Graphical Processing Unit-Accelerated Open-Source Particle Image Velocimetry Software for High Performance Computing Systems

Abstract

Particle image velocimetry (PIV) data processing time can constrain data set size and limit the types of statistical analyses performed. General purpose graphics processing unit (GPGPU) computing can accelerate PIV data processing allowing for larger datasets and accompanying higher order statistical analyses. However, this has not been widespread likely due to limited accessibility to the GPU-PIV hardware and software. Most GPU-PIV software is platform dependent and proprietary, which restricts the computing systems that can be used and makes the details of the algorithm unknown. This work highlights the development of an open-source, cross-platform, GPU-accelerated, PIV algorithm. Validation of the algorithm is done using both synthetic and experimental images. The algorithm was found to accurately resolve the time-averaged flow, instantaneous velocity fluctuations, and vortices. All data processing was done on a GPU supercomputing cluster and notably outperformed the central processing unit version of the software by a factor of 175. The algorithm is freely available and included in the OpenPIV distribution.

PIV Visualization of Flow Around Airfoil Controlled by Synthetic Jet Actuators at Shear-Layer and Wake Instability Frequencies

“…During the experiments, 500 image pairs were captured by each camera for each control case and phase angle. The PIV images were processed by a GPU-based, iterative correlation algorithm developed in part by the authors and the OpenPIV project [10,11]. A resolution of 0.34 mm between vectors was attained using 8 × 8 correlation windows.…”

DPIVSoft-OpenCL: A multicore CPU–GPU accelerated open-source code for 2D Particle Image Velocimetry

“…The first test case will be the (theoretical) spatial wavelength response of the method as originally introduced by Scarano and Riethmuller [25] and later used by other authors as a benchmark [13]. This test allows to evaluate the frequency response of the DPIVSoft-OpenCL code.…”

“…One can use the PyCUDA API to interface the OpenPIV Python source code with a GPU [12]. The same OpenPIV code has been successfully adapted to the use of multiple NVIDIA GPUs [13]. Recently, other massively parallel optical flow methods have been developed that allow PIV processing which produce errors comparable to the aforementioned programs [14].…”

An Enhanced Python-Based Open-Source Particle Image Velocimetry Software for Use with Central Processing Units

“…The growing significance of PIV over the past few decades has led to the emergence of numerous software packages. Among the available open-source packages are [8][9][10], e.g., PIVLab [11], OpenPIV [12], Fluere [13], Fluidimage [14], mpiv [15], JPIV [16], and UVMAT [17]. The advantages of open-source software development include the complete availability of algorithm details, which provides greater flexibility in future developments, especially in the context of community-driven collaboration, and compatibility with highperformance computing systems [8].…”

“…Among the available open-source packages are [8][9][10], e.g., PIVLab [11], OpenPIV [12], Fluere [13], Fluidimage [14], mpiv [15], JPIV [16], and UVMAT [17]. The advantages of open-source software development include the complete availability of algorithm details, which provides greater flexibility in future developments, especially in the context of community-driven collaboration, and compatibility with highperformance computing systems [8]. In this regard, PIVLab and OpenPIV have proven to be popular with the research community with the former being one of Matlab ® 's most popular nonofficial free toolboxes [10].…”