Gas Turbine Engine Health Management Past, Present, and Future Trends

Allan J.Volponi

Pratt&Whitney, 400Main St,MS162-15,

East Hartford,CT06108 e-mail:allan.volponi@https://www.doczj.com/doc/e016013987.html, Gas Turbine Engine Health Management:Past,Present, and Future Trends

Engine diagnostic practices are as old as the gas turbine itself.Monitoring and analysis methods have progressed in sophistication over the past six decades as the gas turbine evolved in form and complexity.While much of what will be presented here may equally apply to both stationary power plants and aeroengines,the emphasis will be on aeropro-pulsion.Beginning with primarily empirical methods centered on monitoring the mechan-ical integrity of the machine,the evolution of engine diagnostics has bene?ted from advances in sensing,electronic monitoring devices,increased?delity in engine modeling, and analytical methods.The primary motivation in this development is,not surprisingly, cost.The ever increasing cost of fuel,engine prices,spare parts,maintenance,and over-haul all contribute to the cost of an engine over its entire life cycle.Diagnostics can be viewed as a means to mitigate risk in decisions that impact operational integrity.This can have a profound impact on safety,such as in-?ight shutdowns(IFSD)for aero appli-cations,(outages for land-based applications)and economic impact caused by unsched-uled engine removals(UERs),part life,maintenance and overhaul,and the overall logistics of maintaining an aircraft?eet or power generation plants.This paper will review some of the methods used in the preceding decades to address these issues,their evolution to current practices,and some future trends.While several different monitoring and diagnostic systems will be addressed,the emphasis in this paper will be centered on those dealing with the aerothermodynamic performance of the engine.

[DOI:10.1115/1.4026126]

Introduction

Cost Benefit.As previously mentioned,engine life cycle cost (and safety,which can be related to cost)has been the primary driving force for engine diagnostic development.While cost is a complex topic in and of itself,for simpli?cation,let us divide this into two general(albeit,not mutually exclusive)categories:opera-tional cost and maintenance cost.To have appreciation of the magnitudes involved,in the(worldwide)commercial aviation sec-tor alone it is estimated that$50billion is spent annually for?eet maintenance activities,and this is expected to rise to$65.3billion by2020[1].The DoD reported spending approximately$1.1bil-lion on engine maintenance in?scal years1992and1993[2]. These seemingly large numbers would be something just short of astronomical if extended to include all military aeroengine?eets worldwide,as well as all of the land and marine propulsion appli-cations,power generation,and oil pumping stations in operation. Improvements in maintenance logistics,reductions in unscheduled events(and their consequences),and improvements in operational ef?ciency can have an enormous impact on reducing these costs. Diagnostics has a signi?cant role to play in contributing to these cost reductions.

Throughout this presentation,I will use the term diagnostics, somewhat liberally,to encompass both detection and identi?ca-tion of(engine related)faults(failures)as well as the monitoring of engine system/subsystem degradation and deterioration prior to actual failure.From a diagnostic perspective,reductions in opera-tional cost can be achieved by avoiding unscheduled events such as UERs,extensive line maintenance for failed subsystems caus-ing secondary damage and IFSDs that can drive an engine re-moval or more catastrophically result in aircraft damage or loss of life,as well as detecting and identifying partial failures or malfunctions of subsystems that contribute to increased fuel con-sumption,such as engine bleed leaks/failures,active clearance control,and variable geometry actuation abnormalities to name a few.Any of these can contribute to an increased number of delays and cancellations(D&Cs)(in the aero sector)that result in cus-tomer dissatisfaction and its associated costs.From a maintenance and?eet management perspective,diagnostics can impact mainte-nance logistics,repair scheduling,spare parts inventory,overhaul and repair work scope,etc.,all of which contribute to the overall life cycle cost of an individual engine as well as the entire?eet as a whole.

A related term,prognostics,deals with predicting the future (health)state of the engine and its accessories and has come into vogue in the last decade or so.Prognostics is the ability to predict the future condition of a component and/or system of components. For the purposes of gas turbine engine prognostics,this de?nition is often further described in terms of hard failures of components or condition/degradation of performance related problems.Failure prognostics is focused on the prediction of damage state or failure rate of a component or system of components in an engine.Failure prognostics is usually affected by the diagnosis of speci?c engine faults,depending on the level of impact the component experien-ces from the fault condition.Prognostic models are required to project to the future condition.Prognostics can also be associated with the slower degradation(wear related)processes that an engine is exposed to throughout its life.It is usually associated with the diagnosis of fault(s)conditions and the capability of pre-dicting when the symptoms of the identi?ed fault(s)will reach an undesirable state in which system operation will be adversely affected.Prognostic models are required to project the future “path”of these identi?ed fault(s)on total system performance or reliability.

An example of prognostics is the calculation of the remaining useful life of a life limited part,component,or subsystem of the engine.Prognostics may or may not be preceded by the detection of a failure precursor.An example of a prognostic technique that is preceded by failure detection includes the detection of bearing

Contributed by the Aircraft Engine Committee of ASME for publication in the

J OURNAL OF E NGINEERING FOR G AS T URBINES AND P OWER.Manuscript received

November20,2013;?nal manuscript received November25,2013;published online

January2,2014.Editor:David Wisler.

Journal of Engineering for Gas Turbines and Power MAY2014,Vol.136/051201-1

Copyright V C2014by ASME

spalling(failure precursor),and then estimation of remaining use-ful life of the bearing.Another example is effective cycle count-ing of life limited parts such as disks,spacers,and shafts.While no failure precursor is evident in this case,a prediction or forecast of the remaining time before the part reaches the end of its useful life is generated.All prognostic methods include a forecast hori-zon and typically associate a con?dence in the forecast.

The preceding paragraphs hopefully provide a glimpse as to why diagnostics(and prognostics)is performed and the role it plays in the operational and maintenance activities that surround gas turbine engines.The question of who engages in these diag-nostic activities is also worthy of https://www.doczj.com/doc/e016013987.html,mercial airlines and power plant operators,military users,original equipment (engine)manufacturers(OEMs)and independent maintenance and overhaul facilities all engage in a variety of diagnostic activ-ities ranging from engine monitoring,health tracking,fault diag-nosis,and?eet health management.Some of this,as we will discuss later,is accomplished onboard,some of it off-board and some of it requiring a mixture of both.

The role that diagnostics(and more notably,prognostics)plays for the engine OEM has changed over time from one of support-ing maintenance troubleshooting to a more proactive activity.The term power by the hour,originally coined by Bristol Siddeley Engines Ltd.,a British aeroengine manufacturer later purchased by Rolls-Royce,describes the concept succinctly.All major engine OEMs offer some form of a?eet management program (FMP)to their airline customers wherein,for a?xed sum per?y-ing hour,a complete engine and accessory replacement service is provided,thus allowing the operator to forecast their operational costs with great accuracy and,thus,relieving them of the need to stockpile spare engines,parts,and accessories.The OEM effec-tively provides a maintenance program over an extended period of time and the operators are assured of an accurate cost projection and avoid the costs associated with breakdowns.Knowing the cur-rent health of each engine in the?eet contributes to the OEM’s ability to con?gure an FMP to support their maintenance business model and derive pro?t and satisfy their customer’s needs.A win-win situation for all parties.

https://www.doczj.com/doc/e016013987.html,rmation is the key element in any health management system.A reasonable analogy can be found in medi-cal practice.When a patient is ailing(or even during a“well vis-it”),the physician gathers information from physical(visual) observations as well as measurements taken with various “sensors”(blood pressure,electrocardiogram(EKG),temperature, blood analysis,X-rays,MRIs,CAT scans,etc.),compares these observational values to known nominal levels,notes deviations that form symptoms,and matches these to known ailments in order to form his/her diagnosis.Much of this medical information is experiential and has been learned by trial and error(and some experimentation)over the centuries.Gas turbine diagnostics is not quite as old as medical diagnostics,but it too has as its key ele-ment,information.This information comes from observations and measurements,from models both physics based and empirical, and analytical methods designed to estimate health indicators that cannot be directly measured or observed.

Information is the key enabler for engine diagnostics.Unfortu-nately,it is also its biggest dif?culty.The dif?culty stems from the fact that there is a considerable information de?cit relative to the number of possible health factors arising in the many compo-nents and subsystems that comprise a modern gas turbine.The number of unknowns simply outweighs(by a very large margin) the number of knowns(i.e.,information)in the diagnostic prob-lem.The addition of more physical sensors,in itself,does not solve the problem.An additional sensor adds an additional prob-lem,i.e.,the health of the sensor itself(is it drifting,is there bias, is precision degrading,etc.).Not to mention,sensors add weight and must be maintained,which adds cost.Thus,the measurement suite used in a gas turbine diagnostic system must provide more information to the diagnostic problem and its attendant cost bene-

?t than the cost introduced by adding the sensor in the?rst place. Fortunately,information can come from other sources,some of which are not directly related to physical sensors,such as:?Constraints:For example,with time and operational use,a gas turbine would be expected to degrade,not get healthier.

This places bounds(or constraints)on the general solution and in effect reduces the number of unknowns.

?Domain knowledge:The laws of physics require certain inter-relationships to exist.For example,if fuel?ow were observed to increase but exhaust gas temperature(EGT)was observed to decrease this would not be generally indicative of an engine fault but would suggest a problem in either the fuel measuring system or the temperature measuring system(or both).

?Assumptions:For example,a sudden(temporal)observed shift in multiple sensed parameters could be indicative of a single subsystem fault(e.g.,a stuck bleed valve)or a simulta-neous fault in all of the sensors or a host of combinations of problem faults.Since experience suggests that it is highly unlikely to have multiple faults or failures,a single fault assumption might be used,thereby reducing the diagnostic problem to something more manageable.

?Negative information can be used to rule out certain potential diagnostic possibilities by the fact that certain observations or patterns were not observed.

We will come back to this repeatedly throughout the paper as we track the evolution of the diagnostic systems in terms of lever-aging as much information as possible with as few sensors as pos-sible,by exploring constraints,domain knowledge,assumptions, experience,information fusion,and the analytical methods that are used to extract and leverage this information.

Engine Health Management(EHM).We end this section with the introduction of a pervasive acronym,EHM.Over three decades ago,this acronym would have been recognized as mean-ing engine health monitoring.Today it has come to mean engine health management.The former refers to passive observations and the latter an active pursuit,with dependencies on the former. Engine health monitoring began in earnest with the advent of commercial high bypass turbofan engines in the1970s.The high cost of repair for these large modular-design machines provided an economic impetus for the development of analysis methods to track the performance health down to the module level in an attempt to reduce maintenance work scope and overall costs.With the advent of the full authority digital engine control(FADEC)in the early1980s,?yable instrumentation intended for control func-tions was now also available for diagnostic purposes.With contin-ued advances in low-cost computing,high speed communication, and more sophisticated sensors,EHM systems are now found in a variety of applications both commercial and military.

Since the need for engine health management systems varies between applications;a one-size-?ts-all solution is not likely to exist and designs must be tailored to each application.EHM can be visualized as a portfolio of building blocks that can be used to create customized architectures that best meet individual user needs.These architectures may include both engine-hosted and ground-based elements as complementary features of an overall integrated system.Engine-hosted elements generate data from onboard sensors and perform basic fault isolation and failure pre-diction,supporting on-wing maintenance,while the ground-based elements support long-term degradation trending,providing plan-ning information that can be used by aircraft?eet managers.The remainder of this paper will explore the different monitoring sys-tems and analytical methodologies that have evolved over time to become the foundation elements of current EHM systems in use today and as well as future trends in these systems and their enhancement potential.

051201-2/Vol.136,MAY2014Transactions of the ASME

Historical Perspective

Engine Health Monitoring.Engine condition monitoring and diagnostic practices are as old as the gas turbine itself.Monitoring exhaust gas temperature(EGT),spool rotational speed and overall engine vibration level(amplitude)marked the beginning of moni-toring methods to insure safe operation by highlighting abnormal levels or exceedances.This rudimentary set of instrumentation was suf?cient to detect deteriorated and undesirable operation, however,it did little to aid in the identi?cation of the root cause of the abnormality.As the gas turbine grew in complexity,along with the attendant cost to maintain and repair,instrumentation expanded.As with most systems,reliability and cost were the main drivers.Engine control functions drove the need for addi-tional sensors(primarily in the engine’s gas path)as well.At?rst this parametric information was only available in the cockpit for the?ight crew to monitor the engine’s operation.During stable cruise operation(for aeroengines),with the engine in a relative steady-state condition,the?ight crew would record parameter lev-els from cockpit gauge observation(typically spool speeds(N1, N2),EGT,fuel?ow(Wf),vibration level,and oil system tempera-ture and pressure and submit a?ight coupon to the ground crew upon landing for subsequent trending by the airline’s power-plant engineers.

In the1970s,electronic engine condition monitoring hardware was introduced on wide body aircraft that would relieve the?ight crew from recording?ight coupons by automatically monitoring and capturing salient engine parameters during takeoff and stable cruise operation.These systems,airborne integrated data system (AIDS),later renamed to airborne integrated monitoring system (AIMS)when the earlier acronym took on a different meaning, consisted of a series of electronic hardware that would monitor a plurality of?ight parameters(e.g.,altitude,Mach,total air tem-perature(TAT and/or engine inlet temperature T2),ambient pres-sure(Pamb,and or total engine inlet pressure P2),aircraft angle of attack,pitch,roll,and yaw rates,etc.)as well as available engine speci?c parameters(e.g.,N1,N2,EGT,Wf,oil system temperature,and pressure and vibration amplitudes,etc.).At?rst only engine control parameters were available in the gas path,but later other gas path parameters were included such as burner pres-sure(Pb),high pressure compressor(HPC)inlet temperature and pressure(T25,P25),HPC exit temperature(T3),and engine pres-sure ratio(EPR?P5/P2),etc.At?rst these additional parameters were added through arrangement between the airline and the OEM on a case by case basis,requiring special sensor instrumen-tation purchase,installation,and multiplexing hardware to route the signals to the AIMS hardware.Much of this was driven and speci?ed by the individual airlines(in concurrence with the OEM) for different aircraft/engine model combinations before the advent of FMPs.Features in the AIMSs differed from airline to airline and the software content and con?guration for the AIMS was dic-tated and controlled by the individual airlines.As such there was no“standard”AIMS con?guration at the onset.Nevertheless, there was some overall functional similarity between systems. The AIMS would monitor the aircraft and engine parameters available to it and record any parameter exceedance from stored thresholds during the?ight.During cruise operation,the system would continuously search for stable conditions(altitude,Mach, spool speeds,etc.,not changing more than prespeci?ed limits), perform a time average across all parameters within this stable frame,and store the data.Likewise,during takeoff,the monitoring system would capture various parameters and record critical val-ues;for example,max EGT and record it for subsequent ground analysis.At the conclusion of the?ight these recordings were retrieved from the quick access recorder(QAR)and loaded into the airline’s ground station computers for trending,EGT margin tracking,and other analysis to support?eet wide health tracking supporting maintenance(line and depot)decisions and logistics. As the digital age took hold in the1980s,with advances in avia-tion electronics and computers,greater sophistication in the AIMS hardware was possible.Full authority digital engine controls (FADECs)replaced the hydromechanical engine controls with digital data buses(ARINC)enabling a wider range of parameters available to the AIMS.With faster central processing units (CPUs)and greater memory capacity available,the monitoring systems could perform a wider range of analysis and storage was capable in real-time during?ight.The airborne communications addressing and reporting system(ACARS)slowly replaced the need for QARs,wherein the recorded data could be radio trans-mitted to the ground-station computers during?ight,and quickly analyzed to provide greater informational direction to line mainte-nance crews in advance of the aircraft landing.The added compu-tational capacity of the AIMS facilitated the generation of aircraft condition monitoring function(ACMF)?ight reports whose pri-mary objective is to obtain indications of upcoming failures ahead of time within aircraft and engine systems.This enables operators to initiate preventative maintenance actions to minimize the risk of impacts on the dispatch reliability.For this purpose the ACMF acquired characteristic system data and provided the operators with performance and trend information as well as indications of incipient deviations from normal system conditions.

Since its inception three decades ago,ACMF has expanded to include more extensive reporting capability and has become a standard practice in current aeroengine monitoring.Typical reports in use today include:

?engine takeoff report

?engine climb report

?engine cruise report

?engine gas path advisory report

?engine mechanical advisory report

?engine start summary report

?ground run-up report

?FADEC maintenance report

?engine divergence report

?engine report on request

The ACMF reports(which are still in use today)provide a wealth of information regarding the general health state of the engines being monitored as well indications of emerging faults and failure conditions.These reports have expanded in scope and complexity as onboard computer capability evolved over time,but this still requires the man-in-the-loop to digest and analyze the in-formation,perform additional off-board analysis,and combine other sources of information to decide upon the appropriate action in maintaining the?ight integrity of the engine.The reports gener-ated in the1970s and1980s were subsets of those listed above and relied more heavily on the ground power plant analyst to pi-ece together the health pro?le of the engine.To aid in this task,a new analysis methodology was introduced in the early1970s known as gas path analysis(GPA).This method took advantage of existing instrumentation common to all gas turbines and pro-vided a means of tracking overall performance degradation of the engine as well as degradation in each of the main engine compo-nents,which for a twin spool turbofan engine included the fan, lower pressure compressor(LPC),high pressure compressor (HPC),high pressure turbine(HPT),and the low pressure turbine (LPT).The data required to support this analysis method were within the capability of the early AIMSs to capture and store(and later transmit)and was generic in its approach to be applied to any con?guration of gas turbine,i.e.,single spool to triple spool, turbojet,and turbofan.Over the subsequent four decades,it has been the topic of much research that has extended its capabilities from its early formulation and remains as one of the central tech-niques supporting health management to this day.

Gas Path Analysis:Early Overview.Gas path analysis is a mathematical process to isolate and assess the magnitude of engine performance shifts to the component level based on observed changes in measurements taken along the engine’s gas

Journal of Engineering for Gas Turbines and Power MAY2014,Vol.136/051201-3

path,such as temperatures,pressures,speeds,?ows,etc.This form of engine diagnostics is a fairly mature methodology,which has been in practice in both commercial and military engine diag-nostic programs for almost four decades.References to this approach ?rst appeared in the literature by one of the early investi-gators and pioneers,Louis A.Urban [3–6].As a result of its pro-longed existence,the method has been referred to by several names,gas path analysis (GPA),module performance analysis (MPA),and performance diagnostics being several of the most popular.In this paper we will use these terms interchangeably.Part of its popularity (and longevity)also stems from the fact that no special instrumentation is required to perform this activity.Bill of material (BOM)sensors typically found on commercial (and military engines)to support control functions will suf?ce to support this analysis.The level of analysis and con?dence in results does,of course,depend on the exact measurement suite in use and additional (noncontrol)gas path instrumentation has been added to engines over the years to further enhance this capability.As a generality,diagnostic systems rely on discernable changes in observable parameters to detect physical faults.Physical faults consist of a variety of problems or combinations of problems such as foreign object damage (FOD),blade erosion and corrosion,worn seals,plugged nozzles,excessive blade tip clearances,etc.If severe enough,these physical faults will induce a change in the thermodynamic performance of the engine and its attendant com-ponents.The underlying precept behind GPA is that engine per-formance depends on the state of these individual components and that furthermore,the condition of these components can be

mathematically represented by a set of independent performance parameters.For the compression modules,it has been customary to use adiabatic ef?ciencies and ?ow capacities and,for the tur-bine modules,adiabatic ef?ciencies and effective nozzle areas.If changes are then observed in the gas path measurements,the prob-lem becomes one of estimating those (module)thermodynamic parameters responsible for the change,with the hope that this will facilitate the subsequent isolation of the underlying physical fault.This concept is summarized in Fig.1below.

