IMPROVING THE RELIABILITY OF MVDC SHIP
POWER SYSTEMS
Technical Report
Submitted to:
The Office of Naval Research
Contract Number: N0014-08-1-0080
Submitted by:
Surya Santoso, Aristotle Arapostathis
University of Texas at Austin
Sherif Abdelwahed, Ranjit Amgai
Mississippi State University
David Cartes, Ruturaj Soman, Tuyen Vu
The Florida State University
Ben Stevens, Jian Shi
September 2013
Any opinions, findings, conclusions or recommendations expressed in this publication are those
of the author(s) and do not necessarily reflect the views of the Office of Naval Research.
Approved for Public Release – Distribution Unlimited
1
MISSION STATEMENT
The Electric Ship Research and Development Consortium brings together in a single entity the
combined programs and resources of leading electric power research institutions to advance
near- to mid-term electric ship concepts. The consortium is supported through a grant from the
United States Office of Naval Research.
2000 Levy Avenue, Suite 140 | Tallahassee, FL 32310 | www.esrdc.com
TABLE OF CONTENTS
1
Executive Summary................................................................................................................. 1
2
Distribution System Reliability and High-Level Network Topology ..................................... 3
2.1
2.1.1
Component Reliability............................................................................................... 4
2.1.2
System Reliability ...................................................................................................... 4
2.2
Component Reliability Indices......................................................................................... 5
2.3
System Reliability Indices................................................................................................ 5
2.3.1
First-Order Interruption Scenarios ........................................................................... 6
2.3.2
Second-Order Interruption Scenarios Not Involving Stuck Breaker ......................... 8
2.3.3
Second-Order Interruption Scenarios Involving Stuck Breaker ................................ 9
2.3.4
System Markov Model .............................................................................................. 9
2.4
Ring Bus ................................................................................................................... 11
2.4.2
Breaker-and-a-Half.................................................................................................. 12
2.4.3
Breaker-and-a-Half with Additional Bus Tie Circuit Breakers ................................. 14
2.4.4
Double Breaker, Double Bus ................................................................................... 14
Topology Reliability Comparisons ................................................................................ 15
2.5.1
Reliability Comparison Procedure .......................................................................... 15
2.5.2
Reliability Comparison Results................................................................................ 15
2.6
Equipment Placement Algorithm ................................................................................... 20
2.6.1
Equipment Placement Algorithm Procedure .......................................................... 20
2.6.2
Equipment Placement Algorithm Results ............................................................... 22
Power System Reliability and Control Systems .................................................................... 26
3.1
4
Distribution System Topologies..................................................................................... 11
2.4.1
2.5
3
Reliability Concepts ......................................................................................................... 4
Reliability Enhancement of SPS .................................................................................... 26
3.1.1
Static Analysis ......................................................................................................... 27
3.1.2
Dynamic Analysis .................................................................................................... 32
Failure mode and effects analysis studies for understanding risks........................................ 39
4.1
The relevance of FMEA for this research ...................................................................... 40
4.2
Two sub-parts of a detailed FMEA ................................................................................ 42
4.2.1
Functional FMEA ..................................................................................................... 42
i
4.2.2
4.3
Hardware FMEA ...................................................................................................... 42
F-FMEA process ............................................................................................................ 43
4.3.1
F-FMEA applied to the overall MVDC zonal SPS architecture ................................ 44
4.4
Application of F-FMEA data for further research ......................................................... 44
4.5
Automating F-FMEA for different modes of the SPS ................................................... 46
4.5.1
4.6
Multi agent systems technology research .............................................................. 54
Natural language processing (NLP) ............................................................................... 58
4.6.1
LSA applied to extract useful information from the F-FMEA database.................. 58
5
Conclusions ........................................................................................................................... 60
6
References ............................................................................................................................. 62
LIST OF FIGURES
Fig. 1: Markov model of a first-order interruption scenario.......................................................... 7
Fig. 2: Markov model of a second-order interruption scenario. .................................................... 8
Fig. 3: Markov model representing an equipment system........................................................... 10
Fig. 4: A shipboard distribution system with ring bus topology. ................................................ 11
Fig. 5: A simple breaker-and-a-half topology.............................................................................. 12
Fig. 6: A shipboard distribution system with breaker-and-a-half topology (version one)........... 13
Fig. 7: A shipboard distribution system with breaker-and-a-half topology (version two). ......... 13
Fig. 8: Comparison of (a) ring bus; (b) breaker-and-a-half; and (c) double breaker, double bus
topologies...................................................................................................................................... 14
Fig. 9: Equipment system interruption rate by distribution system topology.............................. 17
Fig. 10: Equipment system mean time to repair by distribution system topology. ..................... 18
Fig. 11: Equipment system total expected downtime by distribution system topology. ............. 18
Fig. 12: Reliability comparison of ring bus; breaker-and-a-half; and double breaker, double bus
topologies...................................................................................................................................... 19
Fig. 13: Operational procedure of the equipment placement algorithm. ..................................... 22
Fig. 14: Optimal equipment placement within the breaker-and-a-half topology, as determined by
the equipment placement algorithm.............................................................................................. 23
Fig. 15: Changes in the radar system’s reliability indices between the initial and modified
equipment configurations.............................................................................................................. 24
Fig. 16: Changes in the pulsed load system’s reliability indices between the initial and modified
equipment configurations.............................................................................................................. 24
Fig. 17: Changes in the zonal load center system’s reliability indices between the initial and
modified equipment configurations. ............................................................................................. 25
Fig. 18: Framework for Reliable Operation................................................................................. 27
Fig. 19: SPS architecture under study.......................................................................................... 28
Fig. 20: Fuel Consumption Curve fitting for 50Mw Gen with quadratic polynomial function... 30
Fig. 21: Fuel Consumption curve fitting for 5 MW Gen with quadratic polynomial function.... 30
Fig. 22: Shift Factor. .................................................................................................................... 31
Fig. 23: Components of long term system design........................................................................ 32
ii
Fig. 24: Model based Architecture for SPS Dynamic Analysis................................................... 32
Fig. 25: ESHIP under study in 3.1.2.1. ........................................................................................ 34
Fig. 26: Control Work Flow......................................................................................................... 38
Fig. 27: MVDC zonal architecture modeled on the RTDS........................................................... 40
Fig. 28: - Zonal load centers modeled on the RTDS .................................................................... 41
Fig. 29: Subtle difference between F-FMEA and H-FMEA that add up to produce a detailed
FMEA ........................................................................................................................................... 43
Fig. 30: – (a) Ring battle mode, (b) Split plant battle mode ........................................................ 56
Fig. 31: – Each device mapped onto an agent in supervisory control architecture [25].............. 57
Fig. 32: - Proposed approach using FMEA information with NLP and AI-based diagnostics for
decision risk mitigation and decision support............................................................................... 59
LIST OF TABLES
Table 1: Component Failure Reliability Indices............................................................................ 5
Table 2: System Topology Component Count Comparison ........................................................ 15
Table 3: Equipment System Reliability Indices by Distribution System Topology.................... 16
Table 4: Overall Interruption Rate by Distribution System Topology ........................................ 17
Table 5: Object-Slot Swaps ......................................................................................................... 21
Table 6: Equipment Configuration Reliability Index Comparison.............................................. 23
Table 7: Snapshot of Fuel Consumption Metrics ........................................................................ 31
Table 8: Subsections of the architecture with their constituents and functions ........................... 44
Table 9: List of devices in subsections and their respective functions ........................................ 47
Table 10: Energy storage F-FMEA.............................................................................................. 48
Table 11: Radar F-FMEA ............................................................................................................ 49
Table 12: Propulsion motors F-FMEA ........................................................................................ 50
Table 13: Pulsed load F-FMEA ................................................................................................... 50
Table 14: Zone 1 and 2 F-FMEA.................................................................................................. 51
Table 15: DC power ring F-FMEA............................................................................................... 53
Table 16: DC busses F-FMEA...................................................................................................... 53
iii
1 EXECUTIVE SUMMARY
Analysis was performed, by the UT-Austin team, to quantify and compare the reliability of
several different notional shipboard DC distribution system topologies in serving their equipment
loads. Further, the relationship between the relative placement of loads and generators within a
distribution system and the system’s reliability was investigated, resulting in an algorithmicallyderived optimal placement configuration in the best-performing system topology. Using Markov
models and fault-tree analysis, reliability indices were derived from distribution system
component reliability indices, and these values were compared for each topology.
A distribution system based on the breaker-and-a-half topology often used in terrestrial utility
substations was found to be superior in terms of reliability to the currently standard ring bus
topology. Expected rates of service interruptions to equipment systems served by the breakerand-a-half system were reduced overall, in some cases dropping dramatically to below one
expected interruption per 10,000 years. This improvement, however, came at the expense
requiring more circuit breakers in the distribution system’s construction.
Within this breaker-and-a-half distribution system, an optimal placement of loads and generators
was found that further improved the reliability of the system. This improvement over the base
case was marginal, but the optimized placement configuration was able to reduce the expected
interruption rate of the ship’s radar system by over 40%.
The MSU team is working to develop effective measures to improve the overall operational
reliability of the shipboard power system through effective model-based design and controls. The
approach is thus divided into two aspects: a long term static controls approach that supports the
steady state long term planning and a dynamic control approach which handles the dynamic
power system reconfiguration under different scenarios including the load variation and fault
conditions. The proposed power management framework then determines the overall solution
based on the real-time evaluation of these two aspects.
The tool for appropriate assessment includes the evaluation metrics from a system level
perspective. Effective control techniques to support the reliable shipboard power system
operation while upholding the survivability and adequacy is one of the focus area of the study. In
addition, the work eventually includes the dynamic study for transient management, tracking
loads, and system-level fault handling capabilities.
The work is focused on the development of the conceptual management framework relying on
model based design to support the robust and adaptive controls for reliability and survivability.
Coordinated, short term and long term objectives are included in the design along with the
aspects of reliability and Quality of Service (QOS) metrics that have been previously established
in literature or being developed by researchers in ESRDC. Long term design plans are
investigated with the generation dispatch and the unit commitment support. Various mission
conditions are considered for the long term system planning. Minimization of the fuel
consumption for the mission has not only reduced the operation cost, but even supported the
survivability of the SPS. Reliability considerations along with the dispatch controls are being
1
studied. MSU is looking further for the collaboration with the University of Texas at Austin on
this aspect.
The MSU team worked to develop quantitative measures to characterize the performance impact
of failures. An important aspect of improving reliability of the system involves the decision
support system providing reconfiguration strategies necessary to mitigate failure effects and
enable a system to recover. The developed quantification measures are then used to develop
methods to evaluate the quality of service of the power system in terms of observable system
variables relevant to its performance and reliability, and accordingly compute an optimal system
configuration.
Another top priority in the design of shipboard power system applications is the stability and
safety of operation. The evaluation of dynamic responses under transition of operating conditions
or contingency scenarios plays a very important role as shipboard power systems are tightly
coupled and potentially susceptible to the partial dynamic changes within the network which
would cause large disturbance in case of emergencies or damages. Previous research conducted
in the reconfiguration area mainly focuses on the static system performance with regards to
certain optimization functions. There is yet no salient research effort on the dynamic behavior of
the shipboard power system (SPS) under disturbance or operation status change. Stability, load
sharing, fault handling, and load following are some of the problems that are being addressed
dynamically through model based control. This two level control design is expected to foster
more reliable and robust operation. Mission priorities and network connectivity are considered
as a part of the problem statement. The aspects of the reliability are to be included in the cost
function. Load sharing and generation dispatch control are also being considered in the system
level controls architecture.
Additionally, at CAPS-FSU, a well established reliability analysis method namely failure mode
and effects analysis (FMEA) has been employed to understand risks aboard the envisioned SPS.
A thorough understanding of possible known risks would enable efforts to address the issues and
mitigate them, in turn increasing prospects of enhancing and improving stability and mission
safety. The emphasis is laid on a functional FMEA (F-FMEA) which serves as the most
appropriate method to start analyzing failure cause and effects in a novel system such as the
zonal shipboard power distribution architecture studied. The F-FMEA is shown in a tabular
format in this report as an example with the zonal MVDC system as reference. Potential uses of
the information that stem from conducting a sufficiently detailed FMEA are proposed for which
further research is necessary.
2
2 DISTRIBUTION SYSTEM RELIABILITY AND HIGH-LEVEL NETWORK TOPOLOGY
In an electric naval vessel, properly functioning equipment, such as radar, weapons, or
propulsion motors, is of paramount importance to both mission success and personnel wellbeing.
One key component to ensuring continuity of service for a ship’s equipment is the shipboard
electrical distribution system. A failure of the distribution system can result in pieces of vital
equipment being left without power until repairs can be performed, potentially causing serious
threats to the crew and to the mission. Therefore, it is necessary to ensure that shipboard
electrical distribution systems are designed to be as robust as possible in order to minimize the
frequency of service interruption.
During peacetime operations, service interruptions are most often caused by failures of
individual components within the distribution system, such as prime movers, circuit breakers, or
power electronic converters. In addition to peacetime equipment failure, widespread damage to
distribution circuits providing electrical service to equipment during wartime is very likely. The
specific scenarios in which one or more concurrent component failures will lead to a load service
interruption are dependent upon the overall topology of the distribution system, as well as the
relative placement of loads and generation units within the system. Previous work has been
performed to establish metrics for calculating peacetime quality of service (QOS) in shipboard
power distribution systems [1]. This QOS metric has been applied to shipboard power system
design, but these studies have primarily focused on design choices such as generator size and
control interfaces, not on comparisons of overall system topologies [2], [3].
Our work has evaluated system reliability from the perspective of the overall distribution
network topology. That is, the relationship between the reliability of a distribution circuit and
the high-level topology of its connections. System reliability, herein, is defined by indices
quantifying the expected frequency and duration of service interruptions to equipment loads
caused by failures of individual components in a specific network topology. This approach to
quantifying distribution system reliability as a function of system topology has been utilized in
many previous works [4], [5].
Currently, most discussions of electrical power distribution in a notional electric ship utilize a
distribution topology resembling the ring bus configuration used in terrestrial power system
substations. The ring bus is an attractive option for a shipboard distribution topology because it
includes redundancy in distribution paths, requires a relatively small number of circuit breakers,
and is readily scalable. However, reliability studies of terrestrial substations have noted that the
ring bus is not the only distribution topology to possess these qualities. In particular, the
Breaker-and-a-Half (BAAH) distribution topology has been found to be an overall more reliable
substation design with a similar ease of scalability, though at a cost of requiring a greater number
of circuit breakers than the ring bus design [6].
High-level distribution system topologies for use in an electric ship are compared in terms of
reliability using a methodology combining fault-tree analysis and Markov models in order to
derive system reliability metrics. Through these metrics, the relative reliability conferred to
equipment systems by competing distribution topologies can be compared.
3
Several notional distribution system topologies, based upon different arrangements found in
terrestrial utility substations, were designed and compared, ultimately finding the BAAH
topology to be the most reliable. Within the BAAH topology, further explorations were made of
the relationship between system reliability and the relative placement of loads and generators
within the topology. An optimized placement configuration was algorithmically derived,
conferring a marginal improvement in reliability beyond the gains made through altering the
overall system topology.
2.1 Reliability Concepts
Reliability analysis is, in general, the evaluation of how often systems or pieces of equipment are
expected to fail, and how long such a failure is expected to persist before being repaired and
returning to service. In order to quantify reliability into one or more indices, a definition of
failure must be selected, along with the modes by which failure is possible. In the context of
distribution systems, reliability is split into two related concepts: component reliability and
system reliability.
