Page 32 - 《应用声学》2021年第3期

第 40 卷 第 3 期 Vol. 40, No. 3
2021 年 5 月 Journal of Applied Acoustics May, 2021

⋄ 研究报告 ⋄

应用深度学习识别法兰螺栓连接状态

张 洪 † 刘彬彬

(江南大学机械工程学院 无锡 214122)

摘要：针对常规诊断方法对螺栓的连接状态识别效果差、鲁棒性和抗噪性弱等问题，提出了基于深度学习理论
的螺栓检测新方法。首先以 4 种预紧力状态下的法兰螺栓结构产生的声发射信号为研究对象，借助于自适应
噪声的完整集成经验模态分解理论以及梅尔频率倒谱系数特征提取方式，实现了声发射信号的自适应消噪和
最优模态函数分量组的选取，提取到了可以较好分辨螺栓连接状态的梅尔频率倒谱系数特征值。通过训练模
型，较好地对 4 种连接状态下的螺栓进行了识别。结果表明，该模型在法兰螺栓的声发射信号的诊断中，准确
率高，具有较好的抗噪性和鲁棒性。
关键词：法兰螺栓；声发射；梅尔频率倒谱系数；深度学习；状态识别
中图法分类号: TP391.4; TB52+9 文献标识码: A 文章编号: 1000-310X(2021)03-0350-08
DOI: 10.11684/j.issn.1000-310X.2021.03.005

The recognition of flange bolt connection state based on deep learning

ZHANG Hong LIU Binbin

(School of Mechanical Engineering, Jiangnan University, Wuxi 214122, China)

Abstract: Aiming at the problems of conventional diagnostic methods in identifying the connection state of
bolts, weak robustness and noise resistance, a new method of bolt detection based on deep learning theory
is proposed. Firstly, the acoustic emission signals generated by the flange bolt structure under four kinds of
preloading force are taken as the research object, relying on the complete ensemble empirical mode decomposi-
tion with adaptive noise (CEEMDAN), and the feature extraction method of Mel-frequency cepstral coefficients
(MFCC).The adaptive noise reduction of the acoustic emission signal and the selection of the optimal basic
mode component group are realized. The eigenvalues of Mel-frequency cepstral coefficients that can better dis-
tinguish the bolt connection state are extracted. Through the training model, the bolts in the four connection
states are well identified. The results show that the model has high accuracy in acoustic emission diagnosis of
flange bolts, and has good noise resistance and robustness.
Keywords: Flange bolt; Acoustic emission; Mel-frequency cepstral coefficients; Deep learning; State recogni-
tion

2020-07-15 收稿; 2020-12-03 定稿
作者简介: 张洪 (1966– ), 男, 江苏无锡人, 博士, 副教授, 研究方向: 信号与信息处理, 机电检测与控制技术。
† 通信作者 E-mail: 1105399774@qq.com