Você está na página 1de 7

Int J Adv Manuf Technol (2002) 20:277283 Ownership and Copyright 2002 Springer-Verlag London Limited

A Study of the Optimum Reheating Process for A356 Alloy in Semi-Solid Forging
Y. H. Kim1, J. C. Choi1, J. M. Yoon2 and J. H. Park2
1

School of Mechanical Engineering, Pusan National University, Pusan, Korea; and 2Graduate School, Pusan National University, Pusan, Korea

Semi-solid forging (SSF) is compared with conventional casting such as gravity die-casting and squeeze casting. A product without inner defects can be obtained from semi-solid forming with a globular microstructure. Generally speaking, SSF is composed of reheating, forging and ejecting processes. In the reheating process, the materials are heated up to the temperature between the solidus and liquidus line at which the materials exists in the form of liquidsolid mixture. The process variables such as reheating time, reheating temperature, reheating holding time, and induction heating power have much effect on the quality of the reheated billets. It is difcult to consider all the variables at the same time to predict the quality. In this paper, Taguchi method, regression analysis, and neural network were applied to analyse the relationship between processing conditions and solid fractions. A356 alloy was used, and the learning data were extracted by the reheating experiments. The results of a neural network were in good agreement with experimental results. Polynominal regression analysis was formulated by using the test data from the neural network. An optimum processing condition was derived to minimise the grain size, solid fraction standard deviation, and to maximise the average specimen temperature. Consideration is given to the reheating process of the raw material and results are presented of appropriate process variables for correct solid fraction, specimen temperature, and grain size. Keywords: Microstructure; Neural Network; Regression analysis; Reheating process; Semi-solid forming; Solid fraction

1.

Introduction

Semi-solid forging (SSF) is a new forging technology in which the billet is heated to the semi-solid state in a coexisting liquid
Correspondence and offprint requests to: Professor J. C. Choi, Department of Mechanical Engineering, ERC for NSDM at Pusan National University, 30 Changjeon-dong, Kumjeong ku, Pusan 609-735, South Korea. E-mail: jcchoihyowon.pusan.ac.kr

and solid phase. Many studies have been done since Fleming and Suerys proposal in the 1970s [1,2]. It took some 20 years for the thixoforging process to mature from the status of an academic curiosity to that of a nascent production process. A great number of workers have studied semi-solid processing as opposed to the conventional process methods such as die casting and squeeze casting which involve various problems, e.g. blowholes, segregation, and the thermal defects of the die. The SSF process comprises two steps: reheating of feedstock slugs to the semi-solid state and forging of the semi-solid material in a die. Reheating to the semi-solid state is particularly important in the thixoforging process. For the hypoeutectic Al-Si7Mg (A356, A357) type of alloys, the eutectic must be remelted completely in order to obtain good mechanical properties. If this is not done, unmelted, coalesced, polyhedral silicon crystals remain, which have an effect on the rheological properties during die lling and on the elongation of the nished part. A temperature that is slightly higher than the eutectic plateau does not adversely affect the mechanical properties, but does increase metal loss and decrease slug consistency. There are many variables in the reheating process. However, it is impossible to consider all the variables, consequently, eight variables were selected: reheating time (RT1, RT2, RT3), reheating temperature (RTemp1, RTemp2), reheating holding time (RHT), induction heating power (Q) and specimen size (SS). Some work on the reheating process has been reported. Kenney et al. [2] carried out experiments on the solid fraction of billets, 6070%. Hirt et al. [3] assumed that the die temperature could be between 150 and 300 for an optimum semisolid forging condition. Young and Fitze [4] made multistage reheating experiments on AlSi7Mg alloys with a solid fraction of 60%. Cremer et al. [5] controlled the induction heating by using a sensor. Jung and Kang [6] derived the optimal coil dimensions for a commercial induction heating system. However, there have not been any reports about an optimum reheating process using statistical methods and articial intelligence on semi-solid forging. Therefore, the optimum reheating conditions of A356 alloy with globular microstructure made by PECHINEY are proposed in this study. In addition, with

