Proc. of Int. Conf. on Advances in Mechanical Engineering, AETAME

Effects of Squeeze Casting Process Parameters on Density of LM20 Alloy Manjunath Patel G. C1, Robins Mathew2, and Prasad Krishna2 123 1

Department of Mechanical Engineering, National Institute of Technology Karnataka-Surathkal-575025, India Email: [email protected], 2Email: [email protected] and 3Email: [email protected]

Abstract— The present work aims at investigating the effects of squeeze cast process parameters on the casting density of LM20 aluminium alloy. Density considered being one among the major quality characteristics for deciding the mechanical properties and mainly depends on the amount of shrinkage or gas porosities present in the casted samples. The experiments are carried out at different squeeze casting conditions to explore the effects of squeeze cast process parameters such as squeeze pressure, pressure duration, time delay, pouring and die temperatures on density of the casted samples. Time delay has shown negative effects, pressure duration, squeeze pressure has shown positive effects where as pouring and die temperature has shown adverse effects on the density of the LM20 alloy. In this work an attempt was made to develop the process model for the squeeze casting process using artificial neural network. The developed network is trained with levenbergmarquardt algorithm using the data collected from the real experiments and artificially generated data through regression equation. The trained network reduces the mean squared error to a very small value and predicts for the test cases with reasonable accuracy. It has been realized that the use of artificial neural network model helps to predict and select the optimal process parameters for the squeeze casting process. Index Terms— Squeeze casting process, Process parameters, Density and Artificial Neural Networks

I. INTRODUCTION Aluminium alloys are widely used as a casting material from past few decades to meet light weight requirements in automobile industries, save energy by recycling aluminium components, reduce fuel consumptions and better environmental protection [1]. Aluminium silicon (Al-Si) combination alloys have salient features such as excellent fluidity, formability, good castability, high specific strength, wettability, shrinkage reduction, corrosion resistance, wear resistance, low thermal expansion co-efficient, excellent mechanical properties [2-3]. Silicon is used as major alloying element in aluminium alloys, since silicon increases the fluidity, reduces melting temperature, low density, abrasion resistance, cost effective and easily available [2]. The addition of alloying elements such as magnesium (Mg) and copper (Cu) are mandatory to improve the strength of Al-Si alloys [2]. Al-Si-Cu-Mg-Ni family possess better casting characteristics and rate of solidification, minimum porosity, good structural integrity, modifies eutectic silicon particles, excellent mechanical properties and refined micro as well as macro-structure properties, when the casting made of these combination alloys [4]. © Elsevier, 2013

