TY - JOUR
T1 - Interior Wind Noise Prediction and Visual Explanation System for Exterior Vehicle Design Using Combined Convolution Neural Networks
AU - Park, Ha Eun
AU - Jung, Hoichan
AU - Lee, Min Seok
AU - Kim, Doohyung
AU - Lee, Jongwon
AU - Han, Sung Won
N1 - Funding Information:
This research was supported by Brain Korea 21 FOUR. This research was also supported by a Korea TechnoComplex Foundation Grant (R2112651) and Korea University Grant (K2107521, K2107521). Please send any inquiry to the corresponding author, Sung Won Han (School of Industrial Management and Engineering, Korea University, Seoul, 02841, Republic of Korea. e]swhan@korea.ac.kr).
Publisher Copyright:
© 2022, KSAE.
PY - 2022/8
Y1 - 2022/8
N2 - An analytical model configuration, in addition to air pressure analysis and post-processing, was conducted to measure the interior wind noise by changing the exterior vehicular design. Although wind noise can be calculated accurately through the current process, it requires three to five days for each design. In this study, a convolutional neural network (CNN), which is a class of deep neural networks designed for processing image data, was applied to predict the wind noise with vehicle design images from four different views. Feature maps were extracted from the CNN models trained with images of each view and concatenated to flow through a sequence of fully connected (FC) layers to predict the wind noise. Moreover, visualization of the significant vehicle parts for wind noise prediction was provided using a gradient-weighted class activation map (GradCAM). Finally, we compared the performance of various CNN-based models, such as ResNet, DenseNet, and EfficientNet, in addition to the architecture of the FC layers. The proposed method can predict the wind noise using vehicle images from different views with a root-mean-square error (RMSE) value of 0.206, substantially reducing the time and cost required for interior wind noise estimation.
AB - An analytical model configuration, in addition to air pressure analysis and post-processing, was conducted to measure the interior wind noise by changing the exterior vehicular design. Although wind noise can be calculated accurately through the current process, it requires three to five days for each design. In this study, a convolutional neural network (CNN), which is a class of deep neural networks designed for processing image data, was applied to predict the wind noise with vehicle design images from four different views. Feature maps were extracted from the CNN models trained with images of each view and concatenated to flow through a sequence of fully connected (FC) layers to predict the wind noise. Moreover, visualization of the significant vehicle parts for wind noise prediction was provided using a gradient-weighted class activation map (GradCAM). Finally, we compared the performance of various CNN-based models, such as ResNet, DenseNet, and EfficientNet, in addition to the architecture of the FC layers. The proposed method can predict the wind noise using vehicle images from different views with a root-mean-square error (RMSE) value of 0.206, substantially reducing the time and cost required for interior wind noise estimation.
KW - Convolutional neural networks (CNN)
KW - Gradient-weighted class activation map (Grad-CAM)
KW - Image regression
KW - Wind noise prediction
UR - http://www.scopus.com/inward/record.url?scp=85135756730&partnerID=8YFLogxK
U2 - 10.1007/s12239-022-0088-9
DO - 10.1007/s12239-022-0088-9
M3 - Article
AN - SCOPUS:85135756730
SN - 1229-9138
VL - 23
SP - 1013
EP - 1021
JO - International Journal of Automotive Technology
JF - International Journal of Automotive Technology
IS - 4
ER -