[2D4] Use of digital image correlation and machine learning for the optimal strain placement in a full-scale composite tidal turbine blade
J McLoughlin, M Munko, M Valdivia Camacho, F Cuthill and S Lopez Dubon
University of Edinburgh, UK
One of the challenges testing and health monitoring of large structures represents is obtaining as much information as possible from a specimen with a limited number of sensors. In this work, a data-driven approach was pursued to decide the optimal location of single-point strain gauges using machine learning algorithms (MLAs) and information from digital image correlation (DIC) measurements. The optimal strain gauge placement was computed for a range of sensor numbers and the presence of sensors in the high-gradient regions was identified. Strain maps of almost 40,000 measurements were reconstructed successfully with fewer than twenty measured values using the method employed. However, certain loss of image contrast was identified, which is likely to have resulted from the treatment of non-numerical values.
