【作者】 张成梁；

【导师】 冯显英；

【作者基本信息】 山东大学 ， 机械电子工程， 2011， 博士

【摘要】 中国的棉花加工现状相对于世界先进国家有很大差距。当前我国棉花加工行业对不同回潮率、含杂率的籽棉采用单一的轧花模式,或者由操作人员仅凭经验现场手动调整,缺少在线检测和智能控制环节,属于粗放型加工产业。棉花加工技术的落后致使不同品级的籽棉混级、混轧现象严重,降低了皮棉品级,造成了巨大的资源浪费与经济损失。棉花加工过程中影响棉花加工质量的因素众多,涵盖面较广,控制过程复杂,针对加工过程中自动化、智能化方面的不足,本文围绕回潮率在线检测及控制、颜色特征信息提取、杂质自动识别分类、加工设备对棉花性状参数的影响规律、重点设备数控化设计、皮棉质量评价方法及工艺智能优化等关键技术进行了研究。主要工作如下：回潮率是影响棉花加工及安全存储的一个关键因素,研究了棉花回潮率的在线检测方法及控制技术。提出一种基于相对湿度的棉花回潮率在线检测方法,并利用改进BP神经网络建立温度、相对湿度与棉花回潮率的关系模型。为了实现加工过程回潮率的精确控制,建立了籽棉烘干模糊PID控制模型,保证了适于棉花加工的回潮率条件。根据棉包储存理论建立皮棉加湿模糊PID控制模型,实现了皮棉加湿量的精确控制。棉花图像反映了其颜色特征和杂质信息,通过使用彩色CCD工业相机构建了棉花图像在线提取方案。在XYZ空间分析棉花的颜色特征,计算了反射率和黄度,为棉花品级的评定提供了条件。由于机采棉杂质种类较多以及棉花加工设备清杂侧重点不同,提出了基于杂质颜色与形状特征的棉花杂质分类识别统计方法。对于杂质颜色特征方面,首先对彩色图像在HSI空间采用矢量中值滤波,然后通过改进的模糊C均值聚类算法(FCM)对图像进行分割。改进的FCM能够自适应调整初始聚类中心及聚类数目,有效结合了棉花图像的经验知识与FCM的自适应推理功能,并且该算法在迭代过程中采用改进的欧式距离衡量样本点与聚类中心的颜色差别,此方法符合色调、饱和度对视觉的影响随强度O变化的规律。通过杂质图形区域的面积、圆度、形态复杂度、矩形度、伸长度,提取棉花杂质的形状特征,建立基于颜色与形状特征的棉花杂质识别BP神经网络模型,得到了棉花杂质的类型及其含量,为工艺路线的精确优化提供了依据。研究了棉花加工关键设备的清杂机理,针对棉花性状参数与设备工艺参数之间复杂的非线性映射关系,建立了清杂BP神经网络模型。在籽棉清理工艺阶段,分别建立倾斜式籽清、提净式籽清、回收式籽清及喂花轧花BP模型,分析刺钉滚筒转速、刺钉滚筒与格条栅间隙、提净滚筒转速、锯齿滚筒转速及籽棉产量等对籽棉性状参数的影响,并利用正交实验法验证各模型的正确性。针对倾斜式籽清机的结构缺陷,提出了刺钉滚筒与格条栅间隙自动调节的数控化设计方案。在皮棉清理工艺阶段,分别建立锯齿式皮清、气流式皮清BP神经网络模型,分析锯齿滚筒转速、锯齿滚筒与排杂刀间隙、皮棉产量、排杂刀数量及缝隙宽度等对皮棉性状参数的影响,并利用正交实验法验证各模型的正确性。提出了锯齿式皮清机排杂刀数量自动控制方案和锯齿滚筒与排杂刀间隙自动调节方案。为实现棉花加工工艺的智能优化,提出了基于神经网络-遗传算法(BP-GA)的工艺参数优化策略。采用基于BP模型串联设备的棉花性状参数控制方法,为GA算法提供了参数变量空间。棉花加工工艺优化是一个多目标多变量非线性优化问题,利用线性加权和法将该问题转换为基于收益最大化的单目标多变量优化问题,为遗传算法提供了适应度评价函数。棉花质量是评价函数中一个重要的参数,棉花质量依赖于其品级。通过纤维成熟程度、色特征及轧工质量建立了品级评判BP模型。对于遗传算法,提出基于基因组的混合实数编码方法,使得多层遗传操作得以简化；针对加工工艺要求,根据轧花后皮棉中不含重杂的约束条件,使用惩罚函数将其与适应度评价函数建立关联；结合相似度比较、适应度排序、最优保留策略和小范围竞争的思想,提出基于适应度排序的改进遗传算法,该算法具有较强的全局搜索能力及较快的收敛速度。通过本文的工作,提出了棉花加工过程在线检测技术和工艺参数自适应优化策略,该方法能够根据付轧籽棉的回潮率和性状参数等自动优化加工方案,并确定各设备工艺参数,为棉花加工行业实现真正意义上的“因花配车”和精细化作业提供了解决方案。更多还原

