brought about by the proliferation of open source is that CFD, previously requiring expensive commercial software purchases, can now be downloaded and used by anyone without additional cost . Another change is that commercial software, previously limited to the number of purchased units, can now be used as needed, without any restrictions on the number of users or CPU cores .
Recently, the demand for open source has grown as the integration of CFD codes with other technologies has become an important issue due to the demand for utilizing digital twins, AI, and LLM.
A representative open-source software is OpenFOAM, a collection of code that serves as a toolbox for developers. Because OpenFOAM is a developer-centric environment, it can be difficult for general users to use. NEXTFOAM’s BARAM® was developed using OpenFOAM as a user-centric GUI solution, and its user base continues to grow.
DIY CFD
“DIY CFD” is an approach that leverages open-source tools to build your own computational fluid dynamics (CFD) analysis environment. It’s emerging as an attractive option for engineers and researchers seeking to break free from the constraints of commercial software and develop customized analysis tools tailored to their needs.
This strategy is experiencing significant growth in demand due to its advantages, including increased productivity in the CFD analysis process, scalability that allows even non-CFD experts to obtain accurate results, flexibility to continuously modify or extend the software to meet user needs, and cost-effectiveness.
DIY CFD is broadly divided into the following four areas depending on the user’s needs.
– CFD code development : Direct development of specialized physical models, boundary conditions, and governing equations based on open source
– Workflow optimization : Standardization and automation of the entire CFD work process from preprocessing, calculation, and postprocessing
– Minimal User eXperience : Development of UX with minimal input based on workflow optimization results
– User-customized CAD : Development of a simple CAD program that provides only the functions required for a specific problem or development of an automatic geometry generation function using parameters
Data-driven simulation
As CFD begins to be utilized from the product design stage, it has become necessary to build a performance database based on design variables and to quickly derive results for a vast number of conditions for use in optimal design. Furthermore, obtaining real-time results based on changing conditions has emerged as a key issue for building simulation-based digital twins. In response to these demands, data-driven simulation technology is advancing.
Traditional methods of constructing surrogate models using CFD analysis result databases to represent aerodynamic coefficients or values for specific regions, and methods of building reduced-order models (ROMs) using proper orthogonal decomposition (POD) to reconstruct values across the entire CFD analysis domain, have been systematized and are used as real-time simulators in various fields.
Research on data assimilation, which improves the accuracy and speed of simulation by integrating experimental and simulation data, is also actively underway.
AI CFD
AI CFD is an advanced technology that complements or replaces traditional numerical analysis-based CFD. It is an approach that predicts or models physical phenomena using machine learning and artificial intelligence algorithms.
It is being widely studied for its ability to make predictions much faster than CFD, to be applied to problems that are difficult to analyze using traditional methods, and to automate the CFD analysis process. It is being widely studied in fields such as Physics-Informed Neural Networks (PINNs), Deep Learning-based turbulence models, and AI-based mesh generation.
Recently, Large Language Models (LLMs) are increasingly being utilized in the CFD field based on their natural language understanding and code generation capabilities. LLMs can automate setup, code writing, and documentation through interactive user interaction, and future integration with digital twins and AI-based physical models is also expected. Extensive research is underway in the following areas:
– Automatic code generation and assistance: Generate dictionary files or codes required by OpenFOAM with natural language commands
– Support for setting up and modifying physics models: Automate the interpretation and tuning of complex equations and settings such as turbulence models and transition models
– Automated workflow construction : Support for writing scripts that automate the entire CFD process from preprocessing to postprocessing
– Natural language-based simulation control: Set up and run CFD analysis with voice or text without GUI
– AI physics model integration: Explain physics-based neural network models such as PINN and CNN with LLM or suggest structures/hyperparameters
There are also projections that in the future, LLM-based CFD interfaces will replace GUIs and serve as the center of an integrated, automated workflow from CAD to analysis to optimization.
Conclusion
CFD was once limited to expensive tools exclusively for experts. However, the proliferation of open-source tools and advancements in automation technology have significantly improved accessibility. In the future, the convergence of custom program development, AI, data-driven technologies, cloud environments, and AR/VR visualization technologies will transform CFD analysis into a smarter environment.
Open source is at the heart of this transformation. DIY CFD, data-driven simulation, and AI CFD are all impossible to achieve within the closed architecture and expensive licensing costs of commercial software. Start investing in open source today.