You are on page 1of 13

Proceedings of Turbo Expo 2005: 50th ASME International Gas Turbine & Aeroengine Technical Congress June 6-9,

2005, Reno-Tahoe, Nevada

GT2005-68625
Recent Advancements in Aircraft Engine Health Management (EHM) Technologies and Recommendations for the Next Step
Link C. Jaw
Scientific Monitoring, Inc. Scottsdale, Arizona, U. S. A.

ABSTRACT This paper summarizes the findings of a survey of recent advancements in aircraft Engine Health Management (EHM) technologies. The survey has been motivated by the desire to understand new technologies and application trends in EHM, especially in the last few years, when EHM related research and development (R&D) has increased significantly. Although the R&D primarily covers four areas: system partition, system architecture, EHM functionalities, and algorithmic approaches, the latter two have represented the majority of the published work; hence they are the focus of this paper. While recent advancements are providing building blocks for continued maturation of EHM technologies in the future, the survey has revealed a fundamental inconsistency in defining and representing EHM problems. This inconsistency creates barriers in exchanging EHM-related ideas and results; it also undermines the effectiveness of multi-organizational cooperation. Hence another purpose of this paper is to recommend a consistent approach to treating the problems that EHM is trying to address. Finally, the survey recognizes the need for a unified framework for comparing the performance of various algorithms for solving various types of EHM problems. The author hence suggests that an industry review, modeled after the industry review for multi-variable engine controls in late 1970s, be held. This EHM Industry Review is aimed at defining a series of theme problems for EHM and invite experts in the industry and academia to apply their expertise to solving these problems. The Industry Review will be concluded with a conference to present the results and sharing experience. Traditionally, engine condition monitoring has led the way for condition-based maintenance and health management

technologies because of the safety and dispatch requirements of aircraft engines. Hence, the author believes that the results, conclusions, and recommendations presented in this paper can be generalized to all types of equipment, systems, and vehicles, i.e., to EHM (Equipment Health Management), IVHM (Integrated Vehicle Health Management) and ISHM (Integrated System Health Management). TERMINOLOGY ANN or NN BT CBM DFT DOF EAI ECD EDM EDMS EEMS EHM ES FDI FL FOD GA GPA HMM IDMS ISHM IVHM KF MC MCD OSA-CBM (Artificial) neural network Bayesian theory Condition-based maintenance Discrete Fourier transforms Degree of freedom Enterprise application integration Electric chip detector Electro-discharge machine Engine debris monitoring system Electrostatic engine monitoring system Engine (or equipment) health management Expert system Fault detection and isolation Fuzzy logic Foreign object damage Genetic algorithm Gas path analysis Hidden Markov model Ingested debris monitoring system Integrated system health management Integrated vehicle health management Kalman filter Markov chain Magnetic chip detector Open System Architecture for Condition Based Maintenance

Copyright 2005 by ASME

RMS SEM XRF A, B, C, D d t u v x xd y z , , ,

Root mean square canning electron microscope X-ray fluorescence Model matrices Duration in time (or delay) Time Input or control variable Measurement noise State variable Degradation state variable Output variable Measured variable Threshold values

develop more reliable fault libraries, 5) combination of rulebased (e.g., expert system) diagnosis with Artificial Neural Network (ANN or NN) or Fuzzy Logic (FL), 6) knowledge discovery. The current survey builds on the 1999 review [1] and focuses on the research results published in the last few years. Some references before 1999 are cited in the paper to provide continuity for the topics not covered in the earlier review paper but important for this paper. In the last few years, we have seen a prognostic emphasis [2, 3] on the traditional Fault Detection and Isolation (FDI) [4] approach; we have also seen a significant interest in the area of algorithm development. Although advancements have been made in four general areas: 1) system partition, 2) system architecture, 3) EHM functionalities, and 4) algorithmic approaches, the latter two have represented the majority of the published work; hence they are the primary objective of this paper. To present the survey results without losing the big picture, the author starts by describing EHM requirements and summarizing the advancements in all four areas. The author then describes, in more details, the advancements in algorithmic approaches in each of the four EHM functional areas, namely: 1) gas path performance monitoring, 2) oil and debris monitoring, 3) vibration monitoring, and 4) usage and life monitoring. While these advancements provide building blocks for continued maturation of EHM technologies in the future, the survey has found a fundamental inconsistency in defining and representing EHM problems. This inconsistency creates barriers in exchanging EHM-related ideas and results; it also undermines the effectiveness of multi-organizational cooperation. Hence another purpose of this paper is to recommend a consistent approach to treating the problems that EHM is trying to address. Finally, the survey recognizes the need for a unified framework for comparing the performance of various algorithms for solving various types of EHM problems. The author hence suggests that an industry review, modeled after the industry review [5] for multi-variable engine controls in early 1970s, be held. This EHM Industry Review is aimed at defining a series of theme problems for EHM and invite experts in the industry and academia to apply their expertise to solving these problems. The Industry Review will be concluded with a conference to present the results and sharing experience.

INTRODUCTION The traditional Engine Health Management (EHM) approach uses fleet statistical data and signal processing techniques to detect and isolate faults. Modern E HM approaches enhance the traditional approach with physics-based models, individual engine performance tracking, predictive algorithms, and decision support capabilities. A modern approach typically measures key operating variables, compares model-estimated values with the measured values, and applies various algorithms and reasoning logic to make health management decisions; therefore, modern EHM capabilities often include model-based diagnostics (or prognostics) and model-based reasoning. A review of engine monitoring systems for commercial aviation was conducted and reported by Tumer and Bajwa in 1999 [1]. They reported that engine performance monitoring had proven effective in providing early warning and impending failures; however, high number of false alarms had created reluctance among commercial users to rely on the results. Space shuttle and helicopters had more advanced engine monitoring capabilities than commercial aircraft. Although the cost of implementing Health and Usage Monitoring System (HUMS) was still high, the benefits had been steadily increasing. The most commonly measured engine variables are: gas path temperatures and pressures, rotor speed, fuel flow, throttle position, nozzle position, stator position, oil properties, vibration, and life usage. For vibration diagnostics, signal averaging and component-specific vibration signatures had established good health indicators. Tumer and Bajwa identified two practical problems facing EHM: 1) too many false alarms, 2) insufficient sampling and data storage. On-going research areas in the field of EHM were: 1) anomaly detection, 2) replacing standard threshold method with feature extraction, 3) automated fault diagnosis, 4) combination of theory, knowledge, and test information to

