Event
PhD Dissertation Defense: Anjana Hevaganinge De Alwis
Thursday, May 1, 2025
11:30 a.m.
AJC 4104 (4th floor conference room)
Rachel Chang
301 405 8268
rachel53@umd.edu
Title: INTERPRETABLE MACHINE VISION INTELLIGENCE METHODS FOR BIOPHARMACEUTICAL AND AGRICULTURAL MANUFACTURING AUTOMATION
Committee members:
Dr. Yang Tao, Chair
Dr. Maurizio Cattaneo
Dr. Giuliano Scarcelli
Dr. Helim Aranda Espinoza
Dr. Yiannis Aloimonos, Dean's Representative
Abstract:
Within the context of mission critical applications, there is growing concern in the trustworthiness and reliability of Artificial Intelligence (AI) driven automation techniques. In fact, the manufacturing sector is missing adaptable and interpretable AI driven systems that work synergistically with human intelligence and cognition. Thus, the development of such systems will be explored in both biopharmaceutical and seafood industries. (Biopharmaceutical Industry) As of 2016, 5 of the top 10 drugs were manufactured inside of a bioreactor and amounted to 75 billion dollars in annual sales. This market share has only grown since, but there are several upstream/downstream quality issues that stem from the current lack of a continuous control system. The development of a contactless and interpretable sensor to replace current invasive probes will enable adaptive multiplex sensing of critical material attributes (CMAs) and critical product attributes (CPAs) alike for precise real-time control of bioreactor environment to suit even the most delicate cells. On the other hand, (Seafood Industry) robots that can replicate human hand movements with surgical precision have the potential to alleviate manufacturing labor shortages, enhance rehabilitation technologies, and revolutionize procedures in operating rooms. Lump meat extraction from Maryland Blue Crabs represents a class of dexterous human tasks that are difficult to simulate, slow down or tele-operate . Thus, crab picking may be used as an excellent test bed for the development of a precise and interpretable surgical trajectory automation system. Both projects lead to the development of interpretable automation methods for both cell bioprocess cultivation and crab picking surgical motions. Unlike opaque machine learning models, interpretable models foster a trusting and collaborative relationship between the human practitioner and AI enabled machine. So, the developed methods can be extended to other high-stake tasks that directly impact human health and cost of living.