Professional Documents
Culture Documents
Modeling of Power Consumption in Turning of Ferrous and Nonferrous Materials using Artificial Neural Network.
Mr. Mangesh R.Phate#1, Dr. V.H.Tatwawadi*2
Assistant Professor, Department of Mechanical Engineering, TSSMS , PVPIT, Bavdhan ,Pune-411021,Maharashtra,India * Principal, Dr. Babasaheb Ambedkar+ College of Engineering & Research, Wanadongri, Nagpur-441110, Maharashtra, India
AbstractDevelopment of artificial neural network (ANN) for prediction of power consumption in the turning of ferrous and nonferrous materials has been the subject of the present paper. ANN was trained through field data obtained on the basis of random plan of experimentation. Various influential machining field parameters were taken into consideration. The inputs were machine operator, work piece, cutting tool, cutting process parameters, machine specification and the machining field environmental parameters while the output was power consumed during the machining of ferrous and nonferrous materials. It was illustrated that a multilayer perception neural network could efficiently model the power consumption as the response of the network, with a minimum error. The performance of the trained network was verified by further observations.6-5-1 topology has been used for getting simulated result. The results of ANN were compared with the results of conventional turning (CT) observations. Keywords Artificial Neural Network, Mathematical Model, Turning Process, ferrous and nonferrous material & convectional machining.
#
operate sequentially and can do arithmetic computation extremely fast. Biological neurons in the human brain are extremely slow devices and are capable of performing a tremendous amount of computation tasks necessary to do everyday complex tasks, commonsense reasoning, and dealing with fuzzy situations. The underlining reason is that, unlike a conventional computer, the brain contains a huge number of neurons, information processing elements of the biological nervous system, acting in parallel. ANNs are thus a parallel, distributed information processing structure consisting of processing elements interconnected via unidirectional signal channels called connection weights. Although modeled after biological neurons, ANNs are much simplified and bear only superficial resemblance. Some of the major attributes of ANNs are: (a) they can learn from examples and generalize well on unseen data, and (b) are able to deal with situation where the input data are erroneous, incomplete, or fuzzy. II.
PROCESS VARIABLES AFFECTING THE POWER CONSUMPTION
I. INTRODUCTION Artificial neural networks (ANN) have already been applied to various aspects of machining processes, such as optimization of machining parameters [5]-[6], prediction of cutting load [7]-[8], surface roughness modeling[8]- [14], tool wear detection [15]-[16] and estimation of cutting tool stress [17]. Finesa and Agah [18], and El-Sonbaty et al [18] used neural network for positioning error compensation. Hao used ANN and genetic algorithm for modeling the thermal error in turning. This literature survey indicates that responses in cutting operations are well apt to be modeled by neural networks. To the knowledge of the authors, no work can be mentioned to have been done on the application of neural networks to convectional turning (CT). The authors have recently developed an ANN model for prediction of power consumption in the turning of ferrous and nonferrous material in CT. This work has been extended to include S.S.304, EN1A and EN8 as ferrous material and Aluminium 6063 and brass as a nonferrous material for the present study. ANNs offer a computational approach that is quite different from conventional digital computation. Digital computers
A. List of Variables under consideration The list of various parameters is as shown in table 1. B. Reduction of Variables by Buckinghams Pi Therom According to the theories of engineering experimentation by H. Schenck Jr. the choice of primary dimensions requires at least three primaries, but the analyst is free to choose any reasonable set he wishes, the only requirement being that his variables must be expressible in his system. There is really nothing basis or fundamental about the primary dimensions. For this case ,the variables are expressed in mass (M), length (L) , time ( T), temperature ( ) and angle ( ).The final dimensionless pi term id as shown in table 2.
