Skip to content
BY 4.0 license Open Access Published by De Gruyter Open Access October 7, 2022

Determination of flow distance of the fluid metal due to fluidity in ductile iron casting by artificial neural networks approach

  • Çağatay Teke EMAIL logo
From the journal Open Chemistry

Abstract

Ductile irons (DIs) have properties such as high strength, ductility, and toughness, as well as a low degree of melting, good fluidity, and good machining. Having these characteristics make them the most preferred among cast irons. The combination of excellent properties, especially in DI castings with a thin section, make it an alternative for steel casting and forging. But in the manufacture of thin-section parts, fluidity characteristics need to be improved and the liquid metal must fill the mold completely. The fluidity of liquid metal is influenced by many factors depending on the casting processes such as material and mold properties, casting temperature, inoculation, globalization, and grain refinement. In this study, an artificial neural network (ANN) model has been developed that allows for determining the flow distance of the liquid metal in the sand mold casting method under changing casting conditions of DI. Thus, the flow distance was estimated depending on the cross-sectional thickness during the sand casting under changing casting conditions. The experimental parameters were determined as casting temperature, liquid metal metallurgy quality, cross-sectional thickness, and filling time. Filling modeling was performed with FlowCast software. When the results were examined, it was seen that the developed ANN model has high success in predicting the flow distances of the liquid metal under different casting conditions. The calculated coefficient of determination (R 2) value of 0.986 confirms the high prediction performance of the model.

1 Introduction

Although there are many alternative materials, more than 90% of the metallic materials used today are iron alloys, which are divided into two groups steel and cast iron according to the carbon (C) content in the alloys [1]. Cast irons are iron-based alloys with a carbon content high enough to exceed their solubility in iron [2]. In comparison with steel, cast irons are known to be economical materials with relatively low melting temperatures, good fluidity, and castability [3]. The material obtained as a result of inoculation and adding small amounts of spheroidizing additives such as Mg and Ce to the molten iron before the casting process is called ductile iron (DI). It consists of sphere-shaped graphite dispersed in a matrix resembling steel. This situation has brought a different dimension to the engineering applications of cast irons [3]. The most commonly used spheroidizing element is Fe–Si–Mg alloys, which are used in an alloyed form with Fe and Si [4]. DI has a very wide range of applications in the automotive industry, such as engines, suspension components, wind turbines, wheel, bearing, gear manufacturing, pistons, and machine tool bearings [57]. It has properties of high strength, low melting point, ductility, toughness, and good machinability. These features are the main reason why it is the most preferred among cast irons [8,9]. DI casting, especially due to its high strength-to-density ratio, can be lighter, have better mechanical properties, and be more economical compared to aluminum alloys in the production of thin-section materials [10,11]. In addition, it is a very suitable alloy group as an alternative material for steel casting and forging in thin-section applications [12,13].

The production processes of parts by casting involve two important steps, the first is the filling of the melt into a mold, and the second stage is the process of solidification and cooling [14]. In a study conducted on the mold filling process, it was noted that the liquid metal affects the heat transfer and solidification properties, which in turn affects the fluidity of the liquid metal [15]. In addition, it was explained that the fluidity of the liquid metal during mold filling is affected by the thermal properties of the liquid metal and mold, pouring conditions, reinforcing properties, and solidification mechanisms [16]. Fluidity, in casting terminology, is the distance at which a metal will move through the mold without solidifying when casting at a certain temperature, in other words, the molten metal completely fills the inside of the mold cavity. Metals that are not sufficiently fluid can cause insufficient casting, especially in thinner sections of the casting mold [17,18]. Therefore, it is an important property for obtaining sound castings with thin sections. In addition, it is influenced by many factors, such as viscosity, oxide film, chemical composition, melting point, latent heat, melting surface tension, solidification mode, super heating, the mold surface heat transfer coefficient, specific gravity, mold conductivity, and mold temperature [19,20]. The castability of the metal is a parameter that is determined as the distance for the metal flow in the channel of the sand mold before the flow stops with the progressive solidification process [21]. Fluidity in sand molds depends not only on chemical composition but also on casting temperature, flow rate, section thickness, and metallurgical factors.

