System Maintenance

Location:

State College, PA

Posted:

February 14, 2013

Contact this candidate

Resume:

Prognostic Enhancements to Gas Turbine

Diagnostic Systems 1

Carl S. Byington,Matthew Watson Michael J. Roemer Thomas R. Galie, Jack J. McGroarty,

Impact Technologies, LLC Impact Technologies, LLC Christopher Savage

220 Regent Court 125 Tech Park Drive Naval Surface Warfare Center

State College, PA 16801 Rochester, NY 14623 Carderock Division

814-***-**** Philadelphia, PA 19112

****.********@******-***.***

Abstract The development of machinery health 1. INTRODUCTION

monitoring technologies has taken center stage within

Various prognostics and health monitoring

the DoD community in recent years. Existing health

technologies have been developed that aid in the

monitoring systems, such as the Integrated Condition

detection and classification of developing system

Assessment System (ICAS) for NAVSEA, enable the

faults. However, these technologies have traditionally

diagnosis of mission critical problems using fault

focused on fault detection and isolation within an

detection and diagnostic technologies. These

individual subsystem. Health management system

technologies, however, have not specifically focused

developers are just beginning to address the concepts

on the automated prediction of future condition

of prognostics and the integration of anomaly,

(prognostics) of a machine based on the current

diagnostic and prognostic technologies across

diagnostic state of the machinery and its available

subsystems and systems. Hence, the ability to detect

operating and failure history data. Current efforts are

and isolate impending faults or to predict the future

focused on developing a generic architecture for the

condition of a component or subsystem based on its

development of prognostic systems that will enable

current diagnostic state and available operating data is

plug and play capabilities within existing systems.

currently a high priority research topic.

The designs utilize Open System Architecture (OSA)

guidelines, such as OSA-CBM (Condition Based Model-Based Prognostics

Cost & Accuracy

Pr o

Maintenance), to provide these capabilities and (Failure Physics, Virtual

Increasing

Sensing, Functional)

Physical

enhance reusability of the software modules. One such

Models

implementation, which determines the optimal water

sti

wash interval to mitigate gas turbine compressor Evolutionary or

Classification

Trending Models

performance degradation due to salt deposit ingestion, Methods, Fuzzy

(Data Driven, Feature-

Logic, NN, State

is the focus of this paper. The module utilizes

roa

Based Correlation)

Estimation Models

advanced probabilistic modeling and analysis

technologies to forecast the future performance Experience-based

Generic, Statistical Life

characteristics of the compressor and yield the optimal Prognostics

Usage Algorithms (Failure PDFs, Few

Time To Wash (TTW) from a cost/benefit standpoint. sensors or model)

This paper describes the developed approach and

architecture for developing prognostics using the gas Range

Range of System Applicability

turbine module.

Figure 1 Hierarchy of Prognostic Approaches

TABLE OF CONTENTS

In general, health management technologies will

1. INTRODUCTION

2. US NAVY ICAS observe features associated with anomalous system

3. US NAVY GAS TURBINE CBM INITIATIVE behavior and relate these features to useful information

4. INCORPORATING PROGNOSTICS about the system s condition. In the case of

5. PROGNOSTICS CONSIDERATIONS prognostics, this information relates to the condition at

6. GAS TURBINE PERFORMANCE PROGNOSTICS

some future time. Inherently probabilistic or uncertain

7. CONCLUSIONS

in nature, prognostics can be applied to

0-7803-7651-X/03/$17.00 2003 IEEE

Form Approved

Report Documentation Page OMB No. 0704-0188

Public reporting burden for the collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and

maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information,

including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington

VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to a penalty for failing to comply with a collection of information if it

does not display a currently valid OMB control number.

1. REPORT DATE 3. DATES COVERED

2. REPORT TYPE

200*-**-**-**** to 00-00-2003

4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER

Prognostic Enhancements to Gas Turbine Diagnostic Systems 5b. GRANT NUMBER

5c. PROGRAM ELEMENT NUMBER

6. AUTHOR(S) 5d. PROJECT NUMBER

5e. TASK NUMBER

5f. WORK UNIT NUMBER

7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION

REPORT NUMBER

Impact Technologies LLC,2571 Park Center Blvd Ste 1,State

College,PA,16801

9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR S ACRONYM(S)

11. SPONSOR/MONITOR S REPORT

NUMBER(S)

12. DISTRIBUTION/AVAILABILITY STATEMENT

Approved for public release; distribution unlimited

13. SUPPLEMENTARY NOTES

The original document contains color images.

