IEEE AUTOTESTCON 2008

Salt Lake City, UT, 8-11 September 2008

IEEE Standards for Prognostics and Health Management

John W. Sheppard1, Mark A. Kaufman2, Timothy J. Wilmering3
1
The Johns Hopkins University, 3400 N. Charles Street, Baltimore, MD 21218, jsheppa2@jhu.edu
2
NSWC Corona Division, PO Box 5000, Corona, CA 92878, mark.kaufman@navy.mil
3
The Boeing Company, PO Box 516, M/C S270-3800, St. Louis, MO 63166, timothy.j.wilmering@boeing.com

Abstract – Recently, operators of complex systems such as aircraft, parallel the diagnostic data and diagnostic service elements
power plants, and networks, have been emphasizing the need for already contained in the framework.
online health monitoring for purposes of maximizing operational PHM has been defined as “a maintenance and asset
availability and safety. The discipline of prognostics and health management approach utilizing signals, measurements,
management (PHM) is being formalized to address the information
models, and algorithms to detect, assess, and track degraded
management and prediction requirements for addressing these
needs. In this paper, we will explore how standards currently under
health, and to predict failure progression [1].” As defined,
development within the IEEE can be used to support PHM PHM encompasses much more than is currently addressed by
applications. Particular emphasis will be placed on the role of PHM SCC20; however, the AI-ESTATE standard has been found
and PHM-related standards with Department of Defense (DOD) to address many PHM issues related to fault/failure diagnosis.
automatic test systems-related research. The DMC is currently developing standards under the
Software Interface for Maintenance Information Collection
Keywords – Prognostics, PHM, CBM, AI-ESTATE, SIMICA and Analysis (SIMICA) project (P1636) that are likely to
address additional information management requirements for
I. INTRODUCTION PHM. These standards capture historic information that can
be used to analyze maintenance and diagnostic processes and
In 1976, the IEEE established the Standards Coordinating to tie these analyses to system fleets or to individual systems.
Committee 20 (SCC20) for purposes of standardizing on the The result is a collection of standards that can support
Abbreviated Test Language for All Systems (ATLAS). Since diagnostic maturation and PHM process improvement. The
then, SCC20 has expanded its scope to develop standards for focus of this paper is on applying the AI-ESTATE and
larger system-level test and diagnostic related systems. In SIMICA standards in PHM systems. The discussion in this
1989, the IEEE approved a project authorization request paper highlights the recent results in developing these
(PAR) for SCC20 to develop a new standard focusing on standards and focuses on how they can be used to satisfy
diagnostic systems that use techniques from the maturing PHM information management requirements.
field of artificial intelligence—the Artificial Intelligence
Exchange and Service Tie to All Test Environments (AI- II. APPROACHES TO PHM
ESTATE) standard under project P1232. In 1995, SCC20
approved and published the AI-ESTATE standard, IEEE Std Generally, PHM systems incorporate functions of
1232-1995, and in 2002, the standard was updated. Today, condition monitoring, state assessment, fault or failure
SCC20, under the management of its Diagnostic and diagnostics, failure progression analysis, predictive
Maintenance Control (DMC) subcommittee is completing a diagnostics (i.e., prognostics), and maintenance or operational
new update to the AI-ESTATE standard, and this standard is decision support. Ultimately, the purpose of a PHM system is
emphasizing its broad scope by embracing PHM-related to maximize the operational availability and safety of the
issues. target system.
The DOD ATS Framework Working Group is a multi- The primary area of interest in this paper is the impact and
service/industry/academic partnership that is focusing on potential benefit of standardization supporting
defining an information framework and identifying standards interoperability for PHM systems. More specifically, the area
for next-generation automatic test systems (ATS). Based on of interest lies in the “predictive” portion of PHM—the
work in the 1990s when the ATS Research and Development ability to predict from information about some system state
Integrated Product Team defined a set of “critical interfaces” when a significant future event affecting the performance of
for ATS, the current working group has been selecting, the system (such as failure) might occur. Often, this
supporting the development of, and demonstrating prediction is characterized as estimating the remaining useful
commercial standards to be used in ATS. In 2007, the life (RUL) of a component or system [2], [3]. Standardization
working group decided to expand its scope to embrace in information systems involves careful, formal definition of
information requirements for PHM as well and added two concepts and information elements for the target system. We
new “elements” to its framework—an element for prognostic believe RUL is misleading at the system level in that it
data (PROD) and an element for prognostic services (PROS). suggests no repair is possible, thus extending the RUL of the
The working group decided to focus on these elements to system. Therefore, we suggest the term time to fail (TTF) but

