Drone-based Automated Exterior Inspection of an Aircraft using Reinforcement Learning Technique

Visual inspection of aircraft exterior surfaces is one of the most common operations of aircraft maintenance for identifying possible defects such as dents, cracks, leaking, broken or missing parts, etc. Such an operation is not only time consuming as it requires long time preparation but also prone to human errors or injury when manually done. Therefore, automating the operation using a drone to acquire images of the aircraft exterior surface can increase inspection efficiency, effectiveness and safety, while leveraging the computer vision tools to perform anomaly detection from drone-gathered images and human inspector for unexpected or more complex problem diagnosis. This paper presents a drone-based inspection approach to automate visual inspection of a large aircraft in a static environment such as a maintenance, repair, and overhaul (MRO) shop. Based on a pre-existing digital model of the aircraft, a set of candidate viewpoints for fully covering the interested area of inspection is generated. Then, an optimal flight path in the sense of shortest flying distance for full coverage is computed using a reinforcement learning (RL) technique. Finally, the drone follows the computed path to acquire 2D image or 3D point cloud data of the aircraft exterior surface. The data can further be analyzed for detection of possible anomalies and damages. We carried out a digital experiment and the results showed that our approach can inspect 90% of the aircraft surface in one hour, which is much more efficient than manual inspection.