Abstract
This article presents an integrated concept of an aerial robot used for predictive maintenance in the construction sector. The latter can be remotely controlled, allowing the localization of cracks on wall surfaces and the adaptive deposit of the material for in situ repairs. The use of an aerial robot is motivated by fast intervention, allowing time and cost minimizing of overhead repairs without the need for scaffolding. It is composed of a flying mobile platform positioned in stationary mode to guide a soft continuum arm that allows to reach the area of cracks with different access points. Indeed, some constructions have complex geometries that present problems for access using rigid mechanical arms. The aerial robot uses visual sensors to automatically identify and localize cracks in walls, based on deep learning convolutional neural networks. A centerline representing the structural feature of the crack is computed. The soft continuum manipulator is used to guide the continuous deposit of the putty material to fill the microscopic crack. For this purpose, an inverse kinematic model-based control of the soft arm is developed, allowing to estimate the length of the bending tubes. The latter are then used as inputs for a neural network to predict the desired input pressure to bend the actuated soft tubes. A set of experiments was carried out on cracks located on flat and oblique surfaces, to evaluate the actual performances of the predictive maintenance mechatronic robot.