Copyright 2005 by ASME

RESULTS OF SURVEY The major requirements of EHM are [6]: 1) automated monitoring, analysis, and decision support; 2) accurate results with high confidence; 3) robust capabilities against noise and faulty information; 4) wide coverage of fault conditions; 5) predictive capabilities; 6) using existing, or as few as possible, sensing instruments; 7) flexible, mo dular, and open architecture; and 8) user friendliness. SUMMARY OF EHM CAPABILITIES System Partition A widely accepted partition for EHM systems consists of two subsystems: 1) on-board or embedded subsystem, and 2) offboard or ground-based subsystem. Typically, the on-board system collects data, performs built-in-tests (BITs) and in-flight monitoring functions, such as detection, assessment, prediction, and recommendation/alert. The in-flight data and reports are then transferred to the ground-based system for further analysis and decision-making [7, 8]. System Architecture System architecture consists of primarily two types: 1) legacy or proprietary architectures, and 2) open architectures such as MIMOSAs (Machinery Information Management Open System Association) EAI (Enterprise Application Integration) framework [9], OSA-CBMs (Open System Architecture for Condition Based Maintenance) seven-layer data model [10, 11], or ISOs (International Organization for Standardization) sixlayer model for condition monitoring [12]. Aside from the data collection (or acquisition) and the data presentation layers, all of the above three open-architecture standards embrace the concept of five-layer monitoring functions, and they are: data analysis, state detection, health assessment, prognostics, and recommended actions. Web-based software architecture has become the trend in remote monitoring and engine life cycle data management [13, 14, 15]. Aside from OEMs services [16, 17, 18, 19], third-party monitoring software is also available to provide alternatives to engine operators and airlines [20]. Functionality Gas turbine engine is complex machinery that involves diverse disciplines such as aerodynamics, thermodynamics, mechanics, fluid mechanics, and chemistry. Managing the health of the engine requires different approaches and technologies for different disciplines. In general, EHM is divided into four functional areas: 1) gas path analysis and performance trending, 2) oil and debris monitoring, 3) vibration monitoring, a nd 4) usage and life monitoring. Effective EHM combines the monitoring results from more than one functional area, i.e., information fusion, to make the most educated decision.

Algorithmic Approaches Algorithms are used in a monitoring system to calculate capabilities and margins; algorithms are also used in reasoning processes to make decisions. Certain algorithmic approaches may be more effective for a given EHM functionality than others, but typically, algorithms fall into two categories: datadriven algorithms and model-based algorithms, while a combination of the two, i.e., a hybrid algorithm, may offer the benefits of both. The best way to think of algorithms is that they are embedded calculation procedures in an EHM task. Figure 1 illustrates this concept of embedded procedures in these EHM tasks. Each task is an information processing node. In the figure, each column represents an EHM functionality, and each row represents an EHM information aggregation layer. Algorithms typically reside at intersections.
Gas Path Oil/Deb. Vib. Usage

Recomm. Algorithm Predict Assess Detect Analyze

Figure 1: An algorithmic view of the EHM capability

The rest of the paper specifically discusses technology advancements in the four EHM functional areas, respectively. Although engine parameter sensing technologies are essential parts of the overall EHM capability, the author has decided not to cover sensing technologies in this paper. However, a detailed study of the sensor needs for intelligent engines can be found in the paper by Simon. [21]. GAS PATH PERFORMANCE MONITORING Gas path performance monitoring is commonly called the Gas Path Analysis (GPA). Gas path performance monitoring represents the area where the majority of EHM-related work has been published. Gas path monitoring relies on measured air/gas flow properties to infer the problems in engine components. Engine component problems, commonly called faults or failures, include: erosion, corrosion, fouling, dirt buildup, foreign object damage (FOD), worn seals, excessive tip clearance, burned or warped turbine stator or rotor blades, partially or wholly missing blades, plugged fuel nozzles, rotor disk or blade cracks induced by fatigue or operation outside normal intended limits, etc. [22]. Since most faults in gas

Copyright 2005 by ASME

turbine engines show up (or leave a signature) in the performance of the components exposed to the air/gas flow passing through the engine, flow-path related analyses have been effective in detecting faults in turbo machinery. The goal of gas path performance monitoring is to relate the observed changes in measured variables to internal changes in engine (or system) parameters. These internal parameter changes increase in time or usage and may be accelerated by the operating environment. There are many algorithmic approaches to gas path performance monitoring; the most frequently cited approaches are described below. Parameter Sensitivity Analysis The original concept of GPA was based on parameter sensitivity analysis and simultaneous evaluation of a combination of primary faults and sensor faults [23, 24, 25]. Basically, sensitivity analysis provides a means of assessing the relationship between independent (non-measurable) engine parameter deviations and the dependent (measurable) variables. The power of the sensitivity analysis is that it can detect single or multiple faults while isolating faulty components at the same time. For most engines, compressor efficiencies can be computed directly from measured temperatures and pressures; however, the efficiencies of turbines must be decided based on certain assumptions due to lack of measured data [26]. This approach is applicable to both long time-scale changes (or deteriorations) and short time-scale shifts or sensor errors. Model-based Estimation The most commonly used algorithm for state estimation is the Kalman Filter (KF), which is more applicable to long time-scale degradations. If a long time-scale change, such as component efficiency degradation, is modeled as a state variable, then the value of the state variable can be estimated by KF [27] and used to indicate alarming degradations. The same technique has been applied to detection of leaks and blockages in air flow passages [28]. More complex KF-based fault detection and isolation techniques have also been studied [29, 30]. The greatest challenge in applying the sensitivity analysis or the estimation approach to GPA is excess degree-of-freedom (DOF), i.e., the number of parameters that need to be estimated is greater than the number of independent measurements, hence unique identification of faults and their sources can not be achieved in most production engines. To overcome this problem, approximation techniques have been used. Examples of these techniques are: 1) multi-point (or discrete operating condition) GPA [ 31]; 2) to select the parameter-measurement relationship carefully to discern most possible faults [32]; 3) to

