Applied Thermal Engineering xxx (2014) 1e10

Contents lists available at ScienceDirect

Applied Thermal Engineering

journal homepage: www.elsevier.com/locate/apthermeng

Optimization of operating conditions for steam turbine using an

articial neural network inverse
Y.El. Hamzaoui a, J.A. Rodrguez b, *, J.A. Herna
ndez b, Victor Salazar b
a
Instituto de Ingeniera y Tecnologa, Dpto. Ingeniera El n, CU, Universidad Auto
ectrica y Computacio noma de Ciudad Juarez (UACJ),
Av. Del Charro # 450 Norte, CP 32310, AP 1594-D Ciudad Jua rez, Chihuahua, Mexico
b
Centro de Investigacion en Ingeniera y Ciencias Aplicadas (CIICAp-UAEM), Av. Universidad #1001, Col Chamilpa, CP 62209 Cuernavaca, Morelos, Mexico

h i g h l i g h t s

The failure assessment in blades is optimized using articial neural network inverse (ANNi).

(ANNi) is a very effective modeling the useful life in blades of steam turbines.

Failure assessment in blades is optimized using articial neural network inverse.

a r t i c l e i n f o a b s t r a c t

Article history: The useful life (UL) of the failure assessment in blades of steam turbines is optimized using the articial
Received 30 June 2014 intelligence. The objective of this paper is to develop an integrated approach using articial neural
Accepted 23 September 2014 network inverse (ANNi) coupling with a Nelder Mead optimization method to estimate the resonance
Available online xxx
stress when the UL of the blades is required. The proposed method ANNi is a new tool which inverts the
articial neural network (ANN). Firstly, It is necessary to build the articial neural network (ANN) that
Keywords:
simulates the output parameter (UL). ANN's model is constituted of feedforward network with one
Inverse neural network
hidden layer to calculate the output of the process when input parameters are well known, then
Optimal parameters
Optimization
inverting ANN. The ANNi could be used as a tool to estimate the optimal unknown parameter required
Steam turbine failure (resonance stress). Very low percentage of error and short computing time are precise and efcient,
Life cycle assessment in blades make this methodology (ANNi) attractive to be applied for control on line the UL of the system and
constitutes a very promising framework for nding set of good solutions.
2014 Elsevier Ltd. All rights reserved.

1. Introduction performance of steam turbines [6]. Therefore, the critical compo-

nents which determine the useful life of the turbine should be
Steam turbines have many applications in various industrial evaluated to determine the rehabilitation or replacement of them.
sectors. However, by common experience blade failures are the The critical components are the blades of steam turbines [7]. Most
main origin of operational breakdowns in these machines, causing of the existing analytical models used to predict the useful life of
great economic lost in turbo machinery industry. The turbines are the failure assessment in blades of steam turbines are based on
designed to work in stable operation condition [1e4]. Nevertheless, analysis using analysis of vibrations for the construction of the di-
failure in blades has been present after a short time period of work. agram of Campbell, which shows the natural frequencies of the
These failures commonly attributed to resonance stress of the blades like a function of the speed of the rotor (RPM) [8]. These
blades at different stages to certain excitation frequencies. The models do not provide reliable predictions for useful life (UL). This
expense of downtime and repair is about the millions of dollars [5]. is caused by the complexity of solving the equations that involve
The useful life (UL) is a very important variable for determining the the radiant energy balance, the spatial distribution of the absorbed
radiation, mass transfer, and the mechanisms of steam turbines [5].
Moreover, in the light of the rapid development witnessed by
the modern world in different elds of knowledge, science and
* Corresponding author. Tel.: 52 7772677638. technology, due to the increased speed of complexity of the system,
E-mail address: [email protected] (J.A. Rodrguez). in response to the issues requiring urgent attention of the people, in

http://dx.doi.org/10.1016/j.applthermaleng.2014.09.065
1359-4311/ 2014 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Y.E. Hamzaoui, et al., Optimization of operating conditions for steam turbine using an articial neural network
inverse, Applied Thermal Engineering (2014), http://dx.doi.org/10.1016/j.applthermaleng.2014.09.065
2 Y.El. Hamzaoui et al. / Applied Thermal Engineering xxx (2014) 1e10

