连铸坯枝晶臂间距检测方法的优化与应用实践

介绍了连铸坯凝固组织中树枝晶臂间距检测方法的优化试验及其应用效果。通过对连铸坯凝固冷却过程的分析及批量数据的采集,确定了测定连铸坯树枝晶臂间距的代表性区域,建立了一种科学测量枝晶臂间距的方法。将该方法应用于连铸圆坯生产检验中,能快速准确地找出铸坯枝晶臂间距随连铸二冷强度变化的规律,采取调整连铸二冷强度的措施,来控制一、二次枝晶臂间距大小,达到了减轻或消除铸坯开裂缺陷、提高连铸质量的目的。     

BP神经网络在连铸板坯质量在线诊断中的应用

为了更好地应用BP神经网络对连铸板坯质量进行在线诊断,基于连铸生产特点,利用采集的过程数据建立了符合生产实际的均一化函数.通过分析BP神经网络中各参数对网络性能及诊断准确率的影响,对BP神经网络的结构及学习算法进行修正,使该网络有选择和有区分地学习铸坯质量知识.结合某钢厂连铸现场数据,以黏结为例,建立了6种网络模型,对各模型算法进行了比较测试.结果表明:采用自定义函数均一化样本或采用提出的差异性算法训练神经网络,均可明显提高诊断准确率;采用选择性算法可确保诊断准确率不变的同时,提高学习速度;修正的算法更能很好地符合连铸生产实际.     

基于模式识别和人工神经网络建立的板坯连铸二冷水模型

介绍了一个种铸坯表面测温的方法,并依据铸坯表面温度,应用模式识别与人工神经网络相结合的方法,建立了板坯连铸的二冷水模型。该模型既可根据模式识别的分类图预报优化的二次工艺参数,亦可根据钢种、中间包钢水温度、拉速及希望的二次区不同部位的铸坯表面温度由LS顾的输入神经元网络,预测报连铸不同回路的水量。该模型运用于生产实际后,铸坯表面裂纹大为减少,取得了较好的效果。     

基于案例推理的钢水温度在线管控模型

钢水温度的精确管控有利于提高铸坯质量和降低生产成本。针对目前炼钢—连铸区段钢水温度在线管控方面存在的不足,在分析钢水温度影响因素的基础上,建立了基于案例推理的炼钢—连铸区段钢水温度在线管控模型。同时通过调整案例推理算法的相似度计算方法、权重计算方法、重用案例个数等参数提高模型的精度。结果表明:转炉出钢温度预定模型平均降低转炉出钢温度6℃,精炼结束温度预定模型提高连铸开浇温度命中率2.33%。精炼开始温度,精炼结束温度和连铸开浇温度预报模型误差小于10℃的命中率分别达到75.33%、98.33%和95.67%,且均高于神经网络模型。     

Modeling on Dendrite Growth of Medium Carbon Steel during Continuous Casting

Dendrite growth is an important phenomenon during steel solidification. In the current paper, a numerical method was used to analyse and calculate the dendrite tip radius, dendrite growth velocity, liquid concentration, temperature gradient, cooling rate, secondary dendrite arm spacing, and the dendrite tip temperature in front of the solid/liquid (S/L) interface for the solidification process of medium carbon steels during continuous casting. The current model was well validated by published models and measurement data. The results show that in the continuous casting process, the dendrite growth rate is dominated by the casting speed. Dendrite growth rate, liquid concentration at the S/L interface, temperature gradient and cooling rate decrease with proceeding solidification and solid shell thickness growth, while other parameters such as dendrite tip radius, secondary dendrite arm spacing, and dendrite tip temperature in front of the S/L interface become larger with solidification progress and solid shell thickness growth. Parametric investigations were carried out. The effects of the stability coefficient, temperature gradient and casting speed on the micro‐structural parameters were discussed. Under the same conditions, higher casting speed promotes coarser secondary dendrite arm spacing and enlarges the dendrite tip radius, while decreasing temperature gradient, reducing the dendrite growth rate and making the solute distribute more uniform.     

基于改进BP算法的连铸板坯中心偏析预报模型

结合某厂连铸生产数据,采用带有附加动量项的改进BP算法,建立了连铸板坯中心偏析的BP人工神经网络预测模型。应用结果表明,其预测准确率为90%,可满足连铸生产中对铸坯中心偏析预报精度的要求。分析导致预报偏差的主要原因是,网络模型隐含层节点较多、网络结构复杂、中心偏析等级为1.0的样本学习次数较多和噪音样本剔除不彻底等。     

Investigation of solidification and microsegregation of near-net-shape cast carbon steel

Different near-net-shape casting techniques are investigated in terms of solidification parameters, microstructure, and microsegregation of manganese in carbon steels based on experimental simulation methods: ingot casting of thick and thin slab samples, strip samples, and thin strip. The as-cast thicknesses were between 1.9 and 150 mm. By calculations and measurements data have been determined which lead to the solidification structure like solidification- and cooling-rate, secondary dendrite arm spacings, and the concentrations of microsegregations. Finally, some literature data of material properties of steel strip produced by the investigated methods are given.     

