Журнал - The Engine Yearbook 2005

Подождите немного. Документ загружается.

ENGINE YEARBOOK 2005

The Leader in RVI

Remote Visual Inspection

TO SOME PEOPLE,

New

Small-space video borescope

AT EVEREST VIT, IMAGE IS THE ONLY THING

IMAGE

IS EVERYTHING

ADVANCED PORTABLE MODULAR

To schedule a product demonstration, please call

North and South America 1-888-332-3848 or +1 973-448-0077

France +33 (0) 2 282 23 0800

Germany, Eastern Europe, Middle East and Africa +49 (0) 7471 98820

Hong Kong and Asia +852 (0) 2877 0806

Italy +39 0309 651341

Japan +81 (0) 3 5607 5670

UK and Ireland +44 (0) 1993 822613

or log on to www.everestvit.com

Imaging and XL PRO are trademarks of Everest VIT, Inc. All other product names are trademarks of their respective owners.

■

CompactFlash

removable storage media

■

Over 2 hours of MPEG2 full-motion

audio/video recording and playback

■

Playback video on any Windows

■

USB streaming digital video output

■

Integrated temperature warning system

■

Five interchangeable probe diameters from 3.9

mm (0.153 in.) to 8.4 mm (0.331in.)

■

Integrated viewing, recording,

measuring, comparing and playback

■

On-screen time/date, temperature warning,

articulation position and character generation

■ Life extension, based on individual

engine/component condition and

usage profile.

■ Exchange of ‘life-expired’ engines to

other uses to absorb residual creep

or fatigue lives;

■ Reduced need for spares holding;

■ Availability management to limit the

need for unplanned maintenance.

Improved ‘departure’ statistics.

Reduced in-flight shutdown rates

and maintenance away from base.

Enhanced airline reputation;

However, as a senior aero-engine gas

turbine industrialist once famously

stated, “the aero engine business is

like the razor business: you can give

away the engines because the money

is made on the blades!” The total 20-

year value of the aero-engine market

is approximately $750 billion, of

which 45 per cent is judged to be in

the aftermarket. This is large by any

measure. Further, the aftermarket’s

importance to the engine

manufacturers is critical and twofold.

Firstly, the margins are higher than

in original equipment sales. Next,

whilst new equipment sales can be

deferred during economic

downturns, the aftermarket may

sustain the business until the next

upturn.

Aero-engine and gas path

diagnostics

Engine condition monitoring and

engine diagnosis have been

recognised, for some time, as

important assets in making more

informed decisions on the usage,

maintenance, overhaul or replacement

of the engine or one of its

components. The importance of such

techniques has been re-emphasised by

the changes in market positioning

discussed earlier. Additionally,

improvements in instrumentation

quality, information technology and

web-based systems have resulted in

large quantities of data being

routinely gathered from the operation

of fleets of engines. Industry has yet

to obtain the full extent of the added

value that advances in gas turbine

diagnostic systems offer, particularly

when, in these changed circumstances,

they are coupled with business

objectives. Gas path diagnostics is an

important element of such future

ambitions. That gas turbines routinely

deliver high availability and long life

is now broadly accepted. The question

that remains is whether the

availability and life is achieved by

using relatively large “safety

margins”, as these imply additional

maintenance, shorter component lives

and hence higher costs. Among the

quantifiable benefits from the use of

appropriate gas path diagnostics are:

■ Definition of work packages based

on actual diagnosed condition,

instead of the ‘average’ engine;

■ Clarity in defining cost-effective

aftermarket agreement objectives.

Scope and resource management;

■ Performance management to include

‘thrust rating’ and adaptive control;

■ Instrumentation selection against

usage objectives.

The next sections offer an

introduction to the underlying theory

of gas path diagnostics and a brief

EYB2005_2 9/9/04 3:12 pm Page 49

highlight of some of its latest

developments.

Overview of gas path diagnostics

In a fundamental sense,

performance monitoring and fault

diagnostics involves the processing of

engine measurements. In all cases,

some performance parameters of the

investigated engine are compared to

the corresponding values of an

engine considered to be ‘healthy’.

The parameters used and the way of

deriving them characterise each

different diagnostic method. Broadly

speaking, all these techniques rely on

what is known as gas path analysis

(GPA). A practically useful technique

should be able to take into account

the measurement noise and a possible

sensor bias, while at the same time

preserving the non-linearity of the

system.

