See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/347927586

Prediction of Remaining useful life of Mechanical Components – A Review.

Article in International Journal of Engineering Education · October 2014

CITATIONS READS

25 1,368

3 authors, including:

S. B. Kivade Nagendra Iranna Jamadar

Sri Jayachamarajendra College of Engineering Dr. D.Y. Patil Vidya Prathishthan Society
43 PUBLICATIONS 253 CITATIONS 14 PUBLICATIONS 48 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Nagendra Iranna Jamadar on 05 January 2021.

The user has requested enhancement of the downloaded file.

 ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 3, Issue 6, November 2014

Prediction of Remaining Useful Life of

Mechanical Components-A Review
Tushar Salunkhe, N.I. Jamadar, Dr. S. B. Kivade

Abstract— The concept of Remaining Useful Life (RUL) is utilized to predict life-span of components with the purpose
of minimizing catastrophic failure events. As customer demands for dynamically controlled systems it is necessary to have a
continuous monitoring system which tracks, identifies trends and sources of component degradation prior to failure. The
early warning capability seeks to detect, isolate and estimate the severity of faults based on fault propagation and detected
degradation of a machine or a component to predict its RUL. Typically RUL is random and estimated from available
sources of condition and health monitoring information. RUL of a device is also essential in remanufacturing engineering
in the decision for parts to be taken out of service for remanufacturing. This review is centered on various methods of
predicting RUL of a machine or a component. Some of the methods for predicting RUL are analyzed with the case studies
such as Automotive Component, Rotating machinery, aero engine, Electro-hydraulic servo valve, Electronic systems, Low
methane compressor, Bearings etc.

Key words— Remaining Useful Life, Continuous Monitoring System, Fault propagation, Diagnostics, Prognostics.

I. INTRODUCTION
Operational safety, maintenance cost effectiveness and asset availability has a direct impact on the component of
organization and nations. Today's complex and advanced machine demand highly sophisticated and costly
maintenance strategies. Therefore there is a pressing need to continuously develop and improve current
maintenance programs [4]. High maintenance cost and unwanted downtime in industries are often caused by
machine fault problems. Appropriate maintenance techniques are needed to overcome these problems. The prime
objective of maintenance technique is to keep machinery and plant equipments in good operating condition to
prevent failure and production loss [5]. Maintaining equipment in operational condition is an industrial,
economical and societal requirement. This requirement can be satisfied by implementing appropriate maintenance
strategies which can be curative or preventive. But for more efficiency one can implement Condition Based
Maintenance (CBM)[10]. The relentless competition among the industries and increased demands from customers
for dynamically controlled safety systems are creating the need for a continuous monitoring system that tracks and
identifies trends and sources of component degradation prior to failure. This is because conventional maintenance
strategies, such as corrective and preventive maintenance are not adequate to fulfill the needs of customers
demands [18]. The theory of degradation assessment is the key foundation of safety service of the machine parts
and equipment. Now the degradation assessment are widely used for machinery equipment in various engineering
fields [15]. The degradation techniques used by various engineering fields are automotive industries, aircraft
industries, manufacturing industries, military services, etc. For example, accurate life prediction for turbine
engines has become increasingly important in recent years as military and commercial aircraft fleet age and engine
component failures cause unplanned maintenance and repair. Therefore increasing the life prediction confidence
is of utmost importance to increase the service life of aircraft turbo machinery components [3].

Condition-based maintenance (CBM) is a maintenance program that recommends maintenance decisions based
on the information collected through condition monitoring [1]. There are two substantial factors to support CBM,
machine fault diagnosis and machine condition prognosis. Machine fault diagnosis is defined as a method for
detecting, isolating and identifying a failure condition of a part or system, while the affected components are still
operating even though they are in a degradation mode. Prognosis is the ability to assess the current condition of a
part or system, observe the future condition of the part or system and predict the time left before catastrophic
failure occurs [5]. Both diagnostics and prognostics originally come from the medical field. As machinery
maintenance technology emerged diagnostics and prognostics gradually permeated all areas of mechanical
engineering [16]. CBM is an alternative that uses embedded diagnostics and prognostics to determine systems
health. Failure prognosis involves forecasting of system degradation based on observed system condition [17].
CBM aims at achieving reliable and cost-effective operation of engineering systems such as aircraft systems, wind
turbine generators, hydro power plants and manufacturing systems [9]. Prognostics is to predict how much time is
left before a failure occurs (or, one or more faults) given the current machine condition and past operation profile.

