Construction of a multi-class discrimination matrix and systematic selection of areal texture parameters for quantitative surface and defect classification

Classification of Visual Smoothness Standards Using Multi-Scale Areal Texture Parameters and Low-Magnification Coherence Scanning Interferometry

“…The fitted decision tree can intuitively communicate how the selected surface texture parameters and associated thresholds can be employed as rules for classifying a newly measured surface, providing an objective basis for assessing surface quality in accordance with user-defined grades of surface quality. Readers are referred to the original paper [21] for complete details, including examples of the selection procedure, construction of the d ′ matrix, and graphs related to the expected classification outcomes versus d ′ values of the selected parameters.…”

“…To circumvent the need for subjective visual analysis, a previously published systematic framework, SQuID™ [21], will be built upon to derive quantitative descriptions using areal maps subdivided into multiple length scale measurements and a set of logical rules to objectively classify the aforementioned set of subjectively graded visual smoothness panels. Section 2.1 includes an overview of the standardized visual smoothness tiles investigated in this study.…”

“…Instead of relying on preconceived notions of parameter selection for accurately classifying the PCI tiles, an algorithmic method outlined in [21] called SQuID™ (Surface Quality and Inspection Descriptors) is used to down-select optimal surface parameters for classification. SQuID™ uses a selection method that breaks down a user-defined multi-class classification task into a series of binary classification tasks.…”

“…There is a strong desire to replace HVI with automated systems and a longstanding goal of definitively linking functionality to surface topographies. However, there remains a need to make it easier to find suitable metrics that can reliably discriminate between surfaces grouped or ranked by functionality, subjective appearance, or, indeed, processing-induced differences [21][22][23]. Information on surface quality can be obtained by multiple modes of acquisition and processing, giving different outputs.…”

“…Datasets from these systems can provide a wealth of geometrical topographical data with the scale of resolvable features dependent on system-limiting factors, tip radii for tactile measurements, and numerical aperture considerations for optical systems [31]. While the wealth of the available data is advantageous, effectively processing the data to give useful metrics still poses challenges [11,21]. Statistical parameters used to describe surfaces within the manufacturing community include height, spatial, hybrid, functional, and other parameters detailed in the ISO 25178-2 standard [32].…”