order to check for access with high performance and reliability. In

recent years, computer science and articial intelligence have
received increasing attention proving effectiveness in solving many
issues and outstanding challenges [6]. Articial intelligence has
been used in a wide range of elds including industrial application
system has already proved successful, both on the technical level
and as reservoirs of expertise [7,8].
Therefore in practice an articial intelligence tools such as an
articial neural network (ANN) can provide a new approach to
process without take into account any previous assumptions [5].
ANN is a collection of interconnecting computational elements
which function like neurons in biological brain. It has the ability to
model processes by learning from input and output data, without
mathematical knowledge of the process. The (UL) could be
calculated on-line, when the input variables are well known in
blades of steam turbines. Indeed, the problem is that this (UL)
computed by ANN is not ideal in the system, and therefore it is
necessary that its input variables are well known when a given Fig. 1. General neural network inverse model.
(UL) is required, that means Find the effect of this cause. How-
ever, the term of neural network inverse can be misleading
sometimes. However, within the scope of this paper is to show the
new strategy which is proposed in our current paper based on learning algorithm, hyperbolic tangent sigmoid transfer-function,
ANN model about the use of articial neural network (ANN) for linear transfer-function and several neurons in the hidden layer
modeling the useful life of the failure assessment in blades of (due to the complexity of the process) are considered in the built
steam turbines. The main idea is coupling the neural network model. As soon as the model was validated, the second step was to
model with optimization algorithms to estimate the adequate invert the model. With the required output and some input pa-
value of a selected input to obtain the desired output. On the other rameters it is possible to calculate the unknown input parameters.
hand, to understand very well the difference between inverse and However, it is important to note that the analytical solution with one
optimization problem, Marcelo J. Colao and others [9] presented neuron in the hidden layer neural model exists, and it is described
basic concepts of inverse and optimization problems, and illus- in Section 4. Nevertheless, in the case that a proposed ANN model
trated the inverse and optimization problems are conceptually has more than one neuron in the hidden layer it is necessary to use
different, despite their similarities. According to Marcelo J. Colao an optimization method. On the other hand, in many cases, when an
et al., inverse problems are concerned with the identication of optimal output is required, the optimal input parameters are un-
unknown quantities appearing in the mathematical formulation of known, that's why, for this reason, we found that the inverse arti-
physical problems, by using measurements of the system cial neural network (ANNi) is a fundamental strategy to estimate
response. On the other hand, optimization problems generally deal the optimal operation condition. Rodriguez et al., have performed
with the minimization or maximization of a certain objective or the sensitivity analysis to show which parameters have the most
cost function, in order to nd design variables that will result on inuence on (UL) [4] in order to optimize them by means of inverse
desired state variables [9e15]. neural network (ANNi).
The problem was developed through inverting an articial The paper is organized as follows second section will give an
neural network (ANNi) to estimate the optimum input variables on overview about articial neural network inverse (ANNi), third
a required (UL) in the system. However, articial neural networks section will discuss the Nelder Mead method, fourth section is
inverse (ANNi) have been used successfully in different applications devoted to the optimization approach applied on ANNi, then the
in which an output desired parameter is selected and then the input fth section is assigned to the results and discussions. Whilst, the
parameters values are determined [6e8,16e29]. comparative study is showed in Section 6. Finally, the conclusions
The proposed method ANNi is a new tool which inverts articial on this work are drawn.
neural network (ANN) and it uses an optimization method to nd
the optimum parameter value (or unknown parameter) for given
required conditions in the process. In order to do so, rst, it is 2. Articial neural network inverse (ANNi)
necessary to build the articial neural network (ANN) model that
simulates the output parameters of the failure assessment in blades A general neural network is shown in Fig. 1 which is constituted
of steam turbines is constituted of a feedforward network with one by hyperbolic tangent (tanh) or sigmoid function (tansig) in the
hidden layer to simulate output, considering one or more well- hidden layer and linear transfer functions in the output layer. Then
known input parameters of the process. LevenbergeMarquardt the output is given by,

" !!# !
X
S X
K
Outputk yfkg purelin LWfk;sg $ tansig IWs;k $Ink b1s b2k (1)
s1 k1

Please cite this article in press as: Y.E. Hamzaoui, et al., Optimization of operating conditions for steam turbine using an articial neural network
inverse, Applied Thermal Engineering (2014), http://dx.doi.org/10.1016/j.applthermaleng.2014.09.065
 Y.El. Hamzaoui et al. / Applied Thermal Engineering xxx (2014) 1e10 3

According to the denition of purelin and tansig functions are

used, y(k2) is given by,

2 0 13

XS 66
B
B
C7
C7
6LW2;s $B 2
y2 6 B !!  1C 7
C7 b22 (2)
s1 4 @ PR A5
1 exp 2 r1 IWs;r $pr b1s

Let p(r3) be the input to be estimated and k 2 the required

output y(2) then

2 3

XS XS 66 2LW2;s
7
7
y2 b22  LW2;s 6 !!7 (3)
6 7
s1 s1 4 PR 5
1 exp  2 IWs;3 $p3 rs3 IWs;rs3 $pfrs3g b1s

where subscripts s is the number of neurons in the hidden layer; method approximately nds a local optimal solution with N vari-
subscripts r is the number of neurons in the input layer; l is the ables when the objective function varies smoothly. NeldereMead
number of neurons in output layer; S is the number of neurons in generates a new test position by extrapolating the behavior of the
the hidden layer; R is the number of input; Tansig is the hyperbolic objective function measured at each test point arranged as a sim-
tangent sigmoid transfer function; Purelin is linear transfer func- plex. Then, the algorithm chooses to replace one of these test points
tion; and IW, LW and b1s, b2l are the input and output weights and with the new test point. Thereby, a new simplex is generated with a
the biases, respectively. single evaluation of the objective. The numerical algorithm of the
The Eq. (4) is to be minimized to zero to nd the optimal input(s) NeldereMead simplex method has been described in detail by
parameter(s) in a general ANN, in this case, x is the p3 value to be Nelder and Mead [30].
computed to zero by an optimization method.

