Assessment of bending fatigue limit for crankshaft sections with inclusion of residual stresses

The fatigue limit of rolled, ductile cast iron crankshaft sections under bending was investigated experimentally and analytically. The relationship of different failure criteria - specifically, surface crack initiation, resonant shifts, and two-piece failures - on the fatigue limit of the crankshafts was explored from testing. A comparison showed that using the surface crack failure criterion decreased the apparent fatigue limit of the crankshaft significantly. This result supported the theory of crack arrest in the fillet area of crankshafts and raises a significant challenge to the use of the surface crack failure criterion due to its inconsistent correlation to two-piece failure. Also an engineering practice for residual stress simulation and fatigue limit estimation due to a fillet rolling process on a crankshaft was explored. The complete analytical process was validated by the resonant bending fatigue test results. The normal stress distribution from the fillet surface-to-depth of a crankshaft section under bending was first obtained by a three-dimensional elastic finite element analysis. The residual stress distribution near the crankshaft fillet induced by a fillet rolling process was then determined by a three-dimensional elastic-plastic finite element analysis. A resultant normal stress distribution along any possible crack growth plane originating from the fillet surface and extending into the component was defined by superimposing the normal stress components due to the bending moment and the rolling load. Based on these calculated normal stresses, two fatigue models (such as the SWT equivalent stress and the effective stress intensity factor range) used to estimate the fatigue limits on surface cracks and in-depth cracks were examined by the experimental data. The results showed that the SWT equivalent stress was not a good fatigue damage parameter for surface cracks and the effective stress intensity factor range derived on the technique of crack modeling did correlate reasonably to the experimental data.