2.1.1 Component Reliability
Component reliability analysis assesses the expected frequency and duration of physical failures
of distribution system components, such as circuit breakers, buses, and power converters. In this
study, component failures are grouped into three types: passive failures, active failures, and stuck
breakers.
Passive failures cause the failed component to act as an open circuit, preventing power from
flowing through the component. Passive failures only affect the failed component. An example
of a passive failure is a circuit breaker false trip.
Active failures, also referred to as short-circuit faults or overcurrents, not only disable the failed
component, but also cause all adjacent overcurrent protective devices (i.e., circuit breakers) to
trip and isolate the fault. Faults propagate through buses, stopping only at each successfullyopened circuit breaker. Examples of active failures include a bus short circuit or insulation
breakdown in a circuit breaker or cables.
A stuck breaker occurs when a circuit breaker is called upon to isolate a fault but fails to operate.
When this occurs, the fault propagates through the stuck breaker and must be contained by
upstream breakers.
2.1.2 System Reliability
System reliability analysis, the focus of this work, assesses the expected frequency and duration
of service interruptions to equipment loads served by the distribution system caused by
component failures. Here, a service interruption to an equipment load is defined as the load
being electrically isolated from all generation units.
The notional shipboard distribution systems used here (see Section 2.4) serves five equipment
systems: propulsion, energy storage, radar, pulsed loads (e.g., weapons systems), and zonal load
4
centers (secondary distribution circuits serving lighting, refrigeration, etc.). The reliability of
each equipment system is evaluated separately.
2.2 Component Reliability Indices
Component reliability is quantified through two indices: failure rate, denoted », and mean time to
repair (MTTR). The failure rate is defined as the expected number of failures a given component
will experience over the course of one year. The MTTR is defined as the expected length of
time, in hours, that the component failure will persist before it is repaired. The inverse of MTTR
is called the repair rate, denoted À.
With the exception of stuck breakers, which by definition must occur simultaneously with an
adjacent active failure, component failures are assumed independent of one another. Failure and
repair rates are assumed to be constant, making component failures and repairs Poisson
processes. In other words, the waiting times to a failure or a repair are given by exponential
probability distributions.
Each type of component has one set of reliability indices for each type of applicable component
failure. The values for the component failure reliability indices used in this analysis are shown
in Table 1, taken either from manufacturer data or from independent testing [7], [8], [9]. Note
that the failure rate of stuck breaker failures is modeled differently than other failures. This is
explained in further detail in Section 2.3.3.
Table 1: Component Failure Reliability Indices
» (failures
per year)
0.010
0.010
0.010
0.006
0.006
5%
Component Failure
Circuit Breaker – Passive
Circuit Breaker – Active
Bus – Active
Converter – Passive
Converter – Active
Circuit Breaker – Stuck
MTTR
(hours)
4
4
8
1
1
1
2.3 System Reliability Indices
Equipment system reliability is quantified through two indices: the service interruption rate,
denoted μ, and the system MTTR. The service interruption rate is defined as the expected
number of service interruptions that the equipment system will experience due to component
failures over the course of a year. The system MTTR is defined as the expected number of hours
that a service interruption will persist before service is restored through repairs to failed
components. A third index, total expected downtime, is the product of the interruption rate and
MTTR, defined as the expected number of hours per year that the equipment system will spend
in an interrupted state. The derivation of these system reliability indices for each equipment load
is accomplished through a two-part process.
5
First, fault-tree analysis is used to identify a complete list of interruption scenarios for a given
equipment load [10]. An interruption scenario is a minimal set of one or more concurrent
component failures that cause the load in question to become disconnected from all generators.
The number of individual component failures involved in an interruption scenario is called the
scenario’s order. Interruption scenarios up to second-order are considered, as third- and higherorder failures are exceptionally rare and therefore do not greatly affect reliability indices [4], [5].
Second, reliability indices are derived for the equipment system through the use of Markov
models [10]. In a Markov model, the system is assigned a set of states that it can potentially be
found in, along with a set of rates of flow between states. A flow rate represents the rate of
change of the probability of the system being found in a given state. Flows out of a state lower
the probability of being found in that state, while flows into the state raise this probability. The
load's reliability indices are derived through such a model from the component reliability indices
(failure rate » and MTTR) shown in Table 1. Each interruption scenario is simulated in a
Markov model, with each state of the model representing a combination of working and failed
components. Flow rates between these states are defined by the applicable component failure
rates » and repair rates À, as described in Section 2.2.
There are three types of Markov models used to model an equipment system’s various
interruption scenarios: those representing first-order scenarios, second-order scenarios that do
not involve a stuck breaker, and second-order scenarios that do involve a stuck breaker.
2.3.1 First-Order Interruption Scenarios
In the case of a first-order interruption scenario, there are only two states: component functioning
(state 1) and component failed (state 2). The system will be interrupted in state 2. At time t = 0,
we assume the component begins in a functioning state. In other words, p1(t = 0) = 1 and p2(t =
0) = 0, where p1 and p2 are the probabilities of the system being in states 1 and 2, respectively, as
functions of time. As time progresses, these probabilities will change, governed by the
differential equation
‫݌‬ሶ ௜ (‫ = )ݐ‬෍ ior z udwh(݆ ՜ ݅) ‫݌ כ‬௝ (‫ )ݐ‬െ ෍ ior z udwh(݅ ՜ ݆) ‫݌ כ‬௜ (‫)ݐ‬.
௜ஷ௝
(1)
௜ஷ௝
In this two-state Markov model, the flowrate from state 1 to state 2 is the component failure rate
», while the flowrate from state 2 to state 1 is the component repair rate À, as shown in Fig. 1.
Thus, the system of differential equations governing the behavior of this model can be expressed
as
൤
‫݌‬ሶଵ (‫)ݐ‬
െߣ
൨=ቂ
‫݌‬ሶ ଶ (‫)ݐ‬
ߣ
ߨ ‫݌‬ଵ (‫)ݐ‬
ቃ൤
൨.
െߨ ‫݌‬ଶ (‫)ݐ‬
(2)
6
Fig. 1: Markov model of a first-order interruption scenario.
As component failure rates tend to be very small, on the order of years between failures, the
long-term behavior of the model must be considered. As t approaches infinity, the state
probabilities will tend to steady-state values, P1 and P2. These values can be obtained by setting
the differential terms in (2) to 0.
0
െߣ
ቂ ቃ=ቂ
0
ߣ
ߨ ܲଵ
ቃ൤ ൨
െߨ ܲଶ
(3)
As states 1 and 2 are mutually exclusive, it is also known that
ܲଵ + ܲଶ = 1.
(4)
Substituting (4) for the bottom equation of (3), we obtain a system of two linearly independent
equations and two unknowns.
0
െߣ
ቂ ቃ=ቂ
1
1
ߨ ܲଵ
ቃ൤ ൨
1 ܲଶ
(5)
Solving for P1 and P2, we obtain
ߨ
ߣ+ߨ
൥ ൩ = ൦ ߣ ൪.
ܲଶ
ߣ+ߨ
ܲଵ
(6)
The total scenario interruption rate μs is given by the component failure rate (i.e., the flowrate
from state 1 to state 2) times the probability of being in state 1, divided by the probability of not
being in state 2 (in other words, the conditional rate of transition from a working state to a failed
state, given that the system is not already in a failed state). From (4), we know that 1-P2 = P1.
Thus, the total scenario interruption rate is given by
ߤ௦ =
ߣ ‫ܲ כ‬ଵ ߣ ‫ܲ כ‬ଵ
=
= ߣ.
1 െ ܲଶ
ܲଵ
(7)
The total scenario repair rate Às is similarly given by
ߨ௦ =
ߨ ‫ܲ כ‬ଶ ߨ ‫ܲ כ‬ଶ
=
= ߨ.
1 െ ܲଵ
ܲଶ
(8)
The total scenario MTTR is thus given by MTTRs = Às-1 = À-1 = MTTR.
7
2.3.2 Second-Order Interruption Scenarios Not Involving Stuck Breaker
In a second-order interruption scenario, there are four states: both components functioning (state
1), component 1 failed and component 2 functioning (state 2), component 1 functioning and
component 2 failed (state 3), and both components failed (state 4). As interruption scenarios are
assumed to be minimal sets of component failures, the system will be interrupted only in state 4.
The Markov model for a second-order interruption scenario is visually represented in Fig. 2.
Fig. 2: Markov model of a second-order interruption scenario.
From (1), the system of differential equations governing the behavior of this system is
‫݌‬ሶ (‫)ݐ‬
‫)ݐ( ݌‬
െ(ߣଵ + ߣଶ )
ߨଶ
0
ߨଵ
‫ ۍ‬ଵ ‫ۍ ې‬
‫ ۍې‬ଵ ‫ې‬
(‫)ݐ‬
)
(‫)ݐ‬
‫݌‬ሶ
‫݌‬
െ(ߣ
+
ߨ
ߣ
0
ߨ
ଵ
ଶ
ଵ
ଶ
‫ ێ‬ଶ ‫ێ=ۑ‬
‫ ێ ۑ‬ଶ ‫ۑ‬.
ߨଵ
ߣଶ
0
െ(ߣଵ + ߨଶ )
‫݌ێ‬ሶ ଷ (‫ێ ۑ)ݐ‬
‫݌ێ ۑ‬ଷ (‫ۑ)ݐ‬
0
ߣଶ
ߣଵ
െ(ߨଵ + ߨଶ )‫݌ۏ ے‬ସ (‫ے)ݐ‬
‫݌ۏ‬ሶସ (‫ۏ ے)ݐ‬
(9)
Following the same procedure used above, steady-state probabilities P1, P2, P3, and P4 are
calculated using a software script for a given set of component failure and repair rates. Using
these values, the scenario interruption and repair rates are determined by (10) and (11). For the
approximation made in (10), note that P4 is very small relative to the other state probabilities due
to the fact that component failure rates are much smaller than their respective repair rates.
ߤ௦ =
ߣଵ ‫ܲ כ‬ଷ + ߣଶ ‫ܲ כ‬ଶ
ൎ ߣଵ ‫ܲ כ‬ଷ + ߣଶ ‫ܲ כ‬ଶ
1 െ ܲସ
(10)
8
(ߨଵ + ߨଶ ) ‫ܲ כ‬ସ
(ߨଵ + ߨଶ ) ‫ܲ כ‬ସ
=
= ߨଵ + ߨଶ
1 െ (ܲଵ + ܲଶ + ܲଷ )
ܲସ
(11)
ିଵ
P WWU௦ = ߨ௦ିଵ = (ߨଵ + ߨଶ )ିଵ = (P WWUଵିଵ + P WWUିଵ
ଶ )
(12)
ߨ௦ =
2.3.3 Second-Order Interruption Scenarios Involving Stuck Breaker
Because stuck breaker failures must occur in conjunction with an active failure, second-order
interruptions involving a stuck breaker failure are modeled differently than the method described
in 2.3.2. It is assumed that a given breaker, when exposed to an active failure, has a 5% chance
of being stuck [9]. Therefore, the interruption rate of a second-order interruption scenario
involving a stuck breaker and an active failure with failure rate »1 is given by
Ɋ௦ = 0.05 ‫ כ‬ɉଵ .
(13)
Because these two failures must occur simultaneously, the scenario MTTR is simply the lesser of
the two component MTTRs. From the values in Table 2, this value will always equal 1 hour.
2.3.4 System Markov Model
Once the reliability indices for each interruption scenario of a given system interruption have
been derived, the overall system reliability indices can be calculated. For a system with n
associated interruption scenarios, the system can be represented by a Markov model with n+1
states, as shown in Fig. 3. State 1 represents the functioning system, while states 2 through n+1
each represent one of the system’s interruption scenarios. The flowrate from state 1 to state j is
the scenario interruption rate of the interruption scenario associated with state j, while the reverse
flowrate is that interruption scenario’s repair rate.
9
Fig. 3: Markov model representing an equipment system.
Being in any state other than state 1 represents a system interruption, so the system interruption
rate is equal to the total flowrate out of state 1, the sum of the scenario interruption rates.
௡
ߤ௦௬௦
σ௡௜ୀଵ ߤ௦௜ ‫ܲ כ‬ଵ
=
= ෍ ߤ௦௜
ܲଵ
(14)
௜ୀଵ
In (14), μsi is the interruption rate of the ith interruption scenario. Meanwhile, the total system
repair rate is given by the sum of the scenario repair rates weighted by the relative likelihood of
being in a given scenario’s state, divided by the likelihood of being in any of the interruption
scenario states.
ߨ௦௬௦ =
σ௡௜ୀଵ ߨ௜ ‫ܲ כ‬௜ାଵ
= P WWUିଵ
௦௬௦
σ௡௜ୀଵ ܲ௜ାଵ
(15)
Both of these values are calculated for each equipment system interruption through a software
script.
10
2.4 Distribution System Topologies
2.4.1 Ring Bus
The ring bus topology is the basis of most current shipboard electrical distribution systems. As
the name suggests, it consists of a ring of conducting busbar, usually arranged in a rectangular
shape, with several incoming or outgoing conducting lines attached. Incoming lines are attached
to power sources, while outgoing lines are attached to loads. Each pair of adjacent lines is
separated using a bus tie circuit breaker.
In a shipboard DC distribution system based on ring bus topology, the ring of busbar runs around
the perimeter of the ship. Incoming and outgoing lines are connected to buses running along the
port and starboard sides of the ship, with two cross-hull buses connected at the bow and stern to
complete the ring. Some loads are connected to one point on the ring bus, while others (those
spanning the width of the ship, such as zonal load centers) are connected at two different points.
The notional ring bus-based DC distribution system analyzed in this study is shown in Fig. 4.
=
AC Circuit
Breaker
DC
Circuit
Breaker
=
AC Circuit
Breaker
=
у
=
DC/DC
Converter
Drive Inverter
=
=
=
Zone 3 Load
Center
Zone 2 Load
Center
Zone 1 Load
Center
DC/DC
Converter
у
=
Radar Zone 4 Load
Center
AC/DC
Converter
AC Circuit
Breaker
у
AC Circuit
Breaker
Auxiliary AC
Generator 2
DC/DC
Converter
Starboard
Propulsion
Motor
Main AC
Generator 2
у
=
Drive Inverter
Port
Propulsion
Motor
=
=
=
у
DC/DC
Converter
AC/DC
Converter
у
Energy Storage
Auxiliary AC
Generator 1
=
Main AC
Generator 1
AC/DC
Converter
AC/DC
Converter
Fig. 4: A shipboard distribution system with ring bus topology.
11
=
Pulsed
Load
2.4.2 Breaker-and-a-Half
The BAAH topology consists of two parallel lengths of busbar connected by several conducting
lines, called bays. Each bay is attached to two lines, either incoming or outgoing, and is
protected by three circuit breakers. One, called the common breaker, separates the two attached
lines from each other. The other two, called the outside breakers, separate each line from its
adjacent bus. As there are three circuit breakers for every two incoming or outgoing lines, each
line is said to be protected by a “breaker-and-a-half”. A simple representation of the BAAH
topology is shown in Fig. 5. Note that each circuit breaker is outfitted with two disconnect
switches. These are for maintenance purposes, and will not be drawn in subsequent figures.
Source
Load
Load
Load
Fig. 5: A simple breaker-and-a-half topology.
