Reliability of Emergency and Standby Diesel Generators: Impact on Energy Resiliency Solutions

Jeffrey Marqusee and Don Jenket

National Renewable Energy Laboratory, Golden Colorado United States

Abstract
Emergency diesel generators are the most common form of backup power for critical loads when the
grid fails and are most often deployed as stand-alone generators (<2000 kW) tied to individual buildings
for hospitals, emergency services, military bases, ports, airports, industries, and commercial facilities.
Understanding the finite reliability of emergency diesel generators during continuous operation is
crucial for energy planners, managers, and end-users. A new analysis of two large non-public emergency
diesel generator operational data sets shows that commonly used reliability metrics are inadequate to
predict the performance during a grid outage. Additional metrics addressing the likelihood of an
emergency diesel generator to be available at the time an outage starts, successfully start and carry
load, and constantly run for the duration of the outage are needed to predict the continuous run-time
reliability of emergency diesel generators during a grid outage. This is the first analysis of the reliability
of modern generators that follow standard maintenance protocols commonly used for backup power
and these new metrics are shown to be dependent on the level of emergency diesel generator
maintenance with values that can vary by more than an order of magnitude. Even using the new metrics
for well-maintained emergency diesel generators, the analysis shows that the single emergency diesel
generators configurations are only 80% likely to provide power for the duration of a two-week grid
outage.

Keywords
Emergency Diesel Generators, Standby Diesel Generators, Reliability, Energy Resilience, Generator
Reliability

1. Introduction
Energy resilience is a critical issue facing the nation.1 The President’s National Infrastructure Advisory
Council (NIAC) recently issued a report (1) calling for a recognition of this profound risk and a new
national focus. Continuous and reliable electric power is essential to modern society. We depend on it
for health, safety, economic vitality, and national security. The risks of blackouts and loss of electric
power are not new concepts. Outages of just a few hours are well known, but longer duration outages
are becoming more frequent (2). Today, the commercial electric grid faces increasing risks from three
threats: (1) natural disasters and extreme weather; (2) physical attacks; and (3) cyberattacks. Each of
these threats will likely increase in frequency in the future and utilities are already seeing a statistically
significant increase in major event days (2).
Hospitals, emergency services, military bases, ports, airports, industries, commercial facilities, and
others rely on backup power systems to provide electricity for their critical loads during grid outages. In
2009, 170 GW of backup power was installed in the United States (3), and this value has likely increased
in the past decade. A subset of the electrical power backup is captured through the Department of
Homeland Security’s (DHS’s) Enhanced Critical Infrastructure Program (ECIP) Initiative (4). The Initiative

1 Resilience is defined as the ability to prepare for and adapt to changing conditions and withstand
and recover rapidly from disruptions. In the context of energy disruptions, it applies to our ability to respond to
long duration outages.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 reports that 85% of backup power for the critical infrastructure is provided by emergency or standby
diesel generators.2 Going forward, we will refer to these generators collectively as emergency diesel
generators (EDGs).
Emergency power systems based on diesel generator sets are installed at facilities where a loss of utility
power would result in an unacceptable impact to operational capability or present a risk to life, safety,
or property. These generator sets are typically configured to start automatically upon electrical utility
failure and assume essential facility loads until utility power is restored. They are commonly deployed as
stand-alone generators tied to a single critical-load building. For outages of a few hours at most, these
generators have proved effective and easy to deploy. But for longer outages, their reliability and
configuration become critical issues. The failure of an EDG during a grid outage can cause severe
economic consequences to a company, increased community risk due to loss of emergency services,
fatalities in medical facilities, and can endanger national security if it occurs at a military installation (5).
Increasing redundancy (N+1, N+2, and so on), or networking EDGs in a microgrid, can improve the
system’s reliability, but the need for redundancy or networking, and its costs and benefits, depends on
the reliability of the individual generators, the length of the outage, and the value of the lost load.
Standards, regulations, and guidance provide direction on the use, design, and maintenance of EDGs for
providing backup power for critical loads (6) (7) (8) (9) (10). These documents require or recommend the
deployment of EDGs for critical facilities. Some hospital functions require backup power for four days
(11), while military base requirements vary from one to two weeks of backup power. The reliability of
any backup power system based on EDGs is dependent on the reliability of each powered EDG in the
configuration; however, the standards and guidance documents do not provide information (or
appropriate reference) about typical reliability metrics for modern, well-maintained EDGs. We show, for
the first time, that the reliability metrics commonly used today by both government and commercial
sectors are both inadequate and inappropriate for assessing the performance of EDGs for providing
backup power during a grid outage.
Despite how common EDGs are as a source of backup power, the literature on their reliability is very
limited and primarily focused on generators that provide backup power to nuclear power plants (12)
(13) (14) which are typically much larger than EDGs that are used by hospitals, emergency services,
military bases, ports, airports, industries, and commercial facilities. In this paper we compare our results
with the most recent data on EDGs at U.S nuclear power plants. There is published research on older
EDGs (15) (16) in service during the 1970s, 1980s and early 1990s but emissions and environmental
restrictions in the 1990s led to significant design changes in the diesel engines that make these results
no longer relevant for EDGs fielded today. Attempts have been made to calculate reliability from first
principles (17) but the complexity of EDGs and the large number of failures mechanisms make such
attempts uncertain. One might assume that with the large literature on backup power systems using
microgrids that the reliability of the most common distributed energy resource, EDGs, would have been
carefully examined. But a recent review (18) of microgrid design and optimization approaches shows
that the published work ignores the reliability of the EDGs or treat it as a parameter that is unknown.
Even recent optimization work (19) that is designed to assess power avaibility makes assumptions on
the reliability of EDGs that are nor supported by data and lead to different results than we present here.
Thus, in the absence of data driven analysis on EDG reliability, energy resilience studies most often
assume 100% reliability.

