Mid-span displacement and damage degree predictions of RC beams under blast loading using machine learning-based models

Civil engineering has extensively employed RC beams in various applications. However, these beams are susceptible to damage and failure under blast loading. Assessing damage of RC beams from blasts requires real measurement data and time-intensive numerical simulation, making rapid evaluation a significant challenge. In this study, machine learning (ML) models were developed to swiftly evaluate the damage degree and predict the mid-span displacement of RC beams, comprising three models for each problem. To compensate for the lack of real experiment data, a total of 1041 data points are generated through numerical simulation using ABAQUS to create the training dataset. Additionally, for displacement prediction models, 39 collected measurement data was integrated. Initially, the ML models were established to predict the peak mid-span displacement of RC beams under explosion, with inputs including twelve RC beam structural parameters and two blast load parameters, and the output being the maximum min-span displacement of the beam. Subsequently, the ML models were developed using the same fourteen parameters as inputs, with the output being the beam's damage degree and failure mode. These models achieved adequate performance, demonstrated by evaluation metrics such as R2, RMSE, MAE, MAPE, with values of 0.95, 5.468 mm, 3.16 mm, 17.7 %, respectively, for regression problem. For the classification problem, metrics such as accuracy, precision, recall, and f1-score were employed, with an overall value of 0.837 for the best model. The results show that the developed ML models can precisely and rapidly assess the damage degree, failure mode and predict the mid-span displacement of the RC beam under blast load, which provides a good method for the anti-explosive protection design of RC beams. The graphical user interface tool was also established for ease of use.