As shown in Fig. 5, bays may be attached to two loads or to a load and a source. The placement
of loads with respect to sources is especially important in a BAAH topology, as loads sharing a
bay with a power source will be more reliable than those sharing a bay with another load. To
illustrate this, observe the results of active faults occurring concurrently on both buses of the
system in Fig. 5. While the two loads on the right-hand bay will see an interruption, the load
sharing a bay with the source will remain operational. In this way, vital loads can be afforded a
higher level of protection by placing them on the same bay as a generator.
In a shipboard DC distribution system based on BAAH topology, the two buses run along the
port and starboard sides of the ship, with the bays running across the hull. A BAAH topology
requires roughly 1.5 times as many circuit breakers as a ring bus topology with the same number
of incoming and outgoing lines. As space and cost are often of great concern when designing a
naval vessel, two versions of a BAAH-based DC distribution system are analyzed in this study.
Version one, shown in Fig. 6, contains roughly the same number of circuit breakers as the ring
bus configuration in Fig. 4. This is achieved by eliminating some of the redundant connections
used in the distribution system shown in Fig. 4 (specifically, those of the radar and zonal load
centers). Version two, shown in Fig. 7, is connected with the same amount of redundancy as the
system in Fig. 4, but uses a greater number of circuit breakers.
12
Main AC
Generator 1
у
=
у
Port
Propulsion
Motor
Drive Inverter
=
AC Circuit
Breaker
Auxiliary AC
Generator 1
AC/DC
Converter
=
AC/DC
Converter
AC Circuit
Breaker
у
Zone 3 Load
Center
DC
Circuit
Breaker
Energy
Storage
Radar
Zone 2 Load
Center
Zone 4 Load
Center
=
=
=
DC/DC
Converter
Zone 1 Load
Center
=
DC/DC
Converter
DC/DC
Converter
AC/DC
AC Circuit
Converter
Breaker
у
Drive Inverter
у
=
Auxiliary AC
Generator 2
Starboard
Propulsion
Motor
у
=
AC Circuit
Breaker
=
=
Pulsed
Load
=
AC/DC
Converter
Main AC
Generator 2
Fig. 6: A shipboard distribution system with breaker-and-a-half topology (version one).
Main AC
Generator 1
у
у
Drive Inverter
=
AC Circuit
Breaker
DC
Circuit
Breaker
Auxiliary AC
Generator 1
AC/DC
Converter
=
у
DC/DC
Converter
=
AC/DC
Converter
AC Circuit
Breaker
Port
Propulsion
Motor
=
Pulsed
Load
DC/DC
Converter
=
Energy
Storage
=
Radar
Zone 4 Load
Center
Zone 3 Load
Center
Zone 2 Load
Center
=
=
у
Starboard
Propulsion
Motor
Drive Inverter
AC Circuit AC/DC
Breaker Converter
Main AC
Generator 2
у
=
Auxiliary AC
Generator 2
=
DC/DC
Converter
DC/DC
Converter
AC/DC AC Circuit
Converter Breaker
Zone 1 Load
Center
=
=
Fig. 7: A shipboard distribution system with breaker-and-a-half topology (version two).
13
=
у
=
2.4.3 Breaker-and-a-Half with Additional Bus Tie Circuit Breakers
The BAAH topology’s reliability can be improved by the addition of another pair of circuit
breakers used to sectionalize the buses into two halves. With this modification, active faults on a
bus or an outside breaker will be confined to one side of the bus rather than propagating through
the entire bus. Modified versions with added bus tie breakers of both BAAH topologies
discussed above are also analyzed. In version one (Fig. 6), the additional breakers are placed
between bays 3 and 4, counting from the stern. In version two (Fig. 7), the additional breakers
are placed between bays 4 and 5, counting from the stern.
2.4.4 Double Breaker, Double Bus
Like the BAAH topology, the double breaker, double bus (DBDB) arrangement consists of two
parallel buses connected by cross-hull bays. In the DBDB topology, however, each bay contains
only one connecting line to a generator or load, protected by two circuit breakers, one adjacent to
each bus. This topology, therefore, contains an even greater number of circuit breakers than are
found in the BAAH topology. A simple comparison of the ring bus, BAAH, and DBDB
topologies is shown in Fig. 8.
(b)
(a)
(c)
Fig. 8: Comparison of (a) ring bus; (b) breaker-and-a-half; and (c) double breaker, double bus topologies.
Note that the DBDB topology was not explored with the expectation that it might be a superior
topology in terms of reliability, but rather as a way to strengthen our assertion that the BAAH
topology represents the most reliable configuration available. As version two of the BAAH
topology (Fig. 6) requires a greater number of circuit breakers than the ring bus, one may be led
to believe that the increased reliability observed in the BAAH topology is merely a result of this
increased number of protective devices, rather than an advantage conferred by the topology
itself. The reliability indices of the DBDB topology will serve to demonstrate that the reliability
of a distribution system is not merely a function of the number of protective devices present.
14
For the consideration of cost and space concerns, a comparison of the number of distribution
system components required for each of the topologies mentioned here is shown in Table 2.
Table 2: System Topology Component Count Comparison
System
Topology
Ring Bus
BAAH v1
BAAH v2
BAAH w/ bus
breaker v1
BAAH w/ bus
breaker v2
DBDB
AC
Breaker
4
4
4
Number of Components Required
DC
AC/DC DC/DC
Drive
Breaker
Conv.
Conv.
Inverter
36
4
4
2
34
4
3
2
45
4
4
2
4
36
4
3
2
4
47
4
4
2
4
54
4
4
2
2.5 Topology Reliability Comparisons
2.5.1 Reliability Comparison Procedure
The reliability index derivation process described in Section 2.3 is performed through a software
script that represents the distribution system topology with a binary incidence matrix, each row
or column representing a generator, load, or distribution system component. Fault-tree analysis
is performed by simulating component failures through changes in the incidence matrix, and
reliability indices are derived using the equations of the Markov model approach described in
Sections 2.3.1-2.3.4.
In order to facilitate simple comparisons between different topologies, a single overall
interruption rate is calculated as a weighted sum of each load's interruption rate. The weights are
used to reflect the relative severity of an interruption to each load. For example, an interruption
to the radar or weapons system can be potentially catastrophic to crew safety or mission success,
while the zonal load centers of the ship are designed such that each has a level of redundancy
with respect to one another [1]. This overall interruption rate is calculated as follows:
ߤ௢௩௘௥௔௟௟ = 1.5 ‫ כ‬൫ߤ௥௔ௗ௔௥ + ߤ௣௨௟௦௘ௗ ൯ + ߤ௣௥௢௣௨௟௦௜௢௡ + 0.5 ‫ߤ( כ‬௦௧௢௥௔௚௘ + ߤ௭௢௡௘௦ )
(16)
2.5.2 Reliability Comparison Results
Comparisons of equipment system and overall reliability indices derived for each distribution
system topology discussed in Section 2.4 are shown in Table 3 and
15
Table 4, respectively. Reliability index comparisons between the ring bus and various BAAH
topologies are shown in Fig. 9, Fig. 10, and Fig. 11.
Table 3: Equipment System Reliability Indices by Distribution System Topology
Zonal Load Centers
Pulsed Load
Radar
Energy Storage
Propulsion
Equipment
System
Distribution System
Topology
Ring Bus
μ (interruptions
per year)
0.108413344
MTTR
(hours)
3.21388100
Total Downtime
(hours per year)
0.348427587
BAAH v1
0.113011050
3.12380331
0.353024293
BAAH v2
0.115015708
3.08683032
0.355033973
BAAH w/ bus breaker v1
0.110007854
3.18174544
0.350016987
BAAH w/ bus breaker v2
0.111009909
3.16206913
0.351021005
DBDB
0.154034520
2.55834012
0.394072693
Ring Bus
0.068501808
3.59849955
0.246503726
BAAH v1
0.056508847
3.12375107
0.176519571
BAAH v2
0.057512922
3.08675201
0.177528129
BAAH w/ bus breaker v1
0.055254589
3.17186077
0.175259863
BAAH w/ bus breaker v2
0.055756005
3.15235618
0.175762786
DBDB
0.087034703
2.37920108
0.207073059
Ring Bus
0.012904640
3.32508920
0.042909080
BAAH v1
0.056501911
3.12385345
0.176503690
BAAH v2
0.000003960
1.86764208
0.000007395
BAAH w/ bus breaker v1
0.054751150
3.19175534
0.174752274
BAAH w/ bus breaker v2
0.000003092
1.80095904
0.000005569
DBDB
0.020036927
1.00199000
0.020076801
Ring Bus
0.068201808
3.60992960
0.246203726
BAAH v1
0.056008128
3.14272392
0.176018083
BAAH v2
0.057512922
3.08675201
0.177528129
BAAH w/ bus breaker v1
0.054754064
3.19170405
0.174758767
BAAH w/ bus breaker v2
0.055756005
3.15235618
0.175762786
DBDB
0.087034703
2.37920108
0.207073059
Ring Bus
0.012708581
3.36136204
0.042718141
BAAH v1
0.175010411
3.74276599
0.655023014
BAAH v2
0.000015799
2.12721907
0.000033608
BAAH w/ bus breaker v1
0.170006484
3.82346524
0.650013882
BAAH w/ bus breaker v2
0.000007603
2.01802703
0.000015342
DBDB
0.026041278
1.00172538
0.026086209
16
Table 4: Overall Interruption Rate by Distribution System Topology
Distribution System Topology
Overall Interruption Rate
Ring Bus
0.270678211
BAAH v1
0.397536484
BAAH v2
0.230055346
BAAH w/ bus breaker v1
0.386896210
BAAH w/ bus breaker v2
0.222530357
DBDB
0.371179956
Fig. 9: Equipment system interruption rate by distribution system topology.
17
Fig. 10: Equipment system mean time to repair by distribution system topology.
Fig. 11: Equipment system total expected downtime by distribution system topology.
As can been seen, the changes in reliability between topologies varies greatly with the type of
equipment under consideration. For the propulsion system, all five topologies have roughly the
same level of reliability. For energy storage and pulsed loads, breaker-and-a-half topologies are
somewhat more reliable, with almost no variation between the two different versions of the
BAAH topology. This is to be expected, as these loads have no additional redundancy in
conducting lines in version two.
18
For radar, version one of the BAAH topology is significantly less reliable than the ring bus
topology. This is due to the removal of one of the redundant conducting lines from the ring bus
topology. In contrast, version two of the breaker-and-a-half topology represents an improvement
in reliability over the ring bus topology. For zonal load centers, the comparison is similar,
though more exaggerated, to the radar system described above.
For each version of the breaker-and-a-half topology, the addition of bus tie circuit breakers
improved overall reliability, but the improvement was relatively minor. In some cases, the
addition of a bus tie circuit breaker increased the MTTR of the system, usually because the
interruption scenarios prevented by the added breakers tend to have low MTTRs (i.e., stuck
breaker failures).
A reliability comparison between the ring bus, BAAH version two (which will herein simply be
referred to as BAAH), and DBDB topologies is shown in Fig. 12. As shown in Table 2, the
DBDB topology contains a greater number of circuit breakers than both the ring bus and BAAH
arrangements. If one suspected that the superior reliability over the ring bus design
demonstrated by the BAAH described above was merely a matter of an increased number of
protective devices, one would expect the DBDB arrangement to be more reliable still. However,
as can be seen in Fig. 12 and Table 4, this is not the case. In fact, the DBDB topology is less
reliable than both the Ring Bus and BAAH topologies for every equipment load. This supports
the notion that the topology of the BAAH is the cause of its high level of reliability, not merely
the number of components it contains. Therefore, the BAAH topology will be used in the further
exploration of the effects of load and generator placement on reliability.
Fig. 12: Reliability comparison of ring bus; breaker-and-a-half; and double breaker, double bus topologies.
19
2.6 Equipment Placement Algorithm
2.6.1 Equipment Placement Algorithm Procedure
It is important to note that placement of equipment loads within a distribution system is a key
factor in determining service interruption rates. If, for example, the radar and pulsed loads were
to be switched in position in each topology examined in Section 2.5 (holding all other
connections constant), their reliability indices would also be exchanged, approximately.
Therefore, it should be noted that the reliability indices of each equipment load need not apply
specifically to that load, but rather to any load that is placed in that load’s position. These
exchanges will not be necessarily exact, however, due to different power electronic converters
used with different loads, and some equipment systems, such as propulsion, consisting of several
loads.
This relationship between placement of loads and generators and system reliability suggests that,
for any given system topology, there should be an optimal placement configuration for the
system’s loads and generators, for which the system’s overall interruption rate will be at a
minimum.
As mentioned in Section 2.4.2, in the BAAH topology (Fig. 7), loads and generators are
connected to cross-hull lines called bays, which are themselves connected to the buses running
along the port and starboard sides of the ship. There are nine bays, each with two connection
slots, for a total of eighteen slots. There are thirteen generators and loads, collectively called
objects. Eight objects are connected to a single slot (the generators, energy storage unit,
propulsion motors, and pulsed load), while five are connected to two slots each (the radar and the
zonal load centers).
Using the configuration shown in Fig. 7 as a starting point, the equipment placement algorithm
makes a series of swaps, exchanging the slots of two or three objects at a time. The equipment
placement algorithm is based on the particle swarm optimization (PSO) algorithm, in which
several candidate solutions, or "particles", are simulated concurrently. The PSO algorithm tracks
each particle's best solution to date, as well as the global best solution to date, and each particle's
iterative motion in the solution space is influenced stochastically by these best solutions. All
particles eventually converge to the global optimized solution [11].
While the traditional PSO algorithm deals with continuous input functions, the algorithm can be
modified to fit discrete functions in general, and optimal placement problems in particular [12],
[13].
The equipment placement algorithm also simulates several candidate solutions concurrently,
namely as vectors specifying which objects are connected to which slots in each candidate
placement configuration. In each iteration of the algorithm, each candidate solution undergoes
one swap, switching the slot positions of either two one-connection objects, two two-connection
objects, or one two-connection object and two one-connection objects. Once the swap is made,
the overall interruption rate of the new configuration is calculated.
20
The objective of the algorithm is to find the most reliable configuration of objects, thus the goal
is the smallest possible value of the overall interruption rate. The algorithm, then, keeps track of
the lowest overall interruption rate achieved so far by each candidate solution (the individual best
configurations), as well as the lowest overall interruption rate achieved by any candidate solution
so far (the global best configuration).
Before each candidate solution makes its swap, the algorithm checks to see if it is already in its
individual best configuration or in the global best configuration. If it is both, the swap is
random.
If it is in its individual best configuration but not in the global best configuration, the swap will
either be random or be drawn from the global best solution. That is, an object is found in the
global best configuration that is not in the same position in the candidate solution, and that object
is swapped to the same position that it occupies in the global best solution.
If the candidate solution is in neither its individual best nor the global best configuration, the
swap is either random, drawn from the global best configuration, or drawn from its individual
best configuration (in the same manner as described above for the global best solution). The
chances of each kind of swap occurring are summarized in Table 5.
Table 5: Object-Slot Swaps
Swap Type
Random
From Global Best
From Individual Best
Chance of Iterative Swap if Candidate
Configuration is Currently
Global
Individual
Neither
Best
Best
100%
40%
30%
60%
35%
35%
The algorithm, therefore, tends to push the candidate solutions in the direction of the best
solutions found so far, while still allowing for random exploration of the solution space. In order
to expedite the algorithm's progress, several configurations are prohibited, either because they
obviously will not confer an improvement in reliability or because they do not make sense from a
ship-design standpoint. The prohibited configurations are any in which:
x
Two generators are connected to the same bay,
x
Both propulsion motors are connected to the same bay,
x
Both propulsion motors are connected on either the port or the starboard side of the ship,
x
Two zonal load centers are both connected to the same two bays,
x
A two-connection object has both connections on the same bay, or
x
A two-connection object is connected to two non-adjacent bays.
