Inception Time Model for Structural Damage Detection Using Vibration Measurements

Structural health monitoring (SHM), facilitating the detection and identification of damages at an early stage in any important large Engineering structures, can aid significantly in proper functionality and maintenance, thereby enhancing structural safety to a great extent. In recent times, autonomous approaches using deep learning (DL)-based models have played a critical part in the solution of various SHM problems such as structural damage detection (SDD). However, the majority of these DL-based approaches are not capable of capturing time series information with limited computational complexity. To handle this, a deep learning model based on the inception module has been investigated to perform global-level structural damage detection by using acceleration response (time series) data under the ambient vibration scenario. An ensemble of deep convolutional neural networks is a major component of the inception module and it is inspired by the inception module to capture the time series information. This module focuses on increasing network width rather than depth, thereby reducing the problem of over-fitting. The proposed methodology is validated numerically using data of ten storied building structures and the ASCE benchmark structure considering multiple damage-case scenarios. The results are found to be encouraging and considerably better in terms of performance accuracy and computation as compared to other well-known approaches. The proposed methodology thus establishes itself as an interesting SHM tool for SDD.