Special automatic spraying system for civil aircraft parts based on visual recognition

[Objective] For a long time, manual operation is the only option for spraying aircraft parts. Uncertainty in a manual operation leads to problems including uneven paint film thickness, low spraying efficiency, and significant paint waste. More than 50 000 types of parts are available for various civil aircraft. With a wide variety, no fixed placement, and no clamping device, all these problems are far from being solved using existing robotic systems. To target the problems, an automatic spraying system for civil aircraft parts based on visual recognition is developed. [Methods] The system combines a two-dimensional (2D) camera and a three-dimensional (3D) camera to scan the spraying platform to realize the recognition and pose matching of parts. The 2D camera is used to obtain a 2D image of the parts on the spraying platform. Parts are preliminarily located and classified using the images, thereby reducing the interference of the environment on the point cloud. The 3D camera simply scans the partial space marked using the 2D camera, which effectively shortens the time compared to scanning the entire spraying platform space. The viewpoint feature histogram (VFH) descriptor is used to describe the point cloud features to recognize the type of parts. The k-dimensional tree (kd-tree) is used to establish the feature index. The search performance is much better than the global nearest neighbor search method. To solve the problem that the particle swarm optimization (PSO) algorithm is easy to fall into local optimal extremum, an improved PSO algorithm is proposed, where stagnation coefficient and the concept of decoupling are combined with the proposed algorithm. The spraying trajectory of the robot is planned, including the path of the spray gun and the trajectory of the robotic arm joints. The spray gun path planning means planning the path that the spray gun takes in the Cartesian space. Spraying paths are planned using the surface slicing method to realize the full coverage spraying of random poses and complex surface parts. Joint trajectory planning means planning the angle trajectory of each joint of the robot. The Bézier curve is used to plan the joint space trajectory to ensure the robot operated smoothly. [Results] The improved PSO algorithm performed better than the traditional PSO algorithm in convergence speed and accuracy. With an average accuracy of 2.11 mm, the automated point cloud registration for complex parts was completed in 240 s. Spray paths were simulated with Robot Studio, and simulations verified the method's effectiveness. [Conclusions] The spraying issue of numerous civil aircraft parts at arbitrary placements is resolved using the robot automatic spraying system based on visual recognition described in this study. Work efficiency and spraying quality of the system meet the production requirements. It is also found that the ambient light interferes with the work of segmenting parts in 2D images during the experiment. Manual adjustment of parameters cannot produce good results for complex image analysis tasks with high noise and shadow, so an adaptive parameter method should be proposed. The greedy algorithm is used for spraying path combinations, and the paths' global optimization method needs to be improved.