125
 ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 3, Issue 6, November 2014
The time left before observing a failure is usually called Remaining Useful Life (RUL) [1]. Prognosis is the
capability to provide early detecting of the precursor and/or incipient fault condition of a component and to have
the technology and means to manage and predict the prognosis of this fault condition to component failure [6].
The early warning capability seeks to detect, isolate and estimate the severity of faults based on models that
include cross-subsystem fault propagation effects and relate detected degradation to accurate RUL predictions of
replaceable components [18]. To fulfill the goal of prognosis, two crucial steps are needed: failure prediction and
RUL estimation [5].

RUL prognostics is a fundamental premise to perform CBM for a system subject to performance degradation [14].
The RUL of an asset or system is defined as the length from the current time to the end of useful life [7]. RUL also
called remaining service life or remnant life refers to the time left before observing a failure given the current
machine age, condition and the past operation profile [1]. The remaining life prediction of engineering systems
has gained some momentum over the last decade owing to the increasing demand for higher efficiency, lower cost
and longer service time. Remaining life prediction has become a major scientific challenge and bottleneck issue
for the implementation of prognosis and health management (PHM) activities [13]. RUL of a mechanical device
or equipment is the lifetime from current time to the time that the device fails. RUL modeling and online
assessment provide the timely message for system decision making in production management and CBM. RUL of
a device is also essential in remanufacturing engineering in the decision for parts to be taken out service for
remanufacturing [21].

II. LITERATURE REVIEW

Mei Wen et. al [2] presented RUL predictions for EB-PVD thermal barrier coatings (TBCs) on a Pt-modified NiAl
bond coat cycled at two temperatures. They made the predictions based on the data using photoluminescence
piezospectroscopy without knowledge of the test temperature. Further they compared the RUL predictions using
regression methods and neural network methods and it was found that the neural-network methods are superior.

Wahyu Caesarendra et al. [5] discussed a prognosis algorithm, based on Sequential Monte Carlo method (also
known as a particle filter) which can be used in nonlinear systems. They used the real trending data of low
methane compressors acquired from condition monitoring routines for evaluating the proposed method. The
results showed that the proposed method has a potential to predict the trending data in real systems of machine
condition prognosis.

Hack-Eun Kim et al. [8] proposed a technique for accurate assessment of the remnant life of machines based on
health state probability estimation and historical knowledge embedded in the closed loop diagnostics and
prognostics system. To provide long term prediction they used Support Vector Machine (SVM) classifier as a tool
for estimating health state probability of machine degradation process. To validate the feasibility of the proposed
model, the real life historical fault data from bearings of High Pressure-Liquefied Natural Gas (HP-LNG) pumps
are analyzed and RUL is estimated. The obtained results showed that the proposed prognosis system based on
health state probability estimation has the potential to be used as an estimation tool for RUL prediction in
industrial machinery.

Zhigang Tian [9] developed an artificial neural network (ANN) method for prediction of more accurate RUL of
equipment subjected to condition monitoring. They validated the proposed ANN method by using the condition
monitoring data collected in the field from bearing on a group of Gould pumps. Experimental results showed that
the proposed ANN method can produce satisfactory RUL prediction results, which can assist the CBM
optimization. Also a comparative study is performed between the proposed ANN method and the Modified Wu's
method and the results demonstrated the clear advantage of the proposed approach in achieving more accurate
predictions.

K. Medjaher et al. [10] discussed a data driven prognostics method based on the utilization of data provided by
monitoring sensors to estimate the RUL of an industrial plant by focusing on its most critical component. The
bearings in any of the machinery are considered as most failure components as per Electric Power Research
Institute (ERPI) and researchers who worked in the reliability of electrical machines. In this literature, the
degradation of this component is modeled using Mixure of Gaussians Hidden Markaov Models (MoG-HMMs)
represented by Dynamic Bayesian Networks (DBNs), which permits to present the evolution of the degradation

126
 ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 3, Issue 6, November 2014
using hidden states. They estimated the RUL in two main phases a learning phase to derive the degradation's
behavior model and an exploitation phase to estimate the RUL and its associated confidence.

A. Loron et al. [11] presented the prognosis of a simplified spring-mass system and a pneumatic valve for
aeronautical application using stochastic model-based techniques. In this case prognosis is consists of computing
the distribution of the RUL of the system. The proposed method addressed three main challenges; first, the model
should unify the two classical approaches to describing complex systems, second, the prognosis must integrate
online information to provide a specific result for each system depending on their life events. Third, the prognosis
must supply ready meaningful numerical results the error of which must also be under control.