introduce a priori information on typical engine performance through the use of a regularization term in the (weighted) least square method [33]. Parameter estimation techniques using detailed componentlevel engine models, such as a cycle deck, have been studied widely. These techniques typically try to eliminate the error (or the discrepancy) between model-estimated and measured variables by adjusting model parameters [34, 35, 36, 37, 38, 39]. Data-driven Filtering and Smoothing Filtering and smoothing refer to the process of finding the best values for the present and the past, respectively [40]. Datadriven filtering and smoothing represent the earliest methods used in condition monitoring and fault detection. Both the sensitivity analysis and the estimation approaches discussed previously are considered model-based approaches, because they rely heavily on the physical model, although filtering is an inherent part of all model-based approaches to improve the data quality before they are used by the model. Data-driven filtering is applied to measured variables directly. A filter can be one of two types: 1) linear filter, and 2) non-linear filter. Linear filters typically include the low-pass filter, the mean filter, and the exponential filter; whereas non-linear filters include neural network, and combinations of different filters and statistical inferences [41, 42]. Artificial Neural Network An introductory survey of Artificial Neural Network (ANN) applications to aircraft turbine engines was conducted by SAE Aerospace Engine Condition Monitoring Committee [43]. Recent applications reflect three basic types of ANN for gas turbine applications: 1) feed-forward neural network, 2) probabilistic neural network or radial basis function neural network, and 3) auto-associative neural network [44]. These ANNs have been used for two broad purposes: 1) modeling of parametric relationships, and 2) Fault detector or classifier [45, 46, 47, 48, 49, 50, 51, 52, 53]. Relative strengths and weaknesses of ANN and KF techniques as applied to detecting single fault conditions was also studied [54]. ANN is effective in modeling long time-scale degradations and using the model for short time-scale shift detection. Hybrid or Other Approaches Fuzzy Logic (FL) approach lends itself for fault detection with incomplete or imprecise information. FL is typically applied to detecting short time-scale degradations [55, 56]. Genetic Algorithm (GA) has attracted attention as a global search technique that has the potential to escape local optima. Seemingly random, GA follows specific rules to reproduce subsequent cases (or strings) to minimize an

Copyright 2005 by ASME

objective function, and it does not require the calculation of derivatives [57, 58]. Markov Chain (MC) is a discrete-time, linear model in the time domain. It combines both deterministic and stochastic components in a single model. MC has been used for engine model estimation [59]. Hidden Markov Model (HMM) is a stochastic finite-state model. Each state only depends on the state of the previous time step. HMM can be applied to both steady-state and transient fault detection problems [60]. Bayesian Theory (BT) uses conditional probability to identify the most likely faults or fault states [61, 62]. Although elegant in mathematical formulation, BT is often difficult to apply in practice because of the uncertainties associated with the conditional probabilities, especially for the situation where a large number of possibilities (or faults) are present. Expert System (ES) and Decision Tree (DT) approaches have been used as effective troubleshooting tools for fault diagnosis [63, 64]. Hybrid, or cooperating, approach to EHM has the benefit of collecting the strengths of individual approaches; it can also reduce the sensitivity of individual approach for a specific type of data or problem. Examples of hybrid approaches include: GPA and classification [65], ANN and physics -based model [66], ANN and GA [67], GA and gradient search [68], ANN and KF estimation [69]. Even with a hybrid approach, unique identification of faults and their sources often can not be achieved due to lack of information. This lack of information is primarily due to: 1) insufficient sensors or measurements, and 2) similar fault symptoms for a given set of measurements. The end result is the presence of ambiguity (or indeterminate fault conditions). To reduce the ambiguity, GPA diagnostic results should be qualified, or combined, with those from other monitoring functions such as vibration and oil analysis. A comprehensive review of diagnostic techniques used in GPA was given by Li [70]. In this paper, Li presented a qualitative assessment of the computation speed and the model complexity of various algorithms. Using Lis speed-complexity map, the author has added additional algorithms and recast the results in Table 1 under various performance regions.

Table 1: GPA algorithmic performance assessment [70] High Medium Low (Speed) Linear GPA ANN KF BT Param. Est. Low Medium (Complexity) GA High HMM Hybrid FL, ES Nonlin. Est.

OIL AND DEBRIS MONITORING Debris formation is a result of surface failures. Oil and debris monitoring detects abnormal size or number of particles representing excessive wear or fatigue failures of engine components. Oil and debris monitoring can be performed in two approaches: 1) the traditional approach of lube oil debris monitoring, and 2) gas path debris monitoring. Oil debris monitoring can further be divided into two categories: 1) offline inspection and analysis; 2) on-line (or in-line) monitoring. Off-line inspection and analysis is typically performed for (magnetic) debris collectors and oil/debris samples, while online mo nitoring uses sensors to detect ferrous and non-ferrous particles, particle size, and total mass count. In addition, online monitoring of water contamination, viscosity, oxidation, total acid number is possible [71]. On-line oil debris monitoring uses mainly two types of sensors: 1) magnetic chip detector (MCD), and 2) electric chip detector (ECD). The MCD requires scheduled inspection, while the ECD provides immediate indication in the cockpit without the need for scheduled inspection. Newer generation inductive ECDs can collect and count ferromagnetic particles, especially for rolling-contact-fatigue failures; some of them even count nonferrous metals. Debris particles are typically analyzed off-line with an energy-dispersive scanning electron microscope (SEM/EDX) or X-ray fluorescence (XRF) instrument to determine the material [72]; therefore isolating the origin of the particles. On the other hand, gas path debris monitoring can only be performed when the engine is running, i.e., on-line monitoring. The Engine Debris Monitoring System (EDMS), or Electrostatic Engine Monitoring System (EEMS), technology was introduced in late 1970s [73]. The technology is applied to monitoring debris in both engine inlet and exhaust. The multination, multi-role Joint Strike Fighters (JSF) engine incorporates an Engine Distress Monitoring System (EDMS), which monitors the electrostatic charge associated with debris present in the exhaust gas of a gas turbine or aero-engine. The

