Chang Woo Choi, Hyo-eun Kang, Yoonyoung Hong, Yong Su Kim, Guem Bo Kim, Aji Teguh Prihatno, Jang Hyun Ji, Seungdo Hong, Ho Won Kim
{"title":"Indoor Space Flow Analysis Based on Deep Learning","authors":"Chang Woo Choi, Hyo-eun Kang, Yoonyoung Hong, Yong Su Kim, Guem Bo Kim, Aji Teguh Prihatno, Jang Hyun Ji, Seungdo Hong, Ho Won Kim","doi":"10.1109/ICAIIC57133.2023.10067105","DOIUrl":null,"url":null,"abstract":"It is essential to perform flow analysis in all spaces where people live. For example, designing the shape of the wing by analyzing the flow flowing through the wing of an airplane, or finding an appropriate air conditioner installation location by analyzing the flow according to the location of the air conditioner in the indoor space. In this study, we propose a deep learning model that performs real-time flow analysis assuming an indoor space that is relatively smaller than outdoor space. Computational Fluid Dynamics (CFD), a traditional method used for flow analysis, is not suitable for this task because it takes a long time to derive simulation results. Thus, the application of deep learning to flow analysis is considered in the present study because deep learning technology for physics, i.e., fluid mechanics and thermodynamics, can be applied to real spaces. We have constructed a deep learning model based on the TransUnet model that can learn data relationships and capture spatial information. Unlike the existing TransUnet model, our model contains a dense layer to reflect operating and spatial information. train and test data were collected using the ANSYS FLUENT commercial program. On 11 test data cases, the average R2 score between the actual and predicted value was 0.884, and the RMSE was 0.047, which are significant results. We used the image of the entire space as well as a cross-section to see how similar the predicted values were to the actual ones, Although a slight error occurred inside the space, It was confirmed that the flow tendency was accurately learned under the given operating conditions. Flow analysis through simulation based on existing numerical analysis methods requires a minimum of 8 hours for processing. However, our proposed deep learning model significantly reduces the time cost of flow analysis as it requires less than 3 seconds.","PeriodicalId":105769,"journal":{"name":"2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2023-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/ICAIIC57133.2023.10067105","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
It is essential to perform flow analysis in all spaces where people live. For example, designing the shape of the wing by analyzing the flow flowing through the wing of an airplane, or finding an appropriate air conditioner installation location by analyzing the flow according to the location of the air conditioner in the indoor space. In this study, we propose a deep learning model that performs real-time flow analysis assuming an indoor space that is relatively smaller than outdoor space. Computational Fluid Dynamics (CFD), a traditional method used for flow analysis, is not suitable for this task because it takes a long time to derive simulation results. Thus, the application of deep learning to flow analysis is considered in the present study because deep learning technology for physics, i.e., fluid mechanics and thermodynamics, can be applied to real spaces. We have constructed a deep learning model based on the TransUnet model that can learn data relationships and capture spatial information. Unlike the existing TransUnet model, our model contains a dense layer to reflect operating and spatial information. train and test data were collected using the ANSYS FLUENT commercial program. On 11 test data cases, the average R2 score between the actual and predicted value was 0.884, and the RMSE was 0.047, which are significant results. We used the image of the entire space as well as a cross-section to see how similar the predicted values were to the actual ones, Although a slight error occurred inside the space, It was confirmed that the flow tendency was accurately learned under the given operating conditions. Flow analysis through simulation based on existing numerical analysis methods requires a minimum of 8 hours for processing. However, our proposed deep learning model significantly reduces the time cost of flow analysis as it requires less than 3 seconds.