基于EMD的神经网络空耦超声储油罐液位检测  

Neural network air coupled ultrasonic oil tank liquid level detection based on EMD

作　　者：陈志恒 罗文斌 常俊杰[1,3,4] 钟海鹰 吴中权 郑阳 CHEN Zhiheng;LUO Wenbin;CHANG Junjie;ZHONG Haiying;WU Zhongquan;ZHENG Yang(Key Laboratory of Nondestructive Testing Technology,Ministry of Education,Nanchang Hangkong University,Nanchang 330063,China;China Special Equipment Inspection and Research Institute,Beijing 100029,China;Nippon Probe Co.,Ltd.,Yokohama 232-0033,Japan;Yangtze Delta Region Institute of Tsinghua University,Jiaxing 314000,China)

机构地区：[1]南昌航空大学无损检测技术教育部重点实验室,江西南昌330063 [2]中国特种设备检测研究院,北京100029 [3]日本探头株式会社,日本横滨232-0033 [4]浙江清华长三角研究院,浙江嘉兴314000

出　　处：《中国测试》2021年第1期9-14,共6页China Measurement & Test

基　　金：国家自然科学基金(11464030);南昌航空大学研究生创新专项资金项目(YC2019050)。

摘　　要：由于仅从信号时域幅值的大小信息虽然能够判断储油罐中不同介质的液位,但是获得的特征信息非常有限,为获得更多储油罐中不同介质信号的特征信息来提高液面识别率,针对储油罐罐壁厚度为5 mm的钢制储油罐为对象,采用空气耦合超声兰姆波同侧相向检测法,并使用A0模态对储油罐进行检测。利用经验模态分解(EMD)对采集储油罐中的不同介质信号进行EMD处理,求得各阶本征模函数(IMF)。通过分析各阶IMF分量的时域、频域信号与原始信号的相关性,并且以各IMF分量的时域、频域信号为特征值输入到BP神经网络进行决策。实验结果表明,使用该方法能准确地对储油罐不同介质液位在10 mm范围进行识别与分类,识别率可达99%,其检测范围满足实际检测需求。Since only the signal amplitude information in the time domain can determine the liquid level of different media in the storage tank, the obtained characteristic information is very limited. In order to obtain more characteristic information of signals of different media in the storage tank to improve the liquid level recognition rate. As for the steel oil tanks with wall thickness of 5 mm, the ipsilateral detection method of aircoupled ultrasonic Lamb wave was adopted, and the A0 mode was used to detect the oil tanks. Empirical mode decomposition(EMD) was used to process signals from different media in the collected storage tank, and IMF was obtained. By analyzing the IMF component of each order time domain and frequency domain signal and original signal correlation, and with each IMF component characteristics of time domain and frequency domain signal input to the BP neural network to make decisions, the experimental results show that using this method can accurate liquid level in storage tanks of different media in the range of 10 mm for identification and classification, recognition rate can reach 99%, the detection range in the results meet the practical needs.

关 键 词：储油罐 液位检测 兰姆波 EMD 神经网络 

分 类 号：TB302.5[一般工业技术—材料科学与工程]