Function Variability Optimization in Parameter Design

Abstract

A previously reported algorithm for deterministic parameter design is discussed and improved. The method enables the determination of a nominal set of design parameters, which minimize variability in one or more performance measures, in the presence of uncertainties in the design parameters, whilst maintaining the design concept and nominal performance. The advantages and limitations of a previous algorithm are considered and an improved alternative is proposed. The method uses secondary optimization procedures to determine maximum and minimum performance function values within the tolerance range of each nominal point as the primary optimization of design parameter values proceeds, which removes the need for gradient updating and renders the algorithm more reliable for cases of large tolerance values when a performance function gradient could change within the tolerance zone. The algorithm has been programmed using a simulated annealing method for the primary optimization and a modified Newton algorithm for the secondary procedure. Some numerical examples are given and compared with previous results.