基于支持向量机回归的铸坯表面目标温度设定

提出了一种多输入多输出支持向量机回归算法,利用冶金技术人员计算的目标温度设定表,设定实时二冷区铸坯表面目标温度。200 mm×1 534 mm 16Mn钢板坯连铸试验结果表明,在训练样本相同时,支持向量机训练时间为3.2 s,预测目标温度误差为±1℃,BP神经网络训练时间为23.5 s,预测目标温度误差为±2℃,多输入多输出支持向量机回归算法优于BP神经网络算法,能够根据工艺变化情况,实时改变目标温度,为实现连铸动态控制提供了条件,有助于提高铸坯的质量。     

基于极限学习机(ELM)的连铸坯质量预测

针对传统基于BP神经网络建立的连铸坯质量预测模型训练速度慢、适应能力弱、预测精度低等问题,本文提出一种基于极限学习机的连铸坯质量预测方法,对方大特钢60Si2Mn连铸坯中心疏松和中心偏析缺陷进行预测,并与BP和遗传算法优化BP神经网络预测模型的预测结果进行分析对比.结果表明:BP及GA-BP神经网络预测模型对连铸坯中心疏松和中心偏析缺陷的预测准确率分别为50%、57.5%、70%和72.5%;而基于极限学习机的连铸坯预测模型预测准确率更高,对连铸坯中心疏松和中心偏析缺陷的预测准确率分别为85%和82.5%,且该模型具有极快的运算时间,仅需0.1 s.该模型可对连铸坯质量进行迅速准确地分析,为连铸坯质量预测的在线应用提供了一种新的方法.      

结晶器专家系统在莱钢宽厚板连铸机中的应用

为减少漏钢事故,莱钢在宽厚板连铸机应用结晶器专家系统,使用神经元网络模块建立振动摩擦力分析模型,通过拉钢过程中摩擦力的变化特征来预报不同类型的漏钢;将实时采集到的热电偶温度、拉速、液位、摩擦力等作为模型的基本参数,钢种、水垢厚度、保护渣、铸坯宽度等作为辅助参数,建立结晶器热力学模型,采用离线方式训练神经元网络。应用表明,连铸过程漏钢预警预报频率逐渐减少。     

连铸板坯表面质量预报专家系统

介绍了国内外连铸板坯质量预报系统的发展情况和研究开发的预报专家系统的工作原理，基本构成。本系统使用神经元网络模块和逻辑模块对结晶器热电偶温度模式进行识别，同时采集连铸机各项设备参数和工艺参数，利用冶金学理论知识和现场专家的经验知识，对发生铸坯表面质量的可能性给出判定。     

基于改进人工神经网络的LF钢水终点温度预报

采用改进的人工神经网络算法,开发了40t钢包炉精炼时钢水终点温度预报模型。与传统BP网络算法相比较,改进算法可提高预测速度和精度。生产现场实验表明,传统BP神经网络算法,钢水温度预测误差±5℃的炉次仅为77%,用改进的BP神经网络算法,其误差±5℃的炉次为90%。     

宝钢在线二冷控制模型的研发与应用

针对连铸生产过程中的二冷配水问题,建立了铸坯的凝固传热数学模型。通过实时模拟计算铸坯的温度场,并与PID控制技术相结合,开发了在线二冷控制模型。模型能自动根据钢种、铸坯规格及工艺参数的变化动态调整二冷控制水量,将铸坯的表面温度控制在工艺目标值附近。通过设计合理的控制系统架构,确保了二冷控制系统的稳定性及可靠性。在线测温结果表明,模型具有很高的计算精度。当拉速、浇注钢水过热度变化时,模型能快速将水量调整到目标值,速度快且超调小,从而确保铸坯的表面温度跟踪误差始终限制在较小范围内;当浇注过程处于相对稳态时,铸坯的表面温度保持在目标值。目前,模型已经应用于宝钢内外的多台连铸机,应用效果良好。     

不锈钢铸坯宽度优化控制模型研究及应用

 铸坯宽度的精确控制是连铸生产过程中保证铸坯质量的重要手段之一,过宽或过窄都会给轧钢工序带来不利影响。目前,铸坯宽度主要通过人工经验在线调整二冷区参数来进行控制,存在铸坯宽度合格率低、波动大的问题。因此,实现不锈钢铸坯宽度的自动优化控制就极为重要。针对不锈钢铸坯宽度控制过程中影响因素多且具有非线性、多变量耦合等特性,提出一种基于随机森林-差分进化(RF-DE)算法的铸坯宽度优化控制模型。首先,根据专家经验从大量连铸生产工艺参数中选取不锈钢铸坯宽度的关键影响因素,进而采用随机森林算法(random forest,RF)建立不锈钢铸坯宽度预测模型。接着,以连铸二冷工艺参数为决策变量,以宽度模型预测值与铸坯目标宽度的绝对误差为目标函数,构建带工艺约束条件的不锈钢铸坯宽度优化控制模型,进而采用差分进化算法对上述模型进行优化求解,获得可用于连铸生产控制的二冷工艺参数设定值。最后,通过某钢厂实际生产数据对模型进行验证。结果表明,相较于梯度提升树、支持向量回归、多层感知机等模型,随机森林模型泛化能力强且精确度高,更适用于铸坯宽度的预测,其平均绝对误差MAE为0.047 2 mm;且采用宽度优化控制模型后的不锈钢铸坯宽度合格率显著提高,新模型具有较高的宽度控制精度。     

