Algorithms Research And System Design For Vector Field Visualization

Computer simulations are playing an increasing role in understanding the world we live in. With the developments of high performance computing,scientific and engineering computations can generate data of unprecedented size and complexity. Analyzing and visualizing these large-scale scientific data present new challenges to the researches of scientific visualization. Furthermore,the rapid developments of the internet/intranet techniques and applications,Web-based visualization approaches do fulfill more usable visualization systems.Vector fields provide an important concept in science and engineering,which describe a wide variety of phenomena like fluid flow,electromagnetic fields,etc. Visual methods can help people to understand complex structures of vector fields. Combined with real application background,the algorithms and system for representing complex and large vector fields are presents in this thesis.1) A new texture-based 2D vector field visualization method is proposed:Streamline Texture Synthesis(STS). STS maps a ID texture to each streamline,and accelerate computing by rendering streamline thickly. STS can simulate animations and visualize 3D surface flows. STS produces substantially better visual results. STS is faster than LIC(Linear Integral Convolution)texture-based method.2) In order to overcome the LIC's shortcoming that only presents direction of the vector field,Multi-granularity LIC and image blend technique which can present the magnitude and direction at one time are proposed.3) The instruction-level parallel calculation of streamlines on 3D curvilinear grids has been implemented firstly by using the Streaming SIMD Extensions(SSE),which are a set of extensions of the Intel Pentium HI/4 processor. Compared with the conventional algorithm,SSE-based algorithm coded by vector class library enhances performance about 55%,and coded by inlined-assembly is about 75%.4) This thesis presents two PC-based streamline visualization methods for large unstructured grids:Memory-Mapped File Method(MMFM)and MultiThread Out-of-core Method(MTOM). MMFM constructs streamlines like in-core methods through mapping large data file into virtual RAM,and reduces page faults by organizing data in spatial blocks. MTOM implements out-of-core streamline construction with multithread technology. MTOM can overlap CPU and I/O,and can process in parallel with multiple CPUs. MMFM or MTOM can make streamline visualization on large unstructured grids with less memory,and has more efficient performance.5) This thesis introduces the process of error within the dataflow model and analyzes the error in the particle tracing algorithms for vector field visualization. With explicitRunge-Kutta methods for particle tracing,Richardson extrapolation and re-integration(with a smaller tolerance)are used to estimate global error of the tracing path. The methods of visualizing error are also described.6) A visualization prototype system for CFD,simulation of high-speed impact,and so on,is designed with Visualization toolkit(VTK).7) This thesis discusses the reference models for web-based visualization and has designed two Visualization Web Server prototype architectures:VRML visualization server and CORE A-based visualization server.