978-1-4244-2226-5/08/$25.00 ©2008 IEEE

Authorized licensed use limited to: Central Queensland University. Downloaded on August 10,2022 at 02:41:03 UTC from IEEE Xplore. Restrictions apply.
 Preprocessing

Organization

Assessment
Estimation
Extraction

Prediction
Feature

Impact
State

State
Data

Models

Figure 1. Notional CBM/PHM System

note possible confusion with the TTF measure as defined in at time t [8]. Usually, reliability predictions are used to
[4]. Specifically, Vachtsevanos et al. define TTF as “the estimate future failure based on current test results by
duration between initiation of the fault and the time when the applying a probability distribution such as the exponential
failure occurs.” As an alternative, we define TTF to be “the distribution (i.e., P(Di) = 1 – exp[–λit]). One of the principal
time from a measurement of system state to some failure of shortcomings of using the exponential distribution is that it
interest in the system.” imposes a “Markov” assumption, meaning that the future
One can think of PHM as being applied in an operating prediction of a failure is independent of the history of the unit
environment in which there is interaction with and feedback given the current measurement. Given the strength of this
to the system being monitored (Figure 1). When building a assumption, alternative reliability methods have applied the
PHM system, three components are necessary for prognostics Weibull distribution for the predictions since it relaxes the
to be effective (which are highlighted in Figure 1)—the assumption of constant failure rates as well as the Markov
ability to estimate the current state of the system, the ability assumption [9].
to predict future state, and thereby time to fail, and the ability Data-Driven Methods: In a sense, POF and reliability-
to determine the impact of the assessment on system based methods form end-posts along a spectrum of
performance and the need for corrective or mitigating action. techniques for prognostic methods. POF methods depend on
In all three cases, system-specific models must also be high-resolution models but do not scale well. Reliability
provided. In support of these components, several approaches methods rely on statistical characteristics of populations of
are being applied. systems and do not handle idiosyncrasies of specific systems.
Physics Model-Based Methods: Perhaps the most As an attempt to provide a compromise approach, data-driven
effective method in terms of high-fidelity prediction of methods such as regression models [10], time series analysis
system degradation is the application of physics-of-failure [11], and neural networks [12] are being applied. Each offer
(POF) models to structural degradation and structural health an advantage of being able to learn models based on
monitoring systems [5]. POF methods focus on issues such as empirical data but also suffer from the inability to learn
material deformation, fracture, fatigue, and material loss. portions of the model where no such data exists.
Recent attempts at applying POF methods to electronic Probability-Based Methods: Lessons drawn from signal
prognosis have focused on the material degradation of processing, target tracking, and state estimation have
interconnects and substrates [6], [7]. While highly accurate, identified a number of probabilistic models showing promise
POF approaches tend to be computationally prohibitive to for PHM. Specifically dynamic Bayesian network (DBN)
apply at the system level. This limitation has led to architectures such as hidden Markov models (HMM) [13] and
alternative approaches being developed and applied, Kalman filters [11] have been suggested as methods for using
sometimes in combination with POF methods. historical, sequential data to predict future failure. The
Reliability-Based Methods: Perhaps the simplest concern with these models relates to the so-called “diffusion
approach to predicting failure is based on statistical reliability of context” phenomenon where, because of conditional
models of component failure. Recall that reliability is defined independence, the affect of past experience diffuses the
as the probability that a component or unit will be functioning

