Experimental validation of a semi-active fuzzy control strategy based on deep reinforcement learning for a piezoelectric smart isolation system

The high-tech industry, commanding a substantial portion of the global market, remains highly susceptible to disruptions caused by earthquakes because its products rely heavily on vibration-sensitive equipment. Moreover, the characteristics of seismic ground motions are usually difficult to predict; therefore, seismic protection of high-tech equipment is a challenging task. In this context, this study leverages the robustness and adaptability of artificial intelligence techniques, particularly deep reinforcement learning (DRL) and a piezo-electric smart isolation system (PSIS) to protect high-tech equipment. The control objective of the PSIS is to dynamically optimize the isolation performance of the PSIS in real time, accommodating diverse ground motion characteristics. The proposed control strategy for PSIS involves a DRL framework integrated with a fuzzy inference system (FIS) and non-striking friction (NSF) control technique. In the DRL framework, a Deep Deterministic Policy Gradient (DDPG) algorithm is utilized to train the DRL model, employing tailored reward functions to achieve diverse control objectives under varying seismic conditions. The efficacy of the proposed PSIS control strategy is evaluated through numerical simulations and experimental validation. The proposed DRL control module demonstrated superior isolation performance compared to other control algorithms, particularly in reducing isolation displacement by approximately 24% during near-fault earthquakes and superstructure acceleration of 39% reduction under far-field earthquakes. The research outcomes underscore the effectiveness of the proposed approach, positioning it as a promising solution for fortifying the seismic resilience of vibration-sensitive equipment and a strong support of advancing the current nano-scale manufacturing process in the high-tech industry.