Fuqiong Zhao et al. [12] developed an integrated prognostics method for predicting RUL of gear by utilizing both
gear physical models and real-time condition monitoring data. They described a gear FE model for gear stress
analysis, a gear dynamics model for dynamic load calculation and a damage propagation model by using Paris
law. Also they used a bayesian method to fuse the collected condition monitoring data to update the distributions
of the uncertainty factors for the current specific unit and to achieve the updated RUL prediction. The
effectiveness of the proposed approach is demonstrated on the basis of simulated degradation data thus it leads to
more accurate RUL predictions.

Zhaoqiang Wang et al. [13] presented a case study of condition-based remaining storage life (RSL) prediction for
gyros in the inertial navigation system (INS) on the basis of the condition monitoring (CM) data and the influence
of the CM data taking process. They developed a stochastic-filtering-based degradation model to incorporate both
into the prediction of the RSL distribution. The results showed that the predicted RSL depend on not only the CM
data but also the testing process of taking the CM data which the existing prognostic techniques and algorithms
did not consider.

LIU Junqiang et al. [14] developed a novel approach to RUL prognostics of aeroengine subjected to
non-stationary degradation and the state-space method, Kalman filtering algorithm are used for calculation. The
case study showed that an accurate prognosis of RUL is achieved which provides a guide for operators to realize
CBM thereby assisting in guaranteeing civil aviation safety and minimizing operation cost.

Jianhui Luo et al. [17] developed an integrated prognostics process based on data collected from model-based
simulations under nominal and degraded conditions. They constructed prognostic models based on different
random load conditions and Interacting Multiple Model (IMM) is used to track the hidden damages. They
performed prediction of RUL by mixing mode-based life predictions with time-averaged mode probabilities.

Chaitanya Sankavaram et al. [18] discussed a methodology of a hybrid model-based, data-driven and knowledge
based integrated diagnosis and prognosis framework with the application to automotive suspension and electronic
(Li-Ion) systems. The prognostic process is designed with a real-world perspective that allows to experiment with
model-based/data-driven techniques.

Yimin Shao et al. [19] proposed a fault prognosis and diagnosis method based on combination of radial basis
function (RBF) and back-propagation (BP) neural networks. The problems occurring in detection and diagnosis of
the faults in an automotive rear axle gear due to non-linear characteristics of gear vibration under the fluctuating
load and speed conditions are analyzed. Finally they concluded that the forecasting of gear operating conditions
can be used for practical fault prognosis and diagnosis with the help of RBF-BP network.

Dawn An et al. [20] presented a Matlab-based tutorial for model-based prognostics which combines a physical
model parameters to predict the RUL. Author used a battery degradation and a crack growth models as example
for RUL prediction. For the estimation of RUL they used model-based prognostics particle filter algorithms
method. To illustrate the results, the RUL at an arbitrary cycle are predicted in the form of distribution along with
the median and 90% prediction interval.

Yawei Hu [21] proposed an analytical decision method for a remanufacturing program that is based on the RUL
online assessment. They used SVM model to estimate the RUL and the performance reliability based on the
predefined threshold for failure and the remanufacturing characteristics. Also they established the support vector
regression model for performance degradation modeling and the forecasting of RUL.

127
 ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 3, Issue 6, November 2014
III. PROGNOSTIC APPROACHES
Numerous methods and tools regarding failure prognostics have been proposed and reported. These methods and
tools are subjected to several classifications. Three main approaches are distinguished on the basis of complexity,
cost, precision and applicability: model-based prognostics, data-driven prognostics and knowledge-based
prognostics [10].

The model-based prognostic approach uses a mathematical representation of the system and thus incorporates a
physical understanding of the system into the monitoring scheme. These methods use statistical estimation
techniques based on residual generated using observers (e.g., Kalman filters, reduced-order unknown input
observers interacting multiple models, particle filters) and uniform relations (dynamic consistency checks among
measured variables) to track the component degradations [18]. These models are derived from configuration,
usage and historical "run-to-failure" data and applicable to maintenance decision making [22].

Fig. 1 Illustration of the model-based prognostics process [20]

The process of model based prognostics is illustrated in Fig. 1. Where, the degradation model is expressed as a
function of usage conditions U, elapsed cycle or time t and model parameters .