The solution to the problem requires

the search for a best match between

some simulated performance parameters

(such as temperatures, pressures and

speeds) and the corresponding values

from the deteriorated engine. This is

done by defining a suitable error or

objective function. The objective

function should correctly represent the

problem and should be easy to compute.

Gas path diagnostics techniques may

differ in:

■ The definition of the error function

or objective function that define the

above-mentioned best match.

■ The number of operating points at

which the measurements are taken.

■ The number of measurements (level

of instrumentation).

■ The quality of the measurements

(quality of instrumentation).

■ The algorithm for searching through

the vast search space effectively and

in a reasonably short time.

Cranfield University has substantially

contributed to the latest developments

of gas path diagnostics, particularly

through the application of novel

numerical tools, such as neural

networks, fuzzy logic and genetic

algorithms. There follows a description

of some of the research being

undertaken in this direction.

Neural networks

Artificial neural networks (ANN),

one of the artificial intelligence

techniques, were introduced into gas

turbine diagnostics in the late 1980s.

An ANN is a massively parallel,

distributed processor with simple

processing units. It simulates the

functional relationship between

dependent and independent variables

by storing experimental knowledge

in the network (training phase) and

making it available for use

(application phase). An ANN is

especially useful when there is no

model at all to describe the physical

phenomenon under analysis, or when

the model itself is either too poor or

too complex to be used. In gas

turbine diagnostic applications, the

inputs to the network are the

deviations of the gas path

performance parameters such as

pressures and temperatures, while

the outputs are the shifts of some gas

turbine component characteristics,

such as changes in flow capacity and

efficiency. The functional

relationship is stored in the weights

(or synapses), which are obtained by

training the ANN with training

samples.

The features which make ANN

amenable for engine diagnostic tasks are:

● ANN can cope with the large

amount of noise affecting gas

turbine measurements, even though

some parameters have to be chosen

at the design stage and at the

beginning of training;

● ANN do not require the setting of

critical parameters, such as the ones

required in Kalman filter-based

techniques to fix the standard

deviation of each performance

parameter;

● ANN could be trained online to

monitor the engine health in real-

time;

● ANN is capable of dealing with the

large non-linearity that characterises

the correlation between

measurements and performance

parameters in a gas turbine;

● ANN could be used to perform

diagnostics using different data

sources — vibration, aero-

thermodynamic results and gas-path

debris data represent, amongst

others, comprehensive inputs to an

ANN-based system.

ENGINE YEARBOOK 2005

EYB2005_2 7/9/04 11:03 am Page 50

ENGINE YEARBOOK 2005

For complicated gas turbine

diagnostic problems, a single neural

network may not be enough to get

robust and accurate results. The

diagnostic task can be better done if it

is divided and shared with a nested

neural network approach [1]. An

example of such a developed technique

and system is shown in figure 3.

Fuzzy logic

A fuzzy logic system is a non-linear

mapping of an input feature vector into

a scalar output [2, 3]. The flexibility of

fuzzy logic systems in handling

uncertainties has played a key role in

their wide usage for various

engineering applications.

A typical Multi-Input Single-Output

fuzzy logic system [3] performs a

mapping from to using four basic

components: rules, fuzzifier, inference

engine and defuzzifier.

f : V

W

Where V=V

x...xVn

is the input

space and is the output space.

The rules are expressed as IF-THEN

statements (for instance: “if the

exhaust gas temperature and the

compressor delivery pressure are

high then the fault could be in the

compressor module”). Such rules are

either obtained from experts in the

field or from numerical data obtained

by performance simulation in the

case of gas turbines. Once the rules

driving the fuzzy logic system have

been decided, the system can be

expressed as a mapping of input data

to output data. The fuzzifier maps

crisp input members into fuzzy sets.

This is needed to activate rules that

are expressed in terms of linguistic

variables. The role of an inference

engine is to determine the way in

which the fuzzy sets are combined.

The defuzzifier has the opposite role

of the fuzzifier and converts the

fuzzy values to crisp ones.

The brain of the fuzzy logic system

is in the rules and these must be

carefully formulated. Like ANN,

fuzzy logic requires a substantial

amount of information in the form of

training data. This data can be

obtained either from actual engine

runs or from the utilisation of

suitable performance models. While

the first option is time- and resource-

consuming, the latter requires an

accurate model. More recent research

[4] shows further promise of this

technique.

