25. Aeroengine monitoring

Fig. "Connection of operation costs and reliability" (Lit. 25.1-1): The left diagram shows schematically the connection between reliability and costs. Expecially in the civil sector the operator demands a high reliability at the possible lowest costs through the foreseen operation time. These rise, depending from the claim at the reliability. This is composed as follows:

• Monitoring costs.
• Introduction of improvements (hardware, work, organisation).
• Rising effort for inspections.
• Increased effort for repairs.

For problems and failures, sufficient reliability acts as „problem costs“. The highest effect is observed at the beginning of an improvement of the reliability. The higher already the reliability, the lower is the cost advantage that can be expected. Typical cost intense problems are:

• In flight shut down (IFSD) of the aeroengine e.g., by dumping of fuel in emergency cases.
• Bringing `stranded' airplanes back to operation and care for the passengers.
• Costs of maintenance and overhaul of material, repair, inventory/equipment.

The reliability is operator specific caused by influences like

• Differences in the operation (e.g., short haul or high, hot airfields),
• Organisation: Concept, procedure,
• Standards,
• Company policy.

From the cooperation of „problem costs” and costs of the reliability the curve of the „total costs“ arises. It passes at a certain cost optimal reliability a cost minimum and is usually pursued.
The optimization of these influences enables to maintain or even improve the reliability during sinking costs. Such an approach must occur carefully in an iterative process, because of the complexity of the influences and reciprocative influencing of each other.
For an optimal cost/reliability relationship the following activities are of special interest:

• Introduction of improvements against a design weaknesses e.g., for a better deterioration behaviour (volume 2, Ill. 7.0-2) and a extension of the parts lifetime.
• Monitoring and controlling of the applied technics.
• Organisation, concept and practices during maintenance/overhaul.

The diagramm right shows the costs per opreation hour for aeroengines of the same type but significant different introduction dates. It can be seen, how the operator benefits from a markedly lower cost level of later introduced aeroengine versions.
The rise of costs is especially steep during introduction of a new aeroengine type. After several years (here about four years) a maximum,in which the weak points are fully effective, is reached. Here remedies are not yet sufficient introduced. After once more several years, the curve levels. This shows, that a certain reliability optimum was reached with the remedies. From these improvements obviously benefit the later aeroengine modifications. Also profits the cost trend from the whole fleet of new and old aeroengine versions.

Fig. "Reasons for use of engine monitoring" (Lit. 25.1-3): These informations refer to the proving of a monitor system of a big military helicopter.
The gas temperature crucial influences the lifetime of the hot parts. It enables conclusions at the condition of the aeroengine. With worsened efficiencies (deterioration, volume 2, Ill. 7.0-2) the gas temperatur rises at constant power.
Usually it is limited on the basis of displays from exhaust gas temperature at the cockpit instruments. Maximimum limiting temperatures must only be expected at maximum power, i.e. during start. The operating stays most the time below these temperature limits. With this, the chance to indicate failures with the passing of temperature limits at the cockpit instruments, is rather low (diagram left).

An information to the maintenance does not take place before the pilot recognizes a problem at maximum power or markedly power drop.
The use of electronic memories enables the digital recording of all relevant data from the flight aud the aeroengines. To these belong besides further gas temperatures

• rotation speeds of the main shafts.
• Output torque.
• Temperature and pressure of the environment.
• Pressures in the main gas stream.
• Vibrations.

These data are only analyzed with the use of computers at once in the flight (real time) and afterwards on ground (during maintenance). So beginning problems can better be identified. This is also true for the information of the pilot. This increases the flight safety markedly.

To this here some examples from the testing of the monitoring system:

• Clearance of the causative influences of compressor surge with aeroengine fire during flight. It could be recognised, that this problem already occurred some minutes before the crew noticed it. With this ground runs with extensive costs could be avoided.
• Identification of an aeroengine overload caused by indication error of the cockpit instruments for gas temperature and rotation speed of the gas producer.
• The identification of a drop of the compressor efficiency succeeded from the evaluation of the output torque. Initially the fuel control unit and the actuator of the variable guide vanes have been unsuccessfully exchanged, based on the monitoring data. A further investigation showed, that in the compressor a foreign object damage (FOD) with blade fracture existed.

