The research presented in the linked article introduces a digital twin (DT) simulation framework designed for smart kitchens to facilitate improved kitchen and usability design.
A digital twin is a simulation process that integrates multidisciplinary and multiscale simulation through multiple sensors collaborative sensing of physical reality. In the context of a smart home, DT technology uses sensors, internet of things (IoT) devices, deep learning (DL), and artificial intelligence (AI) algorithms to collect and analyze data in real-time, creating a virtual model of the home environment.
Digital twin frameworks typically integrate data processing, flow field simulation, equipment monitoring, interactive elements, and visualization techniques to provide real-time insights into operations. For a smart kitchen, DT offers significant benefits for designers, enabling them to design appropriate kitchen appliances, improve development efficiency, and make more accurate predictions. For users, the DT offers valuable insights into the current state and dynamic evolution of the kitchen environment, thereby elevating the overall user experience to a higher level.
One of the most computationally expensive components of the proposed digital twin framework are computational fluid dynamics (CFD) flow simulations. In order to overcome this limitation, the authors implemented an online and cloud based residual-based Fourier neural operator (RFNO) simulation method to achieve accurate results faster.
RFNO works by encoding input data, such as initial conditions, boundary conditions, and physical parameters, along with geometry and location information. The input is then processed through multiple residual modules, each containing Fourier layers. Within these layers, RFNO applies a Fourier transform to extract flow field information, filters high-frequency modes using a linear transform. Given new inputs, an inverse Fourier transform can then be used to convert the data back without performing costly simulations. The proposed RFNO method could be shown to be 1000 times faster than traditional CFD simulations on a flow around a cylinder benchmark.
In the presented work, the native MATLAB and Python programming API available with FEATool Multiphysics, as well as easy to use OpenFOAM GUI for mesh and case file generator, made it a uniquely suitable tool to script, automate, and programmatically generate thousands of sets of simulation data for training the RFNO digital twin framework.
See the linked references for more detailed information about this research.
References
- Sa G., Wu C., Liu Z., et al. A visual digital twin framework based on residual-based fourier neural operator online simulation method, Journal of Simulation, 1-21, 2024, doi: 10.1080/17477778.2024.2394063.