Genetic algorithms (GA)

GA have recently emerged as a

powerful optimisation tool, finding a

wide range of applications in different

fields. From a diagnostic perspective,

GA are particularly suitable for

identifying the minimum of the

objective function.

The metaphor underlying the

genetic algorithm is that of natural

evolution. In evolution, the problem

that each species faces is that of

searching for beneficial adaptations

to a complicated and changing

environment. GA follow the natural

principle of survival of the fittest.

The GA nomenclature is also

borrowed from the vocabulary of

natural genetics [7]. In the context of

this technique, a string refers to a

possible solution and a collection of

possible solutions or strings is called

a population. The fitness of the

string is a function of the objective

function and is inversely

proportional to it. The best string

would therefore have the highest

fitness, which means that the value

of objective function would be

minimised.

A diagnostics algorithm based on a

GA typically starts with a population

that is created at random;

subsequently the objective function

is calculated for each of the strings in

the population. The objective

function is then mapped to a fitness

function and the larger the fitness,

the higher the probability of

survival. This mapping can be linear

or non-linear. The GA then works

over a number of iterations or

generations, each containing three

fundamental operators: selection,

crossover and mutation. The selection

operator chooses the strings to be

used in the next generation

according to a “survival of the

fittest” criterion. The crossover

operator allows information exchange

between strings, in an attempt to

generate fitter strings. Crossover is

carried out by swapping parts of two

parameter vectors. Mutation is used

to introduce new or prematurally-lost

information in the form of random

perturbations to the values of a

parameter vactor, without exceeding

the fixed upper and lower thresholds.

Figure 4 shows a schematic diagram

of a typical generation.

The diagnostic techniques using GA

have been tested on both commercial

and military engines and have been

applied on simple cycle engines as well

as on advanced ones, such as the

intercooled and recuperated turboshaft

[5, 6]. The results have shown a high

level of accuracy even in presence of

measurement noise and sensor biases.

Furthermore, such diagnostic systems

are flexible: in the case that sensible

guesses on the maximum number of

faulty sensors are available, the

optimiser can be tailored accordingly.

Conclusions and future of gas path

diagnostics

The changes in the role of the

aftermarket, which are re-defining

the business paradigm for the civil

EYB2005_2 7/9/04 11:11 am Page 51

air transport business, also have a

resonance in the defence sector.

Concepts such as ‘by-the-hour’,

already introduced within the airline

industry, may become a reality for

marine propulsion as navies move

towards lean manning of ships and

reduced manpower in ship yards. Gas

turbines also play a major role in the

energy business and here too changes

are becoming apparent in the

aftermarket, where major users are

setting up as competitors to the

engine manufacturers, changing

industry relationships and

economics.

Some of the diagnostic methods

presented earlier are in regular use

by the industry in managing engine

maintenance and repair, often in

conjunction with other diagnostics

techniques such as vibration and oil

analysis. The combination of such

information with detailed

understanding of the design,

operation usage profile and logistics,

provide competitive advantage in the

aftermarket. The next stage for gas

path diagnostics will see the

emergence of powerful hybrid

techniques, combining the most

appropriate features of several gas

path diagnostic techniques and

coupling these to other methods. The

availability of large quantities of

operational data from individual

engines and fleets will provide

statistical databases which, when

taken together with advanced

diagnostic methods, will allow

important advances to prognostics,

further increasing the market value

of these technologies.

The improvements in the in-service

operations of engines have had a

fundamental impact on the industry

[8]. Firstly, engines are getting more

reliable. This is measured by ‘in-

flight shutdown rates’. Another issue

of more long-term consequence is the

expected trend in engine ‘life on-

wing’. With the steadily improving

life of aero engines, it could be that

an engine will not require a major

service for the duration of the

aircraft’s 25 year life. This implies

that, in 50 years time, the oldest

engine in use will not have entered

service until 25 years from now. The

result could be the partial or

complete loss of the engine

manufacturers’ aftermarket business.

Companies would have to make

compensating higher profits on the

original equipment sale. More

interestingly, the loss of the

aftermarket revenues based on

decades of Prime incumbency

eliminates a major market entry

barrier. Perhaps this will allow a new

wave of companies to gain entry to

the business. Another scenario is that

ENGINE YEARBOOK 2005

EYB2005_2 7/9/04 11:13 am Page 52

ENGINE YEARBOOK 2005

technology advances will drive more

appropriate business solutions. The

optimisation of engine management

systems may include internal flows

(sealing and leakage), adaptive

systems and cycles, embedded micro-

and nano-sensors [9] and actuators

coupling performance, cycle, and

environmental impact management.

