Expedited surrogate-based quantification of engineering tolerances using a modified polynomial regression

Abstract Statistical analysis is frequently used to determine how manufacturing tolerances or operating condition uncertainties affect system performance. Surrogate one of the accelerating ways in engineering tolerance quantification analyze uncertainty with an acceptable computational burden rather than costly traditional methods such as Monte Carlo simulation. Compared more complicated surrogates Gaussian process, Radial Basis Function (RBF), Polynomial Regression (PR) provides simpler formulations yet outcomes. However, PR common least-squares method needs be accurate and flexible for approximating nonlinear nonconvex models. In this study, a new approach proposed enhance accuracy approximation power dealing tolerances. For purpose, first, by computing differences between training sample points reference point (e.g., nominal design), we employ certain linear exponential basis functions transform original variable design into transformed variables. A second adjustment made calculate bias true simulation model surrogate’s approximated response. To demonstrate effectiveness approach, provide comparison results conventional employing four practical problems geometric fabrication three-bar truss design, welded beam trajectory planning two-link three-link (two three degrees freedom) robot manipulator. The obtained prove preference over improving significantly lower prediction errors.