The usage conditions and time are given, while the model parameters characterizing the damage behavior should
be identified,. Then the RUL which represents the remaining time to failure is calculated based on the estimated
model parameters. The model parameters are estimated using an algorithm such as particle filter (PF) by
integrating the damage model with the damage data that represents the systems health state at the time data are
obtained. Since damage cannot be directly measured in most cases, a damage quantification process is required
from sensor measurement data, which is called structural health monitoring (SHM) [20].

Model-based approaches are applicable in situations where accurate mathematical models can be construct from
first principles. However, model-based approach may not be most practical approach since the fault type in
questions is often unique, varies from component to component, and is hard to be identified without interpreting
the operation. Statistical techniques are normally used to define thresholds to detect the presence faults. Several
statistical models have been reported recently, such as Recursive Least Square (RLS) , Proportional Intensity
Model (PIM), Proportional Hazards Models (PHMs) and Hidden Markov Model (HMM) [8].

The main advantage of model based approach is the ability to incorporate physical understanding of the system to
monitoring. Another advantage is that, in many situation, the change in feature vector are closely related to model
parameters. Therefore, it can also establish a function mapping between the drifting parameters and the selected
prognostic features [17].

The data-driven prognostic approach to fault diagnosis and prognosis (FDP) is preferred when system models are
not available (e.g., when subsystem vendors do not share models for competitive reasons), but instead system
monitoring data is available. Here, failure prognosis involves forecasting of system degradation and
time-to-failure based on "state awareness" gleaned from monitored data. Neural network and statistical
classification methods are illustrative of this approach [18].

128
 ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 3, Issue 6, November 2014

Fig. 2 SOC, SOH, and RUL estimation framework [18]

The data-driven approaches are derived directly from routinely monitored system operating data (such as
calibration data, calorimetric data, spectrometric data, power, vibration and acoustic signals, temperature,
pressure, oil debris, current and voltages). In many applications, measured input/output data is major source of
knowledge to understand the system degradation behavior. The data driven approaches rely on the assumption that
the statistical characteristics of data are relatively consistent unless a malfunctioning event occurs in the system.
They are built based on past historical records and produce prediction outputs based on condition monitoring
(CM) data [8].

The data-driven prognosis can be explained with the application to Li-Ion Batteries and Electronic Systems. With
unstable oil supply and increased emphasis on alternative energy, hybrid electric vehicles are viewed as the future
of automotive market. In order to implement CBM of Li-Ion batteries, problem to be solved is the dynamic
estimation of its status. A systematic procedure for prognosis of Li-Ion batteries is shown in Fig. 2. It involves
estimating three critical characteristics of a battery, viz., state of charge (SOC), state of health (SOH) and RUL
[18].

The most promising data driven approach is based on Artificial Intelligent approaches. These approaches utilized
statistical and learning techniques, including the theory of pattern recognition such Hybrid SVM-Bayesian
Network (BN) model, Neuro-fuzzy model, Neural Networks, Back Propagation Neural Network (BPNN),
Dynamic Wavelet Neural Networks (DWNN) and Recurrent Neural Networks (RNNs) [8].

The strength of data driven technique is the ability to transform high-dimensional noisy data into lower
dimensional for diagnostic or prognostics decisions. The main drawback of data driven approaches is that there
efficiency is highly on the quantity and quality of systems operational data. The data driven approach is applicable
to systems, where an understanding of first principles of system operation is not compressive [17].

The knowledge based approach uses graphical models such as Petrinets, multi-signal flow graphs and Bayesian
networks (BNs) for system monitoring and troubleshooting. The approach is specially well-suited for prognosis
coupled systems, where the prognostic test outcomes (from a model-based or data-driven approach) can be used to
isolate the root cause to replaceable components. Multi-signal models are preferred because they can be applied to
large scale systems with the thousand of failure models and tests, and can include failure probabilities and
unreliable test as part of the inference process in a way that is computationally more efficient than BNs [8].

This model is a combination of computational intelligence (CI) and experience. The knowledge-based approach
relates to collection store information from subject matter experts and interpretation of rule set. It can be seen as
service performance systems for service delivery based upon the principle service feedback for analysis.

129
 ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 3, Issue 6, November 2014
Parameters reliability is estimated using an experience based approach to information gather from understanding
the asset [22].