2The terms “emergency” and “standby” have different meanings in various building codes but these backup power
systems use the same commercial power generation systems. Both types of backup power systems are governed
by the National Fire Protection Association 110 Standard (6).

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 This paper, based on a new analysis of two large non-public data sets, provides the first known reporting
of the reliability of modern commercial EDGs that follow standard maintenance protocols commonly
used for backup power at typical critical facilities. This paper also addresses the impact of the finite EDG
reliability on energy resilience for facilities during a multiday grid outage. In Section 2, we discuss
currently used reliability guidance, define the reliability metrics needed to assess performance, and
review four existing data sets relevant to estimating EDG reliability. In Section 3, we calculate the critical
reliability metrics based on two data sets that have not previously been fully analyzed and compare
them to the existing literature. In Section 4, we illustrate the impact of these metrics on a few common
energy backup configurations. We conclude with a brief discussion of the importance of considering
finite reliability when assessing energy backup power systems and discuss future directions.

2. Emergency Diesel Generator Reliability Data

Reliability is defined as “a characteristic of an item, expressed by the probability that the item
will perform its required function under given conditions for a stated time interval” (12). In this paper,
we assume that the EDG has passed acceptance testing, was properly engineered and manufactured,
and is not near the end of its life. In terms of the reliability literature’s3 “Bathtub Model” (Figure 1),

Figure 1. A reliability bathtub model showing a low constant failure

rate during the useful life period.

the EDG is in its useful life period and has a constant failure rate. Most reliability assessments are
concerned with systems or components intended to operate continuously. As discussed below, EDGs
run very infrequently and sit in a cold state for most of their lifetimes. Because of this, it is important to
precisely define the required function and time intervals we are considering when specifying reliability
metrics. The reliability metrics discussed below are the standard terms used in the industry. What
differentiates our definitions from common usage is the explicit recognition of the time intervals which
should be considered when treating EDGs.

We assume that the EDG is operating within the useful life period and that failure rates are constant. An
assumption of constant failure rates has been shown to be accurate for many systems4 and is the
method recommended by the IEEE Standard 3006.8 for analyzing reliability of commercial power
systems. If one assumes a constant failure rate, the cumulative failure rate is exponentially distributed.

3 See reference 12 and sources cited.

4 See reference 12 and sources cited

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 Reliability is the probability that the component will perform its function for time t and is designated
R(t). Equivalently it is the probability that the component will have no failures between the time at
which it is required to operate (t=0) and time t. The failure probability is the cumulative distribution
function for failures from t=0 to time t and is given by:

𝐹(𝑡) = 1 − 𝑅(𝑡)

If we take the first derivative of a cumulative failure distribution function, we obtain the failure
probability density function:

𝑓(𝑡) = 𝑑𝐹(𝑡)/𝑑𝑡

Typically, the mean time between failure (MTBF) is used to define the reliability of repairable
components:

𝑙𝑖𝑓𝑒𝑡𝑖𝑚𝑒
𝑀𝑇𝐵𝐹 =
𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑓𝑎𝑖𝑙𝑢𝑟𝑒𝑠

The MTBF is simply the inverse of the failure rate per year or hour (λa):

1
𝑀𝑇𝐵𝐹 =
λ𝑎
This metric is dependent on the number of attempts to maintain, repair, or utilize the EDG which creates
a dependency on the maintenance schedule and number of grid outages. Since EDGs sit idle most of
their lifetimes, the MTBF cannot be used to estimate the probability of an EDG’s runtime failure
however this is the commonly reported reliability metric.

A metric used to describe non-reparable components is the mean time to failure (MTTF):
∞ ∞
𝑀𝑇𝑇𝐹 = ∫ 𝑡 × 𝑓(𝑡) 𝑑𝑡 = ∫ 𝑅(𝑡) 𝑑𝑡
0 0

Although EDGs are repairable, we will use MTTF to define the failure rate as a function of run time.
Upon repair, an EDG may again be started, but that constitutes a new run time interval. Empirically the
MTTF can be calculated as:

𝑡𝑜𝑡𝑎𝑙 𝑟𝑢𝑛𝑡𝑖𝑚𝑒
𝑀𝑇𝑇𝐹 =
𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑓𝑎𝑖𝑙𝑢𝑟𝑒𝑠 𝑤ℎ𝑖𝑙𝑒 𝑟𝑢𝑛𝑛𝑖𝑛𝑔

This reliability metric is independent of the grid’s reliability or the EDG’s testing frequency and is a
function of the EDG run time. The impact of more frequent grid failures requiring more frequent EDG

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 demands are factored out. We separately consider the probability that the EDG fails to start and carry
the load5:

𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑓𝑎𝑖𝑙𝑢𝑟𝑒𝑠 𝑡𝑜 𝑠𝑡𝑎𝑟𝑡

𝐹𝑎𝑖𝑙𝑢𝑟𝑒 𝑡𝑜 𝑆𝑡𝑎𝑟𝑡 𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 (𝐹𝑇𝑆) =
𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑎𝑡𝑡𝑒𝑚𝑝𝑡𝑠 𝑡𝑜 𝑠𝑡𝑎𝑟𝑡

These two metrics, MTTF and FTS, define the inherent reliability of an EDG. Since EDGs are repairable,
we differentiate these from the reported MTBF which does not capture whether the failure occurred
upon start or while running and is dependent on the frequency of demands on the EDG due to testing
and grid outages.