14. ABSTRACT

15. SUBJECT TERMS

16. SECURITY CLASSIFICATION OF: 17. LIMITATION OF 18. NUMBER 19a. NAME OF

ABSTRACT OF PAGES RESPONSIBLE PERSON

a. REPORT b. ABSTRACT c. THIS PAGE

unclassified unclassified unclassified

Standard Form 298 (Rev. 8-98)

Prescribed by ANSI Std Z39-18

system/component failure modes governed by material 3. US NAVY GAS TURBINE CBM INITIATIVE

condition or by functional loss. Like diagnostic

The Navy has formed an open forum working group

algorithms, prognostic algorithms can be generic in

teamed to establish Gas Turbine CBM, with the goal

design but specific in terms of application. Various

to plan and execute integration of CBM technologies

approaches to prognostics have been developed that

into gas turbines on all CG & DDG class ships.

range in fidelity from simple historical failure rate

Installation of FADC (Full Authority Digital engine

models to high-fidelity physics-based models. Figure

Controller) controllers on all gas turbines in the CG &

1 illustrates a hierarchy of prognostic approaches in

DDG classes by the Life Cycle Managers over the next

relation to their applicability and relative costs.

8 years will provide the hardware and computing

This paper will discuss some architectures and specific power required for equipment health assessment and

prognostic implementations for gas turbine engines. monitoring. ICAS will provide the necessary

The ability to predict the time to conditional or connection allowing gas turbine health monitoring

mechanical failure (on a real-time basis) is of systems to provide assessments and recommendations

enormous benefit and health management systems that to ships crew. New algorithms developed by the Navy,

can effectively implement the capabilities presented industry or the other programs will be incorporated as

herein offer a great opportunity in terms of reducing the part of the FADC.

overall Life Cycle Costs (LCC) of operating systems as

The planning phase was necessary to establish a plan,

well as decreasing the operations/maintenance logistics

organize a working group, establish funding

footprint.

requirements for the life of the program with OPNAV

2. US NAVY ICAS sponsors, and develop the complete transition plan.

Basic CBM development phase is designed to use the

The Navy s Integrated Condition Assessment System output of currently installed sensors to change some

(ICAS) [1] is a tool to enable maintenance time-directed maintenance items to condition-directed

troubleshooting and planning for shipboard machinery items. In the advanced phase, a limited number of new

systems. It provides data acquisition, data display, sensors will be used to develop more condition

equipment analysis, diagnostic recommendations, and directed tasks and start turning some corrective

decision support information to operators and maintenance items into condition directed tasks before

maintenance personnel. Additionally, ICAS links to the effected components fail. The last phase, system-

other maintenance-related software programs to wide development will incorporate on-going and new

provide a fully integrated maintenance system. ICAS R&D efforts into the development plan and complete

assesses equipment and system condition for system integration with ICAS. Most of the phases run

maintenance of machinery and equipment. Through concurrently and have parallel timelines.

the use of permanently installed sensors, the ICAS

system monitors vital machinery parameters on a 4. INCORPORATING PROGNOSTICS

continuous basis. ICAS can diagnose the operational

The approach for the PEDS program is to develop

condition of a particular piece of machinery using

prognostic software that is modular and possess the

customer-supplied performance data linked to a logical

capability for multiple transition opportunities. The

diagnostic process.

role of a PEDS module in an existing system is

The ICAS workstation is used for data acquisition, depicted using the diagram in Figure 2. The figure

conditioning, performance analysis, trend and logsheet illustrates the connections and communications

capture, and expert evaluation. Several types of data between existing elements and the system

acquisition devices that process sensor output signals enhancements. In the figure, proprietary interfaces or

augment it. The ICAS workstation is also responsible OSA-CBM middleware are used to glue and hook

for providing all user interface functions and long-term modules together. The figure shows the PEDS

data storage. Within this environment, the Prognostic module s ability to interface directly with the existing

Enhancements to Diagnostics Systems (PEDS) system, it s HSI, and the decision support and logistics

program is focused on demonstrating prognostic system using the proprietary interfaces defined by the

enhancements using demand data interface protocols existing system. This is accomplished using system

and displaying using pseudo sensor inputs or simple specifications, such as a Demand Data Interface (DDI)

web-based interfaces. as in the case of ICAS. In addition, the PEDS module

has the ability to interface directly with any system

that uses OSA-CBM specifications (i.e. OSA-CBM This may be the case in the matured situation, but for