Authorized licensed use limited to: Central Queensland University. Downloaded on August 10,2022 at 02:41:03 UTC from IEEE Xplore. Restrictions apply.
ability to predict. This, in fact, is directly related to the and maintenance system interfaces. These standards,
Markov assumption also inherent in the reliability models developed under the auspices of the IEEE Standards
discussed above [14]. The use of so-called “input-output Coordinating Committee 20 on Test and Diagnosis for
hidden Markov models” has been suggested as an approach Electronic Systems include the Signal and Test Definition
to combat this problem [15]. standard [18], the Automatic Test Markup Language (ATML)
As should be evident from the above review, the “silver family of standards [19], the AI-ESTATE standard [20], and
bullet” for PHM systems has yet to be discovered or the SIMICA standards [21]. Of particular interest to us are
developed. In fact, arguably, PHM technology is still very the AI-ESTATE and SIMICA standards. Within the SIMICA
much in its infancy. Therefore, it is interesting to be family are two additional standards—Test Results [22] and
considering standardization of PHM elements. Even so, the Maintenance Action Information [23].
Machinery Information Management Open Systems Alliance
(MIMOSA) has adopted the development and support of the A. AI-ESTATE
Open System Architecture Condition Based Management
(OSA-CBM) standard that purports to provide a standard IEEE Std 1232 describes the information comprising the
architecture for CBM and PHM systems. diagnostics domain, i.e., information related to system test
OSA-CBM is an architecture standard organized around and diagnosis. The description of the diagnostic domain
seven “hierarchical” layers: sensor/transducer, data enables the exchange of diagnostic information between
acquisition, data manipulation, state detection, health applications. IEEE Std 1232 also supports modular diagnostic
assessment, prognostic assessment, and advisory generation. architectures and interoperability with other test-related
Of particular interest here are the health assessment, software assets. The 1232 standard was developed using
prognostic, and decision support layers [16]. Using the three information modeling practices with the ISO EXPRESS
key components of a PHM system identified above, we see modeling language [24], resulting in the definition of five
the health assessment layer being responsible for health state models addressing static and dynamic aspects of the
estimation, and the prognostics layer being responsible for diagnostic domain.
predicting time to fail. Both layers must address uncertainty Based on the formal information models, AI-ESTATE
management and confidence prediction. These layers are provides two different mechanisms for exchanging diagnostic
shown to be connected with state detection below and information. The historical approach uses the Standards for
advisory generation above. the Exchange of Product model data (STEP) Physical File
Currently, the OSA-CBM standard provides a Unified Format defined in [25]. This format specifies a simple ASCII,
Modeling Language (UML) model identifying key “objects” flat file utilizing tokens within an attribute-value structure
to be defined in a standard CBM system [17]. Unfortunately, and must be used in conjunction with the EXPRESS Schema.
current implementations of the OSA-CBM architecture have SCC20 also plans to use an XML schema consistent with the
not incorporated the means of standardizing the semantics of information model based on ISO 10303 Part 28 [26].
the information being communicated between system Finally, in addition to the information models being
components, This is where the work of SCC20 hopes to developed, AI-ESTATE defines a set of software services to
contribute and is what we discuss next. be used when integrating a diagnostic reasoner into a test
system. The reasoner services are being specified using the
III. STANDARDS IN MAINTENANCE AND Web Services Description Language (WSDL) [27], arising
DIAGNOSTICS mostly due to the increased emphasis on web services and
XML for exchanging information.
Fundamentally, prognosis is an extension of fault or Given both the published AI-ESTATE and the current
failure diagnosis. In addition, given the fact prognosis revision being developed by SCC20, several relationships are
attempts to anticipate and predict impending failure, the apparent between AI-ESTATE information elements and key
nature of the maintenance process under a PHM system is components of PHM systems. For instance, it is sometimes
fundamentally different from a maintenance process based on desirable to qualify test using measures of confidence, and
taking corrective action in response to a reported failure. failure/fault predictions can be provided with associated
Currently, few standards exist of direct relevance to probabilities and levels of confidence. That said, AI-ESTATE
prognostic systems and PHM systems; however, because of is currently limited to assigning discrete outcomes, both to
the close ties between PHM and traditional diagnostic and tests and diagnostic conclusions. Currently, AI-ESTATE is
maintenance systems, several standards for the maintenance also limited to supporting systems that provide state
and diagnostic communities can be applied to PHM. As we assessments at the current point in time, assuming
will discuss below, it is also hoped that these same standards propositional representations of the associated diagnoses.
will serve as a starting point for the development or This is significant because, currently, none of the models
maturation of standards for PHM. support time to fail predictions. Prior proposals have been
Since the mid 1970s, SCC20 has been developing supplied to SCC20 for supporting temporal logic [28] and
standards, originally focused on test specification and test dynamic Bayesian networks that would be useful for
programming, but more recently focusing on test, diagnostics, prognostic algorithms [29]; however, neither was considered