21
A flowchart visualizing the operation of the equipment placement algorithm is shown in Fig. 13.
The algorithm was executed with ten candidate solutions running for three hundred iterations
each.
Base
Configuration
Candidate
Configuration
Is New
Configuration
Individual Best?
Next Candidate
Yes
Store
Is Candidate
Configuration...
Is New
Configuration
Global Best?
Global Best?
Individual
Best?
Neither?
Store
No
Random Swap
Global Best
Swap
Yes
Individual Best
Swap
Yes
Is new
configuration
allowed?
Fig. 13: Operational procedure of the equipment placement algorithm.
2.6.2 Equipment Placement Algorithm Results
The equipment placement algorithm identified a configuration of loads and generators within the
BAAH topology that improves upon the reliability of the initial base case, shown in Fig. 7. This
new configuration is shown in Fig. 14. The modified configuration groups the zonal load centers
closer together in the middle of the ship, putting the radar at the stern and leaving the pulsed load
at the bow. Generators were moved such that the pulsed load and both of the radar’s connections
share their bays with generation units. The improvements conferred by this modified
configuration were relatively small, suggesting that the configuration shown in Fig. 7 was close
to optimal to begin with. This is perhaps not surprising, as the ship design upon which Fig. 7
was based was iteratively developed over time by teams of engineers with reliability as an
important, if not rigorously quantified, concern [1], [2], [3].
22
AC/DC
Converter
Auxiliary AC
Generator 1
=
=
Drive Inverter
у
AC Circuit
Breaker
=
у
DC/DC
Converter
у
AC/DC
Converter
Main AC
Generator 1
AC Circuit
Breaker
Port
Propulsion
Motor
=
=
Pulsed
Load
DC
Circuit
Breaker
DC/DC
Converter
=
Energy
Storage
=
Radar
Zone 4 Load
Center
Zone 3 Load
Center
=
=
=
DC/DC
Converter
AC/DC AC Circuit
Converter Breaker
=
Auxiliary AC
Generator 2
у
Starboard
Propulsion
Motor
AC Circuit AC/DC
Breaker Converter
у
DC/DC
Converter
Main AC
Generator 2
Drive Inverter
=
=
Zone 1 Load
Center
Zone 2 Load
Center
у
=
Fig. 14: Optimal equipment placement within the breaker-and-a-half topology, as determined by the equipment
placement algorithm.
The overall interruption rate of this modified configuration is 0.23003772, compared to the base
case value of 0.23005534. Comparisons of the reliability indices of this configuration to the base
case are shown in Table 6. The changes in reliability indices for the radar, pulsed loads, and
zonal load centers are shown in Fig. 15, Fig. 16, and Fig. 17, respectively. In these figures, a
positive improvement in a reliability index refers to a decrease in the index’s value, and vice
versa.
Table 6: Equipment Configuration Reliability Index Comparison
Equipment System
Modified
Base Case
Propulsion
μ (interruptions/
year)
0.115015708
3.08683031
Total Downtime
(hours/year)
0.355033972
MTTR (hours)
Energy Storage
0.057512922
3.08675200
0.177528128
Radar
0.000003868
1.86449387
0.000007213
Pulsed Loads
0.057512922
3.08675200
0.177528128
Zonal Load Centers
0.000015982
2.12571371
0.000033972
Propulsion
0.115015708
3.08683031
0.355033972
Energy Storage
0.057512922
3.08675200
0.177528128
Radar
0.000002133
1.66866338
0.000003560
Pulsed Loads
0.057502877
3.08690691
0.177506027
Zonal Load Centers
0.000016073
2.12499943
0.000034155
23
Fig. 15: Changes in the radar system’s reliability indices between the initial and modified equipment
configurations.
Fig. 16: Changes in the pulsed load system’s reliability indices between the initial and modified equipment
configurations.
24
Fig. 17: Changes in the zonal load center system’s reliability indices between the initial and modified equipment
configurations.
The reliability indices of the propulsion and energy storage systems are unchanged between the
two configurations. This is to be expected, as only the starboard propulsion motor changed
positions from the initial configuration, and no generators were moved such that they share a bay
with any load in either system.
Radar and pulsed loads both saw reductions of their respective interruption rates. As these two
interruption rates were the mostly heavily weighted in (16), the overall interruption rate that the
algorithm was attempting to optimize, this is to be expected, as well. The improvement in
interruption rate was quite significant for the radar system, while very minor for the pulsed loads.
The radar system saw a significant improvement in its MTTR, as well, reducing the expected
duration of service interruptions by about 12 minutes. The pulsed loads experienced a very
minor increase in MTTR, but this was offset by the lower interruption rate, resulting in an overall
improvement to total expected downtime.
The zonal load centers experienced a small increase in interruption rate and a very small
decrease in MTTR, resulting in a small increase in total expected downtime. As the zonal load
centers were lightly weighted in (16), along with the large number of slots taken up by the load
centers, it is expected that the algorithm would sacrifice some of the reliability of these loads in
order to improve more heavily weighted load systems such as radar and pulsed loads.
25
3 POWER SYSTEM RELIABILITY AND CONTROL SYSTEMS
For the past year the MSU team worked on formulating the static analysis problem for long term
controls approach considering the minimization of fuel to optimally use the multiple generators.
A literature survey for the possible inclusion of the reliability indices along with the static
optimization for SPS has been conducted. On the other hand, the work was focused on
developing modules for analysis of the transient behaviors of the SPS under fault scenarios and
during the restorations. An optimized reconfiguration scheme is selected based on the long-term
efficiency as well as the dynamic performance. For this year, a simplified ESRDC related SPS
model has been assembled for the purpose of transient study. Initial efforts have been placed into
evaluating the system responses under contingencies on different buses and attempts have been
made to implement model predictive control techniques for generation of a restoration solution
in real-time.
The static optimization problem has been formulated as Mixed Integer Problem (MIP) on Matlab
using CPLEX toolbox. Initial results have shown that the generation dispatch problem can be
well formulated to support the long term controls approach for reliable SPS operation upholding
the survivability and availability. Implementation of the formulation on various mission profiles
have demonstrated the potential of the dispatch control tuning approach on saving the fuel along
with the support for combined controls approach.
In the proposed dynamic control framework, a general reconfiguration scheme that considers
both steady-state global performance and dynamic transient responses during the system
reconfiguration is developed. The control approach uses Model-Predictive control to achieve an
overall optimized operation. The objective of the system reconfiguration is to provide rapid yet
reliable system-wide solution for emergencies like battle damages or system faults. Time
constraints (computational efficiency), and the reliability (dynamic transient response evaluation)
are the main targets of this optimization work. In order to achieve the control objective, a list of
available onboard components is used within the framework including: circuit breakers, bus
transfers, protective devices, generator voltage settings, and load shedding among many.
The prospective of including the reliability along with the controls was surveyed from a range of
literatures. Several quality of service indices including Mean-time-between-failure (MTBF),
Mean-time-to-failure (MTTF) are seen as the possible coefficient that can be considered to
merge with the system level controls formulation. The inclusion of the reliability indices along
with the dispatch control will be started in the next quarter. At the same time, the formulation for
the dynamic controls to address the load sharing, and higher level fault handling will also be
initiated from the next quarter. Dynamic analysis will maintain the real-time system conditions
with the model based controls law.
3.1 Reliability Enhancement of SPS
The necessity to increase the reliability of the shipboard power systems, keeping the pace with
emerging technologies, has opened the horizon for both static and dynamic reliability
considerations. Design issues, long term planning considerations, and economic dispatch among
many others can be solved statically. Design specifications considering the quality of service also
plays a vital role in building the reliable ship. Various indices can be characterized along with
26
the controls for the operational reliability. Dynamic analysis for load sharing, fault handling, and
transient management among many, will boost the survivability in various mission conditions.
Fig. 18: Framework for Reliable Operation.
The concept of two different stages is proposed to deal with the short term and long term
reliability considerations. The research work begins with the Static Analysis, which takes long
term design issues for SPS taking controls into consideration. Long term is referred to in terms of
the controls. Merging reliability with the controls supports the Survivability and Quality of
Service (QoS) [2]. Short term issues are handled in a dynamic basis whereas long term issues are
handled through static analysis. Certain variables are passed from the static analysis block to the
dynamic control block for the decision support.
3.1.1 Static Analysis
3.1.1.1 Power Dispatch Control and Fuel Minimization
Fuel consumption metrics play a significant role when it comes to the reduction of the operation
cost and preserving the fuel for emergency mission conditions [14]. Effective handling of the
fuel metrics considering the security constrained SPS operation can help achieve lower cost and
reliable operation for various mission profiles.
This aspect can be formulated to enhance start/stop table function and can be tested with the
available simulation platform for validation. Different mission profiles are of vital importance
for the SPS management study. Optimization of the dispatch values, on/off periods need to be
considered along with security constraints to support reliability. Several other cost functions
27
including excessive starting and stopping, minimizing exhaust gas emission, high/low loads, and
long term reliability factors can be included as well.
The initial work regarding this challenge is focused on minimizing the operational cost for fuel
consumption and at the same time performing the controls planning for security constrained
reliable SPS operation. The following system was explored for the initial work for steady state
and dynamic study. It consists of 12 buses, 2 Main Generators, 2 Auxiliary generators, 2
propulsion motors, and 6 different loads on different zones.
M
Prop_1
MTG1
L2
L1
L3
L4
L6
L5
ATG1
MTG2
ATG2
M
Prop_2
Fig. 19: SPS architecture under study.
Two generators Main Turbine Generator (MTG) and Auxiliary Turbine Generator (ATG) have
two different set of fuel consumption equations representing MTBU/hr. Specific Fuel
consumption (SFC) curves are the function of the generated power as shown in the equation
below.
Ɍ = ߦ଴ +
௉ି௉೘೔೙
(ߦଶ െ ߦ଴ )
ି௠ቀ
ቁ
௉ି௉೘ೌೣ ቉
‫כ‬
െ
݁
ቈ1
ି௠
1െ݁
(17)
Where,
¾= Specific Fuel Consumption
¾0 = Specific Fuel Consumption at lowest power setting
¾2 = Final Specific Fuel Consumption at rated power
m = exponential parameter that depends upon the power rating
P = Power in MW
Pmin = Minimum Generation Capacity
Pmax = Maximum Generation Capacity
The SFC function is different for the MTGs and ATGs. Fuel efficiency of larger turbines tends to
increase if they run at near peak load. Greater fuel consumption is guaranteed for operation of a
large turbine at some percentage of its peak load as compared to the smaller turbines. SFC
remains uniform for conventional gas turbine for nominal load above 80%, however it raises
significantly below the 50%. [15]
28
ௗ
‫ܨ‬௜௧ = ‫ܨ‬௜ଵ ‫ܫ‬௜௧ + ෍ ‫ܨܫ‬௜ௗ ܲ‫ݔ‬௜௧
(18)
ௗ
where, d, i, & t are indexes of segment, unit and time and ‫ܫ‬ଵ௧ , ‫ܫ‬ଶ௧ ‫{ א‬0,1}, ‫ = ݐ‬1,2,3, …
1. Power Balance
ௗ
ܲ௜௧ = ෍ ܲ‫ݔ‬௜௧
(19)
ௗ
2. Generation Limits
ௗ
ௗ
ௗ
P‫ݔ‬௠௔௫,௜
ߜ௜௧ௗ ൑ ܲ‫ݔ‬௜௧
൑ ܲ‫ݔ‬௠௔௫,௜
‫ܫ‬௜௧
݀=1
ௗ
ௗ
ௗ
P‫ݔ‬௠௔௫,௜
ߜ௜௧ௗ ൑ ܲ‫ݔ‬௜௧
൑ ܲ‫ݔ‬௠௔௫,௜
ߜ௜௧ௗିଵ
2 ൑ ݀ ൑ ܰ‫ ܦ‬െ 1
ௗ
ௗ
0 ൑ ܲ‫ݔ‬௜௧
൑ ܲ‫ݔ‬௠௔௫,௜
ߜ௜௧ௗିଵ
݀ = ܰ‫ܦ‬
(20)
3. System power balance with reserve constraints
෍ ܲ௠௔௫,௜ ‫ܫ‬௜௧ ൒ ‫ܦ‬௧ + ܴ௧
(21)
ேீ
0 ൑ ܴ௧ ൑ 5‫ܫ‬௜௧
where, Dt and Rt are mission dependent.
4. Minimum On/Off time
ൣܺon,௜(௧ିଵ) െ ܶon,௜ ൧ൣ‫ܫ‬௜(௧ିଵ) െ ‫ܫ‬௜௧ ൧ ൒ 0
(22)
ൣܺoff,௜(௧ିଵ) െ ܶoff,௜ ൧ൣ‫ܫ‬௜௧ െ ‫ܫ‬௜(௧ିଵ) ൧ ൒ 0
5. Relationship Between Binary variables
‫ݕ‬௜௧ െ ‫ݖ‬௜௧ = ‫ܫ‬௜௧ െ ‫ܫ‬௜(௧ିଵ)
(23)
‫ݕ‬௜௧ + ‫ݖ‬௜௧ ൑ 1
6. Network Constraints
݃(ܲ௧ , ܳ௧ , ߠ௧ , ܸ௧ ) = 0
ܸ௠௜௡ ൑ ܸ௧ ൑ ܸ௠௔௫
(24)
ܳ௠௜௡ ‫ܫ‬௧ ൑ ܳ௧ ൑ ܳ௠௔௫ ‫ܫ‬
29
Fuel Cost curve for 50Mw Gas Turbine
10000
9000
Fuel Consumption (Kg/Hr)
8000
7000
6000
5000
4000
Fuel_Cost vs. P
fit 2
3000
2000
1000
5
10
15
20
25
Power (MW)
30
35
40
45
50
Fig. 20: Fuel Consumption Curve fitting for 50Mw Gen with quadratic polynomial function.
Fuel cost curve for 5 MW gen
1400
1300
Fuel COnsumption (Kg/Hr)
1200
1100
1000
900
800
700
600
Fuel_Cost1 vs. P 5mw
Poly2 fit
500
400
1
1.5
2
2.5
3
Power (Mw)
3.5
4
4.5
5
Fig. 21: Fuel Consumption curve fitting for 5 MW Gen with quadratic polynomial function.
30
Fig. 22: Shift Factor (4 significant digits)
Table 7 shows the initial snapshot of the results from the above formulation. As realistic data is
being pursued from the reliable sources, we have studied on the close to real time data as per our
assumption. Several levels of load are considered for different profiles representing the missions.
Fuel consumption is reduced with the use of the optimized dispatch configuration. Results show
that more savings can be seen in case of maximum utilization of the particular generator set.
Commitment of the generators also supports the redundancy.
More work is being conducted to take account of the several realistic profiles for complete
mission.
Table 7: Snapshot of Fuel Consumption Metrics
Profile
LoadAvg(MW)
1
2
3
4
5
6
40
50
60
70
80
90
Fuel Consumption (Kg/Hr)
Proportional
Optimized Share
Share
8100
7802
1000
9866
1210
1021
1450
1292
1680
1490
1840
1633
MTG1
1
1
1
1
1
1
Genarator Status
MTG2 ATG1 ATG2
0
0
0
1
1
1
0
1
1
0
0
0
0
0
1
0
0
0
3.1.1.2 Power dispatch control with long term Reliability consideration
The idea of handling fuel consumption and quality of service (QoS) is proposed for long term
reliability consideration. QoS metrics including MTBF and MTTR are being studied for different
configurations. This project will be continued with the Researchers from University of Texas at
Austin. System level formulation will be handled with the reliability indices. Reliability related
details will be exchanged with the University of Texas at Austin. This work will be the extension
of the current work explained in Section 3.1.1.1.
