University of British Columbia
Leveraging machine learning and data analysis for the study of complex materials with Raman spectroscopy and image analysis : a bridge between academia and industry
Abstract
dc:descriptionRapid advances in chemical instrumentation now produce massive datasets that often exceed the limits of manual interpretation. Data science, particularly machine learning, provides robust methods to convert these vast, diverse datasets, such as vibrational spectra, images, and discrete measurements, into actionable insights. Despite their apparent differences in format, these data types share fundamental preprocessing and modeling principles. In this thesis, we integrate machine learning and automated systems to streamline the entire data pipeline, from data collection through interpretation. By applying multiple models across various phases of sample analysis and minimizing manual intervention, we enhance reproducibility and accelerate the pace of chemical discovery. The thesis is divided into three parts. The first part addresses industrial questions in the pulp and paper sector, focusing on impurity detection in wood pulp sheets and pulp property prediction for quality control. We developed an imaging technique using a near-infrared (NIR) camera and deep learning to detect impurities, and introduced a data fusion system that combines Raman spectral data with morphological data of cellulose fibers. This approach predicts pulp properties that are challenging to measure using standard wet-lab methods. To further streamline the process, we designed a Graphical User Interface (GUI) that automates the analysis of these two distinct data sources, enabling accurate and efficient property prediction of pulp samples. In the second part, we present a methodology for detecting adulterants in milk samples using conventional Raman spectroscopy combined with advanced machine learning algorithms. These studies focus on identifying three small nitrogen-based adulterants commonly added to increase protein content, as well as quantifying their concentrations. To improve this analysis, we explored Shifted Excitation Raman Difference Spectroscopy (SERDS) and introduced a novel preprocessing technique called Common-Mode Rejection (CMR). This work employed a custom-built Raman scanning system capable of capturing both conventional Raman spectra and SERDS, thereby facilitating automated operation. In the final section, we concentrate on soil analysis, particularly the quantification of carbonate content using SERDS data. By correlating the Raman spectra of soil samples with ground truth carbonate measurements, machine learning models demonstrate a strong relationship, offering a reliable method for soil carbonate analysis.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy - PhD
- Level thesis:degree_level
- doctoral
- Discipline thesis:degree_discipline
- Chemistry
- Grantor dc:publisher
- University of British Columbia
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Poursorkh, Zahra
- Contributors dc:contributor
-
- Grant, E. R.
Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-NoDerivatives 4.0 International
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/2429/91998