Professional Documents
Culture Documents
Benjamin Franklin
We take this opportunity to thank Prof. Rajesh P. Mishra for his constant support and
motivation that drove us forward throughout the project.
Nicholas Burke of Lakehead University Thunder Bay, Ontario was instrumental in our
search for Metropolis-Hastings algorithm.
Special thanks to Mr. Lokesh Patel, BITS Pilani for his insights on CNC machines, and
guidance in formulating a structure for the project.
A big shout-out to all our colleagues for the experiences, suggestions, opinions and ideas
shared over midnight chai from ANC.
For those who have touched our lives in any way since we started this endeavor, you all
know who you are, and we are truly grateful for all you have done.
Thanks GOD!
2
Table of Contents
3
List of Tables
4
List of Figures
5
List of Symbols and Abbreviations
A( ) availability
6
Abstract
7
Introduction
Technological advancements have driven dramatic improvements in industrial
productivity since the beginning of the Industrial Revolution. The steam engine powered factories
in the previous century, and electrification led to mass production in the early phase of the
twentieth century, and industry grew to be automated in the 1970s. In the decades that followed,
however, industrial technological improvements were only limited, especially compared with the
breakthroughs that transformed Information Technology, mobile communications, and e-
commerce. Now, though, we are in the midst of a fourth wave of technological advancement:
the rise of new digital industrial technology known as Industry 4.0, a transformation that is
powered by nine foundational technology advances. In this transformation, sensors, machines,
work pieces, and IT systems will be connected along the value chain beyond a single industry.
These connected systems (also referred to as cyber-physical systems) can interact with one
another using standard Internet-based protocols and analyze data to predict failure, configure
themselves, and adapt to changes. Industry 4.0 will make it possible to gather and analyze data
across machines, enabling faster, more flexible, and more efficient processes to produce higher-
quality goods at reduced costs. This in turn will increase manufacturing productivity, shift
economics, foster industrial growth, and modify the profile of the workforce—ultimately
changing the competitiveness of companies and regions.
Flexible manufacturing systems form the inseparable part of achieving the idea of
Industry 4.0. It is crucial to understand the systems that make up the flexible manufacturing
systems. In that aspect, the importance of computerized numerical control (CNC) machines is
quite high. The CNC turning center is one such machine tool widely used for mass production.
Every CNC machine tools manufacturer is trying to conquer the market by providing reliable
systems to the customers at the lowest cost possible. At the user end, the customer wishes to
manufacture machining components flawlessly without any breakdown caused by the sudden
failure of the manufacturing system. Hence reliability, availability, maintainability and
supportability (RAMS) is crucial in the manufacturing of CNC turning centers to hold on to the
dwindling customers who goes for the next big thing the moment they see one. The optimum life
cycle cost of the CNC turning center is one such factor that grabs the attention of the users. It is
also a measure of the manufacturer’s ability to survive in the technology-driven rapidly changing
market.
A typical CNC machining center consists of mechanical, hydraulic, pneumatic, electrical,
electronic and other sub-systems. Several studies have shown the reliability of CNC assisted
machine centers. Usually, mean time to failure (MTTF) and mean time to repair (MTTR) and
expert opinion data are considered for the reliability and availability analysis. Reliability data can
be used for generating reliability centered maintenance schedules.
8
Pareto analysis will also be done to identify the most critical subsystems in the CNC
turning center. A mathematical model consisting of two CNC turning centers functioning in
parallel will be formulated. Different possible states will be defined. Equations for mean time to
failure and steady-state availability will be formulated based on appropriate assumptions.
Bayesian approach will be used to estimate the unknown parameters that is required to find out
the system characteristics. Metropolis-Hastings algorithm will be used to obtain the distribution
for unknown parameters in MATLAB and system characteristics (mean time to failure and steady-
state failure) will be calculated.
9
Literature Review
10
is utilized to model the reliability characteristics. Some numerical experiments are performed to
illustrate the posterior analysis using Monte Carlo simulation methods.