A lot of research work is done using advance squeeze casting process from the past few decades is due to its following advantages over conventional casting processes such as near net shape manufacturing ability, grain structure refinement, improved mechanical properties, minimum porosity, good surface finish and dimensional accuracy. However the most underlying defects occur in squeeze casting process such as hot tearing, oxide inclusion, under filling/overfilling, cold laps and case debonding and segregations [5]. The aforementioned defects finally affect the casting density of the processed alloy and casting density considered being one among the important quality characteristics in the present study, since it directly relates to the internal casting defects such as porosity, shrinkages and micro-voids. The amount of porosity content present in the casted alloy decreases the available load area, provoke stress concentration and crack initialization resulting in poor tensile strength and ductility of the alloy [6]. Vjian and Arunachalam (2005) studied the influence of squeeze pressures on density, hardness, ductility and tensile strengths of both solid and hollow components of the gun metal [7]. Krishna (2001) reported that high quality squeeze cast products are influenced by squeeze casting process variables and till date there is no universal standard available to obtain optimal process parameters to yield high quality squeeze cast parts [8]. Patel and Krishna (2013) reported artificial neural networks have potential applications in prediction, optimization, control, monitor, identification, modelling, and classification in the field of casting and injection moulding processes [9]. The present study aim for the following two objectives: 1. To study the influence of squeeze casting process variables such as squeeze pressure (Sp), pressure duration (Dp), time delay (Td), die (Dt) and pouring temperature (Pt) on casting density under experimentations. 2. To avoid manufacturing cost incur in selection of optimal process parameter setting under experimentation, The artificial neural network simulation model with Levenberg-Marquardt algorithm was developed to predict the casting density at different squeeze casting conditions. II. MATERIAL AND METHODS Chemical composition of the alloy: The quantitative chemical analysis was performed using optical emission spectrometer with reference standard ASTM E1251-07 and the obtained chemical composition of LM20 alloy is Si-10.41%, Fe-0.287%, Cu-0.177%, Mn-0.526%, Mg-0.175%, Cr-0.017%, Ni-0.016%, Zn-0.347%, Ti-0.175%, and Al-87.84% by weight. Squeeze casting process: The squeeze casting process involves the following three stages, 1. The measured quantity of the molten metal is poured into the pre-heated cylindrical die-cavity. 2. Pressure is applied through punch on the molten metal until the complete solidification takes place. 3. Punch is then retracted and the casting is ejected through ejector pins The major factors in influencing the density of the casting samples namely squeeze pressure, pressure duration, time delay before pressurization, pouring temperature and die temperatures. Higher time delay results in porous casting, where as lower time delay cannot help the inter-dendritic feeding to yield high dense casting components. Higher squeeze pressure and pressure duration affects the die life, low pressure and pressure duration may not be sufficient enough to eliminate all possible gasses affects the casting density. Low die and pouring temperature results in pre-mature solidification, where as high die and pouring temperatures increases the cycle time, amount of flash and affects the die life. High density casting samples can be obtained mainly by controlling the process variables. The choice of process parameters and their respective ranges used for experimental investigation is shown in table I. The selection is done through some pilot experiments conducted in lab and from the past literatures available. TABLE I. PROCESS PARAMETERS AND THEIR R ESPECTIVE LEVELS Process parameters Squeeze pressure, (Sp) Pressure duration, (Dp ) Time delay, (Td ) Pouring temperature, (Pt) Die temperature, (Dt)

Units MPa S S ˚C ˚C

Level-1 0.1 10 03 630 100

Level-2 50 20 05 660 150

Level-3 100 30 07 690 200

Level-4 150 40 09 720 250

Level-5 200 50 11 750 300

Experimental procedure: The melt was prepared using electrical resistance crucible furnace of maximum capacity 5 kg up to 900˚C. Cover flux was used to clean the melt and hexachloroethane (C2Cl6) tablet was used as degasser. Mica strip electric heater was used to pre-heat the die, J-type thermocouples and digital indicators are used to accurately measure the temperature of the melt and die. Water based graphite lubricant 777

is used to protect the die and for ease of removal of casting from the die. The measured quantity of the molten metal is poured into the preheated cylindrical die cavity made of H13 hot die steel. Pressure is applied through punch fitted in the middle of the crosshead of 40 tonne universal testing machine. Punch is then withdrawn and solidified casting is ejected through ejector pins. Experimental plan: Experiments were conducted by varying one parameter at a time and keeping the rest of the other process parameters at the middle level-3. Two replicates were considered for each experimental condition. The obtained squeeze cast samples were cut in the transverse direction, 10±0.1 cm of the cylindrical castings were taken for measurement. The density measurement was performed using Archimedes principle. The sample weights taken both in air and water using weighing balance of accuracy 0.5 mg. The casting sample is rinsed with methanol to ensure no air bubbles adhere to casting surface when the sample weights were taken under water. The obtained casting density results at squeeze casting conditions were tabulated in table III. III. RESULTS AND D ISCUSSIONS Regression analysis is carried out using the data collected from the experimental work as shown in table III (Exp. 1-20) to develop the relationship between the squeeze casting process variables and casting density. Regression analysis is one among the statistical tool helps the investigators/researchers to explore the effects, analyze the behaviour and to obtain the recommended process parameter setting corresponding to the input variables [10]. The analysis of squeeze cast process parameters effects on the casting density was done through surface plots obtained using Minitab software. Significance test was conducted to study the effects, contributions and the significance of the squeeze cast process parameters on the measured density. The results of the significance tests were presented in table II. TABLE II. SIGNIFICANCE TEST RESULTS OF DENSITY Term Constant Td Pd Sp Pt Dt Td *Td Pd *Pd Sp*Sp Pt*Pt Dt*Dt