This type of procedure can be referred to as relative engine per-formance diagnostics,i.e.,it assesses changes in engine perform-ance not absolute performance levels.This makes it possible to track performance of a particular engine against itself (say from when the engine was installed on an aircraft)or how a particular engine has changed relative to some production level or ?eet aver-age.However,the base reference point is chosen,valuable infor-mation can be obtained that can directly impact the maintenance schedule,logistics,and cost.

In order to infer changes in these performance characteristics we must be able to observe discernible changes (from reference)in various engine parameters taken along the engine’s gas path and have a mathematical model interrelating the two sets of varia-bles.The fundamental tenet underlying this approach is that phys-ical faults occurring in the engine (such as blade erosion,corrosion,tip clearance,fouling,etc.)induce a change in compo-nent performance (as modeled by ef?ciencies,?ow capacities,etc.),which in turn produce observable changes in measurable pa-rameters (such as temperatures,pressures,speeds,etc.).Through inverse relationships,it should be possible to estimate the compo-nent shifts responsible for the measurement shifts observed that in turn provides information needed to address the underlying physi-cal fault(s).Typical parameters that could be obtained along the engine’s gas path form a subset of those depicted in Fig.2.

As already mentioned,this form of analysis concerns itself with changes in performance,and the dependent data are changes in the (observed)measurement parameters from a prescribed refer-ence.The reference was typically some manifestation of a so-called nominal engine (e.g.,this could be a production engine standard or a ?eet average of the user’s overhauled engines or the speci?c monitored engine at installation).Whatever the case,it customarily involved the use of an engine “model”that would be normalized (empirically)in some manner to re?ect one of these choices of reference.In the early days of GPA,the model

was

Fig.1Gas path analysis

principle

Fig.2Gas path measurements [7]

051201-4/Vol.136,MAY 2014Transactions of the ASME

merely a table of parameter values at different engine power lev-els (e.g.,EPR or corrected fan speed,N1c2),tabulated at a speci?c altitude,Mach number International Standard Atmosphere (ISA)condition,and interpolated to match the engine power level of the observed data.Before a comparison could be made,the observed data would be corrected to the ISA condition represented by the reference baseline,as well as for speci?c altitude-Mach number (Reynold’s index)and installation effects.It was then possible to compute a d between the observation and the reference value on a parameter by parameter basis.

The data collected for this exercise were typically steady-state,cruise data.When stable conditions were realized,10–60s of data would be collected and the sample time-averaged,thereby produc-ing a vector of steady-state gas path parameters that were stored on the QAR (or transmitted during ?ight through ACARS)for subsequent gas path analysis.Typically this was done once or twice during ?ight and from these snapshots,a series of (com-puted)d vectors from baseline would be generated in the GPA process and trended over time.The form of these trend charts,affectionately referred to as worm charts ,were generated by the ground processing station and visually inspected for anomalies.An early example is given in Fig.3below;these were later replaced by more modern data plots as computerized processing evolved over time.Before the parameter deltas were plotted,some form of data smoothing was performed to reduce the overall data scatter and enhance the analyst’s ability to visually identify pa-rameter changes.This typically required some form of rolling av-erage and statistical manipulation.The parameters appearing in Fig.3represent processed,smoothed data.

When a suspect trend was identi?ed by the analyst,a visual d -d excursion from the main trend line for each parameter was noted (as illustrated in Fig.4).These parameter DD s were then com-pared to a preassembled list of parameter shifts (referred to as a ?ngerprint chart )to determine a closest pattern match.Figure

5

Fig.3Early engine monitoring worm

chart

Fig.4Determine magnitude and direction DD of parameter shifts

Journal of Engineering for Gas Turbines and Power MAY 2014,Vol.136/051201-5

illustrates an abbreviated?ngerprint chart that would be used in the analysis.In the case depicted in Fig.4,it would be deemed that the HPT module had deviant performance and a hot section borescope would have been ordered.Depending on the damage observed during the borescope inspection,a decision would be made as to whether maintenance was required before the engine could return to service.

It should be evident that the process described in the foregoing paragraphs was labor intensive and could be quite subjective depending on the analyst’s experience and competence.These concerns were the catalyst for an attempt to automate the process and remove the subjective uncertainty.The?rst to provide a math-ematical framework for this process in the early1970s was Lou Urban from Hamilton Standard.These?rst attempts were aimed at processing both test stand and?ight acquired data snapshots and provided a?exible and generic framework that continued to evolve through the1980s and1990s(and even today).In the next section we will present an abbreviated description of this mathe-matical framework.

Gas Path Analysis:Mathematical Foundation.As previously stated,the goal of GPA is to infer changes in engine module per-formance health on the basis of observations taken along the engine’s gas path.It is a mathematical technique that estimates individual module and sensor performance shifts from any speci-?ed set of engine measurable parameters,such as temperatures, pressures,rotor speeds,etc.,through the aerothermodynamic rela-tionships that exist between them.For illustrative purposes we will evolve the model by?rst considering a simple approach.The independent(engine)variables,which will constitute the states of the model,we will denote by xe and will consist of the d compo-nent ef?ciencies,?ow capacities,and areas(D g,D FC,D A).For example,we might choose the following engine fault parameters to model:

x e?

D g FAN

D FC FAN

D g LPC

D FC LPC

D g HPC

D FC HPC

D g HPT

D A4

D g LPT

D A45

2

66

66

66

66

66

66

66

66

66

66

64

3

77

77

77

77

77

77

77

77

77

77

75

?

D Fan Efficiency

D Fan Flow Capacity

D Low Compressor Efficiency

D Low Compressor Flow Capacity

D High Compressor Efficiency

D High Compressor Flow Capacity

D High Turbine Efficiency

D High Turbine Nozzle Area

D Low Turbine Efficiency

D Low Turbine Nozzle Area

2

66

66

66

66

66

66

66

66

66

66

64

3

77

77

77

77

77

77

77

77

77

77

75

(1)

The measurement vector(dependent parameters)will be denoted by Z and will consist of percent of point deltas(taken from reference)of various parameters along the engine’s gas path. An example of this might be the following:

Z?

D N1

D N2

D Wf

D T3

D P3

...

D T5

2

66

66

66

66

66

64

3

77

77

77

77

77

75

?

D Low Spool Speed

D High Spool Speed

D Fuel Flow

D HC Exit Temperature

D HC Exit Pressure

...

D Exhaust Gas Temperature

2

66

66

66

66

66

64

3

77

77

77

77

77

75

?100

eN1measàN1refT=N1ref

eN2measàN2refT=N2ref

eWf measàWf refT=Wf ref

eT3measàT3refT=T3ref

eP3measàP3refT=P3ref

...

eT5measàT5refT=T5ref

2

66

66

66

66

66

66

4

3

77

77

77

77

77

77

5

(2)

These deltas are expressed as percents(in lieu of engineering units)as a matter of convenience.Doing so creates uniformity in the numerical values that will contribute to the overall numerical stability of the(inverse)estimation process that will be described shortly.Although the relationship between the independent and dependent parameters is,in general,a nonlinear one,our basic model will be a linear approximation of this relationship for sim-plicity.At a given engine operating point,this will take the form of a matrix of in?uence coef?cients(ICs).The matrix of in?uence coef?cients will consist of partial derivatives interrelating each in-dependent parameter with each dependent parameter.For exam-ple,the in?uence of fan ef?ciency on N1would be the partial derivative

@N1=N1

@g FAN=g FAN

This quantity expresses the effect of a percent change in fan ef-?ciency on N1in terms of percent.As a linear approximation,we have the relationship

D N1%

@N1=N1

@g FAN=g FAN

D g FAN(3)

If we were to evaluate these partial derivatives for each independent-dependent parameter combination at a speci?c engine operating condition we would generate a matrix of engine in?uence coef?cients.This matrix of partial derivatives are eval-uated at a speci?c steady-state engine power operating condition, holding some independent engine power setting parameter,such as EPR or corrected N1,constant.While there are a number of ways to generate these matrices,the most typical is to use a non-linear aerothermodynamic model of the engine through a pertur-bation methodology[8

].

Fig.5Abbreviated?ngerprint chart

051201-6/Vol.136,MAY2014Transactions of the ASME

The procedure of reducing raw engine data to meaningful per-formance change information requires a plurality of intermediate steps and calculations.The basic gas path analysis algorithm can generally be decomposed into several(general)distinct steps:(1) data normalization,(2)reference model generation,(3)estimation of performance shifts,and(4)diagnostic advisories based on the forgoing calculations.Each of these steps are equally important and crucial to establishing the end diagnostic pro?le.The?rst step in the process is to reduce the variation in the raw data informa-tion stream.Variations in gas path parameters due to varying inlet ambient conditions can generally be accomplished by normalizing the raw data.The various parameters appearing along an engine’s gas path,such as?ows,pressures,temperatures,speeds,etc.,vary not only with power condition but also with the ambient condi-tions at the engine’s inlet.Since a change in inlet temperature and/or pressure will contribute to an attendant change in a gas path parameter’s value,it would be dif?cult to characterize the aerothermodynamic relationships between gas turbine engine pa-rameters,(even at a constant engine operating point)unless the ambient conditions are somehow accounted for.This is usually accomplished through the use of corrected engine parameters.The corrected parameter P?(for any gas path pressure,temperature, speed,etc.,P)takes the form

P??

P

H a d b

(4)

where h and d are dimensionless parameters de?ned as

h?T2edeg KT

288:16

;d?

P2epsiaT

14:696

(5)

and a and b are suitably chosen exponents.Further details can be found in Ref.[9].

It should be noted that there are a number of additional consid-erations that have been omitted in this discussion to maintain brevity and simplicity,centering on the normalization of the input data[10,11].Corrections for engine bleed,Reynolds number effect,test cell correlation or nacelle effects(if data were acquired on the test stand or during?ight,respectively),etc.,would also be applied to reduce variation in the computed measurement deltas. We can now construct a simple relationship to model the upper right-hand?ow in Fig.1.In its most basic form,we have the fol-lowing(linear approximate)relationship:

z?H e x et/(6) where Z is the m?I vector of measured parameter deltas,x e is an n e?1vector of(engine module)fault deltas,H e is an m?n e ma-trix of engine fault(in?uence)coef?cients,and h is an m?1ran-dom vector representing the nonrepeatability inherent in the measurement process.Note that this simple formulation assumes that only module performance shifts(plus noise)are responsible for the observed parameter https://www.doczj.com/doc/e016013987.html,ter we will amend this formu-lation to include other faults(e.g.,sensor faults).

For this formulation of the problem,we can consider three possibilities:

(1)n e?m number of unknowns is equal to the number of

equations

(2)n e

equations

(3)n e>m number of unknowns is greater than the number of

equations

For the?rst possibility(n e?m),the matrix H e is square,and if it has proper rank,can be inverted to provide an estimate of the fault D s,^x e,i.e.,

^x e?Hà1e Z(7)For the second possibility(n e

^x e?eH T e H eTà1H T e Z(8) where H T e denotes the matrix transpose.Now,if in addition,we have knowledge of the measurement nonrepeatability and can compute the measurement covariance matrix cov(h,h)?R,we could easily generate a weighted least squares solution for the engine faults,i.e.,

^x e?eH T e Rà1H eTà1H T e Rà1Z(9) In this latter solution,the fault assessments are being determined by weighting the in?uence of each measurement in accordance to the reciprocal of its nonrepeatability.Thus,accurate measure-ments provide a greater weight in terms of their in?uence on the fault estimation then inaccurate measurements.For these three solutions,each of the matrices

Hà1e;eH T e H eTà1H T e;eH T e Rà1H eTà1H T e Rà1(10) form a set of inverse in?uence coef?cients,which illustrates how a percentage point change in each measured parameter affects a percentage point change in each of the engine fault parameters. Through much of the1970s,Eqs.7–9were used to provide the gas path assessment sought in the lower right-hand?ow in Fig.1. It was,however,fraught with problems.The presence of measure-ment error(sensor drift,bias,etc.)as well as apparent measure-ment error induced by model inaccuracies,unmodeled faults, improper data normalization and corrections,etc.most often pro-duced a set of measurement deltas whose error content exceeded the level of performance change detection being sought.In these cases the results obtained from any of these methods could be mis-leading since a measurement error would be erroneously diagnosed as some combination of engine performance fault parameter shifts. To overcome this problem,it would be necessary to amend the ba-sic formulation to include a set of measurement error fault parame-ters.Since there would be a measurement error fault parameter for each measured parameter,we would produce an underdetermined set of equations,bringing us to the last possibility in which n>m. The solution was to amend our basic formulation by introduc-ing an additional vector of(apparent)sensor(measurement d) faults x s[12].

x S?

D N1err

D N2err

D Wf err

D T3err

D P3err

...

D T5err

2

66

66

66

66

66

66

66

4

3

77

77

77

77

77

77

77

5

(11) Augmenting our basic formulation(Eq.(6))as follows:

z?H e x etH s x sth?H

e

...H

s

h i x e

ááá

x s

2

64

3

75th?Hxth

(12) where x??x e..

.

x s T is the concatenation of both the module per-formance deltas and measurement error deltas.The matrix H has

Journal of Engineering for Gas Turbines and Power MAY2014,Vol.136/051201-7

been partitioned into two parts;an m?n e matrix of engine fault in?uence coef?cients(H e)and an m?n s matrix of sensor fault in?u-ence coef?cients(H s).This latter equation relates percent changes in measurement error to percent changes in the(corrected)measure-ment d quantity.The sensor fault IC matrix can be calculated in a manner similar to the calculation of the engine ICs.An approxima-tion can be simply derived as well and can be found in Ref.[13]. While the form of Eq.(12)is similar to that of Eq.(6),we now have an indeterminate situation with the number of unknowns out-weighing the number of equations,i.e.,an information de?cit.The previous solutions given by Eqs.7–9are not possible since the inverse operation in each does not exist.In the late1970s and through the1980s the classical approach used to solve this prob-lem was to formulate a max a posteriori solution that is a simpli?-cation of the well-known Kalman?lter[14].For brevity,we will state the end result,but a complete derivation and discussion can be found in Ref.[15].An estimate for both the module perform-ance shifts and measurement errors can be calculated as follows:

^x?

^x e

^x s

"#

?x0tDezàHx0T

estimate?aàpriori guesstgain?eresidualT

?predictortcorrector(13) As noted,the form of this solution is a predictor-corrector, wherein x0is an a priori guess(predictor)for both the module per-formance and measurement error shifts.For a single snapshot data point analysis,these would typically be taken as zero,whereas if we were trending a time series of snapshot data points,the a priori guess could be the results from the previous analysis,zero,or a combination of the two.The corrector part of the equation,pro-vides the new information inherent in the current data point rela-tive to the a priori information,i.e.,the residual that represents the difference between what you would expect to see for a measure-ment d(Hx0)and the current(measurement)Z,weighted by a gain matrix,D.The gain matrix D,sometimes referred to as the Kal-man gain matrix,is calculated in the following manner:

D?P0H TeHP0H TtRTà1(14) where P0is an n?n matrix(n?n etn s).This is commonly referred to as the state covariance matrix.Its name,suggesting a probabilistic quantity,stems from the(common)derivation of this solution as a max a posteriori solution wherein the state vector x is assumed to be stochastic and normally distributed with mean l x and covariance P0.As noted in Ref.[15],this derivation leads to minimize the following quadratic form to given below:

J?1

ezàHxTT Rà1ezàHxTtexàl xTT Pà10exàl xT

n o

(15)

In this context,we could just as well consider our solution to be a generalized weighted least squares(WLSQ)solution where P0 plays the role of a weighting matrix,prespeci?ed by the analyst. Since it could be argued that the module performance shifts(part of x)are clearly not stochastic in nature,this is not an unreasonable interpretation,and the selection of P0would now become a tuning element to optimize the solution.A complete discussion of this methodology,along with its derivation can be found in Ref.[15]. Despite some shortcomings,which we will discuss in a moment,this methodology has been the foundation for perform-ance diagnostics from the early1980s to the present time[16–19], although it has been amended and expanded over the past three decades to address some of the de?ciencies.While this approach provides a performance health estimate,it is not unique and has some serious shortfalls.It may(or may not)be clear to many read-ers that this solution tends to attenuate the estimated level of the underlying shifts,indeed,if a large measurement bias or module fault(e.g.,foreign object damage(FOD)or domestic object dam-age(DOD))were to occur,the solution would spread that across all of the module performance and measurement error shifts to some degree.This smearing effect has been the topic of discussion in numerous publications,especially those offering an alternate mathematical solution[20–24].Similarly,if an excluded fault such as an engine subsystem element,e.g.,a bleed valve,or a variable geometry actuator,etc.should fail,the entire attendant measure-ment d shift would be interpreted as a combination of module per-formance shifts and measurement error,which would be a totally false diagnosis.Adjoining these additional faults to the state vector x,does not solve the problem either since it makes an already underdetermined problem even more indeterminate.The painful truth is that this problem cannot be solved through mathematical trickery alone but requires that additional information be provided. The myriad of alternate formulations that have emerged in the last 20years,motivated by this de?ciency,work better only when limit-ing assumptions underlying their development are satis?ed.There is no optimal general method.What has emerged as the current state of the art,are systems that leverage different methods in instances where additional information is available to satisfy the assumptions underlying these methods.This fusion of algorithmic methods will be discussed in a later section on current practice. Onboard/Off-Board Applications.As the reader has no doubt already surmised,much of the analyses described above(with the exception of the ACMF reports)were performed off-board using steady-state data.This was driven primarily by the computation and storage capabilities of the onboard electronic hardware (AIMS and FADEC)available in the1970s through the1990s. CPU capability for?ight hardware typically lags what is available on the ground by a decade or more due to environmental issues (temperature,vibration,and radiation)as well as certi?cation con-cerns.This dramatically reduced onboard diagnostic capacity compared to what is available today.Sensors were also limited. Basic gas path sensors were always available due to their use in controlling the engine and with time,additional gas path sensors were added for EHM purposes since they were relatively inexpen-sive and fairly reliable.This explains the early emphasis on gas path diagnostics and its continued longevity.Vibration is another area of concern for turbomachinery.Accelerometers were not as abundant on aeroengines(as gas path sensors),but it was not un-usual to monitor vibration levels on the fan and turbine modules (and sometimes the main gearbox).The signal bandwidth for vibration is considerably higher at2–5k Hz,as opposed to the gas path at5–20Hz,which limited the type of analysis possible (onboard)in the early years due to CPU https://www.doczj.com/doc/e016013987.html,rmation contained in the frequency content of these signals could not be processed onboard and storing the information for off-board anal-ysis was limited due to storage constraints.Consequently,this form of mechanical monitoring was restricted to primarily captur-ing average vibration amplitudes and comparing them to limits to alert on any exceedance that might occur.The exception was engine test stand runs,either as part of a new production or post overhaul acceptance test,or the less frequent inbound test-as-received(TAR)engine runs,where storage and computer through-put were adequate to do the job.

