Parallel Visualization Of Multi-Physics Coupling Field

Multiphysics simulation can reveal the multi-physics process coupling inside a complex system in field of science and engineering,showing the new physical phenomena which cannot be obtained only by single physics simulation.At present,with the rapid increase of the peak speed of supercomputers,many real-world problems can be solved by multiphysics simulation,and the demand for the visualization of multi-physics coupling field has emerged in several important application domains,such as climate change and weather forecasting,electromagnetic environments,fusion and fission energy.However,current multi-physics field visualization faces two bottlenecks.Firstly,it is difficult to visualize the overlapping multi-physics fields because of the low rendering precision of the translucent occlusion objects.Secondly,the interactive rendering performance of large-scale multi-field data is relatively slow,because multiphysics simulation can produce gigabytes to terabytes of data.In addition,the parallel visualization cannot handle the rendering problem of concave,disconnected,and non-manifold regions,which will not only produce incorrect rendering results,but also affect the interactive rendering performance.In order to help domain scientists understand the physical phenomena clearly and accurately,this paper introduces a parallel visualization method for multi-physics coupling field.It can be used to solve the problems both in the multiple field fusion and the large-scale data visualization.The main contributions of this paper include the following:· A multi-channel fusion method for multi-field data visualization is proposed to solve the visual representation problem of the complex physical phenomena.The multi-field data is properly encoded based on the combination of multiple visual channels,so that the domain scientists can correctly decode the physical phenomena contained in multi-physics fields.The method can be used not only to achieve the visual coding for multi-field data,but also to extract the characteristics of the complex field through data analysis,which can effectively reduce the information overload in the visual channel.Secondly,multiple visual channels can separate opaque and transparent objects,thus reducing the number of objects that need to be sorted during translucent rendering,so that the rendering precision can be improved.Finally,multi-channel fusion is performed based on three-dimensional screen space,which can resolve the accuracy problem within the rendering of overlapping multi-physics fields.· A hybrid parallel rendering method for multi-physics coupling field is proposed to solve the interactive rendering performance of large-scale multi-field data.In particular,the different parallel rendering mode is used in different visual channel.Contrast the parallel rendering method in single pass,the hybrid approach can not only improve the performance of large-scale data by take advantage of the data locality in geometric channel,but also can solve the parallel rendering problem of complex regions within multiple cores in optical channel by sorting the sampling points on screen space.At the same time,because the multi-channel fusion tasks take place in the three-dimensional uniform sampling space,our method is not only suitable for CPU based parallel rendering method,but also suitable for GPU based parallel rendering method,which can satisfy the needs of different multiphysics applications.The experimental tests are performed in the typical applications such as climate change and weather forecasting,electromagnetic environments,fusion energy,and fission energy.Based on the parallel visualization method for multi-physics coupling field presented in this paper,the nuclear reactor physicists can understand and analyze the physical process of a full-core high-resolution virtual reactor for the first time.The experts in the field of electromagnetic simulation can explore the electromagnetic problem in real world with the help of multi-physics coupling field visualization.The performance results show that rendering time of our method can be less than 1 second when using GPU based parallel rendering method for 1 billion structured grid data.When visualizing 2 billion unstructured grid data,our CPU based parallel rendering method can reduce the required processor cores from hundreds of cores to 24 cores,and the rendering time can be reduced to 100 seconds.Our method can also perform in-situ visualization of 8 billion structured grid data during simulation time on supercomputer,which can solve the ultra-scale visualization problem.The experimental results show that the proposed method can satisfy the visualization needs of multi-physics coupling field in the typical applications.