Implementation of an Engine Condition Monitoring tool for

Airbus Aircraft

Sergio da Cruz Ribeirosergioribeiro@tecnico.ulisboa.pt

Instituto Superior Tecnico, Lisboa, Portugal

May 2015

Abstract

The growth and development of the aviation industry means an increase in the on-flight dataavailable to perform studies with the objective of enhancing airline operations. Data is collectedthrough systems aboard the aircraft, being posteriorly decoded and stored at the Flight Datadepartment. Statistical analysis, such as the one done by the application developed during the currentdissertation, can be performed using this data. The developed tool allows the analysis of engineperformance parameters through time and identifies situations that require intervention or may indicatemalfunction of an engine’s component. An algorithm that optimizes data processing in terms of timeand computational resources required was also developed, as well as an interactive application thatallows a dynamic analysis of the evolution of said parameters.Keywords: Flight Data, Engine Condition Monitoring, Performance Trends, Corrected Gas TurbineParameters, Stability Points.

1. Introduction

With the fast passed globalization and the growingneed to bring people and goods closer together,airline companies became one of the mostimportant players in the transport industry. Airtransportation evolved from a luxury good to anaccessible good and the number of passengers andflights has been increasing [8]. Even thoughmaintenance costs are usually not the mostsignificant costs of the airline activity [1], they are,usually, the ones most affected by managementdecisions which gives them a high degree ofcontrollability. One way of reducing said costs, isby constantly monitoring the performance of thefleet’s engines to assure they are running in themost safe and efficient way possible. This processis called Engine Condition Monitoring (ECM).ECM also allows for a better scheduling ofmaintenance stops and provides a precursor toactual engine failures that could lead to moreextensive repairs. Technological advancements inboth hardware aboard the aircraft and groundstations allow the use of more powerful statisticanalysis. The increase of recorded data at disposalon flight recorders, the development of bettersensors and the increased computationalcapabilities allow for a more comprehensive andaccurate prognostic of problems of systems aboardthe aircraft. Flight Data Monitoring (FDM)

software is now, more than ever, able to promote amore efficient use of resources and a saferoperation.

2. Condition-based Maintenance

The concept behind Condition-based maintenanceis having the right equipment in the rightcondition at the right time. It is based on theanalysis of real-time data to draw conclusionsabout the health of a given piece of equipment.This concept is based on the fact that mostfailures do not occur instantaneously, but ratherthrough a deterioration process that can bemonitored. This approach allows performingmaintenance only when the equipment is actuallyneeding it thus saving money otherwise spent onunnecessary maintenance stops. Despite itsusefulness, there are several challenges to take intoaccount. First of all, the investment costs ofimplementing the structure needed to perform thedata collection and posterior analysis areconsiderably high. Secondly, introducing thesechanges in the maintenance philosophy of thecompany, will demand a greater level of synergybetween the company’s departments which can bedifficult to accomplish due to the size of theorganization. Also, the technical side of theimplementation is not simple. Even if some typesof equipments can be easily observed by measuring

1

common values such as temperature or pressure, itis not trivial to turn the measured data into actualknowledge about the health condition of theequipment.

2.1. Trend MonitoringDuring normal operation all engines willexperience rubbing, thermal stress, mechanicalstress, dirt accumulation, foreign object ingestionand other events which will eventually result in ameasurable decrease in efficiency. By continuouslymonitoring the evolution of key parameters thatcan translate engine condition, deterioration inengine performance can be detected as well asearly signs of engine faults and thus appropriatemeasures can be undertaken. Parameter trendmonitoring is by definition the process by whichin-flight parameter measurements are processedand then compared to a baseline model thatrepresents the expected behaviour of a givenengine, taking into account ambient and thrustconditions, as if it was newly installed or at itsbest performance. The difference over timebetween measured data and the baseline model iscalled the parameter Delta. Monitoring theevolution of these deltas is what gives knowledgeabout the current state of an engine and allows anestimation of how its performance has deterioratedwith each flight cycle. A summary of the enginecondition monitoring parameters considered in thecurrent document is made in Table 1. Theseparameters are considered to be the performanceindicators, also known as engine PerformanceParameters.

