【作者】 高志伟；

【导师】 刘振宇； 曹光明；

【作者基本信息】 东北大学 ， 材料加工工程， 2020， 硕士

【摘要】 钢铁制造作为典型的流程工业,具有工艺流程复杂的特点,在钢铁材料的生产过程中产生大量强耦合性的工艺参数。针对大量的钢铁工业数据,如何利用机器学习算法准确的建立高强钢组织性能对应关系,已成为钢铁工业发展的重要方向。本文基于国内某钢厂的Q420B、510L、610L、700L、700XL、600XT六个高强钢种的生产数据,对原始数据进行关联分析。结合机器学习算法开发高强钢力学性能预测模型,采用MOEA/D多目标优化算法实现工艺参数的优化。基于以上的方法开发组织性能预测软件,实现力学性能预测及工艺优化技术的工业应用。本文主要研究内容如下:(1)针对数据中存在的异常值、数据分布不均等问题,结合数据处理方法和数据统计方法对数据进行处理。在此基础上,利用主成分分析方法分析参数与力学性能之间的关联性系数,筛选参数贡献率之和大于98%的参数,结合主要影响因素的分析,确定最终的模型输入参数。结果表明:数据处理方法提高数据质量,基于主成分分析方法和影响力学性能的主要因素确定模型的输入参数。(2)针对经验公式和回归模型存在预测精度较低的问题,利用粒子群多目标优化算法对随机森林算法的决策树数量和单棵树的叶子节点深度两个重要参数进行优化,建立高精度力学性能预测模型。将各个化学成分参数和工艺参数进行组合,验证模型参数与力学性能之间的规律,结果表明:PSO-RF(Particle Swarm Optimization Random Forest)算法具有较好拟合能力,采用预测模型对力学性能进行预测,取得了较高的预测精度,也验证了模型的可靠性。(3)基于分解的多目标进化算法,将多目标优化问题分解为单目标优化问题进行逐一优化,将基于分解的多目标进化算法与NSGA-Ⅱ算法的优化结果进行了比较。结果表明:基于分解的多目标进化算法的Pareto前沿具有显著的优越性。(4)以2250生产线为依托,搭建数据平台,结合数据处理方法、高强钢力学性能预测技术及工艺参数优化技术开发组织性能预测软件。针对高强钢Q420B、510L、610L、600XT、700L、700XL六个钢种建立高强钢力学性能预测模型,结合MOEA/D优化算法对工艺参数进行优化,实现610L升级轧制700XL和610L降级轧制Q420B。更多还原

【Abstract】 As a typical process industry,iron and steel manufacturing has the characteristics of complex process flow,and a large number of strong coupling process parameters are generated in the production process of steel materials.In view of the large amount of steel industry data,how to use machine learning algorithms to accurately establish the corresponding relationship between the structure and performance of high-strength steel has become an important direction for the development of the steel industry.Based on the production data of six high-strength steel grades Q420B,510L,610L,700L,700XL,and 600XT from a domestic steel plant,this paper conducts a correlation analysis on the original data.Combine machine learning algorithms to develop high-strength steel mechanical properties prediction models,and use MOEA/D multi-objective optimization algorithms to optimize process parameters.Based on the above methods,the organization performance prediction software is developed to realize the industrial application of mechanical performance prediction and process optimization technology.The main research contents of this paper are as follows:(1)In view of the outliers in the data and uneven data distribution,the data is processed by combining data processing methods and data statistical methods.On this basis,the principal component analysis method is used to analyze the correlation coefficient between the parameters and the mechanical properties,and the parameters with the sum of the parameter contribution rate greater than 98%are selected,combined with the analysis of the main influencing factors,to determine the final model input parameters.The results show that the data processing method improves the data quality,and the input parameters of the model are determined based on the principal component analysis method and the main factors affecting the mechanical properties.(2)In view of the low prediction accuracy of the empirical formula and regression model,the particle swarm multi-objective optimization algorithm is used to optimize the number of decision trees of the random forest algorithm and the depth of the leaf nodes of a single tree to establish high accuracy Mechanical properties prediction model.The chemical composition parameters and process parameters are combined to verify the law between model parameters and mechanical properties.The results show that the PSO-RF(Particle Swarm Optimization Random Forest)algorithm has a good fitting ability,and the predictive model is used to perform mechanical properties.The prediction has achieved high prediction accuracy and also verified the reliability of the model.(3)Based on the decomposition-based multi-objective evolutionary algorithm,the multi-objective optimization problem is decomposed into single-objective optimization problems and optimized one by one.The optimization results of the decomposition-based multi-objective evolution algorithm and the NSGA-Ⅱ algorithm are compared.The results show that the Pareto frontier based on the decomposition multi-objective evolutionary algorithm has significant advantages.(4)(4)Based on the 2250 production line,build a data platform,combine data processing methods,high-strength steel mechanical properties prediction technology and process parameter optimization technology to develop organization performance prediction software.Establish a prediction model for the mechanical properties of highstrength steels of Q420B,510L,610L,600XT,700L,and 700XL,and optimize the process parameters with the MOEA/D optimization algorithm to realize 610L upgrade rolling 700XL and 610L downgrade rolling Q420B.更多还原