Metal fatigue life prediction method based on multi-mode Transformer

Abstract

The invention relates to the technical field of material science and artificial intelligence, in particular to a metal fatigue life prediction method based on multi-mode Transformer, which comprises the following steps of: firstly, loading material attribute data of a metal component and time sequence data of a loading path, and preprocessing; then the model is input, load time sequence features and material features are extracted through a time sequence encoder and a material feature encoder respectively, then the load time sequence features and the material features are input into a cross attention fusion module to generate fusion features, and finally the fusion features are input into a prediction output layer to generate a fatigue life prediction result. The bottleneck that a traditional physical model is limited in precision, low in calculation efficiency and poor in working condition adaptability is broken through. Through a cross-modal data fusion mechanism, non-stationary load time series data and static material attribute characteristics are deeply associated in a unified framework, and high-precision and high-efficiency prediction of the fatigue life of the metal component in a complex service environment is realized.

Images