Page 236
S. N. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44
Process variables affecting the turning process Description Sym Dimensions bol Anthropometric dimensions ratio An M0 L0 T0 0 0 of the operator. Weight of the operator. Wp M1 L0 T0 0 0 Age of the operator. AG M0 L0 T1 0 0 P Experience EX M0 L0 T1 0 0 Skill rating SK M0 L0 T0 0 0 Educational qualifications ED M0 L0 T0 0 0 U Psychological Distress PS M0 L0 T0 0 0 Systolic Blood pressure SBP M0 L0T1 0 0 Diastolic Blood pressure DBP M0 L0 T0 0 0 Blood Sugar Level during BSG M1 L-3 T0 0 0 Working Cutting Tool angles ratio. CT M0 L0 T0 0 0 AR Tool nose radius R M0 L1 T0 0 0 Tool overhang length Lo M0 L1 T0 0 0 Approach angle M0 L0 T0 1 0 Setting angle M0 L0 T0 1 0 Single point cutting tool BH M0 L0 T0 0 0 Hardness N Lip or Nose angle of tool LP M0 L0 T0 1 0 Wedge angle WG M0 L0 T0 1 0 Shank Length LS M0 L1 T0 0 0 Total length of the tool LT M0 L1 T0 0 0 Tool shank width SB M0 L1 T0 0 0 Tool shank Height SH M0 L1 T0 0 0 Work piece hardness BH M0 L0 T0 0 0 NW Weight of the raw work piece. W M1 L0 T0 0 0 Ultimate Shear stress of the sut M1 L-1 T-2 0 workpiece material 0 Density of the workpiece material DST M1 L-3 T0 0 0 Length of the raw workpiece LR M0 L1 T0 0 0 Diameter of the raw workpiece DR M0 L1 T0 0 0 Cutting Speed VC M0 L1 T-1 0 0 Feed f M0 L1 T0 0 0 Depth of cut D M0 L1 T0 0 0 Cutting force FC M1 L1 T-2 0 0 Tangential Force. FT M1 L1 T-2 0 0 Spindle revolution N M0 L0 T-1 0 0 Machine Specification ratio MS M0 L0 T0 0 0 P Power of the Machine motor HP M1 L2 T-3 0 0 Weight of the machine Wm M1 L0 T0 0 0 Age of the machine AG M0 L0 T1 0 0 M Atmospheric Humidity M0 L0 T0 0 0 Atmospheric Temperature DT M0 L0 T0 0 1 Air Flow Vf M0 L1T-1 0 0 Light Intensity LU M1 L0 T-4 0 0 X Sound level DB M0 L0 T0 0 0 Power consumption PC M1L2 T-3 0 0
C. Experimental plan For multifactor experiments two types of plans viz. classical plan or full factorial and factorial plan are available, in this experimentation conventional plan of experimentation is recommended. In all data was collected from total 585 experiments of five material S.S.304, EN1A, EN8, Al 6063 and Brass. The experimental set up is as shown in figure 1. III. MODEL FORMULATION BY ARTIFICIAL NEURAL NETWORK Artificial neural network (ANN) takes their name from the network of nerve cells in the brain. Recently, ANN has been found to be an important technique for classification and optimization problem. Artificial Neural Networks (ANN) has emerged as a powerful learning technique to perform complex tasks in highly nonlinear dynamic environments. Some of the prime advantages of using ANN models are their ability to learn based on optimization of an appropriate error function and their excellent performance for approximation of nonlinear function.
S. N 1 2 3 4 5 6 7
Pi ter m 1 2 3 4 5 6
D1
Dimensionless ratio
An*SBP*SK*Ag*Wp *SPO2 / DBP*PS*EDU*EX*BSG*D3 AR * r * * BHNT * LT*LP*LS / * LO* SW * SH * WG BHNW * W raw* LR * / D * FC * DST * DR f * FT * N * Tempwp* VB Tool / VB Machine * FC*VC SP * PHP * Wm/c / AGM* FC2 HUM*DTO *Vf *DB*VC*FC/ LUX*D3 PC/ FC * V
Nature of basic physical quantities Machine operator data Single point cutting tool Work piece material Cutting process parameters Lathe Machine Environmental data Power Consumption
Page 237
Target
Inputs
Comp are
Output
Adjust Weights
Page 238
(Eqn.1) Model 2 : For turning of Ferrous Material Correlation Coefficient =0.90789834590931 Root Mean Square =0.146175451030486 Reliability =88.87272727 %
S.N. 1 2 3 4 5 6
Parameters Number of hidden layer Learning factor Transfer function used Number of hidden neurons Number of epochs Momentum factor
Model 1: For Turning of Ferrous and nonferrous materials Correlation Coefficient =0.9161179771956546 Root Mean Square =0.157733431267774 Reliability = 87.80989011%
Fig-3: 6-5-1 ANN Model for turning of All Material data.
