Outdoor Field Engineering Measurements, Analysis and Diagnostics Using GPU-based Laptop Computers: checking for cracks in Military Vehicles Damaged by Improvised Explosive Devices (IEDs).
Dr Paul R P Hoole
(from Facebok: Paul Hoole)
The Entertainment Industry is not infrequently ahead of the Engineering Industry or the Scientific Community. The Graphics Processing Unit (GPU) found much of its develo0ment impetus coming from Digital Games, for instance in Play station games, where both two-dimensional (2-D) and three-dimensional (3-D) graphical processing in games require billions of pixels to be operated on in a second, with hundreds of game operations to be performed on each pixel. The GPU (a parallel processor) can be far superior in performing computationally demanding operations which can be parallelized compared to the conventional Central Processing Unit (CPU, a sequential processor), though still using high level languages such as C++.
What is also fascinating is that the mathematical Solvers that are used in engineering and scientific computations may also need rethinking when switching from CPU to GPU (or CPU+GPU) systems. Example: the well-known Gauss-Seidel method (where in each iteration n+1, values of variables in the nth and completed n+1th values are used) which is more efficient than the Gauss-method when using a CPU becomes less efficient than the Gauss method in GPU, since the Gauss method may be readily parallelized, since in the n+1th iteration only values from the nth iteration are used. Similarly the Genetic Algorithm (GA) readily lends itself to parallelization.
When military vehicles are damages in the theater of war, or internal cracks appear in advanced near-shore all-steel patrol ships and submarines, it is expensive and time consuming to bring them into a test station to have it reviewed for war-worthiness. Hence the need is to take the analysis and diagnostics test equipment on to the field – which often means Laptop based measurement, analysis and diagnostics systems. Laptops mean limited amount of hardware and memory, as well as the need to keep power consumption low. In GPU technology research itself, in order to keep power consumption low, seeks to reduce the traffic between GPU and memory thus the need for bandwidth reducing high level algorithms.
To test whether steel vehicles that have been damaged by explosives are still reliable to be used, cracks, crack size and shape and location need to be identified and assessed. The Eddy Current probe generates magnetic field at low frequencies and the surrounding fields are measured and then intensive computational work needs to be performed on a finite element or finite different grid with thousands of pixels or nodes that are cast over the area being analyzed. Not only have the structure and fields to be modeled, but mathematical solvers are used to determine material properties and geometrical shapes of the material around any hidden cracks. The CUDA (Computer Unified Device Architecture) architecture uses a combination of CPUs (for serial executions) and GPU (parallel portion executions) using both types of processors.
Applications of GPU will grow in the fields of electrical, communication, safety, radar and medical systems as researchers forge the way ahead.