基于改进BP神经网络算法的管道缺陷漏磁信号识别

Identification of Magnetic Flux Leak Inspection Signals for Pipeline Detection Using an Improved BP Neural Netwok Algorithm

海底管道漏磁检测信号处理的主要任务是根据霍尔传感器检测到的缺陷漏磁信号来识别缺陷的形态参数.根据漏磁检测原理设计了相关的漏磁检测电路,通过提取信号的主要特征量,利用Levenberg-Marquardt算法在对常用BP神经网络改进的基础上应用其来识别缺陷的尺寸参数,给出了BP神经网络各层数的确定及权值、学习率的调整方法和相应的漏磁信号数据处理过程.漏磁检测数据处理实验表明,该缺陷识别BP神经网络系统具有逼近精度高、收敛速度快等特点.

Recognizing the appearance parameters of defects is the main object in the signal processing of offshore oil pipeline inspection. Based on the feature extraction of magnetic flux leakage （MFL） inspection data, a modified back-propagation （BP） neural network system were presented to recognize pipeline defect and corrosion. The process of teaching and testing of BP neural networks to characterize the defect signals was analyzed. The adjusting method of network output weight and rate of learning based on Levenberg-Marquardt algorithm was also discussed. In order to obtain the perfect structure parameters of network and research each parameter＇s effect on the performance of network, a large number of simulation tests were conducted. The MFL data processing experiment proves that the defect recognition neural network system has high convergence speed and good ability of approaching defect features.