基于知识引导的工业机器人泛化性视觉系统研究与实现
发布时间：2018-11-25 06:49
【摘要】：随着现代科技的发展和自动化生产程度的提高,工业机器人在工程领域已获得了广泛的应用。在实际生产过程中,工业机器人所体现出来的高效性、高精度性和多功能性等特点是普通人力所不能比拟的。相信在不久的将来,工业机器人必将替代普通人力成为工业发展的主要生产力。目前,工业机器人的发展在自主学习和记忆以及柔性加工等方面依然存在一定的缺陷。针对这些缺陷,本文通过在传统算法的顶层引入知识工程,实现先验知识的共享、集成、推理和演绎,开发了机器人视觉知识系统(子系统),以提高机器人的学习、思维记忆和环境感知功能:在此基础上,进一步研发了一套基于知识引导的工业机器人泛化性视觉系统(主系统),用于实现机器人对目标对象的自动识别和精准定位。本文主要以缸体铸件为实验对象,对两套系统中涉及的相关理论和关键技术进行了深入研究,并对该视觉系统的可靠性和泛化性进行了验证。具体研究内容和研究结论总结如下:(1)基于人工神经网络具有非线性学习功能的特性,提出了一种可对机器人视觉知识系统中标定知识自动获取的新标定方法。通过定义手眼标定模型中符号和逻辑表达(隐式知识)的表示方法,可以实现手眼标定模型中隐式知识向显式知识的转变。借助该标定方法的使用,机器人视觉知识系统(子系统)可通过学习机制获取不同工作环境下的最佳标定模型,用于指导工业机器人泛化性视觉系统(主系统)的手眼标定过程,实现机器人对目标对象的精准定位。(2)为了提高图像分割算法在复杂工业图像中分割的自适应性和稳定性,提出了基于图像高层信息的改进型变分水平集分割模型和基于先验知识的自适应阈值分割模型。本文采用无参考图像质量评价分析方法,分析了不同工作环境下导致原始图像质量不稳定的原因和影响图像分割算法稳定性的主要干扰因素。在此基础上,提出了由图像信息能量项、惩罚项和高斯金字塔项共同组成的改进型变分水平集分割方法和基于峰值和灰度统计知识的自适应阈值分割算法。两种分割方法均能快速准确地从复杂背景中分割出包含定位基准的感兴趣区域,且当因环境光强改变造成工件成像质量发生变化时,该分割方法仍然能够准确地分割出感兴趣区域,两种方法的分割耗时分别约为1.3s和0.8s。上述分割方法的应用,可以进一步提高工业机器人泛化性视觉系统(主系统)的稳定性和泛化性能。(3)本文提出了一种基于形状知识的图像语义识别方法。首先,针对工件形状特征,构建了具有平移、旋转和缩放不变的内、外部形状特征图像描述子的形状描述数据库;其次,运用数据挖掘技术中的粗糙集算法对数据库中的数据进行属性约简和识别规则提取,以实现识别知识的自动获取,形成识别知识库;最后,将工件形状的语义信息作为识别规则的前项,将对应的形状描述子作为识别规则的结论,建立了工件形状的语义信息和形状特征图像描述子之间的映射。通过使用该图像语义识别方法,机器人视觉知识系统(子系统)可根据工件形状的语义信息,从识别知识库中自动获取对应的图像描述,从而指导工业机器人泛化性视觉系统(主系统)对目标对象的自动识别,运用上述方法可对已建模形状进行识别,其识别率可达100%。(4)基于Java SSH软件开发平台,通过将建立的标定知识库、事实库和形状识别知识库集成于MySQL数据库管理系统中,开发了基于Web的机器人视觉知识系统(子系统),以获取不同工作环境下的最佳标定模型和不同工件形状语义信息对应的识别知识信息,为工业机器人泛化性视觉系统(主系统)服务。(5)设计了工业机器人泛化性视觉系统的总体框架结构和各功能模块。基于VS2012+QT5.3软件开发平台,研发了一套基于知识引导的工业机器人泛化性视觉系统(主系统)。该主系统可以通过Socket通信方式与子系统进行对接。当用户在浏览器上通过子系统向服务器发送检测请求时,主系统即可根据用户权限获取相关的标定知识和识别知识,从而引导机器人实现对目标对象的自动识别和精准定位。本论文研究了基于知识引导的工业机器人泛化性视觉系统的关键技术,对基于人工神经网络的机器学习标定方法、复杂环境下的图像自动分割算法及基于形状知识的图像语义识别等方面进行了研究,实现了工业机器人对目标对象的自动识别和精准定位。该系统可有效提升工业生产线的柔性加工与连续作业的稳定性,并可实现各类知识的积累和共享,本文的研究成果可为今后工业机器人的智能化研究工作提供一定的理论依据。
[Abstract]:With the development of modern science and technology and the improvement of the degree of automatic production, the industrial robot has been widely used in the field of engineering. In the actual production process, the high efficiency, high precision and versatility embodied by the industrial robot are not comparable to the ordinary human resources. It is believed that in the near future, industrial robots will substitute for common human resources as the primary productivity of industrial development. At present, the development of industrial robot still has some defects in the aspects of autonomous learning and memory and flexible processing. In view of these defects, the paper introduces the knowledge engineering in the top layer of the traditional algorithm, realizes the sharing, integration, reasoning and deduction of the prior knowledge, and develops the robot vision knowledge system (sub-system) to improve the learning, thinking and environment-sensing functions of the robot: On this basis, a set of knowledge-guided industrial robot generalization vision system (main system) is further developed, which is used to realize the automatic recognition and accurate positioning of the target object by the robot. In this paper, the relevant theories and key technologies involved in the two systems are studied in detail, and the reliability and generalization of the visual system are verified. The specific research contents and conclusions are summarized as follows: (1) Based on the characteristic of the artificial neural network with the non-linear learning function, a new calibration method that can automatically acquire the calibration knowledge in the robot vision knowledge system is proposed. By defining the representation of the symbolic and logical expression (hidden knowledge) in the hand-eye calibration model, the transformation of the implicit knowledge to the explicit knowledge in the hand-eye calibration model can be realized. With the use of the calibration method, the robot vision knowledge system (subsystem) can acquire the optimal calibration model under different working environment through the learning mechanism, and is used for guiding the hand-eye calibration process of the industrial robot generalization vision system (main system), and the precise positioning of the target object is realized by the robot. (2) In order to improve the self-adaptability and stability of the segmentation of the image segmentation algorithm in the complex industrial image, an improved split-level set segmentation model based on high-level information of the image and a self-adaptive threshold segmentation model based on the prior knowledge are proposed. In this paper, a non-reference image quality evaluation