Sichuan Provincial Architectural Design and Research Institute (China)

ABSTRACT This paper presents a study on the development of a deep learning-based methodology for rapidly predicting the floor response spectrum (FRS) of multi-story buildings. The proposed approach aims to accurately predict both elastic and inelastic floor response spectra using surrogate-modelling techniques, significantly reducing the computational cost compared to traditional physics-based time-history analyses. Advanced deep learning techniques, such as bidirectional long short-term memory (Bi-LSTM) networks and Transformers are employed. These models are trained on datasets derived from numerical simulations as well as structural health monitoring data collected from real earthquake events. To develop and validate the models, numerical simulations conducted via OpenSees generate datasets comprising FRS data from 16 multi-story buildings modelled as simplified MDOF systems and 10 buildings modelled with detailed finite element representations. A total of 1030 ground motions are considered. In addition, FRS data from 96 multi-story buildings obtained from real earthquake events are incorporated into the model development process. The model’s performance is thoroughly evaluated using metrics, including root mean square error (RMSE), mean absolute error (MAE), mean squared error (MSE), and the coefficient of determination (R2). The proposed surrogate model achieves an accuracy of 0.95, demonstrating its effectiveness in predicting the FRS of multi-story buildings.

This study focuses on the reliability issue of magnetic fluid (MF) in the magnetic fluid sealing technology for the upper guide bearing (UGB) of hydro-generators and proposes selection schemes for MF suitable for different models of hydro-generators. By analyzing the performance indicators of five base fluids and MFs, including the acid value, flash point, oxidation stability, magnetorheological performance, breakdown voltage, dielectric loss factor and volume resistivity, the influencing factors of the insulating performance of MFs and their mechanism in sealing the UGBs of hydro-generators are investigated. The results show that, when the spindle speed is below 27 rpm, the viscosity of the MF is dominated by the magnetic field strength, while, when the speed exceeds 27 rpm, the viscosity of the MF is dominated by the shear rate. In addition, the addition of magnetic nanoparticles (MNPs) causes the breakdown voltage of the base carrier liquid to fluctuate in the range of 31.2–55.9 kV, the dielectric loss factor to fluctuate in the range of 2.5 × 10−4–6.7 × 10−3, and the volume resistivity to fluctuate in the range of 2.8 × 1011–2.6 × 1012 Ω·m. The research results provide a theoretical basis for the application of high-efficiency and stable magnetic fluid sealing technology.

Based on the fundamental theory of vehicle-bridge coupling vibration, this study proposes a novel method for identifying the bridge damping ratio from the vehicle-bridge interaction by using the bridge’s mode shape. This study conducts theoretical derivation via indirect measurement techniques and performs numerical simulation analysis using a finite element model built in MATLAB. First, acceleration sensors are used to measure the bridge response caused by a single-axle vehicle crossing; subsequently, band-pass filtering is employed to extract the vibration component corresponding to the bridge’s first-order frequency, and the Hilbert transform is used to identify the first-order mode shape curve. Ultimately, the bridge damping ratio is rapidly identified by leveraging the analytical relationship between the damping and the location of the mode shape’s peak. The effectiveness of the method is verified by theoretical derivations and numerical simulations, with a relative error of only 2% under the baseline condition. More importantly, parametric analysis shows that the method has strong robustness, with the relative identification error remaining within 5% even at a higher vehicle speed of 8 m/s and in a strong noise environment of 20 dB. Compared with traditional indirect methods that rely on iterative calculations, the method proposed in this study provides a more efficient, direct analytical identification approach, offering a novel idea for structural health monitoring based on indirect measurement.