The Integrated Reconfigurable Intelligent Systems (IRIS) initiative is proposed as a response to the U.S. Navy’s challenges for designing next-generation surface combatants that meet operational goals for increased mission effectiveness, survivability, environmental compatibility, and reduced operating cost. An IRIS designed system is envisioned to integrate several intelligent systems onboard to collect the information about the environment and ship state, assess the situation and then take a best course of action to reconfigure the ship into the state most suitable to handle the situation at hand. This indicates that the IRIS designed system possesses three main functions: the ability to sense, assess and react.The main objective of this research initiative is to develop methods to support the U.S. Navy in the study, design, and operation of complex naval systems.
Unconventional and innovative methods and strategies are investigated in order to capture complex system in a highly dynamic and uncertain environment, including resilience assessment, dynamic modeling and co-simulation, graph-based modeling, hierarchical intelligent control, and human in the loop control. Using traditional systems engineering practices for the early design process followed by an integrated design environment, IRIS seeks to shift ship design to a distributed intelligent control architecture through increased intelligence and automation. Although the IRIS research effort is being applied specifically to the naval complex systems, the IRIS concept is generic in nature and can be applied to any complex intelligent system, such as electrical aircraft, smart building and intelligent automobile.
For more information about IRIS, please contact the project lead Dr. Yongchang Li (email@example.com).
IRIS Design Process
The ultimate goal of IRIS research is to design reconfigurable intelligent naval systems that are capable of self-sensing, self-assessing and self-reacting. The success of such system designs highly depends on the process used to design the system. Traditional design processes fall short of adequately addressing all the functional requirements of IRIS system. The IRIS team at ASDL employed a systematic approach and formulated an integrated design process using Unified Modeling Language (UML). As can be seen, the new design process captures all information necessary for a design process, such as what expertise is needed, what activities need to be completed, what method/tools need to be used, what models need to be developed and how the models interact with each other in each design step.
IRIS Research Efforts
The main objective of the IRIS research initiative is to develop methods to support the U.S. Navy in the study, design, and operation of complex naval systems. Five main research efforts have been identified as critical to the success of this research initiative:
- Dynamic Modeling & and Co-Simulation
- Graph-based Modeling and Dynamic Surrogate Modeling
- Resilience Assessment and Design
- Hierarchical Intelligent Control
- Human-in-the-loop Control
These efforts are synergistic and combined, and produce a system-wide approach to the analysis and understanding of the total ship, as shown in the figure.
Dynamic Modeling & and Co-Simulation
The naval systems are complex systems that are composed of heterogeneous systems, where the interactions and interdependencies are extensive and nonlinear and the behavior of the overall system cannot be inferred from the analysis of the individual portions composing the larger system. In order to find the optimum or robust solution of a ship design for optimal mission effectiveness, the integrated subsystems need to be well studied and understood. Therefore, a Modeling and Simulation (M&S) environment is needed, for the purpose of integrating multiple physics-based models to accurately simulate the dynamic behavior that the systems exhibit.
First, the models representing the major subsystems of the ship need to be identified. In order to gain insight into the identity of the constituent systems, the hierarchical functional architecture of the IRIS ship is explored. An initial functional view of system is conducted to break down different ship functionalities. These functionalities then are organized and refined to obtain a functional decomposition of the system. Based on the functional decomposition, the system models are developed to simulate the functionalities and demonstrate the capabilities of the IRIS designed ship. In current implementation, the simulation environment is mainly used to investigate the behavior of the system and demonstrate the reconfigurability of the system. Thus, the physics based models are identified as fluid mode and electrical model which will provide resources (power and cooling fluid) needed by other subsystems. After the models are identified, they are developed by domain experts using different platforms to simulate the dynamic characteristics the systems exhibit.
In order to investigate the behavior of the ship system, multiple dynamic systems have to be linked into a single integrated environment to represent the overall ship system. This indicates that the dynamic models are not run in isolation but in conjunction with other dynamic models. The linking and integration of different dynamic models into one overall integrated model is referred to as co-simulation. Co-simulation has become a very important enabler for large-scale complex systems with multiple subject matter inputs, multiple third party models, and distributed simulation. In co-simulation, it requires the predetermination of the simulation time steps since these time steps are of critical importance to simulation accuracy and stability.
In order to execute the integrated simulation the sub-models will need to be executed in parallel and, at certain time intervals, need to stop their execution and exchange their data. Similar to simple dynamic models, the criticality of the time step on simulation error and stability directly imports to integrated dynamic systems, with yet an additional level of time stepping, namely that of the data exchange of the sub-models. This becomes an even more important issue if the integrated model is “stiff”. “Stiffness” describes the situation where the sub-models have vastly different dynamics and time scales. In such conditions, the timing must be adapted to cater to the different dynamic behaviors of the sub-models, while at the same time trying to keep the simulation error and computational expense low. The research currently is in the process of determining and applying an algorithm from numerical integration, and will help to determine time steps for co-simulation that will help to reduce errors and computational expenses, both by time stepping and application of further improvements on co-simulation execution, such as extrapolation.