One additional metric we require is the operational availability (OA) of the EDG. This is defined as the
probability that the EDG is in service (or available to attempt to provide power) at the start of a grid
outage and accounts for the EDG being down due to repairs or maintenance:

𝐿𝑖𝑓𝑒𝑡𝑖𝑚𝑒 − 𝑇𝑖𝑚𝑒 𝑜𝑓𝑓𝑙𝑖𝑛𝑒 𝑑𝑢𝑒 𝑡𝑜 𝑟𝑒𝑝𝑎𝑖𝑟𝑠 𝑎𝑛𝑑 𝑚𝑎𝑖𝑛𝑡𝑒𝑛𝑎𝑛𝑐𝑒

𝑂𝐴 =
𝐿𝑖𝑓𝑒𝑡𝑖𝑚𝑒

An EDG could be offline or unavailable due to ongoing repairs initiated due to a failure or due to
scheduled maintenance. These out-of-service times are characterized by the mean time to repair
(MTTR), which is the mean time associated with unscheduled repairs due to failures, and the mean time
to maintain (MTTM), which is the mean time associated with scheduled maintenance activities that
require the system to be taken offline. OA is sensitive to maintenance and repair times, as well as the
failure rates and maintenance schedules.

The standard source for reliability data for equipment used in industrial and commercial power systems
is the Institute of Electrical and Electronics Engineers (IEEE’s) Gold Book (13), recently updated in IEEE’s
3006.8 Recommended Practice for Analyzing Reliability Data for Equipment Used in Industrial and
Commercial Power Systems (14). IEEE provides summary data on key reliability metrics for hundreds of
components. The summary data is based on data from two major collection efforts conducted by the
U.S. Army Corps of Engineers Power Reliability Enhancement Program (PREP). As the IEEE standard
states (14), the PREP’s “efforts created the most comprehensive facility equipment reliability database
in existence.” IEEE’s earlier reports (13) are identical to the reliability data reported in the U.S. Army
standard (15), which also is derived from the PREP database. The Army has not updated their published
guidance since 2006 and now relies on the IEEE documentation of its data collection. Both government
and commercial assessments of reliability are dependent on the recent IEEE reported results.

Older EDG reliability data (16) is of historical interest but not representative of currently deployed EDGs.
Increased scrutiny of air emissions from diesel generators in the early 1990s led to changes in diesel
generator design. Although the focus was not on emergency generators, the 1990 Clean Air Act
impacted the manufacturing of all diesel generators. Therefore, this analysis does not consider reliability
data from generators manufactured before the 1990s to be representative of today’s generator
performance.

5FTS is not usually considered for generators that provide prime power and are on almost all the time. But for an
EDG often run for short time periods, the FTS can be an important reliability driver.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 IEEE summary data provides results for standby diesel generators <250 kW and standby diesel
generators from 250 kW-1,500 kW. These sizes are typical of EDGs used at hospitals, military bases, and
commercial facilities. IEEE and PREP report the following information (17):
• Failure rate per year or hour: λa
• Mean time between failure: MTBF =1/ λa
• Mean time to repair (excluding logistics): MTTRe
• Mean time to maintain: MTTM.

For EDGs, this analysis considers the IEEE- and PREP-reported reliability data to be inadequate and
inappropriate for assessing the performance of EDGs for providing backup power during a grid outage
for the following three reasons.

First, EDGs only run during a grid outage or for testing, which accounts for a very small fraction of the
year. The inherent reliability of an EDG is dependent on the unit’s design, manufacturing, age, and
maintenance, which are independent of the testing schedule and the reliability of the local grid. Using
MTBF, which is based on a failure rate per year, yields the results reported by IEEE of greater than
35,000 hours for 250-kW to 1,500-kW standby generators. This is more than twice the MTBF reported
for continuous-use generators of the same size (14). Annual failure rates are not relevant for assessing
the run time failure rate of an EDG and MTTF as defined previously (as a function of run time) is
required. IEEE and PREP’s MTBF provides a reliability metric, dependent on the EDG’s testing schedule
and the reliability of the grid (more grid outages lead to greater run times per year of an EDG) and is
inappropriate for predicting run time failures during a grid outrage.

Second, emergency generators are turned on and off much more frequently than prime generators.
EDGs are not kept on hot standby and the FTS is an important characteristic usually not considered in
assessing continuous power systems. An EDG is run only for required testing and during a grid outage.
The probability of an emergency generator failing to start is a well-recognized failure event, but IEEE and
the underlying PREP data do not provide this key reliability statistic.

Third, the MTTRe reported by IEEE does not include the logistics associated with a repair (17). It reports
only the time required to make the repair once the needed parts and labor are on-site (i.e. the MTTRe).
The time required to obtain parts and have the appropriate technicians on-site is significant and can be
larger than the time needed to make the repair, as we will show in section 3. In terms of assessing the
OA of a standby generator, it is the full MTTR that includes the logistics time that is critical.

To quantitatively characterize an EDG’s6 reliability during a grid outage, we need to know four metrics:

• OA: The probability an EDG is operationally available (i.e., not out of service due to repair or
maintenance activities) to attempt to generate power when a grid outage occurs;
• FTS Probability: The probability the EDG fails to start and carry load at the beginning of an
outage;
• MTTF: The run time failure rate during an outage. MTTF or run time failure rate requires data on
the EDG run times to calculate; and

6 An EDG is considered a system and includes all components required to output electrical power, including the
transfer switch.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 • MTTR: The likelihood that the failed EDG can be repaired during the outage including the
logistics time to receive parts and/or have skilled labor on-site.

The limited amount of data available for modern EDGs7 creates a challenge for providing information on
any of these key criteria. Given the complexity of an EDG and the large variety of failure mechanisms,8
empirical data is needed to assess their reliability. To our knowledge, only four data sets9 are both large
enough and relevant enough to EDGs to be considered. While none of these data sets provide
information for all four metrics, they can be used together to provide insight on the performance of an
EDG during a long-term grid outage. We briefly review these four data sets (17) (18) (19) (20) below and
summarize their characteristics. We quantitively compare the reliability metrics that can be calculated
based on these four data sets in Section 3.