31
Fig. 23: Components of long term system design.
3.1.2 Dynamic Analysis
A model-based control approach is proposed for the dynamic analysis. This section extends our
previous work [16] to support the overall framework as discussed in Fig. 18. Various parameters
can be tuned dynamically including load sharing variables, fault handling variables and transient
management variables among many. We have listed the Transient Power Management in detail
and will be working on other parts as well in future.
Fig. 24: Model based Architecture for SPS Dynamic Analysis.
32
3.1.2.1 Power management with transient management
3.1.2.1.1 Motivations and outlines
One of the top priorities in the design of shipboard power system applications is the reliability
and safety of operation. The evaluation of dynamic responses under transient operating
conditions or contingency scenarios thus plays a very important role as shipboard power system
is a tightly coupled system and very fragile to the partial dynamic changes within the network
which would cause large disturbance in case of emergencies or damages. Previous research
conducted in the reconfiguration area mainly focuses on the static system performance with
regards to certain optimization functions. There is yet no salient research effort on the dynamic
behavior of SPS under disturbance or operation status change. In most of the research work
based on static analysis, it is just assumed that the system can reach a post-contingency stable
operating point. However, for SPS, this is not necessarily true [17]. If the disturbance is severe,
the system may have instability issues or safety/security degradation. Dynamic analysis of the
transient phase is definitely required for stability margin enhancement.
Based on the prior work, there are two main challenges we need to overcome in this framework.
One of the challenges is that the accurate dynamic calculation requires explicit modeling of
system components not only when working under steady-state operations but also transient
conditions. Another challenge is with the computational burden as with straight-forward time
domain simulation as required by real-time calculation. Any small variations on control
parameters would result in a complete re-computation of the whole system.
In the proposed control framework, a general reconfiguration scheme with consideration of both
steady-state global performance and dynamic transient responses during the system
reconfiguration is developed and combined with Model-Predictive control to achieve an overall
optimized operation. The objective of the system reconfiguration is to provide rapid yet reliable
system-wide solution for emergencies like battle damages or system faults. Time constraints (in
another word computational efficiency) and the reliability (dynamic transient responses
evaluation) are the main targets of this optimization work. In order to achieve the control
objective, a list of available onboard components is used within the framework including: circuit
breakers, bus transfers, protective devices, generator voltage settings, load shedding, etc.
3.1.2.1.2 Methodology
For the control framework, a two level hierarchical baseline structure is proposed. The
management includes a steady-state session where the static optimized states are derived based
on power flow calculations. In addition to the ordinary analysis, the transient responses
following the contingency and during the restoration is also evaluated to determine an optimal
control solution for the system recovery including the system operation point. The candidate
control solution needs to maintain the responses brought by the contingency and restorations
under specified safety margins to assure the system stability during transient phase.
33
3.1.2.1.3 Preliminary Model Development for transient study
The preliminary design of the model used for testing transient management controller is built
based on the notional MVDC NGIPS model, which was originally developed by CAPS at FSU
[18][19]. The original model has been simplified and only parts related to terminal voltage
transient study are selected.
Fig. 25: ESHIP under study in 3.1.2.1.
The function of standard modules in the preliminary model can be listed as follows.
Ship service loads are distributed in four zones from bow to stern along the ship and MVDC
power is fed from both port and starboard DC buses. Cross-hull links are between the port and
starboard DC buses to provide the capability of configuring a ring-bus where the power
generation modules and loads are operated on.
Power generation module: The electrical power is provided by two 36MW (45MVA) main twinshaft gas turbine generators and two auxiliary 4MW (5MVA) single-shaft gas turbine generator.
Power distribution modules: For the MVDC topology, one main generator and one auxiliary
generator are on each side of the network (Port bus and Starboard bus).
Switching gears: The on and off status of the switches determines the power flow directions. To
handle unexpected possible damages, there are switches attached to the cross-hull connections.
For every zonal load, there are also switches connecting and disconnecting the unit from the
power distribution network to provide fast isolation of the area affected by faults.
34
Power conversion modules: convert power from AC to DC/DC to AC, or from AC to AC/DC to
DC. Several power conversion modules exist in the model include PCM1, PCM2.As the main
bus is 5kV DC bus, PCM1 converts 5kV to 800V DC bus for zonal DC loads while PCM2
converts DC to AC converter that converts 800V DC to 450V AC.
Power loads: consumes the power generated by the power generation module and supplied from
either power or starboard bus through switch gear. In the preliminary model, electrical loads are
simplified and classified as constant resistive loads and constant power loads. For the
preliminary model, there are 4 service loads connected to the network, two AC loads supplied
from a 450V AC bus and two DC loads supplied from an 800V DC bus. The service loads are
classified as vital loads and non-vital loads under different operation scenarios. Vital loads can
only tolerate a limited interruption of power supply or cannot tolerate any interruption while nonvital loads can tolerate a wider range of power interruptions.
3.1.2.1.4 Trajectory Sensitivity analysis
For the purpose of calculation burden relief, an approach referred to as trajectory sensitivity [20]
analysis is introduced to help with fast solving of complex system dynamic equations. The main
advantage of this analysis approach is that it provides very precise insights into system dynamic
responses due to parameter(s) changes within a short period and with only negligible extra
computational effort [21]. It has been proven to be an effective tool for situations where
traditional time-domain simulation may lack utility. As the trajectory sensitivity analysis is based
on linear approximation, the error between the actual system performance trajectory and
predicted trajectory are compared to assure the accuracy of this approach.
Basic forms of a system consist of differential algebraic equation (DAE) sets are:
‫ݔ‬ሶ = ݂(‫ݔ‬, ‫ݕ‬, ‫)ݑ‬
(25)
0 = ݃(‫ݔ‬, ‫ݕ‬, ‫)ݑ‬
(26)
where x is the vector of state variables like generator rotor angles and speed, y is the vector of
algebraic variables like voltage magnitude and phase angles, and u is the vector of control inputs.
Suppose u0 is the nominal value of the control variables, the basic forms over a tiny time period
[t0,t1] can be rewritten as:
‫ݔ‬ሶ = ݂(‫ݔ‬, ‫ݕ‬, ‫ݑ‬଴ )
(27)
0 = ݃(‫ݔ‬, ‫ݕ‬, ‫ݑ‬଴ )
(28)
and ‫ݔ‬଴ = ‫(ݔ‬0). Suppose the system described in (27) and (28) has a unique answer x(t,x0,u0),
then the solution of (25) and (26) can be expressed as:
‫ݐ(ݔ‬, ‫ݔ‬଴ , ‫ݐ(ݔ = )ݑ‬, ‫ݔ‬଴ , ‫ݑ‬଴ ) + ‫ݔ‬௨ (‫ ݑ()ݐ‬െ ‫ݑ‬଴ ) + higher order components
(29)
‫ݐ(ݕ‬, ‫ݔ‬଴ , ‫ݐ(ݕ = )ݑ‬, ‫ݔ‬଴ , ‫ݑ‬଴ ) + ‫ݕ‬௨ (‫ ݑ()ݐ‬െ ‫ݑ‬଴ ) + higher order components
(30)
35
Here the term ‫ݔ‬௨ (‫= )ݐ‬
ఋ௫(௧,௫బ ,௨)
ఋ௨
is defined as the trajectory sensitivity of state variables x to
control variables u. In the same way, ‫ݕ‬௨ (‫= )ݐ‬
variables y to control variables u.
ఋ௬(௧,௫బ ,௨)
ఋ௨
is the trajectory sensitivity of algebraic
Now the basic DAE sets with respect to u can be expressed as:
‫ݔ‬௨ሶ (‫݂ = )ݐ‬௫ (‫ݔ)ݐ‬௨ (‫ )ݐ‬+ ݂௬ (‫ݕ)ݐ‬௨ (‫ )ݐ‬+ ݂௨ (‫)ݐ‬
(31)
0 = ݃௫ (‫ݔ)ݐ‬௨ (‫ )ݐ‬+ ݃௬ (‫ݕ)ݐ‬௨ (‫ )ݐ‬+ ݃௨ (‫)ݐ‬
(32)
3.1.2.1.5 The control procedures
The first step is to select a prescribed potential contingency and apply it to the system. Sensors
detect the emergency and initiate some fast local regulations to damp the effects of contingency
to the system, e.g. tripping protection relays, trimming off branches.
Subsequently, in a second step we start recording system states and network topologies when the
"fast" local regulation is finished. The measurements are used to perform the steady-state
evaluation.
Next, we perform power flow with the objective of static performance optimization or load
balancing in the steady-state controller, identify a candidate list of static control parameter sets
according to the power flow results:
The control parameter settings are herein denoted as: ui, where ‫ݑ‬௜ = [‫(ݑ‬1), ‫(ݑ‬2), ‫(ݑ‬3), … , ‫)݇(ݑ‬.
Notice here ui contains a total number of k parameters as it represents a coordination of control
inputs for different components. For example, u(1) can be the initial voltage set point of one
generator and u(2) can be the initial voltage set point of another generator. Elements within ui
need to be considered together to evaluate the global system responses.
In accordance to different sets of control inputs ui and a pre-specified steady-state objective
function, the new operation reference points is defined as:
ܵ௜ = ‫ݑ(ܨ‬௜ ) = ‫(ݑ[ܨ‬1), ‫(ݑ‬2), … , ‫])݅(ݑ‬
(33)
As a next step, we run time domain simulation of the post-contingency system, monitor the
system status and store them in the form of system states matrix J, dynamic states x and algebraic
variables y for future trajectory sensitivity analysis:
‫ܬ[ = )ݐ(ܬ‬ଵ , ‫ܬ‬ଶ , … , ‫ܬ‬௧ ]
(34)
‫ݔ[ = )ݐ(ݔ‬ଵ , ‫ݔ‬ଶ , … , ‫ݔ‬௧ ]
(35)
36
‫ݕ[ = )ݐ(ݕ‬ଵ , ‫ݕ‬ଶ , … , ‫ݕ‬௧ ]
(36)
Here t indicates the number of time steps ” t after the contingency.
As part of the process, we use the system states matrix to perform trajectory sensitivity
calculation for each ui. The predicted trajectories are represented in the form of:
ߜ௜ = ‫ݑ(ܩ‬௜ ) = ‫(ݑ[ܩ‬1), ‫(ݑ‬2), … , ‫])݅(ݑ‬
(37)
The determination of G(ui) after time T0 is based on the recorded data including measurements of
JT0, xT0, and yT0.
In the overall optimization formulation, system transient response trajectories, safe operation
margins are combined with steady-state reference operation points to finalize the final control
actions
௞
௜
ԡܵ௜
min ቌ ෍ ܹி௔௖௧௢௥
௜ୀଵ
௞
௞
െ ߜ௜
ԡଶ
+
௜
෍ ܹி௔௖௧௢௥
ೄ
௜ୀଵ
௜
ԡο‫ݑ‬௜ ԡଶ ቍ
ቛߜ௜ െ ‫ܯ‬௜ೄೌ೑೐ ቛ + ෍ ܹி௔௖௧௢௥
ೆ
(38)
௜ୀଵ
Subject to constraints such as ߜ௜ ‫ߜ[ א‬௦௔௙௘ ], ܵ௜ ‫ܵ[ א‬௦௔௙௘ ].
J is non-sparse.
௜
Here ܹி௔௖௧௢௥
is the weighting factor for the variations between the steady-state optimal
operation point and the predicted state trajectory.
ܹ݅‫ ܵݎ݋ݐܿܽܨ‬is the weighting factor for the variations between the predicted state trajectories and
pre-specified safety operation margin for different states ‫ ݂݁ܽܵ݅ܯ‬. It reflects the priority of stability
enhancement during the transient phase.
ܹ݅‫ ܷݎ݋ݐܿܽܨ‬is the weighting factor for the control variations. Naturally we want fewer changes in
the system structure and this factor is scaled within the objective function as well.
The objective function is calculated for every prediction horizon and the control inputs are feed
back into the system for state updates.
37
Apply contingency
Steady-State power
flow calculation
Store postcontingency system
states matrix
Determine candidate
control parameters for
static operation
Perform Time-Domain trajectory sensitivity analysis
for each of the candidate control parameters based on
the recorded system states
Determine the
optimal control
parameters input
Apply the
control to the
Update the
system states
Fig. 26: Control Work Flow.
38
4 FAILURE MODE AND EFFECTS ANALYSIS STUDIES FOR UNDERSTANDING RISKS
This short report provides an overview of the failure mode and effects analysis (FMEA) studies
undertaken to date at CAPS-FSU for the MVDC shipboard power system (SPS) architecture. It is
intended to highlight the approach, benefits and initial outcomes with respect to research
approaches. The important benefits of the preliminary F-FMEA to date are as follows:
x
Indication of relative vulnerability of certain sections, components and subsystems to
focus studies on their respective faults and failures.
x
An understanding of the need for detailed hardware information which would eventually
lead to a more exhaustive hardware FMEA (H-FMEA). This would help to enhance the
detail of the overall analysis.
x
Help focus diagnostic efforts to manage identified risks using well established AI
techniques or the need to develop novel ones.
x
Aid efforts to enhance decision support by tapping into the data-rich FMEA resources.
The emphasis is laid on a functional (F-FMEA) which serves as the most appropriate method to
start analyzing failure cause and effects in a novel system such as the zonal shipboard power
distribution architecture studied. The F-FMEA will be shown in a tabular format in the report.
To summarize, the contents of this preliminary report are:
x
Introduction to the relevance and need to utilize FMEA as a starting point in this
research.
x
Example of a fundamental functional FMEA conducted on the MVDC system.
x
Potential uses of FMEA data for benefiting future research.
This section explains the F-FMEA process used to begin understanding of risks on the RTDSmodel at CAPS. The relevant outputs are highlighted along with future directions for this
particular research. Observing the merits of beginning at a more superficial F-FMEA are evident,
mainly in the fact that the current research is centered around modeled representative systems
with more generic than specific hardware information. F-FMEA is a logical start to
understanding subsystem and system level risks and studying ways to mitigate their effects with
accurate diagnostics.
Even though an F-FMEA is relatively less exhaustive than an H-FMEA, its outcomes form the
driving force behind further research and eventually aid H-FMEA by narrowing down critical
sections and devices. This aids in focusing and informing further research.
39
4.1 The relevance of FMEA for this research
The aim of this research is to thoroughly understand the risks associated with a notional
integrated SPS comprising MVDC zonal power distribution architecture. The anticipated use of
large number of power electronic equipment onboard poses a challenge to de-risk the system
owing to the fact that it is a relatively new and unproven technology for naval applications. The
lack of benchmark systems further emphasizes the need to understand risks and study methods to
mitigate their effects. The notional MVDC zonal architecture modeled on the real time digital
simulator (RTDS) at CAPS-FSU is shown in Fig. 27.
Fig. 27: MVDC zonal architecture modeled on the RTDS.
40
Fig. 28: Zonal load centers modeled on the RTDS.
