Research
CURRENT PROJECTS
Vulnerability Characterization of Learning-Based Controllers in Networked Cyber-Physical Systems
Intelligent cyber-physical systems (CPS) represent a symbiotic integration of physical systems, sensors, actuators, and learning-based intelligent controllers through communication networks. These systems are increasingly prevalent in diverse applications, including smart grids, robotic swarms, and autonomous vehicles. While learning-based controllers are used to upgrade the capabilities of CPS, providing numerous benefits, the introduction of a learning component adds an additional layer of security challenges, which adversaries can exploit via cyber attacks. This project strives to uncover the characteristics and effects of information patterns that can deceive an intelligent decision-making agent or a learning-based controller, manipulating it into taking biased and unsafe actions. These findings should enable trustworthy secure-by-design solutions for developing real-time learning-based controllers suitable for safety-critical CPS. The research outcomes have direct applicability in remote sensing, smart infrastructure, and robotics, reinforcing the overall safety and reliability of these crucial CPS. The project aligns with efforts to promote inclusivity in computing, workforce development, and education.
The overall objective of this collaborative research is to model, characterize, and synthesize adversarial information patterns to manipulate the performance feedback (rewards) that is fundamental to the learning process of controllers and to develop schemes to detect them reliably.
The primary goals of the collaborative project are to develop a) a real-time reward manipulation scheme for learning-based controllers, b) multi-level attack schemes on reward signals in a distributed control architecture for CPS, and c) data-enabled strategies for their detection. The scientific merit of the project is to gain insight into the information patterns that can stealthily manipulate learning-based controllers in uncertain CPS to increase control costs and threaten their stability. The reward manipulation, from an attacker's perspective, may be formulated as a dynamic-constrained optimization problem. An online approximate solution will be developed to determine the optimal perturbation that can be added to the reward signal by an adversary. The optimization problem will be extended to address multi-level attacks using multiplayer Nash games. From a defender's perspective, attack detection and isolation methods using time-series analysis and perturbation theory will be developed. This research will equip learning-based control schemes with built-in resiliency from their design phase. The success of this research will advance control-theoretic and learning tools, fostering advances that ensure secure and trustworthy autonomy, precise control, and safe operations.
Spatio-temporal Modeling of Lithium-ion Battery Packs
The Battery Management System (BMS) deploys monitoring algorithms to detect abnormal behavior and limit the battery pack output to the lowest cell capacity (a general approach
used by all manufacturers). Detection and identification of defective cells/modules by the BMS in real-time are critical to ensure safe operation and longevity under extreme operating conditions. Therefore, the most urgent requirement of an EV’s BMS is a higher level of autonomy in health-conscious decision-making at a cell/module level. In this project, we will address this issue by 1) developing a high-fidelity pack model that can act at cell/module-level granularity to monitor health and abnormalities and 2) embedding BMS with continual learning capability to enable autonomy in decision-making.
The problem is challenging because the current state-of-the-art health-inclusive Li-ion battery models and learning algorithms primarily focus on the cell level. A pack model is developed by connecting these cell models in series and parallel with simplified dynamics and weak thermal interconnections (couplings) among them. The weak interconnection is a bottleneck to accurately represent the pack's temperature distribution, aging propagation, and health degradation. It also cannot be used to identify latent cell/module abnormalities to prevent their spreading. A pack exhibits a strongly interconnected, nonlinear, spatiotemporal, electro-thermal-aging (ETA) behavior.
The implementation of the model and learning scheme will provide the BMS with the required detection, prediction, and continual learning capability to orchestrate safe operation by precluding failures and thermal runaways.
The overall objective of the fellowship project is to develop a strongly interconnected model of a Li-ion battery pack and deep neural network (DNN)-based learning scheme to learn both the spatial and temporal dynamics.
The holistic growth of an undergraduate student (UG) lies in exposure to an appropriate and valuable education, high-quality research that invigorates critical thinking, and activities that hone interpersonal skills early on. It requires a multidisciplinary research site and educational environment enriched with people from diverse backgrounds.
The REU site will provide the undergraduates with experience in addressing four fundamental challenges in developing Smart Personal Protection Equipment (SmaPP). They are 1) the development of new smart materials, 2) incident heat flux measurement on the surface of SmaPP, 3) intelligent wireless sensing technology for human vital signs and radiation monitoring, and 4) human perception of SmaPP. The REU site will involve nine UG students in the research activities related to the development of SmaPP for ten summer weeks each year. Five students will join the REU site at OSU DET and four at UAH ECE.