and analysis method is used to analyze the causes of the instability of the original image quality under different working conditions and the main interference factors that affect the stability of the image segmentation algorithm. On this basis, an improved split-level set segmentation method, which is composed of image information energy terms, penalty terms and Gaussian pyramid terms, and an adaptive threshold segmentation algorithm based on peak and gray level statistical knowledge are proposed. in that method, the region of interest containing the positioning reference can be quickly and accurately segmented from the complex background, and when the image quality of the work piece is changed due to the change of the environment light intensity, the segmentation method can accurately segment the region of interest, The splitting time of the two methods is about 1. 3s and 0. 8s, respectively. The application of the segmentation method can further improve the stability and generalization performance of the industrial robot generalization vision system (main system). (3) An image semantic recognition method based on shape knowledge is presented in this paper. firstly, aiming at the shape characteristics of the workpiece, a shape description database of an inner and outer shape characteristic image description sub is constructed with a translation, rotation and scaling; secondly, the attribute reduction and the identification rule extraction are carried out on the data in the database by using the rough set algorithm in the data mining technology, so as to realize the automatic acquisition of the identification knowledge and form an identification knowledge base; and finally, the semantic information of the shape of the workpiece is taken as the preceding paragraph of the identification rule, and the corresponding shape description is taken as the conclusion of the identification rule, the mapping between the semantic information of the workpiece shape and the shape characteristic image description is established. by using the image semantic identification method, the robot vision knowledge system (sub-system) can automatically acquire the corresponding image description from the identification knowledge base according to the semantic information of the shape of the workpiece, so as to guide the automatic identification of the target object by an industrial robot generalization vision system (main system), and the method can identify the modeled shape, and the recognition rate can be up to 100 percent. (4) based on the Java SSH software development platform, a Web-based robot vision knowledge system (subsystem) is developed by integrating the established calibration knowledge base, fact base and shape identification knowledge base in the MySQL database management system, so as to obtain the identification knowledge information corresponding to the optimal calibration model and the different workpiece shape semantic information in different working environments, and serve the generalization visual system (main system) of the industrial robot. (5) The general frame structure and function modules of the generalized visual system of industrial robot are designed. Based on the software development platform of VSS2012 + QT5. 3, a set of knowledge-guided industrial robot generalization vision system (main system) is developed. The main system can interface with the sub-system in a socket communication mode. when a user sends a detection request to a server through a sub-system on a browser, the main system can acquire relevant calibration knowledge and identification knowledge according to the user authority so as to guide the robot to realize the automatic identification and accurate positioning of the target object. This paper studies the key technology of the generalization visual system of the industrial robot based on the knowledge guidance, and studies the machine learning and calibration method based on the artificial neural network, the image automatic segmentation algorithm in the complex environment and the image semantic recognition based on the shape knowledge. and realizes the automatic identification and precise positioning of the target object by the industrial robot. The system can effectively improve the stability of the flexible processing and continuous operation of the industrial production line, and can realize the accumulation and sharing of all kinds of knowledge. The research results in this paper can provide some theoretical basis for the intelligent research work of the industrial robot in the future.
