Research On Generalization Of Industrial Robot Vision System Guided By Knowledge

With the development of modern science and technology, and the improvement of automation in industrial field, industrial robots have been widely used in engineering field. In the process of practical production, human is incomparable with industrial robot which has characteristics such as high efficiency, high precision and versatility. It is believed that in near future industrial robots will replace ordinary human and become the main productive forces to drive industrial development. At present, industrial robot is short of autonomous learning, memory function and flexible manufacturing. In order to overcome the above shortcomings, this dissertation introduced knowledge-based engineering technology into the traditional computer vision algorithm to realize the sharing, integration, reasoning and deductive of prior knowledge. This dissertation then developed knowledge-based system of robotic vision (subsystem) to raise learning ability, memory skills and environmental awareness. Meanwhile, the generalization vision system of industrial robot guided by knowledge (primary system) is also developed to identify and locate the target object automatically on the basis of the above theoretical research. In this dissertation, we mainly focus on the key technologies and related theories involved in two systems with cylinder block casting as the research object to carry out a series of research work to verify the reliability and applicability of the proposed generalization vision system. The main research contents and research conclusions can be summarized as follow:(1) This dissertation proposed a new calibration method which can realize automatic knowledge acquisition of knowledge-based system of robotic vision based on the nonlinear learning characteristics of artificial neural network. It redefined knowledge representation method of symbol and logical expression (implicit knowledge) in the hand-eye calibration model, which can realize the transformation from implicit knowledge to explicit knowledge. Based on the proposed calibration method, knowledge-based system of robotic vision (subsystem) can obtain the best calibration model under different working conditions through learning mechanism. This mechanism guides the hand-eye calibration process in the generalization vision system of industrial robot guided by knowledge (primary system) to precisely locate the target object.(2) To improve adaptability and stability of the image segmentation algorithm in complex industrial image, the modified variational level set segmentation model based on high-level information of image and adaptive threshold segmentation based on prior knowledge are put forward in this dissertation. This dissertation analyzed the cause of the unstable quality of original image and the main interference factors influenced the stability of image segmentation algorithm in different work environment by using no reference image quality evaluation method. Based on the above research,a modified variation level set segmentation method consisting of image information energy item,punishment item and the Gaussian pyramid item, also the adaptive threshold segmentation algorithm based on prior knowledge of peak and statistic intensity are put forward. The results are effective, quick and complete when segment original image with complex background and light intensity variation by using the proposed methods.It takes about 1.3s and 0.8s respectively to finish the automatic segementation for the proposed two methods. With the key role of the proposed segmentation methods, the stability and generalization performance of the generalization vision system of industrial robot guided by knowledge (primary system) are considerably improved.(3 ) This dissertation also proposed an image semantic recognition method based on shape knowledge which can transform the semantic information about shape of target object into computer language for the robot to recognize. First of all,methods for constructing the shape description database are introduced. The shape description database contains internal and external shape features descriptors which are invariant to image scaling, translation, and rotation; Secondly, using the rough set algorithm which belongs to data mining technology to obtain reduction attributes set and classification rules from the database. This algorithm can acquire knowledge and form a knowledge base automatically for the knowledge-based system of robotic vision (subsystem);Finally, set up mapping relationship between the semantic information about shape of target object and shape features descriptors by using production rules as knowledge representation method. The production rules are composed of the semantic information about shape of target object and the corresponding shape features descriptors. Using the above image semantic recognition method, the knowledge-based system of robotic vision (subsystem) can obtain the corresponding image description from the knowledge base automatically. The acquired semantic information about workpiece's shape will guide the generalization vision system of industrial robot guided by knowledge (primary system) to identify the target object automatically.(4) This dissertation again developed knowledge-based system of robotic vision(subsystem). This subsystem is a web-based system developed by Java SSH framework and MySQL database management system integrated with calibration knowledge base,extensional database and shape knowledge base. This technology can get the best calibration model under different working conditions and identification knowledge basing on semantic information of different artifacts' shape to serve the generalization vision system of industrial robot guided by knowledge (primary system). The experimental results showed that different modeled shapes could be recognized correctly and the accuracy could reach to 100% by using the above method.(5) This dissertation designed the overall framework and functional modules for the generalization vision system of industrial robot guided by knowledge. This primary system is developed by VS2012 + QT5.3 software and connected with the subsystem through the Socket communication .When a user send inspection request to the server on the browser through subsystem, the primary system accesses the relevant calibration knowledge and identification knowledge according to user's purview, so as to guide the robot to identify and locate target object automatically and accurately.This dissertation introduced the key technologies of the generalization vision system of industrial robot guided by knowledge, including the machine learning calibration method based on the artificial neural network, image automatic segmentation algorithm in complex environment and image semantic recognition based on shape knowledge. The above theoretical research.can help to realize automatic identification and accurate positioning of object for industrial robot. The developped prototype system can effectively promote the flexibility of industrial production line and the stability of continuous operation, and can realize the accumulation and sharing of knowledge. In conclusion, the research results of this dissertation can provide the theoretical reference for subsequent industry robot research being developed for intellectualization.