Experimental study on structural damage identification of multi-sensor separated channel network

Structural damage identification, the core driver of structural health monitoring (SHM), has gradually shifted to data-driven models in recent years. For the characteristics of multi-sensors in the SHM system of buildings, this paper proposes a neural network structure for damage identification called multi-sensor separated channel network (MSSCN). It is dedicated to extracting damage sensitive features from a single measurement channel while maintaining the spatial information of the sensor, thus completely decoupling intra-channel time and interchannel spatial features. The proposed framework is validated by applying to the Qatar University Grandstand Simulator (QUGS) with a decentralized sensor arrangement. Further, we study vertical structures with strong channel correlation via experiments simulating linear and nonlinear multiple complex operating conditions to identify damage in transmission tower. Based on the above, it is demonstrated that the proposed model exhibits good diagnostic capability and performs competitive in noise immunity and small sample tests.