IV. PROGNOSTIC MODELS

Human decision making is not always sufficiently reliable or accurate when dealing with complex equipment with
a multitude of interrelating failure modes. Therefore over recent years a significant amount of research has been
undertaken to develop models that can be used to reduce industry's dependence on key individuals. Selecting an
appropriate prognostic model for a particular application is crucial to the ultimate success of a prognostic
program. Each model type has strengths and weaknesses based on underlying assumptions, data requirements or
complexity that make them better suited to certain applications than to others. Therefore not only technical details
of the physical system and how model variables reflect this system be understood for prognostic modeling
programs to be successful. Generally following prognostic models are used [6];
1. Expert system models
2. Fuzzy system models
3. Stochastic models
4. Statistical models
5. Artificial Neural Networks (ANN)
6. Physical Models
1. Expert System models
These systems are used when the problem area is well understood, stable and narrow. For the implementation of
this system the operating condition must be stable and predictable.
Advantages: Simple to develop and easy to understand
Limitations: Totally depends on knowledge of subject matter experts and requires significant amount of rules.

2. Fuzzy System models

When the mathematical model of the system is not available or not feasible to implement then fuzzy systems
models are used. To apply fuzzy systems one or more variables must be continuous.
Advantages: Fewer rules are required than expert systems.
Limitations: Domain experts required to develop rules.

3. Stochastic models
For the system which has small set of dominant failure modes, the stochastic models are used. When the sample
size is statistically significant and representative of individual sample then such type of models can be used. On
the other hand when the hazard rate is constant these models are not used.
Advantages: Theoretically can be performed at all equipment hierarchy levels, especially when a small number of
failure modes dominate.
Limitations: Failure must be statistically independent and identically distributed.

4. Statistical models
These models are applicable for those systems which has single defined failure modes associated with a single
monitored parameter that can be described with a monotonic trend when the operating conditions are stable.
Advantages: It is the simplest technique to apply and explain. In these models it is easy to set alarm.
Limitations: Availability of confidence limits dependent on amount of data at the different states of failure
development.

5. Artificial Neural Networks (ANN)

When the exact optimal estimation of RUL is required then ANN models are used to predict RUL. This type of
models can handle large amount of noisy, numerical and temporal data. These models are best suited when the
physical, statistical model is not known or impractical to apply.
Advantages: Complex, multi-dimensional and non-linear systems can be modeled. Physical understanding of the
system behavior is not required.
Limitations: Requires a significant amount of data for training data that needs to be representative of true data
range and its variability.

130
 ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 3, Issue 6, November 2014
6. Physical models
To perform the prognosis of the system of which a physical model for each failure mode is available, physical
models can be used. These models can give highly accurate and precise RUL estimation. In this case failure modes
are well understood and defined.
Advantages: It can provide most accurate and precise estimates of all modeling actions.
Limitations: Detailed and complete knowledge of system behavior is required. The accuracy and robustness are
subjected to the experimental conditions under which models were developed.

V. CASE STUDY- BEARING FAILURE OF HP-LNG PUMPS

In this review paper, theoretical approach of RUL prediction is discussed, but for the better understanding of
practical approach for RUL prediction following has been selected and discussed.

Fig. 3 Pump schematic and vibration measuring points [8]

In the literature [8], authors estimated RUL of bearings used in HP-LNG Pumps. For this, RUL prediction tests
were conducted using bearing failure data of HP-LNG pumps to validate the feasibility of integrating health state
probability estimation with historical knowledge for accurate long term failure prediction.

As shown in Fig. 3,. Three ball bearings are installed to support the entire dynamic load of the integrated shaft of
pump and motor. The submerged motor and bearings are cooled and lubricated by a predetermined portion of the
LNG being pumped. For condition monitoring of the pump, two accelerometers are installed on the housing near
the bottom bearing assembly and in two radial directions. It is very difficult to detect the symptom of pump failure
at an incipient stage because certain bearing components can result in rapid bearing failure due to poor lubricating
conditions and high operating speed (3600 rpm).

Vibration data were collected through two accelerometers installed on the pump housing as shown in Fig. 3. The
vibration data from two HP-LNG pumps of identical specification were used for prediction of the RUL. Due to the
random operation of the pumps in order to meet the total production target, there were some restrictions to collect
complete data over the entire life of the pump. The acquired vibration data are summarized in Table 1. From this
table a total 136 vibration samples for P301 C and 120 vibration samples for P301 D were collected during the full
range of operation over the life of the pump, for training and testing of the proposed prognostic model,
respectively.