Table 1: Summary of parameters used in ECMParameter Units

Exhaust gas temperature (EGT) ◦CFuel flow (FF) kg/hr

Low-pressure fan speed (N1) %High-pressure rotor speed (N2) %Engine Pressure Ratio (EPR) N/A

EGT Margin (EGTM) ◦C

2.1.1 EGT Margin

The EGT margin parameter deserves to behighlighted since it is one of the most importantcriteria used for engine removal due to excessiveperformance loss. EGT Take-off Margin is adurability and operability parameter. It representsthe difference between actual (uncorrected) EGTlevels, for a given take-off condition1, and the

1EPR, Mach Number, Altitude, Total Air Temperatureand bleed conditions

EGT limits, also known as redline, established bythe engine’s manufacturer. Gas path temperaturesabove design limits can dramatically reduce hotsection part life. The nefarious effects ofoverheating may include, corrosion of hot sectionmetals, creep and stress rupture resilience ofturbine blades and rotating components may bereduced by a factor of approximately two for afifty degree increase in temperature, andadditional inspection requirements, or prematureengine removal. Figure 1 shows the representationof a hot section and the result of excessive EGTvalues.

Figure 1: Pratt & Whitney engine. Effects ofexcessive EGT values [10]

As the engine deteriorates the EGT gets hotterand its margin decreases. With time, the EGTmargin may become negative, which means it is nolonger possible to reach Take-off thrust at the FlatRate Temperature (FRT). The FRT is thetemperature up to which the engine manufacturerguarantees the engine will produce the ratedthrust. Also note that below the FRT, also calledCorner Point Temperature (CPT), thrust can belimited by software for a more efficient take-off.

Figure 2: EGT Margin on different engineconditions. Adapted from [10].

EGT margin is one of the main indicators ofengine deterioration and normally airlinecompanies have a limit for the number of times it

2

can be exceeded. It also serves as an indicator ofengine performance degradation, allowing theairline to monitor its evolution. After the numberof exceedances is depleted or the EGT margin isconsecutively low the engine will be subject tomaintenance intervention. As the engine ages, dueto normal engine wear, the EGT margin will startto gradually decrease. Indeed, due to a loss ofefficiency, the engine will burn more fuel, whichwill lead to a higher EGT. The rate of degradationof these parameters is highly dependent on therate at which the aircraft accumulates flight hoursand cycles, the environments in which it operatesand the care with which it is operated. The firststep in generating EGT margin estimates is tocollect actual take-off data. The accuracy of thedata collection has a significant impact on theaccuracy of the EGT margin estimate. Toproperly estimate the EGT margin, all relevanttake-off data must be collected at the time EGTreaches a peak. The next step is to scan the datafor the maximum value of EGT that occursbetween the first time CAS2 exceeds 120 knotsand Radio Altitude (RALT) exceeds 1500 feet orclimb is detected. Once the maximum value forthe interval is found it is averaged with dataobtained during the previous three seconds. It isrecommended that during the data averagingperiod all isolation valves are closed and cowl andwing anti-ice valves and pneumatic system discretedo not change. Operators will sometimes try torecover some of the lost EGT Margin byperforming small scale interventions on the engine.These include engine washes which, as it ispossible to see in Figure 3, will recover some of theperformance lost through the accumulation offlight cycles.

Figure 3: Engine Wash and its effects on the EGTMargin.

The red lines represent the dates where the enginewash was performed. Notice how the EGT margin

2Calibrated airspeed is the speed shown by a conventionalairspeed indicator after correction for standard see levelcondition.

instantly raises after these events and how itgradually starts to decay over time until theengine is washed again.