In a study on the investigation of DI fluidity, it was determined that the difference in mold material causes different flow distances of the liquid metal at different section thicknesses [22]. In a similar study, casting temperature was found to be important in the casting of thin section parts. It was also observed that the temperature drop and the increased cooling rate affected the mold filling [23]. In another study, it was reported that the flow distance and the range of solidification temperature were inversely proportional [24]. In a study of DI casting with different section thicknesses, it was observed that the flow distance of Fe–C–2Si cast iron was higher than that of Fe–C–2Al cast iron, regardless of the section thickness [25]. In a study investigating the effect of alloy addition on fluidity, casting experiments were carried out at different temperatures by adding various amounts of Cr and Ni to AISI 1040 steel. In the related study, it was determined that the most important factor in fluidity was temperature, and the addition of Cr and Ni increased the fluidity of the steel [26]. In a study examining the effect of Ni and Si contents on the fluidity of Al–Ni–Si alloys, it is understood that the fluidity of Al–Ni–Si alloys can be increased when the Si content is less than 3% by weight and the Ni content varies between 2 and 6% by weight [27]. There are various studies in the literature on the fluidity and flow distance of the liquid metal of DI and different alloys. However, there is no artificial neural network (ANN) model that evaluates the parameters of metallurgical quality, cross-sectional thickness, casting temperature, and filling time in DI together. For this reason, an ANN model has been developed in the study that estimates the flow distance of the liquid metal by considering the four related parameters.

ANNs are an artificial intelligence technology developed and inspired by the working mechanism of nerve cells in the human brain. The main purposes of use can be expressed as classification, clustering, curve fitting, forecasting, image processing, and the ability to create solutions to nonlinear problems. In addition, it has many advantages, such as the ability to work with incomplete information, have fault tolerance, process unclear information, and has distributed memory. When the literature is examined, it is seen that ANNs are used in many different fields for purposes such as prediction, diagnosis, classification, clustering, and error detection. Studies related to the field of production show that ANN is used to predict experimental results, analyze the effects of process parameters, and predict mechanical properties in manufacturing, such as casting and welding processes [2832]. When the studies conducted in the field of production planning are examined, it is seen that the ANN is used to solve the production redistribution problem, the batch sizing problem, and the labor scheduling problem [3335]. In addition, there are many studies in which ANN is also used in the field of finance and medicine [3642]. On the other hand, there have been no studies in which fluidity has been examined with ANNs in the process of DI casting into sand molds. With this study, it will be possible to use ANN in new application areas by adding research in a specific casting process to the research studies in the ANN literature. In addition, the examination of fluidity using ANN will make a significant contribution to the casting process literature.

2 Materials and methods

2.1 Filling modeling

In terms of study, filling modeling was performed in the sand mold casting method under changing casting conditions of DI material. Thus, the feed distance of the liquid metal was determined depending on the cross-sectional thickness during the castings made into the sand mold under changing casting conditions. Experiment parameters have been identified as casting temperature range between 1,350 and 1,500°C, the metallurgical quality of liquid metal has a value range of 10–90%, cross-sectional thickness between 1 and 5 mm, and filling time between 3 and 9 s (Table 1).

Table 1

Experimental parameters and levels

Level no. Cross-sectional thickness (mm) Casting temperature (°C) Metallurgical quality (%) Filling time (s)
1 1 1,350 10 3
2 3 1,400 50 6
3 5 1,450 90 9
4 1,500

In determining the experimental parameters and in the sand casting method, the parameters have been selected related to the fluidity properties of the alloy which have the most impact on the manufacturing process. Model geometry, a design adapted to the fluidity test model with a width of 20 mm and a length of 500 mm was carried out. In cases where the length of the liquid metal channel has changed, it has been deliberately kept long, so the liquid metal cannot fully progress, thus it is aimed to measure the flow distance of the liquid metal. In addition, it was aimed to determine which thickness castings can be made with the model criterion after the relevant simulation studies by selecting the cross-sectional thickness of 1–5 mm. Figure 1 shows the test bar measurements and the solid model image used in fluidity modeling.

Figure 1 
                  (a) Fluidity test model measurement and (b) solid model image used in fluidity modeling.
Figure 1

(a) Fluidity test model measurement and (b) solid model image used in fluidity modeling.

