基于互信息和MiniRocket网络的CFST脱空识别  

Identification of CFST voids based on mutual information and MiniRocket network

作　　者：覃悦 谢开仲 郭晓 王红伟[4] 王秋阳 彭佳旺 QIN Yue;XIE Kaizhong;GUO Xiao;WANG Hongwei;WANG Qiuyang;PENG Jiawang(School of Civil Engineering and Architecture,Guangxi University,Nanning 530004,China;Key Laboratory of Disaster Prevention and Engineering Safety of Ministry of Education,Guangxi University,Nanning 530004,China;Guangxi Key Laboratory of Disaster Prevention and Engineering Safety,Guangxi University,Nanning 530004,China;Guangxi Xinfazhan Communication Group Co.,Ltd.,Nanning 530029,China)

机构地区：[1]广西大学土木建筑工程学院,南宁530004 [2]广西大学工程防灾与结构安全教育部重点实验室,南宁530004 [3]广西大学广西防灾减灾与工程安全重点实验室,南宁530004 [4]广西新发展交通集团有限公司,南宁530029

出　　处：《振动与冲击》2024年第8期202-212,共11页Journal of Vibration and Shock

基　　金：国家重点研发项目(2019YFC1511103);国家自然科学基金(51868007,51738004);广西研究生教育创新计划项目(YCBZ2023022)。

摘　　要：为提高钢管混凝土(concrete filled steel tube,CFST)脱空检测的效率和精度,本文提出了一种基于快速傅里叶变换(fast fourier transform,FFT)、互信息(mutual information,MI)和MiniRocket神经网络的智能识别方法。首先,采用FFT将待测CFST敲击声波时域信号转换为频域信号;其次,采用MI建立频域信号与脱空状态的相关性,提取相关性最大的前30个特征建立数据集,避免了复杂的数学运算和冗余信息;建立MiniRocket深度学习网络,通过使用更少的参数量和更小的特征尺寸,提高分类的速度和精度。最后,考察了模型的噪音鲁棒性,并与其他算法、特征提取方法和识别方法进行对比。结果表明,在不同脱空深度和脱空宽度下,所提的方法在100次重复试验中获得了100%的平均预测精度。在高信噪比下,该方法受影响较小。此外,与其他算法、特征提取方法和识别方法相比,本方法具有更好的预测性能。因此,所提出的方法在未来实际CFST结构的智能脱空识别中具有较大的应用潜力。In order to improve the efficiency and accuracy of concrete filled steel tube(CFST)void detection,an intelligent recognition method based on fast Fourier transform(FFT),mutual information(MI)and MiniRocket neural network is proposed in this paper.First,the time domain signal of the CFST percussion wave to be measured is converted to the frequency domain signal using FFT.Secondly,MI is used to establish the correlation between the frequency domain signal and the void state,and the top 30 features with the largest correlation are extracted to establish the data set,which avoids complex mathematical operations and redundant information.A MiniRocket deep learning network is established,and by using fewer parameters and smaller feature sizes improving the speed and accuracy of classification.Finally,the noise robustness of the model is investigated and compared with other algorithms,feature extraction methods and recognition methods.The results show that the proposed method achieves 100% average prediction accuracy in 100 repetitions of the experiment for different void depths and void widths.At high SNR,this method is less affected.In addition,compared with other algorithms,feature extraction methods and recognition methods,this method has better prediction performance.Therefore,the proposed method has great application potential in the actual intelligent void identification of CFST in the future.

关 键 词：钢管混凝土(CFST) 脱空 敲击声波 互信息(MI) 深度学习 

分 类 号：TU398[建筑科学—结构工程]