The sample projects are
Strength and impact resistance test and analysis of the auxetic materials
Design optimization and prototype development of auxetic structures for SmaPP.
Study incident heat flux measurement on the surface of SmaPP.
Development of wireless sensors for monitoring human vital signs.
Wireless sensors for monitoring radiation exposure.
Protective mask usage in a campus environment: A critical evaluation of student protective behaviors during the COVID-19 pandemic.
Collaborator: Division of Engineering Technology, Oklahoma State University
SmaPP: Smart personal protection equipment
The overall objective of this REU site is
Provide multidisciplinary fundamental and applied research experience.
Inculcate a research mindset to take the intellectual onus of research activities in a future career.
Broadening participation by facilitating the involvement of women and underrepresented groups.
Provide the necessary impetus for graduate education and interpersonal skill development.
Towards Autonomy in Uncertain Environments: Exploring Vistas Beyond Consensus
Networked autonomous, heterogeneous agents (NAHA) performing collaborative tasks are ubiquitous in diverse applications, e.g., marine exploration and disaster management. To complete these tasks in an uncertain environment, NAHA requires significant communication bandwidth over the wireless channels and computing resources for learning and coordinated control. However, higher communication (data-sharing), e.g., in reconnaissance operations, threatens privacy and risks the security of NAHA performing coordinated tasks. Moreover, once a cyber attack or fault is detected, the massive scale and heterogeneity} of NAHA make it challenging to swiftly isolate the compromised agent to restore (even degraded) operation before it cascades to others.
The recently developed control-theoretic tools and techniques for NAHA (e.g., event-triggered consensus control), while successful in reducing the communication and control overhead, fail to address the key challenges, such as performance guarantee under limited and unreliable communications, data privacy, and isolation of compromised components to recover stability and resume coordination. To reap the advantages of collective behavior for complex task execution in an uncertain naval environment, there is a need for approaches that goes beyond the consensus-based algorithms by addressing the challenges arising from 1) loss of or limited communication, 2) data privacy, 3) isolation of faulty and/or compromised components, 4) adaptability and scalability in the real-time distributed learning for control.
Collaborator: Vignesh Narayan (University of South Carolina)
Networked autonomous, heterogeneous agents
The overall objective of this project is to design efficient aperiodic communication protocols, privacy-preserving and scalable learning algorithms for inference and filtering, and resource-conscious control strategies for NAHA in a distributed setting. We aim to exploit the structural and dynamic properties of the autonomous agents and establish fundamental theories to make the data-driven inference, filtering, and control schemes inherently robust to loss of or limited communication, preserve privacy, and facilitate the isolation of compromised agents.
AI-Enabled Autonomy of Robotic Inspection Platforms for Sustainability of Energy Infrastructure
Funding Agencies
This research will be funded by the Department of Energy (DOE)
Robotic visual inspection (RVI) technologies are employed for inspecting pipes, tanks, and other hard-to-access parts that are critical to energy sectors, hydrogen production, transport, and combustion processes, as well as carbon capture, utilization, and storage (CCUS). The RVI assures the integrity of these critical energy infrastructures, limit fugitive emissions, and ensures efficient operations. Therefore, the RVI technologies are critical to the safety of human life, emission reduction in the current energy value chain, and enable the energy transition in a low carbon manner for the sustainability of energy sectors. The state-of-the-art RVI technologies employed for visual inspection by industries still require human intervention and expertise for operation, data collection, and analysis. This jeopardizes human safety, requires extended time for the inspection, and is subject to human error.
Collaborator: Huaxia Wang (Oklahoma State University) and Baker Hughes ( Oklahoma City)
AI-driven robotic visual inspection technology
The research aims to develop an integrated AI-driven RVI platform with autonomous dynamic path planning and safe navigation capability for closed-loop data collection and real-time defect identification. The research will integrate deep learning-based defect identification models for dynamic and safe path and motion planning in real-time using multimodal data.
