Document Type : Original Article
Authors
Department of Mechanical Engineering, Faculty of Engineering at Shoubra, Benha University, Cairo, Egypt.
Abstract
Keywords
EGTRIB JournalJOURNAL OF THE EGYPTIAN SOCIETY OF TRIBOLOGY VOLUME 17, No. 2, April 2020, pp. 13 - 22 ISSN 2090 - 5882 |
(Received December 22. 2019, Accepted in final form February 07. 2020) www.egtribjournal.com
SURFACE ROUGHNESS PREDICTION IN HARD-TURNING
WITH ANN AND RSM
Ahmed A. Elsadek, Ahmed M. Gaafer and S.S. Mohamed
Department of Mechanical Engineering, Faculty of Engineering at Shoubra, Benha University, Cairo, Egypt.
ABSTRACT
In the below investigation, artificial-neural-network (ANN) and response-surface-methodology (RSM) predictive tools shall be applied for predicting “surface roughness” on hard-turning of AISI H13 hot work steel. The mean relative error shall be utilized for testing the appropriateness of the created predictive models. Also, the influence of hardness of workpiece in addition to speed, feed and “depth of cut” on the surface roughness will be highlighted. The outcomes showed that the mean relative error for the RSM predictive model was 5.07% while the ANN model yielded a mean relative model of 2.21%. Besides, it was revealed that the feed then the workpiece hardness are the terms possessing the greatest influence terms on investigating the surface roughness in hard-turning. Where the feed rate increases the surface roughness while the workpiece hardness reduces it.
KEYWORDS
Response surface methodology, artificial neural network, hard turning, Surface roughness.
INTRODUCTION
Hard-turning is gaining importance as a pre-grinding process, because of its lower costs when compared with grinding owing to avoiding employing the grinding- wheels which costs more than hard-turning tooling, which is done on ordinary turning lathes. [1] Hard-turning is the process of turning workpiece having the hardness of 45–68 HRC [2], into finished components. The best benefit of hard-turning is reducing the machining time as well as improving the products quality in addition to other advantages stated in literature. [3–7] Among the various important applications of hard-turning is molds manufacturing, in which complicated geometries are cut in high hardness materials. Where short lead time along with high quality products are among the challenges that faces manufacturers. The product quality is becoming more significant owing to the strengthened industrial competitions and product quality realization. Thus the major consideration in machining industry is improving the whole performance of the cutting process. Javidi et al. [8] investigated the impact of feed rate and tool nose radius on maximum surface roughness. Ozel [9] “investigated the impact of workpiece hardness and cutting tool geometry on surface roughness on turning AISI H13 tool steel”. Elsadek et al. [10] applied fuzzy logic for roughness and tool wear prediction. Rangwala and Dornfeld [11] employed ANN for predicting the turning performance”. Kant et al. [12] joined the ANN along with genetic-algorithm (GA) for surface roughness prediction and optimization. Mia and Dhar [13] utilized ANN and RSM in predicting cutting temperature in dry and under high pressure coolant. They found that in some cases the RSM showed a higher prediction capability compared with ANN. Other researchers observed that ANN is better than RSM in prediction [14-15] all the time. Consequently, more research is needed to compare between the two models. Thus, the aim of the current work is discussing important aspects including the effect of cutting-parameters in addition to the workpiece hardness on the generated surface finish in hard-turning in addition to utilizing artificial intelligence and statistical techniques in implementing an efficient predictive model that is capable of predicting the surface roughness response based on the input parameters.
EXPERIMENTAL MATERIALS
Round bars of diameter 35 mm and 100 mm length of AISI-H13 tool steel hardened to 45±1, 50±1 and 55±1 HRC were utilized as workpieces. Table 1 shows the chemical composition of the employed alloy. A mixed ceramic insert of designation CNGA 120408 E040 was mounted on a negative rake angled shank. Dry turning tests were performed on a conventional lathe due to environmental concerns [16]. Surftest SJ-310”, Make: MITUTOYO was employed for surface roughness measurement.
