This thesis proposes original contributions aimed at structural durability analysis with random loadings. More precisely, a method to assess the variance of the fatigue damage caused by the randomness of a stationary Gaussian random loading is first presented. After reviewing some methods from the literature, the thesis compares a Monte Carlo simulation study in time- and frequency-domain with such theoretical methods. Best-fitting expressions are then derived to relate the variance of the damage directly to bandwidth parameters of a Power Spectral Density (PSD). The proposed expressions apply to a wide range of PSDs, from narrow-band to wide-band processes. Two theoretical models to assess the variance of the fatigue damage in non-Gaussian case are further proposed. The models extend two solutions existing in the literature and restricted to Gaussian processes. The developed models exploit a non-linear transformation that links Gaussian and non-Gaussian domains based on skewness and kurtosis coefficients, which are used to quantify the deviation from the Gaussian distribution. Monte Carlo numerical simulations in time-domain are performed to confirm the correctness of the proposed non-Gaussian models. As the previous methods require that the random load is stationary, attention is focused on statistical methods to identify stationary and non-stationary loadings. Therefore, a damage-based run test approach is suggested to detect not only changes in the variance and mean levels, but also the frequency content of a random time-history with finite time length. In summary, the non-parametric run test approach takes a sequence of non-overlapping blocks to verify the stationarity of time-histories. For each block, a value is calculated for the statistical parameter under investigation. Rather than using the usual statistical parameters, e.g. the root-mean-square value, as the output calculated in each block, the proposed run test approach considers the damage computed for each block. The correctness of the damage-based run test is first verified by numerical simulations. Afterward, the accuracy of the proposed approach is confirmed by checking the stationarity hypothesis of measured time-history records from a Mountain-bike. The thesis also analyses the variability of the damage when computed from one or few time-histories. For both cases, confidence interval expressions are derived to enclose the exact (but unknown) expected damage. Using a numerical example, which considers a stress PSD in an offshore platform, the confidence interval expressions reveal a good agreement with simulations. The proposed confidence intervals for expected damage is also investigated by measuring the random loadings acting on an instrumented Mountain-bike. As the whole ensemble of an infinite number of time-histories is not available and the expected damage is thus not known a priori, a sort of calibrator sample damage value, which is computed using a large number of measured time-history records, is used to estimate the expected damage. The obtained results confirm the accuracy of the proposed approach also with real measurements. Last, but not least, the thesis presents a new algorithm to implement the Carpinteri-Spagnoli-Vantadori (CSV) multiaxial fatigue criterion for random loading and to shorten the computation time. This goal is achieved after calculating the exact expressions of stress spectral moments in every rotated plane, which allow the maximum variance and expected maximum peak of normal/shear stress to be computed directly. This allows the new algorithm to determine the five rotations of the critical plane without using 'for/end' loops. Two examples are presented to compare the new algorithm with its standard version, which are particularly remarkable when considering the stress output of all finite element model nodes. The approach behind the new algorithm can also be extended to other multiaxial spectral criteria.

La seguente tesi presenta contributi originali nel campo dell'analisi della durabilità strutturale con carichi casuali. Più precisamente, viene prima presentato un metodo per valutare la varianza del danno a fatica provocato dalla casualità di un carico casuale gaussiano stazionario. Dopo aver esaminato alcuni metodi dalla letteratura, la tesi confronta uno studio di simulazione Monte Carlo nel dominio del tempo e della frequenza con tali metodi teorici. Vengono quindi ricavate le espressioni più adatte per mettere in relazione la varianza del danno direttamente ai parametri di banda di una funzione di densità spettrale (PSD). Le espressioni proposte si applicano a un'ampia gamma di PSD, dai processi a banda stretta a quelli a banda larga. Vengono inoltre proposti due modelli teorici per valutare la varianza del danno a fatica nel caso non gaussiano. I modelli estendono due soluzioni esistenti in letteratura limitate ai processi gaussiani. I modelli sviluppati sfruttano una trasformazione non lineare che collega i domini gaussiano e non gaussiano, in base a dei coefficienti che vengono utilizzati per quantificare la deviazione dalla distribuzione gaussiana. Vengono eseguite simulazioni numeriche Monte Carlo nel dominio del tempo per confermare la correttezza dei modelli non gaussiani proposti. Poiché i metodi precedenti richiedono che il carico casuale sia stazionario, l'attenzione è focalizzata sui metodi statistici per identificare i carichi stazionari e non stazionari. Pertanto, si suggerisce l'approccio del run test basato sul danno per rilevare non solo i cambiamenti nella varianza e nel valore medio, ma anche il contenuto in frequenza di un carico casuale con lunghezza finita. In sintesi, l'approccio del run test non parametrico utilizza una sequenza di blocchi non sovrapposti per verificare la stazionarietà del carico. Piuttosto che utilizzare i soliti parametri statistici, ad es. il valore quadratico medio, come output calcolato in ciascun blocco, l'approccio proposto considera il danno calcolato per ciascun blocco. La correttezza del run test basato sui danni viene prima verificata mediante simulazioni numeriche. Successivamente, l'accuratezza dell'approccio proposto è confermata controllando l'ipotesi di stazionarietà dei carichi misurati da una mountain bike. La tesi analizza anche la variabilità del danno quando calcolata da una o da alcune storie di carico. Per entrambi i casi, le espressioni dell'intervallo di confidenza vengono derivate in modo tale che il danno atteso esatto (ma sconosciuto) ricada all’interno dell'intervallo di confidenza. Utilizzando un esempio numerico, che considera carico PSD in una piattaforma offshore, le espressioni dell'intervallo di confidenza sono in accordo con le simulazioni. Gli intervalli di confidenza proposti per il danno atteso vengono anche studiati misurando i carichi casuali che agiscono su una mountain bike. Poiché l'intero insieme di un numero infinito di carichi non è disponibile e il danno atteso non è quindi noto a priori, una sorta di valore di danno del campione del calibratore (calcolato utilizzando un gran numero di carichi) viene utilizzata per stimare il danno atteso. I risultati ottenuti confermano l'accuratezza dell'approccio proposto in riferimento alle misurazioni reali. Infine, la tesi presenta un nuovo algoritmo per implementare il criterio a fatica multiassiale con caricamento casuale Carpinteri-Spagnoli-Vantadori (CSV) e per ridurre il tempo di calcolo. Questo obiettivo viene raggiunto dopo aver calcolato le espressioni esatte dei momenti spettrali di sollecitazione in ogni piano ruotato. Ciò consente al nuovo algoritmo di determinare le cinque rotazioni del piano critico senza utilizzare i loop "for / end". Vengono presentati due esempi per confrontare il nuovo algoritmo con la sua versione standard, che sono più veloce computazionale se si considera l'output di stress di tutti i nodi del modello agli elementi finiti.