278

Y. H. Kim et al.

Table 1. Dimensions of the three kinds of specimen. Specimen size (mm) Li, li Di, di bi (A)(B) (C)(D) 76 70 60 70 76 100

70, 35 76, 30 5 Thermocouple positions

70, 35 60, 24 5

100, 50 76, 30 5

the assumption that the coil is suitable for the reheating process, an optimum reheating process is demonstrated by using reheating experiments, using the Taguchi method, neural networks, and regression analysis. This paper describes the development of a neural network model for the optimum reheating process. The model uses data analysis of the reheating process variables based on the Taguchi method. The developed model is then compared with the traditional polynomial regression model to assess its applicability in practice. The microstructures of semi-solid and the heated materials are investigated with an optical microscope; solid fraction, size, and distribution of grains are measured using an image analysis program.

Fig. 2. Measuring points of the cylindrical specimen.

2. Experimental Condition for the Optimal Reheating Process


The size of the specimen is shown in Table 1 and a photograph of the specimen is shown in Fig. 1. In this experiment, three kinds of the specimen are used and measuring points of the cylindrical specimen are shown in Fig. 2. The reheating processes were carried out using an induction heating system with a capacity of 50 KW and a frequency of 60 Hz. The parameters shown in Table 2 were used to determine the optimal reheating conditions. Generally, the relation between the solid fraction and the temperature is calculated by the following formula [7] or Scheils Eq. [8] which predicts the solid fraction of Al alloys. In semi-solid forming, a ne globular microstructure can be obtained in case of 6070% solid fraction [2]. Therefore, in this experiment, rapid quenching of the reheating specimens was carried out at 566C at a 65% solid fraction.

Fig. 3. Comparison between (a) the general reheating method with 1, 2, and 3 steps, and (b) the proposed temperature prole and application of the variables.

Fig. 1. The three kinds of cylindrical billet.

Figure 3(a) shows a general multi-step reheating method. However, this prole contains a sudden rise in temperature that causes such problems as irregular and non-uniform heating. Figure 3(b) shows the proposed temperature prole and the application of the variables in this study, that make satisfactory temperature control possible. In the present work, the Taguchi method for the reheating process of A356 alloys has been adopted for avoiding too many experiments and for investigating the contribution of the reheating process variables. An L27(313) orthogonal array is adopted to verify the effect of the parameters. The selected parameters consist of reheating time (RT1, RT2, RT3), reheating

Optimum Reheating Process in Semi-Solid Forging Table 2. Range of variables for the reheating process. Input variables Processing value Min. Reheating time, RT1, A (min) Reheating time, RT2, B (min) Reheating time, RT3, C (min) Reheating temperature, RTemp1, D (C) Reheating temperature, RTemp2, F(C) Reheating holding time, RHT, G (min) Induction heating power, Q, H (W) Specimen size, SS, I (mm) 5 4 4 410 530 2 3 7670 Med. 6 5 5 420 540 7 4 670 Max. 7 6 6 430 550 12 5 76100 Fig. 6. The effect of reheating time on the grain size.

279

Fig. 4. The effect of induction heating power on the solid fraction.

Fig. 7. Microstructure of semi-solid aluminium alloy with the same temperature. (a) 65.1% solid fraction. (b) 86.3% solid fraction.

Fig. 5. The effect of reheating temperature on the specimen temperature.

with those from the actual experiments, using neural networks and regression analysis.

temperature (RTemp1, RTemp2), reheating holding time (RHT), induction heating power (Q), and specimen size (SS). Additionally, solid fraction, temperature of the specimen, and grain size are selected for the quality feature. Variance analysis is performed in order to check the effect of the parameters on the reheating process. The properties of the Taguchi method for each reheating process variable are investigated in comparison

3. Results and Discussions


3.1 Taguchi Experimental Design Method

The reheating process variables of the solid fraction, specimen temperature and grain diameter were analysed from experimental conditions using Taguchis L27(38) orthogonal array. If

280

Y. H. Kim et al.

the lower the reheating temperature (RTemp1, RTemp2), the lower the solid fraction and the smaller the gradient of the specimen temperature. Grain size is apt to increase exponentially, as the solid fraction increases.
3.2 Neural Network

Fig. 8. The relation between the solid fraction and the grain diameter average.

