China Mechanical Engineering ›› 2022, Vol. 33 ›› Issue (19): 2325-2330.DOI: 10.3969/j.issn.1004-132X.2022.19.006

Previous Articles     Next Articles

Based on Machine Learning Intelligent Design and Properties Research of Large Sectional High Strength Martensite Steels for Petroleum Equipment

LI Fangpo1;LU Caihong1;ZHAO Jingxiao2;LI Xiucheng2;SHANG Chengjia2   

  1. 1.CNPC Tubular Goods Research Institute,State Key Laboratory for Performance and Structure Safety of Petroleum Tubular Goods and Equipment Materials,Xian,710077
    2.Collaborative Innovation Center of Steel Technology,University of Science and Technology Beijing,Beijing,100083
  • Online:2022-10-10 Published:2022-10-20

基于机器学习的石油装备用大截面高强韧马氏体钢智能设计与性能研究

李方坡1;路彩虹1;赵靖宵2;李秀程2;尚成嘉2   

  1. 1.中国石油集团工程材料研究院有限公司石油管材及装备材料服役行为与结构安全国家重点实验室,西安,710077
    2.北京科技大学钢铁共性技术协同创新中心,北京,10083
  • 作者简介:李方坡,男,1982年生,教授级高级工程师、博士。研究方向为石油管及装备新材料新工艺研发与应用。发表论文20余篇。E-mail:lifangpo@163.com。
  • 基金资助:
    中国石油天然气集团有限公司科研项目(2021ZG14,2019B-4014,2018E-2101);国家科技重大专项(2017ZX05009-003)

Abstract: In order to develop new materials and meet the requirements of ultra-deep petroleum and gas development, four prediction models for composition-yield strength and composition-hardness of large cross-sectional martensitic steel with the highest strength grade for equipment components were established based on machine learning and composition performance big data herein. The results show that the artificial neural network model with 4 layers of neurons and 64 layers depth has the best fitting degree for property predicting, and two optimized martensitic steels chemical composition design with yield strength greater than 1100 MPa, hardness greater than 42HRC and carbon content less than 0.22% are formed based on genetic algorithm. The experimental results show that the hardening distribution curve of designed materials is basically consistent with the predicted values, and the maximum error is less than 3HRC. According to the optimized composition, 35 batches of products were manufactured and testing results show that the materials may meet the performance requirements of 150 mm cross-sectional drilling rig components. More than 95% uniform fine acicular martensite may be obtained in the full cross-sections, yield strength is greater than 1100MPa and impact absorbed energy is greater than 45 J which meets the service requirements of petroleum equipment. The results of prediction performance are consistent with that of experiments. Material big data is combined with machine learning, which provides a new way for developing high performance petroleum equipment materials.

Key words: machine learning, big data, petroleum equipment, large sectional component, martensite steel, optimized chemical composition

摘要: 为了满足超深层油气资源开发需求,针对石油装备用强度等级最高的低碳马氏体钢,结合成分性能大数据,基于四种不同机器学习方法分别建立了大截面高强韧低碳马氏体钢成分强度和成分硬度预测模型,分析表明,神经元层数为4、层深为64的人工神经网络模型的性能预测精度和拟合程度最好。采用遗传算法对材料成分进行智能最优化设计,获得CrNiMo和SiMnCrNiMo两种材料系中屈服强度大于1100 MPa、硬度大于42HRC、碳含量小于0.22%的最优成分,材料的端淬硬度分布曲线与预测值基本一致,最大误差小于3HRC。依据优化设计成分进行多批次产品生产制造后结果表明,150 mm直径的构件全截面获得95%以上的细小针状马氏体组织,屈服强度大于1100 MPa,低温冲击吸收能大于45 J,满足服役性能要求,预测结果与生产实验结果具有较高的一致性。将材料大数据与机器学习相结合实现了材料的智能化设计开发,为高性能材料的开发提供了新途径。

关键词: 机器学习, 大数据, 石油装备, 大截面构件, 马氏体钢, 成分优化

CLC Number: