A multi-task machine learning approach for data efficient prediction of blast loading
Qilin Li, Ling Li, Yanda Shao, Ruhua Wang, Hong Hao
Abstract
Accurate and efficient prediction of blast loading is critical for structural protection and safety management across various industries. With advancement in AI algorithms, various Machine Learning models have been developed recently to predict blast loadings, which are, however, based primarily on Single Task Learning methods despite multiple parameters need be predicted to fully define a blast load . This study introduces a novel Multi-Task Learning (MTL) approach designed to simultaneously predict multiple critical parameters of blast loading, including arrival time , rise time, duration, and peak pressure. By leveraging MTL, the model can also exploit the inherent relationships between these parameters, which enhances the predictive performance. The MTL approach is compared against conventional Single Task Learning (STL) methods using CFD data from 1489 explosion cases. The results demonstrates that MTL consistently outperforms STL in terms of prediction accuracy, computational efficiency, and data utilization. MTL proves especially advantageous in scenarios with limited data, where its ability to share information between related tasks leads to superior performance. This collaborative learning among interconnected tasks is crucial in engineering applications , where acquiring large datasets is often challenging. Additionally, the MTL framework allows for the integration of domain knowledge, enabling engineers to group tasks strategically and further enhance the model's accuracy. The proposed MTL method thus offers a more robust, accurate, and efficient solution for predicting blast loading, ultimately contributing to better structural protection and optimized safety management strategies.