Traditionally, industrial inspection is a costly and dangerous job when performed by human inspectors. Manual inspection is prone to errors which can only be prevented through the implementation and enforcement of a comprehensive inspection methodology [1]. Inspections of infrastructure involved in dangerous industrial processes can also interrupt their operation to provide safety to inspectors. The cost of process interruption, combined with the costs of enforcing standard inspection methods prevent a large challenge to companies hoping to implement inspections of process critical infrastructure.
In order automate the industrial inspection process, mobile inspection robots are emerging to reliably carry out industrial inspections, even in conditions that would be harmful to a human operator. The rise of available robotic inspection platforms can be attributed to advances in multiple technologies. These include mobile motion control and motor drivers, as well as advanced systems for surface inspection and data collection. Understanding the changes in these technologies and the specific challenges in mobile robotics is instrumental in implementing a mobile robot for an industrial inspection application.
Mobile inspection robots must have a robust method of travel to reach interest points within an industrial site. This locomotion is commonly provided by electric motors. Traditionally, the technology used to drive electric motors has been used in precise motion control applications such as controlling CNCs within industrial positioning applications. While mobile inspection robots require powerful and precise motor control similar to common motion control processes, there are unique challenges that mobility presents.
Regeneration and dynamic inertial loads are both associated with mobile robots. [3][4] In order to provide a robust solution, any motor controller providing electric actuation for a mobile robot must be able to handle a higher current draw than a controller for equivalently sized motion control applications. To provide mobility, the footprint of all components on the robot [3] must be taken into consideration, and if possible be minimized. In order to achieve this, it is preferable for a controller to integrate multiple motor outputs on a single unit.
The control methods of the unit should model situations unique to mobile robots. These can be surface slippage, external changing forces due to gravity, or different and changing payload of the inspection platform. The algorithm the motor driver uses to drive the robots’ locomotion should take these into account when determining the output to a motor.
Finally, environments that are subject to inspection are commonly dirtier and more physically demanding than a controlled factory environment where most motion control applications can be found. This requires rugged design that will prove durable in messy industrial environments. Roboteq is a company that aims to provide controllers that meet these requirements, while other providers of mobile inspection robots have developed their own proprietary systems to accomplish this [5].
Actuation is not the only area of mobile robotics that has been advanced by technological research. In order to provide sufficient inspection, inspection engineers and roboticists are taking advantage of improvements in sensor data acquisition, resolution, and processing methods. Traditional use cases of industrial inspection involve monitoring parts coming off production lines for defects. This depends on a combination of computer vision algorithms and laser-based range finding techniques that provide a 3D map of a scannable area. New technology is enabling these techniques to be applied to a much larger surface area in large industrial structures.
New scanners used in industrial inspection platforms are able to continuously scan an area, expanding their application beyond their field of view. Mobile inspection robots use a combination of cameras and LIDAR to continuously generate point cloud data sets of their surroundings – sets of points that exist in 3 dimensions. As these sets are generated, techniques such as Iterative Closest Point are used to construct a 3D model of the robot’s surrounding environment. The Iterative Closest Point algorithm matches the position of two-point clouds by minimizing the distance between their points.
This method is often used in conjunction with a Kalman filter [7] to eliminate noise and produce a reliable result. The deployment of these algorithms on a mobile inspection robot have allowed for the sophisticated virtual reconstruction of its surrounding environment, which the robot can use to localize itself for navigation, as well as perform a quality inspection.
Combing the advances in motor controlling technology with new methods of processing scanned data has allowed mobile roboticists to move surface inspection and automation from the controlled area of a production line, to the unpredictable environments of large industrial processes. The oil and mining industries have used mobile inspection robots to monitor the condition of critical infrastructure, such as oil pipelines and pumps [8]. This technology also has similar applications in mining, public works, energy transmission, and any industry that relies on large infrastructure which cannot fail. Because of these advances, it is time to consider implementing mobile robot platforms for inspection purposes in your application and seeing just how much you can automate.
Sources:
[2]Undproductivity of Current inspection practices
http://ieeexplore.ieee.org/abstract/document/1266853/?reload=true
[3]Regeneration article:
https://www.roboteq.com/index.php/applications/100-how-to/160-understanding-regeneration
[5]GE BIKE Platform
https://inspection-robotics.com/bike-platform-2016/
[6] ICP paper
https://graphics.stanford.edu/~smr/ICP/comparison/chen-medioni-align-rob91.pdf
[7] Kalman filtering Book.
https://www.intechopen.com/books/smoothing-filtering-and-prediction-estimating-the-past-present-and-future
[8] Mobile inspection robots in the energy industry:
https://pdfs.semanticscholar.org/9f63/ed805e36d09868662f950b391a41b5a6b810.pdf