Copyright 2005 by ASME

faults detected by the EDMS are blade r ubs, nozzle vane erosion, combustor burning. The JSF engine also incorporates an Ingested Debris Monitoring System (IDMS), which is based on the same principle but monitors the electrostatic charged debris of the intake air and debris (such as single object, dust or sand) [74, 75]. The electrostatic charge sensing concept can also be applied to sensing derbies in oil lines. The charge-sensing concept has potential to overcome the limitations of current oil debris monitoring devices, namely: 1) insensitivity to fine debris, and 2) inability to detect non-metallic particulates. Charge sensing is a promising technology for detecting failure pre-cursors [76]. VIBRATION MONITORING Vibration monitoring is interested in identifying dangerous vibratory conditions at all engine operating speeds and avoiding the secondary damages caused by engine component degradations. Similar to the gas path performance monitoring, vibration monitoring consists of on-board and off-board parts. The on-board part compares the vibration amplitudes measured at specific locations of an engine with pre-defined (absolute and relative) limits. Exceeding the limit prompts a warning message to be sent to the cockpit, the monitoring system, or the maintenance system. The off-board part processes the on-board data further with more sophisticated algorithms and models, and it monitors the vibration trend to provide early warning of abnormalities such as imbalance, wear, and rub. Vibration data are collected from on-board system through accelerometers that are mounted on engine case, flange, or gearbox. These accelerometers typically cover frequency ranges to several kHz. Alternative sensing technologies, such as eddy current [77] and acoustic sensors, and sensor configuration optimization [78] have been studied for vibration monitoring. Vibration data are sampled periodically and keyed to engine spool speeds. Vibration amplitude is then derived from the sampled data through the Discrete Fourier Transform (DFT), and broadband energy amplitude is extracted from the data through a root-mean-square (RMS) algorithm. The data can be further processed to generate waterfall plots (i.e., 3-D plots of vibration spectrum versus spool speed) and diagnostic indicators, such as a change in engine order, sub-harmonics, side-bands, resonance, and jumps, can be detected and point to potential problems [79]. More advanced algorithms for vibration diagnosis and prognosis have been applied to rotating components and bearings, such as ANN [80], similarity-based modeling [81], stochastic classifier [82], data

fusion [83, 84], and Dempster-Shafers evidence combination [85], have also been studied. LIFE USAGE MONITORING Life usage monitoring consists of on-board tracking of usage or damage and off-board prediction of remaining life of critical rotating components like disks and blades. Usage monitoring by tracking engine operating time and cycles have been practiced since the application of gas turbine engine to airplanes. This approach has inherent uncertainty which leads to overly conservative or potentially unsafe life estimation. With advancements in computational capability and damage accumulation modeling, it is possible (and desirable) to use available (or existing) sensor measurements to assess engine component usage more accurately. Examples of such measurements are: maximum speed, maximum turbine temperature, time-above-temperature, rates of change of critical variables. Usage monitoring, especially the one that monitors the fatigue life for fracture-critical components, is achievable and extends engine safety [86]. Fatigue life usage monitoring typically tracks the accumulated cycle (or Total Accumulated Cycle, TAC) as a function of engine operating time [87]. In general, three dominant damage mechanisms are addressed in life usage monitoring: fatigue, oxidation, and creep; and three factors determine our ability to predict the remaining life of an engine components: 1) interactions among these damage mechanisms, 2) uncertainties due to material or processing variations, and due to stress or temperature variations, and 3) data to facilitate model validation [88]. To improve the reliability of usage monitoring and life prediction, we need to have: 1) more effective sensors, 2) more accurate models, and 3) more accurate detection algorithms. While potential sensing technologies may include: 1) acoustic emission, 2) disk imbalance detection, 3) embedded or attached sensors, 4) blade tip deflection measurement, and 5) torsional vibration sensing [89], models and model-based algorithms present the major opportunity for improving life usage monitoring. Life usage monitoring algorithms are usually derived from damage models with empirical adjustments. These adjustments are derived from operational experience. One example is to quantify the model uncertainties, such as the variations in manufacturing, operating conditions, and controlled variables [90]. This type of probabilistic approach to life prediction [91] is essential for prognostics, because it inherently gives a confidence level for the prediction. It is also simple to implement when the repair and failure histories are available, from which accurate Weibull distributions can be derived [92]. Monitoring algorithms most likely will change over the life cycle of the engine because of possible new algorithmic techniques and potential component redesign/retrofit. Damage

Copyright 2005 by ASME

models that combine structural characteristics with gas path performance characteristics have provided plausible information for remaining life prediction for High Cycle Fatigue (HCF) failures [93]. One way to improve life prediction is to consider residual stresses as a basis for life extension [94]. Another new approach to model improvement incorporates material-based, micro-structural model. Since the conventional characterization of structural damage, such as crack initiation, is based on macroscopic variables such as stress and strain; while this type of models is applicable to long cracks, it is inadequate for predicting crack initiation and small crack growth. Microstructural model has demonstrated increased reliability and potential for predicting crack initiation and growth to failure for high strength materials such as disks and blades [95].

To explain these concepts further, we need to introduce three types of abnormal conditions: 1) fault, 2) anomaly, and 3) failure. The author suggests the following definitions for these abnormal conditions: 1) a fault is a known (or previously analyzed) and repeatable abnormal condition; 2) an anomaly is an unknown (i.e., not seen before, not fully understood, or intermittent) abnormal condition; 3) a failure is the abnormal condition whose performance has degraded such that it is functionally unacceptable. Figure 2 shows the mutually exclusive sets of faults and anomalies. It also illustrates the concepts of known (or modeled) and unknown (or un-modeled) physical behaviors when a system or equipment is operating normally.

REPRESENTATION OF EHM PROBLEMS Through this survey, the author has noticed inconsistency in the definition and representation of EHM problems. The inconsistency might have created barriers in exchanging ideas and sharing research results; therefore, a standard definition of terminologies and consistent use of mathematical expressions could facilitate effective cooperation among different organizations and accelerating the maturation of EHM technologies. The author would like to recommend, by consolidating the practices in the papers being reviewed, the following definitions and expressions to describe EHM problems . DEFINITION OF KEY EHM TERMS An abnormal condition is a sizable deviation of the actual operating condition from the normal condition. The deviation may apply to a single measurement variable or a group of variables which together represent a distinguishable pattern. A sizable deviation means that the difference is larger than a predetermined threshold of tolerance. A deviation is typically called a residual . Abnormal conditions generally result from two common causes: 1) usage-induced degradation, and 2) external factors that exert an abrupt change or an accelerated rate of change (e.g., foreign object damage, or FOD, in gas turbine engines). Two important EHM concepts are developed around the abnormal condition, they are: 1) diagnosis (or diagnostics), and 2) prognosis (or prognostics). In the simplest explanation, diagnosis refers to the process of finding current and past abnormal conditions (or events), while prognosis refers to the process of estimating the future influence of the currentlydiscovered abnormal conditions (or finding the future events that are casually derivable from the current condition).

