Prediction of the drop hammer-derived tear toughness of pipeline steel production lines using literature data and production line data
-
摘要: 管道运输是当前长距离输送石油、天然气等能源最经济的方式之一，具有优异的低温韧性是保证管线钢安全运输的重要特征。落锤撕裂试验（Drop weight tear testing，DWTT）是衡量管线钢低温韧性的最有效的方法。在目前的工作中，根据钢厂提供的产线数据集和文献收集的管线钢数据集，建立了基于机器学习的DWTT剪切面积预测模型。基于纯产线数据和文献数据辅助的产线数据构造了两种机器学习策略方案，测试了不同机器学习算法，效果最好的均是随机森林模型，策略一的纯产线数据模型的性能指标皮尔逊相关系数（PCC）为0.64，策略二的文献数据辅助的产线数据模型的性能指标皮尔逊相关系数（PCC）为0.92，文献数据的增加有效提高了DWTT剪切面积预测精度。机器学习技术为优化和预测DWTT剪切面积提供了一种新的思路。Abstract: Pipeline transportation is the most economical means of transporting oil, natural gas, and other energy sources over a long distance. With the increasingly harsh service environment of pipeline transportation, the requirements of pipeline steel in terms of strength, hydrogen-induced fracture resistance, and corrosion resistance have increased. In areas such as plateaus or deep seas, excellent low-temperature toughness is important to ensure the safe transportation of pipeline steel. Drop weight tear testing is one of the most effective methods for measuring the low-temperature toughness of pipeline steel. The test involves large specimens with full wall thickness. Through the characterization of the ductile–brittle shear area and ligament width of the sample, the toughness and tear resistance of pipeline steel can be better reflected. However, the drop weight tear test is difficult, time-consuming, and laborious, and it consumes a large amount of experimental resources. In this work, a machine learning-based model for predicting the drop weight tear test-derived shear area was established according to production line datasets provided by steel mills and pipeline steel datasets collected from the literature. Different machine learning algorithms were tested using the two datasets. The best models were random forest models. Strategy I included only production line datasets, and the Pearson correlation coefficient (PCC), which is the performance index, predicted by the machine learning model was 0.64. Strategy II involved literature data and production line data, and the PCC predicted by the machine learning model was 0.92. The consideration of literature data effectively improved the prediction accuracy of the drop weight tear test shear area. Moreover, in strategy II, to avoid the overfitting of the machine learning model, a feature screening technique was adopted. Finally, a genetic programming-based symbolic regression approach was developed to establish a formula describing the relationship between the selected features and the target shear area data. The PCC of the precision of this formula was 0.83, which indicates that the formula can be used to estimate the drop weight tear test-derived parameters of pipeline steel. The machine learning technology provides a new method for optimizing and predicting the drop weight tear test-derived shear area of pipeline steel. Moreover, the combination of production line data and literature data remarkably improved the accuracy of the machine learning model, which also allows for the prediction of other material production line data via machine learning techniques.
-
Key words:
- pipeline steel /
- drop weight tear testing /
- machine learning /
- data-driven design /
- symbolic regression
-
图 2 产线数据的机器学习模型结果. （a）6种机器学习模型在产线数据上的皮尔逊相关系数（PCC）和平均绝对误差（MAE）值；（b）实验值与RF模型预测值的比较（所有模型均通过10折交叉验证进行评估）
