Abstract
Robotic inspection offers a robust, scalable, and flexible alternative to deploying fixed sensor networks or human
inspectors. While prior work has mostly focused on single robot inspections, this work studies the deployment of a swarm of
inspecting robots on a simplified surface of an in-orbit infrastructure. The robots look for points of mechanical failure and inspect
the surface by assessing propagating vibration signals. In particular, they measure the magnitude of acceleration they sense at
each location on the surface. Our choice for sensing and analyzing vibration signals is supported by the established position of
vibration analysis methods in industrial infrastructure health assessment. We perform simulation studies in Webots, a physics
based robotic simulator, and present a distributed inspection algorithm based on bio-inspired particle swarm optimization and
evolutionary algorithm niching techniques to collectively localize an a priori unknown number of mechanical failure points. To
perform the vibration analysis and obtain realistic acceleration data, we use the ANSYS multi-physics simulation software and
model mechanical failure points as vibration sources on the surface. We deploy a robot swarm comprising eight robots of 10-cm
size that use a bio-inspired inchworming locomotion pattern. The swarm is deployed on 2.5D (that is curved 2D) cylindrical
surfaces with and without obstacles to investigate the robustness of the algorithm in environments with varying geometric complexity. We study three performance metrics: (1) proximity of the localized sources to their ground truth locations, (2) time to
localize each source, and (3) time to finish the inspection task given an 80% surface coverage threshold. Our results show that
the robots accurately localize all the failure sources and reach the coverage threshold required to complete the inspection. This
work demonstrates the viability of deploying robot swarms for inspection of potentially complex 3D environments.
inspectors. While prior work has mostly focused on single robot inspections, this work studies the deployment of a swarm of
inspecting robots on a simplified surface of an in-orbit infrastructure. The robots look for points of mechanical failure and inspect
the surface by assessing propagating vibration signals. In particular, they measure the magnitude of acceleration they sense at
each location on the surface. Our choice for sensing and analyzing vibration signals is supported by the established position of
vibration analysis methods in industrial infrastructure health assessment. We perform simulation studies in Webots, a physics
based robotic simulator, and present a distributed inspection algorithm based on bio-inspired particle swarm optimization and
evolutionary algorithm niching techniques to collectively localize an a priori unknown number of mechanical failure points. To
perform the vibration analysis and obtain realistic acceleration data, we use the ANSYS multi-physics simulation software and
model mechanical failure points as vibration sources on the surface. We deploy a robot swarm comprising eight robots of 10-cm
size that use a bio-inspired inchworming locomotion pattern. The swarm is deployed on 2.5D (that is curved 2D) cylindrical
surfaces with and without obstacles to investigate the robustness of the algorithm in environments with varying geometric complexity. We study three performance metrics: (1) proximity of the localized sources to their ground truth locations, (2) time to
localize each source, and (3) time to finish the inspection task given an 80% surface coverage threshold. Our results show that
the robots accurately localize all the failure sources and reach the coverage threshold required to complete the inspection. This
work demonstrates the viability of deploying robot swarms for inspection of potentially complex 3D environments.
| Original language | English |
|---|---|
| Number of pages | 8 |
| Publication status | Submitted - Jan-2022 |