Figure 2: Recommended definition for faults and anomalies Because of the continuously changing nature of an abnormal condition, the severity of a fault increases with the usage of the system or equipment. This change in fault severity over time forms a trajectory of degradation in a performance-versustime plot as illustrated in Figure 3. This trajectory is also called the degradation path, because given enough time and usage; the degradation will reach an unacceptable level defined as failure. The figure also illustrates the relative locations of the anomaly, the fault, and the failure in the degradation process. At first, a precursor of performance degradation may be observed as the earliest detectable sign of an abnormal condition. This precursor reflects the sensitivity of the EHM system, although it depends largely on the sensing and modeling capabilities used in the EHM system. As degradation increases, a condition is reached where the degradation can be detected consistently (and reputably), and it justifies alarming the system or equipment operator of the abnormal condition. This condition is called a fault and is usually associated with a specific root cause. We consider a fault less severe than failure, and in most cases, a system can continue to operate with fault conditions present (typically up to a pre-determined time period or usage count). A fault may eventually develop into a failure, where the system can no longer provide

Copyright 2005 by ASME

acceptable functionality and is said to have reached its service life limit. If continue to operate beyond this condition, the system will break down and may cause catastrophic failure. A fault symptom is a collection of individual faults. A fault symptom, in turn, corresponds to one or more failure modes , and each failure mode describes a mechanism that decreases a functionality of the system or vehicle being monitored. If we can measure the functionality quantitatively and replace the word performance in Figure 3 with functionality or capability, then the figure represents the failure progression path of a specific failure mode. Performance Pre-cursor Detectable range Faults

Equations (1) and (2) are the nonlinear form of the state and output equations for the system. It is often desirable t o linearize these equations around a nominal operating condition, e.g., x0, y0, and u 0, to get & t = A t xt +B t u t x (3)

() () () () () y (t ) = C(t )x (t ) + D(t )u (t )

(4)

where A(t) and B(t) are the state-distribution and the inputdistribution matrices for the linear state equation, respectively; and C(t) and D(t) are the state-distribution and the inputdistribution matrices for the linear output equation, respectively. The elements in these distribution matrices are the parameters of the system. Equations (3) and (4) represent a linear time-varying model. In this general form, system parameters are functions of time. When the parameters are constant in time, the model simplifies to a linear, time-invariant (LTI) model as x & t = Ax t + Bu t (5)

Anomaly Failure x Breakdown Time

Functional acceptable limit

() () () y (t ) = Cx(t ) + Du (t )

(6)

where all the matrix elements, a ij, b ip, ckj, and d kp in the matrices A, B, C, and D, respectively, are constants. As a system undergoes internal changes, these changes may be manifested in performance degradations. The performance degradation can be represented in Equations (3) and (4) as time-varying parameters, i.e., these parameters change over time to represent the root cause of the degradation. Since this kind of time-varying models are difficult to analyze, an alternative approach to degradation modeling has been favored. This approach is to treat the time-varying parameters as additional state variables. This is achieved by augmenting the original state and output equations with an additional state variable (or state vector), called the degradation state variable xd(t) , i.e., & A t = AA x A t + Bu t x (7)

Useful life

Figure 3: Graphical depiction of anomaly, faults, and failure

MATHEMATICAL REPRESENTATION The author suggests using the control system formulation to represent EHM problems. The generalized from of the state equation that describes a system or equipment performance is x & t = f x t ,u t ,t (1)

()

( () () )

where x and u stand for the state and the input variables , respectively, and t is the time. The same symbols x and u are used for a variable that is either a scalar or a vector. In steadystate (or quasi-steady) conditions, the left-hand side of (1) is zero, and it degenerates into an algebraic equation (or a system of equations). The generalized form of the output equation that describes system output variable, y, is y t = g x t ,u t , t . (2)

() () () y (t ) = C A x A (t ) + Du (t )
T

(8) (9)

where xA(t) is the augmented state variable and is defined as

x A (t ) = [ x(t ), x d (t )] .

()

( () () )

The state, input, and output variables are illustrated in the basic block diagram in Figure 4. System x(t)

u(t)

y(t)

Note that (7) and (8) can be used as the general form of performance degradation model by defining the state variable as consisting of the non-degraded state x, and the degraded state xd. Also note that Equations (3) to (8) represent the system model in continuous-time. These equations can be expressed in discrete-time to facilitate digital computer implementation. The goal of the diagnostics is to detect alarming performance degradations and identify the root causes of these

Figure 4: Basic block diagram of the system and its variables

Copyright 2005 by ASME

degradations. In terms of the system model described above, the diagnostic goal is to detect one or more output variables, under measurement noise v, that have exceeded the alarm threshold , i.e., z t = y t +v t > (10)

()

() ()

Similarly, the fault isolation aspect of the diagnostic goal is to identify which degradation state variable has exceeded the alarm threshold , i.e., d t > x (11)

()

realizes that the field of E HM is advancing rapidly, and the survey results only represent the information that we have collected till early 2004. E ven with the information that has been collected, it is a daunting task to organize and present the information in an effective form. Because of the large amo unt of published work in EHM, the author recognizes that any survey paper of a finite length inevitably leaves out some references and topics that the author either is not aware of or does not consider relevant enough in the context of this review. As a side benefit of this survey, the author believes that a consistent definition of EHM related terminologies and a unified representation of the EHM problems is needed for the EHM (or the system or vehicle health management) community. The author has recommended a set of definitions, notations, and expressions to facilitate more effective communication and organizational cooperation. As another side benefit of this survey, the author believes that a standardized process to compare the merits and the drawbacks of different modeling and algorithmic approaches is needed. T he author believes the best way to develop this process is to hold an industry-wide forum on EHM. This forum will consist of two parts: 1) a workshop to gather industry experts and EHM researchers to define a theme problem to be solved, and 2) a conference to present the results of different approaches or techniques after the theme problem has been distributed. As an example, the workshop may be held three months after invitations are s ent out to experts and researches. After the workshop, the theme problem can be published on a web site and on professional organizations newsletters/magazines. The conference to share the results may be held in six months after the workshop. Finally the results will be published in a proceeding or a book. Traditionally, engine condition monitoring has led the way for condition-based maintenance and health management technologies because of the safety and dispatch requirements of aircraft engines. Hence, the author believes that the results, conclusions, and recommendations presented in this paper can be generalized to all types of equipment, systems, and vehicles, i.e., EHM (equipment health management), PHM (Prognostics and Health Management), IVHM (integrated vehicle health management), and ISHM integrated system health management).