Compliant Sensors and Processors, PEDS HSI) or the developing open system structure, the

systems that are enhanced to include the OSA-CBM implementations are usually more time-consuming and

specifications, which are represented by the dashed certainly not the path of least resistance. It means

lines. learning new techniques, structuring in different ways,

and in many cases carrying additional baggage. It is

initially a disruptive process. Consider a turn of the

Decision Support & Logistics

century bearing manufacturer having to buy new

System

materials, retool his machines, and change his

finishing process. Also consider the market for the

bearing manufacturer who didn t adapt to the

engineering pressure to standardize. In addition,

S OSA - CBM

MIDDLE-

specifying performance is necessarily just as important

WARE

PEDS

Module

as meeting a standard interface to accomplish true

M S

openness, modularity, and interchangeability.

Existing

The adaptation for software and information systems

Diagnostic

H System &

may be an even more challenging engineering

I PEDS

Database

HSI

N endeavor given the nature of the differences between

bits and atoms: what we can feel and see versus what

we cannot see and must interpret through use cases

Y Existing

and extrapolation. For a particular system integration

Diagnostic Pseudo

System Sensors

task, an open systems approach requires a set of public

HSI

component interface standards and may also require a

DDI

OSA - CBM

separate set of public specifications for the functional

Compliant

Sensors &

behavior of the components. The development of the

Processors

open-systems standards relevant to Condition-based

Figure 2 - PEDS and the Existing Diagnostic Maintenance (CBM) and Prognostics and Health

System Management (PHM) development has been pursued

by an International Standards Organization (ISO/TC

108/SC 5) committee, a consortium of condition

Evolving Open Systems Standards

monitoring companies (MIMOSA), and a DoD Dual-

Openness and open systems architecture is not a new

Use Science and Technology program (OSA-CBM)

concept in most parts of the engineering world. Being

lead by Boeing. These projects were a start down the

able to swap out a gear, bearing, shaft, chain, or even

disruptive path of openness for health management

an engine is made possible through past efforts to

systems.

standardize sizes and performance specifications.

Equivalency is reduced to meeting performance Because the specification deals with the I/O only, the

(strength reliability, etc.) and known functional actual layers or modules can be coded in a manner

interfaces (physical, electrical, etc.). We tend to take as to allow for proprietary approaches, thus protecting

this openness as a given, with little thought that it is the intellectual property of the developer. By applying

usually the case for many situations. Electrical the OSA-CBM specification, one can effectively

components (breakers, wire gauge, outlets, switches) communicate with a module being constructed by

followed a similar path and ultimately so is the another developer without every knowing how the

electronics industry. While it is probably the case that other module operates. Additional information about

these industries may have been slow to adopt these OSA-CBM can be found at http://www.osacbm.org

standards, at least with respect to today s Moore s Law and an OSA-CBM training manual can be found at

expected timescales, in many ways it may be an http://www.osacbm.org/Documents/Training/Train

engineering fait accompli. As the technology matures, ingMaterial/TrainingWebsite/index.html.

there is a desire to box its function, quantitize its form

factor, and structure its interaction with other system

components. Philosophically, it could be argued that

developing open standards minimizes the entropy gain

in the engineering process.

corrector schemes.

5. PROGNOSTICS CONSIDERATIONS

5. Model-Based or Physics of Failure Based

For a health management or CBM system to possess Prognostics: Fully developed functional and

prognostics implies the ability to predict a future physics-of-failure models to predict

condition. Inherently probabilistic or uncertain in degradation rates given loads and conditions.

nature, prognostics can be applied to system/

component failure modes governed by material 6. GAS TURBINE PERFORMANCE PROGNOSTICS

condition or by functional loss. Similar to diagnostic

Benefit of Technology

algorithms, prognostic algorithms can be generic in

design but specific in terms of application. A Fouling degradation of gas turbine engine compressors

prognostic model must have ability to predict or causes significant efficiency loss, which incurs

forecast the future condition of a component and/or operational costs through increased fuel usage or

system of components given the past and current reduced power output. Scheduling maintenance

information. Within the health management system actions based upon predicted condition minimizes

architecture, the Prognostic Module function is to unnecessary washes and saves maintenance dollars.

intelligently utilize diagnostic results, experienced- The effect of the various maintenance tasks (washing

based information and statistically estimated future and overhaul) on gas turbine engine efficiency is

conditions to determine the remaining useful life or shown in the figure below.

failure probability of a component or subsystem.

Prognostic reasoners can range from reliability-based Recoverable losses with Recoverable Losses with

on-line washing Crank Washing

to empirical feature-based to completely model-based.

