Date of Graduation

Summer 2025

Degree

Master of Science in Computer Science

Department

Computer Science

Committee Chair

Razib Iqbal

Abstract

With the advancements of smart assistants in smart spaces, such as smart homes and smart offices, actuators are controlled by voice commands or preconfigured instructions that respond to sensor events triggered by occupant activities requiring human intervention. Complete automation of these predefined instructions, eliminating manual control, hinges on identifying the dependencies between sensors and actuators. Moreover, as the number of sensors and actuators increases, the volume of their events grows significantly, generating large and diverse time series data that must be analyzed to identify actuator dependencies on sensors. This identification task becomes complicated as user preferences and behaviors change over time, such as morning versus evening routines or seasonal variations. This dynamic nature of preferences necessitates adaptive, context-aware rule-setting for actuator control. Additionally, manually configuring actuation rules becomes increasingly complicated since most end users lack expertise in sensor-actuator operations. Hence, in this thesis, I propose a novel approach called Sensor Actuator Mapping, aimed at determining sensor and actuator dependency. This method analyzes time series data consisting of sensor and actuator events that represent the activities of occupants. The proposed approach has two phases: in the first phase, it identifies contextually related sensors; in the second phase, it extracts features from time series data and applies an unsupervised clustering technique to group related sensors and actuators based on temporal correlations that reflect their interaction patterns. By leveraging unsupervised learning, this approach can operate without labeled data or human intervention, thereby enhancing user convenience by eliminating the need for manual rule changes. Experimental evaluations on real-world multiple dynamic smart spaces deployments demonstrate the efficacy of my proposed approach across various deployments with diverse sensor-actuator configurations, underscoring its adaptability and accuracy in autonomous sensor-to-actuator mapping.

Keywords

automatic actuation, clustering, internet of things, operational rules, sensor groups, time series data

Subject Categories

Computational Engineering | Hardware Systems | Other Computer Engineering | Robotics

Copyright

© Fahim Ahmed Irfan

Download

Available for download on Friday, December 31, 2027

Open Access

Share

COinS