Semi-Active Control of MRE Base Isolation Frame Structure Utilizing Online Learning Deep Neural Network Fuzzy Control

The shear stiffness of Magnetorheological Elastomers (MRE) can be rapidly adjusted in response to changes in the magnetic field, making it an ideal material for structural vibration control. This paper presents the development of an anisotropic MRE with a hydroxyl iron powder mass fraction of 27% and the manufacturing of a laminated MRE seismic isolator. The isolator core has a diameter of 60[Formula: see text]mm and consists of 26 layers of 1[Formula: see text]mm thick MRE interleaved with 25 steel plates. Innovative vertical steel rebars were designed to overcome the bottleneck in the vertical load-bearing capacity of MRE materials. Mechanical performance tests were conducted using various sine wave inputs. Parameter identification was then conducted by means of the particle swarm optimization (PSO) algorithm. Based on this, the Bouc–Wen hysteresis model was established to characterize the isolator’s response. A fuzzy logic controller integrated with an online learning deep neural network fuzzy control system (DFNN) was implemented using the MRE isolators, which achieved self-optimization of controller parameters through dynamic error monitoring. Shaking table tests were performed on a three-story steel frame model equipped with MRE isolators. The seismic responses of the fixed base, passive-controlled, ON-OFF controlled, deep neural network fuzzy control (DFNN0) controlled, and DFNN controlled structures were compared and analyzed. The results indicate that the DFNN control strategy demonstrates superior control efficiency and adaptability under various earthquake characteristics and load conditions, significantly reducing structural displacement, floor acceleration, and isolation layer displacement.