The engine oil system(also prone to many malfunctions)suf-fered a similar fate.Flight required measurements such as main oil temperature(OT),oil pressure(OP)(and later oil?lter d pres-sure)were monitored in the same fashion to insure that these quantities did not exceed limits.Oil quantity(OQ)was also meas-ured but was fairly unreliable due to the movement of oil in the main oil tank due to engine operation and?ight maneuvers.OQ was typically performed after?ight and periodic testing of the oil composition was performed on the ground through laboratory analysis(spectrographic oil analysis procedure(SOAP)).More so-phisticated monitoring of the oil system would eventually emerge

051201-8/Vol.136,MAY2014Transactions of the ASME

in the2000s with oil debris monitors(ODM)and oils condition sensors(OCS)but would not become prevalent until very recently (more on that later).

As a?nal note,most of the diagnostic analysis discussed so far has been performed using steady-state data acquired during com-mercial cruise?https://www.doczj.com/doc/e016013987.html,parable analysis capability for transient engine data,albeit a more likely state for things to break,has been more elusive.Spawned perhaps by the early successes garnered by airline users of the steady-state GPA[25–28],some researchers turned their attention to the general problem of processing tran-sient engine data.Among the?rst to publish in this area were Merrington[29],Glenny[30],and Kwon et al.[31]in the mid 1980s,but the research never gained momentum,presumably due to the constraints already mentioned.It was an area that largely remained dormant until the last decade or so as computer technol-ogy,engine modeling,and algorithmic methods became more mature.We will revisit the topic of transient diagnostics in the future trends section of this paper.

In the next section,we will review some of the progress made in the late1990s through the current time.In particular,we will explore some of the advances made in GPA and mechanical sys-tems diagnostics and explore the potential of information fusion. Current Practice

Cost Benefit—A Changing View.As noted in the Introduc-tion—Cost Bene?t section of this paper,diagnostics was primarily a tool used in the aftermath of a failure or to preempt an impend-ing failure of a rotating assembly or engine subsystem.The general health of the engine was,for the most part,maintained through per-iodic maintenance scheduled on operational time or?ight cycles. The monitoring practices and diagnostic developments in the previ-ous decades,as well as the strong motivation on the part of the end user to reduce overall life cycle cost of the engine,drove the indus-try to abandon purely scheduled maintenance practices to mainte-nance performed on condition.The OEMs who primarily made their pro?t through the sale of spare parts required for the scheduled maintenance,adapted to this change by offering?eet management programs(FMPs)that provided a fee-based service to the airlines, at a?xed price per?ight hour,and assumed responsibility for the maintenance of the engine.This power by the hour strategy largely traded spare part pro?t for pro?t derived from services.The longer an engine could remain in service on wing,the larger the pro?t would be in theory.Diagnostics and prognostics now became tools for the OEM to mitigate risk of unscheduled events and failures, occurrences that would bring penalties and hence reduce pro?t. EHM systems are intended to improve the business propositions of engine applications by providing:(1)marketable capabilities that make the product more attractive to the end customers,(2)a means to reduce life cycle costs for the end user,and(3)a means of?nancial exposure reduction for a?eet management services provider.These perceived bene?ts helped fuel an almost exponen-tial growth of EHM related research and development within aca-demia,government,and industry in the mid1990s to the present. Many small businesses were spawned whose primary focus evolved around diagnostics and prognostics and related areas. This growing interest provided a rich variety of new algorithmic approaches to address de?ciencies in the incumbent gas path and mechanical diagnostic systems.It also introduced new sensing technologies that were solely intended for EHM purposes,includ-ing inlet and exhaust debris monitoring,oil debris and condition monitoring,blade tip clearance and blade health monitoring,high frequency vibration monitoring,and emissions monitoring to name a few.In what follows,we will(brie?y)discuss these devel-opments along with the processing and algorithms rendered to transform their raw signals into meaningful EHM information. Extending Gas Path Analysis.As we have seen in the previous section,the matter of determining module performance shifts poses certain mathematical dif?culties.Indeed,with an undetermined system,there cannot exist a mathematically unique solution.The pursuit for greater estimation accuracy can only be achieved through the incorporation of more information regarding the performance state of the https://www.doczj.com/doc/e016013987.html,rmation abounds,how-ever,most of it is heuristic in nature and cannot be applied directly in a quantitative fashion without considerable effort.A few examples of this type information might be:

?Over time,component performance degrades,not improves,i.e.,

D ef?ciencies and?ow capacities tend negative,not positive.?The existence of similar gas path shifts in aircraft companion engine parameters might provide corroborative evidence for com-mon instrumentation problems(e.g.,altitude,Mach,TAT,etc.).?Nongas-path information,such as engine vibration,inlet and exhaust debris monitoring sensors,etc.,might allow a differ-ent perspective on the type of underlying faults that should be assessed.

?Measurement nonrepeatability may differ(for a particular engine)from the assumed(?xed)variances for that engine model installation.

This list is by no means exhaustive.It is this author’s opinion that the fundamental improvements to be realized in GPA accu-racy will be achieved by leveraging this type of information wher-ever possible.The predictor-corrector nature of the estimation algorithm inherent in the Kalman?lter approach,described above, can provide one mechanism for incorporating additional informa-tion in the sense that a good prediction mitigates the estimation accuracy by reducing the(diagnostic information)burden on the limited gas path measurements(corrector)that are available.The use of a constrained Kalman?lter(targeting the?rst bullet) [32,33]provides an additional but marginal improvement. Another aspect of this general information-leveraging problem comes by way of inspecting a time series of points rather than just a single point in isolation.It is concerned with the difference between gradual deterioration and a rapid deterioration,in the temporal sense,and has provided a partial solution to the smearing effect discussed in the Gas Path Analysis:Mathematical Founda-tion section.

Separating Gradual and Rapid Deterioration.Engine per-formance changes can manifest themselves in one of two ways:(a) gradual(long-term)deterioration or(b)rapid(short-term)deteriora-tion.These may affect module performance changes or can be changes in performance of engine subsystems such as bleeds,cool-ing?ows,variable geometry mechanisms,etc.The methodology that we will describe momentarily involves breaking up the general GPA solution described previously into several steps:

(1)estimating module performance and measurement error

shifts via GPA method

(2)perusing the measurement d time series to detect any rapid

and persistent shifts

(3)If a rapid and persistent shift is detected in the time series,a

single fault assumption is made and a separate analysis con-sistent with this assumption is made to identify the underly-ing fault

(4)applying appropriate accommodation logic to reconcile the

analysis made in step1with the results from step3

The single fault assumption made in step3is a concession in the form of heuristic information based on experience,i.e.,it is improbable that more than one failure occurs at the same time. The methodology de?ned below attempts to leverage this infor-mation allowing for both of these processes to operate in concert with one another,automatically,and without corruptive interac-tion.It might be considered an example of algorithm fusion driven by observational information.

A cartoon depicting gradual and rapid“deterioration”as viewed from its effect on an observed measurement d is given in Fig.6 below[34,35].

Journal of Engineering for Gas Turbines and Power MAY2014,Vol.136/051201-9

The ?gure indicates a rapid deterioration event occurring at dis-crete time i and j .If we focus on a time series plot surrounding the ?rst event,we might observe (hypothetically)the situation depicted in Fig.7for a given measurement delta.

The area in the plot that is shaded in pink,indicates the obser-vational emergence of an event,i.e.,rapid deterioration of a mod-ule or failure (full or partial)of an engine subsystem (such as a bleed)or instrumentation fault.The portions of the plot shaded in yellow signify the effect on the measurement d due to long term (gradual)component degradation.Recall,that each point in this plot would signify a single (averaged)snapshot data point taken from steady-state cruise ?ight and the time axis is measured in ?ights (i.e.,days rather than seconds).Thus,such a plot would gradually emerge with each subsequent ?ight and that a decision would have to be made as each data point was processed.In this hypothetical plot,there are four data points depicted in the pink area.The event detection problem (in step 2)must weigh the con-sequences of a false alarm versus the consequences of waiting too long to be assured that an event has actually occurred versus a set of rouge points (outliers).This is the general problem of determin-ing persistency.The detection methodology employed must mini-mize false alarms due to outliers while reducing the risk of catastrophic failure that might occur on subsequent ?ights if the underlying fault was critical and persistency was not established fast enough to allow its determination.This general topic,some-times referred to as anomaly detection ,has attracted much research in the past decades and the literature abounds with a vari-ety of proposed methods [36–39].

Once a determination has been made that an event has occurred,the next step is to identify its nature (step 3).One approach is to assume that a single fault has occurred.The rationale is that it is more likely that only one engine related item has failed (fully or

partially)than several over the (relatively)short period of time needed to make the detection.This is an assumption,but a very powerful one.It effectively transforms our underdetermined esti-mation problem to one that is now overdetermined in the follow-ing sense.A prede?ned list of engine-related faults (that have observability within the engine’s gas path)can be assembled from FMEA and historical information and the thermodynamic signa-tures (i.e.,in?uence coef?cients)can be computed for each of these.The problem now becomes the familiar pattern matching exercise,wherein each signature from the list is compared,one at a time,to the observed temporal delta-delta shift that was observed for each gas path parameter across the event window (i.e.,a measurement DD ).The single fault from the list that matches most closely is proclaimed as the identi?ed fault [34,40].While this appears to be a rather straight forward task,as always,the devil is in the detail.One of the details deals with am-biguity.Since the prede?ned single fault list could be consider-ably large (much larger than the number of parameters in our measurement suite),it would not be surprising that two or more single faults could have signatures that were very similar,wherein the difference between them is contained in the noise band of the measurement deltas.In this situation the aforementioned analysis would not be able to unilaterally distinguish between them,There-fore,it must be recognized beforehand which faults fall into cer-tain ambiguity groups of faults and a given isolation would only be valid to the level of such an ambiguity group [41].As way of illustration,consider a high pressure turbine (HPT)fault caused by severe blade erosion opening the tip clearance that would man-ifest itself as a decrease in HPT ef?ciency.Many high bypass tur-bofan engines are equipped with active clearance control (ACC),sometimes referred to as turbine case cooling (TCC)where,(dur-ing steady-state operation),air is bled from the compressor and circulated through tubes around the external casing of the HPT.As the HPT casing undergoes thermal expansion,tip clearances increase and the TCC is designed to cool the case,restricting its expansion.If the bleed valve that governs the TCC ?ow should fail,the expansion would not be governed and the measurement d signature observed would be very close to that assigned to an HPT blade problem.Thus,within the typical noise band of the mea-surement deltas,these two faults would reside in the same ambi-guity group.Since these two faults carry a signi?cant difference in severity level,i.e.,a TCC fault will increase fuel burn,lower EGT margin,and consume life on life-limited parts (LLPs)at a greater rate if unattended,?ight safety is not immediately impacted,whereas an HPT fault unattended might lead to a DOD event and a catastrophic failure.Clearly what is needed is addi-tional independent information to resolve the ambiguity.In this particular example,such information already exists.As mentioned previously,takeoff (T/O)data are also collected (for ACMF reporting).Another piece of information is that TCC is only active during steady-state operation,i.e.,not during T/O.Thus,if fuel burn is observed to be substantially higher and EGT margin lower than on previous ?ight T/Os then it cannot be a TCC fault and,hence,must be an HPT fault.Likewise if the fuel burn and EGT margin are essentially the same as the previous T/O point then it must be the TCC at fault.This example illustrates the power of in-formation fusion and the current trend in diagnostics systems that use different algorithms for different situations and utilizing both measured and domain knowledge information to arrive at a solution.

Mechanical Systems Diagnostics.While gas path diagnostics continues to be an important topic in support of overall engine health monitoring,it is certainly not the only area to be consid-ered.EHM systems have expanded to include the monitoring of mechanical systems and components such as bearings,gearboxes,lubrication systems,and blade health.This consideration was driven both by the signi?cant percentage of engine problems asso-ciated with these components,as well as the introduction of

new

Fig.6Gradual versus rapid

deterioration

Fig.7Trending and events

051201-10/Vol.136,MAY 2014

Transactions of the ASME

propulsion health management(PHM)-speci?c sensors that pro-vided the requisite information to enable component health moni-toring.In this section,we will brie?y discuss some of these sensors and how they are used.

Vibration.Bearing degradation,ultimately leading to failures, results from unexpectedly rapid and often asymmetrical wear of the rotating parts(e.g.,balls,rollers,etc.)and the surfaces against which they move(i.e.,races).Likewise,gear degradation appears as uneven wear on the surface of the gear teeth,often resulting in audible“chattering.”The most common means of sensing bearing and gear degradation is via vibration monitoring,[42].Typically this is accomplished by mounting single or multiaxis accelerome-ters in the vicinity of the bearings and on the gearbox.Mounting the sensor closer to the source of vibration is preferred since it provides a“more distinct”signal,but often this is not possible due to temperature constraints on the sensor and other factors.Often the accelerometers are simply mounted on the engine case.

The frequency range for the accelerometers depends on a num-ber of factors.One consideration is the attributes of the signal to be sensed,which is generally a function of engine speed.Another is the operating environment,the background noise,which may obscure the signal.A third consideration is how the signal will be used,whether it will be used to provide a“high vibes”indication or used to perform real-time onboard analysis of the signal to aid in fault isolation.One of the classical methods of dealing with vibration signals and the noise inherent in these signals is to per-form what is known as time synchronous averaging(TSA).Here the signal is sampled using a pulse that is synchronized with the signal(typically a tachometer once per rev pulse)and then aver-aged.This process will gradually eliminate any random noise present.A subsequent fast Fourier transform(FFT)can be per-formed to allow detection of abnormal vibration peaks.This is useful in identifying the source of vibrations in bearing and gear systems.Knowing the gear ratios provides aids in determining the presence of chipped gear teeth by observing peaks at the appropri-ate frequency harmonic of the driving spool.

Lubrication System.Sensors generally used to monitor the lubrication system include measures of oil quantity,temperature, delivery pressure,debris,and degradation.Measure of speci?c components,such as the pressure drop across an oil?lter,may provide indications of eminent bypass condition and the need for ?lter replacement.Oil debris monitoring(ODM)is a relatively new technology that was?rst introduced on military aeroengines and has been found to be a highly effective means of early bearing degradation detection.By monitoring the liberated metal particles in the oil it is possible to identify and trend degradation well in advance of a bearing failure[43,44].The use of an ODM can eliminate the need for periodic inspections of the magnetic chip detectors that accumulate ferrous particles in the oil stream.The inspection of a chip detector can be performed only upon indica-tion of abnormal debris by the ODM.Current ODM sensors are capable of detecting nonferrous particle as well,thereby extending the coverage typically provided by chip detectors.As we have seen in our discussion on gas path diagnostics,integrating infor-mation can improve our detection capabilities and subsequent fault identi?cation.Integrating the ODM and vibration informa-tion has likewise been shown to be useful in improving the proba-bility for detecting damage in gear systems[45,46].

Debris Monitoring.Another method for detecting accelerated deterioration in gas turbine engines caused by blade erosion,tip rubbing,and combustion anomalies was introduced in the late 1970s through the analysis of particles present in the engine’s gas stream.This took the form of specialized electrostatic probes that were placed downstream of the engine’s exhaust along with ground test equipment to analyze the signals during a ground run [47].These electrostatic emission monitoring systems(EEMS)would monitor the(electrostatic)charged particles in the engine exhaust,establish a normal engine baseline for the signal over the engine power range,and once established could then be used to detect increased levels of charged debris that could be due to blade rubs,abraded seal wear,blade coating liberation,combustor faults,and the like.Testing during the late1970s and early1980s indicated that these devices offered a means to continuously mon-itor the engine and detect the onset of problems that were not oth-erwise observable with traditional gas path analyses.This naturally led to the development of a?ight-worthy sensor that could be mounted,unobtrusively,in the exhaust nozzle of the engine,accompanied by onboard data acquisition and signal con-ditioning electronics to allow continuous monitoring during?ight. The resultant engine distress monitoring system(EDMS)sensor and electronics,developed by Stewart Hughes,Ltd.[48,49]were deployed on several military aircraft in the early2000s to gain ex-perience and mature the technology.

The original success of exhaust debris monitoring led naturally to the development of similar sensing systems to be place in the inlet of the engine to monitor for potential foreign object damage (FOD).Aeroengines are susceptible to the ingestion of a wide va-riety of materials during taxi,takeoff,and during?ight in the form of dirt,sand,dust,small objects on the tarmac(or?ight deck),and volcanic ash as well as an occasional bird or two. Some of these are potentially damaging to the turbomachinery depending on the size and/or quantity of the material ingested.It is,therefore,important for the system to be able to discriminate between potentially damaging and nondamaging debris in order for the system to be a viable contributor to the overall PHM sys-tem.The sensor system and its attendant signal conditioning elec-tronics,(referred to as the inlet debris monitoring system(IDMS)) is based on the same principles as the EDMS and consists of two concentric rings set a?xed distance away from each other in the inlet of the engine.The use of two rings allows not only for the detection of debris but(in the case of discrete debris)supports the calculation of the speed of the particle ingested and,hence,its energy level to aid in identifying its potential for damage.As the reader might already surmise,the effectiveness of these sensors comes from their collective use and the degree to which sensor/in-formation fusion can be employed.Correlating IDMS,EDMS, vibration data,and traditional gas path monitoring provides a more complete picture to determine a damaging event from long-term deterioration effects as well as identifying the particular modules impacted[50,51].

Blade Health—Usage-Based Lifing.Many engine compo-nents life-limited parts(LLPs)require inspection after some regu-lar intervals and must be replaced well before all usable life has been https://www.doczj.com/doc/e016013987.html,ponent li?ng algorithms that accurately track component usage can allow life-limited parts to remain on wing until their life is more nearly consumed.Safety is actually increased by accurately knowing life consumption on an engine-by-engine basis,while at the same time reducing support costs. The component li?ng system can be implemented either as part of an engine-hosted or ground-based EHM system,with the ground-based implementation supported by on-engine generated,com-pressed,and downloaded data.

Traditional component li?ng methods used to determine the total amount of life consumed are based on either engine operat-ing time or total accumulated cycles,which are correlated to low cycle fatigue for each life-limited component.The determination of these cycles assumes that every engine on every aircraft will be used in a manner consistent with expected?ight pro?les.Mainte-nance of life-limited parts is scheduled based on a very conserva-tive assumption of engine usage in order to maintain required safety levels.In reality,engine usage often falls short of the stand-ard mission pro?les.Yet in order to ensure safety,parts are replaced early,per the conservative schedule.As a result,due to a lack of actual usage knowledge on an engine-by-engine basis,

Journal of Engineering for Gas Turbines and Power MAY2014,Vol.136/051201-11

most components on most engines are replaced with signi?cant life and,hence,value remaining in the part.By implementing methods to accurately track actual component life usage,there is the potential to greatly reduce the width of the usage uncertainty distribution,allowing the scheduled maintenance point to be moved much closer to the center of the distribution.This greatly reduces the number of parts that are scrapped with life remaining, while maintaining or even increasing safety levels.