Table 1 Acquired vibration data of through LNG pump [8]

A. RUL Prediction

Stage no. No. of samples (u) Average operation hours ( ) RUL (%)
1. 1-8 4 99.89 %
2. 25-32 503 85.67 %
3. 41-48 843 75.99 %
4. 81-88 2501 28.77 %
5. 121-128 3405 3.02 %

131
 ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 3, Issue 6, November 2014
In the prediction of bearing RUL, closed and open tests were conducted. The closed test was involved using
identical data sets for model training and prediction. On the other hand, different test data sets were applied in the
open test. Identical training data sets were used in both tests.

In the closed test, the five states were trained using the listed training data sets shown in Table 2, and full data sets
from P301D (120 data sets) were tested to obtain the probabilities of the five degradation states.

Table 2 Classification of training data for the five degradation states using the three selected features [P301 D] [8]

Machine
P301 C P301 D
(HP-LNG Pumps) No.
Total operation hours 4698 3511
Reason of remove and root High vibration and outer raceway High vibration and inner raceway
cause sampling. flaking.
No. of sample data 136 120
Sampling frequency (Hz) 12800 12800

Fig. 4 Comparison of actual RUL and estimated RUL [Closed

Test, P301 D] [8]

Fig. 5 Probability distribution of each health state

[Open Test, P 301 C] [8]

Fig. 6 Comparison of actual RUL and estimated RUL

[Closed Test, P301 D] [8]

132
 ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 3, Issue 6, November 2014
Fig. 4 shows the probability distribution of the five health states of P301 D. The first state probability starts at
100% and decreases as long as the next state probability increases. For example, the first state (solid lines) shows
the probabilities dropping and increasing again until about 90% and eventually dropping to zero (at sample 30).
Simultaneously the second state (dotted lines) reached 100%. Some overlaps between the states and non
uniformity of the distribution can be explained due to dynamic and stochastic degradation processes.

In the open test, similar bearing fault data (P301 C), which consisted of 136 sample sets, was tested to obtain the
probability distribution of each health state using identical training data sets shown in Table 2. Fig. 5 shows the
probability distributions of the five health states of P301 C.
The machine remnant life of bearing failure was estimated using historical operation hours of each training
data sets described in Table 2 and their probabilities evaluated using Eq. (1).
(1)
Fig. 6 shows the results of closed tests in estimating the bearing remnant life and its comparison with the actual
RUL. As shown in Fig. 6, although there are some discrepancies, the overall trend of the estimated RUL follows
the gradient of the actual RUL of the machine.
The average prediction accuracy was 94.4% using Eq. (2) over the entire range of the data set. The estimated RUL
at the final state matched closely to the actual RUL.

Average prediction accuracy (%) = (2)

where N is number of sample, is actual RUL(%), and is estimated RUL(%)

Fig. 7 Comparison of actual RUL and estimated RUL

[Closed Test, P301 C] [8]

Fig. 7 shows the results of open test in estimating the bearing remnant life and its comparison with the actual RUL.
There is a large difference in remnant life prediction at the initial degradation states as shown in Fig.7. In the open
tests, the estimated RUL time was obtained based on the training data sets (P301 D) which had 3511 h in total
operation while the actual operating time of P301C is 4698 h. This causes the discrepancy between the actual RUL
and the estimated RUL at the beginning of the test. However, as it approaches the final bearing failure, the
estimated RUL matched very closely to the actual RUL than those in the initial and middle states.

VI. CONCLUSION
In this paper, an effort has been made to review the different prognostic approaches and models, implemented
practically for estimating the RUL of mechanical components. The following points were taken as the concluding
remarks. The prognostic process is designed with a real-world perspective that allows one to experiment with
model-based or data-driven prognostic approaches.
 The model-based approaches are used when there is a possibility of mathematical modeling of the system.
 The data-driven approaches are used when it is impossible to formulate the mathematical model of the
system.
The selection of an appropriate prognostic model for mechanical components is crucial to the ultimate success of
a prognostic program.
 Expert system models are considered over the fuzzy system models as they are simple to develop and easy
to understand.

133
 ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 3, Issue 6, November 2014
 Furthermore, stochastic models are used when the fault sample size is small but when the fault sample size
is large then the statistical models are used.
 Expert, Fuzzy, Stochastic, and Statistical models are used to estimate approximate RUL.
 However, ANN model is used for accurate estimation of RUL. ANN can handle large amount of fault data
even for the complex systems.
 The physical models are preferred when the detailed physical model of each failure mode is available.
By referring the above facts and figures, the complexity level of the prognostic models goes on increasing from
expert system models to physical models.

