Session: 10-02 Fan and System Optimisation

Paper Number: 152434

Data-Driven AI Surrogate Model for Rapid 3D Flow Approximation in Axial Fans 

Computational fluid dynamics (CFD) simulations are crucial for optimizing engineering designs, but they are often time-consuming and computationally expensive, requiring substantial resources for numerous simulations. This paper presents a data-driven AI surrogate model for axial fans, designed to provide rapid 3D approximations of CFD results, thereby reducing the computational effort needed to generate 3D flow solutions. The AI model utilizes existing CFD data to generalize the relationship between geometry, boundary conditions, and flow solutions, enabling it to predict solutions for similar, yet unseen geometries.

This research extends the applicability of the approach to unstructured data for axial fans in industrial applications, building on previous work with structured data for compressor aircraft cases. Notably, this work expands the scope of AI surrogate modeling beyond its traditional application in high-speed aviation, demonstrating its effectiveness in predicting low Mach number, incompressible flows relevant to industrial axial fans. The AI model architecture is highly flexible and adaptable, capable of producing reliable predictions from both structured and unstructured CFD data from various turbomachinery flows. The model is trained using XYZ coordinates from the CFD mesh, rotational speed, and boundary conditions as input parameters to predict flow field variables such as velocities, pressure, and density. A design of experiments (DOE) generated database is used for training, ensuring the model can effectively handle and predict a wide range of geometries.

The AI model is validated on optimized fan designs, demonstrating strong agreement between its predictions and CFD simulation results. This validation confirms the model’s ability to capture essential flow features and provide reliable approximations, even for optimized geometries that were not included in the training dataset.

Presenting Author: Krishna Srinitha Vithanala German Aerospace center (DLR)

Presenting Author Biography: I hold a Bachelor's degree in Electronics and Communication Engineering from India and a Master's degree in Embedded Systems. Currently, I work as a Research Associate at the German Aerospace Center (DLR), focusing on applying AI to fan development.

Authors:

Krishna Srinitha Vithanala German Aerospace center (DLR)
Marcel Aulich German Aerospace Center (DLR)
Christian Voß German Aerospace Center (DLR)
Aysegül Cavus ebm-papst Mulfingen GmbH & Co. KG
Patrick Buchwald ebm-papst Mulfingen GmbH & Co. KG
Florian Herbst German Aerospace Center (DLR)