【作者】 贾涛；

【导师】 王国栋； 刘振宇；

【作者基本信息】 东北大学 ， 材料加工工程， 2008， 博士

【摘要】 热轧板带TMCP工艺的制定是一项耗时耗力的研究工作。与此同时,由于热轧过程组织演变的复杂性和实际轧制生产中的各种约束条件,传统的TMCP工艺开发方法只能给出较好而并非最优的工艺制度。因此,开发出以产品力学性能为目标优化热轧工艺的新方法是非常有必要的。它的基本概念为建立钢材热轧过程组织演变与性能预测模型,以热轧生产线的设备能力为约束条件,优化计算工艺参数从而达到预设的力学性能目标。在课题的研究中,作者分别采用物理冶金学理论和人工智能技术对热轧过程的组织演变和化学成份-工艺-力学性能关系进行描述,从规律性和预测精度两方面完成建模工作；并结合单目标和多目标粒子群算法,实现了热轧工艺的优化设计。本论文的主要创新性工作如下：(1)连续冷却过程中可加性法则有效性的研究针对四种不同成份的Nb微合金钢,采用热膨胀实验测定了不同冷却速度条件下奥氏体向铁素体的相变动力学。首次将Rios推导的理论方法应用于连’续冷却过程相变建模,利用相变初期数据拟合lnln[1／(1-X)]和1n(|CR|)(CR-冷却速率)获得了很好的线性关系,计算得出了Johnxon-Mehl-Avrami-Kolmogorov (JMAK)方程中的指数n。当结合RIOS方法计算的动力学参数k用于连续冷却相变预测时产生了较大的偏差。而通过假设参数k与温度呈高斯函数关系,采用优化方法计算出的参数k表明了与冷却速率的相关性,用于预测连续冷却相变动力学时也获得了较高的精度。研究表明如果将冷却速率的因素考虑在内,可加性法则基本成立。(2)基于数值分析的连续冷却过程相变建模在RIOS方法的基础之上,开发了用于分析连续冷却热膨胀数据的数值方法,从数学的角度解决了连续冷却相变动力学的建模问题。通过预设JMAK方程中的指数n和温度梯度ΔT,利用可加性法则可以计算出相变任意时刻的参数k。通过不同冷速条件下参数k的对比分析,可得参数k与温度的关系及与已相变份数X的相关性；综合应用RIOS方法和基于JMAK方程与可加性法则的建模新方法,建立了TRIP钢和CP钢相变动力学模型,模型预测值与实测值吻合良好,充分验证了建模方法的有效性。(3)贝叶斯神经网络在热轧组织-性能预测中的应用基于贝叶斯方法的神经网络在网络训练目标函数中引入代表网络复杂程度的惩罚项,融入“奥克姆剪刀”理论防止网络“过训练”的发生,使网络具有较好的泛化性能。Bhadeshia和Mackey等将贝叶斯神经网络广泛的应用于钢材扭转、焊接、裂纹扩展等领域的建模,均获得了突出的效果。本论文将基于贝叶斯理论框架的神经网络程序化,首次将其应用于热轧生产现场化学成份-工艺-力学性能的关系建模,在网络的稳定性、收敛速度和泛化能力上均优于传统的BP神经网络,为热轧工艺优化设计提供了高精度的模型基础。(4)热轧工艺优化设计的实现以连续冷却相变模型为基础,结合单目标粒子群优化方法,实现了针对连续冷却过程的工艺优化设计。通过冷却速度的控制实现钢材内部铁素体、贝氏体、马氏体不同的相组成和铁素体晶粒尺寸。以贝叶斯神经网络模型为基础,结合多目标粒子群算法,实现了针对整个热轧过程的工艺优化设计。该方法能在较短的时间内搜索得到帕累托前沿,解决了在单目标优化中工艺窗口计算困难的问题,且计算精度较高。工业生产数据的验证表明,根据坯料的化学成份和现场的设备约束条件,可以通过工艺参数的优化达到客户提出的力学性能需求,这为柔性生产方式的实现提供了理论指导和技术支持。更多还原

