Machine learning evaluation of PI control effects on neutral equilibrium in bridge virtual pier systems

Area of Science:

• Structural Engineering
• Control Systems Engineering
• Machine Learning Applications

Background:

• Active control systems are crucial for bridge stability under dynamic loads.
• Neutral Equilibrium Mechanisms (NEMs) offer a novel approach to mitigate structural vibrations.
• Optimizing proportional-integral (PI) controller gains is essential for effective vibration control.

Purpose of the Study:

• To investigate the application of NEMs in active bridge control systems.
• To analyze the impact of proportional gain (GP) and integral gain (GI) on vertical displacement stability.
• To develop and validate machine learning models for predicting and optimizing control parameters.

Main Methods:

• Utilized a scaled bridge model with dual NEMs, sensors, and servo motors in a closed-loop system.
• Employed Random Forest Regression and Neural Networks for nonlinear predictive modeling.
• Applied K-means clustering and feature sensitivity analysis for parameter optimization.

Main Results:

• Optimal PI parameters (GP=1.0, GI=0.010) reduced maximum vertical displacement from ~5 mm to ~0.4 mm.
• Stabilization time was decreased to 9.8 seconds with the optimal configuration.
• The Neural Network model achieved high predictive accuracy (R²=0.934, RMSE=0.038).

Conclusions:

• Machine learning-based models are feasible for bridge displacement control.
• Data-driven parameter optimization provides effective guidance for intelligent bridge design.
• Medium-gain settings (GP=1.0, GI=0.010) optimally balance stability and structural symmetry.