Real Time Sub Assembly Identification through Data Fusion of IMU with Vision Sensor for an Inspection System
Indira Gandhi Center for Atomic Research
This work has been accepted for publication in the International Journal for Nuclear Energy, Science and Technology, with the citation -
Thirumalaesh Ashokkumar, N.A. Nibarkavi, S. Joesph Winston, Joel Jose, Rathika P D (in press), Real time sub-assembly identification through IMU data fusion with vision sensor for an inspection system, International Journal for Nuclear Energy, Science and Technology
Problem Description
The PFBR reactor core is made up of a hexagonal lattice, as shown. The inspection of core internals for abnormalities or foreign objects that could compromise the structural integrity is essential.
The Reactor Core Viewing System in Room Temperature (RCVS-RT) is a system that has been developed to aid in this process for the test setup. RCVS-RT is designed to drive a vision probe through the core top and then into the extended sub-assembly.
However,
While the vision probe driven through any (N,H) subassembly, and rotated, it's orientation is unknown.
This implies on detection of any abnormality, the (N,H) of the neighboring sub-assembly which has the abnormality is unknown
Project overview
This work explores fusing noncontact sensors to a vision sensor to achieve orientation recognition of the RCVS-RT, thereby helping in identifying the neighboring sub-assembly that is defective.
Key achievements
Develop an algorithm that exploits the hexagonal lattice structure to output the number tag of the neighboring assembly given the current sub-assembly number tag and orientation
Use of data from an IMU, to detect the orientation of the camera view
Fuse the algorithm along with the data, to sucessfully detect the numbering tag of the sub-assembly with the potential defect
Algorithm for the Number Tag - How?
The algorithm exploits the hexagonal structure, categorising the sub-assemblies into phasal, edge and exceptions. Using this classification, for any given number tag, the neighboring tags can be identified.
Algorithm for the Number Tag - The Result?
The algorithm's classification of the core is shown in the image, along with the flowchart
Now that we have the number tags for all the neighboring sub-assemblies, we can figure out the number tag of the video feed using the IMU(Magnetometer) data, shown below
Project outcome
Fusing the number tag identifiction algorithm with Magnetometer data, the operator can identify the sub-assembly towards which the camera is pointed at. Some Examples are shown here. The algorithm has been tested for robustness through multiple test cases as illustrated below
