AI-driven hybrid deep learning for real-time excavation risk assessment in deep foundation pit engineering
Year: 2025
Type: article
Abstract: Excavation of the deep foundation pit is a challenging organisation due to unstable soil, groundwater seepage, and structural failures. Traditional risk assessments are useful but computationally expensive and inflexible to real-time conditions. This study then proposes a hybrid deep learning model that integrates CNN, LSTM and transformer architectures for improving excavation risk prediction. FEM data are used to capture spatial features, LSTM models sequential deformations, and the transformer incorporating multi-source geotechnical data. The model was validated on a Shanghai excavation project with an RMSE of 2.90 mm, which outperforms FEM, CNN-LSTM and transformer only. In addition, it achieved 94.5% F1 score for failure detection and had reduced inference time to 1.4 seconds. The accuracy and speed of these results provides confidence in the model to be deployed in real-time for safety monitoring and AI based geotechnical risk management.
Author: Xuhui Luo
Language: English
Subfield: Safety, Risk, Reliability and Quality
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4411308918