Statistical Process Control of Stochastic Textured Surfaces

Bui, Anh Tuan

doi:https://doi.org/10.21985/n2-pfsj-b830

Work

Statistical Process Control of Stochastic Textured Surfaces

Public

Download PDF

Download All Files (.zip)

This dissertation develops a new framework and algorithms for statistical process control of stochastic textured surface data that have no distinct features other than stochastic characteristics that vary randomly (e.g., image data of textiles or material microstructures and surface metrology data of metal parts). All methods are general and nonparametric in that they require no prior knowledge of the types of abnormalities that might occur nor the extraction of specific predefined features. The methods are applicable to a wide range of materials and address unsolved problems regarding monitoring and diagnosing quality-related issues that can lead to early damage, reduced lifetime, or compromised aesthetics of the manufactured materials. Specifically, the first problem is detecting defects on the surfaces (e.g., microstructure porosities); the second problem is detecting changes that affect the entire nature of the surface textures (e.g., microstructure morphology changes); and the third problem is characterizing previously unidentified sample-to-sample variation in the surface textures, in a manner that is conducive to conveying an understanding of the physical nature of the variation. To solve these problems, we use supervised learning methods to model the stochastic behavior of the stochastic textured surface samples. For local defects, we propose two spatial moving statistics for detecting local aberrations in the textured surfaces, based on the residuals of the supervised learning model (fitted to an in-control sample) applied to new samples. For global changes, we develop a monitoring statistic using likelihood-ratio principles to detect changes in the surface nature, relative to the in-control one. For understanding surface variation, we derive dissimilarity measures between surface samples and use manifold learning on these dissimilarities to discover a low-dimensional parameterization of the surface variation patterns. Visualizing how the surfaces change as the manifold parameters are varied helps build an understanding of the physical characteristic of each variation pattern.

Creator