【作者】 陈佳佳；

【导师】 丁永生；

【作者基本信息】 东华大学 ， 控制理论与控制工程， 2013， 博士

【摘要】 碳纤维是上个世纪五十年代末异军突起的一种新型的工程材料,它具有理想的工程特性,比如：高强度、高模量,被广泛应用于各行各业,成为一个国家军事、工业和经济等方面的重要影响因素之一。由于我国的碳纤维生产起步较晚,与先进国家生产的碳纤维在性能上具有较大的差距,对其生产过程优化开展研究,具有深远的意义。基于碳纤维纺丝生产线本身流程长,生产环境复杂,影响因素众多,各个环节之间的生产参数又相互耦合,将协同机制及生物机理引入其中,采用人工智能算法对碳纤维的纺丝过程进行建模与优化,对碳纤维的实时生产监控,提高碳纤维产品性能,可以提供相应的指导作用,具有一定的参考价值。本论文的具体贡献归纳如下：(1)以聚丙烯腈基碳纤维的六级牵伸过程为例,通过对搜集整理到的实验数据进行曲线拟合,研究了具有代表性的若干产品性能与牵伸参数之间的相互关系,给出了一个可供参考的碳纤维牵伸过程数学模型；设计了一种具有协同机制、克隆选择和非一致性变异三个主要操作的协同免疫克隆选择算法来解决碳纤维六级牵伸牵伸比分配这个多目标优化问题；由本论文所提算法获得的优化结果可以对碳纤维的实际生产提供一定的参考价值,所提算法也同样可以运用到类似的多目标优化问题中去。(2)以聚丙烯腈基碳纤维的原丝预氧化过程为例,建立了一个基于父代保留策略的免疫遗传加强BP神经网络预测模型,从预氧化过程的生产参数即可以提前评估出碳纤维预氧丝的性能。基于现有的实验数据,通过与已有的碳纤维预氧丝性能预测模型进行比较,结果表明,我们所提出的模型在性能预测方面具有更好的精度,并且能够更加快速的给出预测值,是一个更加可靠的预测模型,能够更好地为碳纤维原丝的预氧化过程提供相应的指导与参考。(3)我们提出了一个基于遗传-加强粒子群混合算法的神经网络(GA-IPSO-RNN)的聚丙烯腈基碳纤维纺丝生产过程双向优化模型。我们先用最近邻聚类算法确定径向基函数神经网络的隐含层所包含的神经元个数,然后提出一种遗传-加强粒子群算法来调整径向基函数神经网络的各个参数：作用函数及权重。通过这个模型,我们一方面给出了一种碳纤维产品性能预测的方法；一方面为新型碳纤维产品的生产方案设计提供了参考。这种新的基于遗传-加强粒子群混合算法引入到径向基函数神经网络中去,它们的结合使用,在一定程度上,推动了人工智能的发展；通过所设计的碳纤维纺丝生产过程双向优化模型,我们可以初步实现对碳纤维纺丝生产过程的在线监控与控制,实时预测碳纤维的产品质量,及时修正碳纤维生产的参数,避免造成较大的经济损失；同样的,基于设计的碳纤维纺丝生产过程双向优化模型,可以在新型碳纤维产品投出生产之前,提供一种或若干种基本符合我们所期望的产品性能的参考生产方案,为实际生产提供一些指导和帮助,防止直接投入大规模生产而引起的时间及金钱上的浪费。最后,对全文的研究内容进行了总结,对其中的不足进行了讨论,对下一步的研究方向提出了建议。更多还原

【Abstract】 Caron fiber is a new engineering material, it has ideal engineering characteristics such as high strength and high modulus. Owning to its superiority, it has been widely used in all works of life and influence national military, industry and economic deeply. Since carbon fiber production starts late in China, there is a large gap on carbon fiber performance between our motherland and advanced countries, in that case, it is significative to focus our attention on the optimization of carbon fiber manufacture process. The production line of carbon fiber is long, it contains several key sub-processes, each sub-process has its own control parameters, and these parameters couples each other. So complex production situation and so many factors make us considering that, introduce collaborative mechanism and biological mechanism into our research, adopt artificial intelligence algorithms to modeling and optimize the spinning process of carbon fiber, give out some reference and guides to real-time monitoring the carbon fiber production and improve the performance of carbon fiber product. The specific contributions of this paper summarized as follows:(1)Take the six steps drawing process of carbon fiber production as an example, based on existing experiments data, we investigate the relationship between representative production parameters and product properties indices, built a mathematical model of the carbon fiber drawing process; design a synergistic immune clonal selection algorithm which includes synergy mechanism, clonal selection and nonuniformty mutation, this algorithm can solve the multi-objective optimization problem, give out a six steps drawing ratio distribution scheme. This optimal result can provide reference to the actual production of carbon fiber, and the algorithm also can be applied to similar multi-objective optimization problem. (2)Take the pre-oxidation process of polyacrylonitrile carbon fiber an a example, set up a immune genetic neural network model based on father keeping scheme, we can estimate the performance of carbon fiber during production process by this model. Based on experiments data, we compare our prediction model to existing model, results show that, our model has better prediction precision, and convergent more quickly, it is a reliable performance prediction model which can provide guidance and reference for pre-oxidation process of carbon fiber better.(3)In this paper, we put forward a genetic-improved particle swarm optimization based neural network (GA-IPSO-RNN) bi-directional optimization model for carbon fiber spinning process. First of all, we use nearest neighbor clustering algorithm to determine the hidden layer nodes of neural network; secondly, we propose a genetic-improved particle swarm optimization algorithm to tune all the parameters of the neural network, which means activate function and weights between layers. For one direction, we can predict properties of carbon fiber by this model; for another direction, we can provide a design tool of new type carbon fiber. Combining this novel GA-IPSO algorithm into neural network plays a role to develop artificial intelligence; base on the bi-directional optimization model of carbon fiber spinning process, we can attempt to realize online monitor and control the manufacture line, real-time predict quality of carbon fiber, adjust the produce parameters in time, avoid to cause serious economical loss; similarly, this model can offer one or several production schemes which approximate meet our expect product properties, provide some guidance and help to actual production, prevent into mass production directly and lead to waste of time and money.At the end, we summary the full text of the research, discuss the deficiency of our investigate results, and propose future research direction.更多还原