Optical-aided unsupervised structural health monitoring of buildings using convolutional autoencoders

As an essential component of smart city development, the integration of optical measurement technologies enables more efficient and non-contact structural monitoring in urban environments. With the rapid development of the low-altitude economy, urban airspace is increasingly being utilized for path planning and autonomous navigation of unmanned aerial vehicles (UAVs). In such environments, high-rise buildings serve not only as typical obstacles to be avoided but also as stable spatial references for vision-based navigation systems. This raises higher requirements for the accurate identification and health status perception of building structures, which directly impact the safety and robustness of flight path planning. Addressing the limitation of traditional structural health monitoring (SHM) methods that require large amounts of labeled damage-state data, this paper proposes an unsupervised SHM method based on convolutional autoencoders (CAE). The method uses only vibration data from structures in healthy states to train the model, enabling accurate reconstruction of similar healthy data. In contrast, input from damaged structures results in significant reconstruction errors, facilitating automatic detection of structural anomalies. A multistory building simulation model is established to verify the method’s effectiveness and its sensitivity to different levels of structural damage. This approach offers a promising solution for providing stable and perceptible spatial references for vision-based navigation systems in urban low-altitude environments, thereby supporting safe UAV path planning and enhancing the operational security of future low-altitude urban traffic systems.