Table 1. Chemical Composition of AISI-H13 workpieces
Element |
C |
Mo |
V |
Mn |
Mo |
Cr |
Si |
S |
P |
% |
0.390 |
1.250 |
0.920 |
0.48 |
1.250 |
4.88 |
1.09 |
0.002 |
0.012 |
Experimental Design
Response surface central-composite-design was employed for experimental design. The factors and factor levels are shown in table 2. Experiments were done and the results were reported. The experimental plans were established for the setting up of linear models for roughness. Table 3 reveals the planned experimental runs as proposed by the design.
Table 2. The factors and factor levels
Factors |
Levels |
||
Cutting Speed “v”(m/min) |
100 |
125 |
150 |
Feed Rate “f” (mm/rev) |
0.05 |
0.10 |
0.15 |
Depth of Cut “d” (mm) |
0.05 |
0.09 |
0.13 |
Hardness “h” (HRC) |
45 |
50 |
55 |
Table 3. Planned Experimental runs
Run |
Speed m/min |
Feed mm/rev |
Depth of cut mm |
Hardness HRC |
Run |
Speed m/min |
Feed mm/rev |
Depth of cut mm |
Hardness HRC |
1 |
100.00 |
0.15 |
0.13 |
45.00 |
16 |
100.00 |
0.05 |
0.05 |
45.00 |
2 |
100.00 |
0.05 |
0.13 |
55.00 |
17 |
125.00 |
0.10 |
0.09 |
50.00 |
3 |
150.00 |
0.05 |
0.13 |
45.00 |
18 |
125.00 |
0.15 |
0.09 |
50.00 |
4 |
125.00 |
0.10 |
0.09 |
50.00 |
19 |
125.00 |
0.05 |
0.09 |
50.00 |
5 |
150.00 |
0.05 |
0.13 |
55.00 |
20 |
125.00 |
0.10 |
0.09 |
50.00 |
6 |
125.00 |
0.10 |
0.09 |
50.00 |
21 |
150.00 |
0.15 |
0.13 |
45.00 |
7 |
150.00 |
0.15 |
0.05 |
55.00 |
22 |
150.00 |
0.05 |
0.05 |
45.00 |
8 |
125.00 |
0.10 |
0.05 |
50.00 |
23 |
125.00 |
0.10 |
0.09 |
50.00 |
9 |
100.00 |
0.15 |
0.13 |
55.00 |
24 |
100.00 |
0.15 |
0.05 |
45.00 |
10 |
150.00 |
0.15 |
0.05 |
45.00 |
25 |
100.00 |
0.05 |
0.13 |
45.00 |
11 |
100.00 |
0.05 |
0.05 |
55.00 |
26 |
100.00 |
0.15 |
0.05 |
55.00 |
12 |
125.00 |
0.10 |
0.09 |
55.00 |
27 |
125.00 |
0.10 |
0.13 |
50.00 |
13 |
100.00 |
0.10 |
0.09 |
50.00 |
28 |
125.00 |
0.10 |
0.09 |
50.00 |
14 |
150.00 |
0.05 |
0.05 |
55.00 |
29 |
150.00 |
0.10 |
0.09 |
50.00 |
15 |
125.00 |
0.10 |
0.09 |
45.00 |
30 |
150.00 |
0.15 |
0.13 |
55.00 |
RESULTS AND DISCUSSION
Analysis of variance (ANOVA) was utilized in testing the adequacy of the proposed model, table 4. The linear-model has an F-Value of 19.67, which indicates its significance. The linear model generated by the “design can be represented” by the following equation:
n |
i=1 |
Yu = βo + ∑ βi xi + e (1)
Where Yu is the required response (roughness), βo, β1... βi are the regression coefficients, ‘x’ is the “independent variables” and ‘e’ is the error. The above mentioned equation can be written to represents the roughness linear-model in terms of the above mentioned factors in the form of actual factors as follows:
Roughness “Ra” = 3.1351861 - 0.0027089 * Speed + 5.29222 * Feed
+ 1.101389 * “Depth of Cut” - 0.024378 * Hardness (2)
Table 4 ANOVA results for roughness
Source |
Sum of Squares |
df |
Mean Square |
F-value |
p-value |
Contribution |
Model |
1.65 |
4 |
0.4113 |
19.67 |
< 0.0001 |
76% |
A-Speed |
0.0826 |
1 |
0.0826 |
3.95 |
0.0580 |
3.80% |
B-Feed |
1.26 |
1 |
1.26 |
60.28 |
< 0.0001 |
58% |
C-Depth of Cut |
0.0349 |
1 |
0.0349 |
1.67 |
0.2079 |
1.61% |
D-Hardness |
0.2674 |
1 |
0.2674 |
12.79 |
0.0015 |
12.32% |
Residual |
0.5227 |
25 |
0.0209 |
|||
Lack of Fit |
0.4515 |
20 |
0.0226 |
1.59 |
0.3214 |
20.8% |
Pure Error |
0.0712 |
5 |
0.0142 |
|||
Cor Total |
2.17 |
29 |