The system enabled many further important findings:

• Operation time,
• Data (e.g., output torque) from which trends can be read.
• Reached maximum data of temperatures and rotation speeds. To these belong also times at which the recorded parameters had surpassed initialised data.
• Power output: Continuously recording of the differences between the two aeroengines of the helicopter.
• Run down times of the gas producer.
• Determination of the consumption of cycles (LCF, Fig. "Failure intercepted by health nonitoring"). Here rotation speeds and temperatures in hight and variation, in time as important parameters, enter the calculation.

A long time analysis enables the safe and cost effective exchange of components. Especally concerned are rotor disks, which reached their lifetime (diagram right). Thereby, the mission specific operation profile is of especial inportance for the `lifing'.
With suitable algorithms so also a monitoring of the turbine blading with regard to thermal fatigue and/or oxidation is possible.

Fig. "Ways of aeroengine monitoring"(Lit. 25.1-4): Engine condition monitoring can take place task specific in different ways (chart below).
Monitoring of aeroengines during flight: It serves the identification and analysis of problems as well as the determination of the component life time. For problem analysis, the data are today transferred by broadcasting for the analysis by computer and/or evaluation on ground.
Usually for the analysis serves a continual recording of parameter data, we speak from „gas path analysis”. These document the aerodynamic and thermodynamic behavior (e.g., compressor behavior/condition).
A monitoring in „real time“ can be registrated by the pilot with the help of the cockpit instruments and, if necessary,by warning signals.

Also the displays of electrical magnetic chip detektors (Fig. "Monitoring particle formation in oil"), which control the oil circuit continuously for magnetic particles, can be couted to the monitoring processes during flight.

The monitoring of aeroengines takes place in time intervals at the ground, often directly after landing. the stored data can be called through a line or an integrated memory/storage or from the removed memory.
Also other processes can be counted to the monitoring of an aeroengine. They stay in connection with the monitoring of deteriorations or the review of failures („health monitoring”, Fig. "Model based diagnosis" and Fig. "Disk lifting system concept for rotors"). The progression of failures can only be controlled for sufficient slow failure mechanisms.

• Borescope inspections can serve as routine determination of the condition from components. To these belong erosion at compressor blades (Fig. "Borescope indications in a compressor") or thermal fatigue at hot parts (Fig. "Borescope findings at turbine blades and vanes").
• Non destructive testing like X-ray (Ill. 25.2.2.2-5), ultrasonic (Ill. 25.2.2.2-9), penetrant inspection and eddy current are mostly used in connection with special actions under safety aspects.

Test runs on the ground in the as mounted condition mostly serve the problem analysis or function proofs. To these belong seal runs, which are used to test the systems of oil and fuel for leaks (Ill. 19.2-2.2 and Fig. "Influencing the O-ring sealing effect").
Checks in the framework of maintenance play an important roll in the aeroengine monitoring. Mostly used are controls of the filters and sieves in the systems of oil and fuel (Ill. 22.2.2-2.1 and Ill. 22.3.3.1-2). Thereby the control for deposits respectively chips (Fig. "Monitoring by magnetic detector") is for the condition of the concerned systems an extremely significant tool.
To this belongs also the monitoring of the lubrication oil (Fig. "Informations of oil analysis") for aging and signs of problems in the oil system. For this also magnetic chip detectors, respectively magnetic plugs (Fig. "What deposits of magnet plugs can tell" and Ill. 22.3.4-8) are used.
With these, failures on components of magnetic materials can be discovered or identified. These are primarily anti friction bearings and gear wheels.

