【作者】 凌玉华；

【导师】 杨欣荣； 桂卫华；

【作者基本信息】 中南大学 ， 模式识别与智能系统， 2010， 博士

【摘要】 铝电磁连续铸轧是近年来我国领衔开发的一种铝金属加工新技术,是一项集多学科、新技术于一体的探索性研究课题。在其生产过程中,能否获得细小且均匀的铝熔体晶粒是能否获得高性能铝材的关键因素。要实现铸轧过程的优化控制,关键是能够在线检测铝熔体的晶粒度。然而,目前的检测手段无法直接实现铝带坯晶粒度的在线实时检测,只能通过离线取样后进行金相分析的方法得到,因此存在很大的滞后,严重影响了铝电磁铸轧产品质量的闭环优化控制。铝电磁连续铸轧过程具有机理复杂、时变、大滞后等特点,难以建立其精确的数学模型。本文从分析铝电磁铸轧工艺机理和铝带坯晶粒度影响因素出发,研究铝带坯晶粒度的软测量及其复合磁场智能控制技术。提出了基于FCM聚类的智能集成模型的晶粒度软测量方案,以及模糊自抗扰技术的复合磁场智能控制策略,设计了基于铝带坯晶粒度软测量模型的复合磁场双闭环控制系统,有效地实现了铝带坯晶粒度的在线检测和复合磁场的智能控制,并据此对铝带坯晶粒度进行闭环控制,成为提高铝带坯产品质量和生产效率的关键。论文主要工作和研究成果体现在以下几个方面：(1)探讨了电磁场作用下铝电磁铸轧技术的工艺机理和复合磁场的形成原理,分析了行波磁场作用下铝熔体内电磁力的特性、铝熔体中电磁力的脉动特性及其分布规律,并对铸轧区复合磁场进行了数值模拟。讨论了电磁因素、铸轧设备因素和工艺因素对铝带坯晶粒度的影响,为晶粒度软测量的辅助变量选择奠定了基础。(2)针对标准FCM聚类算法对初始值敏感、存在局部极值的缺点,提出了一种基于样本点密度指标改进的FCM聚类算法,并对一种人工数据集和经典IRIS数据集进行了聚类仿真分析。结果表明,改进算法得到的初始聚类中心接近实际的中心点位置,从而减少了算法的迭代次数,显著提高了聚类效率,并在一定程度上抑制了算法陷入局部极值。(3)采用基于L-M算法的BP神经网络建立了晶粒度软测量模型,并进行了仿真研究。针对BPNN稳定性和泛化能力问题,利用PSO算法对BPNN的权值／阈值进行了优化。针对神经网络容易出现的局部极小和过拟合问题,研究了基于ε-SVR的晶粒度软测量建模方法,并利用PSO算法对ε-SVR的参数进行优化,解决了常规试凑法和网格法存在的参数非最优和运算量大等问题,减少了模型对数据样本的依赖性,提高了泛化能力。(4)为尽可能提高晶粒度软测量的精度,提出了基于FCM聚类的BPNN和ε-SVR多模型智能集成建模的晶粒度软测量方案。首先利用改进的FCM算法对晶粒度软测量数据样本进行聚类,然后根据各聚类样本数目选择其建模方法,并建立各子模型；最后采用模糊隶属度加权策略进行各子模型的信息融合,得到集成模型的输出。仿真结果表明,集成建模方法的训练精度和泛化能力明显提高。(5)针对电磁铸轧的关键技术,即复合磁场的智能控制问题,提出了一种新型、实用的模糊自抗扰控制器方案。详细介绍了模糊自抗扰控制技术的基本原理和电磁铸轧复合磁场模糊自抗扰控制器的设计方法,并通过仿真验证了采用模糊自抗扰控制器对复合磁场进行智能控制的可行性和正确性。(6)基于晶粒度软测量模型设计了复合磁场双闭环控制系统,研制开发了基于模糊自抗扰技术的复合磁场智能控制器和一套晶粒度软测量软件系统。系统通过试运行,取得了较好的带坯晶粒细化效果。该系统能够显著提高并稳定铝带坯的生产质量,所生产的铝带坯晶粒度指标达到了一级标准。铝带坯晶粒度在线检测和复合磁场控制问题,是目前国内外铝加工行业所面临的难题,论文所作工作具有重要的研究意义和很好的工业应用前景。更多还原