Figure 2. Machine learning (ML) model results for production line data: (a) pearson correlation coefficient (PCC) and mean absolute error (MAE) values of six ML models on production line data; (b) comparison between the experimental values and the predicted values of the RF model (All models were evaluated with 10-fold cross-validation)
图 3 DWTT 36个特征的Pearson相关系数图（在特征集的相关系数矩阵中，颜色越深，相关性越高）
Figure 3. Pearson correlation coefficient diagram of 36 features of DWTT (In the correlation coefficient matrix of the feature set, the deeper the color, the higher the correlation)
图 4 管线钢DWTT的特征筛选过程. （a）13个特征子集的平均重要性得分按升序排列;每个可能的特征子集RF模型的（b）皮尔逊相关系数（PCC）和（c）平均绝对误差（MAE）值分布
Figure 4. Feature screening process of pipeline steel DWTT: (a) the average importance scores of 13 feature subsets are ranked in ascending order; the distribution of (b) pearson correlation coefficient (PCC) and (c) mean absolute error (MAE) values for each possible feature subset RF model
图 5 文献数据辅助的产线数据机器学习模型结果. （a）6种机器学习模型在文献数据辅助的产线数据集上的皮尔逊相关系数（PCC）和平均绝对误差（MAE）；（b）实验值与RF模型预测值的比较（所有模型均通过10折交叉验证进行评估）
Figure 5. Results of the machine learning model on the combination of literature data and production line data: (a) pearson correlation coefficient (PCC) and mean absolute error (MAE) of six machine learning models on the combination of literature data and production line data; (b) comparison between the experimental and predicted values of the RF model (All models were evaluated with 10-fold cross-validation)
表 1 DWTT产线数据集中27个量
Table 1. Twenty-seven values in the DWTT production line dataset
Input/output Abb. Description Max Min Mean Inputs Fe Mass fraction of iron/% 97.952 97.640 97.803 C Mass fraction of carbon/% 0.090 0.050 0.063 Si Mass fraction of silicon/% 0.250 0.150 0.204 Mn Mass fraction of manganese/% 1.700 1.550 1.618 P Mass fraction of phosphorus/% 0.018 0.008 0.013 S Mass fraction of sulfur/% 0.003 0.001 0.002 Ni Mass fraction of nickel/% 0.020 0.000 0.007 Cr Mass fraction of chromium/% 0.220 0.018 0.123 Mo Mass fraction of molybdenum/% 0.090 0.050 0.069 Ti Mass fraction of titanium/% 0.017 0.008 0.012 Cu Mass fraction of copper/% 0.024 0.000 0.010 V Mass fraction of vanadium/% 0.016 0.000 0.002 Al Mass fraction of aluminum/% 0.044 0.016 0.030 N Mass fraction of nitrogen/% 0.008 0.001 0.004 Nb Mass fraction of niobium/% 0.050 0.030 0.039 B Mass fraction of boron/% 0.0003 0.0001 0.0001 Ceq Carbon equivalent 0.404 0.339 0.374 TH Thickness/mm 18 12 16.838 HV Vickers hardness 268 169 200.575 UTS Tensile strength/MPa 746 493 613.817 YS Yield strength/MPa 638 400 535.162 EL Elongation/% 50 20 41.465 YR Yield strength to tensile strength ratio 0.96 0.68 0.872 ST Secondary rolling temperature/°C 1079 654 898.165 FT Final rolling temperature/°C 886 666 837.054 TTE Test temperature/°C −15 −20 −15.441 Output SA Shear area/% 100 88 95.132 
下载: 导出CSV
表 2 DWTT文献数据集中20个量
Table 2. Twenty values in DWTT literature dataset
Input/output Abb. Description Max Min Mean Inputs Fe Mass fraction of iron/% 98.18 95.63 97.680 C Mass fraction of carbon/% 0.105 0.035 0.055 Si Mass fraction of silicon/% 0.40 0.00 0.237 Mn Mass fraction of manganese/% 1.94 1.10 1.657 P Mass fraction of phosphorus/% 0.016 0.00 0.007 S Mass fraction of sulfur/% 0.013 0.00 0.002 Ni Mass fraction of nickel/% 0.40 0.00 0.081 Cr Mass fraction of chromium/% 0.30 0.00 0.051 Mo Mass fraction of molybdenum/% 0.30 0.00 0.090 Ti Mass fraction of titanium/% 0.33 0.00 0.021 Cu Mass fraction of copper/% 0.30 0.00 0.050 V Mass fraction of vanadium/% 0.50 0.00 0.025 Al Mass fraction of aluminum/% 0.04 0.00 0.008 Nb Mass fraction of niobium/% 0.50 0.00 0.036 TH Thickness/mm 38.50 7.50 21.715 UTS Tensile strength/MPa 1023 540 649.327 YS Yield strength/MPa 951 469 559.944 EL Elongation/% 57 8 34.131 TTE Test temperature/°C 0 −80 −29.019 Output SA Shear area/% 100 9 80.014