2.1.2 Baseline model and trend monitoring

Parameter trend monitoring is the process bywhich the recorded engine data is corrected andthen compared to a baseline model of the engine.The difference, over time, between the measuredperformance and the expected performance basedon the engine’s baseline is the parameter Delta,allowing to assess an engine’s condition. In theearly days, baseline models would be provided tothe operators by the manufacturers and trendmonitoring would be conducted by manuallycalculating the parameter Deltas. Figure 4showcases an example of traditional cruisecharacteristic curves for a JT3C-6 turbojet engine.In this case the EGT, FF, N1 and N2 are plottedas a function of the EPR and for different MachNumber. Today’s ECM tools automatically recorddata from the engines and perform trend analysis,making this information available to the operators.

Figure 4: Cruise Performance Characteristics forthe JT3C-6 engine [4].

Baselines allow for an estimation of the expectedengine behaviour for different thrust conditions.The models presented in Figure 4 use the EPR asthrust reference, but this is not mandatory andother references can be used. Enginemanufacturers derive their Baseline models by

3

testing each engine before leaving factory in newcondition, and assuming that an engine is at itsbest performance when new, these Baselines are agood reference for measuring performancedegradation. The fact that manufacturer’sBaselines are no longer available to the operatorsmakes conducting studies like the one made in thecurrent document infeasible unless a Baseline isderived from in-flight data.

2.2. Parameter CorrectionsThe measured engine parameters values vary notonly with the power condition but also withambient conditions at the engine’s inlet [13]. Achange in the inlet temperature and/or pressure iscoupled with a change in the gas path parameter’svalues, which makes it difficult to establish acommon ground for the thermodynamicrelationships between gas turbine parametersunless ambient conditions are accounted for. Thisissue is solved by correcting the engineparameters. For convenience will a two shaftturbofan with station numbers, as shown in Figure5, will be assumed.

Figure 5: Twin spool mixed flow turbofan [13].

2.2.1 Formulation

For a given gas path parameter k, the equivalentcorrected parameter will be represented by k′.Generally speaking a change in the inletconditions, temperature T2 and pressure P2, willimply a change in any downstream gas pathparameters k. The purpose of the correctedparameter k′ is to give a constant approximationto a value regardless of changes in inlet conditions,in other words it represents the value that k wouldhave at some fixed inlet condition. The referenceconditions can be set freely, however, it is commonpractice to select standard day conditions3. It isfair to assume without loss of generality thatk = f(T2, P2, k

′) thus following equations 1 and 2.

dk =

(∂k

∂T2

)dT2 +

(∂k

∂P2

)dP2 +

(∂k

∂k′

)dk′ (1)

3T0 = 288.15 K, P0 = 1013.25 hPa

dk

k=

(∂kk∂T2

T2

)P2=const

k′=const

dT2T2

+

(∂kk∂P2

P2

)T2=const

k′=const

dP2

P2+

+

(∂kk∂k′

k′

)P2=constT2=const

dk′

k′(2)

Assuming that the first two partials are constant(the third clearly is constant and equal to unity)and substituting them for coefficient “a” and “b”,respectively, we can simplify 1 and 2 as follows:

dk′

k≈ adT2

T2+ b

dP2

P2+dk′

k′(3)

If we now define two dimensionless parameters θ =T2/T0 and δ = P2/P0 it is possible to deduce anexpression for k′.

dk′

k′≈ dk

k− adθ

θ− bdδ

δ⇒ k′ ≈ k

θaδb(4)

Equation 4 is the parameter correction expressionin use for the rest of this document. Table 2summarizes some of the common gas turbineparameter corrections and commonly used valuesfor coefficients a and b.

Parameter a b Corrected Parameter

Rotor Speed (N) 0.5 0 N ′ = N√θ

Fuel Flow (FF) 0.5 1 FF ′ = FF√θδ

Temperature (T) 1 0 T ′ = Tθ

Pressure (P) 0 1 P ′ = Pδ

Table 2: Summary of common gas turbineparameter corrections ECM [13].

