In a recent article published in the journal Scientific Reports, researchers presented a novel framework for the generative design of complex functional structures using machine learning and fluid dynamics. They developed a method that leverages evolutionary algorithms to create innovative designs for static mixers, which are crucial components in various industrial processes. Moreover, they showcased the potential of generative design by creating customized reactor elements for applications in flow chemistry.
End to end workflow for the design of novel static mixers. Solutions are represented as strings of floating-point numbers, which are transformed into a solid body geometry whose expression depends on the representation used. Geometries are assessed in a Computational Fluid Dynamics solver, and two performance metrics are measured. The main loop feeds these metrics back into an Evolutionary Algorithm which iteratively improves the set of optimal trade-offs between these objectives. If termination criteria are met, the final designs are 3D printed and their performance experimentally confirmed. Image Credit: https://www.nature.com/articles/s41598-024-62830-5
Background
Flow chemistry and static mixers are essential in industries such as chemical processing, pharmaceuticals, and food production. Flow chemistry allows for continuous chemical reactions, resulting in benefits such as scalability and greener operations. However, designing efficient flow reactors is a complex task that involves optimizing multiple parameters, including mixing, heat transfer, and mass transfer.
Static mixers are used to facilitate the efficient mixing of fluids, gases, or solids continuously. They ensure the uniform distribution of components, which is crucial for various industrial applications. Traditionally, the design of static mixers has relied on manual iteration and trial-and-error methods, which can be time-consuming and limited in their ability to explore the vast design space.
Generative design, a computational method that uses machine learning techniques such as evolutionary algorithms, presents a solution to these challenges. It can explore a vast, complex design space, and produce diverse and innovative designs for both flow reactors and static mixers.
About the Research
In this paper, the authors developed a generative design approach that combines evolutionary algorithms with computational fluid dynamics (CFD) simulations for designing complex functional structures. Their framework consists of three interconnected components: a generative model, a performance model, and a generative design loop.
The generative model was tasked with creating a wide range of bespoke reactor elements. It employs an evolutionary algorithm that leverages mutation and crossover operators to generate new designs from an existing pool. The authors explored tree-like (T) and ribbon-like (RB) geometric representations for the generative design process. T geometries were inspired by bioinspired designs, while the RB geometries were parametrically defined using a series of points in three-dimensional (3D) space, which were then meshed into various RB shapes.
The performance model was responsible for evaluating the reactor elements' efficiency in terms of mixing quality, heat transfer, and mass transfer. It utilizes a scalable fluid dynamics solver based on the lattice Boltzmann method to simulate fluid flow and transport phenomena within the reactor elements. To validate the results, the researchers 3D printed several promising mixer designs in stainless steel and conducted experimental tests in a benchtop setup. The outcomes were then compared to the CSIRO v2 mixer, which represented the current state-of-the-art performance in the target application.
Lastly, the generative design loop iteratively refined the generative and performance models to produce optimal and innovative reactor elements. It adopts a multi-objective optimization approach that strikes a balance between various performance criteria and design complexity.
The study utilized the non-dominated sorting genetic algorithm (NSGA2) to iteratively generate and optimize mixer designs. The algorithm generated a population of candidate solutions, which were then evaluated using CFD simulations based on two objective functions: substrate transport and bulk mixing. The evolutionary algorithm was run for 150 generations, with each iteration producing 16 new candidates. The final Pareto fronts were analyzed to identify high-performing mixer designs.
Research Findings
The authors employed their framework to design bespoke mixers for flow chemistry applications, significantly outperforming existing mixers like T-mixers, Y-mixers, and split-and-recombine mixers by 45% in mixing quality. Additionally, they discovered unique static mixer geometries exhibiting complex, non-intuitive flow patterns such as helical, swirling, and chaotic flows. T geometries showed a 45% improvement in copper extraction compared to the benchmark CSIRO v2 mixer, while RB geometries excelled in bulk mixing, reaching up to 70%, despite slightly lower substrate transport values.
The paper highlighted the adaptable, plug-and-play nature of the framework, which can be easily tailored for various applications. The generated Pareto fronts provide a library of diverse solutions, often eliminating the need for further optimization. The study also showcased the framework's versatility and scalability by designing reactor elements for laminar, transitional, and turbulent flow regimes, achieving efficient mixing and heat/mass transfer across a wide range of Reynolds numbers.
Applications
The proposed generative design framework offers several benefits for industries that depend on complex functional structures, including chemical, pharmaceutical, biomedical, and energy sectors. By automating the design process, the framework can unveil innovative solutions that boost operational performance and sustainability.
For instance, the framework can create flow reactors for complex chemical reactions, such as multistep synthesis, photochemistry, and electrochemistry, with high efficiency and selectivity. It can also optimize heat exchangers for thermal management and energy recovery in industrial processes. Furthermore, the framework can develop biomedical devices for manipulating fluids and cells in diagnostic and therapeutic applications.
Conclusion
In summary, the researchers demonstrated the transformative potential of generative design for creating intricate functional structures that surpass the limitations of traditional design methods. They emphasized the significance of integrating machine learning and fluid dynamics for in silico performance assessment of various design alternatives. Furthermore, they introduced new possibilities for designing customized and optimal structures for flow chemistry and other applications that require efficient fluid flow and transport phenomena.
Boost Machine Learning Trust With HEX's Human-in-the-Loop Explainability