Coef 2.63375 -0.0445 0.0066 0.0178 0.0075 0.0222 0.00108 0.00161 -0.00075 -0.00892 -0.03898

SE Coef 0.002146 0.002781 0.002781 0.002854 0.002781 0.002781 0.003677 0.003677 0.004085 0.003677 0.003677

T 1227.408 -16.001 2.373 6.236 2.697 7.983 0.295 0.439 -0.182 -2.425 -10.6

P 0 0 0.042 0 0.025 0 0.775 0.671 0.859 0.038 0

The terms used in table II is as follows [11]: Coef indicates the co-efficient used in (1) for representing the relationship between the density and the process parameters. The term SE Coef is the standard error for the estimated co-efficient, which measures the precision of the prediction. Smaller the standard error more precise will be the co-efficient. The T-value can be calculated based on the ratio of coefficient and the corresponding standard error. The T-value of the independent variable can be used to test, whether the predictor significantly affects the measured response. The p-value is the minimum value for a pre-set level of significance, at which the hypothesis of equal means for a given factor can be rejected. Considering 95% level of confidence, the significance of different factors and their interaction terms were tested. From the table II the square terms such as Td, Pd and Sp have p-value more than 0.05, hence these terms are considered to have no significant contribution towards the measured casting density at 95% confidence level. From table II, the term Sp and Dt have shown more positive contribution and Td shown negative contribution towards casting density. Experimental data was used to develop the regression equation between the input-output variables utilizing Minitab software. Table III (Exp. No 1-20), depicts the experimental data used to develop regression equation. The developed non-linear regression equation is shown in (1). R-square or co-efficient of determination is term describes how close the data points are fitted to the regression line or curve. In the present work the R2 value obtained for density 98.3%. In general, R2 value always lies between 0-100 %. Higher the R2 value indicates the model fits better with the observed values.

778

 = 1.21121 − 0.0120733 ∗ T + 8.80233E − 05 ∗ P + 0.000270477 ∗ S + 0.00354289 ∗ P + 0.001781 ∗ D + 6.77354E − 05 ∗ T

+ 4.03295E − 06 ∗ P

− 1.32509E − 07 ∗ S − 2.47673E − 06 ∗ P − 3.89751E − 06 ∗ D . (1) A. Effects of squeeze cast process parameters Effects of Time delay: Time delay in bringing the punch to come in contact with molten metal is varied between 3 to 11 seconds in this investigation as shown in Table-III. At 3 seconds of time delay yields near to theoretical density of the LM20 alloy of 2.68 g/cm3. As the Td increases density of the casting decreases because it is not possible to eliminate the gases accumulated on the surface with pressure application. Effects of Pressure duration: Pressure duration is the time up to which the punch is in contact with molten metal. The pressure duration is varied between 10-50 seconds as shown in table III (Exp. No 6-9). Since pressure is applied after time delay of 7 seconds, there is no significant improvement in density after pressure duration of 20 seconds. Short pressure duration may not take full benefit of pressure application leads to shrinkage in castings. Longer pressure duration affects die life, problem with punch retraction, longer cycle time and causes wall cracking [12]. Effects of Squeeze pressure: Squeeze pressure magnitude during experimentation is varied from atmospheric pressure of 0.1 MPa to 200 MPa shown in table III (Exp. No. 21 and 10-12). Theoretical density of the alloy is not reached at time delay of 7 seconds and applied pressure of 200 MPa. This is because at time delay of 7 seconds part of the metal already solidified and may not able to eliminate the gas fully resulted in reduced heat transfer co-efficient. However, casting density still improves with increase in applied pressure. Applied pressure forces molten metal into the die cavity leads to improved metal-mold interface by reducing the airgap between the interfaces results in higher heat transfer rate. The reason for low casting density is might be due to the existence of porosity when casted at atmospheric pressure (0.1 MPa). Higher pressure requires high tonne equipment capacity, increases amount of flash, reduce die life because of combination of high pressure and melt temperature [13]. TABLE III. EXPERIMENTAL OBSERVATIONS OF SQUEEZE CASTING DENSITY Exp. No 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

