Development of machine and deep learning based models for risk and reliability problems

Abstract

Artificial intelligence-based algorithms have evolved dramatically over the last couple of decades. Specifically, Machine Learning (ML) and Deep Learning (DL) models have emerged as solutions for many tasks previously unreachable, bringing innovation to the industry, with autonomous driving cars and smart houses, and revolutionizing the society with applications going from movie recommendation to medical diagnosis. In this context, this thesis proposes and brings discussion to ML and DL methodologies successfully developed for three distinct problems in applications related to risk and reliability engineering. In the first, a drowsiness detection model is developed to avoid accidents caused by inattention in the context of human reliability. The second problem deals with estimations of remaining useful life of bearings in the prognostic and health management context. In the last problem, a system to detect usage of personal protective equipment in the context to support safety monitoring is presented. In ML methodologies, support vector machines are used, while convolutional neural networks are applied to DL models. Considering the availability and accessibility of datasets, the obtained results demonstrate adequation of methodologies as tools to provide valuable information to support decisions.