2.2.2 Limitations

The quality of the parameter corrections can beimproved by using more sophisticated models toestimate the coefficients “a” and “b” instead ofusing typical or empirical values. Reference [7]offers details on how the coefficients may beestimated by running a full thermodynamiccomputer model of the engine over a range of inlettemperatures, pressures and exhaust conditions.The main issue is that the engine model is notaccessible to operators, who do not possessdetailed information to run high end simulations.This can be solved by making use of someempirical methods that use in-service datacollected from the engine and some statisticalanalysis tools. An example of this kind ofapproach can be seen in reference [13]. Note thatequation 4 doesn’t take into account changes inviscosity with altitude. Even tough the effects of

4

differences in the Reynolds number might not besignificant for some engine models, there can beerrors related to the comparison of correctedparameters if this effect is not taken into account.Also, the ratio of specific heat and the gasconstant are assumed to be insensitive to changesin the atmosphere which does not correspond tothe reality even thought this is a commonassumption when performing dimensional analysison a gas turbine engine.

2.3. Performance deteriorationdetection

The simplest way to identify deterioration inengine performance or engine events that need tobe looked into by the maintenance personnel is byanalysing the trend of the engine’s performanceparameters over time and identify the effects ofsaid events. Abnormal situations related to theengine condition, like mechanical changes in theengine that may result in failure or lead to theoccurrence of incidents, normally imply an abruptshift in the corresponding parameters. On theother hand, the effects of accumulated flight cyclesresulting in wearing of engine’s components and adecrement in its performance can be identified bya gradual increase over time of the parameter’sDelta values. In the first situation described, alsoknown as Step Shift, data suffers a shift thatoccurs in a short period of time as opposed to theSlow Drifts that are associated with a slowvariation in the parameter’s trend that occurs overa longer period of time. An effective monitoring ofcruise trends helps minimizing the risks associatedto unexpected engine failures and avoid excessivedegradation of the engine’s performance. Figure 6showcases an example of an actual incident causedby a Low Pressure Turbine Clearance Controlvalve failure.

Figure 6: Delta EGT plot of an A320-200 engine.

Notice the abrupt change in the values of theDelta EGT for the engine in position 1, marked inblue, and another shift when the valve wasreplaced, marked with a vertical dashed red line.This kind of behaviour is what the enginemaintenance personal is looking for when

performing engine parameter trend analysis.Unfortunately the time period represented infigure 6 is not broad enough for a significantperformance deterioration to be evident in theparameter Delta, however as discussed before lossof performance from accumulation of flight cyclesis visible through a gradual increase over time ofthe Delta values.

2.4. Typically observed parametershifts

Detecting abnormal trends in key performanceparameters is an effective way of detecting possibleongoing problems or impending failures in gasturbine engines [11]. But even though it is possibleto detect problems with a given engine by lookingfor Step Shifts in the engine’s parameter deltas, itis also important to identify the source of theproblem so appropriate actions can be undertaken.This can be solved by taking advantage of the factthat some anomalies produce similar shifts everytime they manifest. By recording this standardshifts operators are able to estimate the source ofthe problem just by measuring the key engine’sparameters deltas [12]. These standard shifts arealso called anomaly’s fingerprints and normallyeach engine manufacturer supplies a documentcontaining a register of the fingerprints of themost common anomalies. The operator will usetables containing information from the fingerprintsdocument as a starting point for a closerinterpretation of the anomaly and in conjunctionwith its experience will be able to better assess thesituation.

3. ECM and cruise trend analysisconsiderations

To monitor the performance of an engine it isnecessary to be able to interpret data that can berepresentative of the engine’s health. For data tobe considered valid for trend analysis it needs torespect a set of conditions which makes it eligibleto be considered a Stability Point for a givenflight, as defined in [9]. Depending on the flightconditions there can be several stability points fora specific flight or in the worst case none.

3.1. Stability Point search criteriaThe set of conditions used to search for stableframes can be divided into Basic Conditions andStability Criteria. Their purpose is to avoidtriggering reports when the parameters are of nointerest. Basic Conditions may include ranges forvalues like Altitude(Alt), Mach Number(MN), N1and even contemplate binary parameters like theengine’s bleeds status. They serve as a firstfiltering criteria to exclude values that have nochance of being eligible for a stability point

5

calculation. As for the Stability Criteria they areused to rule out time frames where the values arenot stable enough to translate accurateinformation about the engine’s condition.

