Details
| Originalsprache | Englisch |
|---|---|
| Titel des Sammelwerks | Proceedings of the 31st International Workshop on Intelligent Computing in Engineering |
| Seiten | 146-155 |
| Seitenumfang | 10 |
| Publikationsstatus | Veröffentlicht - 2024 |
| Veranstaltung | 31st International Workshop on Intelligent Computing in Engineering, EG-ICE 2024 - Vigo, Spanien Dauer: 3 Juli 2024 → 5 Juli 2024 |
Abstract
To achieve accurate and fast prediction of air flows with mass and temperature transport in indoor environments for the design phase including exploration and decision-making in the building industry, the high computational effort of computational fluid dynamics (CFD) simulation is a prohibitive barrier. Assistance in decision making in design and engineering requires real-time prediction. Therefore, a method that is able to do model reduction and fast prediction is urgently demanded. This study presents two novel methods, convolutional autoencoder with residual network - supervised neural network (CAER-NN) and convolutional autoencoder without residual network - supervised neural network (CAE-NN), for the real-time reconstruction and prediction of 2D room air ventilation scenarios. Unlike previous studies focusing solely on velocity and temperature variables, this research explores indoor airflow patterns based on varying air inlet heights. The CAER and CAE techniques are employed for dimension reduction, NN serves as a predictor, enabling the entire pipeline to learn the rules and regulations underlying the Navier-Stokes (N-S) equations. Comparative analyses between CAER-NN, CAE-NN, and CFD models are conducted. CAER-NN demonstrates superior performance, with a substantially lower Mean Absolute Error (MAE) compared to CAE-NN, on both training and testing datasets. Additionally, CAER-NN effectively captures airflow characteristics in the main area of the room, which a benefit resulting from the integration of CAE and residual network (ResNet) structure. The study also underscores the importance of tailored evaluation criteria for surrogate models including statistic metrices, flow structures, and physical significance. Last but not least, the prediction speed of this method improved about 105 times.
ASJC Scopus Sachgebiete
- Informatik (insg.)
- Angewandte Informatik
- Ingenieurwesen (insg.)
- Allgemeiner Maschinenbau
Zitieren
- Standard
- Harvard
- Apa
- Vancouver
- BibTex
- RIS
Proceedings of the 31st International Workshop on Intelligent Computing in Engineering. 2024. S. 146-155.
Publikation: Beitrag in Buch/Bericht/Sammelwerk/Konferenzband › Aufsatz in Konferenzband › Forschung › Peer-Review
}
TY - GEN
T1 - Flow Field Prediction with Various Air Inlet Positions Based on Convolutional Neural Network
AU - Wang, Shaofan
AU - Geyer, Philipp
N1 - Publisher Copyright: © 2024 Proceedings of the 31st International Workshop on Intelligent Computing in Engineering, EG-ICE 2024. All rights reserved.
PY - 2024
Y1 - 2024
N2 - To achieve accurate and fast prediction of air flows with mass and temperature transport in indoor environments for the design phase including exploration and decision-making in the building industry, the high computational effort of computational fluid dynamics (CFD) simulation is a prohibitive barrier. Assistance in decision making in design and engineering requires real-time prediction. Therefore, a method that is able to do model reduction and fast prediction is urgently demanded. This study presents two novel methods, convolutional autoencoder with residual network - supervised neural network (CAER-NN) and convolutional autoencoder without residual network - supervised neural network (CAE-NN), for the real-time reconstruction and prediction of 2D room air ventilation scenarios. Unlike previous studies focusing solely on velocity and temperature variables, this research explores indoor airflow patterns based on varying air inlet heights. The CAER and CAE techniques are employed for dimension reduction, NN serves as a predictor, enabling the entire pipeline to learn the rules and regulations underlying the Navier-Stokes (N-S) equations. Comparative analyses between CAER-NN, CAE-NN, and CFD models are conducted. CAER-NN demonstrates superior performance, with a substantially lower Mean Absolute Error (MAE) compared to CAE-NN, on both training and testing datasets. Additionally, CAER-NN effectively captures airflow characteristics in the main area of the room, which a benefit resulting from the integration of CAE and residual network (ResNet) structure. The study also underscores the importance of tailored evaluation criteria for surrogate models including statistic metrices, flow structures, and physical significance. Last but not least, the prediction speed of this method improved about 105 times.
AB - To achieve accurate and fast prediction of air flows with mass and temperature transport in indoor environments for the design phase including exploration and decision-making in the building industry, the high computational effort of computational fluid dynamics (CFD) simulation is a prohibitive barrier. Assistance in decision making in design and engineering requires real-time prediction. Therefore, a method that is able to do model reduction and fast prediction is urgently demanded. This study presents two novel methods, convolutional autoencoder with residual network - supervised neural network (CAER-NN) and convolutional autoencoder without residual network - supervised neural network (CAE-NN), for the real-time reconstruction and prediction of 2D room air ventilation scenarios. Unlike previous studies focusing solely on velocity and temperature variables, this research explores indoor airflow patterns based on varying air inlet heights. The CAER and CAE techniques are employed for dimension reduction, NN serves as a predictor, enabling the entire pipeline to learn the rules and regulations underlying the Navier-Stokes (N-S) equations. Comparative analyses between CAER-NN, CAE-NN, and CFD models are conducted. CAER-NN demonstrates superior performance, with a substantially lower Mean Absolute Error (MAE) compared to CAE-NN, on both training and testing datasets. Additionally, CAER-NN effectively captures airflow characteristics in the main area of the room, which a benefit resulting from the integration of CAE and residual network (ResNet) structure. The study also underscores the importance of tailored evaluation criteria for surrogate models including statistic metrices, flow structures, and physical significance. Last but not least, the prediction speed of this method improved about 105 times.
UR - http://www.scopus.com/inward/record.url?scp=85207181622&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:85207181622
SP - 146
EP - 155
BT - Proceedings of the 31st International Workshop on Intelligent Computing in Engineering
T2 - 31st International Workshop on Intelligent Computing in Engineering, EG-ICE 2024
Y2 - 3 July 2024 through 5 July 2024
ER -