Future engines may have to be

optimised for global warming [10].

Any future solution will have to offer

a high level of diagnostic capability,

integrated to life-cycle management,

including perhaps both economics

and global warming.

In the past, the civil air transport

business has delivered strong long-

term growth by reducing unit cost

and hence allowing more people to

participate in air travel. This

reduction in unit cost has been

achieved both because of market pull

and technology push. A further

contributory factor has been the

technology advances made possible

because of the large number of

technologists this industry employs.

As the business emphasis shifts to

the aftermarket, the changing

paradigm will favour those business

leaders who recognise that this new

market cannot be dominated by focus

on logistics and management, or even

advanced diagnostics. Advantage will

flow to those who recognise the

importance of the contribution that

“intellectual adventure” can make in

the aftermarket as it has, in the past,

in other areas of gas turbine

technology. ■

Reference

1. Ogaji, S.O.T. and Singh, R., 2003:

Gas path fault diagnosis framework for a

three-shaft gas turbine. Proceedings of

the Institution of Mechanical Engineers,

Vol. 217, Part A, pp. 149-157.

2. Ganguli, R., 2001: Application of

fuzzy logic for fault isolation of jet

engines. ASME Turbo Expo 2001, New

Orleans, USA. 4-7 June 2001. ASME

2001-GT-0013.

3. Ganguli, R., 2001: Data rectification

and detection of trend shifts in jet engine

gas path measurements using median

filters and fuzzy logic. ASME Turbo

Expo 2001, New Orleans, USA. 4-7 June

2001. ASME 2001-GT-0014.

4. Marinai, L.; Singh, R. and

Curnock, B., 2003: Fuzzy-logic-based

diagnostic process for turbofan engines.

16th ISABE conference, Cleveland,

Ohio, USA. 31 August - 5 September

2003.

5. Zedda, M. and Singh R., 1999: Gas

turbine engine and sensor diagnostics.

14th ISABE Symposium, Florence, Italy.

6. Sampath, S. Gulati, A. and Singh,

R., 2002: Fault diagnostics using genetic

algorithm for advanced cycle gas turbine.

ASME Turbo Expo 2002, Amsterdam,

The Netherlands.

7. Michalewicz, Z., 1996: Genetic

algorithms + data structures = evolution

programs. Springer Verlag, 3rd Edition.

8. Singh, R., 2001: Civil aero gas

turbines: strategy and technology,

Chairman’s Address, Aerospace

Division, Institution of Mechanical

Engineers, London. April 2001.

9. Campbell, D., 2002: Propulsion for

the 21st Century — NASA Glenn

Research Centre Perspective. Royal

Aeronautical Fedden Lecture, Cranfield

University, UK. November 2002.

10. Whellens, M.W. and Singh, R.,

2002: Propulsion system optimisation for

minimum global warming potential. 23rd

ICAS Congress, Toronto, Canada.

September 2002.

Cranfield University has

substantially contributed to the

latest developments of gas path

diagnostics,particularly through

the application of novel

numerical tools, such as neural

networks,fuzzy logic and genetic

algorithms.

EYB2005_2 7/9/04 11:13 am Page 53

ENGINE YEARBOOK 2005

Are your engines really as

healthy as they seem?

It is a simple fact that most

engine health monitoring

(EHM) systems used by

airlines are unsuitable for the

job. Why? Because they fail to

solve the automation

dilemma - on the one hand

businesses want cost

reductions, but on the other,

high quality health

monitoring demands some

degree of human

judgement. In this article,

Data Systems & Solutions

(DS&S) discusses how it

solved this dilemma with its

EHM solution, which is now

used to monitor more than

4,000 engines.

urrently, EHM quality is

normally judged on the fairly

rudimentary basis of ‘alerts’ -

the number of genuine alerts missed

or misdiagnosed, together with the

number of spurious ones delivered.

With its solution, DS&S is already at

a level of maturity where it is

extremely rare for an alert to be

missed, or for instances of a

particular fault to be incorrectly

diagnosed. Furthermore, spurious

alerts - those resulting from normal

scatter in the data - are virtually

eliminated. This being the case, if

EHM alerting and diagnosis is to

advance still further, a paradigm

shift in the diagnostic system

installed on the engine will be

needed. Such a shift will take place

when the ‘QUICK’ analysis

technology being developed by

Rolls-Royce enters service on all

A380 aircraft in less than two years

time.