3.2. Fixed Window and GlidingWindow stable frame search

There are two methods to perform stable framesearch. The simplest method consists of makingadjacent and consecutive windows with anequivalent length of a stable frame (100 secondsfor instance), and then verify if all basic conditionsand stability criteria are met for each individualwindow. This means that each measured valuebelongs only to a single window, the windows donot overlap each other. Another method forperforming the search is called Gliding WindowSearch and implies overlapping the windowsconsecutively, taking a single window at time nand advancing it consecutively by i, getting Nwindows starting at{n;n+ i;n+ 2i;n+ 3i; ...;n+Ni}, where N isgiven by equation 5.

N ≤ tai,N ∈ N (5)

Note that ta represents the period of time whereanalysis is being performed and expressed in thesame time unit as i. Figure 7 exemplifies thedifference between both methods described.

Figure 7: Fixed Window Vs GLidding Windowsearch method [2]

For the development of the current work theselected method for stable frame search was theGliding Window method with a 1 seconddisplacement.

3.3. Quality NumberUsing the conditions described in the previoussub-sections many stable frames will be found andtherefore many stability points will be calculated.It is then necessary to have a consistent way ofidentifying and selecting only the ones that bestrepresent what is happening to both the aircraftand its engines. It would also not be practical orefficient to store so many values per flight.Reference [3] suggests a criteria for selecting thebest stability points from each flight. Bycalculating the Quality Number (QN)

corresponding to each Stability Point, usingequation 6, it is possible to have a measure of howreliable a stability point is.

QN = Was2aTola

+Wbs2bTolb

+ · · ·+WPs2pTolp

(6)

Wi represents the weight factor for each parameterp ∈ P , being P the set of all analysed parameters,and Toli its tolerance according to the definedStability Criteria. The unbiased sample variance,s2, can be calculated using equation 7.

s2p =n

n− 1σ2p =

1

n− 1

n∑i=1

(xi − x)2 (7)

Note that n is the number of readings, or samples,in the observation window for a given parameter pand xi a reading made at time i. The mean x canbe obtained using the formula below:

x =1

n

n∑i=1

xi (8)

The variance gives a measure of the dispersion ofthe samples to the mean. The stability point willbe of better quality when the overall variance of itsparameters is lower. As stated before a stabilitypoint will then be the set of all calculated xp, overthe duration of a frame, with p belonging to the setof all analysed parameters P .

StabilityPoint = {x1;x2;x3; · · · ;xP } (9)

3.4. Application of parametercorrections

To be able to compare measurements made underdifferent ambient conditions it is first necessary tocorrect these values. After extracting all stabilitypoints of interest but before performing any sortof trend analysis it is first necessary to correct thecalculated values of the considered parameters sothey have a common ground between them. This isaccomplished by first calculating the dimensionlessconstants θ and δ using the following expressions:

θt =T02T0

=T02

288.15(10)

δt =PT02P0

=P02

1013.25(11)

These corrections factors are calculated forstandard day conditions using values extractedfrom each stability point, more precisely the totalair temperature at the inlet and the total pressureat the inlet of the low pressure compressor,represented in equations 10 and 11 as T02 andPT02 respectively. The notation used here is in

6

accordance with the one used in Figure 5. Thesub-index, t, denotes that the values werecomputed with total temperature and pressure. Ifwe assume that the inlet does no thermodynamicwork [5], the free stream total temperature, orTAT, can be used as the total temperature at bothengine inlets.If we consider, for instance, that the parametersbeing trend monitored are the EGT, FF, N2 andN1, which are considered to be the four keyperformance parameters, the corrected values foreach of these parameter is calculated as follows:

EGT ′ =EGT + 273.15

θt[K] (12)

FF ′ =FF√θtδt

[Kg/hour] (13)

N2′ =N2√θt

(14)

N1′ =N1√θt

(15)

The corrected EGT is converted from degreesCelsius to Kelvin. As mentioned before thecorrection factor θt will be the same for bothengines as it is computed from the TAT. As forthe δt it is used to correct the FF and is computedfrom the respective total pressure at the engine’sinlet.