2 3

X
S XS 66 2$LW2;s
7
7
f x b22  LW2;s  y2 6 !!7 (4)
6 7
s1 s1 4 PR 5
1 exp  2 IWs;3 $x rs3 IWs;rs3 $prs3 b1s

Therefore, optimization can be done using the NeldereMead 4. Optimization approach

method.
4.1. Neural network learning
3. NeldereMead method
A learning (or training) algorithm is dened as a procedure that
The NeldereMead method is a generally used nonlinear opti- consists of adjusting the coefcients (weights and biases) of a
mization algorithm. This method is a numerical method to mini- network, to minimize an error function (usually a quadratic one)
mize to zero an objective function in a multi-dimensional space. between the network outputs, for a given set of inputs, and the
This algorithm is a direct search method that does not use nu- correct (already known) outputs as shown in Fig. 2. If smooth non-
merical or analytic gradient [27]. It attends to minimize a sca- linearities are used, the gradient of the error function can be
larvalued nonlinear function of n real variables using only function computed by the classical backpropagation procedure. To deter-
values, without any derivative information. The method uses the mine the best backpropagation training algorithm, ten back-
concept of simplex, which is a polyhedron of N 1 in N dimensions. propagation algorithms were studied. In addition, three neurons
Simplices are a line, a triangle and tetrahedron in one-, two-, and were used in the hidden layer for all backpropagation algorithms.
three-dimensional space, respectively, and so forth [28]. The Table 1 shows a comparison of different backpropagation training

Please cite this article in press as: Y.E. Hamzaoui, et al., Optimization of operating conditions for steam turbine using an articial neural network
inverse, Applied Thermal Engineering (2014), http://dx.doi.org/10.1016/j.applthermaleng.2014.09.065
4 Y.El. Hamzaoui et al. / Applied Thermal Engineering xxx (2014) 1e10

Fig. 2. Numerical procedure used for the ANN learning process, and the iterative architecture used by the model to predict the UL of the failure assessment in blades of steam
turbines (S is the number of the neuron in the hidden layer).

algorithms. LevenbergeMarquardt backpropagation training algo- Consequently, RMSE and R2 were used as the error function
rithm could have smaller mean square error (RMSE), on the other which measures the performance of the network. Therefore, the
hand, we found training with Levenberg Marquardt algorithm can network having minimum RMSE and maximum R2 was selected the
run smoothly in computer with lower expanded memory speci- best ANN model. More detail about neural network training was
cation (EMS), and the training time is quickly, than the other already described by Rodriguez et al. [4].
backpropagation algorithms. Because, the LevenbergeMarquardt
algorithm was designed to approach second order training speed
without having to compute the Hessian matrix. 4.2. Inverse neural network considering one neuron in the hidden
However, the performance of the ANN model was statistically layer in ANN model
measured by the root mean square error (RMSE) and regression
coefcient (R2), which are calculated with the experimental values In the case of one neuron in the hidden layer. The analytical
and network predictions as illustrated in Figs. 3 and 4. These cal- solution is performed as follow:
culations are used as a criterion for model adequacy, obtained as
follows:
If tansig and purelin are considered as the hyperbolic tangent
v sigmoid and linear transfer function in the hidden layer and
u ! output layer, and k 1.
u P  2
u N 0 1
u n1 yn;pred  yn;exp
t 2
RMSE (5) Out1 LW1;1 $@     1A b2
N 1 exp  2$ IW1;r $pr b1

 2 (7)
PN
n1 yn;pred  yn;exp This can be transformed into:
R2 1  (6)
PN  2
0 1
n1 yn;exp  ym
@ 2
Out1  b2 LW1;1 $     1A
where N is the number of data points, yn,pred is the network pre- 1 exp  2$ IW1;r $pr b1
diction, yn,exp is the experimental response, ym is the average of
(8)
actual values and n is an index of data.

Table 1
Comparison of 10 backpropagation algorithms with ve neurons in the hidden layer.

Backpropagation algorithm Function Root mean square Epoch Correlation Best linear equation
error (RMSE) coefcient (R2)

LevenbergeMarquardt backpropagation trainlm 0.00235005 1000 0.990 Y 0.990X 0.306

Batch gradient descent traingd 0.01657932 2000 0.988 Y 0.986X 0.927
Batch gradient descent with momentum traingdm 0.01982303 2000 0.987 Y 0.988X 0.837
PolakeRibiere conjugate gradient backpropagation traincgp 0.03267017 2000 0.979 Y 0.957X 2.53
Scaled conjugate gradient backpropagation trainscg 0.44944913 2000 0.974 Y 1.020X  0.783
BFGS quasi-Newton backpropagation trainbfg 0.48619630 2000 0.971 Y 0.982X 1.23
PowelleBeale conjugate gradient backpropagation traincgb 0.50820237 2000 0.965 Y 0.960X 2.03
One step secant backpropagation trainoss 0.02753381 2000 0.782 Y 0.617X45.3
FletchereReeves conjugate gradient backpropagation traincgf 0.01756329 2000 0.725 Y 0.425X 34.8
Variable learning rate backpropagation traingdx 0.02039637 2000 0.718 Y 0.386X 38

Please cite this article in press as: Y.E. Hamzaoui, et al., Optimization of operating conditions for steam turbine using an articial neural network
inverse, Applied Thermal Engineering (2014), http://dx.doi.org/10.1016/j.applthermaleng.2014.09.065
 Y.El. Hamzaoui et al. / Applied Thermal Engineering xxx (2014) 1e10 5

Fig. 3. Schematic methodology.