Td 3 5 7 9 11 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 --

Squeeze Cast Process Parameters Dp Sp Pt 30 100 690 30 100 690 30 100 690 30 100 690 30 100 690 10 100 690 20 100 690 40 100 690 50 100 690 30 050 690 30 150 690 30 200 690 30 100 630 30 100 660 30 100 720 30 100 750 30 100 690 30 100 690 30 100 690 30 100 690 --0.1 690

Dt 200 200 200 200 200 200 200 200 200 200 200 200 200 200 200 200 100 150 250 300 200

Density, ρ (g/cm3) ρ1 ρ2 2.672 2.678 2.645 2.650 2.632 2.628 2.616 2.606 2.589 2.575 2.621 2.623 2.629 2.626 2.638 2.629 2.638 2.633 2.617 2.613 2.636 2.642 2.646 2.656 2.616 2.603 2.623 2.629 2.629 2.626 2.624 2.631 2.569 2.565 2.597 2.609 2.622 2.620 2.611 2.616 2.549 2.556

Mean ρ (g/cm3) 2.6750 2.6475 2.6300 2.6110 2.5820 2.6220 2.6275 2.6335 2.6355 2.6150 2.6390 2.6510 2.6095 2.6260 2.6275 2.6275 2.5670 2.6030 2.6210 2.6135 2.5525

Effects of Pouring temperature: Low density values at high pouring temperatures of 720 -750˚ C was obtained shown in Table III (Exp. No. 13-16), may be due to the existence of shrinkage pores incur in the casting. High dense component obtained at 690˚ C of time delay 7 seconds, the reason might be pressure is applied when the melt temperature in the die was below its liquidus temperature and just above the temperature required for nucleation initiation. Small amount of flash was observed at high pouring temperatures during experimentation at die interfaces. High pouring temperature affects die life, wherein too 779

low pouring temperature result in cold laps on casting surface, incomplete die-filling due to inadequate fluidity [12]. Effects of Die pre-heat temperature: Die pre-heat temperature shown adverse affects on the casting density. Low and high die temperature resulted in less dense casting values shown in table III (Exp. No. 17-20). Low die temperature and time delay of 7 seconds results in pre-mature solidification before the pressure was applied. Hot spots and shrinkage pores might be the probable reason at high die temperatures for reduction in casting density. At low die temperatures, large temperature difference between the metal and the die interface exist leads to thermal fatigue failures of the die and cold laps on the cast surface [14]. Response surface plots: The surface plots help the researchers/investigators to visualize graphically the relationship of the squeeze cast process variables on the casting density. The surface plots obtained for the response-density were shown in Fig. 1. Following observations made from the surface plots are, 1. Increase in time delay for pressurization and decrease in pressure duration keeping the squeeze pressure, pouring temperature and die temperatures at middle level values reduces the casting density. The response surface seen to be almost a flat one indicates that the casting density have strong linear relationship with time delay and pressure duration as shown in Fig. 1 (a). However time delay parameter contributes more towards casting density compared to pressure duration. As time delay increases the metal starts solidifying from the outer casting surface and the applied pressure may not be possible enough to eliminate all possible gases. As the pressure duration decreases the metal is in contact with the die surface for shorter duration and makes the metal to pull away from the die surface results in lower casting density values. 2. Increase in squeeze pressure and decrease in time delay keeping pressure duration, pouring temperature and die temperatures at middle level values yields higher casting density. The response surface is almost flat as shown in Fig. 1 (b), which indicates squeeze pressure and time delay parameter have strong linear relationship with casting density. Increase in squeeze pressure brings sudden under cooling in the melt, helps the metal to push close enough to the die surface leads to improved metal-die interface results in improved heat transfer co-efficient and casting density. 3. Fig. 1 (c) shows the casting density has adverse affects with time delay and pouring temperatures. The resulting response surface shows slight curvature due to its non-linear relationship of pouring temperature and time delay when the squeeze pressure, pressure duration and die temperature at middle levels. This is because during squeeze casting process, pressure needs to be applied at the midway of solidus and liquidus temperature, which helps inter-dendritic feeding of the metal results in high dense cast components. Too low pouring temperature causes inadequate fluidity leads to pre-mature solidification results in in-complete die fill and cold laps on the casting surface. Whereas too high pouring temperature, in addition to long cycle time may cause shrinkage porosities in the casting. 4. Time delay and die temperature shows non-linear relationship with casting density shown in Fig. 1 (d), when squeeze pressure, pressure duration and pouring temperature was hold at their respective middle level. Higher time delay parameter and low die temperature results in low casting density values. The contribution of time delay parameter is more compared to die temperature towards the casting density. The adverse affects because at low die temperature the metal lose its fluidity very early leads to pre-mature solidification, whereas at high die temperature metal have enough fluidity for the longer duration, may cause hotspots and shrinkage pores in the casting leads to adverse casting density values.