【学位授予单位】：江苏大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.41;TP242.2
[Abstract]:With the development of modern science and technology and the improvement of the degree of automatic production, the industrial robot has been widely used in the field of engineering. In the actual production process, the high efficiency, high precision and versatility embodied by the industrial robot are not comparable to the ordinary human resources. It is believed that in the near future, industrial robots will substitute for common human resources as the primary productivity of industrial development. At present, the development of industrial robot still has some defects in the aspects of autonomous learning and memory and flexible processing. In view of these defects, the paper introduces the knowledge engineering in the top layer of the traditional algorithm, realizes the sharing, integration, reasoning and deduction of the prior knowledge, and develops the robot vision knowledge system (sub-system) to improve the learning, thinking and environment-sensing functions of the robot: On this basis, a set of knowledge-guided industrial robot generalization vision system (main system) is further developed, which is used to realize the automatic recognition and accurate positioning of the target object by the robot. In this paper, the relevant theories and key technologies involved in the two systems are studied in detail, and the reliability and generalization of the visual system are verified. The specific research contents and conclusions are summarized as follows: (1) Based on the characteristic of the artificial neural network with the non-linear learning function, a new calibration method that can automatically acquire the calibration knowledge in the robot vision knowledge system is proposed. By defining the representation of the symbolic and logical expression (hidden knowledge) in the hand-eye calibration model, the transformation of the implicit knowledge to the explicit knowledge in the hand-eye calibration model can be realized. With the use of the calibration method, the robot vision knowledge system (subsystem) can acquire the optimal calibration model under different working environment through the learning mechanism, and is used for guiding the hand-eye calibration process of the industrial robot generalization vision system (main system), and the precise positioning of the target object is realized by the robot. (2) In order to improve the self-adaptability and stability of the segmentation of the image segmentation algorithm in the complex industrial image, an improved split-level set segmentation model based on high-level information of the image and a self-adaptive threshold segmentation model based on the prior knowledge are proposed. In this paper, a non-reference image quality evaluation and analysis method is used to analyze the causes of the instability of the original image quality under different working conditions and the main interference factors that affect the stability of the image segmentation algorithm. On this basis, an improved split-level set segmentation method, which is composed of image information energy terms, penalty terms and Gaussian pyramid terms, and an adaptive threshold segmentation algorithm based on peak and gray level statistical knowledge are proposed. in that method, the region of interest containing the positioning reference can be quickly and accurately segmented from the complex background, and when the image quality of the work piece is changed due to the change of the environment light intensity, the segmentation method can accurately segment the region of interest, The splitting time of the two methods is about 1. 3s and 0. 