下载: 导出CSV
表 3 原子特征列表
Table 3. List of atomic features
Abb. Description Formula References DPEFe PE difference (Fe-based) $\mathop { {\text{d} }A}\nolimits^K = \sqrt {\displaystyle\sum\nolimits_{i = 1}^n {\mathop a\nolimits_i \mathop {\left( {1 - \frac{ {\mathop A\nolimits_i } }{ {\mathop A\nolimits_k } } } \right)}\nolimits^2 } }$ [8,32–35] DPEC PE difference (C-based) DVEFe VE difference (Fe-based) DVEC VE difference (C-based) DARFe AR difference (Fe-based) CDOR Mean concentration of DOR ${\text{CA} } = \frac{ {\displaystyle\sum\nolimits_{i = 1}^n {\mathop a\nolimits_i \mathop A\nolimits_i } } }{ {\displaystyle\sum\nolimits_{i = 1}^n {\mathop a\nolimits_i \mathop { {\text{AN} } }\nolimits_i } } }$ CVE Mean concentration of VE CMR Mean concentration of MR CCS Mean concentration of CS CCR Mean concentration of CR CPE Mean concentration of PE CAM Mean concentration of AM EBE Electron binding energies AN Atomic number $A = \displaystyle\sum\limits_{i = 1}^n {\mathop a\nolimits_i \mathop A\nolimits_i }$ AM Atomic mass PR Poisson’s ratio MP Melting point YM Young’s modulus BP Boiling point NE Number of elements CE Cohesive energy DEN density E1 First ionization energy DOR Waber–Cromer pseudopotential radii TVE Valance electron MR Metallic radius CS Pettifor chemical scale CR Clementi’s atomic radii PE Pauling electronegativity AR Atomic radii MV Molar volume
下载: 导出CSV
-
参考文献
[1] Yoo J Y, Ahn S S, Seo D H, et al. New development of high grade X80 to X120 pipeline steels. Mater Manuf Process, 2011, 26(1): 154 doi: 10.1080/10426910903202534 [2] Li H L, Ji L, Kang T W. Significant technical progress in the West-East Gas Pipeline Projects-Line One and Line Two. Nat Gas Ind, 2010, 30(4): 1李鹤林, 吉玲, 康田伟. 西气东输一、二线管道工程的几项重大技术进步. 天然气工业, 2010, 30(4):1 [3] Ji L K, Feng H, Zhang J M, et al. Strain-hardening behavior of high grade linepipe steel and its influence factors. J Xi’an Shiyou Univ (Nat Sci Ed), 2017, 32(3): 99吉玲康, 封辉, 张继明, 等. 高钢级管线钢形变硬化行为及其影响因素分析. 西安石油大学学报(自然科学版), 2017, 32(3):99 [4] Ren Z P, Li D G, Li X W, et al. Construction and thinking of big data research and development platform for steel enterprises. J Iron Steel Res, 2021, 33(11): 1118 doi: 10.13228/j.boyuan.issn1001-0963.20200276任子平, 李德刚, 李晓伟, 等. 钢铁企业大数据研发平台的建设与思考. 钢铁研究学报, 2021, 33(11):1118 doi: 10.13228/j.boyuan.issn1001-0963.20200276 [5] Wang G D, Liu Z Y, Zhang D H, et al. Transformation and development of materials science and technology and construction of iron and steel innovation infrastructure. J Iron Steel Res, 2021, 33(10): 1003 doi: 10.13228/j.boyuan.issn1001-0963.20210053王国栋, 刘振宇, 张殿华, 等. 材料科学技术转型发展与钢铁创新基础设施的建设. 钢铁研究学报, 2021, 33(10):1003 doi: 10.13228/j.boyuan.issn1001-0963.20210053 [6] Li X R, Ban X J, Yuan Z L, et al. Review on deep learning models for time series forecasting in industry. Chin J Eng, 2022, 44(4): 757李潇睿, 班晓娟, 袁兆麟, 等. 工业场景下基于深度学习的时序预测方法及应用. 