) /(1 e
1*Sum (layer1Cell 0 ) 1*Sum (layer1Cell 0 )
X
Where
sum
1,1
(1 e
)
* X * X
( layer
1 cell * X * X
0 ) 3 . 92600
0 ,3 0 ,6
* X * X
0 ,4
0 ,1
0 . 95984
0 ,2
1 . 15693 0 . 32753
3 . 95621 2 . 0614
10 . 17656
0 ,5
X
sum
1, 2
(1 e
1* Sum( layer1Cell1)
) /(1 e
1* Sum(layer1Cell 1)
Where
( layer 1 cell 1 ) 1 . 16225 * X * X
0 ,3 0 ,6
* X * X
0 ,1 0 ,4
0 . 48958 3 . 15324
* X * X
0 ,2 0 ,5
4 . 48538 0 . 62258
0 . 09865 1 . 5942
X
Where
sum
1,3
(1 e
1* Sum(layer1Cell 2 )
) /(1 e
1* Sum(layer1Cell 2 )
( layer
1 cell * X * X
2 ) 3 . 42859
0 ,3 0 ,6
* X * X
0 ,1
0 . 43029
* X * X
0 ,2
4 . 62208 3 . 31096
9 . 71165 1 . 1261
0 ,4
3 . 66868
0 ,5
X
Where
sum
1, 4
(1 e
1* Sum(layer1Cell 3 )
) /(1 e
1*Sum (layer1Cell 3 )
Fig-4: Comparison between Actual and ANN model for All material.
)
X
1 cell * X * X 3 ) 4 . 93480
0 ,3 0 ,6
1,1
(1 e
1*Sum (layer1Cell 0 )
) /(1 e
1*Sum (layer1Cell 0 )
( layer
* X * X
0 ,1
1 . 11872 1 . 27769
* X * X
0 ,2 0 ,5
Where
sum ( layer 1 cell * X * X 0 ) 2 . 21789
0 ,3 0 ,6
2 . 94682 1 . 27769
3 . 43607 1 . 0003
0 ,4
* X * X
0 ,4
0 ,1
0 . 31074
* X * X
0,2
1 . 93307
1*Sum( layer1Cell 4 )
5 . 72529 3 . 9346
15 . 17661
0 ,5
1,5
(1 e
1*Sum( layer1Cell 4 )
) /(1 e
0 . 23898
Where
sum ( layer 1 cell * X * X 4 ) 1 . 16632
0 ,3 0 ,6
X
* X * X
0 ,1
1, 2
(1 e
1* Sum( layer1Cell1)
) /(1 e
1* Sum(layer1Cell 1)
0 . 21854 0 . 47970
* X * X
0 ,2 0 ,5
Where
sum ( layer 1 cell 1 ) 2 . 30318 * X * X
0 ,3 0 ,6
0 . 20694 0 . 19439
0 . 46583 0 . 9168
0 ,4
* X * X
0 ,1
2 . 30318 1 . 72566
* X * X
0,2 0 ,5
0 . 78437
1*Sum( layer 2Cell 0)
1 . 90586 0 . 8833
0,4
PC (1 e
1*Sum(layer 2Cell 0)
) /(1 e
)
* X * X
1 . 44385
Where
sum ( layer 2 cell * X
1,3
X
0 ) 1 . 3815 1 . 04069 * X * X
1 ,1 1,4
1,3
(1 e
1* Sum(layer1Cell 2 )
) /(1 e
1* Sum(layer1Cell 2 )
1 . 25564 3 . 02215
1,2 1,5
0 . 46085 0 . 00000
Where
Page 239
* X
0 ,4
0 ,1
0 . 19367 * X
* X
0 ,5
0 ,2
0 . 18648 0 . 76082
2 . 51526 0 . 3423
* X
1 . 54239
X
Where
sum
1, 4
(1 e
1* Sum(layer1Cell 3 )
) /(1 e
1*Sum (layer1Cell 3 )
X
Where
sum * X * X
0 ,2 0 ,5
1,1
(1 e
1*Sum (layer1Cell 0 )
) /(1 e
1*Sum (layer1Cell 0 )
( layer
1 cell * X * X
0 ) 4 . 44212
0 ,3 0 ,6
* X * X
( layer
1 cell * X * X
3 ) 0 . 94223
0 ,3 0 ,6
* X * X
0 ,1
4 . 45814
0 ,1 0,4
14 . 01318 7 . 59484
* X * X
0 ,2 0 ,5
16 . 36121 7 . 76097
11 . 51879 5 . 7216
0 . 73124 4 . 35527
1 . 83730 2 . 0429
0 ,4
0 . 24247
X
sum
1,5
(1 e
1*Sum( layer1Cell 4 )
) /(1 e
1*Sum( layer1Cell 4 )
X
* X * X sum
0 ,2
1, 2
(1 e
1* Sum( layer1Cell1)
) /(1 e
1* Sum(layer1Cell 1)
Where
( layer 1 cell * X * X 4 ) 1 . 74585
0 ,3 0 ,6
Where
* X * X
0 ,1
0 . 31055 1 . 08793
( layer
