Politecnico di Torino (logo)

Clustering strategies in automated operational modal analysis for structural health monitoring

Luigi Sibille

Clustering strategies in automated operational modal analysis for structural health monitoring.

Rel. Rosario Ceravolo, Luca Zanotti Fragonara, Marco Civera. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Civile, 2021

PDF (Tesi_di_laurea) - Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives.

Download (6MB) | Preview

In the last decades the interest in studying the dynamic properties of the structures in the field of civil engineering soared rapidly due to the fast development of large??scale civil structures. The analysis of the dynamic behaviour can reveal progressive damages as well as estimate the residual service life. Indeed, modal parameters and their evolution in time can be used as indicators and features of structural weaknesses or deficiencies induced by unforeseen events. The impossibility to excite large size structures with artificial devices and to record data in continuous fully controlled testing environments, leads to the use of output??only records. For this reason, different methods based on vibration data collected during working conditions have been proposed over the years such as the Operational Modal Analysis. However, some challenges in improving the accuracy and the computational cost-effectiveness are still open. In this study, a novel multi??stage clustering approach for Automated Operational Modal Analysis is proposed. When parametric system identification methods are used, the candidate modes are represented by means of a stabilization diagram. In the proposed method the system identification method used is the stochastic subspace identification (SSI). Two-stage clustering techniques are implemented for interpreting such diagram. In the first stage, the cleansing out of certainty mathematical poles from the diagram is addressed through the use of the K-means clustering algorithm, whereas the DBSCAN clustering is developed for detecting columns of stable poles and identify the remaining outliers. Contrarily to the existing approaches, DBSCAN clusters the poles and detect the outliers in only step. The parameters needed to perform the clustering algorithm are automatically estimated using a cluster validation criterion and a very popular heurist method in clustering analysis. Extensive validation case studies illustrate the robustness and the performance of the proposed AOMA method. The Z24 bridge benchmark, an experimental case study regarding a helicopter blade and a numerical case are indeed analyzed.

Relators: Rosario Ceravolo, Luca Zanotti Fragonara, Marco Civera
Academic year: 2021/22
Publication type: Electronic
Number of Pages: 123
Corso di laurea: Corso di laurea magistrale in Ingegneria Civile
Classe di laurea: New organization > Master science > LM-23 - CIVIL ENGINEERING
Aziende collaboratrici: Cranfield University
URI: http://webthesis.biblio.polito.it/id/eprint/20653
Modify record (reserved for operators) Modify record (reserved for operators)