Authorized licensed use limited to: Central Queensland University. Downloaded on August 10,2022 at 02:41:03 UTC from IEEE Xplore. Restrictions apply.
sufficiently mature to be included in the standard. In fact, it is information (MAI), being standardized under IEEE P1636.2
unclear that industry consensus exists on the semantics of an [23].
elementary TTF metric at this point in time. The MAI standard is not intended to support a PHM
process directly, as the Test Results standard or AI-ESTATE
B. Test Results might. Instead, MAI captures what has been done with
respect to a system of interest, either in response to a failure
The current draft of the SIMICA standard is focused on or during preventative maintenance. Nevertheless, the
providing a top-level information model for maintenance information captured in an MAI document can be used to
information. This model will provide an “umbrella” perform data mining and data analysis to support diagnostic
representation to correlate the semantics of several lower- and maintenance system maturation as well as to assist in
level partitions of the system operational and maintenance developing prognostic models and systems.
information domain. While that information model was being
completed, two of those lower level “component” standards D. Related Non-IEEE Standards
within the SIMICA family have been under development as
well. The first—SIMICA Test Results—has been approved This paper has focused on the IEEE standards that have
and published as a trial use standard. potential in supporting PHM. Currently, no IEEE standards
The Test Results standard provides an XML schema and exist that are dedicated to PHM; however, standards exist
accompanying information model to specify a means for outside of the IEEE, in addition to OSA-CBM, that focus on
exchanging test measurement information. The focus of the issues such as health monitoring and condition-based
standard is to capture historical information about the actual maintenance (CBM).
conduct of tests and includes information such as UUT ISO 10303 Part 239 defines an “application protocol” for
identification, measurements, specified test limits, and product life cycle support (PLCS) [32]. The purpose of this
information specific to test session such as setup, test standard is to facilitate exchange of information about
sequence, and fault indictment information [22], [30]. complex “engineering assets” for the purposes of life cycle
Typically, PHM requires systems that perform online support. The data exchange is accomplished through the
monitoring of the system of interest. The 1636.1 standard definition of data exchange specifications (DEX) tied to
provides direct support for a PHM system in that it captures specific domains and derived from the PLCS information
the history of the monitored data. Measurements, test limits, model.
outcomes, and calibration information, coupled with time With recent emphasis being placed on modifying product
stamps for when the data was collected, enable offline maintenance practices from reactive to condition-based or
processing of the data to determine system state, perform “just-in-time,” product-specific data has become critical. The
diagnosis, and when coupled with a prognostic model, Organization for the Advancement of Structured Information
contribute to the prediction of future system state. Standards (OASIS) has developed a PLCS DEX specifically
Since the 1636.1 standard emphasizes data exchange targeted at aviation maintenance [33]. The Aviation
through XML, real-time applications of the standard are not Maintenance DEX focuses on sharing information in four
directly supported. A PHM system can, however, make use of categories:
the information model to determine the relevant types of
information to be captured and the definitions, relationships, 1. Historical maintenance activities;
and constraints on that information necessary to ensure 2. Activities that, while not maintenance-specific, may
interoperability between other components that may require impact future maintenance (e.g., flying sorties);
the data (such as a diagnostic system). By applying a similar 3. Estimation of system state (e.g., fault state, and
“service-oriented” architecture for the PHM system as that serviceability); and
proposed in AI-ESTATE, online processing of the test results 4. Activities affecting product inventories.
could be supported in a standardized way.
The PLCS architecture uses multiple DEXs to support
C. Maintenance Action Information information exchange between agents needing the
information. DEXs related to the Aviation Maintenance DEX
Recently, a new initiative was undertaken that was initially include:
intended to support the capture and processing of historical
maintenance data for military systems. This process involved 1. Product as-designed structure;
surveying maintenance processes for each of the US military 2. Specific product as-delivered structure;
services to identify common, essential maintenance 3. Maintenance plan;
information that is captured on maintenance action forms 4. Faults related to the product; and
(MAF). SCC20 then proceeded to generalize the information, 5. Specific information on maintenance tasks.
to address maintenance processes in non-military
applications. The result was the development of an ISO also provides a collection of standards focusing on
information model and XML schema for maintenance action condition monitoring and diagnosis of machines [34]. These