连铸板坯三维二冷动态配水与精准压下研究与应用

在现有工艺条件下,校验和完善二冷区铸坯凝固传热计算数学模型,开发三维二冷配水模型,解决目前设备状况下冷却水分布不均匀对铸坯温度的影响,从而控制铸坯表面质量,特别是铸坯的角部裂纹,同时对板坯连铸二冷配水制度进行改进和优化,使之满足高效连铸生产条件和改善铸坯质量的需要。提出压下参数计算公式,结合所开发三维二冷配水模型,优化现有压下工艺,提出并应用精准可控单段压下、非稳态压下控制,集中解决连铸板坯中心偏析、中心疏松和缩孔等内部质量问题。同时优化模型数据库,使之数据更加完备,模型计算更加准确,同时模型具备异钢种混浇过程二冷及压下控制功能,能够进行凝固终点W形预测与控制,可进一步提高模型适用性和准确性。模型开发并成功在多家钢厂现场应用,有效改善了铸坯裂纹和偏析等铸坯表面和内部的质量问题。      

连铸板坯表面温度控制器的设计

以某钢厂引进的连铸板坯二冷动态控制系统为研究对象,提出了基于改进粒子群BP算法的板坯二冷区表面温度神经网络控制器.以实际生产现场的设备、工艺参数为基础进行了仿真研究,结果表明表面温度神经网络控制器的输出结果与实际生产参数的误差小于2%,研究结果对引进的同类连铸板坯二冷控制系统的升级改造具有指导意义.     

Characteristic Parameters for Dendritic Microstructure of Solidification During Slab Continuous Casting

 The operating conditions during the continuous casting process have a great effect on the microstructure of slab solidification, including primary and secondary dendritic arm spacing. On the basis of the analysis of available work, a revised expression for describing secondary dendritic arm spacing during solidification of slab continuous casting was presented, and the relation between the ratio of primary dendritic arm spacing to secondary dendritic arm spacing and the cooling rate was obtained.     

Modelling of Heat Transfer,Dendrite Microstructure and Grain Size in Continuous Casting of Steels

The dendrite arm spacing and grain size in continuous casting has been studied by mathematical modelling and experimental measurements. Two in‐house tools have been used in the study. The heat transfer is calculated by the model called TEMPSIMU and the solidification as well as the microstructural phenomena by the thermodynamic‐kinetic software called IDS. The models are validated by comparison the calculated results with experiments from steel plants. In continuous casting, the solidification structure is also influenced by process parameters. In this study the effect casting speed, superheat and secondary cooling on arm spacings and grain size is also studied. The in‐house models and the obtained results are presented in this paper. Using the developed models, the heat transfer and microstructure can be controlled more accurately.     

基于机器学习的Cu-Ni-Co-Si合金固溶处理晶粒尺寸预测

Cu-Ni-Co-Si合金在固溶处理后的晶粒尺寸会影响其服役性能。采用3种不同的机器学习方法——BP神经网络、随机森林、长短期记忆网络,分别建立了Cu-Ni-Co-Si合金固溶温度和固溶时间对固溶后晶粒尺寸影响的机器学习预测模型。对比分析3种不同机器学习模型的预测精度,发现BP神经网络模型预测精度最高,其平均相对误差为8.55%。随后采用遗传算法优化BP神经网络。结果表明,所建立的BPGA模型平均相对误差比BP神经网络模型降低了6.47个百分点,其平均相对误差为2.08%,能够有效地为Cu-Ni-Co-Si合金固溶处理工艺参数的选择提供指导。     

Simulation of macroscopic solidification with an incorporated one-dimensional microsegregation model coupled to thermodynamic software

In this article, a numerical model is presented which predicts phase distributions and dendrite arm spacings for a realistic casting within suitable CPU time. Three software components are coupled to perform calculations: (1) an FEM simulation package for the macroscopic temperature field, (2) an FDM code for the microstructure parameters, and (3) a thermodynamic software package for equilibrium calculations at the interfaces. The macrosoftware provides the micromodule with the present temperature at each node of a finite-element grid. At each such node, the micromodule calculates the dendrite arm spacings, the phase amounts, and the diffusion-controlled segregation profiles for the current time-step using equilibrium information from the thermodynamic software. A change of solid fraction and of the phase concentrations results in the release of latent heat and in the change of the heat capacity. These values are used as input parameters in the macrosoftware for the temperature calculation in the next time-step. Simulations have been performed for the ternary alloy AlCu4Mg1 and the results have been compared to “traditional” temperature calculations and to experimentally determined phase fractions and dendrite arm spacings. Measurements have been done by means of an interactive image analysis system over the entire breadth of an ingot casting at four different heights as well as at three different longitudinal cuts.