Designing Semi-autonomous Networks of Miniature Robots for Inspection of Bridges and Other Large Infrastructures
Visual identification of structural flaws is quite valuable not only to predict an imminent collapse of a bridge, but also to determine effective precautionary measures and repairs.
Statement of objectives: In this project, we are pursuing a three-year basic research program to establish new design and performance analysis principles, and technologies for the creation of a semi-autonomous network of small mobile robots to aid visual inspection of civil infrastructure. This network will aid a human surveyor to remotely and routinely inspect structure areas such as a typical girder assemblage that supports the decks of a suspension bridge. Methods to be used: The goals mentioned above are being addressed via a multidisciplinary basic research effort in hardware, algorithm design and performance analysis. In order to achieve this goal, our team includes one researcher (CTO of Resensys LLC) in the area of bridge monitoring, and 3 faculty from 2 departments. Our team’s expertise covers all the key basic research areas of the proposal.
Broader Impacts: A) The activities funded by this grant will assist in the visual monitoring and guide the maintenance of bridges, which will improve public safety and reduce cost. B) The technological outcomes of this grant will be applicable to other infrastructure, such as tunnels and buildings. C) Proposed educational activities include STEM activities that will help attract and retain young talent to engineering. Here, we will prioritize the inclusion of underrepresented students. D) This grant will have a major impact not only in fostering multidisciplinary research among the PIs and beyond, but it may also lead to further investments by the University of Maryland (UMD) to promote research and education on CPS. E) This grant will promote a solid collaboration between the UMD, Resensys and the Maryland State Highway Administration. F) Our team is formed by a junior entrepreneur and faculty at the associate level whose research programs will be significantly impacted by this grant.
Summary of Technical Approach: We are investigating fundamental principles and theories on algorithm design, performance evaluation, electroadhesion, dynamic locomotion and system integration. The following is a summary of our current and recent research:
1. To move around on complex bridge environments, a new robot has been designed and constructed. A key part of the robot design has been adding compliance to the robot for traversal across rough surfaces (e.g. rivets) and to allow for transition between horizontal and vertical surfaces. The chassis is 3D printed on an Objet Connex with multiple materials to create viscoelastic joints that help reduce vibrations and provide smooth traversal on rough surfaces. Adhesive properties between the robot and bridge are currently simulated with magnetic wheels, and electronics include a 32-bit 48 MHz ATSAMD21 chip, radio, motor drivers, a camera, and additional sensors. The electronics are programmable through the Arduino IDE making these robots accessible to undergraduate researchers. To move beyond ferrous surfaces, the last year has focused on understanding how electrostatic force and polymer adhesion work together to affect electroadhesion. Electrostatic force was adapted into the JKR framework and demonstrated that the critical energy release rate can be a superposition of the polymer’s adhesive property and electrostatic force. Work has also been done to apply these theories into design of textured electroadhesives made of arrays of spherical bumps. These arrays of bumps take advantage of contact splitting, increasing electrostatic force linearly. Adhesion can also be further improved by developing textured devices with multi-tiered contact areas that can be engaged either through mechanical load or electrostatic load. Additional work will be done to explore different designs of these multi-tiered areas to maximum the adhesion range and how to incorporate interdigitated electrodes.
2. We also initiated the development of a robot that can move along non-magnetic vertical surfaces. It uses a powerful fan to press it against the surface with enough force to allow steering using four independently controller wheels. A raspberry pi 3 is used as onboard computer, which is tasked with control of the fan, wheels, communication modules and a camera. The ultimate goal is to use this robot in a semi-autonomous mode for infrastructure inspection.
3. We studied the problem of planning the deployments of (mobile) robots for bridge inspection. The robots are assumed to be initially stationed at multiple depots placed throughout the bridge. The problem is formulated as a min-max cycle cover problem in which the vertex set consists of the sites to be inspected and robot depots, and the weight of an edge captures either (i) the amount of time needed to travel from one end vertex to the other vertex or (ii) the necessary energy consumption for the travel. In the first case, the objective function is the total inspection time, whereas in the latter case, it is the maximum energy consumption among the deployed robots. We proposed a novel approximation algorithm with an approximation ratio of 4+(d-1)/k where d is the number of base stations, and k is the number of robots. In addition, the computational complexity of the proposed algorithm is shown to be O(n3), where n is the number of vertices.
4. We formulated a new class of problems in which a human operator assists a semi-autonomous inspection system. New tasks arrive to a queue (according to a statistical model), which are eventually assigned to a human operator who can service them within a fixed time interval, with a certain probability that depends on the workload level. We adopt a model in which the workload level is likely to increase when a new task is assigned and is expected to decrease after a rest period. We are investigating policies to decide when to assign a new task to the operator or allow for a rest period, such as maximizing the rate at which tasks are successfully completed. We show that given any policy that depends on the queue length and workload state, there is another policy whose performance is no worse that assigns tasks based solely on a threshold on the workload state. Analysis and design of optimal policies can be greatly simplified by considering the aforementioned class of threshold policies. We have also solved the case in which there are multiple types of tasks and similar considerations on operator workload. Here, we discovered a new class of optimal policies that is adaptive and does not require knowledge of the arrival rates of each task type. We also have preliminary results on the new problem of, given a desired service rate, determining policies that minimize workload according to various linear metrics.
5. We consider a remote sensor with energy harvesting ability (such as a solar panel) that monitors an LTI system. The sensor is equipped with a battery and communicates with the base station by sending wireless packets. The packet-drop probability is affected by the battery level which is influenced by the history of the current and past energy harvest and transmission request. A controlled Markov chain models this dependence and a given function of the battery level governs the packet-drop probability. Every randomized stationary transmission policy is specified by a function that determines the probability of a transmission request in terms of the battery level. Two theorems provide necessary and sufficient conditions for the existence of a randomized stationary policy that stabilizes the estimation error, in the second-moment sense. We also show that it suffices to search for deterministic stabilizing policies and identify an important case in which the search can be further narrowed to threshold policies.
6. We considered a common surveillance scenario in which a vehicle is tasked with hovering at a desired position that is inside a region where on-board localization methods are not available. Localization within the denied region is costly and can only be provided by another entity when requested by the agent. In our formulation, the state of the agent evolves according to a Markov decision process (MDP) in discrete-time and, while actuation effort is considered cost-free, no control is applied in the absence of a measurement. At each time-step, the agent has the authority to decide whether to wander like a random walk for one more step, or to request a measurement and use it to move towards the reference point. We show that the analysis of this problem simplifies by recasting it as a renewal reward process, for which the maximum wait time between the most recent renewal and a measurement request parametrizes all policies. Our analysis concerning wait-time optimization enabled us to establish conditions under which any local minimum (if it exists) is also global, thus facilitating the search for a minimizer. Our results are discussed for the cases in which the agent's loci the integers or a finite-dimensional Euclidean space.
7. The research team at Resensys deployed SenSpot sensors on different highway bridges in. The sensors monitor steel girder, piers, and bearings of bridges. In particular, the devices monitor thermal strain applied to bridges in order to determine correct operation of expansion joints on bridges. For this purpose, data analysis methods using linear regression cross power spectral density of strain readings versus temperature readings are used to detect change in mechanical properties of the members which are being monitored. Change is such characteristics will be used to determine likelihood of structural issues.
8. Research team at Resensys has also developed solar powered communication gateways for the purpose of sending command to SenSpot sensors and mobile robots. The gateways use energy saving schemes to provide reliable and uninterrupted operation for long periods of time, even if there is little access to sunlight for up to four weeks. The gateways use cellular data service to send data from robots and the SenSpot sensors to a cloud-based server.