Authorized licensed use limited to: Central Queensland University. Downloaded on August 10,2022 at 02:41:03 UTC from IEEE Xplore. Restrictions apply.
standards are developed by the technical committee on for prediction and are based on discrete test and diagnosis
mechanical vibration and shock and focus on test design, outcomes.
measurement, and data processing focused specifically in The SIMICA family of standards (including Test Results
these areas. and MAI) focuses on historical information but provide a
means for using that information to improve both diagnostics
IV. STANDARDIZING PHM and prognostics. Recognizing that PHM systems based
entirely on discrete, outcome-based testing will be severely
In this paper, we have been focusing on standards for limited in their ability to predict future states, we find that the
information exchange. We recognize that standards are also Test Results standard, while providing outcome-based data,
necessary for communications, form-fit-function of devices, also provides a method for capturing actual measurement
physical interfaces, timing, calibration, etc. Many standards data. This standard is oriented towards automatic test
exist to support such elements, and such standards also need systems; however, the supporting information model provides
careful consideration when building a PHM system. In the a means for transitioning to a real-time health monitoring
previous section, we reviewed the principal standards system as well.
supporting maintenance and diagnosis and suggested how The MAI standard also provides both an XML schema and
these standards might support the PHM enterprise. In this an information model. MAI focuses on the maintenance
section, we will discuss the key PHM characteristics that are process rather than the test process; therefore, it is easily
present or that need to be included in standards to meet PHM generalizable to PHM.
requirements.
B. Enhancements Needed to Current Standards
A. PHM Characteristics in Current Standards
The primary challenge facing SCC20 is in determining
Fundamentally, PHM systems incorporate functions of how best to extend or adapt current standards to PHM. We
condition monitoring, state assessment, fault or failure believe that, at a minimum, the following must be provided as
diagnostics, failure progression analysis, predictive enhancements or additions to the current set of standards:
diagnostics (i.e., prognostics), and maintenance support. Thus
these functions must be supported in any collection of PHM- 1. A means for representing graded health information
related standard. Specifically, PHM-related standards must be rather than limiting diagnostics to discrete outcomes.
able to represent and exchange measured, observed, and 2. Given the ability to represent graded health
inferred information about the target system and its information, a means to “roll up” failure progression
operational environment, information about the current state information to higher levels in the system hierarchy.
as well as either an estimate of some future state or an 3. Relaxation or augmentation of the outcome-based
estimate of when some target state might be reached, approach to diagnosis to support state estimation
historical information about the operation and maintenance of based on real-valued test results.
the target system, and various models of the system. 4. The ability to support periodic measurements and
The OSA-CBM standard provides a detailed object model, correlation between time series.
represented in UML that identifies key data items, objects, 5. Incorporation of usage, operational, and
and their relationships within a CBM system. Note that the environmental data in performing state assessment
algorithms are referenced via MIMExtTypes which are and diagnosis.
extensions to MIME types. No facility is provided for 6. Representation of failure progression/degradation
standard algorithm specification. In addition, OSA-CBM information for specific systems.
provides detailed facilities for exchanging logical 7. A framework for integrating and combining
propositions about the system. These are supported by information from multiple models and model types
detailed health assessment data items that include health level (e.g., physics-based, reliability-based, and data
of the system and health grade of an item. driven) to exploit the specific advantages of each
The IEEE standards provide sound support for information type.
exchange supporting the process of test and diagnosis of the
target system. The AI-ESTATE standard provides a As a specific example, SCC20 is currently exploring a
foundation for diagnostic assessment and includes four change to the AI-ESTATE standard that would address the
alternatives for diagnosis—static fault trees (or decision need for capturing “grayscale” (i.e., graded) health
trees), D-matrix-based systems (e.g., dependency models), information. Including grayscale health supports reasoning
logic-based models (e.g., rule-based expert systems), and about current state of degradation and projecting future
Bayesian networks. All four alternatives have been failure progression. It is also applicable to incipient fault
demonstrated to provide effective and accurate diagnostics. In detection. To address grayscale health, changes will be
addition, the standard has been defined to address all test required in the Common Element Model to specify that a
environments, thus there is no implicit or explicit focus on diagnosis outcome need not be discrete as well as the
ATS. As stated earlier, they do not currently provide facilities Dynamic Context Model to record the inferred grayscale

