Author ORCID Identifier
Year of Publication
Doctor of Philosophy (PhD)
In the last few years, we witnessed the revolution of the Internet of Things (IoT) paradigm and the consequent growth of Cyber-Physical Systems (CPSs). IoT devices, which include a plethora of smart interconnected sensors, actuators, and microcontrollers, have the ability to sense physical phenomena occurring in an environment and provide copious amounts of heterogeneous data about the functioning of a system. As a consequence, the large amounts of generated data represent an opportunity to adopt artificial intelligence and machine learning techniques that can be used to make informed decisions aimed at the optimization of such systems, thus enabling a variety of services and applications across multiple domains. Machine learning processes and analyses such data to generate a feedback, which represents a status the environment is in. A feedback given to the user in order to make an informed decision is called an open-loop feedback. Thus, an open-loop CPS is characterized by the lack of an actuation directed at improving the system itself. A feedback used by the system itself to actuate a change aimed at optimizing the system itself is called a closed-loop feedback. Thus, a closed-loop CPS pairs feedback based on sensing data with an actuation that impacts the system directly. In this dissertation, we propose several applications in the context of CPS. We propose open-loop CPSs designed for the early prediction, diagnosis, and persistency detection of Bovine Respiratory Disease (BRD) in dairy calves, and for gait activity recognition in horses.
These works use sensor data, such as pedometers and automated feeders, to perform valuable real-field data collection. Data are then processed by a mix of state-of-the-art approaches as well as novel techniques, before being fed to machine learning algorithms for classification, which informs the user on the status of their animals. Our work further evaluates a variety of trade-offs. In the context of BRD, we adopt optimization techniques to explore the trade-offs of using sensor data as opposed to manual examination performed by domain experts. Similarly, we carry out an extensive analysis on the cost-accuracy trade-offs, which farmers can adopt to make informed decisions on their barn investments. In the context of horse gait recognition we evaluate the benefits of lighter classifications algorithms to improve energy and storage usage, and their impact on classification accuracy. With respect to closed-loop CPS we proposes an incentive-based demand response approach for Heating Ventilation and Air Conditioning (HVAC) designed for peak load reduction in the context of smart grids. Specifically, our approach uses machine learning to process power data from smart thermostats deployed in user homes, along with their personal temperature preferences. Our machine learning models predict power savings due to thermostat changes, which are then plugged into our optimization problem that uses auction theory coupled with behavioral science. This framework selects the set of users who fulfill the power saving requirement, while minimizing financial incentives paid to the users, and, as a consequence, their discomfort. Our work on BRD has been published on IEEE DCOSS 2022 and Frontiers in Animal Science. Our work on gait recognition has been published on IEEE SMARTCOMP 2019 and Elsevier PMC 2020, and our work on energy management and energy prediction has been published on IEEE PerCom 2022 and IEEE SMARTCOMP 2022. Several other works are under submission when this thesis was written, and are included in this document as well.
Digital Object Identifier (DOI)
Teaching Assistantship from Computer Science Department at University of Kentucky - Fall 2018, Spring 2019
Research Assistantship from Computer Science Department at University of Kentucky - Summer 2019 NSF (National Science Foundation) grant EPCN (Energy, power, Control, and Networks) 1936131 since Fall 2019 NIFA (National Institute of Food and Agriculture) grant 2021-68014-34139 - additional 1% of salary since Fall 2021
Casella, Enrico, "Machine-Learning-Powered Cyber-Physical Systems" (2023). Theses and Dissertations--Computer Science. 130.