Robust design by variability optimization

A deterministic method of robust design against design parameter uncertainty is described. The method consists of the definition of a variability function for the performance measure under consideration, based on known or assumed uncertainties in the parameters of the measure and its known dependency on these parameters. It is assumed that the design parameters are subject to constraints on their individual values and interactions and that the performance measure (or output variable) is required to have a specified nominal design point value when the design parameters are at their chosen nominal values. An algorithm is then described which determines that set of nominal parameter values which results in a minimum value of the variability function given the specified constraints. The algorithm is applied to two examples, firstly in analytic form to demonstrate the procedure and then numerically to specific design problems. The results are discussed in relation to a probable design strategy and with regard to the advantages and limitations of the method employed.