1 cell 1 ) 3 . 68775 * X * X
0 ,3 0 ,6
* X * X
0 ,4
0 ,1
0 . 40033
* X * X
0 ,5
0,2
3 . 88275 1 . 78056
3 . 04742 1 . 4393
0 . 71806 1 . 79097
1 . 29215 0 . 2704
0 . 95935
0 ,4
0 ,5
PC (1 e
1*Sum(layer 2Cell 0)
) /(1 e
)
* X * X
1,5
X
Where
1,2
1,3
(1 e
1* Sum(layer1Cell 2 )
) /(1 e
1* Sum(layer1Cell 2 )
Where
sum ( layer 2 cell * X
1,3
0 ) 2 . 07441 0 . 31121
* X * X
1,4
1 ,1
0 . 64091
sum
( layer
1 cell * X * X
2 ) 0 . 12400
0 ,3 0 ,6
* X * X
0 ,1 0 ,4
2 . 76638 15 . 223226
* X
0 ,2 0 ,5
1 . 97879 0 . 00000
1 . 04151
6 . 04964 2 . 97805
15 . 95501 3 . 3621
* X
(Eqn.2)
1, 4
(1 e
( layer
1* Sum(layer1Cell 3 )
) /(1 e
1*Sum (layer1Cell 3 )
)
* X * X
Where
sum 1 cell * X * X 3 ) 4 . 74921
0 ,3 0 ,6
* X * X
0 ,1
0 . 14112 0 . 43721
0 ,2
0 . 40964 2 . 07350
1 . 68634 0 . 51199
0 ,4
0 ,5
1,5
(1 e
1*Sum( layer1Cell 4 )
) /(1 e
1*Sum( layer1Cell 4 )
Where
sum ( layer 1 cell * X * X 4 ) 14 . 15752
0 ,3 0 ,6
* X * X
0 ,4
0 ,1
2 . 07703
* X * X
0,2
6 . 45692 9 . 46460
1 . 73585 3 . 2115
12 . 98890
0 ,5
PC (1 e
1*Sum(layer 2Cell 0)
) /(1 e
)
* X * X
Where
sum ( layer 2 cell * X
1,3
0 ) 0 . 48587 6 . 80064
* X * X
1,4
1 ,1
6 . 89699
1,2
0 . 61445 0 . 00000
0 . 74602
1,5
(Eqn.3)
Fig-6: Comparison between Actual and ANN model for ferrous material.
Page 240
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
Fig8: Comparison between Actual and ANN model for nonferrous material.
[11] [12]
IV. CONCLUSIONS The power consumption for turning of ferrous and nonferrous maternal was modelled by artificial neural network, in the present study. The following characteristics of the network and training data could yield sufficiently accurate results: Multilayer perception was used for this purpose. The network was trained with field data carried out on the basis of random plan of experimentation and observation .The trained network was verified with separate observed data. Totally, 546 ( 330 for ferrous and 216 for nonferrous material ) observations were taken for training and testing the network. Three-layered back propagation network was proposed for modelling the power consumption. Sigmoid function and one hidden layer accommodating five neurons could converge to acceptable output accuracy after 950 epochs. The test of the trained networks showed good agreement existing between their predictions and the experimental results. ACKNOWLEDGMENT The author would like to great fully acknowledge the use of R.K. Engineers, Shivne, Pune .M.S, India Machine shop and Department of Mechanical Engineering, Priyadarshni college of Engineering, Nagpur, M.S, India.
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
Page 241