Advanced life estimation techniques use validated models derived from component design tools to allow more accurate tracking of component life used and estimation of life remaining. The algorithms accumulate usage by tracking time-variant changes in parameters such as pressures,temperatures,and speeds to drive the li?ng algorithms,which are in essence real-time mod-els of critical component failure mechanisms.Based on how a family of engines is predicted to be used for a particular vehicle application,and based on how they are actually being used,the life usage estimation can be transformed into a life-remaining estima-tion.Life remaining for each modeled component is accumulated and made available to the engine?eet management system,allow-ing the scheduling of necessary overhaul based on actual usage. Information Fusion Potential.Throughout the paper we have already commented on the potential analytic bene?ts that can be afforded through information fusion.Aircraft gas-turbine engine data are available from a variety of sources,including onboard sensor measurements,maintenance histories,and component models.An ultimate goal of PHM is to maximize the amount of meaningful information that can be extracted from disparate data sources to obtain comprehensive diagnostic and prognostic knowl-edge regarding the health of the engine.Data fusion is the integra-tion of data or information from multiple sources to achieve enhanced accuracy and more speci?c inferences than can be obtained from the use of a single sensor(or information source) alone.The basic tenet underlying the data/information fusion con-cept is to leverage all available information to enhance diagnostic visibility,increase diagnostic reliability and reduce the number of diagnostic false alarms.

The information available,much of which we have already mentioned,includes the following.

Engine Gas Path Measurements.These measurements consist of some subset of interstage pressures and temperatures,spool speeds,fuel?ow,etc.Gas path analysis(GPA)itself can be viewed as a form of information fusion in that consideration of these parameters individually provides far less information than when considered collectively to form signatures that can be corre-lated to known fault categories.

Oil/Fuel System Measurements.These measurements consist of various oil system temperatures,pressures,fuel temperature, and delivery pressure.Advanced sensors indicating oil quality,oil debris monitoring sensors,and oil quantity measurements may be available.

Vibration Measurements.Some form of vibration monitoring is typically performed on most engines.This monitoring is usually on the low spool to measure fan and low-pressure turbine(LPT) vibration,but may include high spool vibration probes,as well as speci?c bearing and gearbox vibration measurements. Structural Assessment Sensors.These sensors aid in assessing structural integrity of the engine.Examples include inlet debris and exhaust debris monitors,acoustic sensors,high bandwidth vibration sensors,multiaxis vibration,and blade tip clearance monitors. FADEC Codes.The full authority digital engine control (FADEC)performs a myriad of performance tests on signal condition and?delity.Cross channel checks can aid in determining whether or not a main engine sensor is drifting,going out of limit, or failing.Checks on bleed valves,active clearance control,and variable geometry can provide independent information regarding engine health and the health of various engine subsystems. Onboard Engine Models.Accurate engine models embedded within the FADEC or residing within a dedicated PHM hardware unit can be used to generate virtual engine measurements to aid in detecting faulty engine instrumentation or con?rming degraded engine performance.Self-tuning onboard real-time model (STORM)systems have been developed for this purpose[52–54]. These models adapt themselves to the changing conditions observed by the engine’s measurement suite,thereby providing virtual sensors that can be used to estimate engine module degra-dation.These types of systems can be thought of as a generaliza-tion,for full?ight streaming data,of the techniques previously described for snapshot data wherein the onboard model becomes the reference level from which degradation(estimated by the Kal-man?lter observer)is tracked.This is a broad analogy,and there are considerable differences in the actual methodology and imple-mentation in order to perform the same functionality in a real-time fashion.This will be explored in more detail in the discus-sion on future trends.

Maintenance/Analysis https://www.doczj.com/doc/e016013987.html,rmation regarding the performance disposition of the major modules that comprise the engine can potentially be used as a priori information to support the identi?cation and estimation of performance changes within a module performance analysis(MPA)program.

Companion Engine Data.On multiengine aircraft,informa-tion from the companion engines might be used to provide addi-tional independent con?rmation of instrumentation problems and engine events.For example,it is not uncommon to perform an EGT divergence check where(snapshot)EGT D values are com-pared with those of a companion engine to note if one engine’s value is diverging from the other.This can offer additional infor-mation to aid in determining a speci?c engine event(or sensor problem)from a systemic problem such as a TAT error or base model error that would affect both engine D values.

Negative Information.This pertains to a reasoning methodol-ogy more than an actual source of information.Negative informa-tion[55]constitutes conditions that were not present but would, or should,have been perceived under the hypothesis that a certain fault scenario exists.In mathematical parlance,it is referred to as proof by contradiction.For example,if active clearance control (ACC)was not enabled(i.e.,there was a faulty operation)then exhaust gas temperature(EGT)should increase.If EGT was not observed to increase,then the original assumption is probably false(i.e.,ACC must be working properly).This type of informa-tion would best be employed in an expert systemlike structure that governs the overall analysis and processing of the engine data. With such a wealth of potential information,the manner by which to combine or fuse information for the stated diagnostic goal must be decided.In general,data can be fused at different levels;for example:

?Sensor level fusion:this is where several sensors measuring the same or correlated parameters can be combined(e.g., rake of exhaust gas temperature sensors).

?Feature level fusion:this combines analysis information resulting from independent analysis methods(e.g.,compo-nent performance changes and event detection).?Decision level fusion:here diagnostic actions can be com-bined,(e.g.,damage assessments,maintenance advisories). The level of fusion that is suitable will depend on many factors, including available sensors,models,analysis algorithms,data

051201-12/Vol.136,MAY2014Transactions of the ASME

monitoring and recording speci?cs(continuous versus discrete data),and computing platform.For engine diagnostics,different levels of information fusion will be required depending on whether the system is for a commercial or military application.In military applications,dedicated PHM systems using independent engine monitoring and analysis hardware and/or direct FADEC involvement are not uncommon.In these scenarios,data are col-lected and analyzed in real time,onboard the aircraft during?ight. In commercial applications,as we have already noted,much of the data collected is discrete in nature(several data points per ?ight,typically at takeoff and cruise).This information is down-loaded to ground-based computer systems for subsequent analysis and trending.In addition,advanced sensors are more common in the military environment than in commercial applications,where a compelling cost bene?t argument must overcome the general tendency to minimize sensors and data collection hardware.Gen-eral generic systems have been devised to leverage multisource information at all levels to enhance diagnostics capability and reduce potential false alarms[56,57].

Models.Models form an integral part of an engine diagnostics and prognostics strategy.They come in many forms,i.e.,physics-based,empirical,and hybrid(the mixture of the two)representing the full engine as well as engine speci?c components and accesso-ries,e.g.,actuator systems,bleed systems,pumps,generators,etc. They provide information,either static or dynamic,relative to the system under consideration in terms of either establishing a refer-ence level(e.g.,nominal performance)for measured quantities from which actual measured parameters can be compared and re-sidual differences(D signatures)can be calculated as well as pro-viding information about unmeasured quantities of interest,e.g., in-?ight thrust,thrust speci?c fuel consumption(TSFC)surge margin,etc.Models leverage our a priori domain knowledge about the operation of the engine(or subsystem)and provide the foundation upon which diagnostic and prognostic systems are built.As such,it should be no surprise that the?delity of the model being used is of paramount importance and de?ciencies and errors in the model will directly translate to de?ciencies and errors in the EHM systems that employ them.

Physics-based models are models based on?rst principles or models directly derived from them and constitute the majority form used in current EHM systems and in PHM research that focus on performance diagnostics.Engine manufacturers maintain such models for their engines that they consider proprietary,how-ever,similar model types have been developed for generic(?cti-tious)engines(such as a high bypass twin spool turbofan of a certain thrust class)to enable research by academia and small business third parties[58,59].These are typically nonlinear com-ponent driven aerothermal models,but depending on the particu-lar application it is sometimes necessary to employ a lower order approximation derived from these decks that do not have conver-gence issues(iteration)and have less CPU and memory require-ments.State variable models(SVMs)are piecewise linear representations and have been a popular derived model for use in real-time onboard applications since they have a(relatively)small execution time and consume minimal memory.While they match fairly well(to their parent nonlinear deck)at steady-state points, they are less faithful during transients and often require additional tuning to align their output to the parent model[52–54]. Because of engine-to-engine variation(inherent in production or overhaul),the use of models as a reference level for diagnostic and prognostic purposes typically requires that we further align the model either to the speci?c engine being monitored or to the average level of a user’s?eet to de?ne the appropriate reference level from which deterioration is to be tracked.This initialization of the model can take many forms but is largely accomplished through the use of correction factors applied to the base model that are empirically derived from direct engine data observations, taken during engine test cell acceptance test runs,or from ?ight data acquired when the engine is introduced(or returned to) revenue service,depending on the speci?c application.This is a critical element in any EHM system to mitigate the in?uences of engine variation,instrumentation biases,and base model in?del-ity,all of which can corrupt our diagnostic and prognostic calcula-tions.We will discuss this in more depth for a speci?c application in the next section.

Future Trends

Engine health management can be considered as a collection of capabilities from which building blocks can be drawn to create customized architectures that best meet individual user needs. Both engine-hosted and ground-based elements are viable tactics, not as competing EHM approaches but rather complementary fea-tures of an overall integrated health management system.Engine-hosted elements generate data from onboard sensors and perform basic fault isolation and prediction,supporting on-wing mainte-nance,while the ground-based elements support long-term degra-dation trending,providing planning information that can be used by aircraft?eet managers(Fig.8).

To frame our discussion of future PHM systems,we will brie?y explore three(interrelated)areas,emerging PHM-speci?c

sensor

Fig.8EHM:the big picture

Journal of Engineering for Gas Turbines and Power MAY2014,Vol.136/051201-13

systems,information fusion and advanced analytical methods and ?nally the concept of the intelligent engine.These three general areas are driving current research activities,aimed at producing the ultimate future state,fully automated,self-diagnosing systems with prognostic capabilities.This last item is perhaps the ultimate EHM potential,the ability to automatically and autonomously forecast,with reasonable accuracy,an impending failure of a part or subsystem and its time remaining before failure.

PHM Sensors.The last decade has seen increased research in the development of new sensors whose primary role is in the sup-port of PHM.We have mentioned many of these in the preceding section such as oil debris monitors(ODM),oil condition monitors (OCM),gas path debris monitoring systems for ingestion(IDMS) and exhaust(EDMS),blade health(ECS),blade tip timing and tip clearance,high frequency accelerometers and acoustic sensors, exhaust emission sensors,and the like.In addition to speci?c function sensors such as these,new sensor technologies are required to meet the demands of our future engines and the EHM systems that support them.One such area is high temperature electronics.The integration of microprocessing circuits as part of a so-called smart sensor currently exists on today’s engines,e.g., pressure sensors to allow individually calibrated,highly accurate devices.These must reside in a relatively benign environment (such as the FADEC),necessitating that the pressure to be meas-ured be piped to the sensor,adding additional unwanted weight to the engine.The need to closely integrate the electronics with the sensing element drives the need for more rugged packaging and higher temperature tolerance.

Microelectromechanical systems(MEMS)technology is a potential contributor to address some of these integration issues. MEMS provides a path for integrating the signal processing elec-tronics with the sensing element to provide miniaturization,lower cost,higher reliability,and reduced power consumption.For example,standard piezoresistive,piezoelectric,and capacitive pressure sensing elements are currently under investigation in MEMS studies to evolve new sensors.Silicon-on-insulator,silicon carbide,and silicon carbide nitride are some of the new technolo-gies being investigated to evolve a high temperature smart sensor capability.

Thermocouples and resistance temperature devices(RTDs) make up the large majority of temperature probes in use today. Optical pyrometers are also currently used in some applications but to a lesser degree.The area of radiation pyrometry is a nonin-trusive technology consisting of a sensing head(which would be integrally mounted within the engine casing),a?ber optic bundle to carry the infrared signal,and electronics to capture the signal and provide the appropriate signal conditioning.These devices could extend the temperature sensing range to2200 R to measure combustion gases and turbine blade temperatures,which can pro-vide salient information for blade health determination.

Going forward,new sensing principles and technologies will need to advance to meet the needs of future EHM systems and their applications.Active combustion control to mitigate combus-tion instabilities,emission sensing to monitor exhaust NO x and CO x to ensure low emission green operation,active(turbine) blade tip clearance control,etc.are just some of the numerous potential applications.Ultimately,a compelling cost bene?t posi-tion must be established to justify the addition of each new sensor that involves more than just the sensing devices,but all of the sup-porting electronics and analytical methods that transform the raw data to useful information and its potential impact on engine maintenance and engine control[60].The interested reader should consult Ref.[61]for an in-depth account of existing and future sensing systems.

Information Fusion and Analytical Methods.As we have stated repeatedly in this paper,that information is the essential currency upon which an EHM system derives its value.Engine sensors provide just one avenue for potential information.Domain knowledge,heuristics,constraints,assumptions,models,analyti-cal inferences,and general methods for combining these pieces are essential elements to complete the picture.The ability to uti-lize all of the information at our disposal to the fullest cannot be overstated.It is the author’s belief that this aspect is the most im-portant consideration in the evolution of our future diagnostic and prognostic systems.With the reader’s indulgence,we will diverge brie?y from engine-speci?c discussion to consider a pure mathe-matical puzzle to illustrate the idea.

The Impossible Problem.Martin Gardner was a20th century mathematician,best known for his Scienti?c American articles and books on mathematical puzzles.In1979he related a puzzle in the Scienti?c American that he called the Impossible Problem. While there are many renderings of this puzzle,the basic problem is stated as follows:

“Two integers,x and y,both greater than1and whose sum is less than100are secretly selected.The Sum(xty)is given only to mathematician S and the Product(xy)is given only to mathe-matician P.The mathematicians know only these facts.The exchange between them goes something like this:

P:I do not know the values of x and y.

S:I knew that you did not know the values.

P:Now I know the values of x and y.

S:Now I know them as well.

After this exchange it is now possible to determine the two numbers x and y.”

The puzzle is aptly named since it appears that there is hardly enough information to determine anything let alone its solution. This is an example,admittedly extreme,of capitalizing on every piece of information available in the problem statement,not only the factual statements,but assumptions,such as the mathemati-cians perform?awless logical deductions have at their disposal knowledge of elementary number theory(domain knowledge)as well the employment of negative information in their reasoning. For example,mathematician P can factor his product(uniquely) into a product of primes(domain knowledge).It must be the case that this factorization is not the product of just two prime numbers otherwise P would know the numbers.S can determine the parti-tion of her sum,i.e.,how many different ways the sum can be decomposed into the sum of two numbers(domain knowledge).It must be the situation that this partition not contain a case where the sum can be represented by the sum of two prime numbers,oth-erwise S would not be able to make her?rst statement(analysis, negative information and assumption—mathematicians are?aw-less).It is this type of reasoning,coupled with domain knowledge, assumptions,negative information,and the fusion of all of the above with the actual observations(problem statement),that ulti-mately leads to the unique solution.For those readers interested in puzzles,the solution is provided in the Appendix.The relevant message here is that if information from all sources combined with analytical methods to collectively manipulate them,the(per-ceived)need for additional sensors can be minimized.Additional sensors would only be necessary if they acquire unique informa-tion(not deducible from the information fusion)or as a redundant corroboration to reduce uncertainty that may be inherent in our observations(existing sensors and other information). Prognostics.As we have already noted,the functionality pro-vided by an EHM system includes the facility to evaluate the cur-rent health of the system(diagnostics)as well as forecasting the future health of the system and identifying emerging problems (prognostics).As a generality,a prognostic algorithm offers the ability to predict the time in which the health state of the engine approaches a prede?ned limit or threshold(e.g.,when will EGT margin of the monitored engine become zero).It is the ability to predict the future condition of the overall engine,engine

051201-14/Vol.136,MAY2014Transactions of the ASME

component,and/or system of components.In its ultimate form,it is the capability to predict component failures before they actually fail.For life-limited parts,LLPs(such as rotor disks)this trans-lates to determining the remaining useful life of the part so that it can be replaced with minimal waste and minimal risk.This is not an easy proposition and will require considerable advances in cur-rent analytic methods,models,advanced sensor systems,and in-formation fusion to ensure accurate and reliable predictions.

As onboard computing platforms are becoming faster and mem-ory capacity more abundant,there is a natural trend to perform more PHM-related functions on board.The autonomic logistics approach being introduced for the military joint strike?ghter (JSF)[62]is an example of an application of onboard diagnostics and prognostic health management capabilities to eliminate sched-uled engine inspections and rely on on-condition health assess-ments.It would be anticipated,however,that our future system would draw from data and analysis from both onboard and ground based systems so that the information best developed by each plat-form could be leveraged to the fullest extent possible.

One area of particular interest is the intelligent engine concept wherein the PHM system will autonomously determine its health state not only to support line maintenance and depot overhaul lo-gistics but provide salient information to the engine control to per-mit more ef?cient operation,life extension,and increased operational reliability.

The Intelligent Engine.One argument for supporting greater onboard PHM capacity would be the decrease in time latency to detect,identify,and possibly accommodate an incipient fault or deterioration.It could also be argued that there would not be much(time)gain between an onboard analysis versus merely stor-ing the full?ight data onboard and performing the analysis once the aircraft has landed,several hours later(assuming that an ef?-cient data infrastructure was in place to transfer it off-board), unless the onboard information could be accommodated in some manner during?ight.Accommodation could be in the form of alerting the pilot of an identi?ed problem or incipient fault requir-ing him to abort his mission or operate the engine/aircraft in a dif-ferent manner.A more futuristic tactic could involve automatic and autonomous accommodation by the FADEC without pilot intervention.These concepts have been explored in what has been termed the intelligent engine concept[63–70].For the remainder of this paper we will explore this latter tactic.

Integrating an onboard PHM system with direct engine control is both an intriguing and a daunting concept.Historically,and generally speaking,engine control laws are formulated for the nominal engine with enough robustness and margin so that the engine can perform adequately and safely(perhaps not as ef?-ciently)as the engine deteriorates over its life cycle.The intrigu-ing question is would you(and could you)control the engine differently if you had full knowledge of its current health state? What are the consequences in terms of technologies required,sys-tem complexity,reliability,and eventual certi?cation implica-tions?It has been suggested that a model predictive control (MPC),[71]can provide the technology to achieve an adaptive control strategy that can tradeoff performance goals with remain-ing useful life on-wing in the presence of known engine deteriora-tion[64].Before concluding this section we will revisit our previous discussion on models within the context of their usage onboard within an MPC.

Models Revisited.At the end of the previous section on Cur-rent Trends,we spoke brie?y about onboard models and the im-portance of the initialization of these models to conform to the installed state of the speci?c engine being monitored(and con-trolled).It was mentioned that this was a critical element to miti-gate the in?uences of engine-to-engine variation,instrumentation biases,and base model in?delity,all of which can corrupt our diagnostic and prognostic calculations.Within the context of the intelligent engine and MPC,this becomes a fundamentally critical factor.