Fig. "Monitoring systems" (Lit.25.1-2 and 25.1-5 ): The scheme of a monitoring system derives from the 90s (ground based engine monitoring). It provides monitoring data for the aeroengines of a large two engined airliner type. During flight, already a primarily analysis of the data takes place. For this a printout was automatically read from an OCR program. These data served a later analysis at the ground. In newer systems, data are transferred from the flight by radio for the analysis on ground (Fig. "Ways of aeroengine monitoring"). If necessary feedbacks take place.

The aeroengine data from the fleet of the operator are collected and secured. They serve the analysis under the aspects of reliability, safety and effectivity/cost minimization. This optimization potential is especially interesting for the logistics. For example the „story of the aeroengine“ serves the plannings of the OEM. Concerned are plannings of overhaul, occupancy of the test rigs and work (overhaul, spare parts, repair).
The measurement data of the acceptance run after the overhaul will be compared with those already available. So measurements at the test rig and efficiencies of the modules can be evaluated.

Monitoring data also serve the daily central problem analysis (trouble shooting).
The chart below shows a military monitoring system. Here the measurement data (23 analogous continuous + 5 temporary) enter the electronic control unit of the fighter aeroengine. Additionally data from outside the aeroengine (anti icing, lubrication) and from the avionics are available for analysis. These are used for the following purposes.

• Diagnosis in real time and for analysis of failures/problems. These are compared with stored criteria. If necessary, a feedback to the memory in the airplane and the instruments of the pilot takes place. More than 80 warning signals are possible.
• Tracing of the remaining lifetime: The data are collected and stored from the processor of the aeroengine monitoring system. With this a ground based system traces the lifetime limited parts for overhaul and logistics (spare parts).
• Trend analysis: During start preselected data are stored. These will be processed by a computer on ground after every flight . A printout enables the visual evaluation of the aeroengine condition (engine health).

First experiences in the 80s, to introduce the system, showed the following advantages:
Flight: The safety, especially for single engined fighters could be markedly improved. This primarily was owned to early warning for the pilot during problems in the oil system. So he could react at the situation more targeted. The experiences show, that false alarm signals could be reduced by a power of ten (chapter 19.2.1).

Analysis of problems/failures: The statements got more certain. That is concerned besides mechanical problems (e.g., at pumps and actuating systems) especially for elektronic (e.g., sensors/probes, ignition devices, control units). For this the analysis program is of crucial importance (Fig. "Increased risk of corrosion during military use").

Advantages for the OEM:

With the help of the data it was possible for him, to identify operation problems and to solve those with targeted measures.

Fig. "Health monitoring structure" (Lit. 25.1-5): Accident statistics (1998) in the civil aviation show, that in 12 % of all cases, problems with the propulsoin have been involved. About 6 % of these could be traced back at causative mechanical problems and maintenance faults of the propulsion (volume 1, Ill. 2-7 and Ill. 2-9). 18 % of the 12% share of the aeroengines have been connected with fragment exit (uncontained). At 15 % the aeroengine problem was combined with an unsuitable reaction of the pilot.

• During too high rolling speed aborted take-offs (volume 1, Ill. 2-12 and Ill. 2-13).
• Loss of control of the airplane, caused from unnoticed thrust asymmetry.
• Retraction of power or shut down of the wrong aeroengine during flight.
• Problems during recognition of a malfunction or the introduction of suitable measures.
  Exiting fragments (uncontained) origin from rotors, primarily of disks. The relative high percentage with fragment exit has different causes (volume 1, Ill. 2-9 and Ill. 2-10):
• Mistakes during overhaul.
• Cyclic fatigue (LCF),
• Foreign object influence.
• Mechanical malfunctions.

In many cases precedes the disk failure a slow cyclic crack growth (Fig. "Disk fracture during test run" and volume 3, Ill. 12.2-1).

There are absolutely cases, which could be prevented with a continuous monitoring of the aeroengine for failure relevant deviations (health monitoring). Thereby it is important, to indicate first signs sufficient early before a dangerous failing . For this offer itself the data of the condition monitoring (Fig. "Ways of aeroengine monitoring" and Fig. "Monitoring systems").

