front-end collision of the car is absorbed by tubular
steel columns installed in the automotive body, which are simply understood as spot-welded hat type
columns. It can be achieved both light and strong structure of the automotive body by increasing
crashworthiness of these energy absorbing members. There are many experimental and theoretical
researches on crashworthiness performance of hat type columns. However, in the previous studies,
influence of spot weld failure during the collision is ignored, even though it is unavoidable and greatly
affect crashworthiness of the member.
Objective of this research is to optimize dimension of the spot-welded hat type member for the
crashworthiness considering spot weld failure during the collision. Spot weld failures shown irregularly
in many experimental results are unpredictable, because of the inherent uncertainties of the spot
welds. In this paper, reliability based design optimization with response surface method is employed
to design reliable hat type member to deal with those uncertainties.
There are many reliability based design optimization algorithms, such as the reliability index
approach (RIA), the performance measure approach (PMA) and the sequential optimization and reliability
assessment method (SORA). These reliability based design optimization algorithms require
- ii -
intensive computation of the probabilistic constraint. Therefore, response surface method (RSM) is
usually used to relieve computational burdens. Consequently the approximation error occurs, which
can make the whole optimization process unreliable. To reduce this approximation error, a new response
surface methodology, “selective sampling method”, collaborated with sequential optimization
and reliability assessment (SORA) is developed. Two optimization problems for spot-welded
hat type member are solved by using this method. It has been demonstrated that the suggested optimization
algorithm with “selective sampling method” can search the most probable point (MPP) accurately
and it can reduce the approximation error of response surface and gives much reliable solution.
The difference between the two optimum designs using deterministic design optimization and
reliability based design optimization is discussed. The optimum mass derived by reliability based design
optimization is slightly greater than the optimum mass derived by deterministic design optimization.
Also, it is shown that the change of constraint has a different effect on those two optimum designs.