Intelligent environment recognition and prediction for NDT inspection through autonomous climbing robot

This paper presents a novel approach to environment mapping prediction with focus on autonomous climbing robot to NDT (Non-Destructive Technique) inspection. In industrial installations, the inspection of non-planar surfaces requires that NDT probes passe on whole surface, while the autonomous robot navigates over an unknown environment based only on its perception abilities. However, the path planning of inspection is not a trivial task specially when there is no precise information about environment. In this work, a special kind of climbing robot is used to inspect large metallic surfaces such as spherical pressure vessels used to store Liquified Petroleum Gas (LPG). The robot has adherence skills that allow it to safely navigate through the internal and external surface of the vessel. As a result, robot mobility suffers from hard magnetic adhesion constraints. A new approach is proposed to environment detailed prediction, including specific characteristic (like weld beads and plates) of inspected surface. The goal is the automatic extraction of some environment characteristics to predict the storage tank dimensions and robot localization, based on a group of 3D perception sources (laser rangefinder, light detection and ranging and depth camera) mounted over a rolling platform to improve its reach. The environment prediction is carried out after the robot visually detects two or more weld beads corners. A multi-measuring environment is firstly build by Fuzzy data fusion of the different perception measurements allowing to estimate plates and weld beads based on design and safety standards. Virtual and real experiments are carried out to illustrate proposed method performance.