A machine learning method for HTLCF life prediction of titanium aluminum alloys with consideration of manufacturing processes

Abstract

Most conventional methods only consider the effects of materials and loading conditions when predicting the high-temperature low-cycle fatigue life of titanium aluminum alloys. However, the effects of manufacturing processes are not considered. To address this limitation, this paper proposes a machine learning method. First, the loading conditions are preliminarily filtered by physics knowledge. Second, the manufacturing processes and loading conditions are analyzed by using the average value of multiple mean impact values. Then, the variables that have little contribution to the output are filtered out, and the remaining variables are used as inputs. Finally, an optimized genetic algorithm-back propagation artificial neural network is established to predict the low-cycle fatigue life of the materials. The optimization methods are Levy flight and an adaptive adjustment mechanism of probability. Min-Max normalization is used to normalize and denormalize the data in the proposed method. A set of experimental data for Ti-685 is used to validate the proposed method. Moreover, the results show that the proposed method has better prediction accuracy than other methods.