Industrial chimneys must be inspected regularly, which causes production downtime and poses dangers to human inspection personnel. Multicopters can carry inspection sensors, such as cameras, but their manual control inside chimneys or outside close to the inspected surface is not feasible. In the project, we will equip a multicopter with multimodal sensors and develop autonomous control to perform inside and outside inspection of chimneys.
Chimneys and smoke pipes in industrial plants have to be monitored regularly, e.g., every two years in Germany. Today, these tasks are performed by human experts in baskets of cranes or by climbing frameworks. Sometimes, industrial climbers are taking photos to be analysed by experts on the ground. Chimney inspection poses risks. In some cases, humans are injured by accidents or even die because of dangerous climbing to access points of interest.
The preparation and installation of means for accessing a chimney cost time and money; nevertheless, depending on the situation, some areas are still out of reach. Multicopters are suitable to carry cameras and other sensors and can hover in the air, making the inspection of otherwise hard to reach places possible. Their manual control inside high and narrow chimneys—at defined distance from the wall—is not feasible due to low visibility caused by dust, the distance between the pilot and the copter, the winds, and the flight in the vicinity of obstacles.
The goal is to realize the inside and outside inspection of chimneys with an autonomous multicopter. Starting from a coarse initial model, a detailed 3D model will be built by registering and aggregating measurements of a lightweight 3D laser scanner which is carried by the copter. Registration of the most recent measurements of the 3D laser scanner with the model will yield a localization estimate. Additional sensors will provide attitude and height information. Due to the rotational symmetry of chimneys and the uniform surfaces, height and yaw will be ambiguous. To overcome this difficulty, a local reference system will be developed based on a pattern of lights on the chimney floor.
Based on the pose and velocity estimates and the 3D chimney model, autonomous navigation and obstacle avoidance will be developed to carry out user-defined observation missions of the chimney surface.
The copter will be equipped with high-resolution RGB-D sensors to capture the surface of the chimney. By registering the recorded RGB-D data, a detailed 3D model of the chimney surface will be created and visualized for the inspection by an expert. Deviations from nominal surface properties will be detected and marked for closer inspection by the expert.
The developed system will be able to cope with a variety of chimney cross sections and heights. A domain expert, who is neither a trained copter pilot nor a robotics expert, will be able to operate it. The use of the autonomous copter will 1) diminish the risk for human inspectors, 2) guarantee financial yield, and 3) deliver high quality semantically classified inspection results. This will revolutionize chimney inspection and open up new markets for end users. The developed methods will be general enough to be applicable in related use-cases.