Modeling of casting processes is a necessary mathematical method that the computer can quickly and accurately predict what is happening in the mold in the duration of filling the mold and after filling it. These programs usually calculate using finite difference or finite element techniques. They have the ability to model the given casting geometry with the thermo-physical properties and boundary conditions of the materials that can also be entered by the users and contained in their own databases for different casting and mold materials. The casting geometry of the model was first created as a solid model in the SolidWorks program. Then, it was converted to STL format and transferred to the casting simulation program. In SolidCast casting simulation software, the type and thermo-physical properties of the casting alloy and mold material are defined in Table 2 according to the specified values.

Table 2

Thermo-physical properties of the casting material and the mold

Material Thermal conductivity (W/m K) Specific heat (J/kg K) Freezing range (°C) Density (kg/m3) Latent heat of fusion (J/kg)
Casting material Perlitic DI 25.944 460.24 44.63 7176.06 230115.6
Mold Silica sand 0.59 1075.288 1521.71

The material properties of the solid model geometry are transferred to the program by granulating. It is ensured that the specified boundary conditions are solved in the simulation program for each element. The filling modeling studies were carried out with the FlowCast program running depending on the SolidCast casting simulation software. According to FlowCast fluid dynamics criteria, it also calculates factors such as turbulence, incomplete filling, cold joining, and pressure when filling liquid metal into the mold cavity. Figure 2 shows the sample images obtained as a result of FlowCast casting filling modeling.

Figure 2 
                  Sample image taken from FlowCast filling modeling software.
Figure 2

Sample image taken from FlowCast filling modeling software.

The approach in a similar study was used to determine the flow distances of the liquid metal from the casting modeling results [43]. In this context, side images were uploaded to the program and the channel length was defined as 500 mm. Subsequently, flow distances were determined.

A total of 108 experiments were carried out depending on the experimental parameters and levels. An example section of the experiment results is given in Table 3.

Table 3

Variation of flow distances of the liquid metal depending on the experimental parameters

Experiment no. Cross-sectional thickness (mm) Casting temperature (°C) Metallurgical quality (%) Filling time (s) Experiment results
1 5 1,400 10 6 200.54
2 5 1,500 90 3 448.1
3 1 1,450 10 3 148.4
4 3 1,350 90 9 87.92
24 1 1,450 90 6 121.95
25 5 1,400 50 6 224.05
50 3 1,450 90 6 161.66
51 1 1,450 50 9 87.92
76 3 1,450 90 9 119.12
77 5 1,450 90 3 371.53
106 1 1,500 90 6 131.59
107 5 1,450 10 6 235.39
108 1 1,500 10 6 129.34

2.2 Development of the ANN model

At this stage, an ANN model has been developed to predict flow distances of the liquid metal in sand mold casting processes of DI with high accuracy.

2.2.1 Determination of input and output variables

The parameters of metallurgical quality, cross-sectional thickness, casting temperature, and filling time were determined as input variables. The output variable is the flow distance of the liquid metal. The general structure of the network is shown in Figure 3.

Figure 3 
                     The topology of the developed ANN model.
Figure 3

The topology of the developed ANN model.

2.2.2 Determining the type of network

Although there are many types of ANNs, it can be said that multilayer perceptron, LVQ network, ART networks, SOM networks, and Elman network are widely used. Multilayer perceptrons are also known as feed-forward back propagation network structures. This network structure has a fairly wide range of uses due to its ability to generate solutions to nonlinear problems and make generalizations [44]. For this reason, multilayer perceptrons have been preferred as the network type.

2.2.3 Determination of the training and test set

While developing ANN models, a data set related to the problem area is needed. This data set is divided into training and test set. A large number of samples in the training and test set will allow training and testing of the network with different samples, thus this will have a positive effect on the performance of the network. As for the problem area, examples may contain a number of non-numeric data. In this case, these data must be digitized. So ANNs work with numeric data [44]. Another aspect in determining the training and test set concerns the size of the training and test set. Usually, 70–80% of the total data is used in training and 20–30% is used in the testing process. This issue was also taken into account while determining the training and test set, and 86 of the total 108 experimental data were used to train the network, also 22 were used to measure the performance of the network. The data in the training and test set are subjected to normalization before being delivered to the network. The applied normalization formula is as follows:

(1) X = ( X i X min ) ( X max X min ) .

The normalized version of the data in the training and test set is given in Tables 4 and 5.