8s, respectively. The application of the segmentation method can further improve the stability and generalization performance of the industrial robot generalization vision system (main system). (3) An image semantic recognition method based on shape knowledge is presented in this paper. firstly, aiming at the shape characteristics of the workpiece, a shape description database of an inner and outer shape characteristic image description sub is constructed with a translation, rotation and scaling; secondly, the attribute reduction and the identification rule extraction are carried out on the data in the database by using the rough set algorithm in the data mining technology, so as to realize the automatic acquisition of the identification knowledge and form an identification knowledge base; and finally, the semantic information of the shape of the workpiece is taken as the preceding paragraph of the identification rule, and the corresponding shape description is taken as the conclusion of the identification rule, the mapping between the semantic information of the workpiece shape and the shape characteristic image description is established. by using the image semantic identification method, the robot vision knowledge system (sub-system) can automatically acquire the corresponding image description from the identification knowledge base according to the semantic information of the shape of the workpiece, so as to guide the automatic identification of the target object by an industrial robot generalization vision system (main system), and the method can identify the modeled shape, and the recognition rate can be up to 100 percent. (4) based on the Java SSH software development platform, a Web-based robot vision knowledge system (subsystem) is developed by integrating the established calibration knowledge base, fact base and shape identification knowledge base in the MySQL database management system, so as to obtain the identification knowledge information corresponding to the optimal calibration model and the different workpiece shape semantic information in different working environments, and serve the generalization visual system (main system) of the industrial robot. (5) The general frame structure and function modules of the generalized visual system of industrial robot are designed. Based on the software development platform of VSS2012 + QT5. 3, a set of knowledge-guided industrial robot generalization vision system (main system) is developed. The main system can interface with the sub-system in a socket communication mode. when a user sends a detection request to a server through a sub-system on a browser, the main system can acquire relevant calibration knowledge and identification knowledge according to the user authority so as to guide the robot to realize the automatic identification and accurate positioning of the target object. This paper studies the key technology of the generalization visual system of the industrial robot based on the knowledge guidance, and studies the machine learning and calibration method based on the artificial neural network, the image automatic segmentation algorithm in the complex environment and the image semantic recognition based on the shape knowledge. and realizes the automatic identification and precise positioning of the target object by the industrial robot. The system can effectively improve the stability of the flexible processing and continuous operation of the industrial production line, and can realize the accumulation and sharing of all kinds of knowledge. The research results in this paper can provide some theoretical basis for the intelligent research work of the industrial robot in the future.
【学位授予单位】：江苏大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.41;TP242.2
【相似文献】
相关期刊论文 前10条
1 Mary Aichlmayr;罗雁;;视觉系统提升产品质量[J];饮料工业;2008年05期
2 刘辉;王瑜辉;尹周平;熊有伦;;一种新型的飞行视觉系统设计[J];光学技术;2007年02期
3 ;康耐视推出在线视觉系统评估工具[J];汽车工艺与材料;2009年03期
4 李俊岭;李岚;孙晓波;杨静;;视觉系统在超声打孔机中的应用[J];电子工艺技术;2009年02期
5 Sarah Sookman;陈廷炯;;智能视觉系统:质量控制盲点的终结者[J];软件;2009年09期
6 寿天德;;视觉系统中反馈投射的功能作用(英文)[J];Neuroscience Bulletin;2010年05期
7 田娜;庞师军;卫晓冲;;视觉系统在硅片检测上的应用[J];电子工艺技术;2010年05期
8 肖明怡;;激光视觉系统使汽车制造自动化[J];激光与光电子学进展;1985年04期
9 童永承;视觉系统电子技术的模拟探讨[J];湖北师范学院学报(自然科学版);1987年01期
10 童永承;视觉系统的自控阻断特性及模拟[J];湖北师范学院学报(自然科学版);1988年01期
相关会议论文 前10条
1 李文元;;昆虫复眼视觉系统的仿真研究[A];加入WTO和中国科技与可持续发展——挑战与机遇、责任和对策（下册）[C];2002年
2 陈勇军;;动态标志在现代视觉系统中的运用[A];节能环保 和谐发展——2007中国科协年会论文集（二）[C];2007年
3 王国建;陶霖密;叶喜勇;邸慧军;;视觉系统分布式计算结构模型的构建与实现[A];第六届和谐人机环境联合学术会议（HHME2010)、第19届全国多媒体学术会议（NCMT2010）、第6届全国人机交互学术会议（CHCI2010）、第5届全国普适计算学术会议（PCC2010）论文集[C];2010年
4 吴繁;潘炼;;工业机器人单目双视的应用研究[A];中国计量协会冶金分会2009年年会论文集[C];2009年
5 ;工业机器人[A];日本机械学会技术路线图[C];2008年
6 罗t，
本文编号：2355160
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/2355160.html
