Efficient and automatic mode separation based on compact-bandwidth regularization and spectral-energy metrics with application to long-time continuous monitoring of bridge structures

Modal parameters are fundamental indicators of the dynamic characteristics of engineering structures. Automatic extraction of the structural modes from measured data is essential for structural health monitoring. This study presents a novel automatic mode separation method, facilitating real-time modal identification in fluctuating environmental conditions. The present method sequentially extracts natural frequencies and mode shapes from measured accelerations via utilizing a blind source separation technique known as the Compact-Bandwidth Regularization approach. To address the uncertainty in the number of active modes during continuous monitoring, criteria grounded in spectral energy and power spectral entropy are formulated, enabling the automatic mode separation and endowing the algorithm’s adaptive capabilities. Additionally, a similarity criterion is employed to differentiate between noise modes and structural modes, ensuring the validity of the identification. The accuracy and automation of the method are confirmed through numerical simulations involving a six-storey shear building model. Emphasizing practical application, the proposed method is applied to a long-span suspension bridge equipped with long-term vibration monitoring system. The one-year identification outcomes demonstrate that the proposed method can autonomously and efficiently extract active modes, fulfilling the requirements for real-time applications. Furthermore, the method’s derivational ability to detect the incipient stages of resonance is highlighted through a case study of vortex-induced vibration.