Shallow underground double-hole detection technology based on Bi-LSTM

Journal Information

J. Hefei Univ. Tech. (Nat. Sci.), Jun.2024, Vol.47 No.6: 778-783

DOI: 10.3969/j.issn.1003-5060.2024.06.009

Authors

LIANG Jing $ ^{1,2} $, ZHANG Hong $ ^{1,2} $, YE Chen $ ^{3} $,

(1. School of Civil Engineering and Transportation, South China University of Technology, Guangzhou 510641, China; 2. State Key Laboratory of Subtropical Building Science, South China University of Technology, Guangzhou 510641, China; 3. Guangzhou Expressway Co., Ltd., Guangzhou 510289, China)

Abstract and Keywords

Abstract: This paper explores a shallow underground hole detection technology based on deep learning to deal with the serious threat of underground holes to the safety of pile foundation construction. Based on the principle of shallow seismic reflection wave method, and taking the pile hammer shock in the foundation construction process as the excitation, a new intelligent hole detection method is established by placing a small number of acceleration sensors on the surface of the detection area to collect the hole reflection signal, and take the reflected signal as the input of deep learning to output the hole information. The results show that the bidirectional long short-term memory neural network (Bi-LSTM) prediction model has a high recognition accuracy for the working conditions of underground double holes. When the tolerance error is 2 m, the accuracy of hole location and diameter prediction can reach 95.3%. This study verifies the feasibility of multi-hole detection technology based on deep learning, and is expected to provide technical support for the assessment of soil geological conditions in the pre-construction period.

Keywords: underground hole detection; pile hammer shock; deep learning; bidirectional long short-term memory neural network(Bi-LSTM); finite element simulation

作者信息

梁靖 $ ^{1,2} $，张红 $ ^{1,2} $，叶晨 $ ^{3} $，周立成 $ ^{1,2} $，刘泽佳 $ ^{1,2} $，汤立群 $ ^{1,2} $

(1. 华南理工大学土木与交通学院, 广东广州 510641; 2. 华南理工大学亚热带建筑科学国家重点实验室, 广东广州 510641; 3. 广州市高速公路有限公司, 广东广州 510289)

摘要和关键词

摘要: 文章探究一种基于深度学习的浅层地下孔洞探测技术,以应对地下孔洞给桩基施工安全所造成的严重威胁。基于浅层地震反射波法的原理,采用基础施工过程中的桩锤激震作为激励源,通过在探测区域地表上布置少量加速度传感器采集孔洞反射信号,并将反射信号作为深度学习的输入,以输出孔洞信息,建立一种新型的智能孔洞探测方法。结果表明,双向长短期记忆神经网络(bidirectional long short-term memory neural network, Bi-LSTM)的预测模型对于地下双孔洞的工况具有较高的识别准确率,在容许误差为2 m的情况下,孔洞位置和直径的预测准确率可达95.3%。该研究验证了基于深度学习的多孔洞探测技术的可行性,有望为施工前期土层地质状况的评估提供技术保障。

关键词: 地下孔洞探测；桩锤激震；深度学习；双向长短期记忆神经网络（Bi-LSTM）；有限元仿真

Funding

国家自然科学基金资助项目（11972162）；广州市科技计划资助项目（201903010046）

合肥工业大学学报(自然科学版)

Navigation