Modeling and Simulation of an Electric Warship Integrated Engineering Plant for Battle Damage Response Aaron M. Cramer Department of Electrical and Computer Engineering University of Kentucky cramer@engr.uky.edu Edwin L. Zivi Weapons and Systems Engineering Department United States Naval Academy zivi@usna.edu Keywords: Complex systems, integrated engineering plants, survivability, electric ships Abstract Novel continuity-of-service metrics are applied to perform proof-of-concept simulation-based design of a complex, dynamically interdependent electro-thermal-fluidspatial-control system subjected to hostile disruptions. The power system models are based on experimentally validated reduced-scale and reduced-complexity testbed models which are representative of U.S. Navy Next Generation Integrated Power Systems. In collaboration with an industry partner, representative thermal and spatial models were incorporated into the layered simulation. This time-domain simulation was used to quantify performance to a specific disruption in terms of a weighted aggregate continuity of service to vital loads. By extension, system vulnerability was quantified using a population of likely threats. Optimization-based early design space exploration was shown to dramatically decrease the notional ship integrated engineering plant vulnerability by improving the performance of a worst-case casualty by a factor of 6. These achievements establish the metrics, methods, and tools to perform quantitative optimization-based early design space exploration for complex, dynamically interdependent systems such as an electric warship. 1. INTRODUCTION As the transition to electric warships evolves, the need to predict and understand the behavior of shipboard integrated engineering plants (IEPs) becomes primary. The IEP of an electric warship is the infrastructure that provides propulsion as well as vital services including electric power and thermal management to mission-critical loads [1]. Moreover, as engineering casualty control and damage control responsibilities shift from crew to automation, the IEP must continue to function under disruptive battle conditions. The IEP is a complex dynamically interdependent system that is capable of exhibiting unforeseen behavior [2]. A particular source of this unforeseen behavior is the dynamic interdependence of subsystems that have Scott D. Sudhoff School of Electrical and Computer Engineering Purdue University sudhoff@purdue.edu traditionally been designed and analyzed independently. Moreover, many traditional design procedures impose large design margins to compensate for uncertainty and neglected system dynamics. For example, there is a mutual dependence between the thermal management system and the electric power system of an electric warship. The power generation, conversion, and distribution equipment requires cooling fluid to maintain safe operating temperatures. Simultaneously, the thermal management system requires electric power in order to operate the pumps required to move cooling fluid through the system. This interdependence cannot be adequately analyzed via static analysis. While the subsystems are dependent on each other, there is a temporal aspect to this relationship. If the cooling system fails, the power system may be able to continue to operate while temperatures rise. Since the thermal time scales are generally slower, the temperatures of power system components will rise over time until the component shuts down or fails. This type of dynamic interaction is best analyzed through time-domain simulation of the shipboard IEP. A particular situation in which the dynamic interdependence of the IEP subsystems must be understood is in battle damage scenarios. Increasing interdependence creates the possibility of cascading failures in which the effects of damage inflicted by a hostile adversary are not limited to the equipment that is directly damaged. These cascading failures give rise to specific vulnerabilities by which the enemy could dramatically impact the ability of the ship to carry out its mission. Herein, a time-domain simulation model of a notional electric warship IEP is described. This model encompasses the electro-thermal-fluid-spatial-control aspects of the IEP and its dynamic performance following hostile disruptions. This model enables worst-case and average continuity-ofservice metrics to be incorporated into conventional early design space exploration. The remainder of this paper is organized as follows. First, the simulation problem, including a description of the notional IEP, is presented. Then, a layered simulation approach that is used to implement the simulation model is described. Next, simulation model results are described and Figure 1. Electric system overview Figure 2. Seawater network overview specialized algorithms are used to explore the design space. Finally, future research directions are discussed. 