0 1
2$LW1;1    LW
Out1  b2 @     LW1;1 A 1;1  Out1 b2
1 exp  2$ IW1;r $pr b1 exp  2 IW1;1 $p1 b1 (13)
Out1 LW1;1  b2
(9)
2$LW1;1 !
     LW1;1  Out1 b2
Out1  b2 LW1;1
1 exp  2$ IW1;r $pr b1 2 IW1;1 $p1 b1 ln (14)
Out1 LW1;1  b2
(10)
!
   1 LW1;1  Out1 b2
2LW1;1 IW1;1 $p1 b1  ln (15)
1 exp  2$ IW1;r $pr b1 2 Out1 LW1;1  b2
Out1  b2 LW1;1
(11)
!
1 LW1;1  Out1 b2
   LW IW1;1 $p1  ln  b1 (16)
1;1  Out1 b2 2 Out1 LW1;1  b2
exp  2 IW1;r $pr b1 (12)
Out1 LW1;1  b2
Let p{r1} would be the input parameter to be calculated when
As k 1, R 1, So one output parameter is required. Then:

Please cite this article in press as: Y.E. Hamzaoui, et al., Optimization of operating conditions for steam turbine using an articial neural network
inverse, Applied Thermal Engineering (2014), http://dx.doi.org/10.1016/j.applthermaleng.2014.09.065
6 Y.El. Hamzaoui et al. / Applied Thermal Engineering xxx (2014) 1e10

  LW1;1
exp  IW1;1 $p1 b1 1 (21)
Out1  b2

 
LW1;1
IW1;1 p1 ln  1  b1 (22)
Out1  b2

 
LW1;1
b1  ln Out1b 1
2
p1 (23)
IW1;1

Therefore, according to Eq. (1), that it is possible to simulate the

outputs values, when input parameters are well known. However,
in many cases, the problem is that ANN predicted output values
which are not satisfactory in the system, and therefore, it is
necessary that its inputs variables are well known when giving a
required or satisfactory output. Consequently, the new control
strategy which is proposed here using ANN model applied to en-
ergy systems. The proposed strategy uses an inverse of neural
Fig. 4. Performance plots of ANN during training, validating and testing of the
network. network and the NeldereMead optimization algorithm to nd the
optimal input values for the required output value. Then in this
! ANNi methodology, as mentioned above, the required output value
1 LW1;1  Out1 b2 b1 is well-known.
p1  ln  (17)
2$IW1;1 Out1 LW1;1  b2 IW1;1
5. Results and discussion

If logsig and purelin are considered as the logistic tangent sig- The experimental set up used in this work has been previously
moid and linear transfer function in the hidden layer and output described in detail elsewhere [4]. Failures of turbine blade usually
layer, and k 1 initiate at the zone of high stress concentration which occur in
metallurgical discontinuities or where corrosion is present or even
0 1 in regions of excessive wear. During operation of the turbine, the
1 cracks are frequently caused by erosion, corrosion or small imper-
Out1 LW1;1 @   A b2 (18) fections and then propagate into the fracture. However, these im-
1 exp  IW1;1 p1 b1
perfections increase the fatigue stress concentration factor and of
LW1;1 course the stresses themselves. Turbine inspection revealed that
Out1  b2   (19) sets of 10 blades failure of the L-0 in low pressure stage of a 110 MW
1 exp  IW1;1 $p1 b1
steam turbine were illustrated within 15 cm from the root. The
blades are signicantly affected by a crack practically is a fracture
  LW1;1 observed in a blade root is shown in Fig. 5. The L-0 stage had 110
1 exp  IW1;1 p1 b1 (20) blades of 0.6 m in length with groups of 10 blades. The blades of
Out1  b2
each group are connected in their top end by a shroud and two
wires as shown in Fig. 6. In addition, a visual inspection combined
with a revision of the turbine operation history was carried out into
system description and experimental data. Furthermore, a turbine

Fig. 5. The cracked leading edge of the L-0 blade. Fig. 6. Discrete models of blade group for the calculation of natural frequencies.

Please cite this article in press as: Y.E. Hamzaoui, et al., Optimization of operating conditions for steam turbine using an articial neural network
inverse, Applied Thermal Engineering (2014), http://dx.doi.org/10.1016/j.applthermaleng.2014.09.065
 Y.El. Hamzaoui et al. / Applied Thermal Engineering xxx (2014) 1e10 7

ow analysis was also conducted together with verication of the

blade dimensions and tolerances. The natural frequencies and vi-  
bration modes analysis of the 10 blades group and the stress 2LW1;1 2LW1;2 2LW1;3
UL
analysis of blades were realized in experimental mode and a nite 1 eX1 1 eX2 1 eX3 (25)
 
element program called ANSYS [2,31e34].  LW1;1 LW1;2 LW1;3 b21
The operation history of the turbine was carefully studied and
revised. The turbine was operated by approximately 1800 h in
where
intermittent mode, with a record of 650 start-ups, during a period
of 5 years with at least ve replicates of information acquisition to
ensure the verication of measurements a database about 2500 
X1 2 IW1;1 V1 IW1;2 V2 IW1;3 V3 IW1;4 V4
samples was obtained According to the work developed by Garcia

