Mathapelo Sekwai (PhD candidate)
Lecturer
University of Johannesburg, South Africa
Date: January 9, 2026
Time: 12:00 p.m. – 1:00 p.m.
Location: ED 312
Zoom: https://uregina-ca.zoom.us/j/92843721310?pwd=QznwGHCg8UoLcZncLz7haHa76Lj5jN.1
Meeting ID: 928 4372 1310
Passcode: 498995
Bio: Mathapelo Sekwai is a lecturer and PhD candidate in the Department of Mechanical Engineering Science at the University of Johannesburg, South Africa. She is a multidisciplinary engineer with experience in knowledge transfer, design strategy development, and sustainable engineering solutions. Her research focuses on finite element analysis of failed components, with an emphasis on identifying failure modes and establishing methods to prevent future failures. She is expanding her work into hydrogen-related failure analysis and mitigation and is currently a visiting researcher at the Energy Systems Engineering, University of Regina under the supervision of Dr. Jacob Muthu.
Abstract: Hydrogen embrittlement (HE) poses a threat to the integrity of metallic infrastructure, particularly pipelines, where intricate couplings between alloy chemistry, microstructure, mechanical state, and environmental variables influence failure probability. This presentation reviews recent studies applying machine learning (ML) to predict HE vulnerability and suggests preventive measures. The review also covers the types of data sources, and feature spaces used in current studies, including composition and processing variables, microstructural descriptors, mechanical properties, hydrogen transport metrics, and service-environment parameters as well as the preprocessing and augmentation approaches used to address data scarcity and class imbalance. It also summarizes the supervised and ensemble models commonly used, such as logistic regression, random forests, gradient-boosted trees, support vector machines, and neural networks. Evaluation approaches are compared, with a focus on ROC-AUC, precision-recall analysis, calibration, cross-validation strategies, and external validation using separate datasets. Key findings in the literature show that (1) ML models can achieve strong distinguished embrittlement-prone conditions when trained on representative, quality-controlled datasets; (2) features related to hydrogen diffusivity, tensile strength/hardness, specific microstructures, and aggressive environmental exposure consistently rank highly in importance; and (3) explainability tools improve interpretability and facilitation. Persistent challenges include the lack of publicly available, standardized datasets, restricted model transferability between laboratories and service environments, and insufficient integration of physics-based constraints. The review concludes by recommending best practices for future research, including transparent data sharing and benchmarking, hybrid physics-ML modeling, uncertainty quantification, and deployment-focused validation, to expedite reliable, data-driven HE prevention in pipeline systems.