where the ^ symbol stands for the estimated value of the degradation state variable that has exceeded the threshold, hence indicating the root cause of a degradation. The goal of the prognostics is to predict how performance degradation will deteriorate to an unacceptable level. This goal can be expressed as z t+ d = y t+ d +v t +d > (12)

d (t + d ) > x

(13)

where is the acceptable performance bound for useful service and is the corresponding state variable degradation bound; d is the time-to-failure (measured from the current time t) or remaining useful life of the system. CONCLUSIONS AND RECOMMENDATIONS A survey of recent advancements in EHM has been conducted. The survey indicates that engine gas-path performance monitoring represents the majority of EHM related research; however, the value of oil/debris monitoring and vibration monitoring in refining gas-path monitoring results has been recognized. The approach to combining different monitoring results, i.e., data fusion, is becoming an active area of research. Furthermore, usage and life monitoring for fatiguecritical or life-limited parts is increasingly important to an integrated logistics (or autonomic logistics) system. For all four types of monitoring functions (i.e., gas-path, oil/debris, vibration, and life usage monitoring), model-based reasoning and advanced diagnostic/prognostic algorithms have formed the trend. These models and algorithms try to accomplish five goals: 1) data analysis, 2) fault detection, 3) fault isolation, 4) failure prediction, and 5) fault-failure accommodation. From the modeling standpoint, both physicsbased and data-driven models show benefits. From the algorithmic standpoint, both traditional filtering as well as statistical techniques and artificial intelligence techniques are used, although a combinatorial (or hybrid) approach has seems more promising. In presenting the survey results, the author has emphasized algorithmic approaches to solving EHM problems. The author

ACKNOWLEDGMENTS The author should like to thank NASA Glenn Research Center for the privilege to conduct this survey and for providing financial support. The author is grateful for the guidance and critique of Ten-Huei Guo of NASA during the course of this

Copyright 2005 by ASME

project. The author would like to thank Sanjay Garg, Johnathan Litt, Don Simon, and the ASME paper review team for their suggestions. REFERENCES 1. Tumer, I. and Bajwa, A.; A Survey of Aircraft Engine Health Monitoring Systems; 35th Joint Propulsion Conference; June 1999; AIAA-99-2528. Smith, G, Schroeder, J. B., Navarro, S., and Haldeman, D., Development of a Prognostics & Health Management Capability for the Joint Strike Fighter, Proc. of AUTOTESTCON, IEEE, 1997, pp. 676-682. Scheuren, W., Joint Strike Fighter Prognostics & Health Management, paper presented at the 34th AIAA/ASME/ASEE Joint Propulsion Conference & Exhibit, Cleveland, Ohio, July 1998, AIAA 98-3710. Frank, P. M., Fault Diagnosis in Dynamic Systems Using Analytical and Knowledge-based Redundancy A Survey and New Results, 1990 International Federation of Automatic Control, pp. 459-474. Sain, M. (Editor); Alternatives in Linear Multivariable Control; National Engineering Consortium, Chicago, 1978. Tsalavoutas, A. et al.; Combining Advanced Data Analysis Methods for the Constitution of an Integrated Gas Turbine Condition Monitoring and Diagnostic System; Proceedings of ASME Turbo Expo 2000, Munich, Germany. Henley, S., et al.; Autonomic Logistics the Support Concept for the 21st Century, paper presented at the 2000 IEEE Aerospace Conference, Big Sky, MT; Paper No. ZF11_0701. Zoller, T.; Advanced Engine Monitoring and Diagnosis Systems: Actual System for the EJ200 Engine of the EuroFighter 2000 Aircraft and Future Trends; NATO AVT Symposium on Aging Mechanism and Control, Manchester, UK, 2001. MIMOSA web site: www.mimosa.org. OSA-CBM web site: www.osacbm.org. Followell, D., Gilbertson, D., and Keller, K.; Implications of an Open System Approach to Vehicle Health Management; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2004. ISO web site: www.iso.org. Forsyth, G. and Delaney, J.; Design Diagnostic Expert Systems for Long-term Supportability; Proceedings of ASME Turbo Expo 2000, Munich, Germany. Jaw, L., et al.; ICEMS: Intelligent Condition-based Engine/Equipment Management System;

15.

16. 17.

2.

18. 19. 20.

3.

4.

21.

5.

6.

22.

23.

7.

24.

8.

25.

26.

9. 10. 11.

27.

12. 13.

28.

29.

14.

30.

Proceedings of ASME Turbo Expo 2001; 2001-GT0015. Callan, R. and Larder, B.; The Development and Demonstration of a Probabilistic Diagnostic and Prognostic System (ProDAPS) for Gas Turbines; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2002. GE Remote Diagnostics web site: www.geae.com/services/information/diagnostics. PW Engine Monitoring Program web site: http://www.pratt-whitney.com/serv_comm_ main_monitor.asp. RR Engine Health Monitoring web site: http://www.rolls -royce.com/service/civil/totalcare. Honeywell Engine Services web site: https://www.eengines.honeywell.com. OKeeffe, N; The State-of-the-Play in Engine Condition Monitoring; Aircraft Technology & Engineering Management Magazine, OctoberNovember Issue, 2003. Simon, D., et al.; Sensor Needs for Control and Health Management of Intelligent Aircraft Engines; Proceedings of ASME Turbo Expo 2004; GT200454324. Urban, L. A.; Gas Path Analysis Applied to Turbine Engine Condition Monitoring; J. of Aircraft , Vol. 10, No. 7, July 1973. Urban, L. A.; Gas Path Analysis Applied to Turbine Engine Condition Monitoring; AIAA Paper 72-1082, December 1972. Urban, L. A.; Parameter Selection for Multiple Fault Diagnostics of Gas Turbine Engines; ASME Paper 74-GT-62. Volponi, A. J.; Gas Path Analysis an Approach to Engine Diagnostics; Proceedings of 35th Symposium, Mechanical Failure Prevention Group, April 1982. Doel, D. L.; Interpretation of Weighted-LeastSquares Gas Path Analysis Results; Proceedings of ASME Turbo Expo 2002, Amsterdam, The Netherlands. Luppold, R. H., et al.; Estimating In-flight Engine Performance Variations Using Kalman Filter Concepts; paper presented at the 25th Joint Propulsion Conference, Monterey, CA, 1989; AIAA89-2584. Narendra, J. et al.; Model-based Detection of Leaks and Blockages in Pipes; Proceedings of ASME Turbo Expo 2000, Munich, Germany. Kobayashi, T. and Simon, D. L.; Application of a Bank of Kalman Filters for Aircraft Engine Fault Diagnostics; Proceedings of ASME Turbo Expo 2003, Atlanta, USA; GT2003-38550. Simon, D. and Simon, D. L.; Aircraft Turbofan Engine Health Estimation using Constrained Kalman

10

Copyright 2005 by ASME

31.

32.

33.

34.

35.

36.

37.

38.

39.

40.

41.

42.

43.