[31], Kubiak [32] and Rodrguez [34]. IW1;5 V5 IW1;6 V6 b11 (26)
Rodriguez et al. [5] proposed a neural network model which
demonstrating an efciency of 99% in predicting useful life (UL) of
the failure assessment in blades of steam turbines. This developed 
ANN model has three neurons in the hidden layer (21 weights and 4 X2 2 IW2;1 V1 IW2;2 V2 IW2;3 V3 IW2;4 V4
biases) and considering 6 inputs parameters (resonance stress, 
frequency ratio, dynamic stress, damping, fatigue strength, mean IW2;5 V5 IW2;6 V6 b12 (27)
stress). The proposed equation developed by Rodriguez et al. [5] is

2 0 13

XS 66
B
B
C7
C7
6LW1;s $B 2
UL 6 B !!  1C7
C7 b2 (24)
PR  
s1 4 @ A5
1 exp  2$ r1 IWs;r $pr b1s

where the number of neurons in the input layer is 6. According to 

the model given by Eq. ( 24), it is possible to simulate the useful life X3 2 IW3;1 V1 IW3;2 V2 IW3;3 V3 IW3;4 V4
of the failure assessment in blades of steam turbines, while the 
input parameters are well known. Since we found, that the reso-
IW3;5 V5 IW3;6 V6 b13 (28)
nance stress is the most inuential parameter [5]. Therefore, it is
important to know in this process, what optimal resonance stress is
V1 Resonance stress [MPa]
needed for a required UL (input number 1). Consequently, we have
V2 Frequency ratio
developed a strategy to estimate the optimal resonance stress in
V3 Dynamic stress [MPa]
the failure assessment process about blades of steam turbines from
V4 Damping
the inverse articial neural network (ANNi). In the meantime,
V5 Fatigue strength [MPa]
Hern andez et al. [23] and El-Hamzaoui et al. [18] have been applied
V6 Mean stress [MPa]
ANNi with ANN model prediction in order to different processes of
UL Useful Life [Min]
R2 0.915 and R2 0.986, respectively. The authors mentioned that
it is possible to use ANN model with predictions of R2 > 0.985 for
At this step, we have obtained the function which has to be
implementing ANNi. In our case the ANN model is of R 0.99.
optimized to get the optimal input parameter:
Consequently, we believe that it is possible to developed ANNi in
this process. The results have been showed that the UL's error
2LW1;1 2LW1;2 2LW1;3
evaluation between the experimental and simulated by ANNi is f V1 A
0.7%. As we can see very small. In addition, the neural network 1 eX11 3:58V1 1 eX22 34:1V1 1 eX33 5:5V3
model developed by Rodriguez et al. has an efciency of 99%. (29)
Therefore, with this motivation mentioned above, we can use this
model to perform ANNi model.
Table 2
The proposed method (ANNi) inverts the articial neural Adjustable parameters obtained (weights and bias) in the proposed model with
network. Then we have the following equation that calculates UL S 3, K 6.
during the failure assessment process in blades of steam turbines. IW(s,k) Wi(1,1) Wi(1,2) Wi(1,3) Wi(1,4) Wi(1,5) Wi(1,6)
The key information (optimal performance) for useful life of the 1.79 1.14 2.53 1.21 2.37 1.17
failure assessment in blades of steam turbines, when controlling Wi(2,1) Wi(2,2) Wi(2,3) Wi(2,4) Wi(2,5) Wi(2,6)
the required output is to know the optimal input parameters. An 17.05 7.45 43.37 5.38 27.57 10.29
Wi(3,1) Wi(3,2) Wi(3,3) Wi(3,4) Wi(3,5) Wi(3,6)
inverted ANN could be considered as a model based method of
2.75 0.17 3.58 0.59 0.69 12.76
supervisory control, the control action in which the unknown input Wo(s) Wo(1) Wo(2) Wo(3)
parameters are obtained by solving an on line optimization prob- 0.19 0.13 0.81
lem for the desired output. The inverse of the articial neural b1(s) b1(1) b1(2) b1(3)
network deduced from Eq. (24) is the following Eq. (25) that cal- 123.17 17.15 6.39
b2 0.29
culates UL in the system.

Please cite this article in press as: Y.E. Hamzaoui, et al., Optimization of operating conditions for steam turbine using an articial neural network
inverse, Applied Thermal Engineering (2014), http://dx.doi.org/10.1016/j.applthermaleng.2014.09.065
8 Y.El. Hamzaoui et al. / Applied Thermal Engineering xxx (2014) 1e10

Table 3
Some samples of the experimental and simulated information of the system.

Test number 500 1000 2000 3000 4000 5000 6000 7000 8000 9000

Input
Mean stress 557.12 655.97 483.77 562.34 602.87 500.71 634.83 492.83 673.60 566.39
Frequency ratio 0.91 1.052 1.017 0.902 0.916 1.029 0.966 0.889 1.036 0.937
Dynamic stress 24.12 35.41 101.75 31.66 30.184 85.985 60.516 21.95 60.59 32.159
Damping 0.018 0.020 0.0201 0.023 0.0188 0.0201 0.0185 0.0201 0.0207 0.0198
Fatigue strength 1295.36 967.85 1186.33 977.33 923.95 1070.33 971.45 1128.079 883.3121 1038.347
Resonance stress_Exp 244 221 208 208 211.5 216 219 221 221.5 220
Output
[Useful Life]EXP 4.67E11 4.27E17 4.7855E10 5.38E14 6.87E14 3.19E14 1.54E14 4.18E14 6.03E12 3.04E13
[Useful Life]SIMANN 4.67E9 4.27E15 4.7855E8 5.38E12 6.87E12 3.19E12 1.54E12 4.18E12 6.03E10 3.04E11