It is clear from table 4 that both feed and hardness are statistically significant terms with contribution percent of 58% and 12.32% respectively while speed and depth of cut are insignificant terms. Similar findings were obtained by [17, 18] Inspecting the residuals is essential for testing the established model for adequacy. This analysis of residuals is necessary for confirming that assumptions for ANOVA are being met. Figure 1, reveals the residuals against run test. The figure shows that the value points are randomly distributed. The value points do not take a definite shape which is desirable. Furthermore, Figure 2 shows the predicted response vs the actual values in which the points of the predicted response and the actual value points are randomly distributed along a 45° line. This also indicates that the established model is adequate and there is no reason to suspicious any constant variance [19]. Then the suggested model is satisfactory and can be utilized as it comply with previous works.
Fig. 1 Residuals against run tests for surface roughness.
Fig. 2 Predicted response vs the actual values for surface roughness.
Figure 3 (a-b) reveals the 3D response surface curves for surface roughness. Where Figure 3a reveals that the best roughness value can be obtained at the combination of the least feed rate and depth of cut values. While Figure 3b reveals that surface finish is improved by utilizing the maximum values of speed and workpiece hardness. It is clear that the outcomes of the 3D plots agrees with statistical analysis.
(a) |
(b) |
Fig. 3 (a, b) 3D response surfaces curves for the surface roughness.
Artificial Neural Network
ANNs, are used for developing models in a similar way in that brains of humans treat information. The architecture, training algorithm, number of neurons, functions, weights and biases impact the accurateness of the ANN model. In the present work, feedforward back propagation network was utilized. Four nodes in the input layer representing speed (v), feed (f), “depth of cut” (d) and workpiece hardness (h) and one node for the output layer representing the predicted response that is the roughness. Figure 4. Reveals the structure. ANN predictive model is based on trial and error for finding the best results. The chosen parameters after trial and error are shown in Table 5. MATLAB R2015a ‘nntool’ toolbox was employed for training and testing the ANN. Testing the adequacy of the ANN predictive model RSM was employed. RSM predicts results based on the linear model illustrated in equation 2. Table 6 reveals the experimental and the corresponding RSM, ANN predicted values as well as the relative error for each.
Fig. 4 Surface roughness neural network structure.
Table 5 Selected ANN parameters for surface roughness prediction
Chosen ANN parameter |
Value |
Network structure |
4-9-1 |
Training/testing data |
24/6 |
Network algorithm |
Feedforward back propagation |
Transfer function |
Tansig, Purelin |
Training function |
Traingdx |
Learning function |
Learngd |
Performance function |
MSE |
Momentum constant |
0.9 |
Learning rate |
0.01 |
Regression plot between the experimental and predicted values for roughness by ANN and RSM is shown in Figure 5a and Figure 5b respectively. The value of regression coefficient is 94.946% for ANN and 87.212% for RSM that shows that ANN is more adequate in predicting surface roughness in hard-turning compared with RSM. The mean relative error for the ANN model was 2.21% while that for RSM is 5.07%. Consequently, the generated ANN model can effectively predict the response with a slight error. Figure 6 shows the scatter plot comparing the experimental and predicted values for both RSM and ANN respectively. It shows that ANN predicted value points at different runs are much closer to the experimental values than the value points predicted by RSM.
