Muck identification and anomaly detection in earth pressure balance shield based on multi-task deep learning

1. Home
2. Publications
3. Online Library
4. Muck identification and anomaly detection in earth pressure balance shield based on multi-task deep learning

Publication Title
Muck identification and anomaly detection in earth pressure balance shield based on multi-task deep learning

Abstract
A lack of understanding of geological conditions and improper muck disposal often leads to serious accidents, such as ground settlement or machine damage, during shield tunneling. The image data of muck on the conveyor belt reflects the general soil type of the excavated strata and the discharge condition, providing abundant and valuable information for EPB shield construction decisions. This paper aims to use a multi-task deep learning method to establish an image recognition model that can track the type and location of the muck and estimate whether the muck is discharged smoothly at the same time. The first task is an object detection task and the second is an image classification task. A convolutional neural network for a multi-task learning setting has been designed, and then used to build a multi-task muck recognition model using a specially constructed labeled muck dataset. The mAP for the muck identification task reaches 92.72%, and the accuracy for the anomaly detection task reaches 96.26%. The model is applied to Metro Line 14 in Shanghai, and the results show that the model performs well on both target tasks, meeting the real-time requirements at 45 frames per second.
Authors
L. Fu; D. M. Zhang; H. W. Huang; A. G. Cohn; Z. Q. Liu

Published In

18th European Conference on Soil Mechanics and Geotechnical Engineering (ECSMGE2024)

Theme

D - Current and new construction methods

Download PDF