stress. Whilst, the required output is the useful life. However, the
where resonance stress will be estimated by ANNi.
The optimization method nds the minimum of a scalar func-
A UL  b21 LW1;1 LW1;2 LW1;3 (30) tion of several variables, starting at an initial estimate. However,
this is generally referred to as unconstrained nonlinear optimiza-
 tion. However, the optimization method starts at the initial value
X11 2 IW1;2 V2 IW1;3 V3 IW1;4 V4 IW1;5 V5 and nds a local minimum V1 described in Eq. (29). On the other
 hand, Table 3, shows some samples of the experimental and
IW1;6 V6 b11 (31) simulated information of the system as shown in Fig. 7.
An example of this application is shown to calculate the
 required resonance stress (V1) considering the experimental data
X22 2 IW2;2 V2 IW2;3 V3 IW2;4 V4 IW2;5 V5
reported by Rodriguez et al. [34,35], which values are obtained

from experimental test database showing in test number 1000,
IW2;6 V6 b12 (32)
thus only, we want to calculate the resonance stress value (V1):
Mean stress (V6) 655.97 [MPa];

Frequency ratio (V2) 1.052;
X33 2 IW3;2 V2 IW3;3 V3 IW3;4 V4 IW3;5 V5
 Dynamic stress (V3) 35.41 [MPa];
IW3;6 V6 b13 (33) Damping (V4) 0.020;
Fatigue strength (V5) 967.85 [MPa];
The weights (IW and LW) and biases (b1 and b2) of ANN's model Resonance stress (V1) ?
are showed in Table 2 and the input parameters are reported by And an output value, UL 4.27E 17 [Min]
Rodriguez et al. [4] in order to minimize to zero the Eq. (29), an According to the weights and biases of Table 1 and optimization
optimization method (NeldereMead Simplex Method) is used to method of the NeldereMead, it is possible to calculate the optimum
calculate the resonance stress (V1). Where, the Input of ANN are Resonance Stress (V1), which is: Resonance Stress
frequency ratio; dynamic stress; damping; fatigue strength; mean ANNi V1ANNi 224 [MPa].

Fig. 7. Architecture of the articial neural network inverse for determining the optimum resonance stress.

Please cite this article in press as: Y.E. Hamzaoui, et al., Optimization of operating conditions for steam turbine using an articial neural network
inverse, Applied Thermal Engineering (2014), http://dx.doi.org/10.1016/j.applthermaleng.2014.09.065
 Y.El. Hamzaoui et al. / Applied Thermal Engineering xxx (2014) 1e10 9

Fig. 8. Experimental resonance stress vs resonance stress by ANNi.

However, to validate this value, a test was performed with 6. Comparative results
different data to optimize the resonance stress in different condi-
tions to demonstrate the feasibility of this method about ANNi. The remarkable thing is that, according to Fig. 9, there is good
However, the simulation outcomes were then compared with agreement between the predicted values for useful life of the fail-
experimental data in order to check the accuracy of ANNi. This error ure assessment in blades of steam turbines by ANN and ANNi
is given by: models with experimental data. Indeed, it has been an outstand-
ingly successful models in predicting the experimental results.
jExp  Simj
Err 100 (34) Consequently, the UL error between the experimental and simu-
Exp lated by ANNi is 0.7%.
That means Resonance Stress estimated by ANNi is compared to These models: articial neural network (ANN) and articial
experimental resonance stress of the experimental data using Eq neural network inverse (ANNi) prove to be very effective in
(34). In Table 2, within the test number 1000, the experimental modeling the useful life of the failure assessment in blades of steam
value of Resonance Stress is V1EXP 221 [MPa]. Whilst, the error is turbines. The smaller RMSE and larger R2 mean better performance
given by Eq (34). Therefore, in this case, an error of 1.5% is obtained [5]. However, the performance of the ANN and ANNi on modeling
which is very acceptable. On the other hand, the elapsed time to UL of the failure assessment in blades of steam turbines is pre-
calculate this Resonance Stress from this methodology (ANNi with sented in Fig. 9, where the two models are trained using the same
Nelder Mead Simplex) is only 35.17 s. It's seem that this time is good training datasets and validated by the same datasets (fresh data). In
enough to control the process. In addition, Fig. 8, illustrates that practice, however, the calculation required for system are so
there is a good agreement between the experimental resonance complicated, that's why, all the calculations were carried out on
stress and resonance stress estimated by ANNi. In the meantime, LINUX system, Intel D CPU 2.80 Ghz, 2.99 GB of RAM. According
the tting quality is so good. It has been an outstandingly successful to Fig. 9, we can distinguish the following results: The ANN model
models in estimating experimental results by ANNi. has smaller error for datasets than the ANNi (about 0.85%). In this

Fig. 9. UL versus number of test patterns for failure assessment of steam turbines.

