Understanding complex phenomena
in colour and development of a
novel laundry metric

Abstract

Diffuse reflectance spectroscopy (DRS) is a powerful non-contact technique for probing the physical world. In this project it was applied to two main areas of study, colourimetry and chemometrics. Chapter 1 gives a review of the relevant concepts used throughout this study. Chapter 2 first explores the origin of the difference in perceived colour when compounds containing certain lanthanide ions are viewed under sunlight or fluorescent lighting. It was found that salts of neodymium and holmium are subject the the largest change in colour. The cause of this phenomenon was found to be the overlap of the concentrated green component in fluorescent lighting with absorption bands of the lanthanide ions, leading to illuminant metamerism. The phenomenon observed in neodymium chloride was then utilised in order to control the perceived appearance of a sample to any hue using a custom-build spectrally tunable light source, while maintaining near-white illumination. The process was herein termed tunable illuminant metamerism.
The colourimetry models which could accurately describe the complex colour appearance of lanthanide salts were then applied to study the workings of fluorescent whitening agents (FWAs) and hueing dyes (HDs) on the colour correction of naturally degraded fabric (yellowing). A platform was developed which could be used to model the appearance fabric under different lighting conditions, at different stages of yellowing while simulating the effects of FWAs and HDs.
In chapter 3, DRS is applied to study the lipid component of laundry stains, and a technique for the quantification of lipid on fabric substrates by short-wave infrared DRS was devised. A unit system for comparing the concentration of lipid on fabric, termed wt%, was first discussed and calibration samples of uniformly set lipid concentration were prepared and measured using DRS to yield a dataset for machine learning. A combination of preprocessing techniques, principal components regression and Gaussian process regression models were used to model lipid concentration from DRS spectra. The validity of the best performing model was tested on an external sample set with consistent results, confirming that the proposed technique is capable of non-contact quantitative measurements of lipid concentration on fabric. The approach was applied to point-scan imaging where the concentration of lipid across a fabric could be quantitatively mapped and results were verified by gravimetry. Further validation of the developed machine learning models is still required as the method of producing calibration samples of low lipid concentration was found inaccurate at low levels of staining. A revised method for producing uniform calibration samples using inkjet printing was investigated to address this issue and found to be feasible.