Authorized licensed use limited to: Central Queensland University. Downloaded on August 10,2022 at 02:41:03 UTC from IEEE Xplore. Restrictions apply.
 minX
maxX

midPoint memberType
REAL *MembershipFunction FunctionClass

shape
softOutcome
REAL STRING

confidence (ABS) valueDomain

*ConfidenceValue AssignedValue
Outcome
1
QualifyerType
*qualifier

*ActionOutcome *DiagnosisOutcome *TestOutcome

(Test.allowedOutcomes)
(Action.allowedStatus) *(INV) forTest
*(INV) forAction
4,3(2) 4,1(3) 4,2(2)

2,2 Action 3,3 Diagnosis 2,1 Test

(Diagnosis.allowedOutcomes)
*(INV) forDiagnosis

Figure 2. Soft Outcome Definition for PHM in AI-ESTATE

health estimate. In addition, at least one of the inference

models (i.e., fault tree, D-matrix, logic, and Bayes) will have V. CONCLUSION
to be updated to indicate inferences of grayscale health from
test results or outcomes. Two of the primary reasons for standardization are to
Figure 2 illustrates one possible change to the definition of reduce cost by improving interoperability and minimizing
the Outcome entity in the Common Element Model. The only repeated design of similar systems. The IEEE SCC20
difference from the definition in [35] is the addition of the standards focus on promoting information interoperability
“softOutcome” attribute. The purpose of this attribute is to between components of a test or health monitoring system.
associate a soft or gray-scale function to the outcome where The emphasis by the DOD on acquisition reform based on
the domain of the function is specified but the range is [0, 1]. commercial standards for ATS, combined with declining
Six types of functions have been identified—triangular, budgets mandates the need for more affordable health
trapezoidal, radial (e.g., Gaussian), hard limiter, linear management system development and operation. Cost must
threshold, and sigmoidal. The first three correspond to be reduced. Interoperability must be achieved. Information
common “membership” functions from fuzzy logic, and the must be shared.
latter three correspond to common transfer functions for The cost of reactive maintenance has become prohibitive,
function approximators such as neural networks. The especially for complex systems. Even with the development
attributes, “minx,” “maxX,” “midpoint,” and “shape” provide of comprehensive standards focusing on interoperability and
key parameters for the function. The minimum and maximum reuse, the way systems are maintained must and is changing.
X values specify the domain of the function. All of the Providers are required to implement interoperable systems
functions have a point about which the functions are (usually) based on these standards. With the recent focus on CBM and
symmetric. This is identified with “midpoint.” Finally, the prognostics, the need exists for exchanging more robust
“shape” parameter is specific to the type of function. For information in a timely way that will enable identifying and
example, for the radial function, the shape parameter correcting a fault before it occurs. As with traditional testing,
represents the spread (or variance) of the function where, for standards for CBM and PHM are also required.
the sigmoid function, it represents the slope at the point of Technological advances in CBM and PHM have identified
inflection. Given the soft outcome, a corresponding attribute core types of information needed for health monitoring
can then be added to “ActualOutcome” in the AI-ESTATE systems. What we have found is that there is a strong overlap
Dynamic Context Model to assign a value to the outcome between the type of information needed for health monitoring
based on the associated function. and the type of information used in traditional diagnosis.
Therefore, the IEEE has been working to identify and expand