In the meantime, in order to

enhance the benefits of EHM on

today’s engines, DS&S is applying

significant research and development

resources to create new computing

techniques designed to solve difficult

pattern recognition, diagnostic and

forecasting problems. It is also

working with maintenance

management systems providers to

develop advanced asset management

capabilities that will capitalise on its

existing diagnostic and forecasting

technology. In fact, DS&S has

recently completed major

programmes of research into

improved engine modelling and

diagnostic techniques, which are

already embodied in its existing EHM

service system.

Adding the X-factor — eXperience

The key characteristic of DS&S’

EHM service system is that it is

capable of continuous, fully automatic

operation, whilst allowing manual

intervention in the case of a specific

alert and when system training is

required. It has been operational since

early 2003, and already processes an

estimated 10,000 health monitoring

‘snapshot’ reports per day.

The main features of the system are:

● high integrity;

● near real-time operation;

● full end-to-end automation;

● the system is ‘data-driven’ — the

system starts in response to data

arriving;

● automatic adherence to the specific

business rules of each organisation;

● flexible data input;

● absolute segregation of data for each

customer;

● complete audit trail of the data-

processing cycle;

● it is easily scaleable, without

software change;

● central control; and

● access using commonly-available

web-browsing tools.

The EHM system’s modular design

allows functionality to be hosted on

one server or to be distributed across

a number of servers, so that it can be

“grown” smoothly to manage any

number of engines simply by adding

more modules onto existing or

additional servers. The system’s

inherent flexibility also means that it

EYB2005_2 7/9/04 11:14 am Page 54

framecheck71 12/8/04 11:34 am Page 3

ENGINE YEARBOOK 2005

determine the priority in which data

will be processed, the details of the

processing to be performed, the

destinations of the various forms of

output and so on.

The progress of each file is tracked

at every stage and if processing is

interrupted for any reason it can be

restarted with no loss of data. System

performance can also be monitored

from the individual file level to that

of the whole system, whereby

statistics such as average end-to-end

processing time can be obtained.

The console module provides

system supervision functions, so the

status of the system can be

monitored, individual files can be

tracked and information on particular

airlines or fleets can be reviewed

(such as time of last data receipt,

when a particular aircraft last sent

data, and so on) or it can be used to

review processing errors.

While the EHM system is designed

to operate with minimal supervision,

it must be kept running at all times

to prevent backlogs, and data should

not be discarded indiscriminately. To

overcome any possible conflict in

can be used on aircraft and aircraft

system applications (such as APUs).

It can also be used in non-aerospace

environments such as utilities,

railways and marine applications.

Taking control of your business

At the heart of DS&S’ EHM system

is the administrative database, which

holds the business rules specific to

each organisation. These rules

these directives the system is

designed to actively alert the system

supervisor of impending problems

such as ‘bad data’ or the failure of a

system module to start, via an

automatic e-mail alert. The

supervisor can then log in, remotely

if necessary, to perform the required

remedial action.

Turning data into decisions...

One of the most significant

challenges for any health monitoring

system is the fact that data files can

be sent by airlines in a number of

formats and by different means of

transmission. For example, most files

arrive at the gateway to the system

as e-mail or an FTP transmission, in

ASCII format or a compressed form.

So, DS&S’ EHM solution includes e-

mail and FTP file-handlers, which

will accept the incoming data

according to the method of

transmission. Upon receipt of the

incoming communication, the file-

handler automatically determines

how the data is being carried

(embedded, attached, compressed,

whatever) and converts it into a data

file. It also records the identity of

the data sender.

The files are subsequently passed to

the data-handler module where their

contents are read. The data-handler

can perform operations on the data

such as patching known errors (a

report may have no year or date

specified), according to the business

rules in place for each airline and

fleet. It may also link or split data

files, as required, to optimise

processing.

At this point, the system

understands what needs to be done

with the incoming data, and the ‘job’

is passed on to the scheduler module,

where it is placed in a queue. The

scheduler assesses the resources

available and queues the jobs to

ensure that customer service-level

agreements are complied with. If a

large quantity of data arrives in a

short time, the scheduler determines

the best strategy to minimise service-

level deviations and it can look

ahead to predict how the queue is

likely to progress. This queue

Rolls-Royce takes EHM technology into the 21st

century

The Rolls-Royce QUICK system, developed in association with Oxford University,

is an advanced engine vibration analysis system based on that used in Rolls-

Royce’s engine test facilities. Rolls-Royce is developing an on-engine version of

the system for the Trent 900 engine for the Airbus A380 aircraft, which will be

available for retrofit to other Trent engines.

