Optimization of a shot peening process

The statistical design and analysis of an experiment has been described to establish an empirical regression relationship between parameters of a typical shot peening process and measures of residual stress. In this study, practical constraints led to the choice of a non-orthogonal design for the experiment, the implications of non-orthogonality are discussed, and a multiple regression model was fitted to the results. The residuals from the regression model were better modelled by a Laplace distribution than a Gaussian distribution, so Huber robust estimates of the coefficients in the regression model were substituted for the least-squares estimates. The robust regression relationship is used to optimize the settings of process parameters. The optimization has resulted in an estimated increase of 15 per cent in the maximum residual stress.