Graph-based, Dynamic Surrogate Modeling for Damage Analysis
A distinct goal of the IRIS designed system is that its engineering system network be capable of performing self-healing or reconfiguration by intelligent, distributed control layers, in order to achieve the system’s utmost fight-through capability and survivability against various damage situations. This unique design challenge leads to the formulation of a new, highly flexible modeling approach that can answer the question of how the system’s survivability can be impacted by different numbers and locations of component redundancy and distributed control agents, and different subsystem-connection topologies for the thermal management and electrical power networks in the system.
As enablers of filling these gaps in modeling capabilities, two essential elements were developed in the formulation of the modeling method. The first one is the graph-based topological modeling method, which is employed for rapid model reconstruction and damage modeling, and the second one is the recurrent neural network-based, component-level surrogate modeling method, which improves the affordability and efficiency of modeling and simulation (M&S) computations of the dynamic systems. The integration of the two methods delivers computationally efficient, flexible, and automation-friendly M&S which creates an environment for more rigorous damage analysis and design space exploration of the reconfigurable engineering flow networks.
Resilience Assessment and Design
In the presence of operational uncertainty, one of the greatest challenges in ship design is to ensure that complex systems are going to be safe for their operators and their environment. The IRIS designed ship is envisioned to be self-sensing, self-assessing and self-reacting. Thus, the IRIS initiative, has been advocating for a paradigm shift from passive, reactive and diagnosis-based approaches to the development of architectures that will autonomously manage safety and survivability through active, proactive and prognosis-based engineering solutions. The shift aims to bring safety considerations early in the engineering design process, in order to reduce retrofitting and additional safety certification costs, increase flexibility in risk management, and essentially make safety “built-in” the design.
However, current design methods may not be adequate to implement the new vision for safety, as it is not only about sizing and optimizing an architecture for safety, but it also goes back to foundations of safety engineering, e.g. understanding the mechanism of accident occurrence, extend the concept of risk and redefine the risk assessment process. The need for alternative understanding of various elements of safety has driven the safety management community towards an emerging initiative, known as Resilience Engineering. Resilience engineering is pertinent to safety engineering and is expected to offer insight and research directions regarding the design of more safe and survivable system architectures. Historically, resilience engineering has been first discussed in 1973 by C.S. Holling, through a publication that discusses the resilience and stability of ecological systems. It then proliferated to other disciplines of engineering and science, addressing similar global issues in system stability, robustness, adaptability, and survivability.
System Resilience is still a relatively new concept and not surprisingly, there are several unanswered questions on its definition, relevant concepts, and applications. Several scientific communities are embracing the concept in their own way, for addressing their unique safety and survivability challenges. For instance Hollnagel defines system resilience as a system characteristic, depending on how well a system anticipates disturbances, monitors its status and responds to its mission expectations.
As part of the research within the IRIS initiative, it has been investigated how system resilience is positioned within the big picture of improving system effectiveness through survivability. A framework for system resilience investigation and assessment is being formulated, including standard definitions, the supporting resilience functions along with quantitative deterministic and probabilistic techniques for total system resilience evaluation, based on dynamic event-based scenarios.
Hierarchical Intelligent Control
Since the naval ship is operated in an environment with continuously changing situations and various uncertainties, the reactions decisions are made in a sequential manner and based on the assessment of large amount of uncertain information changing over time. The complexities of the decision making in ship operation are always a challenge to human decision makers since it is usually difficult for human being to manage and organize the time-dependent information and make wise decisions based on the probabilistic assessment of the acquired information. However, currently many critical ship systems are operated manually, which causes increased ownership cost and non-optimized operational actions, as a result, often leading to reduced survivability and mission effectiveness. The gap between the Navy’s requirements and current situation on ship operation inspires a need for an advanced approach to facilitate the decision making tasks in the ship operation. Therefore, a control system is required to enable the system to make autonomous decisions on reconfiguring the system into the mode best deal with the situation at hand.
A hierarchical control architecture was proposed, consisting of three levels – autonomous decision making, distributed agent-based control and reactive control. All the control levels collaborate together to fulfill the control functions.
Autonomous Decision Making
Advanced approaches capable of making autonomous decision under uncertain conditions were investigated, and a Multi-Agent Markov Decision Process (MAMDP) was formulated to perform sequential decision making for multiple agents in a stochastic process. MAMDP can generate an optimal policy specifying the best course of action that a ship can take on to maximize the objective of its operation.
