Deep learning-based structural health monitoring through the infusion of optical photos and vibration data

This paper reports an investigation of deep learning techniques in structural damage identification that can overcome the limitations of traditional visual inspection. First, a vibration-based deep learning model is established to locate the damage in a beam and a truss structure. Then an optical photo-based model is established and used to classify different defects. Based on the satisfactory outcomes of these two models, a new structural health monitoring technique is proposed through the infusion of optical photos and vibration data. Vibration signals and true structural images for a truss are used to demonstrate the capability of the proposed method. It was found that the infusion of vibration data and optical photos can enhance damage identification significantly and overcome the drawbacks in the existing deep learning models due to incomplete vibration signals or blurred optical photo inputs.