The data collected by the PREP that forms the basis for all reported IEEE reliability results was collected
from over 200 sites in the United States and Canada. The sites include military facilities, hospitals, and
universities. PREP collects data by surveys from facilities and follows up with site visits when possible.
The PREP data for EDGs is divided into two size classes: <250 kW and 250 kW-1,500 kW. The PREP data
does not include information on the number of attempted starts or run time of the EDGs. Thus,
estimates for FTS and MTTF based on run time cannot be constructed. PREP data includes the number of
failures as a function of the observation time or, equivalently, a failure rate per year or MTBF. PREP data
also includes detailed data on the time required for maintenance activities and the time to repair in case
of failures which can be used to estimate availability. While this data set does not provide metrics for
FTS or MTTF, it will be used later in this study to estimate maintenance and repair time to calculate OA.

Maintenance frequency and practices affect an EDG’s availability and reliability. PREP rates each site
according to the quality of maintenance employed and categorizes the sites into three tiers:

• Above average maintenance is reserved for facilities that followed a scheduled preventative
maintenance policy equivalent to the manufacturer’s suggested policy; meets National Fire
Protection Association (6) or DoD’s Unified Facility Criteria (10) recommended maintenance
practices; uses specialized equipment tests (thermograph, vibration analysis, oil analysis); and
has complete spare parts kits for the equipment. 25% of the PREP sites employ above average
maintenance.
• Average maintenance also rigorously follows recommended maintenance schedules but does
not use specialized equipment or have complete spare parts kit on hand. 57% of PREP sites
employ average maintenance practices.
• Below average maintenance either has no formal maintenance policy and schedule or fails to
follow one. 17% of PREP sites employ below average maintenance.

7 By modern, we mean typical of EDGs deployed today.

8 Starting failures include battery failures, motor starting failures, starting solenoid/relay failures, fuel pump
failures, fuel solenoid and starting control failures, and others. Running failures include voltage regulator failures,
speed governor failures, cooling system failures, fuel supply system failures, and others.
9 Given the low probability of failure, a data set must be large enough to yield a result in which the confidence

intervals for the key metrics are not meaningless.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 For the purpose of this study, we partition maintenance practices into two classes: well-maintained
EDGs, which include both average and above average maintenance practices; and poorly maintained
EDGs, which are equivalent to below average maintenance.

A study (18) conducted in Hong Kong reported data on 147 EDGs monitored for an average of five years.
The data was collected via a generator reliability survey followed up by site visits when feasible. The
EDGs were used in commercial, residential, industrial, and institutional settings to provide backup power
during a grid failure. They ranged in size from 80 kW to 1,500 kW, which is typical of EDGs for the
purposes of this study. They were reported to have poor maintenance practices, which resulted in a high
reported FTS of 1.65% and an MTTF based on a run time of only 61 hours. This data set provides a
benchmark for EDGs in the below average or poorly maintained category.

In the United States, the Nuclear Regulatory Commission (NRC) requires that the performance data on
EDGS that support nuclear power plants be reported routinely. Like all EDGs, those at nuclear power
plants do not operate all the time. They are required to operate when the grid power is down and
during shutdown periods. The demands and run hours are reported on a quarterly or semi-annual basis,
and existing regulations established the requirements for testing of these on-site power sources.
Therefore, an extensive database on these EDGs exists (21). Recent analysis of this database (19) has
calculated the EDGs’ reliability metrics. All demand types on the EDGS are considered, including both
testing, as well as operational. These EDGs range in size from 50 kW to 499,999 kW, and most are
considerably larger than those used in hospitals, emergency facilities, military bases, and commercial
facilities. The sizes of the EDGs in this database are summarized in Figure 2.

180
160
140
Number of EDGs

120
100
80
60
40
20
0
50 kW - 249 250 kW - 5,000 kW - 100,000 kW -
kW 4,999 kW 99,999 kW 499,999 kW
Size of EDGs

Figure 2. EDG size distribution in NRC database

Since this data set represents all EDGs used at U.S. nuclear power plants, it provides insight into an
industry that requires high consistency, and the EDGs are assumed to be well-maintained. They reported
an FTS of 0.66% and a MTTF base on run times of 636 hours, considerably better than the results for the
smaller EDGs in the Hong Kong study. Due to the EDG size distribution of this data set, direct comparison
of their reliability performance for our applications cannot be achieved.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 The final data set we consider was collected in support of a Ph.D. thesis (20) supported by the U.S. Navy.
The research was intended to provide facility managers with the qualitative data to optimize the staffing
level and generator maintenance. The scope of the study was limited to modern, high-efficiency, low-
emission generator sets. The mean age of EDGs in this database is 11.2 years. Maintenance logs that
followed current government regulations were collected and entered into a structured database. The
sample population included EDGs between 10 kW and 2,000 kW with a 324-kW mean generator set
rating. Figure 3 shows the EDG size distribution for this database.

The database contains information on run times, as well as attempted starts and failures. Detailed
information on the maintenance practices were recorded but do not include data on downtime due to
maintenance time or repair time due to failures. The FTS and the MTTF are reported in section 3. This
data set will be used to provide a benchmark for EDGs that are well-maintained.

160

140
Number of EDG in Database

120

100

80

60

40

20

0
<250kW 250 to <500kW 500 to <1000kW 1000 to 2000 kW
EDG Size

Figure 3. EDG size distribution in the Fehr database

The NRC data set includes mostly EDGs larger than typically used in hospitals, emergency facilities,
military bases, and commercial facilities, while the other three data sets report on EDGs of a size
typically used in these facilities. We report the NRC results only to qualitatively compare them to EDGs
of concern at typical critical facilities. The other three databases represent two different levels of
maintenance: well-maintained and poorly maintained. The characteristics for these four data sets are
summarized in Table 1.