Resource-aware Intelligent Control of Cyber-Physical Systems
Funding Agencies
The project is funded by the University of Alabama in Huntsville research start-up
Cyber-physical systems (CPS) are pervasive in many applications, such as multi-robotic systems, power grids, smart transportation systems, etc. The intertwined connectivity involved complexity, and a significant amount of communication and computation demand by the CPS make the concurrent real-time interactions between the cyber and physical components more challenging. The difficulty in the real-time interactions amplifies the bottlenecks, namely, asynchronous and limited feedback, network imperfections (delays and packet dropouts, limited bandwidth and power), and computational complexity, to the implementation of intelligent control of CPS. It is well known that these bottlenecks deteriorate the learning accuracy and control effectiveness of intelligent controllers.
Our goal is to develop self-learning distributed and intelligent control schemes that are computationally efficient, inherently secure, and can perform optimally under the inevitable network constraints in a complex cyber-physical system.
Cyber-physical system architecture
Selected Publications:
A Sahoo, V. Narayanan, and S Jagannathan, “Resource aware learning-based optimal control of cyber-physical systems,” IEEE Technical Committee on Cyber-Physical Systems, vol.6, no.1, pp.24-34, Mar. 2021.
A. Sahoo ab, V. Narayanan, and Q. Zhao, “Adaptive gain observers for distributed state estimation of linear systems,” in Proceedings of the American Control Conference, New Orleans, USA, May. 2021.
A. Sahoo and V. Narayanan, “Differential-game for resource-aware approximate optimal control of large-scale nonlinear systems with multiple players,” Neural Network, vol. 124, pp. 95-108, 2020.
H. Niu, A. Sahoo, C. Bhowmick, and S. Jagannathan, “An optimal hybrid learning approach for attack detection in linear networked control systems,” IEEE/CAA Journal of Automatica Sinica, vol. 6, no. 6, pp. 1404-1416, 2019.
COMPLETED PROJECTS
Meta and Multimodal Learning for Smart Visual Borescope Inspection
Funding Agencies
This research is funded by OCAST and Baker Hughes
A video borescope is an optical instrument designed to assist visual inspection of narrow, difficult-to-reach cavities, consisting of a rigid or flexible tube linked by an optical or electrical system in between. Visual inspections are crucial to reduce the risk of equipment failures. To further improve the detection accuracy and efficiency, an artificial intelligence (AI)-driven automatic detection system is a desirable solution for the next-generation borescope.
Specifically, deep neural networks have been shown to perform well in many classical machine learning problems (image classification tasks). Developments in image processing and AI have significantly improved the capability of visual techniques for defects (i.e., crack, spots, corrosion, etc.) identification. However, successful implementation of state-of-the-art deep learning algorithms heavily depends on the training dataset quality (i.e., number of labeled images, image resolution, etc.). Due to the small training dataset, it is difficult to obtain a well-trained image classification network for defect identification. To achieve a better defect identification neural network, given very limited training samples, we are focusing on developing novel algorithms that can use both labeled and unlabeled data for training the classifier model. In this research, we will explore meta-learning and graph neural network-based architectures.
Collaborator: Huaxia Wang, Oklahoma State University and Baker Hughes, Oklahoma City
Smart borescope architecture
Intelligent Incipient Fault Detection System for Electric Vehicle Battery
Funding Agencies
This research is funded by Transportation Consortium of South-Central States
Lithium-ion (Li-ion) batteries are the primary power source for electric vehicles (EVs) due to their high energy and power density and long life cycle. The recent variants of the high-end plug-in EVs, with Li-ion battery packs, offer a range of approximately 300 miles on a single full charge, close to their gasoline counterparts. Further, to bridge the gap between the fueling time of gas-powered vehicles and the charging time of EVs, high-power chargers have also been introduced, reducing the charging time to less than 30 minutes. The Li-ion battery packs operate at maximum limits to deliver the required power to achieve these optimal performances.
Extreme operating conditions and abusive operations may lead to internal and external faults, such as short circuits, cell internal temperature rise, lithium plating and loss of lithium, and mechanical failure due to vibration. These internal faults have a cumulative effect on the battery’s health, aggravating the vulnerability to thermal runaway. Although various external safety technologies are employed in the battery monitoring system and battery management system (BMS) to protect the battery from external fault conditions, it is still challenging to detect internal faults from the available measurements (e.g., voltage, current, and surface temperature).
We aim to develop, implement, and validate an intelligent fault detection scheme capable of detecting a Li-ion battery’s internal faults in its incipient stage. This involves significant intellectual challenges related to root-cause analysis for determining the interrelation between internal parameters and type of fault and developing a computationally efficient neural network algorithm for hardware implementation.