As a means of illustrating the potential problem of not initializ-ing the onboard model,we consider a segment of?ight data from large high bypass engine.The?ight pro?le in terms of altitude and Mach number is depicted in Fig.9.This segment consists of a climb and low altitude cruise condition.The system consists of a derived piecewise linear state variable engine model(SVM) coupled with a Kalman?lter observer to track component deterio-ration similar to that described in[52,53].This system,known as a self-tuning onboard real-time model(STORM),adapts itself to changing conditions observed in the engine’s measurement suite, providing virtual sensors that can be used to estimate engine mod-ule degradation.The output of the Kalman?lter is a set of health parameter tuners,consisting of a subset of engine component ther-modynamic performance deltas,(typically ef?ciencies and?ow capacities)that are used to explain the differences between the SVM output and the actual observations in the gas path and tun e the model to force alignment.This provides a set of virtual(model generated)sensor values that will track the engine as it deterio-rates.The overall architecture is depicted in Fig.10.

An implicit assumption that is being made is that the SVM will be a reasonably faithful representation of the speci?c engine being monitored.Engine-to-engine variations,instrumentation biases, and model in?delity(0D model for a3D engine)can easily

violate

Fig.9Typical commercial aeroengine?ight pro?le

Journal of Engineering for Gas Turbines and Power MAY2014,Vol.136/051201-15

this assumption and the net effect is that these differences are absorbed by the tuners corrupting their interpretation as true health parameters.What is needed is an initialization process to align the model at installation independent of the tuning so that the output of the Kalman ?lter ceases to be a mathematical artifact and can faithfully track deterioration changes as they occur across the engine’s life.An empirical tuning (initialization)process has evolved to address this need [72–76],wherein an empirically derived difference model is autonomously constructed (onboard)during the initial ?ights after engine installation and mated with the physics model to yield a hybrid representation of the moni-tored engine.The architecture,referred to as enhanced STORM,or eSTORM,depicted in Fig.11,indicates that the empirical tun-ing (initialization)is performed as an additive bias to the output of the onboard model.The advantage of this approach is that it becomes independent of the particular physics model being employed (e.g.,an SVM or a more complex nonlinear component engine model).The consequence of performing this empirical tuning can be seen on its effect on the performance health parameter D s (tuners)for the ?ight segment depicted in Fig.9.Figure 12portrays the health parameters as they appear in the uninitialized system.The mismatch between the engine model and this speci?c engine causes the tuners to become biased and vary during the ?ight seg-ment in order to adapt to the difference and drive the engine pa-rameter residuals to zero (i.e.,to force a match between the model outputs and the engine measurements).The state of the tuners ren-ders them useless for purposes of tracking actual deterioration and as such cannot be effectively used for EHM purposes or as an input to an MPC within the intelligent engine context.Figure 13depicts the tuners after empirical initialization is in force for the same ?ight segment.The tuners are now less erratic and centered around zero.This allows deviations (from installed state)to be tracked over time as the engine degrades as well as admitting greater visibility for the detection of anomalous events and faults (e.g.,FOD,DOD,etc.)as illustrated in the (?ctitious depiction)in Fig.14

.

Fig.10STORM

system

Fig.11Enhanced STORM system

051201-16/Vol.136,MAY 2014Transactions of the ASME

The establishment of a good initialized model to serve as a ref-erence is a critical element for onboard PHM but by no means the only element.Analysis methods,capable of operating in real time,to accurately perform (gas path)anomaly detection and fault iden-ti?cation across full ?ight operation,including transients ,have yet to advance to a satisfactory level.Although proposed in the late 80s,progress in this area has been slow despite a renewed interest this last decade [77–82].This remains one of the notable chal-lenges for future work in performance diagnostics.

Other Challenges.In our discussion we have highlighted many areas that offer challenges that face EHM in the next dec-ade.Several areas we have not discussed are consideration for system veri?cation and validation,software development,and cer-ti?cation.Historically,veri?cation and validation typically begins with simulation trials followed by real data validation.Since EHM deals with abnormal conditions and since seeded fault trials,being prohibitively expensive,are very rare,the real data valida-tion of an EHM system is opportunistic at best.For ground-based portions of the EHM system,it has been possible to assemble an ongoing database of known historical fault cases that can be used to test new algorithms being applied to a new engine application,the ef?cacy of which is made by inference.For onboard systems this is more dif?cult and the veri?cation activities combine simulation,bench testing,and ?ight trials during the engine development cycle.In any event,it is a long,time-consuming pro-cess wherein true validation evolves over the life cycle of the engine.With the trend toward a more intelligent engine,an improved and timely veri?cation and validation processes will need to emerge.

Software development of EHM systems is another area under-going change.Early ground-based systems were considered to

be

Fig.12STORM system

tuners

Fig.13Enhanced STORM system

tuners

Fig.14Empirical tuning effect on anomaly visibility

Journal of Engineering for Gas Turbines and Power

MAY 2014,Vol.136/051201-17

an informational power assist to ground analysts who made the ultimate decisions relative to line and depot maintenance and as such were given the lowest software classi?cation.Over time,the reliance on EHM to direct inspections and prescribe maintenance on-condition has necessitated an ever increasing importance to its software development and classi?cation[83].This is magni?ed for onboard systems especially those that have or will have greater interaction with the control of an aircraft,either through pilot alerting or direct FADEC interaction.In these cases the software is elevated to the highest classi?cation of being?ight critical.This latter classi?cation has dramatic certi?cation implications.Indeed the whole area of using intelligent methods,(e.g.,neural networks or any method involving adaptation)is an area that is just begin-ning to receive much needed attention[84–87].

Summary

We have provided a broad overview of engine health manage-ment as it has evolved over the last four decades,commenting on the cost bene?t drivers,monitoring and data acquisition methods, analytic methodologies,strengths,and weaknesses and concluded with a brief discussion on areas of research being explored as we move forward into the decades ahead.

There are several central themes emphasized throughout this paper.The?rst is that engine maintenance,whether scheduled or unanticipated,is very costly and the prime goal and bene?t of EHM is to reduce cost and mitigate risk in moving to a totally condition-based maintenance strategy.At its inception,EHM was primarily a monitoring activity,and we have provided a brief his-torical account of how gas path performance monitoring,analysis and reporting evolved to its present day capability.

We have stressed the importance of information in the success-ful execution of diagnostic and prognostic processes and that it goes beyond what we measure with sensors to include our domain knowledge,assumptions,constraints,heuristics,etc.Most impor-tantly,we need to combine the disparate sources of information at our disposal(fusion)to extend our diagnostic coverage while min-imizing the need for additional sensors.Making the most use of every piece of information at our disposal is pivotal to evolving EHM and remains one of the future challenges.We have also dis-cussed the use of PHM-speci?c sensors and their impact on me-chanical system diagnostics as well as discussing some future sensor needs to extend that coverage.

Finally,we discussed the notion of the intelligent engine.This future concept extends the role of EHM from a purely maintenance-centric activity to include adaptive control of the engine based on its estimated state of health.It moves much of the analysis to an onboard con?guration,thereby allowing extended (control)accommodation for estimated deterioration and detected faults as well as generating and off-loading diagnostic/prognostic information to support line maintenance and overhaul logistics. Central to achieving this is the ability to construct an accurate onboard model that can be calibrated to the particular engine being monitored/controlled(to remove engine-to-engine varia-tion),which was illustrated by way of example in an empirical process termed eSTORM.

Whatever its ultimate form,the future state of EHM will need to integrate and balance its onboard and off-board capabilities to maximize cost bene?t and operational reliability. Acknowledgment

I would like to acknowledge my longtime friend,colleague, and mentor,Louis A.Urban,for his pioneering efforts in the?eld of gas turbine engine diagnostics and for introducing me to an in-triguing discipline.I would also like to thank Pratt&Whitney and its DPHM team,past and present,and most notably Rob Luppold, Dr.Ravi Rajamani,Bruce Wood,Steve Butler,Dr.Jun Liu,Dean Kavalkovich,Angelo Martucci,Hans DePold,Jason Siegel,Dr. Danbing Seto,Dr.Liang Tang,Bobby Allen,Bill Donat,and others too numerous to name,as well as Don Simon and his team at NASA GRC,for numerous years of privileged acquaintance and collaboration.

Nomenclature

ACARS?airborne communications addressing and report-

ing system

ACMF?aircraft condition monitoring function

AIDS?airborne integrated data system

AIMS?airborne integrated monitoring system

ARINC?Aeronautical Radio,Inc.

BOM?bill of material

C-MAPSS?Commercial Modular Aero-Propulsion System

Simulation

DPHM?diagnostics,prognostics,health management

DoD?Department of Defense

DOD?domestic object damage

D&C?delays and cancellations

EDMS?exhaust debris monitoring system

EHM?engine health monitoring/management eSTORM?enhanced STORM

FADEC?full authority digital engine control

FMEA?failure modes and effects analysis

FMP??eet management program

FOD?foreign object damage

GPA?gas path analysis

IC?in?uence coef?cient

IDMS?inlet debris monitoring system

IFSD?in-?ight shutdown

LLP?life-limited part

MEMS?microelectromechanical systems

MPA?module performance analysis

MPC?model predictive control

OBIDICOTE?onboard identi?cation,diagnosis,and control of

gas turbine engines

ODM?oil debris monitor

OEM?original equipment manufacturer

PHM?prognostics/propulsion health management

QAR?quick access recorder

RTD?resistance temperature device

STORM?self-tuning onboard real-time model

SVM?state variable model

TAR?test as received

TAT?total air temperature

TCC?turbine case cooling

TSFC?thrust-speci?c fuel consumption

UBL?usage-based li?ng

UER?unplanned engine removal

Appendix

Impossible Problem Solution.Paraphrased from Ref.[88]. Mathematician P’s?rst statement that he does not know the two numbers(even though he knows the product)implies that x and y are not both prime numbers.Mathematician S asserts that she already knew that mathematician P did not know the values. So,before P spoke,S must have known,based on the value of the sum of x and y,that they could not both be primes.This means that,from the list of possible sums between3and100,we can eliminate every sum that can be formed by adding two primes. Following this elimination,we are left with the following set of 24possible values for the sum:

11;17;23;27;29;35;37;41;47;51;53;57;59;65;67;71;77;79;

83;87;89;93;95;97:

Since every sum is odd,we now know that x must be odd and y even,or vice versa.

051201-18/Vol.136,MAY2014Transactions of the ASME

Of the possible values for the sum,16of these24possible cases can be written in at least two different ways as2mtq,where m is at least2,and q is an odd prime.As an example,notice that 11?4t7?8t3.If one of these quantities is the sum,then P can now determine the values of x and y,because the product will be2m*q,which can only be broken one way into an odd and an even factor.But if one of these quantities is the sum,S can never know which of the representations2mtq corresponds to the val-ues of x and y.Since S does determine the values,these16entries in the sum list must be eliminated.

Following this elimination,we are left with the following set of eight possible values for the sum:

17;29;41;53;65;89;97

In fact,we can also eliminate any value in this list of possible sums that can be written in two different ways as the sum of a power of2and the product of any number of odd primes.(This is again because x and y must be one odd and one even).And yet we have

29?2t27?16t13

41?4t37?16t25

53?16t37?32t21

59?16t43?32t27

65?4t61?32t33

89?16t73?64t25

and,therefore,the only possible value for the sum is17!

Now we need to consider every possible partition of17:

S E O P E O

17?2t152?15?30?6?5

17?3t1414?3?42?2?21

17?4t134?13?52

17?5t1212?5?60?2?30

17?6t116?11?66?2?33

17?7t1010?7?70?2?35

17?8t98?9?72?2?36

and note that,for most partitions of17,we can come up with at least two distinct factorizations into an odd and even factor,which have the same product.Therefore,given that the sum is17,the product cannot be any of those ambiguous products.It must be the only product that can only be divided one way into an even factor and an odd factor,namely4t13.

Therefore,x?4and y?1.

References

[1]IATA,2011,“Airline Maintenance Cost Executive Commentary,”International

Air Transport Association,Montreal,Canada.

[2]Denman,J.,1996,“Depot Maintenance:Opportunities to Privatize Repair of

Military Engines.”U.S.General Accounting Of?ce,Washington,DC,Report No.GAO/NSIAD-96-33.

[3]Urban,L.A.,1972,“Gas Path Analysis Applied to Turbine Engine Condition-

ing Monitoring,”J.Aircraft,10(7),pp.400–406.

[4]Urban,L.A.,1973,“Gas Path Analyzer,”U.S.Patent3,731,070.

[5]Urban,L.A.,1974,“Parameter Selection for Multiple Fault Diagnostics of Gas

Turbine Engines,”AGARD Conference Proceedings No.165.

[6]Smetana, F.,1974,“Turbojet Engine Gas Path Analysis—A Review,”

AGARD-CP-165.

[7]Simon,D.L.,2010,“Propulsion Diagnostic Method Evaluation Strategy(Pro-

DiMES)User’s Guide,”NASA/TM—2010-215840.

[8]Doel,D.L.,2003,“Development of Baselines,In?uence Coef?cients and Sta-

tistical Inputs for Gas Path Analysis”(von Karman Lecture Notes),von Karman Institute,Rhode-Saint-Genese,Belgium.

[9]Volponi,A.J.,1999,“Gas Turbine Parameter Corrections,”ASME J.Eng.Gas

Turb.Power,121(4),pp.613–621.

[10]Mathioudakis,K.,2002,“Gas Turbine Test Parameters Corrections Including

Operation With Water Injection,”ASME Turbine and Aeroengine Congress, Amsterdam,Netherlands,June3–6,ASME Paper No.GT2002-30466.

[11]Mathioudakis,K.,Aretakis,N.,and Tsalavoutas A.,2002,“Increasing Diagnos-

tic Effectiveness by Inclusion of Fuel Composition and Water Injection Effects,”ASME Turbine and Aeroengine Congress,Amsterdam,Netherlands, June3–6,ASME Paper No.GT2002-30032.

[12]Volponi,A.J.,1983,Gas Path Analysis:An Approach to Engine Diagnostic,

Time-Dependent Failure Mechanisms and Assessment Methodologies,Cam-bridge University Press,Cambridge,UK.

[13]Volponi,A.J.,1994,“Sensor Error Compensation in Engine Performance

Diagnostics,”ASME Turbine and Aeroengine Congress.The Hague,Nether-lands,June13–16,ASME Paper No.94-GT-58.

[14]Gelb,A.,1974,Applied Optimal Estimation,MIT Press,Cambridge,MA.

[15]Volponi,A.J.,2003,“Foundations of Gas Path Analysis I&II”(von Karman

Lecture Notes),von Karman Institute,Rhode-Saint-Genese,Belgium.

[16]Doel,D.L.,1992,“TEMPER—A Gas Path Analysis Tool for Commercial Jet

Engines,”ASME Turbine and Aeroengine Congress,Cologne,Germany,June 1–4,ASME Paper92-GT-315.

[17]Doel,D.L.,1993,“An Assessment of Weighted-Least-Squares Based Gas Path

Analysis,”ASME Turbine and Aeroengine Congress,Cincinnati,OH,May 24–27,ASME Paper93-GT-119.

[18]Stamatis,A.,and Papailiou,K.D.,1991,“Jet Engine Fault Detection With Dis-

crete Operating Points Gas Path Analysis,”J.Propulsion,7(6),pp.1043–1048.

[19]Volponi,A.J.,and Urban,L.A.,1992,“Mathematical Methods of Relative

Engine Performance Diagnostics,”SAE Technical Paper922048.

[20]Provost,M.J.,1994,“The Use of Optimal Estimation Techniques in the Analy-

sis of Gas Turbines,”Ph.D.thesis,Cran?eld University,Bedford,UK.

[21]Borguet,S.,and Leonard,O.,2010,“A Sparse Estimation Approach to Fault

Isolation,”ASME J.Eng.Gas Turb.Power,132(2),p.021601.

[22]Kobayashi,T.,and Simon,D.L.,2003,“Application of a Bank of Kalman Fil-

ters for Aircraft Engine Fault Diagnostics,”ASME Turbine and Aeroengine Congress,Atlanta,GA,June16–19,ASME Paper No.GT2003-38550.

[23]Aretakis,M.S.,2004,“Identi?cation of Sensor Faults on Turbofan Engines

Using Pattern Recognition Techniques,”Control Eng.Pract.,12.

[24]Lee,Y.K.,Volovoi,V.V.,Yuan,M.,and Fisher,T.,2010,“A Fault Diagnosis

Method for Industrial Gas Turbines Using Bayesian Data Analysis,”ASME J.

Eng.Gas Turb.Power,132(4),p.041602.

[25]Dienger,G.,and Rheilander,H.,1979,“Economic Review on ATLAS JT9D

Gas Path Analysis Application,”17th ETMT Working Paper,Hamburg IF26, Rel604.

[26]Danielsson,S.G.,1977,“Gas Path Analysis Applied to Pre and Post Overhaul

Testing of the JT9D Turbofan Engine,”SAE Technical Paper770993.

[27]Stroobach,J.,1980,“GPA Evaluation of On-Wing Recorded Data With a Limited

Number of Sensors on the Engines Installed PH-DTA,”KLM CED Report G-218.

[28]Stroobach,J.,1981,“GPA Testcell Evaluation Performed on CF6-50C

Engines,”KLM CED Report G-239.

[29]Merrington,G.L.,“Fault Diagnosis of Gas Turbine Engines From Transient

Data,”ASME Turbine and Aeroengine Congress,Amsterdam,Netherlands, June5–9,ASME Paper No.88-GT-209.

[30]Glenny,D.E.1988,“Gas Path Analysis and Engine Performance Monitoring in

A Chinook Helicopter,”AGARD:Engine Condition Monitoring,North Atlantic

Treaty Organization,Brussels,Belgium.

[31]Merrington,G.,Kwon,O.-K.,Goodwin,G.,and Carlsson,B.,1991,“Fault

Detection and Diagnosis in Gas Turbines,”ASME J.Eng.Gas Turb.Power, 13(2),276–282.

[32]Simon,D.,and Chia,T.L.,2002,“Kalman Filtering with State Equality Con-

straints,”IEEE Trans.Aerosp.Electron.Syst.,39(1),pp.128–136.

[33]Simon,D.L.,and Simon,D.,2003,“Aircraft Turbofan Engine Health Estima-

tion Using Constrained Kalman Filtering,”ASME Turbine and Aeroengine Congress,Atlanta,GA,June16–19,ASME Paper No.GT2003-38584.

[34]Volponi,A.J.,2003,“Extending Gas Path Analysis Coverage for Other Fault

Conditions”(von Karman Lecture Notes),von Karman Institute,Rhode-Saint-Genese,Belgium.

[35]Marinai,L.,2004,“Gas-Path Diagnostics and Prognostics for Aero-Engines

Using Fuzzy Logic and Time Series Analysis,”Ph.D.thesis,Cran?eld Univer-sity,Bedford,UK,p.50.

[36]Jaw,L.C.,and Wu,D.N.,2002,“Anomaly Detection and Reasoning With Em-

bedded Physical Model,”2002IEEE Aerospace Conference,Big Sky,MT, March9–16.

[37]Cleary,D.J.,Yu,L.,and Cuddihy,P.E.,2003,“A Novel Approach to Aircraft

Engine Anomaly Detection and Diagnostics,”GE Global Research Report No.

2003GRC117.

[38]Brotherton,T.,and Johnson,T.,2001,“Anomaly Detection for Advanced Mili-

tary Aircraft Using Neural Networks,”2001IEEE Aerospace Conference,Big Sky,MT,March10–17.

[39]Markou,M.,and Singh,S.,2003,“Novelty Detection:A Review,”Signal Pro-

cess.,83(12),pp.2481–2496.