ACKNOWLEDGMENT
The author would like to thank for the valuable suggestions and guidance rendered by Prof. N. I. Jamadar, Dr. S.
B. Kivade. And special thanks to Dr. V Singh (HOD-Mechanical Engineering) and Dr. R. K. Jain (Principal) for
their extreme support to complete this assignment.

REFERENCES
[1] Andrew K.S. Jardine, Daming Lin and Dragan Banjevic, "A review on machinery diagnostics and prognostics
implementing condition-based maintenance", Mechanical Systems and Signal Processing, Vol. 20, pp.1483-1510, 2006.
[2] Mei Wen, Eric H. Jordan and Maurice Gell, "Remaining Life Prediction of Thermal Barrier Coatings Based on
Photoluminescence Piezospectroscopy Measurement", Journal of Engineering for Gas Turbines and Power, Vol. 128,
pp.610-616, 2006.
[3] Girija Parthasarathy, Sunil Menon, Kurt Richardson, Ahsan Jameel, Dawn McNamee, Tori Desper, Michael Gorelik and
Chris Hickenbottom, "Neural Network Models for Usage Based Remaining Life Computation", Journal of Engineering
for Gas Turbines and Power, Vol. 130, pp.012508-1 to 012508-7, 2008.
[4] Aiwina Heng, Sheng Zhang, Andy C. C.Tan and Joseph Mathew, "Rotating machinery prognostic: State of the art,
challenge and opportunities", Mechanical Systems and Signal Processing, Vol. 23, pp. 724-739, 2009.
[5] Wahyu Caesarendra, Gang Niu and Bo-Suk Yang, "Machine condition prognosis based on sequential Monte Carlo
method", Expert Systems with Applications, Vol.37, pp. 2412-2420, 2010.
[6] J.Z. Sikorska, M. Hodkiewicz and L. Ma, “Prognostic modeling options for remaining useful life estimation by industry",
Mechanical System and Signal Processing, Vol. 25, pp. 1803-1836, 2011.
[7] Xiao-Sheng Si, Wenbin Wang, Chang-Hua Hu and Dong-Hua Zhou, "Remaining useful life estimation-A review on the
statistical data driven approaches", European Journal of Operational Research, Vol. 213, pp. 1-14, 2011.
[8] Hack-Eun Kim, Andy C.C. Tan, Joseph Mathew and Byeong-Keun Choi, "Bearing fault prognosis based on health state
probability estimation", Expert Systems with Applications, Vol. 3, pp. 5200-5213, 2012.
[9] Zhigang Tian, "An artificial neural network method for remaining useful life prediction of equipment subject to condition
monitoring", Journal of Intelligent Manufacturing, Vol. 23, pp. 227-237, 2012.
[10] K. Medjaher, D.A. Tobon-Mejia and N. Zerhouni, "Remaining useful life estimation of critical components with
application to bearings", IEEE Transactions on Reliability, Vol. 61, pp. 292-302, 2012.
[11] A. Lorton, M. Fouladirad and A. Grall, "A methodology for probabilistic model-based prognosis", European Journal of
Operational Research, Vol. 225, pp. 443-454, 2013.
[12] Fuqiong Zhao, Zhigang Tian and Yong Zeng, "Uncertainty Quantification in Gear Remaining Useful Life Prediction
Through an Integrated Prognostic Method", IEEE Transactions on Reliability, Vol. 62, pp. 146-159, 2013.
[13] Zhaoqiang Wang, Changhua Hu, Wenbin Wang, Zhijie Zhou and Xiaosheng Si, "A case study of remaining storage life
prediction using stochastic filtering with the influence of condition monitoring", Reliability Engineering and System
Safety, Vol. 132, pp. 186-195, 2014.
[14] Liu Junqiang, Zhang Malan, Zuo Hongfu and Xie Jiwei, "Remaining Useful Life Prognostics for Aero engine Based on
Superstatstics and Information Fusion", Journal of Aeronautics, 2014.
[15] Kun Zhang, Jinyong Yao and Tongmin Jian, "Degradation assessment and life prediction of electro-hydraulic servo valve
under erosion wear", Engineering Failure Analysis, Vol. 36, pp.284-300, 2014.
[16] Jay Lee, Fangji Wu, Wenyu Zhao, Masoud Ghaffari and Linxia Liao, "Prognosis and health management design for
rotary machinery systems-Reviews, methodology and applications", Mechanical Systems and Signal Processing, Vol.42,
pp. 314-334, 2014.

