Virtual encoder: a two-dimension visual odometer for NDT

Abstract

Odometer information is an important feature of NDT systems for inspection procedures that involve mechanical scanning. Knowledge of transducer position during the inspection allows for the localization of flaws and the fusion (stitching) of several images into more sophisticated representations of the inspected object. Commercial encoders provide NDT systems with accurate realtime displacement information that can be integrated to obtain odometry. However, this information is typically limited to a single axis. Although composite schemes with more than one encoder can be built to provide 2-D or 3-D spatial information, they are mechanically intricate and lack flexibility and ease of use. We propose a 2-D position-tracking solution that is based on image processing. A miniature camera continuously captures images of the external surface of the inspected object, which are fed to an algorithm that detects and stores 2-D displacement between each pair of consecutive images. Additionally, the orientation quaternions provided by an Inertial Measurement Unit are stored, allowing for posterior 3-D path reconstruction over objects of known geometries, such as oil pipes. Besides logging position and orientation histories, the device also provides real-time displacement information to the NDT system, where it is perceived as a set of single-axis encoders, thus termed “the virtual encoder”. We demonstrate the applicability of the device to both contact and immersion ultrasonic inspections. The results show that the concept is promising, despite being based on simple principles and relatively easy to implement. The source code is provided as additional material in https://github.com/thiagokalid/VirtualEncoder-ECNDT-2023.