[40]Volponi,A.J.,DePold,H.,Ganguli,R.,and Daguang,C.,2003,“The Use of

Kalman Filter and Neural Network Methodologies in Gas Turbine Performance

Journal of Engineering for Gas Turbines and Power MAY2014,Vol.136/051201-19

Diagnostics:A Comparative Study,”ASME J.Eng.Gas Turb.Power,125,pp.

917–924.

[41]Butler,S.W.,2006,“A Performance Visualization,Assessment,and Tuning

Methodology for Model-Based and Data-Driven Diagnostic Classi?cation Sys-tems,”Master’s thesis,University of Connecticut,Storrs,CT.

[42]Witt,C.,1984,“AIR1839—A Guide to Aircraft Engine Vibration Monitoring

Systems,”SAE Technical Paper841458.

[43]Muir,D.M.,and Howe,B.,1996,“In-Line Oil Debris Monitor(ODM)for the

Advanced Tactical Fighter Engine,”SAE Technical Paper961308.

[44]Miller,J.,and Kitaljevich,D.,2000,“In-Line Oil Debris Monitor for Aircraft

Engine Condition Assessment,”2000IEEE Aerospace Conference,Big Sky, MT,March18–25.

[45]Dempsey,J.P.,2000,“A Comparison of Vibration and Oil Debris Gear Dam-

age Detection Methods Applied to Pitting Damage,”NASA Report No.TM-2000-210371.

[46]Dempsey,J.P.,and Afjeh,A.,2002,“Integrating Oil Debris and Vibration

Gear Damage Detection Technologies Using Fuzzy Logic,”NASA Report No.

TM-2002-211126.

[47]Couch,R.P.,1978,“Detecting Abnormal Turbine Engine Deterioration Using

Electrostatic Methods,”J.Aircraft,15(10),pp.692–695.

[48]Prowie,H. E.G.,1997,“Gas Path Condition Monitoring During

Accelerated Mission Testing of a Demonstrator Engine,”AIAA Paper No.

97-2904.

[49]Fisher,C.,2000,“Gas Path Debris Monitoring—A21st Century PHM Tool,”

2000IEEE Aerospace Conference,Big Sky,MT,March18–25.

[50]Fisher,C.,2000,“Gas Turbine Condition Monitoring Systems—An Integrated

Approach,”The Power Engineer,1(4),pp.3–16.

[51]Fisher,C.,2001,“Data and Information Fusion for Gas Path Debris Monitoring,”

2001IEEE Aerospace Conference,Big Sky,MT,March10–17.

[52]Luppold,R.H.,1989,“Estimating In-Flight Engine Performance Variations

Using Kalman Filter Concepts,”AIAA Paper No.89-2584.

[53]Gallops,G.W.,1992,“In-Flight Performance Diagnostic Capability of an

Adaptive Engine Model,”AIAA Paper No.92-3746.

[54]Kerr,L.J.,1991,“Real-Time Estimation of Gas Turbine Engine Damage Using

a Control Based Kalman Filter Algorithm,”ASME Turbine and Aeroengine

Congress.Orlando,FL,June3–6,ASME Paper No.91-GT-216.

[55]Hall,D.L.,Hansen,R.J.,and Lang,D.C.,1996,“The Negative Information

Problem in Mechanical Diagnostics,”ASME Turbine and Aeroengine Con-gress,Birmingham,UK,June10–13,ASME Paper96-GT-035.

[56]Volponi, A.J.,Brotherton,T.,Luppold,R.,and Simon, D.L.,2003,

“Development of an Information Fusion System for Engine Diagnostics and Health Management,”JANNAF39th Combustion/27th Airbreathing Propul-sion/21st Propulsion Systems Hazards/3rd Modeling and Simulation Joint Sub-committee Meeting,Colorado Springs,CO,December1–5.

[57]Volponi,A.J.,2005,“Data Fusion for Enhanced Aircraft Engine Prognostics

and Health Management,”Glenn Research Center,Cleveland,OH,NASA Report No.CR-2005-214055.

[58]Liu,Y.,Frederick,D.K.,DeCastro,J.S.,and Chan,W.W.,2012,“User’s

Guide for the Commercial Modular Aero-Propulsion System Simulation(C-MAPSS),”NASA/TM—2012-217432.

[59]Kriener,A.,and Lietzau,K.,2000,The Use of Onboard Real-Time Models for

Jet Engine Control,Project Report OBIDICOTE,MTU Aero Engines,Munich, Germany.

[60]SAE,2013,“Determination of Costs and Bene?ts From Implementing an

Engine Health Management System.”SAE Standard ARP4176.

[61]Culley,D.,2009,“More Intelligent Gas Turbine Engines,”RTO Technical

Report,TR-AVT-128.

[62]Hess,A.,2004,“PHM a Key Enabler for the JSF Autonomic Logistics Support

Concept,”2004IEEE Aerospace Conference,Big Sky,MT,March6–13. [63]Ring, D.,2003,“Control Interaction With Diagnostics and Fault Tolerant

Control,”(von Karman Lecture Notes”),von Karman Institute,Rhode-Saint-Genese,Belgium.

[64]Garg,S.,2004,Controls and Health Management Technologies for Intelligent

Aerospace Propulsion Systems,”NASA/TM—2004-212915.

[65]Simon,D.L.,Garg,S.,Hunter,G.W.,Guo,T.,and Semega,K.,2004,“Sensor

Needs for Control and Health Management of Intelligent Aircraft Engines,”

NASA/TM—2004-213202.[66]Litt,J.S.,2005,“A Survey of Intelligent Control and Health Management

Technologies for Aircraft Propulsion Systems,”NASA/TM—2005-213622. [67]Wiseman,M.,2005,“Intelligent Engine Systems:Work Element1.2:Malfunc-

tion and Operator Error Reduction,”NASA/CR—2005-213964.

[68]Fletcher,C.C.,2004,“A Vision for the Intelligent Pipeline System and the

Role of the Intelligent Engine,”GMRC—2004Gas Machinery Conference,Al-buquerque,NM,October4–6.

[69]Turso,J.A.,2008,“Toward an Intelligent,Deterioration Accommodating Con-

troller for Aging Turbofan Engines,”Aeronaut.J.,112(1137),pp.641–652. [70]Rausch,R.,2004,“Towards In-Flight Detection and Accommodation of Faults

in Aircraft Engines,”AIAA1st Intelligent Systems Technical Conference,Chi-cago,IL,September20-22,AIAA Paper No.2004-6463.

[71]Van Essen,H.,1998,“Modeling and Model Based Control of

Turbomachinery,”Ph.D.thesis,Technische Universiteit Eindhoven,Eindhoven, Netherlands.

[72]Brotherton,T.,Volponi,A.J.,Luppold,R.,and Simon,D.L.,2003,“eSTORM:

Enhanced Self Tuning On-Board Real-Time Engine Model,”2003IEEE Aero-space Conference,Big Sky,MT,March8–15,Paper No.1023.

[73]Volponi,A.J.,2005,“Use of Hybrid Engine Modeling for On-Board Module

Performance Tracking,”ASME Turbine and Aeroengine Congress,Reno,NV, June6–9,ASME Paper No.GT2005-68169.

[74]Volponi,A.J.,2007,“Adaptive On-Wing Gas Turbine Engine Performance

Estimation,”2007IEEE Aerospace Conference.Big Sky,MT,March3–10. [75]Volponi,A.J.,Brotherton,T.,and Luppold,R.,2008,“Empirical Tuning of an

On-Board Gas Turbine Engine Model for Real-Time Module Performance Estimation,”ASME J.Eng.Gas Turb.Power,130(2),p.021604.

[76]Borguet,S.,2006,“A Way to Deal With Model-Plant Mismatch for Reliable

Diagnosis in Transient Operation,”ASME Turbine and Aeroengine Congress, Barcelona,Spain,May8–11,ASME Paper No.GT2006-90412.

[77]Li,Y.G.,DeWallef,P.,and Leonard,O.,2003,“Gas Turbine Diagnostic

Approach With Transient Measurements,”Proc.IMechE A J.Power Energy, 217(A2),pp.169–177.

[78]Ogaji,S.O.T.,Li,Y.G.,Sampath,S.,and Singh,R.,2003,“Gas Path Fault Di-

agnosis of a Turbofan Engine From Transient Data Using Arti?cial Neural Networks,”ASME Turbine and Aeroengine Congress.Atlanta,GA,June 16–19,ASME Paper No.GT2003-38423.

[79]Sampath,S.,Ogaji,S.O.T.,Li,Y.G.,and Singh,R.,2003,“Fault Diagnosis of

a Two Spool Turbo-Fan Engine Using Transient Data:A Genetic Algorithm

Approach,”ASME Turbine and Aeroengine Congress.Atlanta,GA,June 16–19,ASME Paper No.GT2003-38300.

[80]Simmons,J.C.,and Danai,K.,2012,“In-Flight Isolation of Degraded Engine

Components by Shape Comparison of Transient Outputs,”ASME J.Eng.Gas Turb.Power,134(6),p.061602.

[81]Borguet,S.,DeWallef,P.,and Leonard,O.,2010,“A Study on Engine

Health Monitoring in the Frequency Domain,”ASME Turbine and Aeroen-gine Congress,Glasgow,Scotland,June14–18,ASME Paper No.GT2010-22635.

[82]Kamboukos,P.,and Mathioudakis K.,2007,“Turbofan Engine Health Assess-

ment by Combining Steady and Transient State Aerothermal Data,”7th Euro-pean Turbomachinery Conference,Athens,Greece,March5-9,Paper NO.137.

[83]SAE,2013,“Guidelines for Engine Health Management System Software

Assurance Levels,”SAE Standard ARP5987.

[84]Taylor,B.J.,ed.,2006,Models and Procedures for the Veri?cation and Valida-

tion of Arti?cial Neural Networks,Springer Science,New York.

[85]Menzies,T.,and Pecheur,C.,2005,Veri?cation and Validation and Arti?cial

Intelligence,Advances in Computers,Vol.65Elsevier,Amsterdam.

[86]Jacklin,S.A.,Schumann,J.M.,Gupta,P.P.,Richard,M.,Guenther,K.,and

Soares,F.,2005,“Towards Development of Advanced Veri?cation and Valida-tion Tools for the Certi?cation of Learning Systems in Aerospace Applications,”AIAA Infotech,Arlington,VA,September26–29.AIAA Paper No.2005-6912.

[87]Jacklin,S.,2004,“Veri?cation,Validation and Certi?cation Challenges for

Adaptive Flight Control Systems Software,”AIAA Guidance,Navigation,and Control Conference and Exhibit,Providence,RI,August16–19,AIAA Paper No.2004-5258.

[88]The Problem Solving Group of Bern,Switzerland,1976,“Impossible Problem,”

Math.Mag.50(5),p.268.

051201-20/Vol.136,MAY2014Transactions of the ASME

50万吨年煤气化生产工艺

咸阳职业技术学院生化工程系毕业论文（设计） 50wt/年煤气化工艺设计 1.引言 煤是由古代植物转变而来的大分子有机化合物。我国煤炭储量丰富，分布面广，品种齐全。据中国第二次煤田预测资料，埋深在1000m以浅的煤炭总资源量为2.6万亿t。其中大别山—秦岭—昆仑山一线以北地区资源量约2.45万亿t，占全国总资源量的94%；其余的广大地区仅占6%左右。其中新疆、内蒙古、山西和陕西等四省区占全国资源总量的81.3%，东北三省占 1.6%，华东七省占2.8%，江南九省占1.6%。 煤气化是煤炭的一个热化学加工过程，它是以煤或煤焦原料，以氧气（空气或富氧）、水蒸气或氢气等作气化剂，在高温条件下通过化学反应将煤或煤焦中的可燃部分转化为可燃性的气体的过程。气化时所得的可燃性气体称为煤气，所用的设备称为煤气发生炉。 煤气化技术开发较早，在20世纪20年代，世界上就有了常压固定层煤气发生炉。20世纪30年代至50年代，用于煤气化的加压固定床鲁奇炉、常压温克勒沸腾炉和常压气流床K-T炉先后实现了工业化，这批煤气化炉型一般称为第一代煤气化技术。第二代煤气化技术开发始于20世纪60年代，由于当时国际上石油和天然气资源开采及利用于制取合成气技术进步很快，大大降低了制造合成

气的投资和生产成本，导致世界上制取合成气的原料转向了天然气和石油为主，使煤气化新技术开发的进程受阻，20世纪70年代全球出现石油危机后，又促进了煤气化新技术开发工作的进程，到20世纪80年代，开发的煤气化新技术，有的实现了工业化，有的完成了示范厂的试验，具有代表性的炉型有德士古加压水煤浆气化炉、熔渣鲁奇炉、高温温克勒炉（ETIW）及干粉煤加压气化炉等。 近年来国外煤气化技术的开发和发展，有倾向于以煤粉和水煤浆为原料、以高温高压操作的气流床和流化床炉型为主的趋势。 2.煤气化过程 2.1煤气化的定义 煤与氧气或（富氧空气）发生不完全燃烧反应，生成一氧化碳和氢气的过程称为煤气化。煤气化按气化剂可分为水蒸气气化、空气（富氧空气）气化、空气—水蒸气气化和氢气气化；按操作压力分为：常压气化和加压气化。由于加压气化具有生产强度高，对燃气输配和后续化学加工具有明显的经济性等优点。所以近代气化技术十分注重加压气化技术的开发。目前，将气化压力在P＞2MPa 情况下的气化，统称为加压气化技术；按残渣排出形式可分为固态排渣和液态排渣。气化残渣以固体形态排出气化炉外的称固态排渣。气化残渣以液态方式排出经急冷后变成熔渣排出气化炉外的称液态排渣；按加热方式、原料粒度、汽化程度等还有多种分类方法。常用的是按气化炉内煤料与气化剂的接触方式区分，主要有固定床气化、流化床气化、气流床气化和熔浴床床气化。 2.2 主要反应 煤的气化包括煤的热解和煤的气化反应两部分。煤在加热时会发生一系列的物理变化和化学变化。气化炉中的气化反应，是一个十分复杂的体系，这里所讨论的气化反应主要是指煤中的碳与气化剂中的氧气、水蒸汽和氢气的反应，也包括碳与反应产物之间进行的反应。 习惯上将气化反应分为三种类型：碳—氧之间的反应、水蒸汽分解反应和甲烷生产反应。 2.2.1碳—氧间的反应 碳与氧之间的反应有： C+O2=CO2（1）

煤气化工艺流程

煤气化工艺流程 1、主要产品生产工艺煤气化是以煤炭为主要原料的综合性大型化工企业，主要工艺围绕着煤的洁净气化、综合利用，形成了以城市煤气为主线联产甲醇的工艺主线。 主要产品城市煤气和甲醇。城市燃气是城市公用事业的一项重要基础设施，是城市现代化的重要标志之一，用煤气代替煤炭是提高燃料热能利用率，减少煤烟型大气污染，改善大气质量行之有效的方法之一，同时也方便群众生活，节约时间，提高整个城市的社会效率和经济效益。作为一项环保工程，（其一期工程）每年还可减少向大气排放烟尘万吨、二氧化硫万吨、一氧化碳万吨，对改善河南西部地区城市大气质量将起到重要作用。 甲醇是一种重要的基本有机化工原料，除用作溶剂外，还可用于制造甲醛、醋酸、氯甲烷、甲胺、硫酸二甲酯、对苯二甲酸二甲酯、丙烯酸甲酯等一系列有机化工产品，此外，还可掺入汽油或代替汽油作为动力燃料，或进一步合成汽油，在燃料方面的应用，甲醇是一种易燃液体，燃烧性能良好，抗爆性能好，被称为新一代燃料。甲醇掺烧汽油，在国外一般向汽油中掺混甲醇5?15勉高汽油的辛烷值，避免了添加四乙基酮对大气的污染。 河南省煤气（集团）有限责任公司义马气化厂围绕义马至洛阳、洛阳至郑州煤气管线及豫西地区工业及居民用气需求输出清洁能源，对循环经济建设，把煤化工打造成河南省支柱产业起到重要作用。 2、工艺总流程简介： 原煤经破碎、筛分后，将其中5?50mm级块煤送入鲁奇加压气化炉，在炉内与氧气和水蒸气反应生成粗煤气，粗煤气经冷却后，进入低温甲醇洗净化装置，除去煤气中的CO2和H2S净化后的煤气分为两大部分，一部分去甲醇合成系统，合成气再经压缩机加压至，进入甲醇反应器生成粗甲醇，粗甲醇再送入甲醇精馏系统，制得精甲醇产品存入贮罐；另一部分去净煤气变换装置。合成甲醇尾气及变换气混合后，与剩余部分出低温甲醇洗净煤气混合后，进入煤气冷却干燥装置，将露点降至-25 C后，作为合格城市煤气经长输管线送往各用气城市。生产过程中产生的煤气水进入煤气水分离装置，分离出其中的焦油、中油。分离后煤气水去酚回收和氨回收，回收酚氨后的煤气水经污水生化处理装置处理，达标后排放。低温甲醇洗净化装置排出的H2S到硫回收装置回收硫。空分

煤化工产业概况及其发展趋势

煤化工产业概况及其发 展趋势 集团标准化办公室：[VV986T-J682P28-JP266L8-68PNN]

我国煤化工产业概况及其发展趋势 煤化学加工包括煤的焦化、气化和液化。主要用于冶金行业的煤炭焦化和用于制取合成氨的煤炭气化是传统的煤化工产业，随着社会经济的不断发展，它们将进一步得到发展，同时以获得洁净能源为主要目的的煤炭液化、煤基代用液体燃料、煤气化—发电等煤化工或煤化工能源技术也越来越引起关注，并将成为新型煤化工产业化发展的主要方向。发展新型煤化工产业对煤炭行业产业结构的调整及其综合发展具有重要意义。 1 煤化工产业发展概况 1． 1 煤炭焦化 焦化工业是发展最成熟，最具代表性的煤化工产业，也是冶金工业高炉炼铁、机械工业铸造最主要的辅助产业。目前，全世界的焦炭产量大约为～亿t/a，直接消耗原料精煤约亿t/a 。受世界钢铁产量调整、高炉喷吹技术发展、环境保护以及生产成本增高等原因影响，工业发达国家的机械化炼焦能力处于收缩状态，焦炭国际贸易目前为2500万t/ a。 目前，我国焦炭产量约亿t/a，居世界第一，直接消耗原料煤占全国煤炭消费总量的14%。 全国有各类机械化焦炉约750座以上，年设计炼焦能力约9000万 t/a，其中炭化室高度为4m～5.5m以上的大、中型焦炉产量约占80%。中国大容积焦炉（炭化室高≧6m）已实现国产化，煤气净化技术已达世界先进水平，干熄焦、地面烟尘处理站、污水处理等已进入实用化阶段，焦炭质量显着提高，其主要化工产品的精制技术已达到或接近世界先进水平。 焦炭成为我国的主要出口产品之一，出口量逐年上升，2000年达到1500t/a，已成为全球最大的焦炭出口国。 从20世纪80年代起，煤炭行业的炼焦生产得到逐步发展，其中有的建成向城市或矿区输送人工煤气为主要目的的工厂，有的以焦炭为主要产品。煤炭行业焦化生产普遍存在的问题是：焦炉炉型小、以中小型焦炉为主，受矿区产煤品种限制、焦炭质量调整提高难度较大，采用干法熄焦、烟尘集中处理等新技术少，大多数企业技术进步及现代化管理与其他行业同类工厂相比有较大差距。 1．2 煤气化及其合成技术 1.2.1 煤气化 煤气化技术是煤化工产业化发展最重要的单元技术。全世界现有商业化运行的大规模气化炉414台，额定产气量446×106Nm3/d，前10名的气化厂使用鲁奇、德士古、壳牌3种炉型，原料是煤、渣油、天然气，产品是F-T合成油、电或甲醇等。 煤气化技术在我国被广泛应用于化工、冶金、机械、建材等工业行业和生产城市煤气的企业，各种气化炉大约有9000多台，其中以固定床气化炉为主。近20年来，我国引进的加压鲁奇炉、德士古水煤浆气化炉，主要用于生产合成氨、甲醇或城市煤气。

