Research On Motion Modeling And Behavior Control Of Virtual Human In Maintenance Simulation

With the development of computer simulation, artificial intelligence and robotics technology, virtual human-based maintenance simulation becomes to be a feasible way. A general way for virtual human application is to substitute virtual prototype for physical mockup, and substitute virtual human for real personnel. In this way, virtual maintenance scenery can be constructed, and disassembly or assembly activities can be animated.There are two methods to introduce virtual human into the virtual environment:"human-in-loop"method and"full virtual"method. Current virtual reality hardware, such as data glove or data suit, are still not good enough to satisfy all requirements during human-computer interaction, especially on aspect of precision. Thus, it is very hard to popularize the"human-in-loop"method. To ensure the effect of maintenance simulation, discussion is focused on how to construct a virtual human represented and controlled by computer programs.A Virtual Human's Action Database (VHAD) based on virtual human technologies is designed and realized in this thesis. Some key issues are researched detailedly, such as maintenance task decomposition, maintenance action classification, parameterized modeling and encapsulation of maintenance actions, key frame extraction of captured motion data, and ground locomotion planning of virtual human.The main contributions and innovations are as following:1) A new task decomposition model is proposed, which can decompose any maintenance task into task units, action units and motion elements. Based on virtual human's kinematics model, an action classification criterion is provided, which is especially suited for maintenance simulation. Based on parametric modeling and encapsulation of maintenance actions, a Virtual Human's Action Database (VHAD) with a hierarchical architecture is designed and developed. Aiming at the reusability difficulty of captured motion data, a motion retargeting method based on path transformation is developed to reuse captured data, and several kinds of periodical ground locomotion actions are realized.2) In motion data compression, a key frame extraction algorithm based on kernel clustering method is proposed. Using Mercer kernel functions, motion data is mapped from the original space to the high-dimensional feature space, where the clustering process can be realized more efficiently. 3) A Petri Net-based locomotion-planning model is proposed to realize virtual human's locomotion planning and walking control in dynamic environment. Firstly, an original optimal path is planned out based on C-Space method. When walking toward goal position, local obstacle-avoiding walking direction is calculated and updated based on real-time collision detection. The Petri Net is used to coordinate the planning and obstacle-avoiding behaviors, and control the human figure moving from the original position to the goal position without any collision.4) To accelerate the generation of maintenance animation, a Virtual Maintenance Simulation System (VMSS) is designed and implemented based on the VHAD. To realize automatic metabolic energy expenditure calculation and fatigue degree evaluation, a human fatigue analysis method based on maintenance action simulation is developed.Based on the work above, a demonstration experiment about replacement of spaceship battery is designed and realized. Experiment results show that the VMSS system can be used to accelerate animation creation of maintenance task, improve interactive level of human-computer interface, and bring up virtual human's intelligence degree. More importantly, working efficiency of software users can be improved 5 to 7 times higher than before. The system has been successfully used in maintenance simulation and ergonomics analysis for virtual design of some digital prototype. It will also lay sold foundation for the future development of a virtual human's simulation platform.