The model consists of a two zone system (Fig. 27. shows more than two zones to illustrate a
network with more number of segregations) with 4 primary power sources. The power is fed to
longitudinal busses on the port and starboard sides with an option to connect them together to
form a ring. A number or DC breakers are lined along the busses and form part of the protection
system. More details on the modeled system’s constituents can be seen in Table 8 and Table 9.
Fig. 28 shows the zonal loads modeled on the RTDS. Each zone is fed by two dc-dc step-down
converters (buck converters) which draw power from the main dc-buses on either side. The
zones have AC loads which are fed via an AC-DC inverter.
In an effort to understand causes and effects of risks associated with the novel SPS architecture,
FMEA is the most logical starting point. FMEA is an established reliability analysis process
aimed at studying ways in which failure occurs in a system. A thorough FMEA provides a
database of known failures, their known causes, effects and in the process can aid in assessing
the severity of each thereby identifying the most pertinent disturbances. An FMEA could be used
at any stage during system development and is an appropriate starting point to assess risks in a
novel system with limited prior understanding regarding fault manifestations. It makes FMEA
ideal to be used for the notional SPS architecture.
Previous research using this approach has been published in [22] with further applications of
reliability analysis techniques explained in [23]. In both these papers, a robust research
methodology beginning at FMEA is explained. FMEA helps outline pertinent issues, in turn
helping focus further research into identifying and diagnosing disturbances. This methodology
not only helps enhance the risk assessment process for the novel SPS, but also channels efforts
into effective fault diagnostics capabilities.
41
4.2 Two sub-parts of a detailed FMEA
FMEA as a detailed process can be divided into two parts of differing levels of technicality.
These two parts are F-FMEA and H-FMEA and are elaborated in [24]. The fundamental
differences between F-FMEA and H-FMEA are described in this section.
4.2.1 Functional FMEA
This type focuses on the functions that a system, process, or service is to perform rather than on
the characteristics of the specific implementation. When developing a functional FMEA, a
functional block diagram is used to identify the top-level failure modes for each functional block
on the diagram. For example, a heater’s two potential failure modes would be: “Heater fails to
heat” and “Heater always heats”. Another example of a functional FMEA would consider that a
capacitor is intended to regulate voltage and then analyze the effects of the capacitor failing to
regulate voltage. It would not analyze what would occur if the capacitor fails because of an opencircuit or shorted-circuit. As FMEAs are best begun during the conceptual design phase, long
before specific hardware information is available, the functional approach is generally the most
practical and feasible method by which to begin a FMEA, especially for large, complex systems
that are more easily understood by function than by the details of their operation. When systems
are very complex, the analysis for functional FMEAs generally begins at the highest system level
and uses a top-down approach.
4.2.2 Hardware FMEA
This type examines the characteristics of a specific implementation to ensure that the design
complies with requirements for failures that can cause loss of end-item function, single-point
failures, and fault detection and isolation. Once individual items of a system (piece-parts,
software routines, or process steps) are identified in the later design and development phases,
component FMEAs can assess the causes and effects of failure modes on the lowest-level system
items. H-FMEA is also referred to as piece-part FMEAs, and are more common than F-FMEAs
since usually in a system, the individual components are well known and altogether novel
components as such are rare. H-FMEAs generally begin at the lowest piece-part level and use a
bottom-up approach to check design verification, compliance, and validation.
For complex systems, a combination of (a) and (b) may be required which constitutes a “Detailed
FMEA”. In the case of the novel SPS, the combination of F-FMEA and H-FMEA is necessary as
it is a system still in the conceptual phase, without the presence of any hardware based
benchmarks. Fig. 29. illustrates the difference in scope between F-FMEA and H-FMEA showing
that both together constitute a detailed FMEA. Also, FMEA is iterative in nature, needing regular
exchange of and updating of data on failure causes and effects. This is shown by bi-directional
arrows in both F-FMEA and H-FMEA in Fig. 29.
42
Detailed FMEA
Novel system
F-FMEA: Top to
bottom iterative
analysis that may
include devices
Subsystem
(sub-sections, zones etc.)
Constituent devices
Individual components
H-FMEA: Detailed
bottom to top iterative
analysis from component
level going upwards to
device level
Fig. 29: Subtle difference between F-FMEA and H-FMEA that add up to produce a detailed FMEA.
F-FMEA applied on the notional zonal SPS provides information on critical sections and devices
in the network. This output in turn guides the more intensive H-FMEA to focus on such critical
devices for fault studies. Outputs of H-FMEA in turn narrow down vital components whose
faults and failures may lead to disturbances in the sub-system or system that could be termed as
catastrophic (or highly severe). This progressive filtering provides a list of pertinent faults on
which further studies could be centered. The next logical progress would be into testing known
diagnostic methods to differentiate faults or develop novel techniques. Another outcome could
be the development of prognostics techniques to help predict failure times in order to prevent
major faults if possible.
F-FMEA is a logical start to understanding subsystem and system level risks and studying ways
to mitigate their effects with accurate diagnostics. Even though an F-FMEA is relatively less
exhaustive than an H-FMEA, its outcomes form the driving force behind further research and
eventually aid H-FMEA by narrowing down critical sections and devices. This aids in focusing
and informing further research.
4.3 F-FMEA process
F-FMEA is a relatively superficial analysis compared to the more detailed Hardware FMEA (HFMEA). F-FMEA considers the fundamental capability of the system under scrutiny to perform
its function. An inability to do its regular or required “job” is deemed a functional failure. Such
functional failures however could vary in severity which can be assessed during the analysis
process. H-FMEA on the other hand considers individual component-level faults and failures for
which detailed information on parts of a device that in turn make up a device/system are needed.
Such level of detail can get cumbersome and exhaustive; hence F-FMEA forms a reasonable
starting point to the study the novel SPS architecture from view of risk assessment.
Before breaking down a system to perform a failure analysis, it is useful to list out various subparts and constituents in a tabular format and highlight the function of each.
43
4.3.1 F-FMEA applied to the overall MVDC zonal SPS architecture
Fig. 27 shows the representative system. Two zones are modeled on the RTDS for analysis. The
system can be segregated into progressively smaller parts on which F-FMEA can be conducted.
Table 8 shows various subsections that the overall system can be broken down into, listing their
respective functions and constituents.
The subsections in turn can be segregated into constituent devices. The list of devices, their type
and function is shown in Table 9.
4.4 Application of F-FMEA data for further research
F-FMEA helps identify various disturbances that could occur in the system by analyzing known
failures, their causes and effects. This helps understand dependencies of parts of the system on
each other from the risk assessment point of view. The information is used for considering
techniques to diagnose the disturbances that could occur. These techniques need to be intelligent
to provide accurate diagnosis of faults and failures with decision support to onboard crew.
Table 8: Subsections of the architecture with their constituents and functions
Subsection
Energy
storage
subsection
Radar
subsection
Propulsion
motors
subsection
Function(s)
To enable
efficient
charging of
ES such that
it can be
operated and
utilized as
desired.
To provide
continuous
power to
RAD
enabling it
to be
operated
within
acceptable
parameters.
To provide
required
power to
propulsion
motors to
enable their
acceptable
operation as
desired.
Relevant
information
Draws power
from DC bus
which is
processed by
DCDC
converter and
then fed into
ES device.
RAD is
supplied by
DCDC
converters
drawing power
from either of
the two DC
busses. This
subsection is
similar to the
arrangement of
a zone.
DCAC
inverters
convert DC
from the DCbus to AC to
supply
corresponding
motors on
either side.
Constituent
devices
DCDC_P_E
S, ES
DCDC_P_R
AD,
DCDC_S_R
AD, RAD
DCAC_S,
DCAC_P,
PM_S,
PM_P
44
Representative diagram
Pulsed
load
subsection
Zone 1 &
2
DC power
ring
section
To provide
desired
power
enabling
charging of
the pulsed
load
allowing it
to be used as
desired.
To provide
continuous
power to
DCLL and
ACLL.
To provide
continuous
power at
required
rating and
quality to
DC busses.
This
subsection
consists of a
dedicated
DCDC
converter and
a charging
circuit (which
is included as
part of the PL).
Total two
conversions,
DCDC (dc-dc
step down)
conversion
followed by
DCAC (dc-ac)
conversion.
Multiple
conversions
could cause
harmonics and
power quality
issues within
zones in the
power
supplied to
both DC and
AC loads.
Interconnected
power sources
through
distribution
switchboards.
Back-up
power
generators also
connected
through same
ring
DCDC_S_P
L, PL
DCDC_S_Z
1,
DCDC_S_Z
2,
DCDC_P_Z
1,
DCDC_P_Z
2,
DCLL_Z1,
DCLL_Z2,
DCAC_Z1,
DCAC_Z2,
ACLL_Z1,
ACLL_Z2
MTG1,
MTG2,
ATG1,
ATG2,
ACDC_P_
MTG1,
ACDC_S_
MTG2,
ACDC_P_A
TG1,
ACDC_S_A
TG2, DCD
45
DC busses
section
To have
continuous
DC power
flow.
ACDC
rectifiers form
part of this
subsection as
well as DCDC
converters
which further
process the DC
input power to
supply to DC
loads and
zones. DCAC
inverters that
supply
propulsion
motors are also
part of this
subsection as
they draw
power from
the DC bus.
ACDC_P_
MTG1,
ACDC_P_A
TG1,
ACDC_S_
MTG2,
ACDC_S_A
TG2,
DCDC_P_E
S,
DCDC_P_R
AD,
DCDC_S_R
AD,
DCDC_S_P
L,
DCDC_S_Z
1,
DCDC_S_Z
2,
DCDC_P_Z
2,
DCDC_S_Z
2, DCAC_P,
DCAC_S,
DCBUS
4.5 Automating F-FMEA for different modes of the SPS
The F-FMEA Table 8 through Table 16 show a generalized outcome with standard connectivity
information, where the particular mode of the naval vessel is not considered. It is understood that
the SPS will have varying network configuration to cater to different scenarios. During these
modes, the connectivity between devices and subsystems will vary, leading to differences
between dependencies that would eventually change the cause-effect deductions of a general FFMEA.
Such information about network connections during each operational mode, if utilized for an
automated computation of F-FMEA, could provide real-time information about known risks,
causes and effects. Knowing mode-specific F-FMEA information can further be utilized for
aiding the AI based diagnosis performed by the FACS as well as decision support during
different scenarios. An example of different operational modes of the ship is;
1. Ring battle mode (RBM) shown in fig.30(a)
2. Split –plant battle mode (SPBM) shown in fig.30(b)
In the RBM, a bus connection at either end makes a continuous path between all generators
(power sources). This ensures redundancy in the supply for vital subsections such as radar,
pulsed power load, propulsion motors etc.
46
The SPBM as the name suggests, divides the power network. In this case, an equal division is
made and each of the two busses (starboard and port) receives the same amount of power from
connected power sources. The vital loads on either bus can be fed only through their respective
sources. The zonal loads however in both modes mentioned, receive supply from either bus, as
seen from the connectivity diagram in Fig. 30 (a) and (b).
Table 9: List of devices in subsections and their respective functions
Device name
Main turbine generator
Special
abbreviations
MTG1
Type
Main turbine generator
MTG2
Auxiliary turbine
generator
ATG1
Auxiliary turbine
generator
ATG2
Primary power
source
AC-DC rectifier
ACDC_P_MTG1
PEC
AC-DC rectifier
ACDC_S_MTG2
PEC
AC-DC rectifier
ACDC_P_ATG1
PEC
AC-DC rectifier
ACDC_S_ATG2
PEC
DC-DC converter
DCDC_P_ES
PEC
DC-DC converter
DCDC_S_PL
PEC
DC-DC converter
DCDC_P_RAD
PEC
DC-DC converter
DCDC_S_RAD
PEC
DC-DC buck converter
DCDC_P_Z1
PEC
Primary power
source
Primary power
source
Primary power
source
47
Function
Provide continuous power at the specified
rating and quality.
Provide continuous power at the specified
rating and quality.
In case main generators fail, then to provide
continuous power at specified rating and
quality. Provide continuous power in case
general power demand increases.
In case main generators fail, then to provide
continuous power at specified rating and
quality. Provide continuous power in case
general power demand increases.
Convert AC power from generator side input to
DC power at output fed into the DC bus at
specified rating and quality
Convert AC power from generator side input to
DC power at output fed into the DC bus at
specified rating and quality
Convert AC power from generator side input to
DC power at output fed into the DC bus at
specified rating and quality
Convert AC power from generator side input to
DC power at output fed into the DC bus at
specified rating and quality
Convert DC power from bus at input to DC
power at specified values. Provide rated DC
power in a continuous manner to zonal loads
and other vital loads.
Convert DC power from bus at input to DC
power at specified values. Provide rated DC
power in a continuous manner to zonal loads
and other vital loads.
Convert DC power from bus at input to DC
power at specified values. Provide rated DC
power in a continuous manner to zonal loads
and other vital loads.
Convert DC power from bus at input to DC
power at specified values. Provide rated DC
power in a continuous manner to zonal loads
and other vital loads.
Convert DC power from bus at input to DC
power at specified values. Provide rated DC
power in a continuous manner to zonal loads
and other vital loads.
DC-DC buck converter
DCDC_S_Z1
PEC
DC-DC buck converter
DCDC_P_Z2
PEC
DC-DC buck converter
DCDC_S_Z2
PEC
DC-AC inverter
DCAC_P
PEC
DC-AC inverter
DCAC_S
PEC
DC-AC inverter
DCAC_Z1
PEC
DC-AC inverter
DCAC_Z2
PEC
High power radar
RAD
Load
Capacitor banks
ES
High power pulsed
load
PL
Secondary
power source
Load
Propulsion motor
PM_S
Load
Propulsion motor
PM_P
Load
Zonal DC lumped load
Zonal DC lumped load
Zonal AC lumped load
Zonal AC lumped load
DCLL_Z1
DCLL_Z2
ACLL_Z1
ACLL_Z2
Load
Load
Load
Load
Convert DC power from bus at input to DC
power at specified values. Provide rated DC
power in a continuous manner to zonal loads
and other vital loads.
Convert DC power from bus at input to DC
power at specified values. Provide rated DC
power in a continuous manner to zonal loads
and other vital loads.
Convert DC power from bus at input to DC
power at specified values. Provide rated DC
power in a continuous manner to zonal loads
and other vital loads.
Invert DC to AC at required rating. Supply
continuous power to AC motors.
Invert DC to AC at required rating. Supply
continuous power to AC motors.
Invert DC to AC at required rating. Supply
continuous power to AC loads.
Invert DC to AC at required rating. Supply
continuous power to AC loads.
To perform tasks related to navigation and
tracking.
To act as back-up for providing additional
power.
To provide high power weapons capability
under special circumstances and missionmodes.
Propel the vessel at required speed and in the
required direction as needed.
Propel the vessel at required speed and in the
required direction as needed.
General ship loads.
General ship loads.
General ship loads.
General ship loads.
Table 10 through Table 16 show F-FMEA for the various subsections of the overall systems
considered.
Table 10: Energy storage F-FMEA
Function:
Constituents:
No.
1
2
No.
Energy storage subsection
To enable efficient charging of ES such that it can be operated and utilized as desired.
DCDC_P_ES, ES
Constituent
devices
(general)
DCDC
1
ES
1
Failure
mode
Quantity
Type
Power delivery device (specific PEC)
Secondary power source
Cause
Effect
48
Severity and remarks
1
Faulty or
inadequate
charging of
ES.
2
No charging
of ES.
Faulty power output
from DCDC device due
to internal fault.
Faulty power output
from ACDC devices due
to internal fault.
Faulty power output
from primary power
source(s) due to internal
fault.
