Prediction of concrete wear resistance under wind erosion based on the LSTM deep learning model

Concrete structures are highly susceptible to severe erosion and spalling caused by wind, which posing a considerable threat to their service life. Gaining insight into the temporal variation of mass loss in concrete under wind-sand erosion is vital for ensuring timely maintenance and safe operation. To achieve this, wind erosion experiments were conducted, and a dataset capturing the temporal mass loss caused by wind erosion was established. Building upon the LSTM model as a baseline, an enhanced deep learning model called Attention-LSTM was introduced by integrating an attention mechanism to improve the predictive performance in estimating concrete mass loss. The results revealed that the Attention-LSTM model achieved an R2 of 0.99, RMSE of 0.012, and MAPE of 0.49 %, indicating a high degree of accuracy in predicting erosion-induced concrete mass loss. Comparative evaluations were performed against other models, including LSTM, GRU, RF, XGBoost, and 1DCNN, on the same dataset. The findings confirm the superior performance of the Attention-LSTM model and underscore the effectiveness of incorporating the attention mechanism into the LSTM architecture. Moreover, the applicability of natural language processing (NLP) techniques in predicting concrete wear resistance under wind erosion was also verified, illustrating their potential for such predictive tasks.