Table 6 Predicted values and corresponding relative errors for RSM and ANN modelling techniques for surface roughness.
Run |
Surface roughness (μm) |
Relative Error (%) |
|||
Experimental |
RSM |
ANN |
RSM |
ANN |
|
1 |
2.63 |
2.7 |
2.624 |
2.662 |
0.152 |
2 |
1.83 |
1.93 |
2.085 |
5.464 |
14.122 |
3 |
1.82 |
2.04 |
1.818 |
12.088 |
0.165 |
4 |
2.11 |
2.21 |
2.296 |
4.739 |
8.815 |
5 |
1.81 |
1.8 |
1.809 |
0.552 |
0.110 |
6 |
2.31 |
2.21 |
2.296 |
4.329 |
0.390 |
7 |
2.21 |
2.24 |
2.204 |
1.357 |
0.136 |
8 |
2.3 |
2.16 |
2.303 |
6.087 |
0.000 |
9 |
2.3 |
2.46 |
2.298 |
6.957 |
0.087 |
10 |
2.51 |
2.48 |
2.507 |
1.195 |
0.000 |
11 |
1.8 |
1.84 |
1.843 |
2.222 |
2.275 |
12 |
2.09 |
2.08 |
1.962 |
0.478 |
6.124 |
13 |
2.5 |
2.27 |
2.503 |
9.200 |
0.080 |
14 |
1.8 |
1.71 |
1.798 |
5.000 |
0.167 |
15 |
2.63 |
2.33 |
2.615 |
11.407 |
0.457 |
16 |
1.92 |
2.09 |
1.913 |
8.854 |
0.157 |
17 |
2.06 |
2.21 |
2.296 |
7.282 |
11.565 |
18 |
2.41 |
2.47 |
2.339 |
2.490 |
2.946 |
19 |
2.07 |
1.94 |
2.066 |
6.280 |
0.193 |
20 |
2.29 |
2.21 |
2.296 |
3.493 |
0.394 |
21 |
2.63 |
2.57 |
2.553 |
2.281 |
2.928 |
22 |
1.67 |
1.95 |
1.671 |
16.766 |
0.120 |
23 |
2.36 |
2.21 |
2.296 |
6.356 |
2.753 |
24 |
2.52 |
2.62 |
2.436 |
3.968 |
3.180 |
25 |
2.25 |
2.18 |
2.404 |
3.111 |
7.082 |
26 |
2.33 |
2.37 |
2.325 |
1.717 |
0.215 |
27 |
2.38 |
2.25 |
2.387 |
5.462 |
0.084 |
28 |
2.26 |
2.21 |
2.296 |
2.212 |
1.413 |
29 |
2.19 |
2.14 |
2.188 |
2.283 |
0.319 |
30 |
2.2 |
2.33 |
2.200 |
5.909 |
0.045 |
(a) |
(b) |
Fig. 5 Regression Plot for a. ANN and b. RSM.
Fig. 6 Experimental versus the predicted values of surface roughness
using ANN and RSM.
CONCLUSIONS
This research aims to highlight the influence of workpiece hardness in addition to speed, feed and depth of cut on the surface roughness in “hard-turning”. Furthermore, predictive models for predicting the values surface roughness in “hard-turning” were formulated utilizing ANN and RSM. Linear model was employed and tested by ANOVA. And the below can be concluded:
1. Workpiece hardness plays an important role in improving the surface roughness in hard- turning. On the other hand, feed rate was found to possess the major negative effect on it, such that to hard turn a part with the best possible surface quality, the feed rate values must be kept as low as possible.
2. ANN predictive model proved its appropriateness and adequacy over the RSM model for each individual response, based on the vales of the mean relative error and the regression coefficient values.
3. ANN can be used effectively in implementing an efficient predicting system so as to be able to predict precise values of the outcome responses based on the input values of cutting parameters and workpiece hardness, in hard-turning operations.