Structural durability analysis with random loadings

ENZVEILER MARQUES, Julian Marcell
2021

Abstract

This thesis proposes original contributions aimed at structural durability analysis with random loadings. More precisely, a method to assess the variance of the fatigue damage caused by the randomness of a stationary Gaussian random loading is first presented. After reviewing some methods from the literature, the thesis compares a Monte Carlo simulation study in time- and frequency-domain with such theoretical methods. Best-fitting expressions are then derived to relate the variance of the damage directly to bandwidth parameters of a Power Spectral Density (PSD). The proposed expressions apply to a wide range of PSDs, from narrow-band to wide-band processes. Two theoretical models to assess the variance of the fatigue damage in non-Gaussian case are further proposed. The models extend two solutions existing in the literature and restricted to Gaussian processes. The developed models exploit a non-linear transformation that links Gaussian and non-Gaussian domains based on skewness and kurtosis coefficients, which are used to quantify the deviation from the Gaussian distribution. Monte Carlo numerical simulations in time-domain are performed to confirm the correctness of the proposed non-Gaussian models. As the previous methods require that the random load is stationary, attention is focused on statistical methods to identify stationary and non-stationary loadings. Therefore, a damage-based run test approach is suggested to detect not only changes in the variance and mean levels, but also the frequency content of a random time-history with finite time length. In summary, the non-parametric run test approach takes a sequence of non-overlapping blocks to verify the stationarity of time-histories. For each block, a value is calculated for the statistical parameter under investigation. Rather than using the usual statistical parameters, e.g. the root-mean-square value, as the output calculated in each block, the proposed run test approach considers the damage computed for each block. The correctness of the damage-based run test is first verified by numerical simulations. Afterward, the accuracy of the proposed approach is confirmed by checking the stationarity hypothesis of measured time-history records from a Mountain-bike. The thesis also analyses the variability of the damage when computed from one or few time-histories. For both cases, confidence interval expressions are derived to enclose the exact (but unknown) expected damage. Using a numerical example, which considers a stress PSD in an offshore platform, the confidence interval expressions reveal a good agreement with simulations. The proposed confidence intervals for expected damage is also investigated by measuring the random loadings acting on an instrumented Mountain-bike. As the whole ensemble of an infinite number of time-histories is not available and the expected damage is thus not known a priori, a sort of calibrator sample damage value, which is computed using a large number of measured time-history records, is used to estimate the expected damage. The obtained results confirm the accuracy of the proposed approach also with real measurements. Last, but not least, the thesis presents a new algorithm to implement the Carpinteri-Spagnoli-Vantadori (CSV) multiaxial fatigue criterion for random loading and to shorten the computation time. This goal is achieved after calculating the exact expressions of stress spectral moments in every rotated plane, which allow the maximum variance and expected maximum peak of normal/shear stress to be computed directly. This allows the new algorithm to determine the five rotations of the critical plane without using 'for/end' loops. Two examples are presented to compare the new algorithm with its standard version, which are particularly remarkable when considering the stress output of all finite element model nodes. The approach behind the new algorithm can also be extended to other multiaxial spectral criteria.
TOVO, Roberto
BENASCIUTTI, Denis
TRILLO, Stefano
File in questo prodotto:
File Dimensione Formato  
STRUCTURAL DURABILITY ANALYSIS WITH RANDOM LOADINGS.pdf

accesso aperto

Descrizione: STRUCTURAL DURABILITY ANALYSIS WITH RANDOM LOADINGS
Tipologia: Tesi di dottorato
Dimensione 4.72 MB
Formato Adobe PDF
4.72 MB Adobe PDF Visualizza/Apri

I documenti in SFERA sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11392/2478820
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact