Study On Multilevel And Multistage Humanoid Intelligent Control And Recognition

Although the study on the control and recognition of complex systems and their process control has become a trend for decades, intelligent control and recognition are still in the initial stage with concepts and theories under development. Intelligent control and recognition as a whole is an emerging academic area that still lacks of solid and systematic theoretical foundation and commonly agreed technology. There are still no unified description of the definitions, theories, methodologies, and structures. Nevertheless, it can be predicted that intelligent control and recognition will achieve further development and wide application with the advance in system theory, artificial intelligence and computer technology.The main contents of the dissertation are:1. Study on the Theory and Methodology of Intelligent Control and Recognition.Conclusion is achieved by analysis and discussion that the purpose of the study onintelligent control and recognition is not the controlled objects but the ways of the controller or recognizer to imitate the experience, activity and knowledge structure of experts, and the aim of recognition and modeling is not the mathematical models of the controlled objects but the dynamic characteristic model of the whole system, as well as the qualitative and quantitative knowledge model of the controller. Hierarchical control, expert system, pattern recognition and system theory are combined for the study on the theory and method of designing a multilevel and multistage intelligent control and recognition structure that imitates the modes of human's macroscopic thinking and microscopic action. A three-level-three-stage intelligent controller/recognizer is constructed, design specification and operational mechanism are drawn up, and the presentation and the acquisition of the knowledge model are defined.2. Design of Intelligent States ObserverWith the thoughts of combining the methods of three-level-three-stage intelligent control/recognition and classic observer, a discussion is made on the basic requirements of designing an intelligent state observer with the specification and examples developed. The observer is applied to an AC vector controlled system with re-structured load torque as feed-forward compensation and lead to satisfactory laboratory results. The prominent merit of this method is that the aim of the study is not the observer's mathematical models of the controlled objects but the qualitatively and quantitatively identified model of expert's knowledge. With the strategy of multi-mode open-looped observation, the problem of observation accuracy is solved in avoidance of the problems of poles-assignment and stability. The method might be extended to the state re-structure of various practical engineering objects such as the design of the observers of complex objects with time delay, tightly coupled variables, nonlinearity and time-variance.3. Detecting and Tracking Technique of Computer VisionWith the thoughts of combining the detection and recognition of supervised objects and Kalman's predictive tracking, a discussion is made on the recognition and control algorithm for the automatic supervision and tracking of mobile robots. Design examples and laboratory results are obtained. The prominent merit of this method is that the qualitatively and quantitatively characterized knowledge model of supervised moving objects can effectively adapt to environmental change and capture the moving objects without the need of background image. With the state prediction of moving object by Kalman filter and an incremental two-step tracking algorithm, fast and smooth driving control for the moving robot's object tracking can be realized.4. Recognition of Chinese Character ImagesAiming at imitating the process of human to cognize Chinese characters and by combining the design method of three-level-three-stage intelligent control/recognition with the theory of pattern recognition, a discussion is made on the basic requirement of the design method of intelligent Chinese character recognition. Design specification and examples are given and applied to the recognition of handwritten characters and the characters on vehicle's license-plate. The prominent merits of this method are that the qualitative and quantitative expert knowledge model based on the recognition of Chinese characters' integrated and decomposed property is constructed; the strategy of pattern recognition by the fusion of multiple properties is adopted; . The Chinese character samples library based on priori information is built up; and the fault-tolerant recognition of Chinese characters is realized.