基于LMBP神经网络的连续油管疲劳寿命预测方法  

Coiled Tubing Fatigue Life Prediction Method Based on LMBP Algorithm of Neural Network

作　　者：彭嵩 张全立 王宏伟 侯福祥 马汝涛 PENG Song;ZHANG Quanli;WANG Hongwei;HOU Fuxiang;MA Rutao(CNPC Drilling Research Institute,Beijing 112200,China)

机构地区：[1]中国石油集团钻井工程技术研究院,北京112200

出　　处：《石油管材与仪器》2018年第6期36-40,共5页Petroleum Tubular Goods & Instruments

基　　金：2006年中国石油集团重大项目;国家863项目(编号为2006AA06A106);项目名称为"连续管技术装备"

摘　　要：弯曲和内压作用下产生的疲劳是造成连续油管失效的主要原因。由于传统的疲劳方法对于复杂的多轴加载情况下连续油管疲劳寿命估算不理想,提出了采用LMBP人工神经网络(ANN)的方法,利用其优越的非线性逼近能力和泛化能力来建立连续油管疲劳寿命的预测模型。以连续油管的管径、壁厚及内压作为网络训练的输入向量,以标准实验机实验循环数作为训练目标向量,建立3层LMBP神经网络并对网络进行训练,并通过学习样本和测试样本对该神经网络模型进行验证。结果表明,预测值与实际值拟合较好,预测精度明显优于常规建模方法,验证了用LMBP神经网络预测连续油管疲劳寿命的可行性。该方法对开发连续油管疲劳寿命跟踪和监测软件及提升连续油管现场应用水平具有一定的借鉴作用。The fatigue under bending and internal press is the main reason for coiled tubing( CT) failure. Because the conventional life prediction methods for estimating CT fatigue life under multiracial loading are not ideal,a new method is proposed to predict the CT fatigue life by using its nonlinear approximation and generalization abilities based on the Levenberg-Marquardt back propagation( LMBP) artificial neural network( ANN). The CT size,wall thickness and internal pressure are used as ANN training input vectors,while the experimental cycle times of the CT are used as training target vectors. A three-layer LMBP neutral network is established and trained by training samples. And the ANN model is verified by learning samples and testing samples. The results show that the predicting data is comparatively consistent well with the experimental data; moreover,the precision of this method is superior to that applied by the conventional life prediction model. The feasibility of predicting the CT fatigue life by applying the LMBP ANN is proved. The above method provides a new route for developing CT fatigue life tracking and Monitoring software and improving the application level of CT for the field use.

关 键 词：连续油管 疲劳寿命 反向传播 LMBP神经网络 预测 样本 

分 类 号：TE933[石油与天然气工程—石油机械设备]