Robust surface fitting—Using weights based on à priori knowledge about the measurement process

Modern surface layouts like automotive cylinder liners, turbine blades or seal faces need high information content. This high information content can only be reached with modern 3D-surface measurement techniques like confocal microscopy or white light interferometry. For an analysis of the surface properties, an antecedent surface fitting is necessary. This surface fit has to be robust and must be based on trustworthy data. According to the optical measurement techniques there are many known effects, which lead to wrong or insecure measuring data. Using à priori knowledge about the measurement process leads to knowledge about surface structures, which otherwise tend to be unsure or wrong. Examples for the confocal microscopy are “bat-wings” at sharp edges, multiple peaks because of oil films or surface coating. White light interferometry also has problems with speckles, when the surface structures have the size close to the interference length of the white light interferometer. Using this knowledgebase for a pre-analysis of the surface data, a confidence level for every single data point could be calculated. That leads to a weighting function, which is usable with the commonly known surface fitting methods. In this work different weighting methods are introduced. Some weighting methods are based on the original measured data and the à priori knowledge about the measurement method. Other weighting methods also use information about the measurement process, for example, the sharpness and skewness of a confocal peak or the signal to noise ratio. The weights could also be based on à priori knowledge about the surface and the structures on the surface, for example, sharp edges or surface areas with bad reflectivity properties. There are combinations with each other, as well as with the already known weights from the common surface fitting methods from the regression analysis. This leads to a regression analysis which based on measured data with a higher reliability. The reducting of the measurement device influence provides a better comparability of surface data.