Although these data sets are the largest available, their size is limited. Failures are relatively rare events
and it is important to understand the uncertainty in estimates for the reliability metrics due to sample
size. In the next section we address this issue by providing 90% confidence intervals for the two key
reliability metrics, MTTF and FTS and the impact on a single EDG’s reliability is provided. In section 6 the
impact on system level reliability is discussed.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 Source (Ref) # EDGs EDG Years Available Metrics Comments
of
Observation
PREP (17) 304 2,298 • MTBF • Representative size EDG
• Repair time (<1,500 kW)
• Maintenance time • Mixed maintenance
Hong Kong 147 790 • MTBF • Representative size
(18) • MTTF EDGS (80 kW-1,500 kW)
• FTS • Poorly maintained
NRC (19) 232 1,790 • MTTF • Large EDGS (most >> 1
• MTBF MW)
• FTS • Well-maintained
Fehr (20) 239 1,281 • MTTF • Representative size
• MTBF EDGS (10 kW-2,000 kW)
• FTS • Well-maintained
Table 1. Summary of EDG data set characteristics

3. Emergency Diesel Generator Reliability

MTTF:

The MTTF based on run time is the most important EDG reliability parameter when looking at EDG
performance periods from days to weeks. Over long duration outages, failures while running dominate
the overall reliability.

We have analyzed the recently collected Fehr dataset (20) to determine the MTTF for well-maintained
EDGs. As shown in previous work (20), this data contains no statistically significant evidence that the
generator’s make, model, or size (10 kW-2,000 kW) has any significant impact on reliability. Using a
simple frequentist analysis (21), the MTTF and its 90% confidence intervals are provided in Table 2. This
information is compared to results from the NRC and Hong Kong data sets described above that provide
information on failures as a function of run time (the PREP data does not contain information on run
times).

Data Source MTTF Low Value MTTF Mean Value MTTF High Value
Fehr (20) 1,180 hours 1,662 hours 2,410 hours
NRC10 (19) 568 hours 636 hours 714 hours
Hong Kong (18) 53 hours 61 hours 71 hours
Table 2. MTTF data including mean and 90% confidence intervals for the three data sets

The 90% confidence intervals do not overlap. The Fehr and Hong Kong data sets involve similar size and
types of EDGs and are used to represent EDGs with significantly different levels of maintenance. The
well-maintained EDGs in the Fehr data set have MTTFs over 20 times longer than seen in the poorly

10The published study on EDGs at U.S nuclear power plants reports only failures to run for run times greater than
one hour. The underlying data is not available for analysis outside the nuclear industry. This constraint is unlikely
to affect the calculated MTTF significantly and would decrease the reported MTTF value.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 maintained Hong Kong data set. The NRC data set includes much larger EDGs. Whether their relative
MTTF (between the other two data sets) is due to the size of the generators or the maintenance
practices in the nuclear industry cannot be determined.

FTS Probability

As stated above, EDGs are not kept on hot standby and must start and transfer power to the load when
called upon during a grid outage. Failures to start is a rare phenomenon but significant enough to
warrant its inclusion in reliability assessments. The NRC and Hong Kong data sets report number of
attempted starts and FTS. The NRC divides its FTS into two classes: immediate FTS and FTS and carry
load.11 For the Fehr data set, 44 FTS were observed for the 239 EDGs monitored. The number of
attempted starts was only recorded only for 35 of the 239 EDGs in the data set. Three of these EDGs
were installed for less than two months and were still undergoing initial testing. For the 35 EDGs, the
average number of starts per year was 26.7, consistent with common practices. We can use this rate to
estimate the total number of attempted starts for all 239 EDGs. Applying this simple estimate yields
34,134 attempted starts over the observation period for the 239 EDGs resulting in an estimated FTS of
.13%.

Statistics on the FTS with 90% confidence intervals are shown in Table 3 for all three data sets (the PREP
data contains no information on attempted starts). Similar to the MTTF metrics, well-maintained EDGs
are much more reliable with the FTS probability an order of magnitude lower than the poorly
maintained EDGs. The larger EDGs used in the nuclear industry have FTS and carry load roughly midway
between the other two data sets. Again, the cause for this difference may be due to the size of the EDGs
or maintenance practices.

Data Source Low Value FTS Mean Value FTS High Value FTS
Fehr (20) 0.10% 0.13% 0.17%
NRC12 (19) 0.26% 0.66% 1.20%
Hong Kong (18) 1.44% 1.65% 1.88%
Table 3. Mean FTS Probabilities and the 90% Confidence Intervals

Operational Availability

The OA of an EDG is dependent on the annual failure rates, the time it takes to repair a failure, and the
time the EDG is out of service due to scheduled maintenance activities. The PREP database contains
information on the repair and maintenance times. The published (14) mean time to repair does not
include the logistics time (MTTRe) and is not relevant for calculating an OA; however, the underlying
database does include the needed information, and Figure 3 shows the distribution of repair times,
including logistics (MTTR) for all packaged EDGs in the PREP database. Due to PREP reporting from
earlier data collection efforts, PREP characterized a subset of observations by the subset’s mean, which
falls in the 16-24-hour interval. Thus, the data artificially appears as a bimodal distribution.

11 Failure to carry load includes any failure that occurs within 1 hour of starting.
12 These are the FTS values that include both the immediate FTS and the failures to carry load.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 100
90
80
Frequency of Repair Times

70
60
50
40
30
20
10
0
<1hr 1 to <2 2 to < 4 4 to < 8 8 to < 16 16 to <24 24 to <48 48 to < 72 72 to < 120 120 < 240 240 +

Hours to Repair

Figure 4: Distribution of repair times including logistics for EDGs after a failure 13

The MTTR is 37 hours if we eliminate a single outlier, which was nearly 2,000 hours and more than twice
the next-longest repair time. The MTTR is over 12 hours longer than the reported MTTRe (14). The
MTTM is the average time an EDG is offline due to scheduled maintenance and has been previously
reported by PREP (15) and IEEE (14). The MTTM is narrowly distributed, with a mean value of 1.7 hours;
however, most outages are due to extreme weather events, and scheduled maintenance is often
delayed when severe weather is expected. For assessing the performance of an EDG to provide power
during extended outages, we will ignore this short duration of an EDG being unavailable due to
scheduled maintenance. Providing power for very short outages (typically less than 15 minutes) is
accomplished by an Uninterruptable Power Supply (UPS) rather than an EDG, which is not the subject of
this study.

