Estimation of the Error of Data Processing by the Method of Linear Regression Analysis in Fatigue Tests

Abstract

A method for determination of the confidence limits of the fatigue curve based on linear regression analysis is considered to provide forecasting of the durability of samples at a given probability of non-destruction. A specific example of using the method demonstrated that at fixed stresses the real laws of the durability distribution of the samples are described rather poorly by parametric statistics and the number of strain cycles prior to destruction thus forecasted does not correspond to the probability of their non-destruction which leads, in turn, to significant error in calculation of the confidence limits. We present an original algorithm of determining the confidence limits of fatigue curve using nonparametric statistics. The developed algorithm provides estimation of the systematic error of fatigue test data processing by the methods of linear regression analysis.