|M.Sc Student||Samuel Elioz|
|Subject||Retrieving Fatigue Loading History by Fractographic Image|
|Department||Department of Mechanical Engineering||Supervisor||Professor Emeritus Eli Altus|
|Full Thesis text - in Hebrew|
High significance is attributed to discovering the root causes and failure mechanisms of structures under fatigue loading. In most cases, even after considering design criteria and stress analyses, the loading history is only partially known; therefore unforeseen failures might occur prematurely. This history is calculated nowadays by detecting fatigue striations and measuring the distances between them manually. The challenge is to automatically correlate fracture surface elements with the crack growth rate (CGR).
In this study, a new methodology, based on detection of fatigue striations from fracture surface images, for calculating automatically the CGR is proposed.
Nine fatigue CGR experiments were conducted, in various stress levels, on standard ASTM specimens. Crack lengths as a function of cycle number were measured from the notched specimen until fracture.
Numerous methods for analyzing the topography of fracture surface images were examined. For example the scattering of the local absolute peaks was compared with a simulation of Voronoi tessellation. The stochastic field was found to be without specific order, meaning that the absolute peaks information has no correlation with the CGR.
The most effective method to detect and classify fatigue striations is the groups of directional peaks (crack growth direction) with two consecutive equal directional spaces. However the result did not match the reference experimental curve, since each group of striations had different density. CGR based on the average density was found to be higher than the experimental curve and the densest group is of small size and did not represent the fracture surface.
It is known that not every load cycle forms a striation on the fracture surface and therefore the CGR should be measured on the densest striations group in each image.
Based on the above insights a window based algorithm was developed. For each window size the algorithm locates the densest area in the image and derives a CGR measurement from it. The dependence of the CGR vs. window size was classified by two trends:
a. An increase until an asymptotic value for which an optimal CGR calculation was made.
b. An unordered scatter. These images were found uncorrelated to the CGR.
For every specimen two sets of images in X200 and X400 were taken. It was found that each set contains CGR data which characterizes different stress intensities: in high stress intensities the low magnification more appropriate and vice versa.
To summarize, the research conclusion is that it is possible to automatically calculate the fatigue load history, applied on a failed part, in a specific range of stress intensity factors. To widen the prediction range additional sets of images in various magnifications should be used.
The CGR prediction based on the above method was found to be more accurate than the manual count and requires no calibration experiment, as in other common methods. However the proposed method is not valid for low stress intensities in which striations are not visible.