 |
 |
 |
 |
 |
 |
 |
||
|
|
|||||||||
Remaining Useful
Life Estimation Using Hybrid Monte-Carlo Simulation and Proportional Hazard
Model
Daming Lin1,
Naaman Gurvitz2, Murray Wiseman1
1OMDEC Inc., Toronto, Canada (daming@omdec.com, murray@omdec.com)
2Clockwork Solutions, Austin TX (naaman.gurvitz@clockwork-solutions.com)
PS PLUSTM software offers power providers
the ability to implement intelligent gas turbine life cycle management
processes. Operators wish to achieve higher availability by reducing
unnecessary scheduled outages for either inspection or repair. PS-PLUS is a
SPARTM-based application that uses the Monte-Carlo (MC) method to
estimate machinery remaining useful life. The method predicts the scope and
schedule of maintenance associated with important failure modes. Other works
have explored the Proportional Hazard Model (PHM), using EXAKTTM
software to accurately forecast the probability of failure of gas turbine
components. A PHM quantitatively measures the relative importance of each
influential risk factor (covariate) that affects life estimation. The
propensities for failure are modeled as a function of both time dependent
covariates and an item’s working age. The hybrid PHM-MC prototype application
demonstrates Remaining Useful Life Estimation in conjunction with time
dependent covariates such as (a) key operational duty cycle profile factors
i.e. load, fuel type, starts, trips, etc, (b) sensor readings, and (c)
borescope inspection data indicative of component health and state. This paper
presents a conceptual design, data requirements and analysis techniques needed
to fuse PHM and Monte-Carlo simulation techniques. The hybrid system should
generate accurate remaining useful life predictions. Those predictions form the
basis of cost-effective condition-based maintenance (CBM) of gas turbines.
Effective CBM, in contrast to time-based maintenance (TBM), profoundly improves
life cycle performance and cost. The paper demonstrates the superiority of PHM
analysis compared to traditional Weibull analysis in predicting lower-end
failure probabilities, for example B1 and B5 lives. Because of the serious
economic consequences of critical failures, such reliability estimates must be
considered in business decisions related to gas turbine operation and warranty
management.
Traditional gas turbine maintenance policy is
primarily comprised of time-based (or duty-cycle-based) maintenance (TBM). The
search to avoid unnecessary scheduled maintenance and to reduce failure risk is
shifting attention away from planned maintenance of gas turbines and towards
advanced condition-based maintenance (CBM) (Chen et al., 1994; Reebe, 2003;
Al-Bedoor et al., 2003). Current gas turbine CBM policy, however, is based
mainly on conservative experience-derived engineering judgment. There is a
growing interest among operators to investigate opportunities for reducing
overall costs by supplementing that judgment with rigorously calculated
Remaining Useful Life Estimations (RULE).
In turbine asset management, the term
"inspections" refers both to information gathering (as in condition
based maintenance) and scheduled renewal. “Standby” and “running inspections”
are carried out to allow for minor adjustments. They can provide much recorded
information that is related to maintenance cost and reliability. A disassembly
inspection, of which there are three types (Combustion Section, Turbine Section,
and Turbine Rotor), is a costly event.
Standby inspections apply mostly to backup and peaking units.
In gas turbine operations, failure can be
catastrophic and preventive maintenance is expensive. These factors alone
provide ample incentive for driving decisions from all possible information
sources. Information abounds in gas turbine operations, and its very volume
challenges operators in using it to greatest possible effect. The information
intensive nature of gas turbine operation and maintenance and the scale of the
impact of less than optimum decisions encourage the examination of novel data
interpretation methodologies. Two such important decisions are 1) When to do a
turbine section inspection? and 2) When to do a rotor inspection?
Information gathered during standby, running,
and combustion section inspections contains potential knowledge useful for
optimally planning and scheduling future Turbine Section and rotor inspections.
Such information may be further supplemented with day-to-day sensor and
operational profile information. Running inspections provide steady state
operating parameters such as load versus exhaust temperature, vibration, fuel
flow and pressure, lube oil pressure, exhaust gas temperatures, exhaust
temperature spread variation, and startup time. Deviations from the norm relate
to calibration errors and equipment health.
Combustion section inspections are relatively
short duration disassembly inspections where the opportunity is taken to make
CBM borescope and visual inspections the results of which are highly related to
risk and remaining useful life, thus bearing heavily on the optimal schedule of
a subsequent turbine section or rotor inspection. The combustion section
inspection includes:
1. Visual inspection of first-stage turbine
nozzle partitions.
2. Borescope inspect turbine buckets to mark the
progress of wear and deterioration of these parts. 1st, 2nd, 3rd buckets +
nozzle. (Data related to turbine section component failure.)
3. Borescope inspection of compressor, intermediate
compressor rotor stages
4. Borescope observation of the condition of
blading in the aft end of axial-flow compressor.
5. Visual inspection of the compressor inlet and
turbine exhaust areas, checking condition of inlet guide vanes (IGVs), IGV bushings,
last stage buckets and exhaust system components.