【Abstract】 The technology of aluminum electromagnetic roll-casting which was developed by China in recent years is a new kind of technology of metallic materials processing. It is an exploratory research subject that integrates multidisciplinary and new technologies. During its production process, whether high-performance aluminum can be gotten depend on whether fine and uniform grains can be obtained. The key to achieve optimal control of roll-casting process is that the grain size of aluminum melt can be detected online. However, the aluminum strip’s grain size can not be detected online and real-time with current detection methods, and it is detected through offline metallographic analysis by manual sampling. Therefore, there is great delay which affects aluminum products quality of the electromagnetic roll-casting seriously.It is difficult to establish accurate mathematical models for the process of aluminum electromagnetic roll-casting due to its charac-teristics of complicacy, time-varying, large delay and so on. Based on the analysis of mechanism of aluminum electromagnetic roll-casting and the affecting factors of aluminum strip’s grain size, the soft-measuring of aluminum strip’s grain size and intelligent control technology of composite magnetic field were researched in this paper. The intelligent integrated soft-measuring model of aluminum strip’s grain size based on FCM clustering and fuzzy active disturbance rejection control strategy for composite magnetic field were proposed. Based on soft-measuring model of aluminum strip’s grain size, a double closed loop control system of composite magnetic field was designed. The online measuring of aluminum strip’s grain size and the intelligent control of composite magnetic field are realized effectively. With these methods, the closed loop control of aluminum strip’s grain size was realized, which is the key to improve product quality and production efficiency. The main work and contributions in the paper are as follows:(1) The process mechanism of the aluminum electromagnetic roll-casting technology and the forming principle of composite magnetic field are discussed. The characteristics of the traveling electromagnetic force and the pulsating electromagnetic force in the molten aluminum are analyzed. In addition, the composite magnetic field of roll-casting area is analyzed by the numerical simulation method. Then, the roll-casting factors and electromagnetic factors of the influencing factors of grain size are described. Therefore, the foundation of the auxiliary variables in the grain size soft-measuring is established.(2) For the disadvantages of the standard FCM clustering algorithm, as the sensitivity to initial value and the existence of local extremum, such an improved FCM clustering algorithm is proposed based on the sample point density index. An artificial data set and the classic IRIS data sets are simulated and analyzed by using the algorithm. The results show that the improved algorithm has many advantages, including reduction of the iteration number, improvement of the clustering efficiency, and so on.(3) Based on the BPNN of using the L-M algorithm, the soft-measuring model of grain size is established and the simulation is researched. For the problem of BPNN’s stability and generalization, the parameters of weight and threshold of BPNN are optimized by the particle swarm optimization algorithm. For some disadvantages of BPNN, theε-SVR-based soft-measuring model of grain size is researched and the parameters are optimized by the particle swarm optimization algorithm. The simulation results show that the generalization of the model is enhanced highly.(4) To improve the precision of soft-measuring model, the intelligent integrated model is established based on BPNN, s-SVR and FCM clustering algorithm. Firstly, the data samples are clustered by the improved FCM clustering algorithm. Then, the different model for each cluster is selected in accordance with the number of its samples. Finally, using the fuzzy membership weighted strategy, the output of integrated model is calculated. The simulation results show that the training accuracy and the prediction accuracy are improved in the model.(5) For the key technology of electromagnetic roll-casting, the problem of intelligent control of composite magnetic field, a new style and practical controller based on fuzzy ADRC is proposed. The basic principles of fuzzy ADRC and the design methods of fuzzy ADRC for composite magnetic field were discussed in details in this paper. And the simulation results show that fuzzy ADRC for the composite magnetic field is feasible and correct.(6) Based on the soft-measuring model of grain size, the double closed loop control system of composite magnetic field is developed. A composite magnetic intelligent controller using the fuzzy ADRC and a software system of grain size soft-measuring are developed. Though the experimental debugging, the strip’s grain is refined and the grain size reaches the first-class level.The strip’s grain size’s online measuring and the control of composite magnetic field are the issues in the aluminum industry. These achievements have important research significance and good prospects in the industrial applications.更多还原