【Abstract】 The design of TMCP (Thermal Mechanical Controlled Processing) parameters for hot rolled plate/strip is a time-consuming and exhausting research work. At the same time, because of the complexity of microstrual evolution and constraints of equipments in hot rolling process, it is difficult to determine the optimal TMCP parameters in traditional way. Therefore, development of new ways, which is to optimize the hot rolling process with the mechanical properties set up as the target, is of great necessity. The mathematical model for description of microstructure evolution and mechanical properties prediction, and the optimization of hot rolling process under the constraints of equipments to achieve the required mechanical properties constitute the basic concept of optimal design of hot rolling process. In the present paper, the physical-metallurgical theory and the artificial intelligence were applied in the modeling work for the microstructural evolution and the relationship between chemical compositions, process parameters and mechanical properties, accounting for the regularity and prediction precision. The optimal design of hot rolling process was finally realized by the employment of single-and multi-objective particle swarm optimization method. The chief original work of this paper is as follows.(1) Research work concerning the effectiveness of additivity rule applied to the modeling of phase transformation in the continuous cooling processThe dilatometric test was employed to measure the transformation kinetics of austenite to ferrite in four kinds of Nb bearing steels with different niobium contents. The theoretical method developed by Rios was for the first time applied to the modeling of phase transformation in the continuous cooling process. The relationship between lnln[1/(1-X)] and ln(|CR|) (CR-cooling rate) was plotted using the data at early stage of transformation, which successfully fitted a linear relationship to calculate the exponential values of n in the JMAK equation. The values of k in the JMAK equation obtained with the Rios method, however, have led to big discrepancies when the isothermal equations were used to predict the transformation kinetics during cooling. By assuming a Gaussian dependence of temperature, k was calculated by using an optimization method based on the rule of additivity and exhibited cooling rate correlation. The isothermal transformation model was used to predict the transformation kinetics during cooling, showing good agreements with the measured data. It has been proved that even though the rule of additivity has to be relaxed to take into account the effects of cooling rates, precise conversion between non-isothermal and isothermal kinetics can still be realized.(2) Modeling of phase transformation during continuous cooling process based on mathematical anlysisBased on the Rios’method, a new mathematical method designed to analyze the dilation data of phase transformation in the continuous cooling process was developed. The modeling work of phase transformation kinetics was solved mathematically. According to the preset of exponential n and temperature step, the parameter k at any moment during the transformation was calculated based on the inverse application of additivity rule. Relationship between k and temperature and transformed fraction dependence of k could be revealed by the comparison of k under different cooling rate. Combined with the Rios’method, the transformation model for TRIP and CP steels were developed. Good agreements between measured and predicted value of transformation kinetics have validated the effectiveness of the new method.(3) Application of Bayesian neural network in the microstructure and mechanical properties prediction for the hot rolling processThe punishment item, based on the "Occam’s razor" theory, was introduced into the objective function for the training of neural network in order to prevent the occurrence of "over-fitting" and improve the generalization ability. Extraodinary results have been obtained when the Bayesian neural network was applied to the modeling work in the field of hot torsion, welding and fatigue crack growth rate by Bhadeshia and Mackey. In the present work, the Bayesian neural network was programmed and used to model the relationship between chemical composition, process parameters and mechanical properties, wich has exhibited better stability, fast convergence rate and generalization ability when compared with the traditional BP neural network. It could provide model basis with high precision for the optimal design of hot rolling process.(4) The realization of optimal design of hot rolling processCombined with the single-objective particle swarm optimization method, the phase transformation model was used to realize the optimal design of continuous cooling process. Different ferrite grain size and combination of ferrite, bainite and martensite phases were obtained according to the control of cooling rate. And based on the Bayesian neural network, the multi-objective particle swarm optimization method was emplyed to build up the optimal design system for the whole hot rolling process. The process window was determined through the calculation of Pareto front with a high precision, which could be difficult for single-objective optimization. The production data has suggested that, the required mechanical properties can be achieved through the optimization of process parameters based on the chemical composition and constraints of equipment. It could provid theoretical guidance and technical support for flexible production mode.更多还原