Inspection-while-flying: An autonomous contact-based nondestructive test using UAV-tools

Abstract

Abstract Today's modern cities depend on basic services such as water and sewer systems including tunnels, which are supported by a pervasive infrastructure. The structural condition of these places is naturally prone to deterioration. Therefore, regular inspections are required for early detection of damage and material failure. It is envisaged that robotic systems may provide an up-to-date solution for this inspection task. Considering the challenging conditions in the surrounding environments, unmanned aerial vehicles (UAVs) may be a potential candidate to meet the need for both purely visual (contactless) and subsurface (contact-based) inspection functionalities. When the interaction with an uncooperative target is required, the UAV needs to be aware of the forces arising during the contact phase. This stage brings additional challenges including a variable center of gravity as well as the moment of inertias due to the shift in the attached parts, sliding on the target because of the surface characteristics and reaching desired interaction levels out of the equilibrium point without crashing the dedicated sensors while creating counter forces (reactions) in a desired manner. At the same time, ultrasonic inspection requires uninterrupted contact in certain force ranges. In order to address these factors in a compact and feasible manner, an optimization algorithm consisting of nonlinear moving horizon estimation (NMHE), which is a part of nonlinear model predictive control (NMPC) is proposed. In this algorithm, the baseline model of the UAV is augmented by the external forces where uncertainties, modeling mismatches and disturbances are lumped. Therefore, the NMHE has estimated these values in an online manner. In a simultaneous fashion, the identified external forces are fed into the NMPC to physically interact with the ceiling during the contact phase. The experimental inspection data is collected while autonomously staying in predefined constraint limits. It is observed that the complete inspection data is obtained and streamed via Bluetooth when the force on the ultrasonic tool reached 5.5 N level. The external force update coming from the estimation allows the proposed approach to reach these interaction levels.