The decision of when to do disassembly
inspections is based on conservatively pragmatic and simplified engineering
approximations of the combined effect of diverse operational factors that are
known or assumed to influence component life. The major ones are:
·
Cycle effects
(the number of starts)
·
Firing
temperature (power setting)
·
Fuel type (gas,
light, crude, residual)
·
Level of steam or
water injection used to increase power and control NOx emissions.
High cycles of "peaking machines"
are associated with the failure mode "thermal mechanical fatigue".
However continuous duty machines' dominant failure modes are creep, oxidation,
and corrosion leading to rupture, erosion, and deflection. Both types of duty
cycles have certain failure modes in common. They are: high cycle fatigue,
rubs/wear, and foreign body damage.
In this paper, we propose an approach to
Remaining Useful Life Estimation of gas turbines using PHM and Monte-Carlo
simulation. The combination of the aforementioned multiple factors can be
included in a PHM. The proposed RULE approach will provide an advanced
estimation of machinery remaining useful life and outage scope/schedule
requirements associated with important failure modes. It will eventually
contribute to achieving higher availability by reducing unnecessary scheduled
outages for either inspection or repair.
In this hybrid
PHM-MC prototype application, the
PHM is first explored to depict the failure mechanism of the component
associated with key failure modes. It quantifies the propensity for
failure as a function of both time dependent covariates (e.g. operational
factors, sensor readings, inspection information) and the working age. Then stochastic models are used to describe
the behavior of covariates. The covariate behavior models are necessary since
the RULE depends on future covariate values while some future covariate values
are unknown and have to be forecasted. Finally, PS-PLUS Monte-Carlo simulation
model will provide the RULE. These three steps will be discussed in detail in
the following sections.
[Contact OMDEC
for body of paper]
In this manuscript, we have presented a conceptual design of hybrid PHM-MC prototype application for advanced Remaining Useful Life Estimation of gas turbines. For each critical component in a gas turbine, a PHM is established in EXAKT to relate both the working age and condition information (covariates) of the component to its risk of failure. Then a stochastic model is developed to describe the behavior of covariates included in the PHM. This covariate behavior model is required in the calculation of RULE when the future values of some covariates are unknown. Finally the SPAR Monte-Carlo simulation engine is used to simulate the remaining useful life distribution of the gas turbine based on its system structure (described by reliability block diagrams in SPAR), the PHMs built for its critical components, and the covariate behavior model built for associate covariates.
The potential benefits of utilizing the proposed hybrid PHM-MC prototype are:
a) Providing more accurate remaining useful life estimation, with more confidence;
b) Achieving more effective and adaptive condition-based maintenance;
c) Reducing unnecessary costly maintenance and overhauls, and hence achieving higher availability at lower cost.
The authors are most grateful to Dr. Dragan
Banjevic at the CBM Lab, University of Toronto for his assistance in preparing
this manuscript.
Agresti, A. (1990), Categorical Data
Analysis, Wiley.
Al-Bedoor, B.O., Ghouti, L., Adewusi, S.A.,
Al-Nassar, Y. and Abdlsamad, M. (2003), “Experiments on the extraction of blade
vibration signature from the shaft torsional vibration signals”, Journal of Quality in Maintenance
Engineering, 9, 144-159.
Beebe, R. (2003), “Condition monitoring of
steam turbines by performance analysis”, Journal
of Quality in Maintenance Engineering, 9,
102-112.
Bilmes, J.A. (1998), “A gentle tutorial of the
EM algorithm and its application to parameter estimation for Gaussian mixture
and hidden Markov models”, International Computer Science Institute, University
of Berkeley, California (1998).
Brockwell,
Peter J. and Davis, Richard A. (2002). Introduction to Time Series and
Forecasting, 2nd. ed., Springer-Verlang.
Chen, W., Meher-Homji, C.B. and Mistree, F.
(1994), “Compromise: an effective approach for condition-based maintenance
management of gas turbines”, Engineering
Optimization, 22, 185-201.
Doron, A. (2003), PS-PLUS v1.0 User Guide,
Clockwork Solutions, Beer-Sheva, Israel,
Dubi, A. (1999), Monte Carlo Applications in Systems Engineering, Wiley.
Dubi, A. and Goldfeld, A. (2002), SPAR v.
5.0 User Guide, Department of Nuclear Engineering, Ben Gurion University of
the Negev.
EXAKT v. 3.30 Guide and Manual (2003), CBM laboratory, Department of
Mechanical Industrial Engineering, University of Toronto, Canada.
Hoeft, R. and Gebhardt, E. (2000), GE Energy
Services Atlanta, GA Heavy-Duty Gas Turbine Operating and Maintenance
Considerations, GER-3620G (9/00).
Jardine, A.K.S., Anderson, P. M. and Mann, D.
S. (1987), “Application of the Weibull proportional hazard model to aircraft
and marine engine failure data”, Quality & Reliability Engineering
International, 5, 77-82.
Kalbfleisch, J.D. and Prentice, R. L. (1980),
The Statistical Analysis of Failure Data, Wiley.
Lawless, J. F. (1982), Statistical Models and Methods for Lifetime Data, Wiley.
MacGregor, John F., et al. (1994), Multivariate Statistical Process Control of Batch Processes Using PCA and PLS, IFAC ADCHEM 1994 Preprints, May 4-27, Kyoto, Japan.
|