Please cite this article in press as: Y.E. Hamzaoui, et al., Optimization of operating conditions for steam turbine using an articial neural network
inverse, Applied Thermal Engineering (2014), http://dx.doi.org/10.1016/j.applthermaleng.2014.09.065
10 Y.El. Hamzaoui et al. / Applied Thermal Engineering xxx (2014) 1e10

way, the ANN achieves better performances than the ANNi model. [10] M.J. Colao, G.S. Dulikravich, T.J. Martin, Optimization of wall electrodes for
electro-hydrodynamic control of natural convection effects during solidica-
Therefore, ANN is a good choice for modeling the useful life of the
tion, Mater. Manuf. Process 19 (4) (2004) 719e736.
failure assessment in blades of steam turbines. It is believed also, [11] M.J. Colao, G.S. Dulikravich, T.J. Martin, Optimization of wall electrodes for
that ANN and ANNi could be used to handle many other types of electro-hydrodynamic control of natural convection effects during solidica-
problems about the failure assessment in blades of steam turbines. tion, in: Proceeding ASME International Mechanical Engineering Congress &
Exposition, 2003.
[12] M.J. Colao, G.S. Dulikravich, T.J. Martin, Reducing convection effects in so-
7. Conclusions lidication by applying magnetic elds having optimized intensity distribu-
tion, in: Proceeding ASME Summer Heat Transfer Conference, 2003.
[13] M.J. Colao, H.R.B. Orlande, G.S. Dulikravich, F.A. Rodrigues, A comparison of
The useful life (UL) of the failure assessment in blades of steam two solution techniques for the inverse problem of simultaneously estimating
turbines is optimized using articial neural network inverse in order the spatial variations of diffusion, in: Proceeding ASME International Me-
to calculate an ideal input value from an ideal (UL) and taking into chanical Engineering Congress & Exposition, 2003.
[14] M.J. Colao, H.R.B. Orlande, Inverse forced convection problem of simulta-
account the above well known input values excepting required input neous estimation of two boundary heat uxes in irregularly shaped channels,
value as resonance stress. Then, NeldereMead method is applied in Numer. Heat Transf. A Appl. 39 (2001) 737e760.
the inverse problem to optimize the optimal operating condition is [15] J. Marcelo, M.J. Colao, H.R.B. Orlande, Inverse natural convection problem of
simultaneous estimation of two boundary heat uxes in irregular cavities, Int.
tested for a single parameter. Thanks to this method, it is possible to J. Heat Mass Transf. 47 (2004) 1201e1215.
nd any unknown input variable on line in the engineering failure [16] O. Cortes, G. Urquiza, J.A. Hern
andez, Optimization of operating conditions for
analysis in blades of steam turbines. Indeed, it is very important to compressor performance by means of neural network inverse, Appl. Energy
86 (11) (2009) 2487e2493.
note that the elapsed time to calculate the optimum input param-
[17] J.A. Hernandez, A. Bassam, J. Siqueiros, D. Juarez-Romero, Optimum operating
eter is only a few seconds (<40 s), thus it is feasible to get optimal conditions for a water purication process integrated to a heat transformer
parameters on line and is sufciently suitable to direct control of with energy recycling using neural networks inverse, Renew. Energy 34 (4)
steam turbines. Briey, ANNi integrated with NeldereMead method (2009) 1084e1091.

