A deep learning‐based crack damage detection in exterior finish of timber houses in Japan for automated damage diagnostic system

Abstract

Abstract To ensure the speed, accuracy, and fairness of the building damage survey, a deep learning diagnostic support technique for crack damage that occurs around openings such as windows on the exterior finish is studied. Firstly, using training images based on fake crack images generated by Deep Convolutional Generative Adversarial Network (DCGAN), a deep learning model that can detect with high accuracy was constructed. Generally, in order to clarify the relationship between exterior finish damage and inter‐story drift ratio, it is necessary to analyze a wide range of intervening parameters. Secondly, this paper begins by focusing on “the relationship between exterior finish surface damage and the deformation of the exterior finish itself”, and this relationship is investigated by an optical measurement method applied for the in‐plane shear force tests. Finally, to analyze the relationship between the deep learning model and the image acquisition conditions, outdoor camera shooting test of damaged specimens was conducted.

Keywords

• deep learning
• image acquisition conditions
• image diagnosis
• post‐earthquake damage inspection
• timber house