Noise Reduction in Flexible-Array-Inspection Images with Machine Learning for Aerospace Applications

Advancements in additive manufacturing and computer-aided design have resulted in the production of highly-complex mechanical structures. Within the aerospace industry, this can increase aeronautical efficiency whilst reducing production costs. However, the market is heavily regulated, and these complex parts require regular inspection to meet strict safety standards. While designed for optimal performance, these parts are challenging for ultrasonic inspection via conventional probe technology. The low-profile and fully-flexible “Novosound Kelpie” has been demonstrated to conform effectively to complex geometries like those found on turbine blades and other aircraft components. Although, intricate surface changes still give rise to low signal to noise ratios (SNR) due to sound being scattered from within the part under test. In this work, a turbine blade was inspected using a 64-element, 20 MHz linear ultrasound array and images were recorded using a conventional flaw detector. The ultrasound data showed that it is possible, through contact measurement, to resolve back-wall reflections across a dynamic test surface with varying contours and wall thickness. Novel machine-learning (ML) algorithms were subsequently used to improve SNR and reconstruct a 3D-representation of the turbine blade, from data extracted.