基于深度置信网络的工字钢梁损伤识别方法  

Damage Identification Method of I-beam Based on Deep Belief Network

作　　者：程海根[1] 胡钧剑 胡晨 CHENG Haigen;HU Junjian;HU Chen(East China Jiaotong University,Nanchang,Jiangxi 330013,China;Jiangxi Expressway Investment Group Co.Ltd,Nanchang,Jiangxi 330025,China)

机构地区：[1]华东交通大学,南昌330013 [2]江西高速公路投资集团有限公司,南昌330025

出　　处：《铁道工程学报》2021年第11期67-73,共7页Journal of Railway Engineering Society

基　　金：国家自然科学基金项目(51368018,51968024)。

摘　　要：研究目的:工字钢梁作为一种重要的受力结构,在建筑结构、桥梁支架、轨道交通、仓储厂房等领域应用广泛。由于其在服役期间极易受到外部荷载、自然环境、材料性质等影响而发生严重的损伤,危害结构的耐久性和安全性,因此有必要采取相应的措施及时发现工字钢梁的损伤状况。基于深度置信网络的工字钢梁损伤识别方法,提取结构发生振动时的竖向加速度响应值作为损伤指标,通过深度置信网络对损伤指标的特征分析,识别结构的损伤位置和损伤程度。研究结论:(1)该方法在损伤位置和损伤程度方面的识别准确率均在90%以上,且随着噪声程度不断提高,识别准确率能够保持稳定的水平,抗噪性能好;(2)SVM支持向量机在损伤位置的识别准确率低于20%,BP神经网络在损伤程度的识别准确率低于70%,随着噪声程度不断提高,两种传统的方法识别准确率不能保持稳定的水平,识别能力差;(3)本文所提出的基于深度置信网络的损伤识别方法具有一定的可行性,在实际工程应用中可研究性较强,具有一定的研究前景。Research purposes: As an important force-bearing structure, I-beam is widely used in building structures, bridge supports, rail transit, storage plants and other fields. Because it is extremely vulnerable to external loads, natural environment, material properties during service, serious damage occurs, which endangers the durability and safety of the structure. Therefore, it is necessary to take corresponding measures to discover the damage of the I-beam in time. The I-beam damage identification method based on the deep belief network extracts the vertical acceleration response value when the structure vibrates as the damage index, and uses the deep belief network to analyze the characteristics of the damage index to identify the damage location and damage degree of the structure.Research conclusions:(1) The recognition accuracy of the method in terms of damage location and damage degree is more than 90%, and with the continuous increase of the noise level, the recognition accuracy can maintain a stable level, and the anti-noise performance is good.(2) The recognition accuracy rate of SVM support vector machine at damage location is less than 20%, and the recognition accuracy rate of BP neural network at damage level is less than 70%. With the continuous increase of noise level, the recognition accuracy rate of the two traditional methods cannot maintain a stable level, and the recognition ability is poor.(3) The damage identification method based on the deep belief network proposed in this paper is feasible, and it can be researched in practical engineering applications and has a certain research prospect.

关 键 词：工字钢梁 损伤识别 深度置信网络 竖向加速度响应值 特征分析 

分 类 号：TU317[建筑科学—结构工程]