FINITE ELEMENT OPTIMIZATION FOR NONDESTRUCTIVE EVALUATION ON A GRAPHICS PROCESSING UNIT FOR GROUND VEHICLE HULL INSPECTION

Shape reconstruction for nondestructive evaluation (NDE) of internal defects in ground vehicle hulls using eddy current probes provides a rationale for determination of when to withdraw vehicles from deployment. This process requires detailed finite element optimization and is computationally intensive. Traditional shared memory parallel systems, however, are prohibitively expensive and have limited central processing units (CPUs), making speedup limited. So parallelization has never been done. However, a CPU that is connected to graphics processing units (GPUs) with effective built-in shared memory provides a new opportunity. We implement the naturally parallel, genetic algorithm (GA) for synthesizing defect shapes on GPUs. Shapes are optimized to match exterior measurements, launching the parallel, executable GA kernel on hundreds of CUDA™ (Compute Unified Device Architecture) threads to establish the efficiencies.