Fig. 9. The back-propagation neural network with 5-4-3 network structure.

the induction-heating power is small (3 kW), the solid fraction does not reach the target value equivalent to 65%, as shown in Fig. 4. As the reheating holding time (G) of the specimen temperature becomes longer, a more uniform distribution of temperature can be obtained. There are cases in which the experimental value has a narrow range of temperature for a comparatively low reheating temperature (D, F) as shown in Fig. 5. When the reheating time (C), reheating temperature (F), and specimen size increases, generally, a large grain size can be obtained as shown in Fig. 6. Table 3 shows reheating process experiments and the data of the solid fraction based on the design of experiments method. The microstructure of experiments 13 (65.1%) and 17 (86.3%) for the reheating process of semi-solid aluminium alloy are shown in Fig. 7. Both specimens, as shown in Fig. 7, were heated up to 566C in different ways. However, the microstructures show a clear difference in solid fraction average which were 65.1% and 86.3%, respectively, which indicates that the microstructure is affected not only by the temperature, but also by the other process variables. Figure 8 shows the graph of the solid fraction and average grain diameter. It is found that as the solid fraction reduces, the accuracy of globularisation becomes better and the boundary between the solid and liquid becomes clearer. Moreover,

Using the Taguchi method and an articial neural network, we propose an optimal reheating condition for optimum solid fraction control of the reheated specimen. A three-layer neural network is used and a back-propagation algorithm is employed to train the network. The training of the neural network was performed using the back-propagation method by using experimental data sets sampled in the Taguchi method. This approach saves computing time and storage space. In addition, it provides easy extendibility as new data become available. The performance of the trained neural network for the reheating process is evaluated to examine how well it predicts the reheating process variables. Furthermore, an optimal solid fraction that satised both a small temperature gradient and a ne grain size is determined by applying the ability of data approximation of neural network. The neural network reduces the number of experiments for determining the optimal solid fraction in the reheating process. 5-4-3 back propagation neural networks were used, with full interconnection, consisting of an output layer, an input layer, and a hidden layer. Each layer has a set number of nodes that are chosen to t the problem at hand, as shown in Fig. 9. Network learning is performed with the training data to nd the weight matrix and node biases. In order to cope with the long training time of back propagation neural networks, some measures for improving efciency are taken. The input patterns and target outputs are normalised and scaled within the range 0.5 and 0.5. The learning rate and the momentum are adjusted during the training process for speeding up the convergence. Typical values of learning are 1.03.0 for the small training data set size, and 0.10.3 for the large training data set size. In the case of the optimum reheating process, a learning rate of 0.2 was used. The values of momentum were kept at less than 1.0. Results of the 5-4-3 neural networks and 8-5-3 neural networks learning are shown in Table 4. Figure 10 shows a comparison of the solid fraction between experiments and neural networks (5-4-3, 8-5-3). Even though the root mean square of the 8-5-3-neural network is smaller than that of the 5-4-3-neural network, the test results show that the neural network for the reheating process is able to predict the reheating process parameter values within an acceptable error. Figure 11 shows the microstructure of the specimen enlarged 100 times, after the reheating process. This microstructure was obtained by using the result of the analysis in Table 4, which shows very ne and globular grains.
3.3 Regression Analysis

Statistical analysis of the reheating variables can assess alterations in the reheating conditions. It then will become possible to set up parameters of the equipment correctly and to predict