Efficiency

Some of the information that may be required

depending on the type of prognostics approach used in

the system include:

Time

Engineering Model and Data Recoverable Losses with

Hot Section Overhaul

Degradation Rate with no

Failure History on-line Waterwashing

Degradation Rate with Degradation Rate with both

Past Operating Conditions on-line washing only on-line and crank washing

Current Conditions

Identified Fault Patterns Figure 3. Effects of Washing on Efficiency and

Transitional Failure Trajectories Overhaul

Maintenance History

Currently, washes are performed on a preventative

System Degradation Modes

schedule of 50 hours for on-line washes and 500 hours

Mechanical Failure Modes

for crank washes. This maintenance task is performed

Examples of prognostics approaches that have been with no engineering assessment of conditional need or

successfully applied for different types of problems optimal time to perform. In addition to the loss of

include: availability and maintenance time incurred,

unnecessary washes generate an environmental impact

1. Experience-Based Prognostics: Use

with the used detergent. Clearly operating with a

statistical reliability to predict probability of

module that assesses condition and predicts the time to

failure at any point in time. May be augmented

wash more appropriately would benefit the Navy.

by operational usage information.

2. Evolutionary/Statistical Trending Data and Symptoms for Development

Prognostics: Multi-variable analysis of

The compressor wash prognostic model was developed

system response and error patterns compared

using data from fouling tests taken at NSWCC in

to known fault patterns.

Philadelphia, PA and is an example of evolutionary

3. Artificial Intelligence Based Prognostics:

prognostics approach. It is based upon specific system

Mechanical failure prediction using reasoners

features and a simple model for compressor efficiency.

trained with failure data.

In order to simulate the amount of salt the typical

4. State Estimator Prognostics: System

Navy gas turbine is exposed to on a normal

degradation or diagnostic feature tracking

deployment, a 9% salt solution was injected into the

using Kalman filters and other predictor-

engine intake. Over the course of the entire test (3

days) approximately 0.0057m3 of salt was used to Compressor Performance Prognostics Module

induce compressor degradation at four different load The compressor performance prognostic module

levels (1/3, 2/3, standard and full load levels or consists of data preprocessing and specific prognostic

bells ). This method of testing was performed on algorithms for assessing current and future compressor

both Allison 501 and LM2500 Units. Figure 4 shows conditions. The data preprocessor algorithms examine

a borescope image of the salt deposits on the LM2500 the unit s operating data and automatically calculate

1st stage blading. key corrected performance parameters such as pressure

ratios and efficiencies at specific load levels in the

fashion already described. As fouling starts to occur

in service, probabilistic classifiers match up

corresponding parameter shifts to fouling severity

levels attained from these tests with corresponding

degrees of confidence. The techniques employed and

processing in the module are shown in detail in Figure

5. As can readily be seen, the consideration of

uncertainty is carried through the entire process to

produce a confidence in the prediction.

Figure 4 - Borescopic Image of Salt deposits: 1st

stage

9.5

Efficiency

Double

Sensor

8.5

Model

7.5

Exponential

Inputs

6.5

d Meas

Smoothing

5.5

Uncertainty

0 5 10 15 20 25

d Slope Buffer

+ (slope of

Analysis

Sensor degradation) (Sliding Window)

Uncer-

d Model

tainties

Clean

Confidence

Interval

SlopeUncertainty

Monte Carlo

t-Statistics Simulation SlopeMean

(TTW Model)

TTW Distribution

Slope Distribution

Confidence

TTWMean

Historical Combination NPV

NPV

Actual

TTW

Degradation Curve

% Parameter

TTW

TTWHistory Averaged Historical

Degradation Curve

Limit Line

Effective Operating

Hours Fused Prognosis

(weighted average)

DES Prognosis

Figure 5 - Processing Flow for Compressor Performance Prognostics

A probabilistic-based technique was developed that (1)

CDPT

utilizes the known information on how measured 1

CIPT

parameters degrade over time to assess the current

adb

severity of parameter distribution shifts and project

CDT

their future state. The parameter space is populated by CIT 1

two main components. These are the current condition

and the expected degradation path. Both are multi-

In the event that total pressure measurements are not

variate Probability Density Function (PDFs) or 3-D

available, other methods can be used to approximate

statistical distributions. Figure 6 shows a top view of

this efficiency calculation with other specific sets of

these distributions. The highest degree of overlap

sensors.

between the expected degradation path and the current

condition is the best estimate of compressor fouling. Once the statistical performance degradation path is

realized along with the capability to assess current

Shifting Fuel Flow Shifting PR

7 14

degradation severity, we needed to implement the

6 12

predictive capability. The actual unit-specific fouling

Frequency

5 10

rate is combined with historical fouling rates with a

4 8

double exponential smoothing method. This time

3 6

2 4 series technique weights the two most recent data

1 2

points over past observations. The following

0 0

0 2 4 6 8

0 2 4 6 8 10 12 equations give the general formulation. (Bowerman,

PR Fuel Flow 1993).

