TY - GEN
T1 - A transfer learning framework for crashworthiness analysis based on sphere projection and PointNN
AU - Colella, G.
AU - Lange, V. A.
AU - Duddeck, F.
N1 - Publisher Copyright:
© 2024 Proceedings of ISMA 2024 - International Conference on Noise and Vibration Engineering and USD 2024 - International Conference on Uncertainty in Structural Dynamics. All rights reserved.
PY - 2024
Y1 - 2024
N2 - In the early phase of vehicle development, car manufacturers need to ensure the compliance with strict safety requirements while facing the problem of low data availability. Aiming to support the engineers, we propose a transfer learning (TL) framework for crashworthiness. The design of the new product can be viewed as a broad modification of past ones. For this reason, an attractive concept to investigate is the development of a machine learning (ML) approach able to learn from past designs, transferring the acquired knowledge to the new ones. TL can serve to this aim. Here, we propose a TL framework and apply it to explicatory industrial crash examples. To allow the ML model to learn the topological differences between old and new designs, we propose two approaches: sphere projection- and PointNN- based TL. Both approaches enhance TL, which results in a powerful technique to extract valuable information from limited datasets.
AB - In the early phase of vehicle development, car manufacturers need to ensure the compliance with strict safety requirements while facing the problem of low data availability. Aiming to support the engineers, we propose a transfer learning (TL) framework for crashworthiness. The design of the new product can be viewed as a broad modification of past ones. For this reason, an attractive concept to investigate is the development of a machine learning (ML) approach able to learn from past designs, transferring the acquired knowledge to the new ones. TL can serve to this aim. Here, we propose a TL framework and apply it to explicatory industrial crash examples. To allow the ML model to learn the topological differences between old and new designs, we propose two approaches: sphere projection- and PointNN- based TL. Both approaches enhance TL, which results in a powerful technique to extract valuable information from limited datasets.
UR - http://www.scopus.com/inward/record.url?scp=85212209420&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:85212209420
T3 - Proceedings of ISMA 2024 - International Conference on Noise and Vibration Engineering and USD 2024 - International Conference on Uncertainty in Structural Dynamics
SP - 4151
EP - 4160
BT - Proceedings of ISMA 2024 - International Conference on Noise and Vibration Engineering and USD 2024 - International Conference on Uncertainty in Structural Dynamics
A2 - Desmet, W.
A2 - Pluymers, B.
A2 - Moens, D.
A2 - del Fresno Zarza, J.
PB - KU Leuven, Departement Werktuigkunde
T2 - 31st International Conference on Noise and Vibration Engineering, ISMA 2024 and 10th International Conference on Uncertainty in Structural Dynamics, USD 2024
Y2 - 9 September 2024 through 11 September 2024
ER -