UNEQUAL INTERVAL STRESS SPECTRUM COMPILATION METHOD BASED ON CLUSTER ANALYSIS AND SUPPORT VECTOR MACHINE

A method of unequal interval stress spectrum compilation based on fatigue damage was proposed to ensure the stress spectrum compilation from the measured stress-time history to reflect the actual effect characteristics of the load. According to the damage equivalence principle, a mathematical model of the relationship between the stress spectrum classification numbers and fatigue damage were established by adding unequal interval adaptive ratio coefficients. Considering the attribute characteristics of each stress cycle, cluster analysis was used to classify all stress cycles, while the classification accuracy and prediction map under different classification numbers were obtained based on the support vector machine multi-classification. Considering the relative error of damage obtained from the mathematical model and the classification accuracy of support vector machine, the corresponding unequal interval stress spectrum were compiled by means of the optimal number of stress spectrum, which was compared with the equal interval spectrum and unequal interval spectrum of 8 levels. The results show that the stress spectrum of classification based on the damage equivalent principle, cluster analysis and support vector machine considers the overall characteristics of all stress cycle and the local information and properties of the stress, and the results of the fatigue cumulative damage theory are significantly lower than the actual damage, so that the measured data of stress characteristic and the effects of fatigue are accurately reflected. This method can be used for other measured load-time history to unequal interval classification and load spectrum compilation.