CUDA (compute unified device architecture) acceleration and keypoint tracking for aero-engine blade deformation measurement.

This paper presents a compute unified device architecture-accelerated visual method for aero-engine blade deformation measurement, addressing inaccurate checkerboard control-point localization and high digital image correlation latency. The approach employs Devernay-operator-based concentric circle edge extraction for precise binocular camera calibration and a parallel-optimized initial value search for real-time subpixel matching. Experiments using a binocular-stroboscopic system on a rotating blade show a maximum unidirectional displacement error of ≤2 μm. At 6000 rpm, peak deformations along the X, Y, and Z directions are 234, 258, and -193 μm, respectively, with a total maximum deformation of 486 μm.