A study of creep-fatigue life prediction using an artificial neural network

In this study, using AISI 316 stainless steel, creep-fatigue tests were carried out under various test conditions (different total strain ranges and hold times) to verify the applicability of the artificial neural network method to creep-fatigue life prediction. Life prediction was also made by the modified Coffin-Manson method and the modified Ostegren method using 21 data points out of a total 27 experimental data points. The six verification data points were carefully chosen for the purpose of evaluating the predictability of each method. The predicted lives were compared with the experimental results and the following conclusions were obtained within the scope of this study. While the creep-fatigue life prediction by the modified Coffin-Manson method and the modified Ostegren method had average errors of 35.8% and 47.7% respectively, the artificial neural network method had only 15.6%. As a result, the artificial neural network method with the adaptive learning rate was found to be far more accurate and effective than any of the others. The validity of the artificial, neural network method for life prediction checked with the six verification data points also proved to be very satisfactory.