Under these assumptions, the OA of an EDG or the probability an EDG can attempt to provide backup
power can be calculated from annual failure rates and repair times (ignoring scheduling maintenance
downtime) from:

𝑇𝑜𝑡𝑎𝑙 𝑡𝑖𝑚𝑒 − 𝑇𝑖𝑚𝑒 𝑜𝑓𝑓𝑙𝑖𝑛𝑒 𝑑𝑢𝑒 𝑡𝑜 𝑟𝑒𝑝𝑎𝑖𝑟𝑠

𝑂𝐴 =
𝑇𝑜𝑡𝑎𝑙 𝑡𝑖𝑚𝑒
If we divide both the numerator and denominator by the number of failures, we estimate:

13The most common frequency for 16-24 hours is an artifact of the inclusion of large numbers of repairs from an
early PREP data collection for which only the average value for that set was recorded.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 𝑀𝑇𝐵𝐹 − 𝑀𝑇𝑇𝑅
𝑂𝐴 =
𝑀𝑇𝐵𝐹
Table 4 shows estimates for availability for modest size EDGs (<2,000 kW) that are well- or poorly
maintained. The MTTR is taken from the PREP data and applied to the Fehr and Hong Kong data sets.
The differences in OA are due to the failure rates or, equivalently, the number of required repairs.

Maintenance OA
Well-Maintained 99.98%
Poorly Maintained 99.84%
Table 4. Availability Estimates for Different Levels of Maintenance

These high availabilities reflect the limited number of runs per year of an EDG, and, thus, the small
number of potential failures per year.

4. Impact of EDG Reliability on Resiliency

The resiliency of a backup power system during a long duration outage depends on the repairability of a
failed EDG during the outage. The MTTR of 37 hours is already relatively long and reflects repairs during
short outages, which is the dominant type of outage generators experience. During a multiday outage, it
is unlikely that staff and equipment will be available to make the needed repairs. Thus, we treat the
EDGs in our analysis below as unrepairable for the duration of an outage.14

The reliability (12) of a single EDG at time t is given by:

−𝑡⁄
𝑅(𝑡) = 𝑂𝐴 × (1 − 𝐹𝑇𝑆) × 𝑒 𝑀𝑇𝑇𝐹

We use the estimates listed above from the Fehr and Hong Kong data sets to model modestly sized
EDGs (<2,000 kW) that are well-maintained or poorly maintained and compare these to the results for
larger EDGs used in the nuclear industry. Figure 5 shows the expected reliability for a single EDG for
outages that range from one hour to two weeks (336 hours). Also included is the predicted reliability if
one uses the MTBF provided by IEEE (14) instead of the MTTF. As discussed above this is incorrect but a
common mistake that results in major overestimate in EDG reliability.

14In estimating availability, we treated the EDGs as repairable. Most failures during an EDG’s lifetime that lead to
the unavailability of an EDG occur during testing or short duration outages.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 1
0.9
0.8
0.7
Reliability

0.6
0.5
0.4
0.3
0.2
0.1
0

180

270
105
120
135
150
165

195
210
225
240
255

285
300
315
330
0

90
15
30
45
60
75

Outage Duration (hours)

Well-Maintained NRC Poorly Maintained IEEE MTBF

Figure 5. The reliability of a single EDG over two weeks

In Figure 6, the same reliability results using the MTTF values for outages up to 12 hours are shown to
clearly illustrate the impact of different probabilities for FTS.

1
0.98
0.96
0.94
Reliability

0.92
0.9
0.88
0.86
0.84
0.82
0.8
0 1 2 3 4 5 6 7 8 9 10 11 12
Outage Duration (hours)

Well-Maintained NRC Poorly Maintained

Figure 6. The reliability of a single EDG for outages less than half a day

Figure 7 shows the reliability range expected for a well-maintained EDG with its 90% confidence
intervals.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 1

0.95
Reliability

0.9

0.85

0.8

0.75

156

192

276

312
108
120
132
144

168
180

204
216
228
240
252
264

288
300

324
336
0

36
12
24

48
60
72
84
96

Outage Duration (hours)

Mean 90% CI 90% CI

Figure 7. Mean and 90% confidence interval reliabilities for a well-maintained

EDG for outages up to two weeks

The impact of our sample size can be seen in figure 7. Uncertainty in the reliability metrics has a minor
impact on the performance of well-maintained EDGs for outages of only few hours but it grows as the
outage duration grows. Well-maintained EDGs have a reliability between 92% and 96% for providing
backup power for four days (96 hours) into an outage (current requirements for critical loads at
hospitals), between 90%-95% at one week, and between 75%-87% at two weeks (requirements for
critical loads at military facilities). These uncertainties should be considered when evaluating
performance for long duration outages. Not surprisingly, a below average maintenance EDG is unlikely
to provide power for durations longer than a day and have reliabilities of only 80% at 12 hours. These
figures reinforce the importance of following the current guidance on EDG maintenance. Using the IEEE
MTBF value yields incorrect predictions. Thus, single EDGs tied to individual loads do not provide a
robust source of emergency power for critical loads over multi-day outages.

One can substantially increase the reliability as seen in Figure 7 by using two EDGS for a single critical
load (a backup to the backup or N+1). This model represents a best-case reliability because it ignores
failures due to shared components, such as switch gear, that could still represent a potential single point
failure. The failures of two EDGs are independent; one achieves higher reliability but at double the cost.

