Abstract: The boom system is a key load-bearing component of construction machinery, and its structural safety determines the safety of engineering construction. Structural fatigue damage prediction can ensure safe service throughout the life cycle of the boom system, but the stress data used for prediction is difficult to obtain reliably in the long term. Indirect signal backpropagation using deep learning is an effective method, but there are problems such as difficulty in grasping the equivalent relationship between input signals and fatigue damage, and low prediction accuracy. In response to the above issues, a fatigue damage prediction model is proposed for boom systems based on embedded physical information and attention mechanism BiLSTM neural network. Firstly, the overall framework of the model is introduced; Then, through a small number of actual working condition experiments, a data model of BiLSTM neural network based on a ttention mechanism is established, overcoming the problem of high-precision equivalent mapping between input signals and fatigue damage. Finally, by regressing and integrating the results of physical and data models, a novel physical guided loss function is innovatively proposed, significantly improving the model's fatigue damage prediction ability. The research results indicate that the prediction model has high prediction accuracy for fatigue damage of the boom system under different working conditions.

Key words: physical information, deep learning, boom system, fatigue damage, attention mechanism