11
Model Description
CNC Model
A CNC turning center is a complex system, with high-level automation and complicated
structure, which employs mechanics, electronics, hydraulics and so on. It is mainly composed of
the mechanical system; CNC system and hydraulic or/and air feed system. Fig. 1 is the system
block diagram of a typical CNC lathe. The mechanical system includes spindle and its transmission
(fixed in a headstock), two slide axes (named X, Z or U, W in turns), carriage apron, turret or tool-
holder, tailstock, bed and pedestal and so on. The spindle, with continuous or stepped continuous
speed change, is driven by AC or DC spindle motor directly or through main transmission, and
there is a photoelectric encoder on the spindle for thread turning. Both of X and Z-axes are driven
by AC or DC servomotors through ball lead screws, and controlled simultaneously. The turret or
tool-holder may exchange tools automatically. All of these are controlled by CNC system. A CNC
system, as the heart of a lathe, is usually comprised of power supply, main printed circuit board
(PCB) (usually a micro-computer), programmable logic controller (PLC) I/O PCB (which connects
the control panel, limit switch, button, magnets, turret and so on), axis PCB (which controls the
slide axes and the spindle through semi-closed or closed loop electronic control–motor drive and
photoelectric encoder), memory PCB (which connects additional encoder, CRT/MDI (manual data
input), manual pulse generator (MPG), backup battery and RS-232 serial communication device).
The CNC system and some electronic components, such as contactor switches, relays, regulators,
buttons and so on, are fixed in a cabinet. Other electronic components, such as limit switches,
proximity switches, encoders and so on, are located on the machine.
12
Taking into consideration the main subsystems as detailed above, meta data was collected on
based on the failure frequency over the multiple sub systems. All failure modes of CNC turning
center and their frequency of failures were found out and are detailed in table 1.
Pareto diagram was plotted to find out the most critical failure modes. It can be seen that the
main failure subsystems are the electric and electronic system, turret, CNC system, chuck and
clamping fixture, power supply, servo unit and so on. The electric and electronic system includes
contactor switches, relays, magnets, buttons, limit switches and so on, which are located on the
machine or in the cabinet. It can also be seen that the main failure subsystems for the mechanical
system are the turret and chuck.
13
Figure 2: Pareto diagram for failure modes
We consider a repairable manufacturing system with two identical CNC turning centers in
the context of imperfect coverage, which work independently and simultaneously. A detailed
description of the two-unit repairable system is given as follows.
A CNC turning center fails independent of the state of the other turning center and has
an exponential time to failure distribution with rate parameter λ. When a CNC turning center
fails, it is immediately detected, located, and recovered. We also assume that the coverage factor
for a turning center failure is denoted by c. That is, in case of failure, recovery can successfully be
performed with probability c. Typically, recovery takes a brief period of time and applications
may not be susceptible to such short interruptions (i.e., recovery can be accepted and tolerated).
Sometimes, however, the system does not successfully recover from a turning center failure and
suffers from a more severe impact. In this case, the system needs to be rebooted with longer
averaged duration. The recovery times are assumed to be exponentially distributed with rate
parameter Ө. The reboot times are exponentially distributed with rate parameter β. Failed
turning centers need to be repaired, having an exponential distribution with parameter μ. Only
one turning center can be repairable at a time and one turning center under repair does not
affect the proper working of the remaining unit. If no turning center is running correctly, the
whole system is out of service until the finish of the repair of the first turning center. Neither
reboot nor recovery is performed when the last turning center fails. In addition, various processes
(failure, repair, recovery, reboot) are assumed to be independent to each other.
Finally, let n denotes the states corresponding to the number of operating turning centers
in the repairable system (i.e., n=2,1,0) and N(t) denote the state of repairable system at time t.
Let RC and RB represent the recovery and reboot state of the failed turning centers respectively.
Therefore, n=2,1,0, RC, RB.
14
Figure 3: (a) The state transition diagram of reliability model for two-unit repairable system; (b) The state
transition diagram of availability model for two-unit repairable system.
Assume that the process is initially in state 2, so that P2(0) =1, P1(0) =0, P0(0) = 0, PRC (0) =0, and
PRB (0) = 0. Let Pn(s) be the Laplace transform of Pn(t). The system of differential equations using
Laplace transforms are obtained in terms of λ, Ө and μ as follows:
sP2 ( s ) − 1 = −2P2 ( s ) + P1 ( s ),
sP1 ( s ) = −( + ) P1 ( s ) + PRC ( s ),
sP0 ( s ) = P1 ( s ),
sPRC ( s ) = −PRC ( s ) + 2cP2 ( s ),
sPRB ( s ) = 2(1 − c)P2 ( s ).