As shown in Figure 8, two poorly maintained EDGs still yield very limited reliability beyond 24 hours
(<90%), but two well-maintained EDGs achieve relatively high reliabilities over 96% out to two weeks.
This is still considerably less than the multiple nines often quoted as the desired or required reliability.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 1
0.9
0.8
0.7
Reliability

0.6
0.5
0.4
0.3
0.2
0.1
0

168
180
192
108
120
132
144
156

204
216
228
240
252
264
276
288
300
312
324
336
0
12
24
36
48
60
72
84
96

Outage Duration (hours)

Well-Maintained Poorly Maintained

Figure 8. Reliability of two EDGs (backup to a backup) as a function of
maintenance

The situation is complicated if one considers a campus-like environment (hospital complexes,

universities, military installations) where critical loads occur in multiple buildings. The most common
architecture for providing backup power is to place a single EDG at each building with a critical load.
Figure 9 illustrates the probability that all critical loads across 1, 8, and 80 buildings can be met even
with well-maintained EDGs.

1
0.9
0.8
0.7
Probability

0.6
0.5
0.4
0.3
0.2
0.1
0
180
192
108
120
132
144
156
168

204
216
228
240
252
264
276
288
300
312
324
336
0
12
24
36
48
60
72
84
96

Outage Duration (hours)

1 Building 8 Buildings 40 Buildings 80 Buildings

Figure 9. Probability critical load can be met using one well-maintained EDG at
each critical building across a campus with 1, 8, 40, or 80 buildings

Clearly the likelihood of meeting the load across many buildings becomes very small in a few days. This
is simply due to the greater opportunity for failures when multiple independent generators are involved.
Results for EDGs that have poor maintenance are worse.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 An alternative way to view these results is to calculate the number of buildings that will lose power
during an outage. Assume we have N buildings with critical loads on a campus, each with an EDG with a
reliability R(t) and a cumulative failure probability given by F(t)=1-R(t). The probability that k buildings,
each with one EDG, have power during an outage at time t is given by the binomial distribution:
𝑁
𝑃(𝑘, 𝑁) = ( ) 𝑅𝑘 𝐹 𝑁−𝑘
𝑘
The mean number of buildings at time t that still have power can be determined by calculating the mean
value of k at time t:

𝑀𝑒𝑎𝑛(𝑘) = 𝑁 × 𝑅(𝑡)
The number of buildings without power is simply the number of buildings, N, minus these numbers. Or,
equivalently, the mean fraction of buildings on a campus without power during a grid outage is 1-R(t),
which is independent of the number of buildings.

In assessing the risks associated with the loss of power to buildings, one should consider not only the
expected average performance but the range of failures likely to occur. Deviation from the mean
performance is expected due to two causes. First the uncertainty in the key reliability metrics (OA, FTS,
and MTTF) due to the limitations in the empirical data sets, which is captured by the reported 90%
confidence intervals illustrated in Figure 6. Second, the inherent probabilistic nature of outages as
expressed by the binomial distribution. This is measured by the standard deviation of the binomial
distribution.

𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛(𝑘) = 𝑆𝑄𝑅𝑇(𝑁 × 𝑅(𝑡) × (1 − 𝑅(𝑡)))

The magnitude of these deviations from the mean is dependent on the number of buildings. The relative
magnitude of the standard deviation of the binomial distribution decays as one over the square root of
the number of buildings. For a small number of buildings, the probabilistic effect is significant, but for
campuses with many buildings (>40), the uncertainty in the reliability metrics dominates.

Figure 10 shows the mean fraction of buildings that do not have power plus one standard deviation
from both the uncertainty in the underlying reliability metrics and the probabilistic nature of failures as
a function on the number of buildings with one well-maintained generator tied to each building. This
estimates the 90% worst-likely scenario as a function of number of buildings.

For long outages, this can be a significant number of buildings. For all campuses, the mean fraction of
buildings expected to be without power is roughly one-fifth of facilities during a two-week outage. For
large campuses with greater than 40 critical-load buildings, one quarter of the buildings can be without
power and over a third of buildings for small campuses.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 40.0%
Buildings Without Power 35.0%
30.0%
25.0%
20.0%
15.0%
10.0%
5.0%
0.0%

180

216

252

288

324
108
120
132
144
156
168

192
204

228
240

264
276

300
312

336
0
12
24
36
48
60
72
84
96

Outage Duration (hours)

Mean N=4 N=8 N=40

Figure 10. Mean fraction of buildings that lose power plus potential positive
deviations when one EDG supports each building

5. Conclusions

There is an absence of realistic estimates for the reliability of the primary technology used for backup
power, an emergency diesel generator, despite the amount of literature for modeling energy resilience
during outages. Using IEEE reported mean time between failure results in incorrect predictions. We have
presented estimates for the reliability of modern commercial emergency diesel generators that are
commonly used based on empirical data sets. The uncertainty of these estimates is determined by the
size of the available data sets, which results in 90% confidence interval roughly equal to 50% of the
mean values. Reliance on poorly maintained emergency diesel generators leads to a very significant risk
of power loss at a facility during a grid outage. Even for well-maintained emergency diesel generators,
less than perfect reliability is important to consider when estimating a backup power system’s
performance. For multiday outages, the reliance on single emergency diesel generators for supporting
backup loads is unwarranted and leaves the end user with a significant risk not usually recognized.

For campus-like environments, where multiple buildings house loads supported by independent
individual emergency diesel generators, the number of buildings without power can be significantly
higher than often recognized. Increasing the redundancy of EDGs provides much higher reliabilities but
at a significant cost.

The analysis presented here does not take into account the impact of hybrid backup power systems that
combine emergency diesel generators with other distributed energy resources or variable load on
backup systems which allow networked distributed energy resources. In future papers we will show
that networked or microgrid-based solutions utilizing a set of emergency diesel generators can provide
much higher levels of reliability. We will also report on the reliability contribution provided by
intermittent renewable energy coupled to battery storage.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 Acknowledgements

We wish to thank Peyton Hale and Chris Thompson of the U.S. Army Corps of Engineers for providing the
PREP data as of December 2018 and generously answering our questions, and Dr. Fehr for generously
providing NREL with a copy of his database and answering all our questions.