The engine mounted QUICK unit continuously monitors engine rpm, vibration

and other signals to give early warning of problems and accurate diagnostic

information. QUICK’s advanced signal processing and neural network

techniques extract the maximum amount of information from vibration and

other sensors fitted to an engine to identify anomalous behaviour and generate

fault diagnoses.These diagnoses are compared with a database of known

conditions and then transmitted to the ground for analysis and action. The

initial database is generated using test-bed data and is updated with

knowledge acquired in service.

DS&S is working with Rolls-Royce to seamlessly integrate the additional

information available from QUICK into its EHM system in time for the entry into

service of the Trent 900 in early 2006.

DS&S’approach to EHM is based

on the use of a number of

techniques to trend equipment

parameters, but to ensure the

most reliable results,it will always

default to using the one that is

highest in the accepted hierarchy

of trending techniques.

EYB2005_2 7/9/04 11:15 am Page 56

ENGINE YEARBOOK 2005

prediction can be viewed via the

console.

... and decisions into actions

Once the queue of jobs has been

defined by the EHM system’s

scheduler, the dispatcher module

executes them using the system’s

trending and computational

intelligence (CI) tools to identify

likely causes and solutions.

Specifically, the trending tool, which

is a version of DS&S’ engine

monitoring system COMPASS

Navigator™, normalises the observed

data for any variation in operating

conditions and compares it to an

engine performance model, thereby

providing optimum trending

information on engine condition.

In parallel, the CI tool performs

three main functions: it cleans any

systematic errors from the trended

output; identifies any anomalies in

the data; and ascribes a diagnosis to

anomalies that are found. As such,

the development of the CI tool has

proved to be an essential enabler to

the cost effective delivery of a 24/7

engine health monitoring service

since previously all trend inspections

had to be performed manually. In

addition, the CI tool has the

capability to manage multi-parameter

alerting, which is a significant step

forward from traditional single-

parameter alerting used in most EHM

systems. Refining the detection

techniques employed and combining

the information from a number of

related parameters vastly increases

the probability of correctly

identifying impending problems. For

example, a deviation in a single

parameter is most likely to result

from a sensor problem, but

consistent deviations in a number of

related parameters would be

indicative of a genuine engine fault.

As a final step in the management

process, the EHM system’s ‘web

uploader’ module uploads newly

processed trend and alert data to the

web database for access by

authorised personnel through

CoreControl™, DS&S’ predictive

services web portal. In addition, if an

alert has been generated the

‘customer notify’ module will send a

notification by e-mail or SMS text

message. Typically, the entire health

monitoring process takes less than 10

minutes from receipt of incoming

data to updated trends being

available to the customer.

A question of technique

DS&S’ approach to EHM is based

on the use of a number of techniques

to trend equipment parameters, but

to ensure the most reliable results, it

will always default to using the one

that is highest in the accepted

hierarchy of trending techniques (see

figure 2).

Ideally, all monitored engine

parameters that vary in a prescribed

way relative to a set of independent

variables should be compared to a

background model. For example,

engine gas path parameters vary

depending on engine thrust setting,

altitude, airspeed, total inlet air

temperature and other independent

parameters. Gas-path parameters

should, therefore, be trended relative

to a model that embodies their

relationship with the independent

variables.

However, some monitored

parameters, such as broadband and

tracked order vibration signals, have

a very weak relationship or none

whatsoever. In such cases, a

background model is of minimal

benefit, and the parameter would be

trended ‘raw’, or monitored relative

to a constant reference value.

Where high fidelity analytical

models are available or can be created

from knowledge of the way that

equipment works they represent the

best quality trending solution. These

models have a known validity range

and provide a basis for further

analysis and understanding of the

problem when an anomaly has been

detected. However, if these models

need to be created specifically for

trending work, such a solution can

also be the most expensive.

Where no analytical model is

available, engineering-based

parametric models can be created

from observed performance data,

together with a general

understanding of engine

performance. These models exhibit

linear behaviour outside their

derivation domain and provide some

level of understanding of the anomaly

detected. DS&S has recently

completed a research and

development project to create a rapid

method of producing models of this

type for any engine.