15
( s + )(s + + )
P2 ( s ) = ,
( s + 2 )(s + )(s + + ) − 2c
2c
P1 ( s ) = ,
( s + 2 )(s + )(s + + ) − 2c
2c2
P0 ( s ) = ,
s[( s + 2 )(s + )(s + + ) − 2c ]
2c ( s + + )
PRC ( s ) = ,
( s + 2 )(s + )(s + + ) − 2c
2(1 − c) ( s + )(s + + )
PRB ( s ) = .
s[( s + 2 )(s + )(s + + ) − 2c ]
Assume both states 0 and RB are system-down states. After inverting P0(s) and PRB(s), we obtain
P0(t) and PRB(t), the probabilities that the system fails at time t. Let Z be the random variable
representing times to failure of the system; then P0(t)+PRB(t) is the probability that the system
fails at or before time t. Thus, the reliability function can be expressed as:
RZ (t ) = 1 − P0 (t ) − PRB (t ), t 0.
dRZ (t ) d ( P0 (t ) + PRB (t ))
Z (t ) = − =
dt dt
can then be written:
Using these equations, we obtain the MTTF using the derivative of Z(s) with respect to s evaluated
at s=0;
This section investigates availability behavior of the repairable system. From the state transition
diagram, the balance equations for the steady-state of the stochastic process are given by:
16
2P2 = P1 ,
PRC = 2cP2 ,
( + ) P1 = PRC + PRB + P0 ,
PRB = 2(1 − c)P2 ,
P0 = P1 .
2
P2 = ,
(2 + 2 + + c) + 2 [ + (1 − c) 2 ]
2
P1 = ,
(2 + 2 + + c) + 2 [ + (1 − c) 2 ]
2c 2
PRC = ,
(2 + 2 + + c) + 2 [ + (1 − c) 2 ]
22
P0 = ,
(2 + 2 + + c) + 2 [ + (1 − c) 2 ]
2(1 − c) 2
PRB = .
(2 + 2 + + c) + 2 [ + (1 − c) 2 ]
Thus, availability is
(2 + 2 + + c)
A() = .
(2 + 2 + + c) + 2 [ + (1 − c) 2 ]
An appropriate prior distribution for λ is a gamma distribution G (w1, v1) with density
1
p( ) = 1 1−1e = 1 / (1), 0
Denoted by λ ~ G ((w1, v1), where Г(.) is the gamma function andw1>0, v1>0 are specified
parameters, E (λ)= w1/v1 and Var(λ)= w1/v12. According to Bayesian theory and using (24) and
(25), the posterior distribution of λ given T1 is given by
17
which is the density of a gamma distribution with parameters n1+w1 and T1+v1.
Similarly, G ((w1, v1), i=2,3,4 are assumed as prior distributions for Ө, μ and β respectively. We
assume the prior distributions of all the system parameters are independent. Thus the joint
distribution of λ, Ө, μ and β is taken to be the product of prior distributions of each parameter.
Proceeding analogously, we obtain the joint posterior distribution which is given by
If we use standard gamma density G((wi,1), (i=1,2,3,4) as the prior distributions of λ, Ө, μ and β,
respectively, then the joint posterior distribution in (27) becomes
Monte Carlo simulation methods are used to obtain a sequence of random sample from a
probability distribution where it is impossible to obtain the variable directly. The general
algorithm for a MCMC Metropolis Hasting algorithm is as follows
18
Simulation study and comparisons
The 2-unit CNC system was considered for reliability where in the rate parameters for the system
were unknown. In this scenario, a Bayesian approach was adopted for formulation of a posterior
distribution which in term provides information on rate parameters λ, Ө, μ and β . The results of
the posterior are used to estimate the availability as well as the MTTF of the system. As all the
temporal variables such as failure time, reboot time, recovery time and standby time are
estimated from independent exponential distribution having the respective rate parameters. The
random variables are as follows:
𝑈𝑖𝑗𝑛 = (𝑈𝑖1 , 𝑈𝑖2 , 𝑈𝑖3 , … , 𝑈𝑖𝑛 )
19
The simulations are done for standard gamma prior in which the shape parameter of the gamma
distribution is taken as 1 in all cases. Posterior plots are plotted for values of n = 50,100,500,1000
to give a sense of how the mean propagates in the simulation over 10,000 iterations.
20
Figure 6: Posterior distribution for availability; standard gamma prior (n=500)
We see that as the n increases, the mean shifts to the true value and the distribution changes
from gamma to a normal distribution with the mean estimate. The use of Metropolis Hastings
for MCMC ‘s also shows that as the sample size taken from the random variable increases, the
mean tends to the true value.
21
Figure 8: Posterior distribution for MTTF; standard gamma prior (n=50)
22
Figure 10: Posterior distribution for MTTF; standard gamma prior (n=50o)
Figure 11: Posterior distribution for MTTF; standard gamma prior (n=1000)
The above figures give the plot of posterior distribution of MTTF over varying sample sizes as
followed in A( ) The results follow the same trend of increasing precision over large sample
sizes.