(a)

(b)

780

(c)

(d)

(e)

(f)

(g)

(h)

(i)

(j)

Figure 1. Surface plots of casting density with (a) time delay and pressure duration, (b) time delay and squeeze pressure, (c) time delay and pouring temperature, (d) time delay and die temperature, (e) pressure duration and squeeze pressure, (f) pressure duration and pouring temperature, (g) pressure duration and die temperature, (h) squeeze pressure and pouring temperature, (i) squeeze pressure and die temperature, and (h) pouring temperature and die temperature

5. Fig. 1 (e) shows a rapid increase in casting density with increase in pressure duration and squeeze pressure when the time delay, pouring temperature and die-temperature at their respective middle level. The response surface found to be almost a flat one, which indicates a strong linear relationship of these input parameters with casting density. This increase in casting density is due to the improved contact area between the diemetal interfaces. 6. Casting density increases with increase in pressure duration and pouring temperature when time delay, squeeze pressure and die temperature holds at the middle level of their respective range as shown in Fig. 1 (f). The response surface shows there is slight curvature with increase in pouring temperature indicating that there is an nonlinear effects on casting density. However pouring temperature contribution is more towards casting density compared to pressure duration. Increase in casting density is due to improved metal-die interface because fluidity of the molten metal generally improves with high pouring temperatures. 7. Casting density increases marginally with increase in pressure duration and increases dramatically with die temperature initially and decreases slowly when squeeze pressure, time delay and pouring temperatures hold at their middle level (Refer Fig. 1 (g)). The response surface shows curvature indicating casting density has 781

non-linear relationship with pressure duration and die temperature. However, die temperature shown major contribution compare to the pressure duration towards casting density. Minimum casting density was observed at low values of pressure duration and dies temperature due to premature solidification and reduced heat transfer co-efficient. 8. Higher casting density was observed with increase in squeeze pressure and pouring temperature when time delay, pressure duration and die temperature were hold at their respective middle level as shown in Fig. 1 (h).The response surface shows squeeze pressure contribution more towards the casting density compared to pouring temperature. Increase in squeeze pressure improves the heat-transfer co-efficient by eliminating all possible gasses between the die-metal interfaces when the metal has higher fluidity at high pouring temperatures. 9. An increase in squeeze pressure increases the casting density, whereas increase in die temperature has shown initial increase in casting density and then decreases with increase in die temperature when the time delay, pressure duration and pouring temperature holds at their middle level. The Fig. 1 (i) show the contribution of die temperature towards the response is more compared to the squeeze pressure. 10. Fig. 1 (j), the casting density is seen to be decreased with decrease in pouring and die temperature when the squeeze pressure, pressure duration and time delay parameters hold at their middle level. Minimum casting density was observed at a combination of low die and pouring temperature. The surface plot shows die temperature contribution is more towards casting density when compared to pouring temperature. Confirmation experiment: From the experimental study the optimum combination of squeeze cast process parameter levels were identified. The confirmation experiment was conducted corresponding to optimal process parameter setting and yielded the casting density equal to the theoretical density of the LM20 alloy shown in table IV (Exp. No 22). B. Artificial Neural Network Simulation Model Artificial neural networks are the simplified models of our biological nervous system. Our biological nervous system consist of large number of inter connected processing units called as neurons [15]. The neurons of one layer connected to neurons of the other layer through connecting strength called as weights. The connection pattern formed within and between the layers termed as network architecture. The weighted inputs to a neuron are then passed to the transfer function which determines the neuron output. There may have multiinputs for a single neuron but having only single output. One of the most important interesting features of the artificial neural networks is its ability to learn from the training data and apply the past experience of the training process and produce the approximate results when presented with the data which was never experienced before. The schematic diagram of multi-layer feed forward neural network for predicting the casting density is shown in Fig. 2. Levenberg-Marquardt approximation algorithm was used to train the network. ANNs need to be trained using a large data base. Hence regression equation is used to generate the huge training data using (1). The data is normalized between -1 to 1 to avoid the numerical overflows and complex computation. Tan-sigmoid activation function was used in the hidden layer and pure linear function was used in the input and output layer. Total 250 data was used for training the network which includes 16 experimental data and 236 data obtained through regression equation. The experimental data used for training and testing is shown in table IV. Selection of optimum number of hidden neurons is done though experimentation based on minimum mean squared error. The final architecture with minimum mean squared error was obtained at twenty one hidden neurons shown in Fig 3. Hence in the present work 5-21-1 network architecture was used to check the network prediction accuracy which was never experienced during the training process. The results shown network training completed with the prediction accuracy of 0.04544% MAPE. Once the weights are optimized the five test subset were passed to the network. It was found that the network predicts with a reasonable accuracy of about 0.176524% MAPE, which is slightly superior compared to regression prediction as shown in table IV.

