Autonomous systems are an essential step toward the industry of the future because they can optimize risky and repetitive tasks such as structural maintenance inspections and without risking human personnel. Micro-aerial vehicles (MAVs) are the ideal platform to reach elevated and confined spaces, but they require robust navigation capabilities. There are currently two main limitations to making structural inspections a mainstream technology: most of the cutting-edge inspection systems today are semi-autonomous and are vulnerable to degraded signals as they rely on Global Positioning System (GPS) for navigation. We propose a novel inspection system for industrial structures that overcomes the limitations of GPS-denied conditions by utilizing a combination of GPS and visual-inertial odometry (VIO), or exclusively VIO depending on the signal quality. A thermo-visual payload enables online multi-spectral image acquisition, and a customized convolutional neural network (CNN) architecture is trained using acquired data to intelligently detect rust and surface cracks through online inference. We explore a new paradigm for autonomous navigation inspired by computer-aided manufacturing (CAM) and additive manufacturing (AM), where the inspection path is planned out of the structure of interest’s computer-aided design (CAD) model. Our system benefits from exploratory mapping missions in case a CAD model is not available. We extensively validated autonomous navigation, structural damage detection, and three-dimensional (3D) map generation capabilities of our inspection system throughout various stages of the design process in laboratory-controlled and industrial production environments.
Reyes‐Munoz et al. (Tue,) studied this question.