工程科学学报, 2022, 44(4):757 [7] Geng X X, Mao X P, Wu H H, et al. A hybrid machine learning model for predicting continuous cooling transformation diagrams in welding heat-affected zone of low alloy steels. J Mater Sci Technol, 2022, 107: 207 doi: 10.1016/j.jmst.2021.07.038 [8] Chen Y M, Wang S Z, Xiong J, et al. Identifying facile material descriptors for Charpy impact toughness in low-alloy steel via machine learning. J Mater Sci Technol, 2023, 132: 213 doi: 10.1016/j.jmst.2022.05.051 [9] Fürnkranz J, Gamberger D, Lavrač N. Foundations of Rule Learning. Lodon: Springer Science & Business Media, 2012 [10] Jiang X, Jia B R, Zhang G F, et al. A strategy combining machine learning and multiscale calculation to predict tensile strength for pearlitic steel wires with industrial data. Scripta Mater, 2020, 186: 272 doi: 10.1016/j.scriptamat.2020.03.064 [11] Zhao J H, Wang X Q, Kang J, et al. Crack propagation behavior during DWTT for X80 pipeline steel processed via ultra-fast cooling technique. Chin J Mater Res, 2017, 31(10): 728 doi: 10.11901/1005.3093.2016.055赵金华, 王学强, 康健, 等. 超快冷工艺下X80管线钢的DWTT裂纹扩展行为. 材料研究学报, 2017, 31(10):728 doi: 10.11901/1005.3093.2016.055 [12] Jiang C L, Lin T, Zhu J Y. Research on low-temperature DWTT of 33 mm thick-wall X80 pipeline steel plate. Iron Steel Vanadium Titanium, 2019, 40(4): 158 doi: 10.7513/j.issn.1004-7638.2019.04.029蒋昌林, 林涛, 诸建阳. 33 mm壁厚X80管线钢板的低温落锤撕裂研究. 钢铁钒钛, 2019, 40(4):158 doi: 10.7513/j.issn.1004-7638.2019.04.029 [13] Lu C J, Shao C J, Zhang L, et al. Experimental research on DWTT performance of high-strength pipeline steel. J Iron Steel Res, 2019, 31(12): 1065陆春洁, 邵春娟, 张磊, 等. 高强度管线钢落锤撕裂性能的试验. 钢铁研究学报, 2019, 31(12):1065 [14] Huo X X, Zhou P, Huang S W, et al. Effect of the microstructure on the DWTT properties of X70 pipeline steel. Shandong Metall, 2011, 33(5): 99 doi: 10.3969/j.issn.1004-4620.2011.05.036霍孝新, 周平, 黄少文, 等. X70管线钢微观组织结构对落锤性能的影响. 山东冶金, 2011, 33(5):99 doi: 10.3969/j.issn.1004-4620.2011.05.036 [15] Liu B, Wei F, Zhao Y, et al. Development and performance study of X80 Φ1422 mm × 38.5 mm thick wall SAWL pipe. Welded Pipe Tube, 2021, 44(1): 1刘斌, 韦奉, 赵勇, 等. X80钢级Ф1422 mm×38.5 mm大壁厚直缝埋弧焊管的开发及性能研究. 5焊管, 2021, 44(1):1 [16] Su D X, Xu W C. Research on influence of specimen thickness and drop weight test method on X70 pipeline steel DWTT performance. Welded Pipe Tube, 2018, 41(7): 21苏大雄, 徐惟诚. 试样厚度及落锤试验方式对X70 DWTT性能影响的研究. 焊管, 2018, 41(7):21 [17] Gao Y, Jiang J X, Zuo X R, et al. Microstructures and fracture behaviors in heavy thick multi-phase X80 pipeline steel. J Netshape Form Eng, 2018, 10(6): 43 doi: 10.3969/j.issn.1674-6457.2018.06.008高燕, 姜金星, 左秀荣, 等. 厚规格多相组织X80管线钢组织控制与断裂行为研究. 精密成形工程, 2018, 10(6):43 doi: 10.3969/j.issn.1674-6457.2018.06.008 [18] Agrawal A, Deshpande P D, Cecen A, et al. Exploration of data science techniques to predict fatigue strength of steel from composition and processing parameters. Integr Mater Manuf Innov, 2014, 3(1): 90 doi: 10.1186/2193-9772-3-8 [19] Strnadel B, Ferfecki P, Židlík P. Statistical characteristics of fracture surfaces in high-strength steel drop weight tear test specimens. Eng Fract Mech, 2013, 112-113: 1 doi: 10.1016/j.engfracmech.2013.10.001 [20] Hwang B, Shin S Y, Lee S, et al. Effect of microstructure on drop weight tear properties and inverse fracture occurring in hammer impacted region of high toughness X70 pipeline steels. Mater Sci Technol, 2008, 24(8): 945 doi: 10.1179/174328406X148732 [21] Sha Q Y, Li D H. Microstructure, mechanical properties and hydrogen induced cracking susceptibility of X80 pipeline steel with reduced Mn content. Mater Sci Eng A, 2013, 585: 214 doi: 10.1016/j.msea.2013.07.055 [22] Hwang B, Kim Y G, Lee S, et al. Effective grain size and charpy impact properties of high-toughness X70 pipeline