Machine learning driven damper for response control in vehicle–bridge interaction systems

Abstract

The implementation of machine learning for the real-time prediction of the suitable value of the damping ratio of a semi-active tuned mass damper (SA-TMD) is investigated to ensure enhanced vibration control in vehicle–bridge interaction (VBI) problems. The response assessment of the uncontrolled, tuned mass damper (TMD)-controlled, and SA-TMD-controlled bridge models is performed under the Japanese Shinkansen (SKS) train model. The energy-based predictive (EBP) control algorithm is implemented for the bridge fitted with the SA-TMD. The EBP algorithm-controlled SA-TMD results in more effective suppression of the bridge vibration as compared to the passive TMD. However, the effectiveness of the EBP algorithm reduces for more complex VBI systems because of the increased computational time delay. To circumvent the effect of the delay, a control strategy is proposed based on the weighted random forest (WRF) algorithm. The WRF algorithm is trained based on the data obtained from the EBP algorithm-controlled bridge and implemented to suppress the vehicle-induced vibration of the bridge using SA-TMD. The results demonstrate that the implementation of the newly proposed WRF algorithm-based control strategy nullifies the effects of the computational time delay. Furthermore, it is established that the WRF algorithm suppresses the bridge vibration more effectively than the EBP algorithm.