A0971
Title: Gauge reproducibility and repeatability for matrix-variate data with application to forensic fracture surface-matching
Authors: Carlos Llosa-Vite - Sandia National Laboratories (United States)
Ranjan Maitra - Iowa State University (United States) [presenting]
Abstract: Three-dimensional (3D) microscopy can be used to analyze the unique microscopic patterns present in fractured surfaces and, therefore, help forensic match-analysts reduce the subjectivity involved in comparative microscopy. Yet, the repeatability and reproducibility of these 3D microscopy-generated features have seldom been studied, and little is known about the effects that the microscope operator or sample alignment has on the measurement system. To study the quality of the measurement system, three inexperienced microscope operators are trained, and six overlapping images are repeatably obtained from a set of 10 steel rods that were generated and fractured under controlled conditions. A novel gauge R\&R model is developed for matrix-variate data and proposed an expectation-maximization algorithm for maximum likelihood estimation to quantify the multiple sources of variability affecting the obtained features. After the third imaging repetition, each operator used a fixture to align the two fractures during the imaging process. While all the matches and non-matches were classified correctly regardless of the imaging fixture, gauge R\&R showed that the fixture helped us improve the measurement system by keeping the within-operator as the smallest source of variability. The importance of assessing 3D microscopy measurement systems is shown with tools such as gauge R\&R, as it can help improve the measurement system.