Filtering; Proceedings of ASME Turbo Expo 2003, Atlanta, USA; GT2003-38584. Stamatis, A., et al.; Jet Engine Fault Detection with Discrete Operating Points Gas Path Analysis; Journal of Propulsion and Power, Vol. 7, No. 6, 1991. Pinelli, M. and Venturini, M.; Improvement of the Accuracy in Gas Turbine Health Determination; Proceedings of ASME Turbo Expo 2001, New Orleans, USA. Doel, D. L.; An Assessment of Weighted-LeastSquares-based Gas Path Analysis; ASME J. of Engineering for Gas Turbines and Power; Vol. 116, pp. 366-373, 1994. Mathioudakis, K., Kamboukos, Ph., and Stamatis, A.; Turbofan Performance Deterioration Tracking using Non-linear Models and Optimization Techniques; Proceedings of ASME Turbo Expo 2002, Amsterdam, The Netherlands. Ogaji, S. O. T. and Singh, R.; Study of the Optimization of Measurement Sets for Gas Path Fault Diagnosis in Gas Turbines; Proceedings of ASME Turbo Expo 2002, Amsterdam, The Netherlands; GT2002-30050. Razak, A. M. Y. and Dosanjh, M. S.; Application of an Advanced Performance Monitoring System to Detect an Implanted Fault on a Twin Spool Aero Derivative Gas Turbine; Proceedings of ASME Turbo Expo 2002, Amsterdam, The Netherlands. Srli, A., et al.; Performance Monitoring on the Asgard A FPSO as Part of an Integrated Monitoring Strategy; Proceedings of ASME Turbo Expo 2002, Amsterdam, The Netherlands. Gorinevsky, D, et al.; Model-based Diagnostics for an Aircraft Auxiliary Power Unit; Proceedings of IEEE Conference on Control Applications; September 2002. Marcos, A., Balas, G. J., and Mylaraswamy, D.; Robust Identification and Residual Generation Application to a Turbine Engine; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2004. Bryson, A. E. and Ho, Y. C.; Applied Optional Control Optimization, Estimation, and Control; revised printing, Hemisphere Publishing Corporation, 1975. Ganguli, R.; Data Rectification and Detection of Trend Shifts in Jet Engine Gas Path Measurements using Median Filters and Fuzzy Logic; Proceedings of ASME Turbo Expo 2001, New Orleans, USA. DePold, H., et al.; Validation of Diagnostic Data with Statistical Analysis and Embedded Knowledge; Proceedings of ASME Turbo Expo 2003, FT200338764. Jaw, L. - Editor and Working Group Chairman for SAE Aerospace Recommended Document (ARD) on

44.

45.

46.

47.

48.

49.

50.

51.

52.

53.

54.

55.

Neural Network Applications to Engine Condition Monitoring; ARD50069, SAE International, 1998. Mattern, D. L., et al.; Using Neural Networks for Sensor Validation; paper presented at the 34th Joint Propulsion Conference, Seattle, WA, 1998; AIAA 983547. Bin, S., Jin, Z, and Shaoji, Z.; An Investigation of Artificial Neural Network (ANN) in Quantitative Fault Diagnosis for Turbofan Engine; Proceedings of ASME Turbo Expo 2000, Munich, Germany. Embrechts, M. J., et al.; Neural Network Modeling of Turbofan Parameters; Proceedings of ASME Turbo Expo 2000, Munich, Germany. Lu, P. J., et al.; An Evaluation of Engine Faults Diagnostics using Artificial Neural Networks; Proceedings of ASME Turbo Expo 2000, Munich, Germany. Romessis, C., Stamatis, A., and Mathioudakis, K.; A Parametric Investigation of the Diagnostic Ability of Probabilistic Neural Networks on Turbofan Engines; Proceedings of ASME Turbo Expo 2001, New Orleans, USA; 2001-GT-0011. Chiras, N., Evans, C., and Ress D.; Nonlinear Gas Turbine Modeling using Feedforward Neural Networks; Proceedings of ASME Turbo Expo 2002, Amsterdam, The Netherlands; GT-2002-30035. Romessis, C. and Mathioudakis, K.; Setting up of a Probabilistic Neural Network for Sensor Fault Detection Including Operation with Component Faults; Proceedings of ASME Turbo Expo 2002, Amsterdam, The Netherlands. Dambrosio, L., et al.; Feed Forward Neural Networkbased Diagnostic Tool for Gas Turbine Power Plant; Proceedings of ASME Turbo Expo 2002, Amsterdam, The Netherlands. Gluch, J. and Krzyzanowski, J.; Application of Preprocessed Classifier Type Neural Network for Searching of Fault Components of Power Cycles in Case of Incomplete Measurement Data; Proceedings of ASME Turbo Expo 2002, Amsterdam, The Netherlands. Kim, K., Ball, Charles, and Nwadiogbu, E; Fault Diagnosis in Turbine Engines using Unsupervised Neural Network Technique; Proceedings of SPIE, Vol. 5439, April 2004, pp. 150-158. Volponi, A. et al.; The Use of Kalman Filter and Neural Network Methodologies in Gas Turbine Performance Diagnostics: a Comparative Study; Proceedings of ASME Turbo Expo 2000, Munich, Germany. Ganguli, R.; Application of Fuzzy Logic for Fault Isolation of Jet Engines; Proceedings of ASME Turbo Expo 2001, New Orleans, USA; 2001-GT-0013.

11

Copyright 2005 by ASME

56.

57.

58.

59.

60.

61.

62.

63.

64.

65.

66.

67.

68.

