| Abstrakt: |
Purpose: This study investigates the application of Artificial Neural Networks (ANNs) for controlling Active Tuned Mass Dampers (ATMDs) in seismic response reduction. The objective is to develop an AI-based ANN controller that effectively reduces structural vibrations during earthquakes. This approach offers a key advantage: achieving good response reduction with fewer sensors compared to traditional methods like Linear Quadratic Regulators (LQR), leading to increased practicality and cost-effectiveness. Methods: A supervised learning approach with the Levenberg–Marquardt backpropagation algorithm trains the ANN controller. The performance of the ANN-controlled ATMD is compared with that of an LQR-controlled system. Additionally, the ANN controller's robustness under signal time delay and noise contamination is evaluated. The ATMD with both controllers is implemented on a 10-story benchmark building subjected to near-field and far-field seismic records. Results: The obtained results indicate significant reductions in peak displacement, acceleration, velocity, inter-story drift, maximum drift, base shear, and structural energy. Notably, the ANN controller achieves this with a reduced sensor requirement compared to the LQR method. Further, the ANN showed good robustness against signal time delay and noise contamination. Conclusion: ANNs demonstrated a high potential for controlling ATMDs for seismic response reduction due to their effectiveness and reduced sensor requirements, making them a conceivably more practical and cost-effective solution. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink