This thesis is devoted to the life-cycle structural reliability analysis and optimal maintenance strategy of bridges with time-variant performance due to ageing and structural deterioration. Structural reliability analysis is performed by Monte Carlo simulation, considering the main source of uncertainties involved in the problem. Surrogate models are constructed by using the response surface method (RSM) as an alternative to complex and time-consuming structural analysis. RSM generates approximate results for nonlinear analysis by means of explicit equations for specified inputs, which significantly reduces the computational effort involved in doing structural reliability analysis using a large number of samples. The training sample set used to generate the response surface consists of the inputs generated by various Design of Experiment (DoE) methods and the theoretical outputs obtained from classical mechanical models. Different DoE methods and forms of RSM have been compared for their computational efficiency and accuracy. Data-driven surrogate models are black boxes where the internal implementation process from inputs to outputs is opaque. Explanation of surrogate models is necessary for analysing the reliability and maintenance of structures that may endanger the lives and property of citizens. Explainable techniques help monitor the mapping relationships between inputs and outputs found by a surrogate model and have a human judge whether these relationships apply to the actual situation, increasing society's confidence in the black-box model. Accumulated Local Effects (ALE) plot is one of explainable techniques to visualise the effect of predictor variables on the predicted response. This thesis presents a framework of moving least squares based response surfaces method (MLS-RSM), regulated and explained by ALE plots, for structural life cycle reliability analysis. Meanwhile, the optimal preventive time-based maintenance strategy has been identified in the visualisation of the impact of the design variables of the maintenance strategy on the effectiveness and cost provided by ALE plots.
La tesi è dedicata all'analisi dell'affidabilità strutturale del ciclo di vita e alla manutenzione ottimale di ponti con prestazioni variabili nel tempo a causa dei fenomeni di invecchiamento e degrado strutturale. L'analisi di affidabilità viene eseguita mediante simulazione Monte Carlo considerando le principali incertezze coinvolte nel problema. Si sviluppano dei modelli surrogati utilizzando il metodo della superficie di risposta (Response Surface Method) come alternativa allo svolgimento di analisi strutturali che possono risultare complesse e costose in termini computazionali, soprattutto per un numero elevato di simulazioni. L'approccio proposto viene presentato e discusso, anche in relazione ad altri metodi di letteratura, evidenziando vantaggi e svantaggi. L'accuratezza dei risultati e l'efficacia della metodologia vengono mostrate con una serie di casi prova e applicazioni all'analisi di affidabilità e alla programmazione di interventi di manutenzione di ponti.
Response surface methods for life-cycle structural reliability analysis
Ning, Xihong
2022/2023
Abstract
This thesis is devoted to the life-cycle structural reliability analysis and optimal maintenance strategy of bridges with time-variant performance due to ageing and structural deterioration. Structural reliability analysis is performed by Monte Carlo simulation, considering the main source of uncertainties involved in the problem. Surrogate models are constructed by using the response surface method (RSM) as an alternative to complex and time-consuming structural analysis. RSM generates approximate results for nonlinear analysis by means of explicit equations for specified inputs, which significantly reduces the computational effort involved in doing structural reliability analysis using a large number of samples. The training sample set used to generate the response surface consists of the inputs generated by various Design of Experiment (DoE) methods and the theoretical outputs obtained from classical mechanical models. Different DoE methods and forms of RSM have been compared for their computational efficiency and accuracy. Data-driven surrogate models are black boxes where the internal implementation process from inputs to outputs is opaque. Explanation of surrogate models is necessary for analysing the reliability and maintenance of structures that may endanger the lives and property of citizens. Explainable techniques help monitor the mapping relationships between inputs and outputs found by a surrogate model and have a human judge whether these relationships apply to the actual situation, increasing society's confidence in the black-box model. Accumulated Local Effects (ALE) plot is one of explainable techniques to visualise the effect of predictor variables on the predicted response. This thesis presents a framework of moving least squares based response surfaces method (MLS-RSM), regulated and explained by ALE plots, for structural life cycle reliability analysis. Meanwhile, the optimal preventive time-based maintenance strategy has been identified in the visualisation of the impact of the design variables of the maintenance strategy on the effectiveness and cost provided by ALE plots.| File | Dimensione | Formato | |
|---|---|---|---|
|
2023_10_NING.pdf
non accessibile
Dimensione
7.46 MB
Formato
Adobe PDF
|
7.46 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/10589/210469