2. SIMULATION PROBLEM To investigate new computational approaches for simulation-based design of electric warship IEPs, a notional electric warship IEP is set forth. This system is based on the land-based, reduced-scale Naval Combat Survivability (NCS) testbed at Purdue University [3] and is described in [4]. In order to employ experimentally validated power system models, the notational IEP is composed of the reduced scale, reduced complexity NCS testbed models. As in the NCS testbed, the notional IEP considered herein has two ac power networks and a zonal dc power distribution network with three zones. As shown in Figure 1, one of the ac networks is located forward, and the other aft. In this figure, each ac network is supplied with power from a prime mover (PM) driving the corresponding synchronous machine (SM) and establishing the ac electrical frequency. A brushless exciter/voltage regulator (BE/VR) controls the output voltage of each SM. Each SM is connected to an ac bus that feeds two loads: a propulsion drive (PD), which supplies the propulsion motor, and a power supply (PS), which delivers power to the zonal distribution network. The PSs from each of the ac networks supply a primary dc bus (PDCB) power on each side of the ship. A converter module (CM) is connected to each PDCB in each zone. Opposite CMs are connected to a zonal dc bus (ZDCB), which supplies power to the inverter module (IM) in each zone. The IM provides ac power to the ship service loads located in that zone. Shipboard electrical components are cooled by the seawater network depicted in Figure 2. In this figure, numbers enclosed in squares indicate seawater nodes (SWNs), numbers enclosed in circles indicate seawater valves (SWVs), and unenclosed numbers indicate seawater branches (SWBs). In each zone of the ship, a seawater pump (SWP) provides pressure to the network. Component heat exchangers (CHXs) are used to cool larger loads (particularly SMs and PDs) directly from the seawater network. Freshwater loop systems (FWLSs) are used to cool the smaller loads. As shown in Figure 3, each FWLS contains a fluid heat exchanger (FHX) that transfers heat to the seawater network. Figure 3. Freshwater loop system overview The baseline control scheme used herein is called anarchist because it only relies on local supervisory controllers to determine when to operate each device. There is no communication between devices. Each device is operated when the local measurements suggest that conditions are viable to operate. The overall layout of the reduced scale and reduced complexity notional ship is shown in Figure 4. Supervisory control algorithms and intercommunication models are reserved for future work. The goal of the simulation model presented herein is to assess the effects of hostile disruptions on the performance of the IEP. In particular, the weapons effects are considered and are represented by a spherical region of disruption. Any component that intersects with this sphere is rendered inoperable at the moment of the event. Likewise, any electrical line that intersects experiences a bolted ground fault, and any pipe that intersects ruptures. 3. LAYERED SIMULATION APPROACH To manage the complexity of this multidisciplinary system, a layered simulation approach is implemented. This approach is depicted in Figure 5. The various layers encompass different aspects of the system’s overall behavior. The spatial layer describes the geographic location of each of the components of the IEP. This description and information about a potential weapon event are used to determine which components would survive that event. The automation layer contains models of the supervisory controllers that govern the operation of IEP components. This layer combines information from other layers to determine when a given device can operate. This includes checking whether the device has survived the weapon event. The ac layer contains models of the two ac networks in the system and interacts extensively with the dc layer that models the zonal distribution network. The seawater layer models the pressure and flow of seawater through the seawater network, while the thermal layer models the heat flow through heat exchangers and freshwater cooling loops. Figure 5. Layered simulation approach Figure 4. Layout of integrated engineering plant The simulation itself is implemented in the Advanced Continuous Simulation Language (ACSL) [5]. ACSL provides several useful advantages for this application. The first is a textual model description language including macro capabilities which facilitates the parameterization of models which span the trade space. The second advantage is the ease with which parallel ordinary differential equation solvers can be implemented. This allows for a cosimulation approach in which subsystems with large time scale separation or that interact infrequently can be simulated separately. This increases the speed at which the overall system can be studies can be performed. In this case, the thermal and seawater systems are simulated using 0.5-s time steps, and the electrical systems are simulated using a variable time step algorithm with steps as small as 0.1 ns but typically 0.1 ms. Configuration, interconnection, and parametric data describing the of the IEP system components is maintained in Excel. An ACSL representation of the system can be automatically generated from the data contained in the spreadsheet using