An experimental study of the surface evolution during controlled plastic strain amplitude single-step and two-step loading tests reveals the important damage mechanisms for 316L stainless steel. In the first stage, the cyclic plastic strain is concentrated into persistent slip bands (PSBs) and a surface relief is formed consisting of extrusions and intrusions. The frequency of occurrence and the total density of PSBs has been assessed using systematic observations in a scanning electron microscope. The relative volume occupied by PSBs determines the fatigue damage in this stage. Two-step loading has only a small effect on the PSB damage evolution and nearly equal saturated values (corresponding to the applied plastic strain amplitude) were achieved in the single-step and the two-step loading investigations.

Fatigue crack growth data are usually derived from measurements on laboratory specimens, and a discrete set of data relating crack length a. to the number of loading cycles N. is obtained from experii ment. T~e data contain random and systematlc errors depending on the method of measurement. The problem consists in extracting the crack growth rates from these data. Three methods are currently in use (see, e.g., [1,2,3]): a) plotting a. vs N., drawing the best smooth curve through the data • I.Iand taking gradlents of the slope; b) calculating finite differences between successive values to obtain Aa./AN. ; or c) fi~tin~ an analytical curve through all or a part of the data.The first procedure is both tedious and subject to individual choice and therefore unsatisfactory.The second method usually has great scatter due to experimental errors, however, it is objective and can therefore be taken as a reference. The third method can smooth the data, however, it can also introduce further errors by forcing the experimental data to fit certain functional dependence. This was well illustrated by Davies and Feddersen [2] for polynomials of the nth order. Smith [3] obtained a better result using another type of function and dividing all data into three parts. The division introduces a subjective factor. Moreover, due to different test piece geometries and loading conditions no general equation describing fatigue crack propagation can be found.To smooth the data Munro [4] used the orthogonal polynomial method and fitted cubic expressions to equidistantly spaced crack length measurements By repeating this procedure several times, the smoothing was considerable. However, there is no limit to or criterion for the number of smoothing operations.All methods used so far try to smooth the raw data assuming all deviations from the smooth dependence are due to experimental errors. There are cases, however, in which the real variation of the crack rates can be suppresed by the smoothing procedure. The aim of the ideal method for crack growth rate evaluation is to obtain actual crack growth rates excluding the effect of random errors only.For this purpose the spline technique was used. The spline technique is well documented [5,6] and can be used both for interpolation and data smoothing.It consists in approximating the unknown dependence between successive points by a third order polynomial together with the requirement of a continuous first and second derivative. The degree of fit to the experimental points is determined by the errors of the measurement 6 i and by the parameter S given by the relation Int Journ of Fracture II (1975)