134
 ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 3, Issue 6, November 2014
[17] Jianhui Luo, Madhavi Namburu, Krishna Pattipati, Liu Qiao, Masayuki Kawamoto and Shunsuke Chigus, "Model-based
Prognostic Techniques", AUTOTESTCON 2003 IEEE System Readiness Technology Conference, pp. 330-340, 2003.
[18] Chaitanya Sankavaran, Bharat Pattipati, Anuradha Kodali, Krishna Pattipati, Mohammad Azam, Sachin Kumar and
Michael Pachet,(2009) "Model Based and Data Driven Prognosis of Automotive and Electronic System", 5th Annual
IEEE Conference on Automation Science and Engineering, India, pp. 96-101.
[19] Yimin Shao, Jie Liang, Fengshou Gu, Zaigang Chen and Andrew Ball, "Fault Prognosis and Diagnosis of an Automotive
Rear Axle Gear Using a RBF-BP Neural Network", 9th International Conference on Damage assessment of Structures,
Vol. 305, pp.1-11, 2011.
[20] Dawn An, Joo-Ho Choi and Nam Ho Kim, "A Tutorial for Model-based Prognostics Algorithms based on Matlab Code",
Annual Conference of Prognostics and Health Management Society, 2012.
[21] Yawei Hu, Shujie Liu, Huitian Lu and Hongchao Zhang, "Remaining useful life assessment and its application in the
decision for remanufacturing", 21st CIRP Conference on Life Cycle Engineering, Vol. 15, pp. 212-217, 2014.
[22] C. Okoh, R. Roy, J. Mehnen, L. Redding, "Overview of Remaining Useful Life Prediction Techniques in Through-Life
Engineering Services", Product Services Systems and Value Creation. Proceedings of the 6th CIRP Conference on
Industrial Product-Service Systems, Vol. 16, pp. 158-163, 2014.
AUTHOR BIOGRAPHY

The author, Mr. Tushar Salunkhe, born on 14-08-1989 completed Diploma in Mechanical Engineering from
Pimpri Chinchwad Polytechnic (MSBTE), Pune in the year 2010 and graduated in Mechanical Engineering
from Indira College of Engineering & Management (Pune University), Parandwadi, Pune, Maharashtra,
India in the year 2013 also pursuing his postgraduate study in Mechanical Engineering with specialization in
Design Engineering from Padmashree Dr. D. Y. Patil Institute of Engineering & Technology (Pune
University), Pimpri, Pune, Maharashtra, India.

The author, Prof. N. I. Jamadar, born on 03-03-1976 graduated in Automobile Engineering from P.D. A
College of Engineering, Gulbarga, Karnataka in the year 1998. He did his postgraduate study in Mechanical
Engineering with specialization in Machine Design at U. V. C. E College of Engineering (Bangalore
University), Bangalore, Karnataka in the year 2006. The author has been engaged in active research in the
field of vibration for the last couple of years. The author has 08 research papers to his credit which have been
published in various national and international journals. At present, He is working as Assistant Professor in
the Department of Mechanical Engineering at Padmashree Dr. D. Y. Patil Institute of Engineering &
Technology, Pimpri, Pune- 411018, Maharashtra, India.

The author, Dr. S. B. Kivade, born on 07-06-1970 graduated in Industrial and Production Engineering from P.
E. S. College of Engineering, Mandya, Karnataka in the year 1993. He did his postgraduate study in
Mechanical Engineering with specialization in Production Management at the Basveshwar Engineering
College, Bagalkot, Karnataka in the year 2001.The author has obtained his Ph. D degree in the year 2013 from
National Institute of Technology, Surathkal, Karnataka. The author has been engaged in active research in the
field of Noise Control, CAD, CAM and CAE, Operation Research, Operations Management, for the last
couple of years. The author has 08 research papers to his credit which have been published in various national
and international journals. He has also presented 04 research papers related to noise control at various national
and international conferences held in India and abroad. At present, He is working as Prof. and Head,
Department of Mechanical Engineering at the Basvakalyan Engineering College, Basvakalyan, Karnataka,
India.

135
View publication stats