Robust metrological characterization of circular profiles

Abstract Outliers are likely to corrupt data acquired by optical profiles and ignoring them can badly affect the output of the analysis. The occurrence of very few outliers can completely invalidate the fitting process and uncertainty evaluations. Outliers are unexpected values that often mask each other being not detectable by any pre-screening of the data. Therefore, there is the need to adopt a proper robust approach to fit circular profiles data and identify data inadequacies. Here, it is suggested to resort to impartial trimming to fit the centre and the radius of a circle. The optimization problem is based on a subset of the data after discarding that fraction of points with the largest distance from the fitted centre. Trimmed observation are likely to be outliers. The methodology is illustrated through both synthetic and real data.