3.5. Engine Bleed and its effectsHighly pressurized air is extracted from the engineto feed the aircraft’s pneumatic systems that drivethe air conditioning systems, wing anti-ice,amongst others. This can have major effects onthe performance of an engine and needs to betaken into account. The air is normally bled fromthe intermediate stages of the High Pressurecompressor. Keep in mind that the BasicConditions dictate that the anti-ice valves must beclosed for the duration of the analysis for data tobe valid for ECM purposes. The ideal would be touse values not affected by bled air, but since this isnot possible, as the systems described above arevital for the aircraft normal operation, we can onlymitigate the side effects of engine air bleed in theperformance assessment. Most automated datacollection systems aboard the aircraft are not ableto quantify the amount of air that is being bled,but they can give an estimate of the order ofmagnitude of the bleed effect in the engine’sperformance. This information can be extractedfrom the Engine Cruise Report using the EngineControl Word 1 (ECW1) or, when this is not anoption, it can be inferred from the position of theswitches in the cockpit. From this informationthree different bleed modes can be identified:

• Minimum Pack

• Normal Pack

• Maximum Pack

The Pack Flow Selection regards the functioningmode of the air-conditioning system. Thisinformation is crucial since stability points cannow be separated according to the amount ofpressure being withdrawn from the engine. Theeffects of engine air bleed in the performanceparameters are made clear in Figure 8

Figure 8: High-pressure fan speed vs low-pressurefan speed showcasing different pack flow selection.

The PSK code 1, 2 and 3 correspond to aMinimum, Normal and Maximum bleed statusrespectively. By taking a closer look at Figure 8 itis possible to notice that for a given N1 the valuesof N2 are consistently higher for Maximum PackFlow than for the other modes. A higher N2implies an increase in FF which means a decreasein efficiency and therefore a loss in performance.For the same reason parameter corrections wereneeded to take into account different ambientconditions, Stability Points should also not becompared if calculated under different Pack FlowSelection. The ideal would be to only take intoaccount stability points generated from MinimumPack Flow Selection, since it is the mode thatleast affects engine performance, but the majorityof stability points have a Normal Pack Flow. Thepercentage of Stability Points with a MinimumPack Flow Selection is low, approximately 7%,which is not sufficient to perform trend analysis.For this reason when developing the tool for ECMboth Minimum and Normal Pack Flow Selectionwill be taken into account, as will be described inthe next section, and Stability Points withMaximum Pack Flow Selection will be discarded.

7

3.6. Baseline DefinitionTo perform parameter trend monitoring it isnecessary to define a baseline for the engine tohave a reference to measure the performancedecay. As discussed before, engine’s manufacturersobtain Baseline values by testing each engine whenthey are new. As operators can’t replicate theseconditions the only option to perform this kind ofanalysis is by deriving a baseline from datarecorded in-flight. The best way to do this is byusing data from a period of time when the engine’sperformance is at its peak and use it as reference.The most obvious would be to do this when theengine was new or with few flight cycles and storethe measurements to trace the Baselines, but inmost cases this is no longer an option. One of themajor indicators of engine performance is theEGT margin and it can be used as a parameter tochoose the best period of time to use for baselinecalculation. EGT margin is at its maximum aftershop visits, best case scenario after the engine hasbeen overhauled, and will decay with each flightcycle [6]. For the development of the current thesisthe baselines will be calculated from the bestStability Points found in the first 90 days ofservice after an engine has been overhauled. Thisperiod of time was selected based on empiricalknowledge, taking into account the necessary timeto collect enough data to generate the baselinemodel but keeping in mind that the longer theperiod the more engine loss of performance cancontaminate the data. For the reasons presentedtwo different Baselines are considered, onerespecting a minimum Pack Flow Selection andthe other a normal Pack Flow Selection. Eachparameter delta will be calculated as the differencebetween the corrected Stability Points and thecorresponding Baseline with the same Pack FlowSelection.

