摘要
钻头磨损状态的准确监测对于提高钻井效率、规避钻井事故的发生具有重大意义。针对钻井过程中钻头磨损状态监测困难的问题,结合井下近钻头工程参数短节所测数据,提出基于自适应噪声完备经验模态分解(CEEMDAN)、小波阈值以及卷积神经网络(CNN)的钻头磨损监测模型。首先将近钻头工程参数测量短节测得的振动数据进行基于自适应噪声完备经验模态分解,得到一系列本征模态函数分量,再对本征模态函数分量进行小波阈值去噪并完成信号重构,最后根据重构信号提取钻头磨损特征,完成卷积神经网络模型训练和钻头磨损状态识别。研究结果表明,钻头磨损监测模型精度达到92.3%,即该模型能准确识别钻头磨损状态且识别准确率高。研究结果可为及时调整钻井参数、确定更换钻头时机等提供技术支持。
Accurate monitoring of bit wear plays an important role in improving drilling efficiency and avoiding drilling accidents.In order to solve the problem of difficult monitoring of bit wear during drilling,combined with the data measured by the near-bit engineering parameter nipple,a bit wear monitoring model based on complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN),wavelet threshold and convolutional neural networks(CNN)was proposed.Firstly,the vibration data measured from the near-bit engineering parameter nipple were used for CEEMDAN to obtain a series of intrinsic mode function(IMF)components.Then,wavelet threshold muting was conducted for the IMF components and signal reconstruction was completed.Finally,the bit wear characteristics were extracted based on the reconstructed signal,and the CNN model training and the bit wear recognition were completed.The study results show that the accuracy of the bit wear monitoring model reaches 92.3%,that is,the model can very accurately recognize the bit wear.The study results provide technical support for timely adjusting drilling parameters and determining the timing of bit replacement.
作者
刘奕呈
李玉梅
张涛
李超
Liu Yicheng;Li Yumei;Zhang Tao;Li Chao(Beijing Key Laboratory of High Dynamic Navigation Technology,Beijing Information Science&Technology University;No.2 Drilling Engineering Company,BHDC)
出处
《石油机械》
北大核心
2022年第9期59-65,共7页
China Petroleum Machinery
基金
国家自然科学基金青年项目“干热岩储层双重介质射孔簇内复杂多裂缝起裂及扩展机理研究”(52104001)
北京信息科技大学重点研究培育项目“基于人工智能方法的近钻头高频扭转振动状态识别研究”(2121YJPY220)
北京市教委一般项目“高频复合振动冲击钻井立体破岩机理研究”(KM202111232004)
中国石油天然气集团有限公司-中国石油大学(北京)战略合作科技专项(ZLZX2020-03)。
关键词
钻头磨损
自适应噪声完备经验模态分解
小波阈值
卷积神经网络
近钻头工程参数
磨损监测
bit wear
complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)
wavelet threshold
convolutional neural network(CNN)
near-bit engineering parameter
wear monitoring
