Job Description

Overview

This project aims to enhance the performance and stability of a Neutral Equilibrium Mechanism (NEM)–based virtual pier control method for bridge deformation control by introducing Reinforcement Learning (RL) as an auxiliary adjustment mechanism. The primary objective of the NEM virtual pier is to reduce the deformation at a target location toward zero under service-level loads without adding physical supports. However, in practical implementations, performance improvement using fixed-parameter controllers has reached a bottleneck due to system nonlinearities, mechanical backlash, friction, and other uncertainties inherent in real structural and actuation systems.
Rather than replacing established control theory with artificial intelligence, this project adopts a hybrid control philosophy. A mature and verifiable PID control loop is retained as the main control backbone, while RL is explicitly positioned as a strategy and parameter adjustment layer. This design allows the core control actions to remain interpretable, traceable, and safe, while enabling adaptive compensation for unmodeled effects and variations in operating conditions. Such a separation between the control layer and the learning layer provides a practical pathway for applying learning-based methods to high-risk physical systems.
The proposed methodology will be validated through physical experiments using existing bridge models and NEM mechanisms. A closed-loop system integrating sensing, control, and learning modules will be established, enabling safe execution of training and evaluation processes under multiple loading scenarios. Experimental results will be used to assess deformation reduction performance, stability, and robustness across varying conditions. The implementation will leverage a Python-based ecosystem for data processing, learning, and performance evaluation, ensuring efficient integration with external hardware and real-time control systems.
The expected contributions of this project are threefold. First, from an engineering perspective, the proposed approach is expected to further reduce deformation responses and improve consistency under service loads without modifying existing hardware configurations. Second, from a methodological perspective, the project will establish a verifiable hybrid control framework that clearly delineates the roles of classical control and reinforcement learning when addressing nonlinear and uncertain systems. Third, from a knowledge transfer perspective, the project will produce reproducible training, validation, and experimental procedures in the form of standard operating procedures (SOPs) and implementation examples. These outputs are intended to lower the entry barrier for researchers in structural control to adopt AI-based reinforcement learning methods and to facilitate future extensions to other systems and application domains.

Topic 1  Establish an integrated methodology architecture for deformation control of NEM virtual bridge piers based on reinforcement learning 

Objective:       establish a minimum executable framework

 

Topic 2 Build a digital twin of RL-NEM 

Objective:      focusing on constructing a simulation system for the NEM virtual bridge pier control system and establishing a mechanism for gradually integrating measured data

 

General Requirements:

1. Research log (ref. Samples, daily)
2. Weekly meeting (ppt., summarize, question and discussion)
3. Group meeting (reference review, 2 papers/week)

Final Report includes: 

1. Introduction
2. Literature review
3. Theories
4. Experimental setup
5. Result and discussion
6. Conclusion

Preferred Intern Educational Level

This internship requires applicants to have a solid background in mechanical or civil engineering, strong mathematical and mechanics foundations, and advanced programming language skills. Prior experience in experimental design is preferred. Therefore, this internship opportunity is suitable for PhD and Master's students.

Applicants with a pure information science or AI background will not be considered.

Skill sets or Qualities

1. Programming language skills: C++ or Python
2. Mechanical design, automation control
3. Basic understanding of artificial intelligence
4. Familiarity with Arduino or Raspberry Pi