steels. Metall Mater Trans A, 2005, 36(8): 2107 doi: 10.1007/s11661-005-0331-9 [23] Zhang J M, Sun W H, Sun H. Mechanical properties and microstructure of X120 grade high strength pipeline steel. J Iron Steel Res Int, 2010, 17(10): 63 doi: 10.1016/S1006-706X(10)60185-9 [24] You Y, Shang C J, Nie W J, et al. Investigation on the microstructure and toughness of coarse grained heat affected zone in X-100 multi-phase pipeline steel with high Nb content. Mater Sci Eng A, 2012, 558: 692 doi: 10.1016/j.msea.2012.08.077 [25] Matrosov Y I, Bagmet O, Nosochenko A. Development of modern heavy plate steels for pipelines. Proceedings of the Materials Science Forum, 2007, 539-543: 4756 doi: 10.4028/www.scientific.net/MSF.539-543.4756 [26] Johnson J, Hudson M, Takahashi N, et al. Specification and manufacturing of pipes for the X100 operational trial // Proceedings of 2008 7th International Pipeline Conference. Calgary, 2009: 453 [27] Hillenbrand H G, Liessem A, Grimpe F, et al. Manufacturing of X100 pipes for the TAP project // Proceedings of the International Pipeline Conference. Calgary, 2006: 261 [28] Zha C H, Jiang Z H, Wang W J, et al. Research and development of heavy wall X80 transmission pipeline steel with high deformation characteristics for polar environments at Shougang steel // Proceedings of the International Pipeline Conference. Calgary, 2012: 249 [29] Aihara S, Lange H I, Misawa K, et al. Full-scale burst test of hydrogen gas X65 pipeline // Proceedings of 2010 8th International Pipeline Conference. Calgary, 2010: 415 [30] Jordan M I, Mitchell T M. Machine learning: Trends, perspectives, and prospects. Science, 2015, 349(6245): 255 doi: 10.1126/science.aaa8415 [31] Chai T, Draxler R. Root mean square error (RMSE) or mean absolute error (MAE)? –Arguments against avoiding RMSE in the literature. Geosci Model Dev, 2014, 7: 1247 doi: 10.5194/gmd-7-1247-2014 [32] Pauling L. The nature of the chemical bond. IV. The energy of single bonds and the relative electronegativity of atoms. J Am Chem Soc, 1932, 54(9): 3570 [33] Waber J, Cromer D T. Orbital radii of atoms and ions. J Chem Phys, 1965, 42(12): 4116 doi: 10.1063/1.1695904 [34] Clementi E, Raimondi D L. Atomic screening constants from SCF functions. J Chem Phys, 1963, 38(11): 2686 doi: 10.1063/1.1733573 [35] Rabe K M, Phillips J C, Villars P, et al. Global multinary structural chemistry of stable quasicrystals, high-TC ferroelectrics, and high-Tc superconductors. Phys Rev B Condens Matter, 1992, 45(14): 7650 doi: 10.1103/PhysRevB.45.7650 [36] Zhang Y, Wen C, Wang C X, et al. Phase prediction in high entropy alloys with a rational selection of materials descriptors and machine learning models. Acta Mater, 2020, 185: 528 doi: 10.1016/j.actamat.2019.11.067 [37] He J J, Li J J, Liu C B, et al. Machine learning identified materials descriptors for ferroelectricity. Acta Mater, 2021, 209: 116815 doi: 10.1016/j.actamat.2021.116815 [38] Romanski P, Kotthoff L, Kotthoff M L. Package ‘FSelector’ [J/OL]. Sciencepaper Online (2009-10-29) [2022-12-19]. https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=52ee687db3d94ec2143ab44021a303301ad133b1 [39] Chandrashekar G, Sahin F. A survey on feature selection methods. Comput Electr Eng, 2014, 40(1): 16 doi: 10.1016/j.compeleceng.2013.11.024 [40] James G, Witten D, Hastie T, et al. An Introduction to Statistical Learning. New York: springer, 2013 -
点击查看大图
图(5) / 表(3)
计量
- 文章访问数: 217
- HTML全文浏览量: 75
- PDF下载量: 51
- 被引次数: 0
