The determination of fatigue life to design structures and mechanical components is extremely important. The S-N curve can be affected by different operational conditions, and some factors are more pronounced under ultrasonic fatigue test, depending on the material. The influence of the high frequency in self-heating phenomena and frequency effect are discussed. A relevant aspect in VHCF is the mechanism of crack initiation and propagation. The fish-eye and fine granular area (FGA) phenomena were encountered on the fracture surfaces. This thesis is divided in 3 topics: temperature evolution, microplasticity strain amplitude, and investigation of the FGA region. The materials under study are DIN 34CrNiMo6 and DIN 42CrMo4 steel. Ultrasonic fatigue test was conducted at different intermittent driving and loading ratios, accompanied by an infrared thermographic camera. These results obtained by the thermographic camera were used to developed an artificial intelligence model using machine learning to predict the temperature-number of cycle curves based on the fatigue life. The model was able to predict the temperature and the coefficient of determination values to be above 0.98. In order to predict the fatigue life, parameters were selected based on stress, the traditional S-N curve, slope temperature at the beginning of the test, (Rayleigh ratio), heat dissipation, Qcyc, and gradient temperature. A steady state temperature was reached approximately in 5E+04 cycles with both steels. It is noticeable that the number of cycles to failure increases as slope temperature and heat dissipation decreases. (Rayleigh ratio) provided better agreement with the experimental results followed by Qcyc. Moreover, ultrafine grains in the cross–section of the FGA between 500 - 700 nm within the fracture surface were detected by FIB and EBSD analysis. Local grain refinement was choose the best model to explain FGA formation. The non-metallic inclusions were ultimately responsible for all internal crack initiations of Al2O3.