Battery management system with intelligent fault detection.
Selected Publications:
G. Vennam and A. Sahoo, “Internal fault diagnosis of lithium-ion batteries using SOH-coupled electro-Thermal-Aging Model,” IEEE Transactions on Vehicular Technology, Under Preparation, Sep. 2022.
Geetika Vennam and A. Sahoo, “Core temperature estimation of lithium-ion batteries under internal thermal faults using neural networks,” Submitted to the American Control Conference, 2023.
Smart Battery Management System for Electric Vehicles
Funding Agencies
This research is funded by Transportation Consortium of South-Central States
The efficient and safe operation of Li-ion batteries in Electric Vehicles requires an intelligent and smart battery management system (BMS) capable of learning the health degradation for accurately estimating the state of charge (SOC) and the state of health (SOH). The design of smart BMS requires the development of 1) enhanced SOC and SOH-dependent parameter-varying dynamical model of Li-ion battery and 2) real-time learning algorithms to learn the parameter-varying model.
In this research, we are exploring the effects of both normal and accelerated degradation on battery health to develop a SOC and SOH-dependent parameter-varying electric circuit model and learning algorithms to learn the models. The smart BMS offers several advantages over the existing ones, including a realistic SOC and SOH-dependent model, which can further be used to optimize the Li-ion battery charging, power and energy management functions, and computationally efficient real-time machine learning algorithms for implementation on hardware.
Smart battery management system architecture.
Selected Publications:
G. Vennam and A. Sahoo, “A dynamic SOH-coupled lithium-ion cell model for state and parameter estimation,” IEEE Transactions on Energy Conversion, Under Review, May 2022.
Avimanyu Sahoo, "Smart Battery Management System for Electric Vehicles: Selflearning Algorithms for Simultaneous State and Parameter Estimation, and Stress Detection," Final Project Report, TranSET at Lousiana State University, Baton Rouge, 2021.
G. Vennam, A. Sahoo, and S. Ahmed, “A survey on lithium-ion battery internal and external degradation modeling and state of health estimation,” Journal of Energy Storage, vol.52, part A, Aug. 2022.
Geetika Vennam and A. Sahoo, “A novel coupled electro-thermal-aging Model for simultaneous SOC, SOH, and parameter estimation of lithium-ion batteries,” in Proceedings of the c American Control Conference (ACC), Atlanta, GA, USA, Jun. 2022, pp. 5259-5264.
G. Vennam and A. Sahoo, “Simultaneous state and parameter estimation of lithium-ion battery: An observer based approach,” in Proceedings of the American Control Conference (ACC), 2019, Philadelphia, PA, USA, 2019, pp. 4485-4490.
Model-based Fault Detection System for Induction Motors
Funding Agencies
This research was funded by the DEI Group, Millersville, MD
In this research, we have developed a parameter estimation algorithm to develop an induction motor model. The model is used to deployed to detect the internal fault of the motor. We have experimentally validated the model using our hydraulic testbed and NI CompactRIO (NI CRIO) data acquisition and embedded controller. We have acquired three-phase voltages, three-phase currents, and speed data with simultaneous sampling. The codes were implemented using the onboard FPGA of the NI CRIO.
Hydraulic testbed for experimental validation of the induction motor model
Development of Wireless Heat Flux Sensor
Funding Agencies
This research was funded by the Technology and Business Development Office, Oklahoma State University
Heat flux sensors are the first step towards establishing the fire-fighting front line since it provides information on the material's ignition potential. Wireless connectivity by leveraging the Internet of Things (IoT) technology would help significantly to monitor an area under fire remotely. This information can be used to optimize the plan of firefighting and allocate resources.
Some of the current state-of-the-art technologies consist of water-cooled heat-flux sensors. They have been widely used to measure temperatures and calculate heat flux in fiery conditions. While they are precise, these sensors are limited to applications where there is continuous water flow and adequate tubing. Another issue is that commercially available water-cooled heat flux sensors may not be sturdy enough for firefighters to handle. Our solutions are to develop a Wireless DAQ with that can be displayed remotely.
Our objectives included
Develop an IoT-based wireless heat flux sensing system
Transmit via Cellular networks to The IoT
Gather raw temperature data
Compute heat flux from at least 4 sensors
Reduce the Burden of wire connections and water supply
Must be low data low power.
Collaborators: Dr. Haejun Park, Oklahoma State Univerisity, Stillwater, OK