煤气化技术的现状及发展趋势分析

煤气化技术是现代煤化工的基础，是通过煤直接液化制取油品或在高温下气化制得合成气，再以合成气为原料制取甲醇、合成油、天然气等一级产品及以甲醇为原料制得乙烯、丙烯等二级化工产品的核心技术。作为煤化工产业链中的“龙头”装置，煤气化装置具有投入大、可靠性要求高、对整个产业链经济效益影响大等特点。目前国内外气化技术众多，各种技术都有其特点和特定的适用场合，它们的工业化应用程度及可靠性不同，选择与煤种及下游产品相适宜的煤气化工艺技术是煤化工产业发展中的重要决策。 工业上以煤为原料生产合成气的历史已有百余年。根据发展进程分析，煤气化技术可分为三代。第一代气化技术为固定床、移动床气化技术，多以块煤和小颗粒煤为原料制取合成气，装置规模、原料、能耗及环保的局限性较大；第二代气化技术是现阶段最具有代表性的改进型流化床和气流床技术，其特征是连续进料及高温液态排渣；第三代气化技术尚处于小试或中试阶段，如煤的催化气化、煤的加氢气化、煤的地下气化、煤的等离子体气化、煤的太阳能气化和煤的核能余热气化等。 本文综述了近年来国内外煤气化技术开发及应用的进展情况，论述了固定床、流化床、气流床及煤催化气化等煤气化技术的现状及发展趋势。 1．国内外煤气化技术的发展现状 在世界能源储量中，煤炭约占79%，石油与天然气约占12%。煤炭利用技术的研究和开发是能源战略的重要内容之一。世界煤化工的发展经历了起步阶段、发展阶段、停滞阶段和复兴阶段。20世纪初，煤炭炼焦工业的兴起标志着世界煤化工发展的起步。此后世界煤化工迅速发展，直到20世纪中叶，煤一直是世界有机化学工业的主要原料。随着石油化学工业的兴起与发展，煤在化工原料中所占的比例不断下降并逐渐被石油和天然气替代，世界煤化工技术及产业的发展一度停滞。直到20世纪70年代末，由于石油价格大幅攀升，影响了世界石油化学工业的发展，同时煤化工在煤气化、煤液化等方面取得了显著的进展。特别是20世纪90年代后，世界石油价格长期在高位运行，且呈现不断上升趋势，这就更加促进了煤化工技术的发展，煤化工重新受到了人们的重视。 中国的煤气化工艺由老式的UGI炉块煤间歇气化迅速向世界最先进的粉煤加压气化工艺过渡，同时国内自主创新的新型煤气化技术也得到快速发展。据初步统计，采用国内外先进大型洁净煤气化技术已投产和正在建设的装置有80多套，50%以上的煤气化装置已投产运行，其中采用水煤浆气化技术的装置包括GE煤气化27套（已投产16套），四喷嘴33套（已投产13套），分级气化、多元料浆气化等多套；采用干煤粉气化技术的装置包括Shell煤气化18套（已投产11套）、GSP2套，还有正在工业化示范的LurgiBGL技术、航天粉煤加压气化（HT-L）技术、单喷嘴干粉气化技术和两段式干煤粉加压气化（TPRI）技术等。

煤气化工艺流程

精心整理 煤气化工艺流程 1、主要产品生产工艺 煤气化是以煤炭为主要原料的综合性大型化工企业，主要工艺围绕着煤的洁净气化、综合利用，形成了以城市煤气为主线联产甲醇的工艺主线。 主要产品城市煤气和甲醇。城市燃气是城市公用事业的一项重要基础设施，是城市现代化的重要标志之一，用煤气代替煤炭是提高燃料热能利用率，减少煤烟型大气污染，改善大气质量行之 化碳 15%提 作用。 2 。净化 装置。合成甲醇尾气及变换气混合后，与剩余部分出低温甲醇洗净煤气混合后，进入煤气冷却干燥装置，将露点降至-25℃后，作为合格城市煤气经长输管线送往各用气城市。生产过程中产生的煤气水进入煤气水分离装置，分离出其中的焦油、中油。分离后煤气水去酚回收和氨回收，回收酚氨后的煤气水经污水生化处理装置处理，达标后排放。低温甲醇洗净化装置排出的H2S到硫回收装置回收硫。空分装置提供气化用氧气和全厂公用氮气。仪表空压站为全厂仪表提供合格的仪表空气。 小于5mm粉煤，作为锅炉燃料，送至锅炉装置生产蒸汽，产出的蒸汽一部分供工艺装置用汽

，一部分供发电站发电。 3、主要装置工艺流程 3.1备煤装置工艺流程简述 备煤工艺流程分为三个系统： （1）原煤破碎筛分贮存系统，汽运原煤至受煤坑经1#、2#、3#皮带转载至筛分楼、经节肢筛、破碎机、驰张筛加工后，6~50mm块煤由7#皮带运至块煤仓，小于6mm末煤经6#、11#皮带近至末煤仓。 缓 可 能周期性地加至气化炉中。 当煤锁法兰温度超过350℃时，气化炉将联锁停车，这种情况仅发生在供煤短缺时。在供煤短缺时，气化炉应在煤锁法兰温度到停车温度之前手动停车。 气化炉：鲁奇加压气化炉可归入移动床气化炉，并配有旋转炉篦排灰装置。气化炉为双层压力容器，内表层为水夹套，外表面为承压壁，在正常情况下，外表面设计压力为3600KPa（g），内夹套与气化炉之间压差只有50KPa（g）。 在正常操作下，中压锅炉给水冷却气化炉壁，并产生中压饱和蒸汽经夹套蒸汽气液分离器1

煤化工工艺流程

煤化工工艺流程 典型的焦化厂一般有备煤车间、炼焦车间、回收车间、焦油加工车间、苯加工车间、脱硫车间和废水处理车间等。 焦化厂生产工艺流程 1.备煤与洗煤 原煤一般含有较高的灰分和硫分，洗选加工的目的是降低煤的灰分，使混杂在煤中的矸石、煤矸共生的夹矸煤与煤炭按照其相对密度、外形及物理性状方面的差异加以分离，同时，降低原煤中的无机硫含量，以满足不同用户对煤炭质量的指标要求。 由于洗煤厂动力设备繁多，控制过程复杂，用分散型控制系统DCS改造传统洗煤工艺，这对于提高洗煤过程的自动化，减轻工人的劳动强度，提高产品产量和质量以及安全生产都具有重要意义。

洗煤厂工艺流程图 控制方案 洗煤厂电机顺序启动/停止控制流程框图 联锁/解锁方案：在运行解锁状态下，允许对每台设备进行单独启动或停止；当设置为联锁状态时，按下启动按纽，设备顺序启动，后一设备的启动以前一设备的启动为条件（设备间的延时启动时间可设置），如果前一设备未启动成功，后一设备不能启动，按停止键，则设备顺序停止，在运行过程中，如果其中一台设备故障停止，例如设备2停止，则系统会把设备3和设备4停止，但设备1保持运行。

2.焦炉与冷鼓 以100万吨/年-144孔-双炉-4集气管-1个大回流炼焦装置为例，其工艺流程简介如下：

100万吨/年焦炉_冷鼓工艺流程图 控制方案 典型的炼焦过程可分为焦炉和冷鼓两个工段。这两个工段既有分工又相互联系，两者在地理位置上也距离较远，为了避免仪表的长距离走线，设置一个冷鼓远程站及给水远程站，以使仪表线能现场就近进入DCS控制柜，更重要的是，在集气管压力调节中，两个站之间有着重要的联锁及其排队关系，这样的网络结构形式便于可以实现复杂的控制算法。

煤气化工艺流程

煤气化工艺流程 1、主要产品生产工艺 煤气化是以煤炭为主要原料的综合性大型化工企业，主要工艺围绕着煤的洁净气化、综合利用，形成了以城市煤气为主线联产甲醇的工艺主线。 主要产品城市煤气和甲醇。城市燃气是城市公用事业的一项重要基础设施，是城市现代化的重要标志之一，用煤气代替煤炭是提高燃料热能利用率，减少煤烟型大气污染，改善大气质量行之有效的方法之一，同时也方便群众生活，节约时间，提高整个城市的社会效率和经济效益。作为一项环保工程，（其一期工程）每年还可减少向大气排放烟尘1.86万吨、二氧化硫3.05万吨、一氧化碳0.46万吨，对改善河南西部地区城市大气质量将起到重要作用。 甲醇是一种重要的基本有机化工原料，除用作溶剂外，还可用于制造甲醛、醋酸、氯甲烷、甲胺、硫酸二甲酯、对苯二甲酸二甲酯、丙烯酸甲酯等一系列有机化工产品，此外，还可掺入汽油或代替汽油作为动力燃料，或进一步合成汽油，在燃料方面的应用，甲醇是一种易燃液体，燃烧性能良好，抗爆性能好，被称为新一代燃料。甲醇掺烧汽油，在国外一般向汽油中掺混甲醇5～15%提高汽油的辛烷值，避免了添加四乙基酮对大气的污染。 河南省煤气（集团）有限责任公司义马气化厂围绕义马至洛阳、洛阳至郑州煤气管线及豫西地区工业及居民用气需求输出清洁能源，对循环经济建设，把煤化工打造成河南省支柱产业起到重要作用。 2、工艺总流程简介： 原煤经破碎、筛分后，将其中5～50mm级块煤送入鲁奇加压气化炉，在炉内与氧气和水蒸气反应生成粗煤气，粗煤气经冷却后，进入低温甲醇洗净化装置

，除去煤气中的CO2和H2S。净化后的煤气分为两大部分，一部分去甲醇合成系统，合成气再经压缩机加压至5.3MPa，进入甲醇反应器生成粗甲醇，粗甲醇再送入甲醇精馏系统，制得精甲醇产品存入贮罐；另一部分去净煤气变换装置。合成甲醇尾气及变换气混合后，与剩余部分出低温甲醇洗净煤气混合后，进入煤气冷却干燥装置，将露点降至-25℃后，作为合格城市煤气经长输管线送往各用气城市。生产过程中产生的煤气水进入煤气水分离装置，分离出其中的焦油、中油。分离后煤气水去酚回收和氨回收，回收酚氨后的煤气水经污水生化处理装置处理，达标后排放。低温甲醇洗净化装置排出的H2S到硫回收装置回收硫。空分装置提供气化用氧气和全厂公用氮气。仪表空压站为全厂仪表提供合格的仪表空气。 小于5mm粉煤，作为锅炉燃料，送至锅炉装置生产蒸汽，产出的蒸汽一部分供工艺装置用汽，一部分供发电站发电。 3、主要装置工艺流程 3.1备煤装置工艺流程简述 备煤工艺流程分为三个系统： （1）原煤破碎筛分贮存系统，汽运原煤至受煤坑经1#、2#、3#皮带转载至筛分楼、经节肢筛、破碎机、驰张筛加工后，6~50mm块煤由7#皮带运至块煤仓，小于6mm末煤经6#、11#皮带近至末煤仓。 （2）最终筛分系统：块煤仓内块煤经8#、9#皮带运至最终筛分楼驰张筛进行检查性筛分。大于6mm块煤经10#皮带送至200#煤斗，筛下小于6mm末煤经14#皮带送至缓冲仓。 （3）电厂上煤系统：末煤仓内末煤经12#、13#皮带转至5#点后经16#皮

(能源化工行业)我国煤化工产业概况及其发展方向

（能源化工行业）我国煤化工产业概况及其发展方向

我国煤化工产业概况及其发展趋势 煤化学加工包括煤的焦化、气化和液化。主要用于冶金行业的煤炭焦化和用于制取合成氨的煤炭气化是传统的煤化工产业，随着社会经济的不断发展，它们将进壹步得到发展，同时以获得洁净能源为主要目的的煤炭液化、煤基代用液体燃料、煤气化—发电等煤化工或煤化工能源技术也越来越引起关注，且将成为新型煤化工产业化发展的主要方向。发展新型煤化工产业对煤炭行业产业结构的调整及其综合发展具有重要意义。 1煤化工产业发展概况 1．1煤炭焦化 焦化工业是发展最成熟，最具代表性的煤化工产业，也是冶金工业高炉炼铁、机械工业铸造最主要的辅助产业。目前，全世界的焦炭产量大约为3.2～3.4亿t/a，直接消耗原料精煤约4.5亿t/a。受世界钢铁产量调整、高炉喷吹技术发展、环境保护以及生产成本增高等原因影响，工业发达国家的机械化炼焦能力处于收缩状态，焦炭国际贸易目前为2500万t/a。 目前，我国焦炭产量约1.2亿t/a，居世界第壹，直接消耗原料煤占全国煤炭消费总量的14%。全国有各类机械化焦炉约750座之上，年设计炼焦能力约9000万t/a，其中炭化室高度为4m～5.5m之上的大、中型焦炉产量约占80%。中国大容积焦炉（炭化室高≧6m）已实现国产化，煤气净化技术已达世界先进水平，干熄焦、地面烟尘处理站、污水处理等已进入实用化阶段，焦炭质量显著提高，其主要化工产品的精制技术已达到或接近世界先进水平。 焦炭成为我国的主要出口产品之壹，出口量逐年上升，2000年达到1500t/a，已成为全球最大的焦炭出口国。 从20世纪80年代起，煤炭行业的炼焦生产得到逐步发展，其中有的建成向城市或矿区输送人工煤气为主要目的的工厂，有的以焦炭为主要产品。煤炭行业焦化生产普遍存在的问题是：焦炉炉型小、以中小型焦炉为主，受矿区产煤品种限制、焦炭质量调整提高难度较大，采用干法熄焦、烟尘集中处理等新技术少，大多数企业技术进步及现代化管理和其他行业同类工厂相比有较大差距。 1．2煤气化及其合成技术 1.2.1煤气化 煤气化技术是煤化工产业化发展最重要的单元技术。全世界现有商业化运行的大规模气化炉414台，额定产气量446×106Nm3/d，前10名的气化厂使用鲁奇、德士古、壳牌3种炉型，原料是煤、渣油、天然气，产品是F-T合成油、电或甲醇等。 煤气化技术在我国被广泛应用于化工、冶金、机械、建材等工业行业和生产城市煤气的企业，各种气化炉大约有9000多台，其中以固定床气化炉为主。近20年来，我国引进的加压鲁奇炉、德士古水煤浆气化炉，主要用于生产合成氨、甲醇或城市煤气。 煤气化技术的发展和作用引起国内煤炭行业的关注。“九五”期间，兖矿集团和国内高校、科研机构合作，开发完成了22t/d多喷嘴水煤浆气化炉中试装置，且进行了考核试验。 结果表明：有效气体成分达83%，碳转化率>98%，分别比相同条件下的德士古生产装置高1.5%～2%、2%～3%；比煤耗、比氧耗均低于德士古7%。该成果标志我国自主开发的先进气化技术取得突破性进展。 1.2.2煤气化合成氨 以煤为原料、采用煤气化—合成氨技术是我国化肥生产的主要方式，目前我国有800多家中小型化肥厂采用水煤气工艺，共计约4000台气化炉，每年消费原料煤（或焦炭）4000多万t，合成氨产量约占全国产量的60%。化肥用气化炉的炉型以UGI型和前苏联的Д型为主，直径由2.2m至3.6m不等，该类炉型老化、技术落后。加压鲁奇炉、德士古炉是近年来引进用于合成氨生产的主要炉型。

煤气化制甲醇工艺流程

煤气化制甲醇工艺流程 1 煤制甲醇工艺 气化 a）煤浆制备 由煤运系统送来的原料煤干基（<25mm）或焦送至煤贮斗，经称重给料机控制输送量送入棒磨机，加入一定量的水，物料在棒磨机中进行湿法磨煤。为了控制煤浆粘度及保持煤浆的稳定性加入添加剂，为了调整煤浆的PH值，加入碱液。出棒磨机的煤浆浓度约65%，排入磨煤机出口槽，经出口槽泵加压后送至气化工段煤浆槽。煤浆制备首先要将煤焦磨细，再制备成约65%的煤浆。磨煤采用湿法，可防止粉尘飞扬，环境好。用于煤浆气化的磨机现在有两种，棒磨机与球磨机；棒磨机与球磨机相比，棒磨机磨出的煤浆粒度均匀，筛下物少。煤浆制备能力需和气化炉相匹配，本项目拟选用三台棒磨机，单台磨机处理干煤量43～ 53t/h，可满足60万t/a甲醇的需要。 为了降低煤浆粘度，使煤浆具有良好的流动性，需加入添加剂，初步选择木质磺酸类添加剂。 煤浆气化需调整浆的PH值在6～8，可用稀氨水或碱液，稀氨水易挥发出氨，氨气对人体有害，污染空气，故本项目拟采用碱液调整煤浆的PH值，碱液初步采用42％的浓度。 为了节约水源，净化排出的含少量甲醇的废水及甲醇精馏废水均可作为磨浆水。 b）气化 在本工段，煤浆与氧进行部分氧化反应制得粗合成气。 煤浆由煤浆槽经煤浆加压泵加压后连同空分送来的高压氧通过烧咀进入气化炉，在气化炉中煤浆与氧发生如下主要反应： CmHnSr+m/2O2—→mCO+（n/2-r）H2+rH2S CO+H2O—→H2+CO2 反应在6.5MPa（G）、1350～1400℃下进行。 气化反应在气化炉反应段瞬间完成，生成CO、H2、CO2、H2O和少量CH4、H2S等气体。 离开气化炉反应段的热气体和熔渣进入激冷室水浴，被水淬冷后温度降低并被水蒸汽饱和后出气化炉；气体经文丘里洗涤器、碳洗塔洗涤除尘冷却后送至变换工段。 气化炉反应中生成的熔渣进入激冷室水浴后被分离出来，排入锁斗，定时排入渣池，由扒渣机捞出后装车外运。 气化炉及碳洗塔等排出的洗涤水（称为黑水）送往灰水处理。 c）灰水处理 本工段将气化来的黑水进行渣水分离，处理后的水循环使用。 从气化炉和碳洗塔排出的高温黑水分别进入各自的高压闪蒸器，经高压闪蒸浓缩后的黑水混合，经低压、两级真空闪蒸被浓缩后进入澄清槽，水中加入絮凝剂使其加速沉淀。澄清槽底部的细渣浆经泵抽出送往过滤机给料槽，经由过滤机给料泵加压后送至真空过滤机脱水，渣饼由汽车拉出厂外。 闪蒸出的高压气体经过灰水加热器回收热量之后，通过气液分离器分离掉冷凝液，然后进入变换工段汽提塔。 闪蒸出的低压气体直接送至洗涤塔给料槽，澄清槽上部清水溢流至灰水槽，由灰水泵分别送至洗涤塔给料槽、气化锁斗、磨煤水槽，少量灰水作为废水排往废水处理。 洗涤塔给料槽的水经给料泵加压后与高压闪蒸器排出的高温气体换热后送碳洗塔循环