At the health monitoring, false alarms are especially problematic, because the trigger thresholds, due to the recognizability and with this sufficient safety, should be adjusted as low as possible. So the probability of a false alarm increases by uncritical signals. The picture shows the model (Fig. "Model based diagnosis") of a „health monitoring” systems, that at this time obviously was in the development. From this, it is expected the avoidance of the mentioned problems and a sufficient certain realization chance. This model is distributed on three levels.

In the most cases, fractures of disks preceded by a relatively slow cyclic crack growth. This crack acts at the unbalance. So it is obvious to use the effect as warning signal (Fig. "Identifying a cracked rotor by vibration analysis"). This takes place by vibrations analysis (Fig. "Elements of a vibration monitoring") of the signals from the acceleration probes (Fig. "Condition indicator using vibration sensors"). Most sensitive are sensors/probes, positiooned directly at the outer ring of the bearing. This method is already used successful at industrial gas turbines and cyclic spin test rigs. They have additional the potential to record problems like bearing failures, locally tip rub of rotorblades at the circumference and other unbalances (e.g., oil accumulations, volume 3, Ill. 12.6.3.1-12).

Fig. "Identifying a cracked rotor by vibration analysis" (Lit. 25.1-5): At least in the development are processes, which should enable to suggest from the vibration characteristics of a rotor at possible cracks in disks or rings. Thereby the vibrations are measured at a suitable location with acceleration probes. For this, the outer ring of the main bearing from the rotor in question offer itself.

The model of this monitoring process can be described as follows:
An always existing allowable „original“ rotor unbalance (assembly influence, production tolerances, abrasion/wear) produces a radial force. This increases quadratic with the rotation speed (sketch left). A crack in the outer zone of the disk openes under the tangential stresses from the zentrifugal forces (sketch right). The gaping produces at the large diameter a corresponding unbalance at the opposite side. This grows with the rotor speed, different (faster by the change of streess distribution/stress rearrangement, plastic deformation, crack growth) than the still present „original” unbalance. Both unbalances superimpose to a resultant. Its direction and size differs, depending of the rotation speed, from the „original“ unbalance. The consequences are a shift of the phase and amplitude of the shaft vibrations. This is sufficient for a computer supported analysis as early sign for a crack.

Requirement for a sufficient early detection of the crack is, that it produces an analyzable change of the unblance. In fact this can be expected for a crack in the outer zone/rim of a disk with a sufficient large diameter and elastic flexibility. However, this does not apply for thick/stiff disks with small diameter. They are typically for turbine wheels (mostly integrally casted) of small gas turbines (e.g., helicoper aeroengines). Does a crack start in the region of the hub, the chance of an identification is probalbly rather low, because of the high stress level. At this high stress level already a small crack leads, even with a minimum unbalance,m to a burst. Additionally with the flat stress gradient a fast crack growth must be expected. So an early enough premonition is not guaranteed (volume 3, Ill. 12.2-10).

Such a „health monitoring” (Fig. "Health monitoring structure") can be promising, at least in some cases. This shows the disk fracture in Fig. "Disk fracture during test run". Similar conditions are existent at cases in volume 4, Ill. 16.2.1.1-9.1 and Ill. 16.2.1.6-15. The crack started always at the rim or from an exterior zone.
At industrial gas turbines, the described monitoring process has already proved. It is thinkable, that it can be successfully used, at least in suspicious cases,m during test runs on wing or at test rigs.
This process is not only interesting for aeroengines on test rigs, but also for spintest rigs for better crack analysis/investigation results (fracture analysis) without damges from the fracture.

Fig. "Model based diagnosis" (Lit. 25.1-4, Lit. 25.1-5 and Lit. 25.1-6): The model based (model based controls and diagnostics = MBCD) can be a powerful and information certain tool of the aeroengine monitoring.

