Fast fringe analysis method using graphics processing unit acceleration for dynamic fault identification

Dynamic fault identification from fringe patterns is a challenging problem in optical metrology, and is required for applications such as non-invasive condition monitoring and fracture propagation . The paper addresses this problem by proposing a high speed technique for identifying temporally varying faults using graphics processing unit (GPU) accelerated Wigner-Ville distribution method. For this case, a huge stack of fringe patterns need to be processed and the parallel processing ability of GPU provides high computational efficiency and overall execution time improvement. For testing, we simulated a stack of 100 noisy fringe patterns containing time varying defects to mimic the temporal evolution of fracture in a test material. Each fringe pattern has size 4096 by 4096 pixels and signal to noise ratio of 5 dB, and thus the resulting image stack constitutes a large noisy data set. We demonstrate the performance of the proposed method for high speed detection of defects from the fringe patterns, and also show the comparative advantage of the GPU based parallel approach versus the conventional approach of sequential processing. For the given 16 megapixel image size, the sequential implementation using Python’s Numpy scientific library took about 38 minutes for processing a single fringe pattern whereas the same task could be completed within 4.5 minutes using the GPU based parallel implementation. Cumulatively, the reduction in computational cost for processing the complete fringe pattern data set would be substantial. Overall, our results show that the intensive and tedious task of dynamic fault detection can be efficiently processed using the proposed approach with high robustness against noise.