782

Figure 2. Structure of artificial neural network

Figure 3. Selection of hidden neurons

IV. C ONCLUSIONS Casting density evaluation as a function of squeeze cast process parameters were examined at different squeeze cast conditions and following conclusions were drawn from the present study: 1. Experiments conducted by varying one parameter individually and keeping the rest of the parameters at middle level. Increase in time delay before pressurization of the liquid metal shown negative effect on the casting density due to non-elimination of the gases completely. 2. Pressure duration and squeeze pressure shown strong linear relationship with casting density. Increase in squeeze pressure and pressure duration increases the casting density values because of improved contact area between the metal-die interfaces.

783

TABLE IV. C OMPARISON OF M EASURED CASTING DENSITY WITH REGRESSION AND ANN PREDICTIONS Squeeze casting process parameters Exp. Td Pd Sp Pt Dt No 1 3 30 100 690 200 2 5 30 100 690 200 3 7 30 100 690 200 5 11 30 100 690 200 6 7 10 100 690 200 7 7 20 100 690 200 9 7 50 100 690 200 10 7 30 50 690 200 11 7 30 150 690 200 12 7 30 200 690 200 13 7 30 100 630 200 15 7 30 100 720 200 16 7 30 100 750 200 17 7 30 100 690 100 18 7 30 100 690 150 20 7 30 100 690 300 Mean Absolute Percentage Error (MAPE) 22 3 40 200 690 14 7 30 100 660 19 7 30 100 690 8 7 40 100 690 4 9 30 100 690 Mean Absolute Percentage Error (MAPE)

3.

4. 5.

200 200 250 200 200

Response Density, ρ 2.6750 2.6475 2.6300 2.5820 2.6220 2.6275 2.6355 2.6150 2.6390 2.6510 2.6095 2.6275 2.6275 2.5670 2.6030 2.6135 2.6804 2.626 2.621 2.6335 2.611

Regression Equation Prediction Prediction Residual 2.673315 0.062982 2.650252 0.103962 2.627731 0.086257 2.584315 0.089667 2.622745 0.028399 2.624835 0.101437 2.635945 0.016871 2.615201 0.007702 2.639599 0.022695 2.650804 0.007398 2.611315 0.069556 2.629252 0.066697 2.626315 0.045086 2.566557 0.017267 2.606888 0.149361 2.610956 0.097343 0.060793 2.700091 2.621752 2.629087 2.631435 2.605752

0.734629 0.162137 0.308564 0.078422 0.200995 0.296949

ANN Prediction Prediction Residual 2.6735 0.056075 2.6479 0.015109 2.6278 0.083650 2.5826 0.023238 2.626 0.152555 2.6274 0.003806 2.6359 0.015177 2.6151 0.003824 2.6389 0.003789 2.6475 0.132026 2.6104 0.034489 2.6283 0.030447 2.6258 0.064700 2.5683 0.050643 2.6036 0.023050 2.6126 0.034437 0.04544 2.6831 2.6214 2.6295 2.6305 2.6066

