Date of Award
Open Access Dissertation
Chemistry and Biochemistry
Stephen L. Morgan
This dissertation focuses on the application of both novel and standard chemometric approaches toward societal problems of interest in the areas of forensic science and cultural heritage preservation. Microspectrophotometry (MSP), a technique enabling measurements of absorption of electromagnetic radiation by microscopic materials in the ultraviolet-visible (UV-Vis) region, is widely used by forensic examiners for comparisons of metameric textile fibers. These comparisons are often hindered, however, by the raw or normalized spectra showing little detail or having few points of comparison. Derivative preprocessing can enhance structure in some instances. We have demonstrated through the use of multivariate statistics that derivatives are an effective tool for discriminating dyed textile fibers. The Fiber Spectra Comparison Tool developed in this work is an easy-to-use program designed for comparing multiple fibers simultaneously. Microspectrofluorimetry (MSF) is another useful technique, often used as a follow-up method to MSP, for studying fibers that absorb and emit in the UV-Vis region. Results found after applying MSP and MSF to the same set of fibers suggest that the discrimination power of MSP measurements are slightly higher than those obtained from MSF for most colors and fiber-types. In some instances, MSP and MSF provide complimentary information which can be taken advantage of by fusing the measurements. A low-level (i.e., data level) fusion strategy has been developed which provides increased discrimination over the individual techniques.
The ability to transfer multivariate classification models between laboratories having differing instruments has also been investigated in this work. Such efforts could save time and resources in forensic analyses and help identify variability between examiners. A set of 12 blue acrylic fibers was analyzed by MSP at three academic institutions and two certified forensic laboratories. Using a six-step preprocessing procedure combined with quadratic discriminant analysis, a transferrable classification model was developed which, when tested, produced a classification accuracy of 93.2%. This percentage was only slightly lower than the 96.3% accuracy resulting from intra-laboratory models. This outcome speaks to the consistency of results obtained on the same samples in different laboratories. Multivariate classification strategies similar to those applied to colored fibers have also proven useful for determining the playability of magnetic audio tapes, a popular recording medium from the 1950s to the 1990s. Attempting to play degraded tapes during the digitization process can cause damage to the playback instrument, and to the tapes themselves, often leading to significant downtime for museums and archives. A reliable and non-destructive technique for determining the playability of a tape without ever actually playing the tape would be beneficial. This work has shown that with attenuated total reflectance Fourier transform infrared spectroscopy and machine learning algorithms, playability of quarter-inch magnetic audio tapes can be determined with greater than 90% accuracy. This finding led to the creation of the Magnetic Audio Tape Spectra Analysis program, a user-friendly software program allowing tape custodians to visualize data and determine which tapes need to be subjected to restoration processes.
Fuenffinger, N. C.(2015). Optical Spectroscopy and Chemometrics for Discrimination of Dyed Textile Fibers and Magnetic Audio Tapes. (Doctoral dissertation). Retrieved from https://scholarcommons.sc.edu/etd/3252