[18] Y. El Hamzaoui, J.A. Hernandez, S. Silva-Martnez, A. Bassam, A. Alvarez,
signicantly reduced the computational time with better conver- C. Lizama-Bahena, Optimal performance of COD removal during aqueous
gence for optimal solution for useful life of the failure assessment in treatment of alazine and gesaprim commercial herbicides by direct and in-
blades of steam turbines. Despite its successes, ANNi, is still in its verse neural network, Desalination 37 (2011) 325e337.
[19] Maamar Laidi, Salah Hanini, Optimal solar COP prediction of a solar-assisted
infancy. It's part of the future. In a way it's amazing we have done so adsorption refrigeration system working with activated carbon/methanol as
much with so little, and we have barely begun. However, if there are working pairs using direct and inverse articial neural network, Int. J. Refrig.
many input parameters to be found (solution to multi-parameter 36 (1) (2013) 247e257.
[20] Nour Hattab, Mikael Motelica-Heino, Application of an inverse neural network
problems) then NeldereMead method couldn't be able to solve model for the identication of optimal amendment to reduce copper toxicity
the optimization problem. It would be recommended to use another in phytoremediated contaminated soils, J. Geochem. Explor. 136 (2014)
advanced techniques for solving optimization problem, such as ge- 14e23.
[21] Yasaman Sanayei, Naz Chaibakhsh, Ali Chaibakhsh, Ali Reza Pendashteh,
netic algorithms (GAs) and particle swarm optimization (PSO). Norli Ismail, Tjoon Tow Teng, Long-term prediction of biological wastewater
treatment process behavior via wiener-laguerre network model, Int. J. Chem.
Acknowledgements Eng 2014 (2014) 1e7.
[22] Mohammad Rostamizadeh, Mashallah Rezakazemi, Kazem Shahidi,
Toraj Mohammadi, Gas permeation through H2-selective mixed matrix
J.A. Rodrguez, expresses his gratitude to CONACYT for the membranes: experimental and neural network modeling, Int. J. Hydrogen
project with title is: Experimental and numerical study for evalu- Energy 38 (2) (2013) 1128e1135.
[23] J.A. Hernandez, W. Rivera, D. Colorado, G. Moreno-Quintanar, Optimal COP
ation of reliability and life estimation of turbine blades under
prediction of a solar intermittent refrigeration system for ice production by
resonance conditions [In Spanish] whit reference number: 156757. means of direct and inverse articial neural networks, Sol. Energy 86 (4)
Y.El. Hamzaoui, expresses his gratitude to PROMEP for nancial (2012) 1108e1117.
support (Project PROMEP/103.5/13/7073.UACJ-PTC-289). [24] M. Behloul, H. Grib, N. Drouiche, N. Abdi, H. Lounici, N. Mameri, Removal of
malathion pesticide from polluted solutions by electrocoagulation: modeling
of experimental results using response surface methodology, Sep. Sci. Tech-
References nol. 48 (4) (2013) 664e672.
[25] Ian Buchanan, H.C. Liang, Vahid Razaviarani, Pesticides and herbicides, Water
[1] C.R.F. Azevedo, A. Sin atora, Erosion-fatigue of steam turbine blades, Eng. Fail. Environ. Res. 35 (2012) 1533e1567.
Anal. 16 (2009) 2293e2303. [26] Shahrokh Asadi, Jamal Shahrabi, Peyman Abbaszadeh, Shabnam Tabanmehr,
[2] J. Kubiak, J.A. Segura, G. Gonza lez, J.C. Garca, F. Sierra, J.G. Nebradt, A new hybrid articial neural networks for rainfallerunoff process modeling,
J.A. Rodriguez, Failure analysis of the 350 MW steam turbine blade root, Eng. Neurocomputing 121 (2013) 470e480.
Fail. Anal. 16 (2009) 1270e1281. [27] P.C. Wang, T.E. Shoup, Parameter sensitivity study of the NeldereMead sim-
[3] Y. Jung, P. Papadopoulos, R.O. Ritchie, Constitutive modelling and numerical plex method, Adv. Eng. Soft. 42 (2011) 529e533.
simulation of multivariant phase transformation in superelastic shapememory [28] G. Lepoittevin, G. Kress, Optimization of segmented constrained layer
alloy, Int. J. Numer. Meth Eng. 60 (2004) 429e460. damping with mathematical programming using strain energy analysis and
[4] D. Taylor, Crack modelling: a technique for the fatigue design of components, modal data, Mater. Des. 31 (2010) 14e24.
Eng. Fail. Anal. 3 (1996) 129e136. [29] Xin-She Yang, Engineering Optimization. An Introduction with Metaheuristic
[5] J.A. Rodrguez, Y.El. Hamzaoui, J.A. Herna ndez, J.C. Garca, J.E. Flores, Applications, 2010, ISBN 978-0-470-58246-6.
A.L. Tejeda, The use of articial neural network (ANN) for modeling the useful [30] J.A. Nelder, R.A. Mead, Simplex method for function minimization, Comput. J.
life of the failure assessment in blades of steam turbines, Eng. Fail. Anal. 35 7 (1965) 308e313.
(2013) 562e575. [31] J.C. Garcia, Determination of Induced Forces by Steam Flow in Turbines
[6] A. Chamekh, Mode lisation et Optimisation en Mise en Forme, Une approche (Doctoral thesis), CIICAp, Cuernavaca, Morelos, 2008 (in Spanish).
hybride base e sur la Methode des Elements Finis et les Re
seaux de Neurone [32] J. Kubiak, Diagnostic of Fracture of the Blade of the Steam Turbine, Internal
Articiels: Application  a l'emboutissage et a  l'hydrofomage, Europeenne Ed, report of EM-CIICAp, No 6P/DM CIICAp, University of the State of Morelos,
2010. 2002.
[7] A. Chamekh, H. BelHadjSalah, R. Hambli, A. Gahbiche, Inverse identication [33] J. Kubiak, J.A. Rodrguez, J.A. Segura, J.C. Garca, The effect of cracks in the
using the bulge test and articial neural networks, J. Mater. Process. Technol. steam turbine L-0 blades on their natural frequencies and their remaining life,
177 (1e3) (2006) 307e310. in: Proceeding of the International Conference on Electric Power, Rosemont,
[8] Abdessalem Chamekh, Souad BenRhaiem, Houda Khaterchi, 2007.
Hedi BelHadjSalah, Ridha Hambli, An optimization strategy based on a met- [34] J.A. Rodrguez, Experimentation and Simulation of Fracture in Blades of the
amodel applied for the prediction of the initial blank shape in a deep drawing Last Stage in Steam Turbines (Doctoral thesis), CIICAp, Cuernavaca, Morelos,
process, Int. J. Adv. Manuf. Technol. 50 (1e4) (2010) 93e100. 2009 (in Spanish).
[9] M.J. Colao, H.R.B. Orlande, G.S. Dulikravich, Inverse and optimization prob- [35] C. Cuevas Arteaga, J.A. Rodrguez, C.M. Clemente, J.A. Segura, G. Urquiza, Y.
lems in heat transfer, in: Proceedings the 10th Brazilian Congress of Thermal El. Hamzaoui, Estimation of useful life in turbines blades with cracks in cor-
Sciences and Engineering, 2004. rosive environment, Eng. Fail. Anal. 35 (2013) 576e589.

Please cite this article in press as: Y.E. Hamzaoui, et al., Optimization of operating conditions for steam turbine using an articial neural network
inverse, Applied Thermal Engineering (2014), http://dx.doi.org/10.1016/j.applthermaleng.2014.09.065