Optimum Reheating Process in Semi-Solid Forging Table 3. Reheating process experiments and data of solid fraction based on design of experiments. Array number A 1 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 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 B 2 0 0 0 1 1 1 2 2 2 0 0 0 1 1 1 2 2 2 0 0 0 1 1 1 2 2 2 C 3 0 0 0 1 1 1 2 2 2 1 1 1 2 2 2 0 0 0 2 2 2 0 0 0 1 1 1 D 4 0 0 0 1 1 1 2 2 2 2 2 2 0 0 0 1 1 1 1 1 1 2 2 2 0 0 0 F 5 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 G 6 0 1 2 0 1 2 0 1 2 1 2 0 1 2 0 1 2 0 2 0 1 2 0 1 2 0 1 H 7 0 1 2 0 1 2 0 1 2 2 0 1 2 0 1 2 0 1 1 2 0 1 2 0 1 2 0 I 8 0 1 2 1 2 0 2 0 1 0 1 2 1 2 0 2 0 1 0 1 2 1 2 0 2 0 1 Solid fraction (%) 1 74.0 76.0 76.3 81.2 74.5 66.3 74.9 80.6 69.5 85.6 75.0 78.6 69.3 79.0 66.7 61.4 85.8 78.4 66.6 78.4 79.2 79.4 80.0 74.5 63.6 70.0 79.5 2 76.0 77.0 70.0 81.0 72.4 67.3 72.8 79.8 70.2 85.0 75.2 80.0 62.7 78.3 65.7 63.1 85.5 74.6 67.3 84.2 78.8 79.1 79.7 78.2 65.4 72.3 83.6 3 79.7 75.1 72.1 87.2 73.4 65.2 75.9 79.2 70.1 85.5 74.8 80.1 63.6 78.2 64.2 59.6 87.6 77.9 67.0 81.0 78.4 81.9 76.1 76.2 64.2 71.8 84.4 76.57 76.03 73.13 83.13 73.43 66.27 74.53 79.87 69.93 85.37 75.00 79.57 65.10 78.50 65.53 61.37 86.30 77.00 67.00 81.20 78.80 80.13 78.60 76.30 64.30 71.37 82.50 8.3634 0.9034 9.9234 12.4134 1.1034 1.1034 2.5034 0.4934 0.1434 0.1034 0.0400 0.7034 12.8250 0.1900 1.5834 3.0634 1.2900 4.2650 0.1250 8.4400 0.1600 2.3634 4.7100 3.4300 1.1100 1.4634 6.9100 Average (vi) S/N

281

Table 4. Optimum reheating conditions by the neural network. Experiment 5-4-3 Network 0.00484 2.02 107 C2D0F2G0H1 65.36 559.40 7.45 8-5-3 Network 0.001916 1.98 107 A1B1C2D0F2G0H1I0 65.59 560.07 7.63

Table 5. Selection of the best regression model and regression factors. (reference: X11 X1 X1). Regression model Object variables Selected variables Rsquare 0.2982 0.2312 0.1501 0.2783 0.2218 0.5259

EMS Iteration Array Solid fraction Temperature Grain diameter

A1B1C2D0F2G0H1I0 65.53 560.25 7.70

Solid fraction Selection best Temperature regression model Grain diameter 1st regression model Solid fraction Temperature Grain diameter Solid fraction Temperature Grain diameter

X4, X6, X7 (D, G, H) X4, X6, X8 (D, G, I) X3, X6, X7 (C, G, H) X3, X4, X5, X6, X7 (C, D, F, G, H)

2nd regression model

future trends in their changes. It is also a matter of statistics to nd out whether or not the as constant factors produce a signicant impact on the predictions given by a particular model. Regression analysis is used to illustrate what information can be obtained when the model used for process variables is analysed by linear regression. Quantitative characteristics of the model predictive ability were made in addition to standard statistical F-tests for model adequacy. Table 5 shows a selection of the best regression model and regression factors. The optimum regression model was established in the second regression model. The results of linear regression analysis enhance the predictive ability of the solid fraction, specimen temperature, and grain size.

