|
CSME 2024/12
Volume 45 No.6
:
563-571
|
|
Investigation of Collision Performance Prediction Method for Anti-Collision Beam Based on MI-MDA-Stacking Algorithm
Zhao-Hui Hu a, Da Zheng b and Shao-Wei Chen c
aAssociate Professor, State Key Laboratory of Advanced Design and Manufacturing for Vehicle Body, Hunan University, Hunan 410082.
bGraduate Student, State Key Laboratory of Advanced Design and Manufacturing for Vehicle Body, Hunan University, Hunan 410082.
cGraduate Student, State Key Laboratory of Advanced Design and Manufacturing for Vehicle Body, Hunan University, Hunan 410082.
|
Abstract:
To optimize computing resources and reduce
labor costs associated with finite element analysis in
the traditional anti-collision beam design process,
constructing a vehicle collision surrogate model has
emerged as an efficient and feasible method for
predicting collision performance. In order to enhance
the prediction accuracy of the surrogate model, a
model selection strategy based on mutual information
theory and the random forest algorithm for the
stacking algorithm is proposed. High-precision
surrogate models are developed for estimating both
maximum collision acceleration and maximum
compression of the anti-collision beam structure
during collisions. Firstly, the fundamental principles of
mutual information and random forest are introduced,
and the algorithm framework is proposed. Secondly,
the validity of the algorithm is validated by using
mathematical test functions. Finally, the proposed
algorithm is employed to construct high-precision
surrogate models that accurately predict collision
performance for the anti-collision beam. The results
demonstrate that these constructed surrogate models
enable quick and accurate predictions of anti-collision
beam performance during collisions. This research
holds significant engineering implications for
enhancing safety designs in vehicle structures.
|
|
Keywords: anti-collision beam, frontal collision, ensemble learning, surrogate.
|
|
|
*Corresponding author; e-mail:
|
©
2024
CSME , ISSN 0257-9731
|