3.7. Trend MonitoringThe process of Trend Monitoring implies thecalculation of the parameter’s Deltas. TrendAnalysis corresponds to the evolution over time ofthe parameter’s Deltas calculated from thedifference between the corrected data extractedfrom the Stability Points and the Baseline models.Depending on the parameter being analysed, itsdelta can be expressed in the same units as theparameter or as a dimensionless ratio as seen inthe following equations.

4EGT (t) = EGTStab.Point(t)− EGTBaseline(t)[◦C]

(16)

4FF (t) =FFStab.Point(t)− FFBaseline(t)

FFBaseline(t)(17)

4N2(t) =N2Stab.Point(t)−N2Baseline(t)

N2Baseline(t)(18)

4. Development of an ECM toolThe tool developed had the objectives of beingable to work for different aircraft and enginemodels composing TAP’s fleet and also optimizingthe tool’s ability to process flights in order toincrease the number of flight hours worth of datathat could be processed each day. Theimprovements made allowed for a significantincrease in the capacity to process flight-data withthe intent of calculating Stability Points. The toolhas two core functions: (1) process flights andstore the Stability Points corresponding to thatflight; (2) use said points to perform trend analysison demand for a given engine over a specificperiod of time. The first requisite taken intoaccount, when developing the software, was thepossibility of working with differentaircraft/engine combinations. To do this, afunction was created ro read text files (.txt) withthe information needed to create the BasicConditions and Stability Criteria for each aircrafttype. To analyse flight-data from a specific aircraftthe program needed the configuration files for therespective aircraft, one for the Basic Conditionsand the other for the Stability criteria. Thedevelopment of an User Interface for the ECM toolas well as making it available remotely, were on ofthe most needed implementations. It allows crosschecking data between different departments aswell as increasing the accessibility to an analysistool which will ultimately contribute to anincrease in safety and efficiency in maintenanceprocesses. The development an User Interfacethrough a web-server was done using R and apackaged called Shiny using GGVIS as the plotsgenerator. The application offers the option ofchanging the parameter being plotted through adrop-down menu or changing the engine beingmonitored by selecting a different check-box andalso allows the selection of the period beinganalysed through an interactive calendar.

5. Results - Case StudyTo test the ability of the developed ECM tool tocorrectly identify engine performance loss orsituations affecting its correct functioning, a realcase of a component fault was selected. Theintention was to verify if by using the softwaredeveloped in the current thesis it would bepossible to clearly identify this event as was doneusing the manufacturer analysis tool. The selectedevent corresponded to a High Pressure TurbineClearance Control (HPTCC) valve failure. Thisevent was detected by P&W’s software through

8

the visual analysis of the performance parametersdeltas as seen in Figure 9.

Figure 9: P&W long term trend plot report of aPW4168A engine.

Notice how the EGT values, plotted with G’s, andthe FF , plotted with F’s, register a suddenincrease, after the green line, and return to itsprevious behaviour, after the red line representingthe HPTCC valve replacement. The faulty valvecan be seen in Figure 10.

(a) HPTCC valve lever position (b) HPTCC valve butterfly position

Figure 10: Faulty HPTCC valve.

The problem with this valve was that its butterfly,as seen in 10(b), was opened when it should beclosed according to the valve’s lever position, see10(a). Since the variations detected on the reportare common in HPTCC valve faults, Maintenanceused a borescope to check the butterfly position.

The ECM tool developed in the current thesis wasused to process all flight-data corresponding to theperiod where the incident happened plus theflights needed to calculate the Baselines (corresponding to the 90 days after engineinstallation for both positions). The Delta EGTand Delta FF from engine 2 were calculated andplotted in the interactive ECM tool webapplication. The results obtained are representedin Figure 11 and 12.

Figure 11: Delta EGT[◦C] Eng. 2.

One can easily note the shift in the beginning ofMarch that corresponds to the one detected withP&W’s software. The values return to normalityonce the valve is replaced. This is also noticeable,but in a smaller scale, in the FF Delta, see Figure12

Figure 12: Delta FF[%] Eng. 2.

