Study of diagnosing damages in steel frame structures using improved Artificial Neural Network

Structural Health Monitoring (SHM) is a crucial approach for tracking and providing early warnings of structural degradation in civil infrastructure. However, in real-world applications, vibration data are often incomplete due to sensor limitations, making accurate damage diagnosis challenging. This study proposes an improved artificial neural network model (ANN-PSO-GA-AE), which integrates a hybrid Particle Swarm Optimization–Genetic Algorithm and an Autoencoder to enhance prediction performance under limited data conditions. The model is capable of simultaneously identifying the damaged structural element and estimating its stiffness reduction from incomplete vibration data. Validation on a four-story steel frame structure demonstrates that the proposed model maintains stable accuracy and generalization capability, even when full measurement data are not available. This approach shows strong potential for application in practical SHM systems, particularly in civil and industrial structures where sensor deployment is constrained.