Authorized licensed use limited to: Central Queensland University. Downloaded on August 10,2022 at 02:41:03 UTC from IEEE Xplore. Restrictions apply.
its existing test and diagnostic standard to address PHM [18] IEEE Std 1641-2004, IEEE Standard for Signal and Test Definition,
Piscataway, New Jersey: IEEE Standards Association Press, 2004.
requirements. It is likely new standards will also emerge
[19] IEEE Std 1671-2006, IEEE Trial Use Standard for Automatic Test
where enhancing current standards would constitute a “force Markup Language (ATML) for Exchanging Automatic Test Information
fit.” Fortunately, SCC20 is well-positioned to identify and via eXtensible Markup Language (XML), Piscataway, New Jersey:
address those challenges. IEEE Standards Association Press, 2006.
[20] IEEE Std 1232-2002, IEEE Standard for Artificial Intelligence
Exchange and Service Tie to All Test Environments (AI-ESTATE),
REFERENCES Piscataway, New Jersey: IEEE Standards Association Press, 2002.
[21] IEEE P1636, Draft IEEE Trial Use Standard for Software Interface for
[1] Kalgren, Patrick, Carl Byington, Michael Roemer, and Matthew Maintenance Information Collection and Analysis (SIMICA), D1.6,
Watson, “Defining PHM, A Lexical Evolution of Maintenance and Piscataway, New Jersey: IEEE Standards Association Press, 2008.
Logistics,” IEEE AUTOTESTCON 2006 Conference Record, Anaheim, [22] IEEE Std 1636.1-2007, IEEE Trial Use Standard for Software
California, September 2006, pp. 353–358. Interface for Maintenance Information Collection and Analysis
[2] Hess, Andrew, “Real Prognostics—Challenges, Issues, and Lessons (SIMICA): Exchanging Test Results and Session Information via the
Learned: Chasing the Big 'P',” First International Forum on Integrated eXtensible Markup Language (XML), Piscataway, New Jersey: IEEE
System Health Engineering and Management in Aerospace, NASA, Standards Association Press, 2007.
Napa Valley, CA, November 7–10, 2005. [23] IEEE P1636.2, Draft IEEE Trial Use Standard for Software Interface
[3] Roemer, Michael, Carl Byington, Gregory Kacprzynski, and George for Maintenance Information Collection and Analysis (SIMICA):
Vachtsevanos, “An Overview of Selected Prognostic Technologies Exchanging Maintenance Action Information via the eXtensible
with Reference to an Integrated PHM Architecture,” First International Markup Language (XML), D3.0, Piscataway, New Jersey: IEEE
Forum on Integrated System Health Engineering and Management in Standards Association Press, April 2008.
Aerospace, NASA, Napa Valley, CA, November 7–10, 2005. [24] ISO 10303-11:1994, Industrial Automation Systems and Integration—
[4] Vachtsevanos, George, Frank Lewis, Michael Roemer, Andrew Hess, Product Data Representation and Exchange—Part 11: Description
and Biqing Qu, Intelligent Fault Diagnosis and Prognosis for Methods: The EXPRESS Language Reference Manual, Geneva,
Engineering Systems, John Wiley & Sons, Inc., 2006. Switzerland: International Organization for Standardization, 1994.
[5] Jata, K. V. and T. A. Parthasarathy, “Physics of Failure,” First [25] ISO 10303-21:1994, Industrial Automation Systems and Integration—
International Forum on Integrated System Health Engineering and Product Data Representation and Exchange—Part 21: Implementation
Management in Aerospace, NASA, Napa Valley, CA, November 7–10, Methods: Clear Text Encoding of the Exchange Structure, Geneva,
2006. Switzerland: International Organization for Standardization, 1994.
[6] Kalgren, Patrick, Mark Baybutt, Antonio Ginart, Chris Minnella, [26] ISO 10303-28:2007, Industrial Automation Systems and Integration—
Michael Roemer, and Thomas Dabney, “Application of Prognostic Product Data Representation and Exchange—Part 28: XML
Health Management in Digital Electronic Systems,” Proceedings of the Representation of EXPRESS Schemas and Data using XML Schemas,
2007 IEEE Aerospace Conference, Big Sky, MT, March 2007. Geneva, Switzerland: International Organization for Standardization,