REFERENCES
1. Hard Turning and the Machine Tool, Daniel P. Soroka, Hardinge Inc. One Hardinge Elmira, New York, pp. 1 – 7, (2002).
2. Huddle D., “New hard turning tools and techniques offer a cost-effective alternative to grinding”. Tooling and Production Magazine Vol. 80, pp. 96–103, (2001).
3. Narutaki N., Yamane Y., “Tool wear and cutting temperature of CBN tools in machining of hardened steels”, Annals of the CIRP 28/1, pp. 23 - 28, (1979).
4. Hodgson T., Trendler P. H. H., Michelletti G. F., “Turning hardened tool steels with Cubic Boron Nitride inserts”, Annals of the CIRP 30/1, pp. 63 - 66, (1981).
5. Chryssolouris G., “Turning of hardened steels using CBN tools”, Journal of Applied Metal Working 2, pp. 100 - 106, (1982).
6. Koenig W.A., Komanduri R., Toenshoff H. K., Ackeshott G., “Machining of hard metals, Annals of the CIRP 33/2”, pp. 417 – 427, (1984).
7. Koenig W., Klinger M., Machining hard materials with geometrically defined cutting edges, Annals of the CIRP 39/1, PP. 61–64, (1990).
8. Javidi A., Rieger U., Eichlseder W., “The effect of machining on the surface integrity and fatigue life”, Int. J. Fatigue, Vol. 30, pp. 2050 – 2055, (2008).
9. Ozel T., “Modeling of hard part machining: effect of insert edge preparation in CBN cutting tools”, J. Mater. Process. Technol., Vol. 14. Pp. 284 – 293, (2003).
10. Elsadek A. A., A. M. Gaafer A. A., Lashin M. A., “Prediction of Roughness and Tool Wear in Turning of Metal Matrix Nanocomposites”, J. Eng. App. Sci., 64, pp. 387 - 408, (2017).
11. Rangwala, S. S, Dornfeld D. A., “Learning and optimization of machining operations using computing abilities of neural networks”, IEEE Trans Syst Man Cybern., 19, pp. 299 - 314, (1989).
12. Kant, G., Sangwan, K. S., “Predictive modelling and optimization of machining parameters to minimize surface roughness using artificial neural network coupled with genetic algorithm”, Procedia CIRP 31, pp. 448 – 453, (2015).
13. Mia M., Dhar N. R., “Optimization of surface roughness and cutting temperature in high-pressure coolant-assisted hard turning using Taguchi method”, Int. J. Adv. Manuf. Technol., 88, pp. 739 - 753, (2017).
14. Zerti A., Yallese M. A., Zerti O., Nouioua M., Khettabi R., “Prediction of machining performance using RSM and ANN models in hard turning of martensitic stainless steel AISI 420”, Journal of Mechanical Engineering Science, 233(13), pp. 4439 - 4462, (2019).
15. Sharmin I., Dhar N. R., “Application of ANN and RSM for predicting surface roughness in turning Al based MMC with coated carbide insert”, AIP Conference Proceedings. pp. 1 - 5, (2018).
16. Jacobson M., Dahlman P., Gunnberg F., “Cutting speed influence on surface integrity of hard turned bainite steel”, J Mater Process Technol. Vol. 128, pp. 318 – 323, (2002).
17. Bonhin E. P., David-Müzel S., De Souza J. V. C., Alves M. C. D. S., Ribeiro M. V., “Effect of machining parameters on turning of VAT32® superalloy with ceramic tool”, Mater. Manuf. Process, 34, pp. 1 - 7, (2019).
18. Azizi M.W., Belbah A., Yallese M. A., Mabrouki T., Rigal J. F., “Surface roughness and cutting forces modeling for optimization of machining condition in finish hard turning of AISI 52100 steel”, J Mech Sci Technol., 25(12), pp. 4105 - 4114, (2012).
19. Elsadek A. A, Gaafer A. M. and Mohamed S. S., “Optimization of Roundness Error in Hard Turning of AISI H13 Tool Steel”. EGTRIP, Vol. 17, pp. 53 - 61, (2020).