Distributed Agent-based Control
When the ship takes on the best course of action, in order to coordinate all the distributed parts of a widely distributed and dynamic system in real time, the control system should provide intelligent ways to identify and monitor various situations using incomplete and uncertain information. A distributed dynamic inference engine is formulated using Bayesian Network and employed to monitor the system state and environmental situation in a probabilistic way.
After mid-level control identified the system state, appropriate control commands at component level will be created for low level control to carry out. Reactive control is a rule-based controller that can react to the situation that needs rapid response, such as rupture identification and isolation in a fluid system.
The desire to optimize manning and the functions of the crew require that the ship be autonomously reconfigurable, but it is essential that operators have the ability to override the decision making systems. This is due to the fact that the Navy’s culture does not permit automation systems to work unsupervised, and therefore it is crucial to see how the dynamics of that interaction will play. For this reason, a Human Machine Interface (HMI) has been developed that allows operators to supervise and interact with the system. The HMI model serves to send the data and receive commands from the interface, and is employed to visualize the simulation results and enable interactions between the simulation environment and the human participants.
In order to properly present a simulation intended for human interaction, the simulation environment and the HMI must be able to maintain a real time performance requirement. This helps human operators obtain a feel for the responsiveness of the system to external stimulus. The development of the HMI has created an interesting opportunity in the realm of design science. The HMI has a very tight integration with simulation environments, and can be employed as an application framework for analyzing data. The developed HMI is a web-based interface which is integrated with other models through Model Center. The simulation environment sends information to the remote server for visualization and analysis, and based on the human decision, override commands can be sent through the interface from the server to interact with the simulation. It is critical for the human operator to better understand the overall system behavior and interact with the system dynamically to avoid some automation surprise.
- Balchanos, M., Li, Y., Mavris, D.N., “A Theoretical Framework for the Analysis and Design of Resilient Engineering Systems” To be submitted to the Journal for Safety Research, by Elsevier, 2011.
- Hoepfer, M., Li, Y., Mavris, D.N. "Integration of a Dynamic Modeling and Simulation Environment for Naval Complex System Design”, To be submitted to Simulation Modeling Practice and Theory, by Elsevier, 2011.
- Moon, K and Mavris, D, “Modeling and Simulation for Damage Analysis of Intelligent Self-Reconfigurable Ship Fluid Systems in Early Design Phase,” under review, submitted to Simulation Modelling Practice and Theory, Elsevier in Jun. 10, 2010
- Nairouz, B., Hoepfer, M., Weston, N. and Mavris, D.N., “Investigations for Time Step Settings in a Dynamic System Co-Simulation Environment”, Electric Ship Design Symposium, Jan 2009
- Li, Y., Balchanos, M., Nairouz, B., Weston, N. and Mavris, D.N., “Modeling and Simulation of Integrated Intelligent Systems”, Proceeding of the 2008 Winter Simulation Conference, Dec. 7-10, 2008
- Balchanos, M., Moon, K., Weston, N., and Mavris, D.N., “A Dynamic Surrogate Model Technique for Power Systems Modeling and Simulation”, 2008 SAE Power Systems Conference, Nov. 11-13, 2008
- Nairouz, B., Hoepfer, M., Weston, N., and Mavris, D.N., “An Exploration of Guidelines and Limitations on the Co-Simulation of Integrated Sub-Models”, Advanced Naval Propulsion Symposium, Dec 2008
- Li, Y., Weston, N., and Mavris, D.N., “A Multi-Agent Autonomous Decision Making Process for Resource Allocation”, 2007 ASNE Automation and Control Symposium, Dec. 10-11, 2007
- Balchanos, M., Balestrini, S., Weston, N., and Mavris, D.N., “Multi-Physics Time-variant First-order Model Integration of Complex Systems”, 2007 ASNE Automation and Control Symposium, Dec. 10-11, 2007
- Moon, K., Weston, N., and Mavris, D.N., “A Method for Speeding up the Time-Domain Simulation of a Complex System Using Surrogate Modeling Technique”, 2007 ASNE Automation and Control Symposium, Dec. 10-11, 2007
- Weston, N., Balchanos, M., Koepp, M., and Mavris, D.N., “Strategies for Integrating Models of Interdependent Subsystems of Complex System-Of-Systems Products,” IEEE Proceeding of the Thirty-Eighth Southeastern Symposium on Systems Theory, March 5, 2006
- Weston, N., Balestrini-Robinson, S., Fulmer, D., and Mavris, D.N., “Design of a Resource Advisor for the Next-Generation Surface Combatant,” IEEE Proceeding of the Thirty-Eighth Southeastern Symposium on Systems Theory, March 5, 2006