Power quality issues in
DC busses due to
cabling fault.
Internal fault in ES
device.
No power output from
DCDC device due to
internal failure.
No power output from
ACDC devices due to
internal failure.
No power output from
primary power source(s)
due to internal failure.
No power flow in DC
busses due to cabling
failure or burnout.
Internal failure in ES
device.
Cannot charge
and operate ES
as desired.
Medium – The ES is generally for
back-up during highly specific
scenarios whose occurrence could
be relatively rarer than general
mission-modes.
Cannot operate
ES as desired.
High – This failure mode may be
traced back to serious issues as
stated in the causes unless the ES
itself has a failure in it.
Table 11: Radar F-FMEA
Function:
Constituents:
No.
1
2
No.
1
2
Radar subsection
To provide continuous power to RAD enabling it to be operated within acceptable parameters.
DCDC_P_RAD, DCDC_S_RAD, RAD
Constituent
Quantity
Type
devices (general)
DCDC
2
Power delivery device (specific PEC)
RAD
1
Load
Failure mode
Cause
Effect
Severity and remarks
Cannot operate High – RAD is a vital
Faulty power output from
Faulty or
load and must be in
RAD as
DCDC device due to internal
inadequate
fault.
available at all times.
desired.
operation of
RAD.
Faulty power output from
ACDC devices due to internal
fault.
Faulty power output from
primary power source(s) due to
internal fault.
Power quality issues in DC
busses due to cabling fault.
Internal fault in RAD device.
No operation of
No power output from DCDC
Cannot operate High – RAD is a vital
RAD.
device due to internal failure.
RAD at all.
load and must be in
49
No power output from ACDC
devices due to internal failure.
No power output from primary
power source(s) due to internal
failure.
No power flow in DC busses due
to cabling failure or burnout.
Internal failure in RAD device.
available at all times.
Table 12: Propulsion motors F-FMEA
Function:
Constituents:
No.
1
2
No.
1
2
Propulsion motors subsection
To provide required power to propulsion motors to enable their acceptable operation as desired.
DCAC_S, DCAC_P, PM_S, PM_P
Constituent devices
Quantity
Type
(general)
DCAC
2
Power delivery device (specific PEC)
PM
2
Load
Failure mode
Cause
Effect
Severity and remarks
Cannot operate High – PM is a vital load
Faulty or inadequate Faulty power output from
operation of PM.
DCAC device due to internal PM as desired. and must be in available
fault.
at all times.
Faulty power output from
ACDC devices due to
internal fault.
Faulty power output from
primary power source(s) due
to internal fault.
Power quality issues in DC
busses due to cabling fault.
Internal fault in PM device.
Cannot operate High – PM is a vital load
No operation of PM. No power output from
DCAC device due to internal PM at all.
and must be in available
failure.
at all times.
No power output from
ACDC devices due to
internal failure.
No power output from
primary power source(s) due
to internal failure.
No power flow in DC busses
due to cabling failure or
burnout.
Internal failure in PM device.
Table 13: Pulsed load F-FMEA
Function:
Constituents:
No.
1
Pulsed load subsection
To provide desired power enabling charging of the pulsed load allowing it to be used as desired.
DCDC_S_PL, PL
Constituent devices
Quantity
Type
(general)
DCDC
1
Power delivery device (specific PEC)
50
2
No.
1
PL
1
Failure mode
Faulty or inadequate
operation of PL.
2
No operation of PL.
Cause
Faulty power output from
DCDC device due to
internal fault.
Faulty power output from
ACDC devices due to
internal fault.
Faulty power output from
primary power source(s) due
to internal fault.
Power quality issues in DC
busses due to cabling fault.
Internal fault in PL device.
Internal fault in charging
circuit.
No power output from
DCDC device due to
internal failure.
No power output from
ACDC devices due to
internal failure.
No power output from
primary power source(s) due
to internal failure.
No power flow in DC
busses due to cabling failure
or burnout.
Internal failure in PL device.
Internal failure in charging
circuit.
Load
Effect
Cannot
operate PL as
desired.
Cannot
operate PL at
all.
Severity and remarks
Medium – Though PL is a
vital load, it's availability
is required less frequently
across mission-modes.
High – This failure mode
may indicate more serious
issues towards power
generation side.
Table 14: Zone 1 and 2 F-FMEA
Function:
Constituents:
No.
1
2
3
4
No.
1
Zones 1 and 2
To provide continuous power to DCLL and ACLL.
DCDC_S_Z1, DCDC_S_Z2, DCDC_P_Z1, DCDC_P_Z2, DCLL_Z1, DCLL_Z2, DCAC_Z1,
DCAC_Z2, ACLL_Z1, ACLL_Z2
Constituent
devices
Quantity
Type
(general)
DCDC
4
Power delivery device (specific PEC)
DCAC
2
Power delivery device (specific PEC)
DCLL
2
Load
ACLL
2
Load
Failure mode
Cause
Effect
Severity and remarks
Medium – Depending on every
Cannot operate
Faulty power output
Faulty or
mission, the priority of loads may
from at least one DCDC DCLL as
inadequate
device.
desired and may vary, making it vital to be able to
power input to
affect achieving operate the DCLLs which are
DCLL.
Power quality issues in
needed at the time. Further, faults
mission
DC busses.
in the PEC (in this case DCDC) of
objective.
Internal fault with
zones may cause disturbances in
DCDC device.
51
2
No power input
to DCLL
3
Faulty or
inadequate
power input to
ACLL.
4
No power input
to ACLL
Faulty power output
from ACDC devices.
Faulty power output
from primary power
source(s).
Internal fault in DCLL.
No power output from
at least one DCDC
device.
No power flow in DC
busses.
Internal fault with
DCDC device.
No power output from
ACDC devices.
No power output from
primary power
source(s).
Internal failure in
DCLL.
Faulty power output
from at least one DCDC
device.
Power quality issues in
DC busses.
Internal fault with
DCDC device.
Faulty power output
from ACDC devices.
Faulty power output
from primary power
source(s).
Faulty power output
from DCAC device.
Internal fault in ACLL.
No power output from
at least one DCDC
device.
No power flow in DC
busses.
Internal fault with
DCDC devices.
No power output from
ACDC devices.
No power output from
primary power
source(s).
No power output from
DCAC device.
Internal failure in
ACLL.
52
the DC bus leading to issues for
other zones owing to factors such
as switching harmonics and
current surges.
Possible
system-wide
power outage
and inability to
operate DCLL
in turn
hampering
mission goal(s).
High – This failure mode may
point towards system-wide
disturbances apart from the
obvious hindrance in achieving the
mission goal in case a particular
DCLL is off-line.
Cannot operate
ACLL and may
affect achieving
mission
objective.
High – Similar reasoning to case1. Further, faults in the PEC (in
this case DC-AC converter device)
of zones may cause disturbances
within its zone (propagation of
fault to busses and in turn other
zones is evaded owing to isolation
provided by the DCDC devices)
owing to factors such as switching
harmonics because of the added
DC-AC inversion.
Cannot operate
ACLL and very
likely that entire
zone is without
power.
High – Similar reasoning to case-2
with respect to inability of being
able to operate ACLL for a
particular mission.
Table 15: DC power ring F-FMEA
Function:
Constituents:
No.
1
2
3
No.
1
2
DC ring subsection
To provide continuous power at required rating and quality to DC busses.
MTG1, MTG2, ATG1, ATG2, ACDC_P_MTG1, ACDC_S_MTG2, ACDC_P_ATG1,
ACDC_S_ATG2, DCD
Constituent
Quantity
Type
devices (general)
MTG
2
Primary power source
ATG
2
Primary power source
ACDC
4
Power delivery device (specific PEC)
Failure mode
Cause
Effect
Severity and remarks
Faulty and
High – System wide issue which
Faulty power output
Faulty or
degraded quality
would have impacts on all
from ACDC device
inadequate input
power flow in DC devices and loads.
due to internal fault.
to DC bus.
busses.
Faulty power output
from primary power
source(s) due to
internal fault.
Power quality issues
in DC busses due to
cabling fault.
Internal fault in DC
bus.
High – System wide issue which
No power input
No power output from System wide
power outage.
would have impacts on all
to DC bus.
ACDC devices due to
internal failure.
devices and loads making it
unable to fulfil mission goal(s).
No power output from
primary power
source(s) due to
internal failure.
No power flow in DC
busses due to cabling
failure or burnout.
Table 16: DC busses F-FMEA
Function:
Constituents:
No.
1
2
3
No.
1
DC busses subsection
To provide continuous DC power flow to connected parts.
ACDC_P_MTG1, ACDC_P_ATG1, ACDC_S_MTG2, ACDC_S_ATG2, DCDC_P_ES,
DCDC_P_RAD, DCDC_S_RAD, DCDC_S_PL, DCDC_S_Z1, DCDC_S_Z2, DCDC_P_Z2,
DCDC_S_Z2, DCAC_P, DCAC_S, DCBUS
Constituent
devices
Quantity
Type
(general)
ACDC
4
Power delivery device (specific PEC)
DCDC
8
Power delivery device (specific PEC)
DCAC
2
Power delivery device (specific PEC)
Failure mode
Cause
Effect
Severity and remarks
High – An assessment is needed to
Unable for
Faulty power
Faulty power output
determine whether the fault is
zone to fulfil
flow.
from ACDC device
limited to a zone or has origins
its function.
due to internal fault.
53
2
No power flow.
Faulty power output
from primary power
source(s) due to
internal fault.
Power quality issues in
DC busses due to
cabling fault.
Faulty power output
from zonal DCDC due
to internal fault.
No power output from
ACDC devices due to
internal failure.
No power output from
zonal DCDC due to
internal failure.
No power output from
primary power
source(s) due to
internal failure.
No power flow in DC
busses due to cabling
failure or burnout.
nearer the generation side.
System wide
power outage.
High – System wide issue which
would have impacts on all devices
and loads making it unable to fulfil
mission goal(s).
In both modes, there are fundamental differences to the network architecture, yet there are some
commonalities. These variations change the nature of dependencies that determine functional
failure cause-effect relations in turn modifying the F-FMEA data. Each new mode means the FFMEA information needs to be appropriately altered. While this can be done for a finite number
of modes using a relatively small number of devices and subsections (like on the RTDS model
with 2 zones), it becomes a cumbersome task to produce F-FMEA tables for a real ship system
with its full complement of subsections, devices and over 6 zones typically. In such a case, it
may be prudent to study ways in automating F-FMEA at a more fundamental level such that
network connectivity information could be optimally utilized to produce a real-time F-FMEA.
4.5.1 Multi agent systems technology research
The use of a decentralized multi-agent system (MAS) for reconfiguring the shipboard power
architecture is reported in [25]. The agents are developed in MATLAB while the system is
simulated on a virtual test bed. MATLAB-SIMULINK is extensively used to simulate power
systems in part or whole using the various toolboxes available within the software. The use of
MATLAB to build agents as presented in this paper makes this software in general promising to
be utilized for analyzing and experimenting within this research field.
Feliachi et al propose a distributed scheme with MAS based control agents in [26]. This is to aid
the notion of automated reconfiguration and self-healing in the event of battle damage and other
fault scenarios. With a system utilizing agents, the crucial aspect is the information fed into
individual software agents and its accuracy. Here, the authors aim to implement a graph theoretic
self-stabilizing maximum flow algorithm as the agents’ strategy to ensure efficient power
management which would include considering constraints and load priorities.
54
A MAS with two layers (shown in Fig. 31) for power system reconfiguration is proposed by
Cartes et al in [27]. One layer is the power system layer with a network of devices and the other
layer is the one with software agents. Every device has its agent with whom information can be
exchanged. The communication constraints on every agent are placed such that information
exchange is possible only with a neighboring agent. This paper is one of the first to introduce a
layered MAS where the electrical devices in the hardware layer is mapped onto its respective
agent in the MAS software layer. Simulations are carried out by the authors using a RTDS
model. Cartes et al systematically propose a structured methodology making use of state-of-theart technology to provide a potentially promising intelligent system that may be adopted for the
SPS not only for reconfiguration (as suggested in [27]) but for other tasks such as condition
monitoring, fault diagnosis and perhaps prognosis as well.
A SPS power system restoration scheme using a MAS is proposed by Momoh in [28]. The
rationale given by the author to use a MAS is its decentralized network and local data processing
capability which greatly reduce the computation time and network bandwidth. Another
advantage is the ease of scalability in case newer loads/devices are added to the network and the
subsequent ease of extensibility to carry out required tasks.
55
Vital
loads/subsections
Vital
loads/subsections
Main generators
Bus
connection
Zonal
loads
Zonal
loads
Zonal
loads
Bus
connection
Main generators
Vital
loads/subsections
Vital
loads/subsections
Vital
loads/subsections
(a)
Vital
loads/subsections
Main generators
Bus
connection
Zonal
loads
Zonal
loads
Zonal
loads
Main generators
Vital
loads/subsections
Vital
loads/subsections
(b)
Fig. 30: (a) Ring battle mode, (b) Split plant battle mode.
56
Bus
connection
Software: MAS layer
Agent
Agent
Agent
Agent
Agent
Every device has its
own agent
Hardware: Power
system devices layer
Device
Device
Device
Device
Device
Fig. 31: Each device mapped onto an agent in supervisory control architecture [27].
Cartes and Srivastava et al have published research related to use of MAS for onboard modes of
the ship [25, 29, 30, 31]. The work outlines how a MAS can be deployed and configured to
handle a mesh-structured topology [27] as well as a ring topology [30, 31]. A general overview
of agents being used in large numbers for supervisory control, such that each device is
represented by its own agent is presented in [27, 32].
The idea of using agents which hold operational and functional information of the device it is
mapped onto, for supervisory control decisions, could be exploited further for the purpose of a
real-time F-FMEA. This process taps into a wealth of known information of fault/failure causeeffect relationships and in a real-time manner computes dependencies based on given network
topology. Such real-time updating could potentially also aid the diagnosis system by providing it
up-to-date information on the network’s connections. Also, the F-FMEA information could come
in handy for explaining briefly the cause and effect of a diagnosed fault/failure.
Agent based technology could be harnessed for a real-time F-FMEA during different modes of
the ship. Research listed in the previous paragraphs mention advantages of a MAS being mainly;
x
Extensibility to add modules
x
Flexibility to cater to different scenarios
57
x
Effective use of information
These three benefits could be utilized in an information rich supervisory control environment
tapping data from FMEA documentation. The mode of a ship is a standard baseline for
connected devices and subsections, but during a mission, it is possible some of these constituents
become unavailable because of various reasons. In such an event, a real-time F-FMEA would
need to be performed that comprises the existing connected parts. This incorporates the
extensibility and flexibility features of using agents. Further research into language processing
capabilities (highlighted in 4.6 onwards) may potentially solidify the choice of using MAS.
4.6 Natural language processing (NLP)
NLP forms an important wing of AI with rigorous research prevalent in mainly user-interaction
based systems employing text and/or speech and their combinations. Statistical data mining and
analysis methods are used to process large volumes of linguistic data in documents to find
meaningful means of reproducing the information for human use. A popular technique to preprocess large volumes of data from a corpus to derive meaningful links between words is latent
semantic analysis (LSA).
LSA could be employed on F-FMEA data as a corpus, to build a system capable of answering
fundamental queries. This can potentially form the primary part of a decision support system
which aids onboard crew to react during various situations. The F-FMEA documentation
contains precise information on failure/fault cause and effects, which makes it a resourceful
corpus to run LSA on.
