Evaluation of Fatigue Damage Classification Based on Probabilistic Weibull Analysis

Abstract

This paper focuses on the evaluation of fatigue damage classification based on statistical probability distribution using Weibull analysis. Feature extractions, i.e., kurtosis, wavelet-based energy, and fatigue damage, were calculated from the segments of fatigue strain signal. The feature extractions were then classified using artificial neural network (ANN) approach in order to find the class or level of fatigue damage. Subsequently, the statistical distribution fitting, i.e., Weibull, was applied to evaluate the fatigue damage classification. Based on the results, the accuracy of the ANN classification was found at 92% and a total of five classes or levels of fatigue damages were obtained. Based on Weibull distribution, when the maximum fatigue damage for the first class is inserted at approximately 8.42 × 10−5, around 69% of the coil spring will have a probability of failure. When the higher fatigue damage is inserted in the fifth class at 1.65 × 10−3, about 99% of the component will have probability to failure. The results show that fatigue damage classification is consistent with the Weibull distribution.