Also note that in both figures it is possible tonotice a small increase in the Deltas over time,where the mean Delta value from December toFebruary is lower than from March to May forboth EGT and FF. This is in accordance with theslow degradation of performance resulting in aslow and gradual increase over time of theperformance parameter Deltas.From the results presented one can conclude thatthe objectives of the dissertation wereaccomplished. The ECM tool was able to identifygradual deterioration of engine performance andidentify faults in engine components like theHPTCC valve fault presented.

6. ConclusionsIn a competitive environment such as the airtransport market it is of utmost importance tomake sure the operation is being conductedefficiently while keeping high safety standards.Transportadora Aerea Portuguesa (TAP)) is theleading Portuguese airline and, like other airlines,is affected not only by the severe competition ofthe commercial aviation industry but by the lowprofit margins and high fluctuations of the price of

9

fuel which has made the company seek strategiesthat improve the efficiency of their operation andconsequently reduce operating costs.

6.1. AchievementsEngine condition monitoring is a methodologythat evaluates aircraft engine health condition, itis the main process behind condition-basedmaintenance which helps make maintenanceplanning more efficient. Analysis such as engineperformance parameters trend monitoring allowthe monitoring of performance degradation overtime and any engine events that requireintervention. From the results presented one canconclude that the objectives of the dissertationwere accomplished. The ECM tool was able toidentify gradual deterioration of engineperformance and identify faults in enginecomponents like the HPTCC valve fault presented.The main advantage of the developed program isthat it is dynamic in the sense that anyone canprocess data flights and perform trend analysisinteractively on a web application without havingto be familiarized with statistic tools like R. Alsothe development of the User Interface andweb-server will contribute greatly to theintegration of different departments at TAP thatrequire access to this data.

6.2. Future WorkThe ECM tool developed still relies on thegraphical interpretation of the results obtained, itwould be interesting to integrate some machinelearning concepts to the software. The use ofpattern recognition algorithms would allow thetool to store information about the shifts causedby certain events in the parameter deltas andgradually create a list of the most common eventsand the typical shifts they produce. By crossinginformation of this list with maintenance reportscontaining the reasons for engine removals ormaintenance events and its causes the tool wouldbe able to label these patterns and create its ownfingerprints database for the parameter deltasshifts. Another possible improvement might be thecreation of an algorithm to calculate the take-offEGT Margin and add this performance parameterto the analysis. The data needed to calculate theEGT Margin was not included in the data framesused in the work developed but with theextraction of the correct parameters thiscalculation would be possible.

References[1] S. Ackert. Engine Maintenance Concepts for

Financiers. Macquarie AirFinance, 2011.

[2] Airbus. Aircraft Maintenance Manual.

[3] Airbus. Getting to grips with aircraftperformance monitoring, December 2002.

[4] P. . W. Aircraft. Malfunction analysis throughgas generator comparisons and mechanicaltrends. Aircraft Gas Turbine OpertationInformation Letter, 1965.

[5] J. ao T. Borges. Aulas de Propulsao. InstitutoSuperior Tecnico.

[6] G. Electric. Engine deterioration andmaintenance actions. ICAO / TransportCanada Conference Aircraft Panel, September2006.

[7] J. Kurzke, editor. Model based gas turbineparameter corrections, Atlanta, 2003. in ASMETurbo Expo 2003.

[8] J. Leahy. Global market forecast. TechnicalReport 2033, Airbus, Spetember 2014.

[9] C. J. L. Palmeiro. Estudo e desenvolvimentode um algoritmo de deteccao de pontosde estabilidade. Master’s thesis, InstitutoSuperior Tecnico, December 2009.

[10] Pratt & Whitney. Takeoff EGT MarginEstimates, May 2005.

[11] I. Y. Tumer and A. Bajwa. Learning about howaircraft engines worl and fail. AIAA, 1999.

[12] R. Vingerhoeds, P. Janssems, B. Netten, andM. A. Fernandez-Montesinos. Enhancing off-line and on-line condition monitoring andanalysis, volume 3 of Control EngineeringPractice. 1995.

[13] A. J. Volponi. Gas turbine parametercorrections. Jounal of engineering for gasturbines and power, 121(4):613–621, 1999.

10