Imagine a vehicle that can dynamically adjust its aerodynamic profile in real-time, presenting an innovative concept that could change the automotive sector for the better. Artificial Intelligence (AI) possesses the ability to entirely remodel the design and production process of motor vehicles, by enabling the optimisation of a car’s aerodynamics with regard to numerous flow conditions. This could potentially result in enhanced performance, better fuel economy, and a more exhilarating driving experience.
The efficiency of road vehicles relies greatly on their aerodynamic characteristics. Hence, the incorporation of an aerodynamic package can significantly improve the range of future electric automobiles, bringing us one step closer towards our goal of breaking free from fossil fuels.
Wind tunnel testing and computational fluid dynamics (CFD) provide useful tools for enhancing a vehicle’s performance in wind conditions. By generating a flow condition and experimenting with various configurations, the most efficient solution can be determined. Nevertheless, this selection approach may not always be optimal as it may overlook certain parameters or the most effective configuration for a range of flow situations.
The merging of Artificial Intelligence (AI) with wind tunnel testing enables the creation of a responsive, feedback-driven system that can independently adjust vehicle aerodynamics for any external airflow condition (referred to as Wind Tunnel Experiments, or WTEs). The combination of AI and wind tunnel experimentation presents a potent mechanism for enhancing a vehicle’s aerodynamic capabilities.
Advancements in Turbulent Flow Observation Technology: Introducing NEXTFLOW
UC3M (Universidad Carlos III de Madrid) has secured a Starting Grant from the European Research Council (ERC) to back the NEXTFLOW initiative. The primary aim of this project is to create advanced technologies that provide increased efficiency and reliability when measuring turbulent flows.
The utilisation of data mining and AI techniques may have the potential to enhance the aerodynamics of cars, thereby decreasing their adverse impact on the environment.
Developing more dependable methodologies for characterising and regulating turbulent flows (e.g. the fluid motion around an airplane wing) is a significant obstacle in the field of aerodynamics. Unravelling the intricacies of turbulent flows is an ongoing pursuit, as accurately predicting and comprehending their behaviour continues to be a challenging task. Therefore, progress in the field of aerodynamics is reliant on the development of more proficient approaches for comprehending and managing turbulent flows.
“The intricate dynamics of chaotic turbulence make it challenging to fully comprehend using current techniques.” This remark was expressed by Stefano Discetti, the coordinator of the Department of Bioengineering and Aerospace Engineering at UC3M, whilst discussing the NEXTFLOW project.
In modern industrial operations, measuring turbulent flows has become a crucial aspect owing to its widespread presence in diverse applications. Enhanced understanding of turbulence dynamics can be leveraged to boost performance in industries such as transportation. Revamping our approaches for quantifying and controlling turbulent flows can unveil novel prospects for augmenting efficiency in the transportation industry.
Aerodynamics and hydrodynamics specialists concur that turbulent flows play a crucial role in the amount of resistance experienced by various vehicles (e.g. cars, aircraft, and boats) while in motion. Developing a deeper comprehension and investigation of these turbulent flows may potentially result in the enhancement of their efficiency and reduction of the environmental impact caused by their motion.
As pointed out by Discetti, the existing approaches implemented to detect turbulent flow during experiments offer only a partial depiction of its velocity, temperature, or pressure. The objective of this ERC project is to devise sophisticated measurement equipment that integrates Artificial Intelligence (AI) and data mining methods to gain a more comprehensive understanding of turbulence’s dynamic behaviour and enhance control over it.
Enhancing Particle Image Velocimetry with Artificial Intelligence (PIV)
Volumetric Particle Image Velocimetry (PIV) is a technique employed to quantify and scrutinise the three-dimensional flow of fluids. This method involves tracing the trajectory of particles illuminated by a laser light to accurately depict the movement of the fluid. The approach offers a lucid visualisation of the flow, thereby facilitating a deeper comprehension of the fluid’s dynamics and properties.
Scientists are striving to leverage data from high-frequency point probes to capture temporal dynamics more precisely in 3D spaces. As per a recent publication in the journal Experimental Thermal and Fluid Science, researchers from UC3M (Universidad Carlos III de Madrid) have put forth a new data mining approach to refine the precision of particle image velocimetry (PIV) techniques. This approach holds significant potential to boost the accuracy of PIV techniques.
Fluid mechanics equations will be employed by researchers to extract pressure fields from high-precision time-resolved data. The aim is to generate manageable models that can describe flow behaviour and design efficient control algorithms.
The outcomes of these investigations may form the basis for devising innovative methods for precisely evaluating and governing flows in real-world applications. As per Discetti, this has the potential to be advantageous for multiple industries, including aviation, as it could potentially improve their operations and decrease their environmental footprint.
Various Applications of AI in the Aerodynamics Domain
- The application of AI can potentially enhance computational aerodynamics.
- AI can support the research and development of aerodynamic designs.
- The bicycle industry can explore the use of AI to manufacture frames that offer greater wind resistance.
- It serves as a tool for acquiring new knowledge.
- With the aid of available data, it may be feasible to expedite the process of acquiring numerical solutions to practical aeronautical design and development predicaments.
CNNs May Impact Design Procedures
Deep learning can provide a faster and more effective solution for computing functions in high-dimensional spaces, in contrast to traditional methods. In addition, deep learning frameworks like deep neural networks (DNNs) are frequently used in data mining and are ideal for extracting multi-scale features from voluminous, high-dimensional datasets.
Recent studies (Lecun et al., 1998; Taylor et al., 2010; Zuo et al., 2015) have demonstrated the effectiveness of Deep Convolutional Neural Networks (CNNs) in analysing photographs. These networks have exhibited the ability to acquire high-level features even in the presence of strong spatial and temporal correlations in the input.
The potential advantages of utilising Convolutional Neural Networks (CNNs) in the field of fluid mechanics have attracted the attention of researchers because of their capability to adapt to three-dimensional and transient scenarios, as well as their versatility in representing intricate shapes.
This research has revealed that using Convolutional Neural Networks (CNNs) can aid in the design and optimisation of simulations in almost real-time, providing an efficient way of designing. Nevertheless, it is crucial to keep in mind that the overall prediction capability for unfamiliar airfoil geometries from various families is restricted as only three airfoil shapes were employed for training purposes.
Future studies strive to train on a large dataset that encompasses various airfoil families and to expand the training datasets through techniques like translation and rotation, which can convert a limited quantity of input data into a considerably larger set of significantly varied data (Shijie et al., 2017). This development will not only dramatically enhance efficiency but also safeguard the network from acquiring insignificant patterns. Introducing physical constraints on networks, such as mass and momentum conservation, may be accomplished by investigating physical loss functions.