Vision-Aided Damage Detection Method with DIFun Model for Beam-Like Structures: A Novel Information Fusion of DOG Multi-Scale Space of Mode Shape

Overview: Structural Health Monitoring (SHM) plays a crucial role in maintaining the integrity and safety of aging infrastructures like bridges and buildings. Since the measurement noise is inevitable, structural damage detection with noisy information is still a challenge work. Methods: This paper presents a novel vision-aided damage detection method in noisy environments. For more accurate damage location detection, a vision technique called phase-based optical flow is introduced to measure structural mode shape with high-spatial resolution. Considering the characteristic that measurement noise is stochastic in multi-scale space, mode shape curvature (MSC) is transformed into Difference of Gaussian (DOG) multi-scale space to observe noise and damage features. To further address the problem of scale choosing, a Damage-Information-Fusion neural network (DIFun) with self-attention units are constructed for multi-scale information fusion and damage location. Conclusions: The datasets based on numerical simulations and vision-aided structure vibration measurements of beam-like structure are used for training the proposed fusion neural network. The proposed model maintains remarkable accuracy even under challenging signal-to-noise ratio conditions. The results of both simulations and experiments show that the proposed model can extract damage-sensitive and noise-robust features and perform high-precision damage location in noisy environments.