Damage Detection in Truss Structures Supporting Pipelines and Auxiliary Equipment in Power Plants

In this work we propose an automatic damage detection procedure for truss structures. The procedure exploits the natural frequencies of the structure, which can be estimated from vibrational signals measured by sensors, and provides as output the classification of the structure state as healthy or damaged. The approach developed for anomaly detection is based on the use of Principal Component Analysis (PCA) for the reconstruction of the natural frequencies as they should be in a healthy truss structure. Then, the occurrence of damage is detected by applying the Q-statistic test to the differences (residuals) between the observed natural frequencies and their values reconstructed by the PCA model. The proposed damage detection strategy is applied to a synthetic dataset containing the natural frequencies of healthy and damaged truss structures obtained by finite element simulations. The frequency distributions account for structural properties and boundary conditions variability, possibly introduced by variation in the structure operational conditions (e.g., ambient temperature, fluid flow, environmental noise). The obtained results show that the proposed model is able to correctly recognize the state of the truss structure with a limited number of false and missed alarms.