COMMITTEE CHAIR: Dr. Lijun Qian
TITLE: MODELING AND CONTROL OF SPRING-MASS SYSTEMS WITH SWITCHED DYNAMICS USING RESERVOIR COMPUTING
ABSTRACT: Hybrid spring-mass systems with dynamic state switching present significant chal- lenges for modeling and control due to their complex nonlinear dynamics and dis- crete state transitions. In this thesis research, a novel reservoir computing frame- work based on echo-state networks (ESNs) is proposed to model and control spring- mass systems with two distinct states: F1 (without air friction) and F2 (with air friction). The system switches between states based on velocity thresholds, creat- ing a complex hybrid dynamics requiring sophisticated modeling approaches. The proposed method enables efficient learning of spatiotemporal patterns in hybrid dy- namics, offering accurate system identification and robust control capabilities. The proposed framework is computationally efficient, supporting rapid training (under 5 seconds) and real-time deployment. Experimental results demonstrate that the ESN models achieve high performance across different threshold configurations, with MAPE values of 0.06% to 8.2% for modeling and 2.52% to 7.07% for con- trol tasks. It is also observed that the reservoir computing framework significantly outperforms traditional deep learning methods while maintaining computational ef- ficiency suitable for real-time applications. This thesis research successfully bridges physics-based modeling with machine learning to provide a practical solution for real-world dynamic hybrid system applications.
Room Location: Electrical Engineering Conference Room 315D