This work was authored by the National Renewable Energy Laboratory, operated by Alliance for
Sustainable Energy, LLC, for the U.S. Department of Energy (DOE) under Contract No. DE-AC36-
08GO28308. Support for the work was also provided by the Department of Defense’s Environmental
Security Technology Certification Program (ESTCP) under [Agreement IAG-18-02080]. The views
expressed in the article do not necessarily represent the views of the DOE or the U.S. Government. The
U.S. Government retains and the publisher, by accepting the article for publication, acknowledges that
the U.S. Government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or
reproduce the published form of this work, or allow others to do so, for U.S. Government purposes.

References
1. The President's National Infrastructure Advisory Council. Surviving a Catastrophic Power Outage: How
to Strengthen the Capabilities of the Nation. December 2018.
https://www.dhs.gov/sites/default/files/publications/NIAC%20Catastrophic%20Power%20Outage%20St
udy_FINAL.pdf.

2. Larsen, Peter, Kristina LaCommare, Joseph Eto, and James Sweeney. Assessing Changes in the
Reliability of the U.S. Electric Power System. LBNL - 188741. Berkeley, CA: Lawrence Berkeley National
Laboratory. August 2015. https://emp.lbl.gov/sites/all/files/lbnl-188741.pdf.

3. National Energy Technology Laboratory. Backup Generators (BUGS): The Next Smart Grid Peak
Resource. U.S. Department of Energy. DOE/NETL-2010/1406. April 2010.
https://www.smartgrid.gov/files/bugs_the_next_smart_grid_peak_resource_04-2010.pdf.

4. Phillips, Julia, Kelly Wallace, Terence Kudo, and Joseph Eto. Onsite and Electric Power Backup
Capabilites at Critcal INfrastructure Facilites in the United States. ANL/GSS-16/1. Argonne, IL: Argonne
National Laboratory. April 2016. http://eta-publications.lbl.gov/sites/default/files/onsite-and-electric-
power-backup.pdf.

5. Marqusee, Jeffrey, Dorothy Robyn, and Craig Schultz. Power Begins at Home: Assured Energy for U.S.
Military Bases. January 2017.
https://www.pewtrusts.org/~/media/assets/2017/01/ce_power_begins_at_home_assured_energy_for_
us_military_bases.pdf.

6. NFPA (National Fire Protection Association). NFPA 110 Standard for Emergency and Standby Power
Systems. 2019. https://www.nfpa.org/codes-and-standards/all-codes-and-standards/list-of-codes-and-
standards/detail?code=110.

7. FEMA (Federal Emergency Management Agency). P-1019, Emergency Power Systems for Critical
Facilities: A Best Practices Approach to Improving Reliability. September 2014.
https://www.fema.gov/media-library/assets/documents/101996.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.
 8. IEEE. "IEEE Recommended Practice for Emergency and Standby Power Systems for Industrial and
Commercial Applications." IEEE Std 446-1995. July 1996. https://ieeexplore.ieee.org/document/653237.

9. DoD (U.S. Department of Defense). Unified Facility Criteria (UFC); Engine Driven Generator Systems for
Prime and Standby Power Applications. UFC 3-540-01. October 2017.
https://www.wbdg.org/FFC/DOD/UFC/ufc_3_540_01_2014_c1.pdf.

10. DoD. Unified Facility Criteria (UFC): OPERATION AND MAINTENANCE (O&M): GENERATORS. UFC 3-
540-07. February 2017. https://www.wbdg.org/FFC/DOD/UFC/ufc_3_540_07_2018.pdf.

11. NFPA. NFPA 101 Life Safety Code. 2018. https://www.nfpa.org/codes-and-standards/all-codes-and-

standards/list-of-codes-and-standards/detail?code=101.

12. Birolini, Alessandro. Reliability Engineering Theory and Practice. Springer, 2007.

13. IEEE 493. Design of Reliable Commercial Power Systems. IEEE, 2007.

14. IEEE Standard 3006.8. Recommended Practice for Anayzing Reliability Data For Equipment Used in
Industrial and Commercial Power Systems. October 2018. https://standards.ieee.org/content/ieee-
standards/en/standard/3006_8-2018.html.

15. U.S. Army Corps of Engineers. TM 5-698-5, Survey of Reliability and Availability Information for
Power Distribution, Power Generation, and Heating, Ventilating and Air Conditioning (HVAC)
Components for Commercial, Industrial and Utility Installations. July 2006.
https://www.wbdg.org/ffc/army-coe/technical-manuals-tm/tm-5-698-5.

16. IEEE. 3600 Standards Collection, “ Historical Reliability Data for IEEE 3006 Standards: Power Systems
Reliability.” 2012.

17. Thompson, Christopher, Peyton Hale, and Robert Arno. "Decanting the Data: The Gold Book Presents
Equipment Reliability Refreshment." IEEE Transactions on Industry Applications 482, no. 2 (March-April
2012): 772-776. DOI: 10.1109/TIA.2011.2180873.

18. Du, P.Y., J. Burnett, and S.M. Chan. "Reliability of Standby Generators in Hong Kong Buildings." IEEE
Transactions on Industry Applications 39, no. 6 (November-December 2003): 1592-1595.
DOI: 10.1109/TIA.2003.818978.

19. Schroeder, John A. Enhanced Component Performance Study: Emergency Diesel Generators 1998–
2016,. s.l. : INL/LTD 17-44204, April 2018.

20. Fehr, Stephen John. "Emergency Diesel-Electric Generator Set Maintenance and Test Periodicity."
Ph.D. diss., Old Dominion University 2017. DOI: 10.25777/q2nk-n411.
https://digitalcommons.odu.edu/emse_et, 2017.

21. NRC (U.S. Nuclear Regulatory Commission). “About the Reliability Calculator.” Last modified March
6, 2019. https://nrcoe.inl.gov/radscalc/.

Pursuant to the DOE Public Access Plan, this document represents the authors' peer-reviewed, accepted manuscript.
The published version of the article is available from the relevant publisher.