Such systems have proved since years at industrial gas turbines. The advances of the computing technology now enable aso the use in airplanes.
The architecture is complicated. It needs advanced software and hardware components. Above this, extensive knowledge/experience about certain influences is needed.
This process is not only interesting for aeroengines on test rigs, but also for spintest rigs for better crack analysis/investigation results (fracture analysis) without damges from the fracture. The system uses a software based aero-thermodynamic aeroengine model („2“). This continuous working, electronic evaluation algorithm, respectively monitoring filter („2”), adjusts the model. So unusual behaviour/efficiencies of components, drift of sensors/probes and actuation systems can be recorded.

With this it gets possible

• to trace in real time tends of the aeroengine condition („health“),
• to combine sensor properties for a sensor test (chapter 19.2.1) and
• determine not directly measurable aeroengine parameters, like the thrust or threshold values for the monitoring logic of components.
• Increase of the safety through in time identification and locating of problems and failures. Especially aeroengine in flight shut downs (IFSDs) and compressor surge must be prevented. The logic works within the control system („1”). This should be adjustable to react at results from the check of the sensors/probes and the error detection.

Fig. "Engine health management" (Lit. 25.12-6): In a greater term we speak from engine health management (EHM). It servrs for the identification of failures and problems in time. So an unacceptable risk can be avoided, land suitable cost effective measures can be taken. For this, systems of the aeroengine can be monitored for certain problems and failures. In contrast during condition monitoring trends are observed. In this case we speak about so called health monitoring.

The potential of the process can be realised from the chart. From a networking/combination of the informations (data fusion, Fig. "Data Fusion and information fusion"), from different monitorings and the single steps of the EHM, certain and detailed conclusions can be expected. Particularly the consequences of faults from measurement and transmission errors of the data from sondes/probes may be limited.
This process finds today an application primarily at four main fields:

• Monitoring of the main flow: Gas path performance monitoring or briefly gas path analysis (GPA, Fig. "Reasons for use of engine monitoring") and trend analysis (performance trending). This has the potential to identify problems, respectively faults and failures.
  Typical examples are:
  Worn air seals/gas seals (volume 2, chapter 7.2).
  Too large tip gaps (volume 2, chapter 7.1.2).
  damages of blades in compressor and turbine:
  Erosion (volume 1, chapter 5.3.2),
  foreign object damages (volume 1, chapter 5.2.1.1),
  thermal caused failures (volume 3, chapter 11.2.3.2),
  crack formation in rotor disks (Fig. "Identifying a cracked rotor by vibration analysis").
• Oil- and debris/chips monitoring (Fig. "Informations of oil analysis").
• Vibration monitoring (Fig. "Identifying a cracked rotor by vibration analysis")
• Usage and life monitoring (Fig. "Disk lifting system concept for rotors").

Till now EHM suffers itself from problems, in spite of its high potential.

• False alarms are obviously the biggest problem.
• A further problem is an unsufficient data collection and data storage.
• To this come the costs of such a system.

Traditionally, the trigger thresholds from statistical evaluated data of the fleet operation, serve the failure identification (diagram above left). Further developments use additionally physical (theoretical) models (Fig. "Model based diagnosis"), the tracing of individual aeroengine data, predicted algorithms and systematic decision supports. For such an evaluation the measurement data are compared with thr model results.
The EHM seems till now to identify and define its own problems. Fig. "Illness symptoms for health monitoring" contains to this a compilation of definitions of impotant terms from the EHM field. Usually these are used in the expert literature.
In spite of the difficulties, false alarms can be minimized with an improvement of the identification of a deviation and the use of standard trigger threshold. For this are promising:

• Characteristics,
• automatic failure diagnosis and the
• Combination of theory, knowledge/experience and tests for the creatiion of „failure libraries“.
• Combination of rules from expert systems with suitable computer analysis.
• From the available knowledge, better understanding of the connections .

Fig. "Illness symptoms for health monitoring" (Lit. 25.1-6): Obviously the effectivity and certainty of the health monitoring suffers under different definitions, respectively ambiguities of important terms. Because of this, in the following important definitions are specified. Abnormal condition is a measurably deviation of the current operation condition from the normal. A measurable deviation lays outside the specified tolerance. Because this is a difference, such a deviation is called „residual”.