[7] Scott, Andrew, “Electro-Mechanical Diagnostics/Prognostics,” IEEE 2007.
AUTOTESTCON 2007 Conference Record, Baltimore, MD, September [27] W3C, Web Services Description Language (WSDL), v1.1, W3C Note,
2007, pp. 340–348. March 15, 2001, http://www.w3.org/TR/wsdl.
[8] Klion, J., Practical Electronic Reliability Engineering, New York: Van [28] Sheppard, John W. and Jonas Åstrand, “Modeling Diagnostic
Nostrand Reinhold, 1992. Constraints with AI-ESTATE, IEEE AUTOTESTCON 1995
[9] Groer, P. G., “Analysis of Time-to-Failure with a Weibull Model,” Conference Record, Atlanta, GA, September 1995, pp. 528–533.
Proceedings of the Maintenance and Reliability Conference, MARCON [29] Kaufman, Mark A. and John W. Sheppard, “Bayesian Modeling: An
2000, Knoxville, TN, Amendment to the AI-ESTATE Standard,” IEEE AUTOTETSCON
[10] Carey, M. B. and R. H. Koenig, “Reliability Assessment Based on 2005 Conference Record, Orlando, FL, September 2005, pp. 424–430.
Accelerated Degradation: A Case Study,” IEEE Transactions on [30] Ralph, John, “A Proposed Comprehensive Architecture Utilizing the
Reliability, Vol. 40, No. 5, December 1991, pp. 499–506. Automatic Test Markup Language (IEEE 1671 and 1636.1),”
[11] Luo, M., D. Wang, M. Pham, C. B. Low, J. B. Zhang, D. H. Zhang, and AUTOTESTCON 2007 Conference Record, Baltimore, MD, September
Y. Z. Zhao, “Model-Based Fault Diagnosis/Prognosis for Wheeled 2007, pp. 190–196.
Mobile Robots: A Review,” Proceedings of the 31st Annual Conference [31] Wilmering, Timothy and John Sheppard, “Ontologies for Data Mining
of IEEE Industrial Electronics Society, November 2005, pp. 2267– and Knowledge Discovery to Support Diagnostic Maturation,”
2272. Proceedings of the 18th International Workshop on Principles of
[12] Khiripet, Noppadon, An Architecture for Intelligent Time Series Diagnosis (DX-07), Nashville, TN, pp. 210–217.
Prediction with Causal Information, PhD Thesis, Department of [32] ISO 10303-239:2005, Industrial Automation Systems and Integration—
Electrical Engineering, Georgia Institute of Technology, 2001. Product Data Representation and Exchange—Part 239: Application
[13] Sheppard, J. W., S. G. W. Butcher, and R. Ramendra, Electronic Protocol: Product Life Cycle Support, Geneva, Switzerland:
Systems Bayesian Stochastic Prognosis: Algorithm Development, International Organization for Standardization.
Technical Report JHU-NISL-07-002, September 2007. [33] Organization for the Advancement of Structured Information Standards
[14] Bengio, Y. and P. Frasconi, “Diffusion of Context and Credit (OASIS), Product Life Cycle Support (PLCS) Aviation Maintenance
Information in Markovian Models,” Journal of Artificial Intelligence Data Exchange Specification (DEX), rev 1.145, http://docs.oasis-
Research, 3:249–270, 1995. open.org/plcs/dexlib/R1/dexlib/data/dex/aviation_maintenance/home.ht
[15] Bengio, Y. and P. Frasconi, “Input-Output HMMs for Sequence m, March 10, 2008.
Processing,” IEEE Transactions on Neural Networks, Vol. 7, No. 5, [34] ISO 13374-1:2003, Condition Monitoring and Diagnostics of
September 1996, pp. 1231–1249. Machines—Data Processing, Communications and Presentation—Part
[16] Machinery Information Management Open Standards Alliance 1: General Guidelines, Geneva, Switzerland: International
(MIMOSA), Open Systems Architecture for Condition Based Organization for Standardization.
Maintenance (OSA-CBM) Primer, August, 2006. [35] IEEE P1232, Draft IEEE Standard for Artificial Intelligence Exchange
[17] Machinery Information Management Open Standards Alliance and Service Tie to All Test Environments (AI-ESTATE), D1.0,
(MIMOSA), Open Systems Architecture for Condition Based Piscataway, New Jersey: IEEE Standards Association Press, 2008.
Maintenance (OSA-CBM) UML Specification, v3.1, http://docs.oasis-
open.org/plcs/dexlib/R1/dexlib/data/dex/aviation_maintenance/home.ht
m, December 30, 2006.

Authorized licensed use limited to: Central Queensland University. Downloaded on August 10,2022 at 02:41:03 UTC from IEEE Xplore. Restrictions apply.