The same LSA process that is used to streamline F-FMEA data is used on queries put forth by
crew. This ensures consistency between the word-processing analyses such that a basic level of
accuracy is always achieved. Modifications to conventional LSA could then be experimented
with to seek improvements to accuracy of the decision support capability.
This report demonstrates use of conventional LSA on the F-FMEA generated corpus. The results
of streamlining the data are shown. A sample of general queries that could be asked by onboard
crew is then tested to check if the system answers them correctly. This NLP based system for
decision support arising out of information derived from F-FMEA is a novel approach in the
field of SPS fault management. Combined with automated control decisions based on measured
quantities of the power network, the overall intelligent system could potentially have a very high
accuracy not only to diagnose faults but also to provide credible decision support.
4.6.1 LSA applied to extract useful information from the F-FMEA database
The first step to conduct LSA is to use a set of stop words (SW). These are words that are used
commonly in communication and as such can be omitted from the corpus. An example of SW
are words like “the, a, is, it, and, or” etc. There exists a list of common SW used extensively for
LSA involving news reports on the web. At the moment, as a primary threshold, this common
list of SW is used to perform the first step of LSA on the given data. However, it may be useful
and could enhance accuracy if a dedicated list of SW is created for specific applications such as
SPS in this case.
58
The initial experiments are carried out to test whether basic and general questions about the
system’s state are answered. These questions are in text form and using NLP processes, the
appropriate answer is found from the F-FMEA documentation. Rather than just answering a
simplistic “Yes” or “No”, using F-FMEA documentation, a more detailed explanation or
description to a question could be obtained. The question is processed as per LSA guidelines and
the remaining keywords form the basis for searching the F-FMEA corpus.
Owing to the highly specific nature of the corpus in this case, it may be needed to adopt a
slightly different or novel approach using LSA to process data in order to derive the required
information. This possibility needs to be explored in detail to ensure accuracy of decision
support. A proposed approach to begin experimental studies on an F-FMEA informed automated
decision support system is shown here. The approach is illustrated in Fig. 32.
Information
Fault indication and
diagnosis using local
measurements and AI
User query
NLP based processes
FMEA (functional and/or
hardware) database
Decision support system
using data from diagnostics
engine and FMEA database
Fig. 32: Proposed approach using FMEA information with NLP and AI-based diagnostics for decision risk
mitigation and decision support.
Fig. 32 shows a proposed approach utilizing information from the onboard intelligent diagnostics
and supervisory control system. Combined with information from FMEA databases, one can
enhance the diagnostic capability by adding descriptive elements to it by utilizing NLP. This
may not necessarily improve diagnostic accuracy, but is anticipated to aid in decision support for
the onboard crew by enabling word-queries to be answered satisfactorily.
The following process steps describe the working of an NLP based decision support system
deriving information from both the diagnostic engine and FMEA corpus.
59
3. Initial fault diagnosis – This action is performed by the onboard diagnosis system most
probably utilizing a sophisticated AI based methodology. Information from this system is
combined with the already existing FMEA database.
4. NLP process runs – The combined information from the previous steps is passed through
the selected NLP processes which analyze the input, to produce a relevant output.
5. Response to user query – In case onboard crew requires information about the current
state of operation, the system harnesses the NLP engine’s output to respond to this query.
A sophisticated NLP capability is able to analyze the question asked and produce a
relevant and correct answer.
6. Decision support – The ability of the onboard supervisory system to utilize data from the
diagnosis system, combine it with FMEA database and be processed by the NLP system,
bode well for providing enhanced decision support to onboard personnel.
5 CONCLUSIONS
The results demonstrate that the reliability of a distribution system is fundamentally linked to the
high-level topology of that system, as well as the relative positions of loads and generators
within that system. When redundancy is preserved, the breaker-and-a-half topology was shown
to be the most reliable of the commonly used distribution system designs.
Compared to the ring bus topology, the breaker-and-a-half topologies analyzed here offer
improvements in some aspects of equipment reliability, along with disadvantages in other areas.
For version one of the breaker-and-a-half topology, improvement is seen in the reliability of the
pulsed load and energy storage systems, but at a cost of reduced reliability in radar and zonal
load centers. For the second version of the breaker-and-a-half topology, reliability in all systems
is improved or held constant, but at a cost of requiring additional circuit breakers, which may be
infeasible due to cost or space concerns. Finally, the addition of bus tie circuit breakers makes
slight improvements to the reliability of both versions of the breaker-and-a-half topology, but at
the cost of two additional circuit breakers.
As each topology has different strengths and weaknesses in terms of reliability and cost/space,
this comparison should allow designers of shipboard electrical distribution systems to tailor a
ship’s distribution topology to the equipment requirements of the ship’s future missions, as well
as its physical dimensions and budget constraints.
Further, the results of the equipment placement algorithm show that there are further gains in
reliability that can be achieved through optimized placement of equipment loads within a
distribution system topology. These gains are relatively small compared to those that are
achieved through choice of overall topology, but changes in equipment placement are easier and
less costly design choices to implement than changes in system topology. Placement choices
also do not affect the number of required distribution system components, as can be the case with
choices of system topology.
60
For any distribution system, reliability can be optimized through a combination of overall
topology choice and the placement of loads and generators within the topology. The
demonstration of the breaker-and-a-half topology’s superior reliability and the equipment
placement algorithm developed here is are important tools in the design of any distribution
system in which reliability is a key concern.
The idea of linking the system design policy with the reliability consideration is put forward in
various levels. Both the static and dynamic analyses are considered for long term and short term
controls planning supporting the survivability. Fuel saving for long term system design is
proposed through static analysis. QoS metrics are being studied to be added to the research work.
On parallel, a general reconfiguration scheme with consideration of both steady-state global
performance and dynamic transient responses during the system reconfiguration is developed
and combined with Model-Predictive control to achieve an overall optimized operation. A
preliminary simplified ESRDC related SPS model has been built for the sole purpose of transient
study. Initial efforts have been put into evaluating the system responses under contingencies on
different components and attempts have been made to implement preliminary model predictive
control techniques to generate of a restoration scheme based on chosen optimization objective.
Further analysis of the research issue will be continued on successive quarters.
Another aspect of this research work is the importance and rationale behind starting off with FFMEA for understanding risks within the notional MVDC SPS. A generalized F-FMEA was
demonstrated through the standard tabular format taking into account functional attributes at the
subs-system and system levels. This activity helps understand fundamental risks associated with
the network from a basic functionality point of view without dwelling on the specifics of
hardware related data. From this stage on, the next research could be in the more detailed HFMEA which derives severity and criticality information from the aforementioned F-FMEA.
This however is a matter for further research especially when there is more clarity on the type
and nature of hardware devices used. It would also make it imperative in future research to set up
hardware test beds for conducting failure analysis to enhance understanding gained from RTDS
models. By doing this, research done through simulations can be backed up using actual
experimental data.
It was discussed that a useful application would be the capability to perform an automated FFMEA that feeds off the prevalent network topology data per mission. This ability could supply
information to the diagnostic engine to enhance its accuracy. Further, an automated F-FMEA
could aid in the decision making process as it contains information of well known causes and
effects of faults/failures.
A detailed FMEA database with an NLP capability onboard the SPS could provide the following
advantages to the overall fault accommodating control system.
7. Enhance fault diagnosis – A fault accommodating system is anticipated to have a robust
diagnostic engine, capable of providing accurate fault detection and identification. This
action has sufficient information regarding the particular subsection/device/component
that has caused the disturbance. A feasible remedial action is expected to be taken by the
automated control system. In addition, a detailed description could be provided tapping
the FMEA database to enhance understanding of the failure for the onboard crew.
61
8. Improve decision support – Specific information regarding failure modes, causes and
effects contained in the FMEA database(s) could be accessed to enhance the process of
decision making and taking remedial action when necessary manually. Combined with
the ability to perform an automated real-time F-FMEA, the system remains up to date
with network topologies and dependencies, thereby having the potential to improve upon
aiding onboard crew to assess the scenario in event of a disturbance.
Current research focuses on developing techniques for performing an automated F-FMEA given
a topology information and general F-FMEA information. Further research is centered on
developing NLP based algorithms to automate information retrieval at a user-query from the FFMEA database shown. This part of the work is guided by the methodology shown in Fig. 32,
wherein the NLP components will be used to combine effectively with diagnostic components of
the system.
6 REFERENCES
[1] N.H. Doerry and D. Clayton, “Shipboard Electrical Power Quality of Service,” IEEE ESTS, July 2005.
[2] N.H. Doerry, “Designing Electrical Power Systems for Survivability and Quality of Service,” ASNE Naval
Engineers Journal, Vol. 119, No. 2, 2007, pp. 25-34.
[3] N.H. Doerry and J.V. Amy Jr., “Implementing Quality of Service in Shipboard Power System Design,” IEEE
ESTS, 2011.
[4] M.S. Grover, R. Billinton, “A Computerized Approach to Substation and Switching Station Reliability
Evaluation”, University of Saskatchewan, November 1973.
[5] R.N. Allan, M.F. de Oliveira, “Reliability Modeling and Evaluation of Transmission and Distribution Systems”,
PROC. IEEE, Vol. 124, No. 6, June 1977.
[6] Z. Liu, “Reliability Analysis of Breaker Arrangements in High Voltage Stations: A Fault Tree Analysis,”
Chalmers University of Technology, 2008.
[7] R. E. Brown, G. Frimpong, and H. L. Willis, “Failure Rate Modeling Using Equipment Inspection Data,” IEEE
Trans. Power Systems, vol. 19, no. 2, May 2004.
[8] Y. Wu and W. Chang, “A Study on Optimal Reliability Indices in an Electrical Distribution System,” Proc.
IEEE Power System Technology, vol. 2, 2000, pp. 727-732.
[9] R. E. Brown, J. R. Ochoa, “Distribution System Reliability: Default Data and Model Validation,” IEEE Trans.
Power Systems, vol. 13, no. 2, May 1998.
[10] P. M. Anderson, Power System Protection, IEEE Press, 1999, pp. 1052-1248.
[11] J. Kennedy and R. Eberhart, “Particle Swarm Optimization,” Proc. IEEE International Conference on Neural
Networks, vol. 4, pp. 1942–1948, Piscataway, NJ, 1995.
[12] K. Prakash and M. Sydulu, “Particle Swarm Optimization Based Capacitor Placement on Radial Distribution
Systems,” Proc. IEEE PES GM, 2007.
[13] O. Amanifar, “Optimal distributed generation placement and sizing for loss and the reduction and voltage
profile improvement in distribution systems using particle swarm optimization and sensitivity analysis,” Proc.
IEEE EPDC, 2011.
[14] N. Doerry, “Energy and the affordable future fleet”, 10th International Naval Engineering Conference and
Exhibition (INEC 2010), Portsmouth, UK, May 11-13, 2010
[15] Gas Turbine World Handbook, 2003, Pequot Publishing, Southport, CT
[16] R. Amgai, J. Shi, S. Abdelwahed,”Lookahed Control Based Framework for Power System Applications”,
NAPS, 2012, 1-6
[17] IEEE Recommended Practice for 1 kV to 35 kV Medium-Voltage DC Power Systems on Ships, IEEE Std
1709, 2010
[18]Andrus, M.; Steurer, M.; Edrington, C.; Bogdan, F.; Ginn, H.; Dougal, R.; Santi, E.; Monti, A., "Real-time
simulation-based design of a power-hardware-in-the-loop setup to support studies of shipboard MVDC
62
issues," Electric Ship Technologies Symposium, 2009. ESTS 2009. IEEE , vol., no., pp.142,151, 20-22 April
2009
[19]IEEE Standards Association, “IEEE recommended practice for 1 kV to 35 kV medium-voltage dc power
systems on ships”, IEEE Std 1709TM-2010
[20] Hiskens, I.A.; Pai, M. A., "Trajectory sensitivity analysis of hybrid systems," Circuits and Systems I:
Fundamental Theory and Applications, IEEE Transactions on , vol.47, no.2, pp.204,220, Feb 2000
[21] Hiskens, I.A.; Pai, M. A., "Power system applications of trajectory sensitivities," Power Engineering Society
Winter Meeting, 2002. IEEE , vol.2, no., pp.1200,1205 vol.2, 2002.
[22] Soman, R.R.; Davidson, E.M.; McArthur, S.D.J, "Using functional failure mode and effects analysis to design
the monitoring and diagnostics architecture for the zonal MVDC shipboard power system," Electric Ship
Technologies Symposium, 2009. ESTS 2009. IEEE, pp.123-128, 20-22 April 2009.
[23] Soman, R.R.; Davidson, E.M.; McArthur, S.D.J.; Fletcher, J.E.; Ericsen, T., "Model-based methodology using
modified sneak circuit analysis for power electronic converter fault diagnosis," Power Electronics, IET, vol.5,
no.6, pp.813-826, July 2012.
[24] Savakoor, D.S.; Bowles, J.B.; Bonnell, R.D., “Combining sneak circuit analysis and failure modes and effects
analysis,” Reliability and Maintainability Symposium, 1993. Proceedings., Annual , vol., no., pp.199-205, 26-28
Jan 1993.
[25] Solanki, J.M.; Schulz, N.N.; , “Using intelligent multi-agent systems for shipboard power systems
reconfiguration,” Intelligent Systems Application to Power Systems, 2005. Proceedings of the 13th International
Conference on , vol., no., pp.3 pp., 6-10 Nov. 2005.
[26] Feliachi, A.; Schoder, K.; Ganesh, S.; Hong-Jian Lai; , “Distributed control agents approach to energy
management in electric shipboard power system,” Power Engineering Society General Meeting, 2006. IEEE ,
vol., no., pp.6 pp., 0-0 0.
[27] Kai Huang; Srivastava, S.K.; Cartes, D.A.; Sloderbeck, M.; , “Intelligent Agents Applied to Reconfiguration of
Mesh Structured Power Systems,” Intelligent Systems Applications to Power Systems, 2007. ISAP 2007.
International Conference on , vol., no., pp.1-7, 5-8 Nov. 2007.
[28] Momoh, J.A.; , “Navy ship power system restoration using multi-agent approach,” Power Engineering Society
General Meeting, 2006. IEEE , vol., no., pp.5 pp., 0-0 0.
[29] Zhiping Ding; Srivastava, S.K.; Cartes, D.A.; Suryanarayanan, S.; , “Dynamic Simulation-Based Analysis of a
New Load Shedding Scheme for a Notional Destroyer-Class Shipboard Power System,” Industry Applications,
IEEE Transactions on , vol.45, no.3, pp.1166-1174, May-june 2009.
[30] Kai Huang; Cartes, D.A.; Srivastava, S.K.; , "A multiagent based algorithm for ring-structured shipboard power
system reconfiguration," Systems, Man and Cybernetics, 2005 IEEE International Conference on , vol.1, no.,
pp. 530- 535 Vol. 1, 10-12 Oct. 2005.
[31] Huang, K.; Cartes, D.A.; Srivastava, S.K.; , "A Multiagent-Based Algorithm for Ring-Structured Shipboard
Power System Reconfiguration," Systems, Man, and Cybernetics, Part C: Applications and Reviews, IEEE
Transactions on , vol.37, no.5, pp.1016-1021, Sept. 2007.
[32] Li Liu; Logan, K.P.; Cartes, D.A.; Srivastava, S.K.; , "Fault Detection, Diagnostics, and Prognostics: Software
Agent Solutions," Vehicular Technology, IEEE Transactions on , vol.56, no.4, pp.1613-1622, July 2007.
63