Search Results

Auto components with large manufacturing variation may cause vehicle quality problems after they are assembled. The impact of this variation depends on the assembly process used. If the assembly process is sensitive to the component variation, the impact may be more significant. In this case, an assembly process with lower sensitivity to component variation will solve the problem. This paper presents an example where the component variation largely impacted the quality of the car, and a more robust assembly process solved the problem.

Robust optimization usually requires numerous functional evaluations, which is not feasible when the functional evaluation is time-consuming. Examples in automobile industry include crash worthiness/safety and fatigue life simulations. In practice, a response surface model (RSM) is often used as a surrogate to the CAE model, so that robust optimization can be carried out. However, if the error in the RSM is significant, the solution based on the RSM can be invalid. This paper proposes a method of finding multiple candidate solutions, all of which have similar predicted performances. This approach is effective in finding the close-to-optimum solutions when the model has error, and providing design alternatives. Examples are provided to illustrate the method.

In performing stochastic simulations using computer models, the method of sampling is important. It affects the quality and the convergence speed of the results. This paper discusses one special case: sampling of spot-weld locations from potentially thousands of spot welds on a vehicle body. This study is prompted by the need of evaluating the effect of missed spot welds on the structural integrity, identifying critical welds, and optimizing weld locations. A balanced random sampling algorithm based on the concept of Latin-Hypercube sampling is developed for this application. We also present a case study in which the efficiency of three different sampling methods is compared using a car joint stiffness example. The new method, called the Balanced Latin-Hypercube Sampling (BLHS), has shown significantly faster convergence over the other two.

When using a math model to estimate the failure rate of a product, or the mean and standard deviation of performance characteristics of the product, one important issue is the accuracy of the estimates. All math models have error. This error will be transmitted to the error in the estimates of failure rate, mean, and standard deviation. This paper presents a method to calculate the bounds on the transmitted error, which can then be used to 1) obtain confidence bounds on estimates of mean, standard deviation, and failure rate; and 2) establish accuracy requirements on math models.