Automatic Methods for Modal- Based Structural Health Monitoring
Doctoral thesis
Permanent lenke
https://hdl.handle.net/11250/3135168Utgivelsesdato
2024Metadata
Vis full innførselSamlinger
Sammendrag
It is necessary to know the actual state of a structure to ensure its safe continued operation. Structural health monitoring can provide this information, given a monitoring network of sensors on the structure. A prompt intervention in the case of damage promotes security, easier and cheaper repairs, and prolongs the life of a structure, which is environmentally beneficial. For structural health monitoring to be deployed widely, its methods must be as automatic as possible.
This thesis proposes an automatic approach to modal-based structural health monitoring and damage detection through automatic operational modal analysis (AOMA) and mode tracking (MT).
A comparison of prominent AOMA algorithms shows that near-automatic algorithms perform better than their fully automatic counterparts. By combining the best subroutines from the compared algorithms, a new fully automatic operational modal analysis algorithm is proposed, rivalling the best near-automatic ones in terms of performance. The algorithms are tested using data recorded from the Hardanger suspension Bridge, the Bergsøysund floating pontoon bridge, and the Hålogaland suspension Bridge.
A new mode tracking algorithm is presented, allowing the evolution of modal features from the AOMA to be tracked across datasets. The MT algorithm detects which structural modes should be tracked and tracks them through small and large changes to the modal features, given a few intuitively selected parameters. It applies to any structure and is independent of the choice of AOMA algorithm. The MT algorithm performs very well at detecting structural modes and tracking them for numerical test cases and when using real-world data.
Finally, a damage detection algorithm is suggested based on the Mahalanobis squared distance and natural frequencies. The algorithm is intended to work with missing and randomly available data. AOMA and MT do not detect natural frequencies for each structural mode from every dataset. The suggested damage detection algorithm maximises the use of the available natural frequency data. It functions as intended on numerical test data and can detect small, simulated damages. Its performance is confirmed on data from a real bridge with a controlled imposed damage, being able to detect the damage even though its effect was below the scale of the environmentally driven variations in the bridge’s natural frequencies.
Overall, this thesis proposes new methods to go from structural monitoring data to damage detection following an automatic, modal-based, structural health monitoring framework.