Article: Research on multiphase flow simulation of cave grouting diffusion coupled with phase field method and convolutional neural network Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.94 pp.19 - 41 Abstract: To enable efficient real-time prediction of grouting behaviour in karst cavities, this study proposes a physics-informed framework integrating phase-field multiphase flow simulation with a convolutional neural network (CNN). The phase-field model captures grout-groundwater interfacial evolution and provides physical constraints for data-driven prediction. A dataset of 1,000 samples covering 100 operational scenarios was generated using Latin hypercube sampling, and an encoder-decoder CNN was trained after hyperparameter optimisation. The model achieved a root mean square error of 0.028 and a structural similarity index of 0.942 on the test set. Physical model tests further showed that prediction errors were below 0.2 cm for diffusion radius and 0.5 kPa for pressure field. The proposed framework markedly improves prediction efficiency and accuracy while maintaining physical consistency, providing an effective tool for grouting design and real-time decision-making in karst engineering. Inderscience Publishers - linking academia, business and industry through research

Title: Research on multiphase flow simulation of cave grouting diffusion coupled with phase field method and convolutional neural network

Authors: Yong Bi; Jintao Yao; Zepeng Yao

Addresses: China Railway Tenth Bureau Group Urban Rail Transit Engineering Co., Ltd., Guangzhou 511493, China ' China Railway Tenth Bureau Group Urban Rail Transit Engineering Co., Ltd., Guangzhou 511493, China ' China Railway Tenth Bureau Group Urban Rail Transit Engineering Co., Ltd., Guangzhou 511493, China

Abstract: To enable efficient real-time prediction of grouting behaviour in karst cavities, this study proposes a physics-informed framework integrating phase-field multiphase flow simulation with a convolutional neural network (CNN). The phase-field model captures grout-groundwater interfacial evolution and provides physical constraints for data-driven prediction. A dataset of 1,000 samples covering 100 operational scenarios was generated using Latin hypercube sampling, and an encoder-decoder CNN was trained after hyperparameter optimisation. The model achieved a root mean square error of 0.028 and a structural similarity index of 0.942 on the test set. Physical model tests further showed that prediction errors were below 0.2 cm for diffusion radius and 0.5 kPa for pressure field. The proposed framework markedly improves prediction efficiency and accuracy while maintaining physical consistency, providing an effective tool for grouting design and real-time decision-making in karst engineering.

Keywords: phase field method; convolutional neural network; cave grouting; multiphase flow simulation; coupling model; diffusion prediction.

DOI: 10.1504/IJICT.2026.156126

International Journal of Information and Communication Technology, 2026 Vol.27 No.94, pp.19 - 41

Received: 04 Feb 2026
Accepted: 06 Jun 2026
Published online: 04 Sep 2026 *

Title: Research on multiphase flow simulation of cave grouting diffusion coupled with phase field method and convolutional neural network

Keep up-to-date