【正文】 t crack in the bulk material. Obviously there is no way to detect this occurrence in practice. To be detectable the crack must propagate to the surface and produce a spall of sufficient magnitude to produce a marked effect on an operating parameter of the bearing: for example, noise, vibration, and/or temperature. Techniques exit for detecting failures in application systems. The ability of these systems to detect early signs of failure varies with the plexity of the test system, the type of bearing under evaluation, and other test conditions. Currently no single system exists that can consistently provide the failure discrimination necessary for all types of bearing life tests. It is then necessary to select a system that will repeatedly terminate machine operation with a consistent minimal degree of damage.The rate of failure propagation is therefore important. If the degree of damage at test termination is consistent among test elements, the only variation between the experimental and theoretical lives is the lag in failure detection. In standard throughhardened bearing steels the failure propagation rate is quite rapid under endurance test conditions, and this is not a major factor, considering the typical dispersion of endurance test data and the degree of confidence obtained from statistical analysis. This may not, however, be the case with other experimental materials or with surfacehardened steels or steels produced by experimental techniques. Care must be used when evaluating these latter results and particularly when paring the experimental lives with those obtained from standard steel lots.The ultimate means of ensuring that an endurance test series was adequately controlled is the conduct of a posttest analysis. This detailed examination of all the tested bearings uses highmagnification optical inspection, highermagnification scanning electron microscopy, metallurgical and dimensional examinations, and chemical evaluations as required. The characteristics of the failures are examined to establish their origins and the residual surface conditions are evaluated for indications of extraneous effects that may have influenced the bearing life. This technique allows the experimenter to ensure that the data are indeed valid. The “Damage Atlas” piled by Tallian et al. [] containing numerous black and white photographs of the various bearing failure modes can provide guidance for these types of determinations. This work was subsequently updated by Tallian [], now including color photographs as well. The posttest analysis is, by definition, after the fact. To provide control throughout the test series and to eliminate all questionable areas, the experimenter should conduct a preliminary study whenever a bearing is removed from the test machine. In this portion of the investigation each bearing is examined optically at magnifications up to 30 for indications of improper or outofcontrol test parameters. Examples of the types of indications that can be observed are given in Figs. .Figure illustrates the appearance of a typical fatigueoriginated spall on a ball bearing raceway. Figure contains a spalling failure on the raceway of a roller bearing that resulted from bearing misalignment, and Fig. contains a spalling failure on the outer ring of a ball bearing produced by fretting corrosion on the outer diameter. Figure illustrates a more subtle form of test alteration, `where the spalling failure originated from the presence of a debris dent on the surface. Figure gives an example of a totally different failure mode produced by the loss of internal bearing clearance due to thermal unbalance of the system.The last four failures are not valid fatig