A Hybrid Sensitivity Analysis for Use in Early Design

Sensitivity analyses are frequently used during the design process of engineering systems to qualify and quantify the effect of parametric variation on the performance of a system. Two primary types of sensitivity analyses are generally used: local and global. Local analyses, generally involving derivative-based measures, have a significantly lower computational burden than global analyses but only provide measures of sensitivity around a nominal point. Global analyses, generally performed with a Monte Carlo sampling approach and variation-based measures, provide a complete description of a concept’s sensitivity but incur a large computational burden and require significantly more information regarding the distributions of the design parameters in a concept. Local analyses are generally suited to the early stages of design when parametric information is limited and a large number of concepts must be considered (necessitating a light computational burden). Global analyses are more suited towards the later stages of design when more information regarding parametric distributions is available and fewer concepts are being considered. Current derivative-based local approaches provide a significantly different set of measures than a global variation-based analysis. This makes a direct comparison of local to global measures impossible. To reconcile local and global sensitivity analyses, a hybrid local variation based sensitivity (HyVar) approach is presented. This approach has a similar computational burden to a local approach but produces measures in the same format as a global variation-based approach (contribution percentages). This HyVar sensitivity analysis is developed in the context of a functionality-based design and behavioral modeling framework. An example application of the method is presented along with a summary of results produced from a more comprehensive example.