Probabilistic Prediction of Fatigue Life of Cracked Parts: Linear Elastic Fracture Mechanics based Approach

Abstract

In the context of regular inspections of structures, the presence of cracks can sometimes be revealed. It is therefore interesting to know whether the structure can still be used or if a degraded mode of operation should be considered. It is of concern to assess the scatter of the remaining life of such cracked parts due to the uncertainties on the parameters of the prediction model. Thus for the purpose of the present study, a special attention has been given to the quantification of the uncertainty of each collected data (e.g., material properties, crack measurements) to be integrated in the crack propagation model. Three families of uncertainties have been studied, namely, material properties, geometrical and loading uncertainties. A linear elastic fracture mechanics (LEFM) based approach has been used to predict crack propagation in a widely used pressure vessel steel. The model uses classical Paris’ law where all the controlling parameters have been replaced by statistical distributions obtained from experiment, namely, crack growth tests, hardness tests, tensile tests and Charpy tests. Crack growth tests have also been carried out on thick notched specimen submitted to uniaxial cyclic load to obtain two dimensional cracks that can be representative of those to be found in industrial structures. During these tests, crack lengths have been measured simultaneously by time of flight diffraction (TOFD) ultrasonic method, digital image correlation (DIC) and some markings have been performed to estimate the crack length and crack front shape after the specimen failure. The comparison of data obtained from different observation techniques allowed for the quantification of the crack measurement uncertainty of the industrial TOFD technique. A sensitivity analysis has been carried out on the parameters of the model to evaluate and classify the influence of all sources of uncertainty on the residual life prediction. Last, results from crack growth tests on notched specimen have been used to assess the accuracy of the model prediction.