University of Michigan – Dearborn
Predoctoral fellows have been nominated by their programs and are selected through a competitive review process based on the creativity and impact of the research they are pursuing. The abstracts for recipients at the University of Michigan – Dearborn describe the framework, aims, and significance of each fellow’s dissertation and demonstrate the breadth of Rackham doctoral programs.
Abstracts
Custom AI Techniques for Data Analytics in Industrial Applications
Wael Hassanieh, Industrial and Systems Engineering
Industries have long struggled to integrate artificial intelligence (AI) methods in their applications and operations. The broad adoption of AI is limited due to the lack of custom architectures that can effectively handle the unique complexities of modern datasets and perform robust and accurate estimations/predictions. With the integration of the Internet of Things into industrial operations and the increasing technological advancements, businesses are accumulating vast amounts of data, but lack the knowledge and confidence to extract useful insights for decision-making purposes. Our research focuses on providing novel methods to transform this data into understandable and actionable information, which can benefit a range of industrial challenges, including but not limited to sensor monitoring systems, rail corrugation detection, Lithium-ion battery capacity degradation, and warranty claims forecasting. Data-driven frameworks are built for (i) robust feature selection, (ii) real-time anomaly detection, (iii) identification of trends and patterns for long-term estimations, and (iv) overcoming data maturation.
An Integration of ML and LLMs for Security of Cyber-Physical Systems
Aydin Zaboli, Electrical, Electronics, and Computer Engineering
The objective of this dissertation is to develop and implement advanced machine learning (ML) methodologies and large language models (LLMs), specifically designed to augment the performance and security aspects of cyber-physical systems. This research primarily concentrates on the application of these technologies within the domains of smart grids (SGs) and autonomous vehicles (AVs). The use of ML algorithms for accurate load forecasting behind-the-meter in SGs has been investigated, aiming to improve the efficiency and reliability of electricity supply by predicting consumption patterns with higher precision. This is crucial for balancing energy demand with supply, especially in the context of integrating renewable energy resources. A development of context-awareness methods has been explored to improve the capability of ML algorithms in abnormal scenarios where they struggle to accurately recognize and interpret traffic signs or objects on the road. This research is vital for advancing the safety and reliability of AVs, particularly in complex and uncertain driving environments. Then, this dissertation delves into the use of LLMs for the anomaly detection of GOOSE and SV communication messages, a novel approach in the area of SG cybersecurity. This section of the study emphasizes the importance of robust cybersecurity measures in protecting critical infrastructure from emerging digital threats. Using case studies of multiple cyber-attacks or abnormal scenarios (e.g., DoS or replay attacks in GOOSE and SV messages), it is also demonstrated how digital substations or AVs could be attacked by abnormal scenarios. Further, mitigation strategies (e.g., context awareness methods) are studied that could be used to enhance the reliability of the system. A hardware-in-the-loop (HIL) testbed using different equipment has been set up to generate real-time data for GOOSE and SV messages in the anomaly detection process of SGs. Generally, this dissertation not only highlights the potential of ML algorithms and LLMs in enhancing the operational efficiency and security of cyber-physical systems but also provides valuable insights into their practical applications in the fields of SG management and AV technology.