Due to the high operation frequency and intensity, casting cranes are prone to structural fatigue damage. In this paper, a life prediction method is proposed based on digital twin. A prediction model based on LSTM network is established. By combining real-time monitoring data with finite element simulation data and integrating the Miner linear cumulative damage theory, the life prediction of key structures of casting cranes is achieved. The large amount of data generated by the structural digital twin model is used to make up for the shortage of measured data, and to train the LSTM model. Through the analysis of the training and validation loss curves and the R2 accuracy analysis of the validation data set, it is shown that this proposed model can effectively predict the life of the structure.