Center for Advanced Vehicular Systems

Spotlight on Research - Computationally Inexpensive Reliability-Based Optimization

Reliability-based design optimization plays an increasingly important role in the design of advanced vehicular systems, ranging from automobiles to spacecraft. However, direct application of optimization techniques to reliability-based design is daunting as the reliability computation is very time consuming. Considering the fact that high-fidelity and complex computer models (e.g., crash simulations, see Fig. 1 on the Left) is required, computationally inexpensive models needs to be developed. Figure 1 shows a full frontal impact crash simulation. Finite element mesh is composed of 327 components, 320,998 nodes, and 582,541 elements. Simulation of 100 ms takes 10 hrs with 36 1.266GHz Pentium III processors.

The simulation-based design optimization team of the CAVS is actively working on develop design technologies that can help improving the safety and reliability of automobiles while keeping them fuel efficient. The team has recently proposed a new computationally inexpensive reliability analysis technique (DR+EGLD) that combines dimension reduction method and extended generalized lambda distributions. They showed that the DR+EGLD is more accurate than FORM, and more efficient than MCS for the reliability analysis of structures with moderate level reliabilities. Table 1 shows the accuracy and efficiency evaluation of the DR+EGLD method for a crash problem.