Physically constrained neural network for predicting blast-induced ground vibration
Ground vibration from blasting, measured by peak particle velocity (PPV), threatens urban areas. To address this, physically constrained neural network (PCNN) is proposed for accurate PPV prediction, integrating physical relationships into data-driven learning processes. The core innovation lies in the design of the loss function, which combines data-driven loss with physical loss derived from dimensional analysis. Tested on 152 datasets from Tonglvshan open-pit mine, the PCNN achieves R 2 = 0.925, RMSE = 0.407, and MAE = 0.309, outperforming traditional models. The results highlight the value of embedding physical constraints for accurate PPV prediction, offering a high-precision method that merges data-driven modeling with physical knowledge.
Authors
- Jingwen Zhu
- Shuangshuang Ge
- Shuo Yuan (ORCID: https://orcid.org/0000-0002-5299-4380)
- Wei Gao
- Peiyuan Li
Institutions
- Hohai University (CN)
Publication Details
- Journal
- Journal of Vibration and Control
- Published
- 2026-11-19
- DOI
- https://doi.org/10.1177/10775463261506937
- Primary Topic
- Blasting Impact and Analysis
- Type
- article
- Field-Weighted Citation Impact
- 0.00