摘要
搭建了球阀内漏实验平台,利用声发射技术采集球阀内漏数据,用小波包阈值去噪方法对内漏信号进行降噪处理,及皮尔逊相关系数法筛选出与内漏信息相关程度大的信号特征参数,最后分别利用BP神经网络和卷积神经网络(CNN)对筛选出的内漏声信号特征进行内漏速率反演。结果发现,CNN对球阀内漏速率的反演效果更好,其平均绝对误差MAE为0.0039,均方根误差R为0.004,优于BP神经网络。通过CNN实现了球阀内漏速率的反演,该研究结果为球阀泄露程度的判定和维修作业提供了便利。
The valve is a key control component in the system of oil and gas long-distance pipelines.The internal leakage detection of a valve not only provides safety guarantee for oil and gas production and operation,but also provides convenience for equipment maintenance and repair operations,which has important research significance.By setting up a ball valve internal leakage experiment platform,we use acoustic emission technology to collect ball valve internal leakage data,use wavelet packet threshold denoising method to denoise the signals,use Pearson correlation coefficient method to screen out the signal characteristic parameters with a large degree of correlation with internal leakage information.Finally,BP neural network and convolutional neural network CNN are used to invert the internal leakage rate of the selected internal leakage acoustic signal characteristics.The results showed that the convolutional neural network CNN has a better inversion effect on the internal leakage rate of the ball valve.Its mean absolute error is 0.0039,and the root mean square error is 0.004,which is better than the BP neural network.The inversion of the internal leakage rate of the ball valve is achieved through CNN,and the research results provide guidance for the determination of the degree of leakage of the ball valve.
作者
古丽
贾彦杰
冯兰婷
温皓
高仕玉
张琳
GU Li;JIA Yanjie;FENG Lanting;WEN Hao;GAO Shiyu;ZHANG Lin(PipeChina Southwest Pipeline Company,Chengdu 610036,China)
出处
《实验室研究与探索》
CAS
北大核心
2022年第4期57-62,67,共7页
Research and Exploration In Laboratory
关键词
声发射
球阀内漏
小波包去噪
相关系数法
反演
acoustic emission
ball valve internal leakage
wavelet packet denoising
correlation coefficient method
inversion