0.100731 0.175151 0.324304 0.113917 0.168518 0.176524

Pouring and die temperatures shown adverse effects on the casting density, when the parameters varied between their respective ranges. Higher casting density was observed at middle levels indicating strong interaction effects on the response casting density. Higher die and pouring temperature increases the fluidity of the molten metal results in shrinkage castings where as low temperatures leads to premature solidification yields low density values. Confirmation test was conducted for the obtained optimal process parameter levels, which gives higher casting density values. Artificial neural network model was developed for the squeeze casting process. The developed model was trained with data collected from the experimental work and artificially generated data through regression equation. The trained neural network reduces the mean squared error to a minimum value. The accuracy of the developed network was tested for few test cases which are never experienced during the training process. The trained neural network outperforms the developed regression model in the present work. The developed ANNs can be used to predict and select the optimal process parameter setting by any novice user without having prior background knowledge about the squeeze casting process.

ACKNOWLEDGEMENT The authors wish to thank Department of Applied Mechanics and Hydraulics, National Institute of Technology Karnataka, Surathkal for providing research facilities REFERENCES [1] WS Miller, L Zhuang, J Bottema, AJ Wittebrood, “Recent development in aluminium alloys for the automotive industry,” Materials Science and Engineering: A, vol. 280.1, pp. 37-49, March 2000 [2] S Hegde, KN Prabhu, “Modification of eutectic silicon in Al–Si alloys,” Journal of materials science, vol. 43.9, pp. 3009-3027, May 2008 [3] VA Hosseini, SG Shabestari and R Gholizadeh, “Study on the effect of cooling rate on the solidification parameters, microstructure, and mechanical properties of LM13 alloy using cooling curve thermal analysis technique,” Materials & Design, vol. 50, pp. 7–14, September 2013 [4] A Maleki, A Shafyei, B Niroumand, “Effects of squeeze casting parameters on the microstructure of LM13 alloy,” Journal of Materials Processing Technology, vol. 209.8, pp. 3790-3797, April 2009 [5] DJ Britnell and K. Neailey, “Macrosegregation in thin walled castings produced via the direct squeeze casting process,” Journal of materials processing technology, vol. 138.1, pp. 306-310, July 2003

784

[6] HD Zhao, F Wang, YY Li and W Xia, “Experimental and numerical analysis of gas entrapment defects in plate ADC12 die castings,” Journal of materials processing technology, vol. 209.9, pp. 4537-4542, May 2009 [7] P Vijian and VP Arunachalam, “Experimental study of squeeze casting of gunmetal,” Journal of materials processing technology, vol. 170.1, pp. 32-36, December 2005 [8] P Krishna, “A study on interfacial heat transfer and process parameters in squeeze casting and low pressure permanent mold casting,” Ph.D. Thesis, University of Michigan, 2001 [9] M Patel and P Krishna, “A review on application of artificial neural networks for injection moulding and casting processes,” International Journal of Advances in Engineering Sciences, vol. 3.1, pp. 1-12, January 2013 [10] A. Sykes, An introduction to regression analysis, Law School, University of Chicago, pp. 1-5, 1993 [11] MB Parappagoudar, DK Pratihar and GL Datta, “Linear and non-linear modeling of cement-bonded moulding sand system using conventional statistical regression analysis,” Journal of Materials Engineering and Performance, vol. 17.4, pp. 472-481, August 2008 [12] S Rajagopal, “Squeeze casting: A Review and update,” Journal of applied metal working, vol. 1. No. 4, pp. 3-14, January 1981 [13] MR Ghomashchi and A. Vikhrov, “Squeeze casting: an overview,” Journal of Materials Processing Technology, vol. 101.1, pp. 1-9, April 2000 [14] B Kosec, G Kosec and M Sokovic, “Temperature field and failure analysis of die-casting die,” Archives of materials Science and Engineering, vol. 28.3, pp. 182-187, 2007 [15] DK Pratihar, Soft computing, Narosa publishing house pvt. Ltd, India, 2008

785