Shifting Means

ST= yT+(1- )ST-1 (2)

0.5

PR 4

S[2]T= ST+(1- )S[2]T-1 (3)

0 3

4 (4)

[2]

yT + (T ) = 2 +

1 S T 1 +

(1 ) 1 T

10 12

PR 0-468-*-*-*-*-*-**-**

0 2 Fuel Flow Fuel Flow Analysis of the degradation requires the simulation to

Figure 6 - Prognostic Model Visualization

predict the range of condition that might exist given

To manipulate the data into the form of this model, the the measurement and modeling uncertainties. This is

time dependency of the test results had to be removed accomplished using a Monte Carlo simulation with the

because of the unrealistic fouling rates. The percent mean and 2-sigma uncertainties. The resulting

changes in static pressure ratio, fuel flow, and CDT distribution is the range of Time-to-Wash predictions.

were recast in terms of % pseudo-efficiency drops. Appropriate statistical confidence intervals can be

This increment was chosen because it was the highest applied to identify the mean predicted value. This

resolution that still permitted statistical analysis. With estimate is updated with a weighted fusion from the

the assimilation of the data into these discrete bands, predicted value and the historical degradation level

the statistical parameters (e.g., mean and standard derived from the fouling data. The results of this

deviation) can be ascertained for use in the prognostic process are shown in Figure 7.

model. Figure 6 shows the evolution of the compressor

degradation for the LM-2500 test at 1% pseudo- Actual

% Efficiency Change

efficiency drops (for visual clarity). The top two plots Degradation Curve

illustrate the distributions of pressure ratio and fuel

Averaged Historical

flow respectively while the bottom two provides the Degradation Curve

joint probability distributions.

Allowable

Degradation Limit

The compressor inlet temperature (CIT), outlet Effective

temperature (CDT), inlet total pressure (CIPT) and Operating Hours Fused Prognosis

discharge total pressure (CDPT) can typically be used (weighted average)

DES Prognosis

to find compressor efficiency. (Boyce 1995) However

CDT, CDPT are not standard sensors in most Naval

Figure 7 - Prediction of Degradation Rates

platforms.

PEDS Implementation Stylesheet Language (XSL). Therefore the same

document can be displayed in multiple ways

The PEDS module implementation consists of

depending the consumer of the information. This

translating the engineering code (in Matlab in this

separation of information content from its presentation

case) into an implemented plug and play module.

is especially useful for user-centric interface designs.

The final compiling of the code is somewhat platform

specific, but for Windows based applications the code A major advantage of the PEDS architecture is it s

can be written in C++ and compiled as a dll (dynamic modularity and code re-usability. The figure below

linked library). The module currently supports OSA- shows the two possible deployment opportunities for

CBM compliant XML (eXtensible Markup Language) the Compressor Wash prognostic algorithm, ICAS and

and other documented data structures. Tiger . As shown, the Initialization element is the

only part of the code that is different between the two

XML is an extension of Standard Generalized Markup

implementations. Therefore the other elements of the

Language (SGML) and has been a World Wide Web

module are re-usable between the two approaches.

Consortium (W3C) recommendation since February

This is possible because the code has been written to

1998. XML is focused on describing information

allow for a number of different input possibilities.

content and information relationships. XML is similar

Flags are set in the initialization element that indicates

to HTML (commonly used in most web-based

which inputs to expect for the current implementation.

applications) except that, unlike HTML, XML does

Therefore, this modularity of design allowed easy

not have a predefined structure. The structure of the

modification of the compressor water-wash module to

XML document is defined by a user-generated

interface with different existing monitoring systems,

Document Type Definition (DTD) or schema. The

resulting in faster development time, and lower

display format of an XML document is also specified

incurred costs.

by the user/generator of the document using

eXtensible

Module Coding

Engineering Code Module Architecture

Prognostic

CW_GUI

Prognostic Director

Director

ResetCW

(C++)

MakeK17_ini

MakeLM_ini

Initialization- Initialization-

Initialization Initialization

DefaultHist

ICAS (C++) Tiger (C++)

Contact this candidate