【Abstract】 There’s a giant gap between China’s cotton processing and that of the advanced countries of the world. Currently, a single pattern is employed to seed cotton ginning regardless of moisture regain and various impurity rates; or operators adjust manually by rules of thumb, lack of both on-line detection and intelligent control, which makes cotton processing still in an extensive form. Because of the poor techniques in cotton processing, different grades of seed cotton are ginned on a mixed-level, affecting lint quality, and leading to a great waste of resources and economic loss.There are a variety of factors influencing cotton quality in cotton processing, and control process is complex. In view of insufficient automation and intellectualization of cotton processing, the dissertation carries out research into on-line detection and control techniques of moisture regain, color feature withdrawing, automatic classification of impurities, the laws of influence of processing equipment on the parameters of cotton traits, numerical control design of key equipment, lint quality evaluation approach, and process intelligent optimization. The main research contents of the dissertation are as follows:Because moisture regain is a key factor of cotton processing and secure storage, research on on-line detection and control technology of moisture regain is conducted. On-line detection method based on relative humidity was proposed. Relation model among temperature, relative humidity and moisture regain was established by improved BP neural network. In order to realize accurate control over moisture regain in processing, fuzzy PID control model of seed cotton drying was built, guaranteeing moisture regain suitable for cotton processing. According to cotton bale storage theories, the fuzzy PID control model of lint humidification was set up to reach accurate control over lint humidity.Cotton images reflect color features and impurity information of cotton. A program of extracting cotton images on-line based on color CCD cameras was launched. Through analysis of cotton color features reflectance Rd and yellowness +b were calculated in XYZ space, laying foundations for grading cotton quality. Each piece of cotton processing equipment focused differently in terms of cotton cleaning, so based on the impurity color and shape feature a classified statistics approach of cotton impurity was adopted. For impurity color features, vector median filtering was applied to color images in color space HIS firstly, and then through improved fuzzy C-means image segmentation was conducted. Improved fuzzy C-means adaptively adjusted initial clustering centers and clustering numbers, which drew on both the experience of the cotton images and the adaptive reasoning function effectively. In the iterative process, the fuzzy C-means clustering algorithm adopted improved Euclidean distance to measure color differences between sample points and cluster centers, which was consistent with the pattern that the impact of color saturation on vision varies with I. The shape features of cotton impurities were withdrawn by the area, roundness, complexity, rectangle degree and elongation of impurity graphic region. Based on color and shape features BP neural network model of cotton impurity identification was built to obtain the types of cotton impurities and the respective amount of different types, which laid foundations for accurate optimization of process route.Cleaning mechanism of key processing equipment in cotton processing was studied, and BP neural network model for cleaning was established due to complex non-linear mapping relations between cotton trait parameters and process parameters. At the stage of seed cotton cleaning, BP models were set up respectively in inclined seed cotton cleaners, stripper and stick cleaners, inclined and recovery seed cotton cleaners and saw gin stands. The influence on the parameters of seed cotton traits exercised by the rotational speed of barbed nail rollers, the clearance between barbed nail rollers and lattice grates, the rotational speed of defecation rollers and that of sawtooth rollers and the yield of seed cotton was analyzed. The orthogonal experimental method was used to verify the correctness of each model. Due to the structural defects of inclined seed cotton cleaners, numerical control design scheme of automatic adjustment of the clearance between barbed nail rollers and lattice grates was brought forward.At the lint cleaning stage, BP neural network models were respectively established in saw lint cleaners, flow-through air lint cleaners. Besides, the influence on the parameters of lint traits exerted by the rotational speed of saw type rollers, the clearance between grid bars and saw type rollers, the numbers of grid bars, the slit width and the yield of lint was analyzed. The orthogonal experimental method was employed to verify the correctness of each model. A program of automatic control of the numbers of grid bars in saw lint cleaners and that of automatic adjustment of the clearance between saw type rollers and grid bars were put forward.To achieve intelligent optimization of cotton processing technology, parameter optimization strategies based on BP-GA were proposed. A method of controlling the parameters of cotton traits of series equipment was established based on BP model which provided the genetic algorithm with parametric variables space. Cotton processing optimization is multi-objective, multi-variable and non-linear, which can be converted to single-targeted and multi-variable optimization based on maximizing revenue by the linear weighted sum, providing the fitness evaluation function for the genetic algorithm. Cotton quality, which highly depends on its grade, is a primary parameter of evaluation function. BP model of grade evaluation was set up through cotton fiber maturity, color characteristics and rolling quality. As to genetic algorithm, the genome-based mixed-real-coded method was proposed, making multi-layer genetic manipulation simplified. In order to meet the requirements of processing technology, according to the constraint condition that there is no flotation in ginned cotton after ginning, the penalty function was used to associate this constraint condition with the fitness evaluation function. An improved genetic algorithm was brought forward based on the ideas of similarity comparison, fitness sorting, optimal retention strategies and small-scale competition. The algorithm had strong global search ability and fast convergence speed.On-line detection techniques and adaptive optimization strategies of parameters are proposed in the dissertation. The method can optimize automatically processing programs based on moisture regain and trait parameters of seed cotton, and also determine the parameters of each piece of equipment. The method provides a solution to fine processing and helps to realize "adjusting according to cotton traits".更多还原