Numerical simulation of electromagnetic flowmeter on GPU

摘要

The role of numerical simulations in new device development and optimization sensor properties is steadily increasing. Last years we observe dramatic increase of computational power of GPU based chips with double point precision. With using CUDA architecture process of adapting old numeric code for modern parallel architecture is greatly simplified. In contrast to the modern universal CPU chips graphics accelerators are designed for parallel computing with a large number of arithmetic operations. Steadily increasing number of transistors on GPU works for its intended purpose - processing of data sets. As a result, the basis for the effective use of power of the GPU in scientific is the parallel algorithms for hundreds of execution units available in video chips. In addition, the use of the multiple GPUs greatly increases the processing power of the system at relatively low cost.Authors developed a method of describing the induced electromagnetic field in the channel makes it possible to evaluate the influence of various factors: the extent and heterogeneity of the magnetic field, the spatial distribution of the field coil, the conductivity of the wall and fluid, the thickness of the pipe on the sensitivity of the flowmeter and the linearity of its properties. Previously, similar problems were solved analytically by setting the magnetic field distribution approximation of functions, for example, infinitely extended and homogeneous, or decaying exponentially, or a Fourier series with idealized coefficients, etc. Then obtained approximate theoretical results were refined through pilot studies breadboard and prototype devices. The increasing availability of high-performance resources can get more accurate results in less time. In this paper we use graphics cards for the parallel calculation of the magnetic field given by approximate functions. The accuracy of calculations depends on the amount of these approximate functions. They can be calculated independently of each other, which is very well suited for the parallel architecture of modern graphics accelerators..