Structural damage identification via multi-type sensors and response reconstruction

One outstanding obstacle that hinders robust application of vibration-based damage identification to civil structures is that the number of sensors installed on a large civil structure is always limited, compared with the total degrees of freedom of the structure, so that the limited measured responses may not provide enough information for detecting local damage. Furthermore, developments in sensor technology make installation of heterogeneous sensors on a structure practical and feasible while every type of sensor has its own merits and drawbacks for damage identification. But the benefits of utilizing heterogeneous sensors in vibration-based damage identification have not been fully investigated. This study proposes a damage identification method by combining the response reconstruction technique with the response sensitivity-based finite element model updating method to address these issues. The number and location of heterogeneous sensors, such as accelerometers, displacement transducers, and strain gauges, are optimally and collectively determined in an optimization strategy to obtain the best reconstruction of multi-type responses of a structure using Kalman filter. After damage occurrence, radial basis function network is employed to predict the mode shapes using the modal properties extracted from the measurement data by experimental modal analysis method, and these modal properties are further used to reconstruct responses of the damaged structure. The reconstructed responses are finally used to identify the damage in terms of sensitivity-based finite element model updating. In every updating, the sparse regularization is employed to increase the identification accuracy. A simply supported overhanging steel beam composed of 40 elements serves as a numerical study to demonstrate the procedure and feasibility of the proposed method. The validation of this method is further conducted by laboratory test. Both simulation study and laboratory test show that the multi-sensing approach via response reconstruction does improve the identification accuracy of damage location and quantization considerably.