X1, X2, X3, X4, X5, X6, 0.9998 X7, X8, X11, X22, X33, 0.9440 X44, X55, X66, X77, X88, 0.9967 X18, X25, X26, X27, X28, X35, X36, X38, X48

3.4 Comparison of Analysis Methods

In this study, the Taguchi method, a neural network, and regression analysis were used and experiments were performed. Table 6 shows the results of the reheating process variable optimisation according to the various analysis methods. Figure 12 shows a comparison of the results by NNL (neural network learning) REG, regression analysis, and an 8-5-3-neural network for deriving the solid fraction average. As shown in

282

Y. H. Kim et al.

Table 6. Results of the process variables according to various analysis methods. Analysis method Solid fraction 65.1000 65.3300 65.1841 66.3809 Specimen Grain temperature diameter 554.2500 556.5000 554.3728 555.6469 7.0000 9.2200 7.0274 7.3070

Array A1B1C2D0 F0G1H2I1

Experiment Taguchi method Neural network Regression analysis

Fig. 11. Microstructure obtained by the results of the analysis in Table 4. (a) 100. (b) 50.

Fig. 10. Comparison of solid fraction average in experiments and neural networks. (a) 5-4-3 neural network (r.m.s. 0.005). (b) 8.5.3 neural network (r.m.s. 0.002).

Table 6 and Fig. 12, it was found that the result of the 8-53-neural network was in good agreement with that of the experiment.

4.

Conclusions

This study has demonstrated that the new methods of articial neural networks and regression analysis are useful techniques for predicting the optimum process variables in a reheating process. The conclusions of the present study are as follows. 1. In a reheating process, other process variables as well as temperature should be considered to obtain the desired solid fraction of the semi-solid material. 2. The Taguchi method, articial neural networks, and regression analysis have been implemented for predicting

Fig. 12. Comparison of NNL REG, regression analysis and 8.5.3 neural network for solid fraction average.

the optimum solid fraction, specimen temperature, and grain size and for investigating the effects of the reheating variables on the reheating process. 3. Neural networks have been found effective in the selection of the optimal combination of reheating variables in the reheating process. The proposed method can give more economical and effective means for optimising a reheating process.

Optimum Reheating Process in Semi-Solid Forging

283

Acknowledgement

This work was supported by the Korea Science and Engineering Foundation (KOSEF) through the Engineering Research Center for Net Shape and Die Manufacturing at Pusan National University.

References
1. D. B. Spencer, R. Meharabian and M. C. Flemings, Rheological behavior of Sn-15%Pb in the crystallization range, Meteorological Translations 3, pp. 19251932, 1972. 2. M. P. Kenny, J. A. Courtois, R. D. Evans, G. M. Farrior, C. P. Kyonka, A. A. Couch and K. P. Young, Semisolid metal casting and forming, Metal Handbook 9th edn, vol. 15, pp. 327338, 1988.

3. G. Hirt, R. Cremer, A. Winkelmann, T. Witulski and M. Zillgen, SSM-forming of usually wrought aluminium alloys, Proceedings 3rd International Conference on Processing of Semi-Solid Alloys and Composites, University of Tokyo, pp. 107116, 1994. 4. K. P. Young and R. Fitze, Semi-solid metal cast aluminium automotive components, Proceedings 3rd International Conference on Semi-Solid Processing of Alloys and Composites, pp. 155 189, 1994. 5. R Cremer, A Winkelmann and G Hirt, Sensor controlled induction heating of aluminium alloys for semi solid forming, Proceedings 4th International Conference on Semi-Solid Processing of Alloys and Composites, University of Shefeld, pp. 159164, 1996. 6. H.-K. Jung and C.-G. Kang, A study on induction heating process of Al-6%Si-3%Cu-0.3%Mg alloy for thixoforming, Journal of the Korean Foundrymens Society, 19(3), pp. 225235, 1999. 7. I. Ohnaka, Introduction to Heat and Solidication Analysis by Computer, Marusen Press, pp. 196199, 1985. 8. M. C. Flemings, Solidication Processing, pp. 31 36, McGrawHill, New York, 1974.

Você também pode gostar