Visual Basic for Applications macros. This process is shown in Figure 6. Figure 6. Model automation This simulation approach can be expanded to encompass additional layers that model more complex aspects of system behavior. It also helps to manage the complexity of the simulation by restricting interaction between the layers to well-defined interfaces. Finally, it allows for cosimulation of various subsystems, resulting in reduced simulation time. The response of the system starting at 800 s (100 s before the explosion) is shown in Figure 8. Therein, v bus is the line-to-line rms voltage of the two ac buses. The dotted and solid traces correspond to the forward and aft ac buses. The variable i out , sm is the rms output current of the SMs. Again, the dotted and solid traces correspond to the forward (SM 1) and aft (SM 2) generators. The power supply output voltages and currents are designated v out , ps and i out , ps . In each case, the dotted trace is the forward power supply (PS 1) and the solid trace the aft power supply (PS 2). The output current ( i out , cm ) of the Zone 2 conversion modules is shown next (CM 2—dotted; CM 5—solid). The forward most (IM 1—dotted) and aft most (IM 3—solid) inverter module input voltages are labeled v in , im . Seawater flow rates for SWP 1 in Zone 1 (dotted) and SWP 3 in Zone 3 (solid) are labeled q P . The cold plate temperature of the heat exchangers cooling the generators is designated T hx , chx (CHX 1—dotted; CHX 4—solid). Freshwater flow rates T hx , fwl w cf and cold plate temperatures of the FWLs cooling IMs 1 and 3 are also shown (FWL 4—dotted; FWL 11—solid). The scenario that unfolds as a result of the weapons impact is illustrated in Figure 8. 4. BASIC SIMULATION RESULTS To examine the IEP model, the following scenario is considered. The system has been running for 15 minutes under full load, at which time the IEP has reached steadystate thermal operation. The propulsion drives (PD 1 and PD 2) are each consuming 37 kW and each inverter module (IM 1, IM 2, and IM 3) is providing 5 kW to ship-service loads. Then, a weapon with an explosion radius of 2.00 m detonates at the point (100.00,−4.18,3.38) m as shown in Figure 7. This particular explosion damages pipe SWB 20 in the seawater cooling network. Figure 7. Example event Cascading Damage Scenario 1. Plots begin at t = 800 s when the system has reached thermal steady state. 2. At t = 900 s a weapon with an explosion radius of 2.00 m detonates at the point (100.00,−4.18,3.38) m as indicated in Figure 7. This particular explosion damages pipe SWB 20 in the seawater cooling network. 3. Sea water valves SWV 3, 4, 7, 8, and 14 close isolating this fault Zone 3 from the rest of the seawater network. SWP 3 flow rate increases due to the unsecured rupture. The remainder of the seawater network settles to a new equilibrium point. 4. In Zone 3, the loss of seawater cooling causes the temperature of both CHX 4 and FWL 11 to rise following the weapon detonation. At approximately 1016 s (116 s after the initial event), SM 2 overheats and is forced to shut down. 5. The SM 2 overheat shutdown causes PD 2 and PS 2 to shut down. 6. The PS 2 shutdown causes CM 4, CM 5, and CM 6 to shut down, as evidenced by drop in the input voltage to IMs 1 and 3. The input voltages drop due to the increased output current of CMs 1 and 3 caused by the loss of CMs 4 and 6. In addition, the output currents of CM 2 and CM 5 are shared prior to this point, but at 1016 s, the current that CM 5 is providing has to shift to CM 2. The output currents of both SM 1 and PS 1 increase to cover the load that was previously shared with SM 2 and PS 2. 7. At 3363 s (2347 s after SM 2 shutdown), IM 3 overheats. This causes the pumps in FWLs 8, 9, 10, and 11, and SWP 3 to shut down, as seen in the pump flow rates. The shutdown of IM 3 causes the input voltage of IM 3 to rise as the CM 3 output current decreases. Also, the output currents of SM 1 and PS 1 drop due to the decreased load in the dc system. At this point, the system stabilizes in its new configuration. The following devices are shut down: SM 2, PD 2, PS 2, CMs 4, 5, and 6, IM 3, SWP 3, and FWLs 8, 9, 10, and 11. AC bus rms voltages, forward is dotted 4. SM 2 over temp. shutdown SM AC generator currents, forward is dotted 6. Partial load shift to SM 1 4. SM 2 over temp. shutdown DC PS power supply voltages, forward is dotted 5. PS 2 cascading shutdown DC power supply output currents, forward is dotted 5. PS 2 cascading. shutdown Conversion module currents, CM 2 (dotted) & CM 5 6. Partial load shift to SM 1 5. CM 5 cascading shutdown AC inverter input voltages, IM 1 (dotted) & IM 3 6. Droop due to increased load Seawater flow rates, SWP 1 (dotted) & SW 3 2. SWB 20 pipe rupture, 3. Rupture isolated Generator cold plate temperatures, forward is dotted 4. SM 2 over temp. shutdown Fresh water flow rates, CHX 1 (dotted) + CHX 4 7. IM 3 shutdown causes pump shutdown Inverter cold plate temperatures, IM 1 (dotted) & IM3 1. Steady-state initial conditions Figure 8. Example event scenario 7. IM 3 over temp. shutdown 5. RESULTS OBTAINED THROUGH SIMULATION By defining a metric called operability, it is possible to