Joint Stiffness Estimation of Body Structure Using Neural Network. Estimating The Joint Stiffness for The Influence Factors.

Abstract

In this paper, an application of hierarchical neural networks to joint stiffness estimation of automobile body structure is described. We deal with two simple joint structures, T-or-L-shape models, which are composed of thin walled box beams combined like capital letters T or L, as typical models of actual body structure. Plate thickness, sectional dimension, length of partition, and position of flange, are considered as influence factors on joint stiffness. The joint stiffness is expressed with a joint stiffness matrix. The influence factors and joint stiffness are used as input and output data for a neural network, respectively. The sample data of some factors vs. joint stiffness are calculated by the finite element method as the training data sets for a neural network. The error-back-propagation neural network is trained using the sample data. Finally, it is found that the neural network after sufficiently trained is useful for estimating the value of joint stiffness.