23
Conclusion
The reliability analysis dealt with in this report deals with identify the sub systems that
are prone to occur in a CNC system with Pareto analysis where the Electronic and Electrical
subsystems were identified as the vital few with a combined 47 % of failures attributed to these
systems. We also used a 2-unit CNC model setup to identify the MTTF and Availability
characteristics for systems where system parameters are unknown.
We have modelled the characteristics of the system on a Bayesian approach with a two-
parameter prior and validated the results. We notice that the reliability model can be used when
parameters are unknown by using Monte Carlo Markov Chains and are a good estimate of the
actual parameters of the system.
24
References
1. Hsu YL, Lee SL, Ke JC. A repairable system with imperfect coverage and reboot: Bayesian
and asymptotic estimation. Mathematics and Computers in Simulation. 2009 Mar
1;79(7):2227-39.
2. Patil RB, Kothavale BS. Failure modes and effects analysis (FMEA) of Computerized
Numerical Control (CNC) turning center. International Review of Mechanical Engineering.
2018;12(1):78-87.
3. Wang Y, Jia Y, Yu J, Zheng Y, Yi S. Failure probabilistic model of CNC lathes. Reliability
Engineering & System Safety. 1999 Sep 1;65(3):307-14.
4. Osaki S, Nakagawa T. Bibliography for reliability and availability of stochastic systems. IEEE
Transactions on Reliability. 1976 Oct;25(4):284-7.
5. Abu-Salih M, Anakerh N, Ahmed MS. Confidence limits for steady state availability.
Pakistan Journal of Statistics. 1999;6(2A):189-96.
6. de Almeida AT, de Souza FC. Decision theory in maintenance strategy for a 2-unit
redundant standby system. IEEE Transactions on Reliability. 1993 Sep;42(3):401-7.
7. Billinton R, Pan J. Optimal maintenance scheduling in a two identical component parallel
redundant system. Reliability Engineering & system safety. 1998 Mar 1;59(3):309-16.
8. Chandrasekhar P, Natarajan R, Yadavalli VS. A study on a two-unit standby system with
Erlangian repair time. Asia-Pacific Journal of Operational Research. 2004 Sep;21(03):271-
7.
9. Goel LR, Shrivastava P. Profit analysis of a two-unit redundant system with provision for
rest and correlated failures and repairs. Microelectronics Reliability. 1991 Jan
1;31(5):827-33.
10. Gururajan M, Srinivasan B. A complex two-unit system with random breakdown of repair
facility. Microelectronics Reliability. 1995 Feb 1;35(2):299-302.
11. Mi J. Interval estimation of availability of a series system. IEEE transactions on reliability.
1991 Dec;40(5):541-6.
12. Ke JC, Huang HI, Lin CH. Parametric programming approach for a two-unit repairable
system with imperfect coverage, reboot and fuzzy parameters. IEEE Transactions on
Reliability. 2008 Jul 9;57(3):498-506.
13. Chien YH, Ke JC, Lee SL. Asymptotic confidence limits for performance measures of a
repairable system with imperfect service station. Communications in Statistics-Simulation
and Computation. 2006 Sep 1;35(3):813-30.
14. Masters BN, Lewis TO, Kolarik WJ. A confidence interval for the availability ratio for
systems with Weibull operating time and lognormal repair time. Microelectronics
Reliability. 1992 Jan 1;32(1-2):89-99.
15. Moustafa MS. Reliability analysis of K-out-of-N: G systems with dependent failures and
imperfect coverage. Reliability Engineering & System Safety. 1997 Oct 1;58(1):15-7.
25
16. Pham H. Reliability analysis of a high voltage system with dependent failures and
imperfect coverage. Reliability Engineering & System Safety. 1992 Jan 1;37(1):25-8.
17. Rajamanickam SP, Chandrasekar B. Reliability measures for two-unit systems with a
dependent structure for failure and repair times. Microelectronics Reliability. 1997 May
1;37(5):829-33.
18. Seo JH, Jang JS, Bai DS. Lifetime and reliability estimation of repairable redundant system
subject to periodic alternation. Reliability Engineering & System Safety. 2003 May
1;80(2):197-204.
19. SK Srinivasan, R. Subramanian. Probabilistic analysis of redundant systems: Volume 175
in: Lecture Notes in Economics and Mathematical Systems, Springer, Berlin, 1980, vii+ 356
pages, DM 52.-, soft cover.
20. Trivedi KS. Probability & Statistics with Reliability, Queuing and Computer Science
Applications. PHI Learning Pvt. Limited; 2011.
21. Burke Nicholas Metropolis, Metropolis-Hastings and Gibbs Algorithms, Department of
Mathematical Science, Lakehead University Thunder Bay, Ontario.
26