Yu, L. J., Cleary, D. J., and Cuddihy, P. E.; A Novel Approach to Aircraft Engine Anomaly Detection and Diagnostics; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2004. Cranfield University; Introduction to Genetic Algorithms; Gas Turbine Diagnostics, Prognostics, and Simulation Short Course, London, UK, March 2004. Sampath, S., Gulati, A., and Singh, R.; Fault Diagnostics using Genetic Algorithm for Advanced cycle Gas Turbine; Proceedings of ASME Turbo Expo 2002, Amsterdam, The Netherlands. Arkov, V. Y., Kulikov, G. G., and Breikin, T. V.; Application of Markov Chains to Identification of Turbine E ngine Dynamic Models; Proceedings of ASME Turbo Expo 2002, Amsterdam, The Netherlands. Menon, S., et al.; Incipient Fault Detection and Diagnosis in Turbine Engines using Hidden Markov Models; Proceedings of ASME Turbo Expo 2003, Atlanta, U. S. A., GT2003-38589. Bickford, R. L. et al.; Real-time Flight Data Validation for Rocket Engines; paper presented at the 32nd Joint Propulsion Conference, Lake Buena Vista, FL, 1996; AIAA 96-2827. Bryg, D. and Jaw, L.; Continuous Logic - a Technique for Engine Condition Monitoring Using a Combination of Kalman Filtering, Bayesian Analysis, and Decision Analysis ; Proceedings of COMADEM 1998. Bajwa, A. R. and Kulkarni, D.; Engine Data Analysis Using Decision Trees; paper presented at the 36th Joint Propulsion Conference, 2000; AIAA 2000-3627. Spina, P. R., Torella, G., and Venturini, M.; The Use of Expert Systems for Gas Turbine Diagnostics and Maintenance; Proceedings of ASME Turbo Expo 2002, Amsterdam, The Netherlands. Aretakis, N., Mathioudakis, K., and Stamatis, A.; Non-linear Engine Component Fault Diagnosis from a Limited Number of Measurements using a Combinatorial Approach; Proceedings of ASME Turbo Expo 2002, Amsterdam, The Netherlands; GT2002-30031. Jaw, L.; Neural Networks for Model-based Prognostics; Proceedings of IEEE Aerospace Conference, Aspen, Colorado, March 1999. Takahisa, K. and Simon, D. L.; A Hybrid Neural Network-Genetic Algorithm Technique for Aircraft Engine Performance Diagnostics; paper presented at the 37th Joint Propulsion Conference, Salt Lake City, Utah, 2001; AIAA 2001-3763. Grnstedt, T. U. J.; Identifiability in Multi-point Gas Turbine Parameter Estimation Problem; Proceedings

69.

70.

71.

72.

73.

74.

75.

76.

77.

78.

79.

80.

81.

of ASME Turbo Expo 2002, Amsterdam, The Netherlands. Brotherton, T, et al.; eSTORM: Enhanced Self Tuning On-board Real-time Engine Model; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2003. Li, Y. G.; Performance-analysis -based Gas Turbine Diagnostics: a Review; Journal of Power and Energy, Proceedings of the Institution of Mechanical Engineers, Vol. 216, Part A, 2002. Kristiansen, P. A. and Leeker, R. E.; U. S. Navy Inline Automated Oil Analysis Program; Proceedings of the JOAP International Condition Monitoring Conference, April 2002. Tauber, T. and Johnson, D.; Experience with the Electric Oil Debris Monitoring System of the General Electric GE-90 Gas Turbine Engine on the Boeing 777 Aircraft; Proceedings of the JOAP International Condition Monitoring Conference, April 2002. Couch, R. P.; Detecting Abnormal Turbine Engine Deterioration Using Electrostatic Methods; paper presented at AIAA Aircraft Systems and Technology Conference, August 1978. Lapini, G., Zippo, M., and Tirone, G.; The Use of Electrostatic Charge Measurements as an Early Warning of Distress in Heavy-duty Gas Turbines; Proceedings of ASME Turbo Expo 2001, New Orleans, USA; 2001-GT-373. Powrie, H. E. G. and Fisher, C. E.; Engine Health Monitoring: Towards Total Prognostics; Proceedings of IEEE Aerospace Conference, Aspen, Colorado, March 1999. Powrie, H., et al.; Electrostatic Charge Generation Associated with Machinery Component Deterioration; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2002; Z11_0304. Swan, P. and Betts, R. G.; Improved Vibration Monitoring on Aeroderivative Gas Generators with Internal Transducers; Proceedings of ASME Turbo Expo 2001, New Orleans, USA; 2001-GT-0030. Banks, J. C., et al.; Vibration Sensor Configuration Optimization for the AV-8B F402-RR-408 Engine; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2002; Z11_0506. Barragan, J. A. M.; Engine Vibration Monitoring and Diagnosis Based on On-board Captured Data; NATO AVT Symposium on Aging Mechanism and Control, Manchester, UK, 2001. Peng, P. Y. and Yang, M. T.; Neural Networks Based Diagnostics for Mistuned Bladed Disk; Proceedings of ASME Turbo Expo 2000, Munich, Germany. Wegerich, S. W.; Similarity Based Modeling of Time Synchronous Averaged Vibration Signals for

12

Copyright 2005 by ASME

82.

83.

84.

85.

86.

87.

88.

89.

90.

91.

92.

93.

Machinery Health Monitoring; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2004. Ghiocel, D. M.; Refined Stochastic Field Models for Jet Engine Vibration and Fault Diagnostics; Proceedings of ASME Turbo Expo 2000, Munich, Germany. Brotherton, T., et al.; A Testbed for Data Fusion for Helicopter Diagnostics and Prognostics; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2003. Orsagh, R. F., Sheldon, J., and Klenke, C. J.; Prognostics/Dignostics for Gas Turbine Engine Bearings; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2003. Modgil, G., Orsagh, R. F., and Roemer, M. J.; Advanced Vibration Diagnostics for Engine Test Cells; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2004. Pfoertner, H.; Long-term Operation and Maintenance of Engine Monitoring Systems Recommendations Derived from 15 Years of OLMOS Use by the GAF; NATO AVT Symposium on Aging Mechanism and Control, Manchester, UK, 2001. Pfoertner, H.; The Information Content of Turbine Engine Data a Chance for Recording-based Life Usage Monitoring; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2002; Z11_0406. Akers, D. W. and Rideout, C. A.; Measurementbased Prognostic Models for Service-induced Damage in Turbine Engine Components; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2004. Sarlashkar, A. V., et al.; Model-based Analysis in Support of the TF-41 Seeded-fault Testing; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2003. Guo, T-H and Chen, P.; Sensor-based Engine Life Calculation a Probabilistic Perspective; Proceedings of the XVI International Symposium on Air Breathing Engines (ISABE), 2003. Shende, R. W., et al.; A Probabilistic Lifing Approach for Gas Turbine Disk; Proceedings of the XVI International Symposium on Air Breathing Engines (ISABE), 2003. Sutherland, H., et al.; Prognostics, a Look at Statistical Life Prediction for Condition-based Maintenance; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2003. Suarez, E. L., et al.; Jet Engine Life Prediction Systems Integrated with Prognostics Health Management; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2004.

94.

95.

John, R., et al.; Incorporating Residual Stresses in Life Prediction of Turbine Engine Disks; NATO AVT Symposium on Aging Mechanism and Control, Manchester, UK, 2001. [Nas04] Nasser, L. and Tryon R.; Integration of Material-based Simulation into Prognosis Architecture; Proceedings of IEEE Aerospace Conference, Big Sky, Montana, March 2004.

13

Copyright 2005 by ASME

You might also like