Where there is no engineering

understanding of the operation of a

piece of equipment and the

manufacturer is therefore unable to

supply a model of it, the only

recourse is to derive a numerical

model. DS&S uses computational

intelligence techniques, such as

neural networks, to create this type

of model - usually for monitoring of

equipment other than gas turbines.

As a primary trending technique, the

models are generally good within

their training domain, but tend to

exhibit unpredictable behaviour

when a new set of operating

conditions is encountered and the

model is unable to suggest a reason

for the detected anomaly.

Why it pays to be a model worker

Rather than working at an

individual engine level, DS&S’ EHM

system is designed to use one

trending model to represent each

distinct ‘bill of material’ or ‘model’ of

an engine. Provided the quality of

the model is shown to be satisfactory,

EYB2005_2 7/9/04 11:16 am Page 57

ENGINE YEARBOOK 2005

firmly grounded on a view of how

the equipment is behaving relative to

the design intent.

In comparison, other modelling

methods tend to centre on a view of

how engine health has changed since

the beginning of monitoring, which

unfortunately makes the false

assumption that the engine was

exhibiting satisfactory or ‘normal’

operation during the initial model-

building period.

Now with added intelligence

DS&S identified the value of

computational intelligence techniques

at an early stage. As a result, most of

its current services use a high-fidelity

analytical or parametric model as the

primary trending technique, with a

computational intelligence model as a

secondary ‘line of defence’. This

means that maximum understanding

and linearity is extracted in the

primary trending and further

extraction of unaccounted-for

systematic relations is achieved in the

secondary trending. In this way, the

strengths of both approaches are

cumulative, maximising the

probability of finding and correctly

diagnosing a fault.

Within DS&S’ EHM system, the

computational intelligence tool uses a

secondary neural-net, multi-layer

‘perceptron’ model to remove any

remaining systematic errors in the

trend data following application of

the primary model. Use of this two-

phase approach leads to the

smoothest possible trends, so that

anomalies in the data are identified

this approach permits the same model

to be retained indefinitely, unless

there is a significant change in the

engine design, such as a major

performance-improvement

modification.

This approach yields a number of

advantages relative to individual

engine serial-number-based

modelling. Remodelling is rarely

required but even where it is, it will

normally be to implement

improvements to reduce trending

scatter, so the main datum level will

be retained. This ensures that long-

term changes are detected, proximity

to physical limits exposed and

engine-to-engine and fleet-wide

comparisons are sound.

Consequently, health monitoring is

with greater reliability. The neural-

net model is derived off-line from the

service, using data from a number of

engines.

Having smoothed the trends, the CI

tool uses an improved Kalman

smoothing technique to fit the trend

data and reject outliers. Based on the

resulting fit, the software detects

anomalous data according to

predefined criteria. The anomaly will

be defined in terms of deviations in a

number of parameters as well as their

rates of change and cumulative

change.

The anomalies detected are then

passed to the diagnosis section where

they are compared with the

knowledge base of known diagnoses.

The CI tool determines goodness of fit

with the known events and ranks

them in order of likelihood.

Diagnoses are then either

“remembered”, if they are below the

alerting threshold, or made available

to the ‘customer notify’ module of

the EHM system, to generate an alert.

Closure of alerts (and the health

monitoring process) is achieved by

feedback through the CoreControl™

web portal. Generally the airline will

determine when the alert is closed,

though in some cases this

responsibility lies with the engine

manufacturer, especially if the engine

is on a ‘power-by-the-hour’

maintenance contract. DS&S’ back-

office procedures and software ensure

that the maximum value is derived

from each alert in terms of ‘training’

the EHM system to detect the same

problem as soon as possible the next

time it arises and providing the best

possible diagnosis for each feature

detected.

EHM in action

Allowing an aircraft to keep

generating revenue is one of the most

valuable benefits of EHM, as

demonstrated by a recent in-service

event with a Rolls-Royce Trent 500-

powered A340.

Following a lightning strike, the

aircraft experienced surge messages

during the climb phase en-route

across the Pacific. The Rolls-Royce

operations room was notified of the

One of the most significant

challenges for any health

monitoring system is the fact that

data files can be sent by airlines

in a number of formats and by

different means of transmission.

EYB2005_2 7/9/04 11:16 am Page 58