We distinguish between three conditions:

• Fault is a known respectively identified and repeatable unusual condition. It is considered in the model of the monitoring system (Fig. "Health monitoring structure").
  
  In this case a degradation is for the first time repeatable certain recorded (diagram left). This allows an alarm at the operator.
  In many cases a fault is less engraving than a failure. With this the possibility exists, to still operate the system safe for some time under observation (tracing of the utilisation). Thereby a sufficient safety margin, up to the development of a failure must be guaranteed.
  A fault has usually a specific (main) cause/root cause.
  
  Fault symptom is a coincident feature of sveral individual faults. It can fit to one or several failure modes (failure mechanisms), which influence the functionality. If this is quantitative determinable, we speak about functionality or capability. Is this plotted on the Y axis (diagram left), the curve shows the failure progression path.
• Failure is a function outage as an unusual condition.
• Anomaly is an unknown, respectively before never seen, not fully understood or only temporary unusual condition. In the model it is not considered.

Diagnosis/ diagnostics refer to processes to find unusual conditions.
Prognosis stay for the process, which enables a conclusion at future effects of the unusual condition.
The severity of a fault rises with the frequentness in a system or an equipment (diagramm left).
The diagram shows the position of fault and failure in the failing/deterioration process.
From a precursor we speak, when first signs of the unusual condition are observed. It is an It is an indication for the sensitivity of the monitoring systems (Fig. "Engine health management").
The trajectory of degradation shows with the drop of the operation property/performance through the lifetime. With the failure it reaches than the unacceptable condition.

Fig. "Failure intercepted by health nonitoring" (Lit. 25.1-11): In this case the data are used from the `digital electronic engine control' (= DEEC) for the trend monitoring. The service firm/company analysis with a `gas path analysis'

• program data sets of thousands of aeroengines. These analysis are reviewed

by specalists for hints at failures on components like disks.
Most frequent problems for the shown aeroengine type are in this succession:

• Rubbing of the turbines.
• Leakage in the bleed system.
• Faulty displays (!), see Fig. "Engine health management" and chapter 19.2.1.

In the shown case the data analysis showed changes of important parameters from the gas producer. These can already more than halve the lifetime of the hot parts (volume 3, Ill. 12.5-4). Failures in the hot part area showed during the disassembly of the aeroengine. The estimated repair costs correlated several middle class cars. If the failure would not have been identified in time by monitoring, it could have been tripled. Further costs would have been the expected result of a sudden failing during operation.

Fig. "Monitoring parameters" (Lit. 25.1-2 und Lit. 25.1-4): This example should give an impression of the data from a monitoring. Concerned is a frequent aeroengine type which had at the beginning not yet a `full authority digital electronic control (= FADEC). The „ground based engine monitoring“ (Fig. "Model based diagnosis") serves the „on condition” overhaul. It can be seen, that the data can be assigned four fields. These cover the whole aeroengine with its surrounding. This system has the potential to identify as well trends, as also failures (grey field) and problems.

With the introduction of a FADEC, electronic components of the system could be saved. These executed the signals and adopted the connection to the fuselage.

Fig. "Disk lifting system concept for rotors" (Lit. 25.1-6 and Lit. 25.1-7): The determination of the lifetime consumption respectively residual lifetime of aeroengine components with help of the monitoring is also called `life usage monitoring'. If the monitoring of rotor disks and rings is concerned, we speak about a disk lifing system. Thereby the data of the 'condition monitoring' are used.

Such a system needs a reliable mathematical model (algorithm) for the failure mechanism to look on. To this belong sure knowledges of the material behavior like crack development and crack propagation. For disks are this primarily the cyclic load (LCF, volume 3, chapter 12.6) from start and shut down processes (main cycles) and power changes (mini cycles). Also static loads (creep, volume 3, chapter 12.5) in the stationary operation condition can be considered.

Requirement for the determination of the remaining rest lifetime of a rotor disk is the knowledge of the save lifetime under design according operation conditions.

