A shallow artificial neural network for mapping bond strength of soil nails

Abstract

Soil nails are extensively used to stabilize existing slopes and new cuts in Hong Kong. The current method adopted for estimation of nail bond strength is the effective stress method (ESM), which has been shown to be excessively conservative and highly dispersive in prediction. A more robust and accurate model for estimation of nail bond strength is thus highly desired. The present study provides an artificial neural network (ANN) model for mapping of soil nail bond strength. The ANN model is trained, validated, and tested using 522 nail bond strength data contained in Hong Kong soil nail pullout database and is publically accessible in the literature. The accuracy of the developed ANN model is examined and compared with the default ESM. The results showed that ANN model is accurate on average and the predicted dispersion is low. The practical value of ANN model is highlighted by a reliability-based design example of soil nails against the pullout limit state. This work demonstrates the opportunity of applying machine learning approaches to design of soil nail walls.

Keywords:

• Soil nails
• bond strength
• effective stress method
• artificial neural network
• model uncertainty
• reliability-based design

Acknowledgement

The authors are grateful for financial support from Young Scholar Starting Research Funds provided by Foshan University and Sun Yat-Sen University.

Disclosure statement

No potential conflict of interest was reported by the authors.