Mechanical structures and civil infrastructure are systems prone to suffer damage after long duty times and varying environmental and operational conditions, which might affect their structural behaviour. Maintenance, in general terms, is evolving towards Condition Based Maintenance and Predictive Maintenance, which requires a good knowledge of the health status of the systems to be maintained. In the context of mechanical and/or civil structures, several approaches have been proposed during the years to tackle the Structural Health Monitoring issue and accurately estimate the structure health state. Yet, it remains difficult to diagnose damages and estimate the structural health in the presence of varying operating and environmental conditions. Particle Filters have already been proposed as a time-domain-based method in the field of SHM, showing promising results as an estimator of hidden states. On the other hand, neural networks-based autoencoders have been used for structural damage detection, extracting damage-related features from vibration measurements. In this thesis work is proposed a combination of particle filters with autoencoders in order to obtain an algorithm for structural damage identification and diagnosis, robust to changing environmental conditions and both linear and non-linear damages. An adaptive threshold is used to reduce human evaluation and create an automatic indicator.
Le strutture meccaniche e le infrastrutture civili sono sistemi soggetti a degradazione dopo lunghi periodi di servizio e condizioni ambientali e operative variabili. La manutenzione, in generale, si sta evolvendo verso la Condition Based Maintenance e la Predictive Maintenance, che richiede una buona conoscenza dello stato di salute dei sistemi da mantenere. Nel contesto delle strutture meccaniche e/o civili, diversi approcci sono stati proposti per affrontare il problema di Structural Health Monitoring. Tuttavia, rimane ancora difficile diagnosticare i danni e stimare lo stato di salute strutturale in presenza di condizioni ambientali variabili. I Particle Filters sono stati usati come metodo basato sul dominio del tempo nel campo del SHM, mostrando promettenti risultati come estimatore degli stati. D'altra parte, gli Autoencoder sono stati utilizzati per la rilevazione dei danni strutturali, estraendo le caratteristiche relative con i danni. In questa tesi si propone una combinazione di Particle Filters con Autoencoders per ottenere un algoritmo robusto per diagnosticare e rilevare danni strutturali sotto condizioni ambientali variabili e guasti lineari e non-lineari. Una soglia adattativa è stata utilizzata per ridurre la valutazione umana e creare un indicatore automatico.
Non-linear structural identification and damage diagnosis under changing environmental conditions using a combination of particle filters and autoencoders
FERRATER ROCA, MARC
2018/2019
Abstract
Mechanical structures and civil infrastructure are systems prone to suffer damage after long duty times and varying environmental and operational conditions, which might affect their structural behaviour. Maintenance, in general terms, is evolving towards Condition Based Maintenance and Predictive Maintenance, which requires a good knowledge of the health status of the systems to be maintained. In the context of mechanical and/or civil structures, several approaches have been proposed during the years to tackle the Structural Health Monitoring issue and accurately estimate the structure health state. Yet, it remains difficult to diagnose damages and estimate the structural health in the presence of varying operating and environmental conditions. Particle Filters have already been proposed as a time-domain-based method in the field of SHM, showing promising results as an estimator of hidden states. On the other hand, neural networks-based autoencoders have been used for structural damage detection, extracting damage-related features from vibration measurements. In this thesis work is proposed a combination of particle filters with autoencoders in order to obtain an algorithm for structural damage identification and diagnosis, robust to changing environmental conditions and both linear and non-linear damages. An adaptive threshold is used to reduce human evaluation and create an automatic indicator.| File | Dimensione | Formato | |
|---|---|---|---|
|
2019_04_FerraterRoca.pdf
non accessibile
Descrizione: Testo della tesi
Dimensione
2.57 MB
Formato
Adobe PDF
|
2.57 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/148132