quantitatively assess the effects of a given event on a given IEP design [6]. The operability metric, ranging from 0% to 100%, quantifies the (weighted) degree to which vital engineering services are provided to mission-critical loads following a hostile disruption. Operability is defined as follows tf O ( ) I * t i 1 w i ( t , ) o i ( t ) o i ( t ) dt 0 tf (1) I * t i 1 w i ( t , ) o i ( t ) dt 0 use approximately 31,350 operability evaluations to arrive at this value. The event that causes this worst case is shown in Figure 10. This occurs when the missile strikes (39.82,2.73, 3.83) m, hitting PS 1 and CM 4, as well as the electric line from CM 4 to PDCB 2, the CM 4 part of ZDCB 1, SWB 1, and FWL 2. Because PS 1 is hit, the port dc bus goes down. Hitting the line from CM 4 to the starboard dc bus shorts the starboard bus. Within 10 s, the entire dc system is down. Since all of the cooling pumps require the zonal dc distribution system to operate, after 100 s, both PDs overheat. where w i ( t , ) 0 , o i (t ) , and o i* ( t ) are the weight, operating status, and commanded operating status, respectively, of load i at time t for event . For the event described above, the operability can be calculated from the simulation results as 95.03%. Operability can be averaged over a set of possible events, yielding average system dependability. System dependability is defined as D s O ( ) f ( ) d (2) E where E is the set of events being considered and f () is the probability density function over the set of events. The average system dependability over the set of events uniformly distributed over the volume shown in Figure 9 is 96.38%. This calculation is performed by approximating the integral in (2) as a sum over a grid of 31,350 points contained in the volume shown in Figure 9. The value is relatively close to 100%, but the event set contains many events in which no critical equipment is damaged. Even very small improvements in this value can reflect large improvements in average IEP performance. Figure 9. Set of events It is also possible to use optimization algorithms to locate the worst-case event. The worst-case operability establishes the minimum system dependability: (3) D s , min min O ( ) . E Genetic algorithms or particle swarm optimization are used in [6] to calculate the minimum system dependability of the notional ship as 9.16%. Both of these optimization methods Figure 10. Worst-case event This simulation model can be further combined with advanced optimization algorithms such as those in [7] to improve the design of the IEP with respect to the two system dependability metrics described above. In [8], the average system dependability is improved from 96.38% to 96.42% by modifying the locations of the components in the electrical network and adjusting the settings of the valves in the seawater network. More importantly, the minimum system dependability is improved from 9.16% to 60%. The new worst-case event is shown in Figure 11. This event is a sphere centered at (66.05,−1.65,12.04) m that disrupts IM 2. This causes IM 2 to shut down, but nothing further occurs. This value is absolutely optimal for the loads and load weightings. In essence, the adversary attacks the load with the highest priority. From the perspective of the IEP designer, no greater performance can be obtained. The adversary can obtain no further advantage (e.g., through cascading failures or by interdependencies) by attacking the IEP, so the adversary resorts to attacking the most valuable load. Figure 11. Improved worst-case event Each of these optimizations requires 3,375,000 operability evaluations. These evaluations are performed on the Genetic Optimization Processing Array (GOPA), a 216processor array located at Purdue University. Additionally, reduced-order models are used to reduce the runtime required for operability evaluation. 6. CONCLUSIONS This investigation demonstrates the capability of performing simulation-based design using dynamically interdependent electro-thermal-fluid-spatial-control timedomain simulation. Moreover, a proof-of-concept demonstration of novel operability and dependability continuity-of-service metrics is accomplished. Furthermore, optimization-based early design space exploration is shown to dramatically decrease the ship vulnerability by improving the performance of a worst-case casualty by a factor of 6. These achievements establish the metrics, methods and tools to perform quantitative optimization based early design space exploration for complex, dynamically interdependent systems such as an electric warship. 