Professional Documents
Culture Documents
Abstract---Steam Turbine used in power station is a costly apparent in its early stages the problem can usually be
system. It is used to drive generators for producing electricity. resolved without any of the disruption caused when a turbine
Periodic monitoring and preventive maintenance is mandatory to has to be shut down. By appropriate trending of the various
minimize heavy loss in terms of time and money. The loss can be measurement points and the identification of excessive
sudden breakage of small to large sized mechanical rotating parts due
to improper maintenance. There are many methods of identifying the
vibration or movement, scheduled equipment stoppages or
fault that prevails when the turbine is operating. However, outages can often be utilized to investigate and resolve the
monitoring of vibration of rotating parts is very important as the failure mechanism.
vibration slowly escalates to different mechanical parts leading to Typical vibration measurement include: 1.Absolute
failure of components. vibration of bearing pedestals, 2.Shaft vibration relative to
Many statistical methods, artificial neural network algorithms, bearing, 3.Structural & foundation vibration monitoring,
evolutionary algorithms have been proposed by earlier researchers
Typical location measurement include: 1.Shaft eccentricity,
especially for bearing fault identification, gas turbine condition
monitoring. 2.Differential expansion or shaft movement, 3.Valve position
However, there are very limited publications that discusses about on steam inlet, 4.Casing expansion, both inner and outer,
implementation of ANN in condition monitoring of steam turbine. 5.Speed, including over speed and zero speed and
This paper implements Locally weighted projection regression 6.Temperature
(LWPR) which experimentally recognizes turbine bearings faults. Each of the measurement techniques are used to monitor the
Wavelet coefficients are obtained from the vibration signal and used turbine during its operating cycle, some measurements may be
as feature. The statistical parameters of the wavelet coefficients are
configured to provide warning alarms as well as automated
used to train the LWPR. The turbine working condition has been
categorized as good, satisfactory and bad. RBF with 5 X 7 X 1 shutdown, although these systems tend to operate on a voted
topology has been used to classify the conditions of turbine. principle to ensure maximum system integrity.
condition estimation
45 19.5
18
35
Acc=1, Hor
17.5
izontalal, μ
30 m 17
Acc=2, Ver
Dispalcement, μm
16.5
tical, μm
25
16
0 20 40 60 80 100 120
Test pattern numbers
Acc=2, Hor
20 Fig 4 Performance of LWPR
izontalal, μ
m
V. CONCLUSION
15 Acc=3, Ver This paper shows the extraction of features form the
tical, μm vibration signal followed by training and testing using the
locally weighted projection regression network. The following
10
Acc=3, Hor conclusions are obtained.
izontalal, μ The size of the network can be as small as possible to
5 m learn the features of the non linear signal.
The LWPR network can learn huge number of patterns
Acc=4, Ver unlike other neural networks.
0 tical, μm
1
25
49
73
97
REFERENCES
[1] Alguindigue, I. E., Loskiewicz-Buczak, A. and Uhrig, R. E., 1993,
Training patterns Monitoring and diagnosis of rolling element bearings using artificial
neural networks, IEEE Transactions on Industrial Electronics, Vol.40,
Fig 2 Displacement Values in the Steam Turbine pp.209–216.
[2] Atiya AF and Parlos AG, 2000, New results on recurrent network
0.045
training: Unifying the algorithms and accelerating convergence, IEEE
7 X 1 with receptive fields Trans. Neural Networks, Vol.11, Issue 3, pp.697-709.
0.04 [3] Chebil, J., Noel, G., Mesbah, M., Deriche, M., 2009, Wavelet
Decomposition for the Detection and Diagnosis of Faults in Rolling
0.035 Element Bearings, Jordan Journal of Mechanical and Industrial
Engineering, Vol.3, No.4, pp.260-267.
0.03 [4] Chow, M., Mangum, P. M., and Yee, S. O., 1991, A neural network
approach to real-time condition monitoring of induction motors, IEEE
Error