Nous avons le plaisir de vous convier à la défense de thèse de Monsieur Quang A. MAI, qui se déroulera le vendredi 7 décembre à 10h (salle s.39, B37).
Cette recherche doctorale s'intitule "Updating Failure Probability of a Welded Joint Considering Monitoring and Inspection For Offshore Wind Turbine Support Structures" et a été réalisée sous la direction du professeur Philippe RIGO.
Reliability assessment of existing offshore wind turbine (OWT) support structures taking advantage of historical data from inspection and monitoring is crucial for the optimization of maintenance and life extension. In this thesis, the crack inspection data and the monitoring data are separately considered for updating failure probability of welded joints.
Concerning crack inspection data, this thesis aims at finding advantages of using the fatigue assessment diagram (FAD) in the failure criteria for failure probability estimation and updating. The crack inspection results (no detection or detected) and possible immediate interventions (repair normally or perfectly) are considered. Failure probabilities are calculated using the FAD and then compared with those obtained from the usual critical crack size criteria. The simulation-based approach is used to calculate and update the failure probability. Crack depth and length are simulated simultaneously. The crack propagations are calculated using a bi- linear Paris’ law with stress-range value varying over time. Uncertainties come from the crack growth parameters, initial crack sizes, fracture toughness, yield and ultimate strengths, FAD formula, stress intensity factor, stress-range values, and the detectable crack size.
By combining the fracture toughness with the crack size in the failure criteria, the results show that the estimated failure probability of the welded joint is significantly increased in comparison to the case where only critical crack size is considered in the Limit State Function (LSF). In comparison with the failure criteria which includes both the critical crack size and fracture toughness, the FAD approach gives similar reliability results when the applied peak tensile stress is small. However, when the applied peak tensile stress is high (the ratio between applied peak tensile stress and yield strength is more than 65%), the FAD approach predicts higher failure probability values. The uncertainty in FAD does not significantly affect the predicted failure probability of the joint as compared to the uncertainties in the ultimate and yield strength. This is because the latter affects directly the cut-off location of the FAD curve where the plastic failure is defined. The FAD approach can be used to update failure probability considering crack inspections and intervention actions. The calculation shows that reliability of the welded joint increases when no crack is detected, or when a crack is detected and repaired. In comparison with a perfect repair, a normal repair assumption significantly reduces the reliability of the joint.
With regard to the monitoring data, the research questions how to effectively incorporate the measured strain and the oceanographic data (wind, wave) for updating failure probability of a welded joint in fatigue failure mode. To answer the question, the monitoring data is used to update the characteristics of a random variable in the LSF. Consequently, the updated random variable is then used in reliability analysis to obtain the updated failure probability.
The LSF is based on the Miner’s rule and solved using the first order reliability method (FORM). The random variable used to update failure probability is the joint distribution of wind speed, wave height and wave period. The monitoring data consists of strain, 10-minute mean wind speed, significant wave height, and mean wave period. Fatigue damage is summed up from all load combinations, i.e. from all the discretized components of the joint distribution of wind and wave. The measured strain data is used to calculate fatigue damage in each load combination. The probability of each load combination is calculated using its joint distribution, which in turn can be updated using monitoring data. The 10-minute mean wind speed is assumed to follow a Weibull distribution and can be updated using Bayesian approach. Assuming that the scale parameter is a normally distributed random variable with unknown mean and standard deviation, the predictive distribution of this random variable becomes a student’s t-distribution.
The proposed methodology has been applied to a monopile support structure of a wind farm in Belgium. The measured strain is used to find the potential hot-spot location. Stresses are derived at the hot-spot location for fatigue analyses. It is assumed that stress-ranges in each load combination follow a Weibull distribution. The stress-range distribution parameters are found by performing least squares fitting method on the fatigue damage.
The results show that the Weibull distribution is generally not very good for fitting stress-ranges in each wind-speed bin for the considered data. However, the integrated fatigue damage for the considered load combination is quite accurate since it is the objective of the fitting procedure. The main influence on the remaining fatigue life is the magnitude of stress-ranges at the hot-spot. So the stress concentration factor, the interpolating factor (for example to obtain stresses at under water locations), the correction factor for corrosion effects (if any), and their uncertainties should be estimated with great care. The measured 10-minute wind speed data has a significant effect in adjusting the predicted probability of failure and eventually the remaining fatigue life. The duration of strain measurement should be long enough to be combined with oceanographic data. Longer strain measurement duration, greater number of parameters in the oceanographic data can be considered for the failure probability updating.