Intelligent Structural Health Monitoring (SHM) of bridge is a technology that
utilizes advanced sensor technology along with professional bridge engineering
knowledge, coupled with machine vision and other intelligent methods for
continuously monitoring and evaluating the status of bridge structures. One
application of SHM technology for bridges by way of machine learning is in the
use of damage detection and quantification. In this way, changes in bridge
conditions can be analyzed efficiently and accurately, ensuring stable
operational performance throughout the lifecycle of the bridge. However, in the
field of damage detection, although machine vision can effectively identify and
quantify existing damages, it still lacks accuracy for predicting future damage
trends based on real-time data. Such shortfall l may lead to late addressing of
potential safety hazards, causing accelerated damage development and threatening
structural safety. To tackle this problem, this study designs a deep learning
model based on temporal information to solve the problem of predictive damage
development, achieving early warning and dynamic evaluation effects. This study
focuses on concrete crack development, and the CrackAE model is based on
traditional semantic segmentation models and conditional autoencoder
architecture. The model consists of an encoder and a decoder. The encoder
accepts image data and outputs a feature map. The future map along with the
conditional vector encoded based on physical temporal information, serves as the
input to the decoder. The output of decoder is the development state of the
crack at the specified prediction time. The model achieved an accuracy of 94.6%
in real bending failure tests of concrete beams, indicating that the model meets
high-precision prediction requirements. This validates the feasibility of deep
learning in predicting damage development and provides new ideas for data
collection and prediction in actual bridge maintenance.