Visual-based automated aircraft inspections for 3d skin damage

Abstract

Visual inspection is the common mean to detect damages on the aircraft skin. Standard maintenance programmes require engineers to perform frequent inspections that are costly, time consuming, hazardous and subject to human factors. Engineers are required to inspect all the areas, including crown, wings and vertical stabiliser, and personally evaluate the damage. Dents, in particular, are flaws that are challenging to detect and measure due to undefined boundaries, complex geometry and difficult access. Because of these characteristics, dents cannot be detected by monocular cameras and automation of their inspection has been lacking momentum, generally limited to manually operated 3D scanning tools. Moreover, no solution has been explored to replace the human judgement of the damage. The aim of this work is the design of an automatic system to inspect the aircraft skin, identify dents, measure and report them to the engineer, thus demonstrating the feasibility of such autonomous task via unmanned aerial vehicles. After reviewing the state of the art, data is acquired by means of a single-shot structured-light algorithm for 3D scanning based on Fourier transform profilometry and compatible with the use of un- manned aerial vehicles, yet delivering submillimetre accuracy. Machine learning is then considered for the autonomous identification of dents, implemented through a novel point cloud segmentation algorithm. Finally, a mathematical model is proposed to evaluate dent shapes, replacing the current reporting standards by allowing accurate and comparable dimensional evaluation. The three main contributions operate together to enable autonomous aircraft dent inspections, whose feasibility is demonstrated by experiments with a prototype system. This work paves the way for future automated systems capable to increase safety and advance aircraft inspection reliability, while reducing human workload, downtime and thus costs.