The frame above shows an „early“ approach of a `disc lifing systems'. It serves the determination of the design, according lifetime potential. Thereby, rather a vague evaluation is concerned. Characteristical is the OEM specific use of cyclic spin tests. These can also serve the safeguard with a determination of the remaining lifetime. For this, used disks of aeroengines, are sampled („sampling”) and are cyclic spun (grey field above).

The frame below contains a new version of the `Disc Lifing Systems' with improvements. These refer primarily at the guarantee of quality from the new part (grey field below).

• Quantitative deterioration standards of all relevant influences like welds (disk joints, labyrinth tips).
• `Risk examinations' of the production processes.
• Adjusting of the design at high loads, respectively high strength materials. For this, properties like growth capable failure size, crack growth, crack propagation and critical crack length must be considered (volume 3, Ill. 12.2-3 and Ill. 12.2-7).

With this a recommendation for an `on-condition principle' of individual disks is given. This also includes the possibility of an increase of the lifetime. However for this it must be restrictive mentioned, that the 'retirement for cause' principle (black field) till now, is not used (chapter 25.2-1). This is due to not sufficient sure non destructive testing methods.

Fig. "Data Fusion and information fusion" (Lit. 25.1-8 and Lit. 25.1-9): Monitoring of data, which surpass direct sensor measurements, is also termed propulsion health mangement (PHM). The displayed system enables diagnosis and predictions.
It uses:

• Experiences from maintenance, overhaul and operation as well as the
• Analysis of theoretical models and a
• Failure identifuication modul.
• Also influences at the logistics can be contained in this term. To these belong recommendations for maintenance and measures for identifued problems.

Data fusion is used for the summary of data for monitoring. These come from very different sources (datamining, upper frame). Plausibility considerations and free of contradiction as well as an independent examination increases the analysis certainty for failures and problems. This is especially true for exchangeable aggregates (`line replaceable units =LRUs diagnostics) like gears, pumps and mechanical control systems. Depending from the problem/failure; also statements about the point of time for an exchange can be made (prognostics). In this case we speak about iIntelligent engine health monitoring. Data Fusion can be composed from several steps:

• Sensor level fusion is the combination of each other influencing data. To this counts oil pressure and exhaust gas temperature (e.g., during an oilfire or bearing failure).
• Feature level fusion stands for the combination of features from results/conclusions of independable data analysis.
• Decision level fusion uses the `feature level fusion' for the assessment of failures or for mantenance instructions.

Figure "Engine health monitoring with data fusion" (Lit. 25.1-10): Here a monitoring system for helicopters in military use is concerned (vibration management enhancement program =VMEP). This consists of three `blocks' and monitors at the aeroengines especially vibrations (1st). Measurement data during operation are analyzed on ground by a computer program (2nd), which recommends corresponding maintenance steps. In a 3rd `block' data are stored.

In this manner an identified indication error will be evaluated. This takes place with the assessment of the relation from signal to background disturbance (signal-to-noise ratio = SNR). If the fault is advanced, respectively a failure already occurred, the SNR is high and the failure signal easy to identify. Corresponding the failure definition (Fig. "Illness symptoms for health monitoring") a reaction must take place at once.

Beginning failures are difficult to identify, because of the weak signal. With this the increased probability of a false alarm exists Remedy can bring the combination of weak signals from several sensors with a suitable algorithm. We speak of `anomaly detectors'.

The diagrams have been compiled for three `condition indicators' (= CIs) for the aeroengine of a fighter helicopter. Obviously analysis of measurement data from acceleration sensors/probes in the fleet are concerned. In these, graphically overall evaluations and data of individual helicopters and components are contained. The combination of the three CI leads to the nine diagrams. A diagonal from below left to above right reveals the frequency distribution of the data from the individual sensors. Striking outliers of the measurement values (grey circle) occur at two sensors. Thew are identified by the analysis system as anomalies and therefore not evaluated as mismeasurement.
Other outliers don't show this pecularity.