2026-09-20T09:43:56Z https://www.ideals.illinois.edu/oai-pmh

oai:www.ideals.illinois.edu:2142/129844 2025-10-25 col_2142_5131 col_2142_10761 com_2142_5130 com_2142_10755 com_2142_234

application/pdf en text Submission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2025-10-20 without embargo terms The student, Beitong Tian, accepted the attached license on 2025-07-08 at 13:45. The student, Beitong Tian, submitted this Dissertation for approval on 2025-07-08 at 14:09. This Dissertation was approved for publication on 2025-07-08 at 15:53. DSpace SAF Submission Ingestion Package generated from Vireo submission #22441 on 2025-10-20 at 16:57:30 Patronus: multi-modal sensing, analytics, and localization assistance for heterogeneous working environments Tian, Beitong 2025-07-08 Nahrstedt, Klara Nahrstedt, Klara Caesar, Matthew Soltanaghai, Elahe Shenoy, Prashant Internet Of Things Smart Glasses Multimodal Sensing Condition Monitoring Environmental Monitoring Visual Localization Intelligent Environments eng Today’s “Jarvis-like” AI assistants excel at generic, consumer-oriented tasks, yet they remain ill-suited for heterogeneous working environments—dynamic laboratories and industrial facilities where hands-on professionals must juggle safety-critical processes, rapidly changing context, and a deluge of multimodal data. This thesis argues that effective assistance in these settings hinges on the tight co-design of three pillars: (i) a low-cost, scalable, and evolvable sensing infrastructure, (ii) a trustworthy real-time analytics pipeline, and (iii) an accurate, practical localization layer that enables context-aware humandata interaction. In this thesis, we introduce Patronus, a modular framework composed of five interoperable systems that collectively satisfy these requirements. SENSELET++ deploys a plug-and-play sensor network and anomaly analytics for scalable environmental monitoring. MachineStethoscope enables on-device, unsupervised health monitoring for legacy rotating machinery. WeldMon fuses heterogeneous signals and introduces synthetic fault augmentation to improve failure prediction in ultrasonic welding. GaugeTracker digitizes analog gauges entirely on low-cost IoT hardware, leveraging multiple vision and vision language models for robust transcription. Finally, AnyLoc provides energy-efficient visual localization that operates under low-resolution and low-light conditions in cluttered indoor scenes. Together, these systems power MAINTGlasses, a hands-free smart-glasses interface that delivers spatially relevant insights to professionals in real time. Deployments across cleanrooms, nanofabrication labs, and server rooms demonstrate that Patronus fosters safer, faster, and more efficient workflows. By unifying sensing, analytics, and localization, this work charts a practical path toward intelligent environments that actively understand and support complex human work. 2025-08 Text https://hdl.handle.net/2142/129844 Copyright © 2025 Beitong Tian. All rights reserved. Siebel School Comp & Data Sci Computer Science University of Illinois Urbana-Champaign Ph.D. Dissertation