A small-sample multiaxial fatigue life prediction method for titanium alloys based on a parameter fusion model

Abstract A physics data fusion driven strategy for estimating the multiaxial fatigue life of titanium alloys is proposed. Using TC4 titanium alloy as the research subject, multiaxial fatigue experiments are performed and a physical framework is created by adopting von-Mises equivalent strain as the damage indicator. A Backpropagation (BP) neural network is then developed in which phase difference, normal strain amplitude, shear strain amplitude, normal stress amplitude, shear stress amplitude, and von-Mises equivalent strain serve as inputs, while the life predicted by the physical framework serves as the output. The connection weights between the hidden layer and the output layer of the neural network are subsequently refined through measured fatigue life data to form a parameter fusion model. By using the measured fatigue life as the reference, the predictive performance of the physical framework, the neural network, and the parameter fusion scheme is evaluated. The results reveal that the parameter fusion architecture achieves a substantial improvement in predictive accuracy compared with the single physical framework or the BP neural network, with every predicted life value remaining inside the twofold scatter zone of the experimental results. This approach integrates the foundational understanding of physical behavior with the adaptive learning abilities of data centered models, provides a promising route for reliable multiaxial fatigue life estimation of titanium alloys under limited sample conditions, and contributes theoretical value for fatigue resistant design of titanium alloy structural elements.