A new Gaussian process-based approach for uncertainty propagation in surface metrology profile parameters estimation

Surface texture parameters are important indicators for understanding and controlling manufacturing processes. Deriving these parameters is however beset by ambiguities and uncertainties. The inclusion of confidence bands should provide valuable information on the reliability of the derived surface parameters. Existing methodologies of uncertainty modelling assume nonrandom interpolation functions, which do not adequately allow for the inclusion of interpolation uncertainties. This paper presents a new approach based on Gaussian processes, including a mechanism for the derivation and inclusion of such interpolation-based uncertainties when calculating surface texture parameters. The interpolation-based uncertainties are assumed to be independent of measurement uncertainties and as such, they can be independently modelled and propagated onto the derived parameters. When tested using real machining surface data, validation results show that the newly proposed technique has the advantage over the ISO-based approach of systematically characterising interpolation-based uncertainties in the form of confidence bands in the estimated profile parameters.