7. FUTURE DIRECTIONS Although this investigation represents significant progress towards understanding, quantifying, and optimizing the behavior of dynamically interdependent systems such as an electric warship IEP, this work represents a new beginning rather than a final accomplishment. First, it is possible to incorporate more advanced control strategies into the simulation model. The baseline control method used herein does not represent the current state of practice. More advanced control systems require a communication infrastructure that is itself vulnerable to the effects of disruption. These features can be represented within the layered modeling approach. This will also allow the behavior of the control systems themselves to be studied during battle damage conditions. Second, the electrical models need to become aligned with the current Next Generation Integrated Power System (NGIPS) baseline systems. This requires that the library of average-value models needs to be improved and expanded. Since the power system models bound the computational load, the balance between fidelity and computational complexity should be reconsidered. Third, energy storage modules need to be incorporated to quantify the effects of distributed energy storage and to learn how to optimize the sizing and location of energy storage modules throughout the IEP. Finally, more advanced simulation techniques can be applied. As an example, cosimulation improves simulation performance by allowing slower dynamics to be simulated separately from faster dynamics. The inherent separation of time scales associated with this multidisciplinary system creates this opportunity. However, techniques for event (i.e., zero crossing or root) detection must be employed that provide consistent results when cosimulation is applied. Further, in this system there is not a clear distinction between slow and fast dynamics. While the electrical dynamics are generally faster, the stiffly stable, variable step solver used to simulate the electrical dynamics will take very large time steps when the electrical transients have settled. Sometimes, the electrical solver takes larger time steps than the ―slower‖ fixed step thermal solver. Consistent event detection between solvers is necessary for dependable simulation results. References [1] J.P. Walks and J.F. Mearman, ―Integrated engineering plant,‖ presented at ASNE Reconfiguration and Survivability Symp. 2005. [2] C.N. Calvano and P. John, ―Systems engineering in an age of complexity,‖ Syst. Eng., vol. 7, no. 1, pp. 25–34, Jan. 2004. [3] S.D. Sudhoff, S. Pekarek, B. Kuhn, S. Glover, J. Sauer, and D. Delisle, ―Naval combat survivability testbeds for investigation of issues in shipboard power electronics based power and propulsion systems,‖ in Power Eng. Society Summer Meeting, 2002, vol. 1, pp. 347–350. [4] A.M. Cramer, R.R. Chan, S.D. Sudhoff, Y. Lee, M.R. Surprenant, N.S. Tyler, E.L. Zivi, and R.A. Youngs, ―Modeling and simulation of an electric warship integrated engineering plant,‖ in SAE Power Syst. Conf., 2006. [5] Advanced Continuous Simulation Language (ACSL) Reference Manual, AEgis Technologies Group, Inc., Huntsville, AL, 1999. [6] A.M. Cramer, S.D. Sudhoff, and E.L. Zivi, ―Performance metrics for electric warship integrated engineering plant battle damage response,‖ IEEE Trans. Aerosp. Electron. Syst., vol. 47, no. 1, pp. 634–646, Jan. 2011. [7] A.M. Cramer, S.D. Sudhoff, and E.L. Zivi, ―Evolutionary algorithms for minimax problems in robust design,‖ IEEE Trans. Evol. Comput., vol. 13, pp. 444–453, Apr. 2009. [8] A.M. Cramer, S.D. Sudhoff, and E.L. Zivi, ―Metric optimization-based design of systems subject to hostile disruptions,‖ IEEE Trans. Syst., Man, Cybern. A, Syst., Humans, to be published. Biography Aaron M. Cramer received the B.S. (summa cum laude) degree in electrical engineering from the University of Kentucky, Lexington, in 2003. He received the Ph.D. degree from Purdue University, West Lafayette, IN, in 2007. He is an Assistant Professor at the University of Kentucky. From 2007 to 2010, he was a Senior Engineer with PC Krause and Associates, West Lafayette, IN. From 2004 to 2007, he was a Graduate Research Assistant at Purdue University. His interests include simulation techniques, optimization methods, and power electronics. Having performed the bulk of this investigation as his dissertation research and with years of professional simulation experience, Aaron is ideally positioned to contribute to future work in this area. Edwin L. Zivi received the B.S. degree in engineering science and mechanics from the Virginia Polytechnic Institute and State University, Blacksburg, in 1975. He received the M.S. and Ph.D. degrees in mechanical engineering from the University of Maryland, College Park, in 1983 and 1989, respectively. He is presently an Associate Professor of Systems Engineering at the United States Naval Academy, Annapolis, MD. Prior to 1998, he was a Senior Research Engineer and Technical Advisor at the Naval Surface Warfare Center, Annapolis, MD. His research focuses on the integration of survivable naval automation systems with advanced shipboard engineering and damage control systems. Scott D. Sudhoff received the B.S. (highest distinction), M.S., and Ph.D. degrees in electrical engineering from Purdue University, West Lafayette, IN, in 1988, 1989, and 1991, respectively. Currently, he is a Professor at Purdue University. From 1991 to 1993, he served as Visiting Faculty with Purdue University. From 1993 to 1997, he was a Faculty Member at the University of Missouri-Rolla. He has published over 50 journal papers, including six prize papers. His interests include electric machines, power electronics, and finiteinertia power systems, applied control, and genetic algorithms.