Seismic Control of a Smart Base-Isolated Building with Nonlinear Behavior Using Deep Reinforcement Learning

Reinforcement learning-based adaptive gyroscopic control for torsional response mitigation: a proof-of-concept study

Torsional vibrations induced by seismic excitation can significantly increase structural demand, particularly in systems with geometric or stiffness irregularities. This proof-of-concept study proposes a hybrid adaptive control framework that combines passive gyroscopic damping with reinforcement learning (RL)-based adaptive torque control for torsional response mitigation. A simplified three-degree-of-freedom (3DOF) structural model is evaluated under bidirectional earthquake loading, while the control policy is trained using the Proximal Policy Optimization (PPO) algorithm to generate adaptive torque directly from measured system states without explicit system identification. Numerical simulations using recorded earthquake ground motions show that the proposed approach reduces root-mean-square torsional displacement to 0.012 ± 0.001 rad, representing a 65.7% reduction relative to the uncontrolled case and improved performance over passive gyroscopic control alone. Sensitivity analyses involving variations in mass, stiffness, and damping indicate stable controller performance across the structural parameter ranges considered. The trained controller achieved sub-millisecond inference time with minimal computational overhead, supporting its computational feasibility for future real-time implementation. The results demonstrate the potential of integrating physically grounded gyroscopic damping with data-driven adaptive control for intelligent torsional vibration mitigation.

Deep excavation wall design using reinforcement learning

This study investigates the application of reinforcement learning (RL) for obtaining near-optimal designs of diaphragm walls in geotechnical engineering. A physics-based numerical environment is developed to simulate soil–structure interaction, relying on a Winkler spring formulation with pressure-dependent soil springs to approximate the nonlinear response of the ground. This modelling framework allows the agent to evaluate candidate designs through physically meaningful structural responses rather than surrogate performance indicators. To reflect realistic engineering practice, a specifically designed action space and reward function are formulated, incorporating both discrete design decisions and continuous geometric parameters. During training, the agent iteratively proposes a design configuration, observes the response computed by the physical simulator, and updates its policy based on the resulting reward signal. Several common RL algorithms are investigated, including Proximal Policy Optimization (PPO), REINFORCE, and the Parameterized Deep Q-Network (P-DQN), enabling a comparative assessment of policy-based and hybrid value based approaches for this task. The algorithms are evaluated in terms of learning stability, convergence behaviour, and the quality of the resulting design solutions. The results demonstrate the potential of RL-based methods to explore complex design spaces efficiently while respecting physical constraints, highlighting their suitability for supporting automated or decision-assisted design of diaphragm walls.

An Improved Model Reference Predictive Control Method for Building Mass Dampers

Deep learning empowered stochastic dynamics analysis of single-tower cable-stayed bridge with nonlinear tuned inerter dampers

A ridge-isolated tuned mass damper for seismic rehabilitation of a 3D structure under bidirectional earthquakes.

Efficient and practical seismic control of structures-particularly through isolation systems-remains a significant challenge due to high implementation costs and construction complexity. Moreover, their application in retrofitting existing structures often requires extensive modifications.To address these challenges, this study proposes a novel and cost-effective seismic control system, termed the Ridge-Isolated Tuned Mass Damper (RITMD), which integrates the advantages of tuned mass damping and seismic isolation. A reduced-order two-degree-of-freedom (2DOF) model is developed to capture the coupled dynamic behavior of the primary structure and the RITMD system. An optimization framework based on the Particle Swarm Optimization (PSO) algorithm is employed to determine the optimal tuning parameters. Subsequently, closed-form design expressions are obtained using nonlinear regression analysis to facilitate practical engineering applications .The proposed approach is further extended to three-dimensional structural systems, and its performance is evaluated through nonlinear time-history analyses under bidirectional earthquake excitations. The results demonstrate that the RITMD system significantly reduces structural responses, including displacements, accelerations, inter-story drifts, torsional demands, and their corresponding root-mean-square (RMS) values.Overall, the proposed system provides an efficient, practical, and versatile solution for enhancing the seismic performance of both new and existing structures.

A New Weighting Scheme for Ensemble Learning in Seismic Vibration Control of High-Rise Buildings with Soil–Structure Interaction

This paper presents a novel ensemble learning framework that integrates Adaptive Boosting Regression Threshold (AdaBoost.RT) with weighted extreme learning machines (WELMs) to improve seismic performance in intelligent control systems. In our approach, the Imperialist Competitive Algorithm (ICA) is used to optimize the relative error threshold of AdaBoost.RT, and WELMs are selected as base predictors due to their fast-training times and robustness in handling nonlinear phenomena. The proposed framework incorporates sample weights into the output weights of the ELMs and updates them iteratively, effectively capturing complex dynamic behaviors, including those arising from soil–structure interaction (SSI) under stochastic excitations. Comparative analyses with established ensemble learning techniques demonstrate that the presented optimized, fast, and efficient model achieves superior predictive accuracy, generalization capabilities, and computational efficiency. The methodology is validated through extensive simulations and experimental data, showcasing its potential for addressing smart control challenges in engineering applications. This research not only advances the state of the art in ensemble learning for active control systems but also contributes to the broader field of nonlinear dynamics by providing a reliable, efficient, and robust tool for system identification and vibration control in oscillating systems.