国内煤气化技术评述与展望

2012年 第15期 广 东 化 工 第39卷 总第239期 https://www.doczj.com/doc/e016013987.html, · 59 · 国内煤气化技术评述与展望 付长亮 (河南化工职业学院，河南 郑州 450042) [摘 要]依据煤气化技术的常用分类标准和评价指标，分析研究了国内所用的煤气化技术的优势与不足。综合考虑原料广泛性、技术先进性、投资成本等因素，认为航天炉干粉煤气化技术具有适应的煤种多、气化效率高、生产能力大、碳转化率高、投资省、操作费用低等优势，在未来的煤化工产品生产中将会得到普遍的应用。 [关键词]煤气化技术；评述；展望 [中图分类号]TQ [文献标识码]A [文章编号]1007-1865(2012)15-0059-02 Review and Prospects of Domestic Coal Gasification Technology Fu Changliang (Henan V ocational College of Chemical Technology, Zhenzhou 450042, China) Abstract: According to common classification standard and evaluation index, advantages and disadvantages of domestic coal gasification technology were analyzed and studied. Considering comprehensively the raw material extensive, technology advanced and investment cost, it was thought that HT-L dry powder coal gasification had the vast potential for future development, because of the more quantity of coal type used, higher gasification efficiency, larger production capacity, higher carbon conversion, lower investment cost. Keywords: coal gasification technology ；review ；prospects 1 煤气化及其评价指标 煤气化指在高温下煤和气化剂作用生成煤气的过程。可简单表示如下： +???→高温 煤气化剂煤气 其中的气化剂主要指空气、纯氧和水蒸汽。煤气化所制得的煤气是一种可燃性气体，主要成分为CO 、H 2、CO 2和CH 4，可作为清洁能源和多种化工产品的原料。因此，煤气化技术在煤化工中处于非常重要的地位。 煤气化反应主要在气化炉(或称煤气发生炉、煤气炉)内进行。不同的煤气化技术主要区别在于所用的气化炉的形式不同。 通常，对煤气化技术的评价主要从气化效率、冷煤气效率、碳转化率和有效气体产率四个方面进行。气化效率衡量原料(煤和气化剂)的热值转化为可利用热量(煤气的热值和产生蒸汽的热值)的情况，是最常用的评价指标，标志着煤气化技术的能耗高低。冷煤气效率衡量原料的热值转化为煤气热值的情况，是制得煤气量多少及质量高低的标志。碳转化率衡量煤中有多少碳转化进入到煤气中，是煤利用率高低的标志。有效气体产率衡指单位煤耗能产出多少有效气体(CO+H 2)，是对煤气化技术生产有价值成分效果好坏的评价。这四个指标不完全独立，从不同的方面反映了煤气化技术中人们最关注的问题。 2 煤气化技术的分类 煤气化的分类方法较多，但最常用的分类方法是按煤与气化剂在气化炉内运动状态来分。此法，将煤气化技术分为如下几种。 2.1 固定床气化 固定床气化也称移动床气化，一般以块煤或煤焦为原料。煤由气化炉顶加入，气化剂由炉底送入。流动气体的上升力不致使固体颗粒的相对位置发生变化，即固体颗粒处于相对固定状态。气化炉内各反应层高度亦基本上维持不变。因而称为固定床气化。另外，从宏观角度看，由于煤从炉顶加入，含有残炭的灰渣自炉底排出，气化过程中，煤粒在气化炉内逐渐并缓慢往下移动，因而又称为移动床气化。目前，国内采用此方法的煤气化技术主要有固定床间歇气化法和加压鲁奇气化法。 2.2 流化床气化 流化床煤气化法以小颗粒煤为气化原料，这些细粒煤在自下而上的气化剂的作用下，保持着连续不断和无秩序的沸腾和悬浮状态运动，迅速地进行着混和和热交换，其结果导致整个床层温度和组成的均一。目前，国内属于此方法的煤气化技术主要有恩德粉煤气化技术和ICC 灰融聚气化法。 2.3 气流床气化 气流床气化是一种并流式气化。气化剂(氧与蒸汽)与煤粉一同进入气化炉，在1500~1900 ℃高温下，将煤部分氧化成CO 、H 2、CO 2等气体，残渣以熔渣形式排出气化炉。也可将煤粉制成 煤浆，用泵送入气化炉。在气化炉内，煤炭细粉粒与气化剂经特殊喷嘴进入反应室，会在瞬间着火，发生火焰反应，同时处于不充分的氧化条件下。因此，其热解、燃烧以及吸热的气化反应，几乎是同时发生的。随气流的运动，未反应的气化剂、热解挥发物及燃烧产物裹挟着煤焦粒子高速运动，运动过程中进行着煤焦颗粒的气化反应。这种运动形态，相当于流态化技术领域里对固体颗粒的“气流输送”，习惯上称为气流床气化。属于此类方法的煤气化技术较多，国内主要有壳牌干粉煤气化法、德士古水煤浆气化法、GSP 干粉煤气化法、航天炉干粉煤气化等[1-3]。 3 国内主要煤气化技术评述 3.1 固定床间歇式气化 块状无烟煤或焦炭在气化炉内形成固定床。在常压下，空气和水蒸汽交替通过气化炉。通空气时，产生吹风气，主要为了积累能量，提高炉温。通水蒸汽时，利用吹风阶段积累的能量，生产水煤气。空气煤气和水煤气以适当比例混合，制得合格原料气。 该技术是20世纪30年代开发成功的。优点为投资少、操作简单。缺点为气化效率低、对原料要求高、能耗高、单炉生产能力小。间歇制气过程中，大量吹风气排空。每吨合成氨吹风气放空多达5000 m 3。放空气体中含CO 、CO 2、H 2、H 2S 、SO 2、NO x 及粉灰。煤气冷却洗涤塔排出的污水含有焦油、酚类及氰化物，对环境污染严重。我国中小化肥厂有900余家，多数采用该技术生产合成原料气。随着能源和环境的政策要求越来越高，不久的将来，会逐步被新的煤气化技术所取代。 3.2 鲁奇加压连续气化 20世纪30年代，由德国鲁奇公司开发。在高温、高压下，用纯氧和水蒸汽，连续通过由煤形成的固定床。氧和煤反应放出的热量，正好能供应水蒸汽和煤反应所需要的热量，从而维持了热量平衡，炉温恒定，制气过程连续。 鲁奇加压气化法生产的煤气中除含CO 和H 2外， 含CH 4高达10 %~12 %，可作为城市煤气、人工天然气、合成气使用。相比较于固定床间歇气化，其优点是炉子生产能大幅提高，煤种要求适当放宽。其缺点是气化炉结构复杂，炉内设有破粘机、煤分布器和炉篦等转动设备，制造和维修费用大，入炉仍需要是块煤，出炉煤气中含焦油、酚等，污水处理和煤气净化工艺复杂。 3.3 恩德粉煤气化技术 恩德粉煤气化技术利用粉煤(<10 mm)和气化剂在气化炉内形成沸腾流化床，在高温下完成煤气化反应，生产需要的煤气。 由于所用的原料为粉煤，煤种的适应性比块煤有所放宽，原料成本也得到大幅度降低。得益于流化床的传质、传热效果大大优于固定床，恩德粉煤气化炉的生产能力比固定床间歇制气有较大幅度的提高。由于操作温度不高，导致气化效率和碳转化率都不高，且存在废水、废渣处理困难等问题。此技术多用于替代固定床间歇制气工艺[4-6]。 [收稿日期] 2012-07-21 [作者简介] 付长亮(1968-)，男，河南荥阳人，硕士，高级讲师，主要从事化工工艺的教学与研究。

煤气化技术的现状和发展趋势

煤气化技术的现状和发展趋势 1、水煤浆加压气化 1.1 德士古水煤浆加压气化工艺（TGP） 美国Texaco 公司在渣油部分氧化技术基础上开发了水煤浆气化技术，TGP 工艺采用水煤浆进料，制成质量分数为60%~65％的水煤浆，在气流床中加压气化，水煤浆和氧气在高温高压下反应生成合成气，液态排渣。气化压力在2.7~6.5MPa，提高气化压力，可降低装置投入，有利于降低能耗；气化温度在1 300~1 400℃，煤气中有效气体（CO+H2）的体积分数达到80％，冷煤气效率为70％~76％，设备成熟，大部分已能国产化。世界上德士古气化炉单炉最大投煤量为2 000t/d。德士古煤气化过程对环境污染影响较小。 根据气化后工序加工不同产品的要求，加压水煤浆气化有三种工艺流程：激冷流程、废锅流程和废锅激冷联合流程。对于合成氨生产多采用激冷流程，这样气化炉出来的粗煤气，直接用水激冷，被激冷后的粗煤气含有较多水蒸汽，可直接送入变换系统而不需再补加蒸汽，因无废锅投资较少。如产品气用作燃气透平循环联合发电工程时，则多采用废锅流程，副产高压蒸汽用于蒸汽透平发电机组。如产品气用作羟基合成气并生产甲醇时，仅需要对粗煤气进行部分变换，通常采用废锅和激冷联合流程，亦称半废锅流程，即从气化炉出来粗煤气经辐射废锅冷却到700℃左右，然后用水激冷到所需要的温度，使粗煤气显热产生的蒸汽能满足后工序部分变换的要求。 1.2 新型（多喷嘴对置式）水煤浆加压气化 新型（多喷嘴对置式）水煤浆加压气化技术是最先进煤气化技术之一，是在德士古水煤浆加压气化法的基础上发展起来的。2000 年，华东理工大学、鲁南化肥厂（水煤浆工程国家中心的依托单位）、中国天辰化学工程公司共同承担的新型（多喷嘴对置）水煤浆气化炉中试工程，经过三方共同努力，于7 月在鲁化建成投料开车成功，通过国家主管部门的鉴定及验收。2001 年2 月10 日获得专利授权。新型气化炉以操作灵活稳定，各项工艺指标优于德士古气化工艺指标引起国家科技部的高度重视和积极支持，主要指标体现为：有效气成分（CO+H2）的体积分数为~83％，比相同条件下的ChevronTexaco 生产装置高1.5~2.0 个百分点；碳转化率＞98％，比ChevronTexaco 高2~3 个百分点；比煤耗、比氧耗均比ChevronTexaco 降低7％。 新型水煤浆气化炉装置具有开车方便、操作灵活、投煤负荷增减自如的特点，同时综合能耗比德士古水煤浆气化低约7％。其中第一套装置日投料750t 能力新型多喷嘴对置水煤浆加压气化炉于2004 年12 月在山东华鲁恒升化学有限公司建成投料成功，运行良好。另一套装置两台日投煤1 150t 的气化炉也在兖矿国泰化工有限公司于2005 年7 月建成投料成功，并于2005 年10 月正式投产，2006 年已达到并超过设计能力，目前运行状况良好。该技术在国内已获得有效推广，并已出口至美国。 2、干粉煤加压气化工艺 2.1 壳牌干粉煤加压气化工艺（SCGP） Shell 公司于1972 年开始在壳牌公司阿姆斯特丹研究院（KSLA）进行煤气化研究，1978 年第一套中试装置在德国汉堡郊区哈尔堡炼油厂建成并投入运行，1987 年在美国休斯顿迪尔·帕克炼油厂建成日投煤量250~400t 的示范装置，1993年在荷兰的德姆克勒（Demkolec）电厂建成投煤量2 000t/d 的大型煤气化装置，用于联合循环发电（IGCC），称作SCGP 工业生产装置。装置开工率最高达73％。该套装置的成功投运表明SCGP 气化技术是先进可行的。 Shell 气化炉为立式圆筒形气化炉，炉膛周围安装有由沸水冷却管组成的膜式水冷壁，其内壁衬有耐热涂层，气化时熔融灰渣在水冷壁内壁涂层上形成液膜，沿壁顺流而下进行分

现代煤气化技术发展趋势及应用综述_汪寿建

2016年第35卷第3期CHEMICAL INDUSTRY AND ENGINEERING PROGRESS ·653· 化工进展 现代煤气化技术发展趋势及应用综述 汪寿建 （中国化学工程集团公司，北京 100007） 摘要：现代煤气化技术是现代煤化工装置中的重要一环，涉及整个煤化工装置的正常运行。本文分别介绍了中国市场各种现代煤气化工艺应用现状，叙述汇总了其工艺特点、应用参数、市场数据等。包括第一类气流床加压气化工艺，又可分为干法煤粉加压气化工艺和湿法水煤浆加压气化工艺。干法气化代表性工艺包括Shell炉干煤粉气化、GSP炉干煤粉气化、HT-LZ航天炉干煤粉气化、五环炉（宁煤炉）干煤粉气化、二段加压气流床粉煤气化、科林炉（CCG）干煤粉气化、东方炉干煤粉气化。湿法气化代表性工艺包括 GE水煤浆加压气化、四喷嘴水煤浆加压气化、多元料浆加压气化、熔渣-非熔渣分级加压气化（改进型为清华炉）、E-gas(Destec)水煤浆气化。第二类流化床粉煤加压气化工艺，主要有代表性工艺包括U-gas灰熔聚流化床粉煤气化、SES褐煤流化床气化、灰熔聚常压气化（CAGG）。第三类固定床碎煤加压气化，主要有代表性工艺包括鲁奇褐煤加压气化、碎煤移动床加压气化和BGL碎煤加压气化等。文章指出应认识到煤气化技术的重要性，把引进国外先进煤气化技术理念与具有自主知识产权的现代煤化工气化技术有机结合起来。 关键词：煤气化；市场应用；气化特点；参数数据分析 中图分类号：TQ 536.1 文献标志码：A 文章编号：1000–6613（2016）03–0653–12 DOI：10.16085/j.issn.1000-6613.2016.03.001 Development and applicatin of modern coal gasification technology WANG Shoujian （China National Chemical Engineering Group Corporation，Beijing100007，China）Abstract：Modern coal gasification technology is an important part of modern coal chemical industrial plants，involving stable operation of the entire coal plant. This paper introduces application of modern coal gasification technologies in China，summarizes characteristics of gasification processes，application parameters，market data，etc. The first class gasification technology is entrained-bed gasification process，which can be divided into dry pulverized coal pressurized gasification and wet coal-water slurry pressurized gasification. The typical dry pulverized coal pressurized gasification technologies include Shell Gasifier，GSP Gasifier，HT-LZ Gasifier，WHG (Ning Mei) Gasifier，Two-stage Gasifier，CHOREN CCG Gasifier，SE Gasifier. The typical wet coal-water slurry pressurized gasification technologies include GE (Texaco) Gasifier，coal-water slurry gasifier with opposed multi-burners，Multi-component Slurry Gasifier，Non-slag/slag Gasifier (modified as Tsinghua Gasifier)，E-gas (Destec) Gasifier. The second class gasification technology is fluidized-bed coal gasification process. The typical fluidized-bed coal gasification technologies include U-gas Gasifier，SES Lignite Gasifier，CAGG Gasifier. The third class gasification technology is fixed-bed coal gasification process. The typical fixed-bed coal gasification technologies include Lurgi Lignite 收稿日期：2015-09-14；修改稿日期：2015-12-17。 作者：汪寿建（1956—），男，教授级高级工程师，中国化学工程集团公司总工程师，长期从事化工、煤化工工程设计、开发及技术管理工作。E-mail wangsj@https://www.doczj.com/doc/e016013987.html,。

煤气化工艺资料

煤化工是以煤为原料，经过化学加工使煤转化为气体，液体，固体燃料以及化学品的过程，生产出各种化工产品的工业。 煤化工包括煤的一次化学加工、二次化学加工和深度化学加工。煤的气化、液化、焦化，煤的合成气化工、焦油化工和电石乙炔化工等，都属于煤化工的范围。而煤的气化、液化、焦化（干馏）又是煤化工中非常重要的三种加工方式。 煤的气化、液化和焦化概要流程图 一.煤炭气化

煤炭气化是指煤在特定的设备内，在一定温度及压力下使煤中有机质与气化剂（如蒸汽/空气或氧气等）发生一系列化学反应，将固体煤转化为含有CO、H2、CH4等可燃气体和CO2、N2等非可燃气体的过程。 煤的气化的一般流程图 煤炭气化包含一系列物理、化学变化。而化学变化是煤炭气化的主要方式，主要的化学反应有： 1、水蒸气转化反应C+H2O=CO+H2 2、水煤气变换反应CO+ H2O =CO2+H2 3、部分氧化反应C+0.5 O2=CO 4、完全氧化（燃烧）反应C+O2=CO2 5、甲烷化反应CO+2H2=CH4 6、Boudouard反应C+CO2=2CO 其中1、6为放热反应，2、3、4、5为吸热反应。 煤炭气化时，必须具备三个条件，即气化炉、气化剂、供给热量，三者缺一不可。 煤炭气化按气化炉内煤料与气化剂的接触方式区分，主要有： 1) 固定床气化：在气化过程中，煤由气化炉顶部加入，气化剂由气化炉底部加入，煤料与气化剂逆流接触，相对于气体的上升速度而言，煤料下降速度很慢，甚至可视为固定不动，因此称之为固定床气化；而实际上，煤料在气化过程中是以很慢的速度向下移动的，比

较准确的称其为移动床气化。 2) 流化床气化：它是以粒度为0-10mm的小颗粒煤为气化原料，在气化炉内使其悬浮分散在垂直上升的气流中，煤粒在沸腾状态进行气化反应，从而使得煤料层内温度均一，易于控制，提高气化效率。 3) 气流床气化。它是一种并流气化，用气化剂将粒度为100um以下的煤粉带入气化炉内，也可将煤粉先制成水煤浆，然后用泵打入气化炉内。煤料在高于其灰熔点的温度下与气化剂发生燃烧反应和气化反应，灰渣以液态形式排出气化炉。 4) 熔浴床气化。它是将粉煤和气化剂以切线方向高速喷入一温度较高且高度稳定的熔池内，把一部分动能传给熔渣，使池内熔融物做螺旋状的旋转运动并气化。目前此气化工艺已不再发展。 以上均为地面气化，还有地下气化工艺。 根据采用的气化剂和煤气成分的不同，可以把煤气分为四类：1.以空气作为气化剂的空气煤气；2.以空气及蒸汽作为气化剂的混合煤气，也被称为发生炉煤气；3.以水蒸气和氧气作为气化剂的水煤气；4.以蒸汽及空气作为气化剂的半水煤气，也可是空气煤气和水煤气的混合气。 几种重要的煤气化技术及其技术性能比较 1.Lurgi炉固定床加压气化法对煤质要求较高，只能用弱粘结块煤，冷煤气效率最高，气化强度高，粗煤气中甲烷含量较高，但净化系统复杂，焦油、污水等处理困难。 鲁奇煤气化工艺流程图