Table 4

An example section of the normalized training set

Data no. Cross-sectional thickness (mm) Casting temperature (°C) Metallurgical quality (%) Filling time (s) Experiment results
1 1 1 1 0 1
2 0.5 0 1 1 0.047973991
3 1 0.333333333 1 1 0.407792139
30 1 0 0 0.5 0.257870113
31 0 0.333333333 0 0.5 0.097454603
53 0 1 1 1 0.071207676
54 1 0 0 1 0.13792192
84 0.5 1 0.5 1 0.1701689
85 0 1 1 0.5 0.163402321
86 1 0.666666667 0 0.5 0.437765972
Table 5

An example section of the normalized test set

Data no. Cross-sectional thickness (mm) Casting temperature (°C) Metallurgical quality (%) Filling time (s) Experiment results
1 1 0.333333333 0 0.5 0.465494792
2 0 0.666666667 0 0 0.266276042
3 0.5 0 1 0 0.348889803
10 0.5 0.333333333 1 0.5 0.203125
11 0 1 0 0.5 0.163274397
20 1 1 1 1 0.620990954
21 1 1 0.5 0.5 0.659847862
22 0.5 0.666666667 0.5 0.5 0.261410362

2.2.4 Selection of the training algorithm and the transfer function

Although there are many different training algorithms used in the ANN, the Levenberg–Marquardt training algorithm was used because it is faster and more reliable than other training algorithms [45]. As a transfer function, the Log-Sigmoid function formula is used as given below:

(2) a = logsig ( n ) = 1 1 + e n .

2.2.5 Determination of the number of neurons in the hidden layer

Regarding the ANN, there is no clear approach that expresses what kind of network topology should be created in which situations. In multilayer perceptrons, the number of neurons in the input and output layers can be determined according to the type of problem, but the number of neurons in the hidden layer cannot be determined clearly. For this reason, the performances of multilayer perceptron models with different neuron numbers in the hidden layer have been studied. As a result of these studies, the number of neurons in the hidden layer of the model with the least error value is determined as the most appropriate neuron number of the hidden layer [44].

2.2.6 Measuring the performance of the network

The performance of a developed ANN model is measured by the correct estimation rate of the samples in the test set. In some cases, the network can predict the data in the training set with a very high accuracy rate, but this prediction rate may remain very low in the test set. In this case, it is concluded that the network memorizes instead of learning. In order to avoid such situations and to achieve high prediction accuracy, the network is tested with a test set consisting of samples that are not included in the training set. The high performance of the network depends on the minimum difference between the estimated values produced for the samples in the test set and the output values in the test set. At this stage, the performance measurement of the ANN models will be performed with Mean Absolute Percentage Error (MAPE) and R 2:

(3) MAPE = 100 % n t = 1 n | At Ft At | .

3 Results and discussion

The Matlab software was used in the development of ANN models. In order to determine the ANN model with the best prediction performance, ANN models with different hidden layer neuron numbers were created. Each ANN model was subjected to a training and testing process. The prediction performances of these models are shown in Figure 4.

Figure 4 
               MAPE values of the developed ANN models.
Figure 4

MAPE values of the developed ANN models.

As Figure 4 is examined, it is seen that the prediction performance of these models changes as the number of neurons in the hidden layer of ANN models changes. Prediction performances vary approximately between 81 and 95%, and the highest prediction performance was achieved when the number of neurons in the hidden layer was six. Thus, the ANN model that gives the best prediction performance was determined, and the prediction performance of this model was determined as 95.12%. Then, regression analysis was performed for this model. The result of the analysis is included in Figure 5.

Figure 5 
               Regression analysis of the ANN model with six neurons in its hidden layer.
Figure 5

Regression analysis of the ANN model with six neurons in its hidden layer.

As a result of the analysis, the regression equation and the coefficient of determination (R 2) were obtained. The coefficient of determination calculated expresses the harmony of the actual flow distance of the liquid metal and the estimated flow distance of the liquid metal produced by the ANN model. The closer the obtained value is to one, the better the fit. Here, the R 2 value is calculated as 0.986. This obtained value emphasizes that the developed ANN model makes quite successful predictions as it was shown in some other previous works [4654].

4 Conclusion

In this study, it is aimed to estimate the flow distance of the liquid metal in the sand casting process of DI depending on the parameters of cross-sectional thickness, casting temperature, metallurgical quality, and filling time. In this context, ANN models with different numbers of neurons in the hidden layer have been developed. The data obtained from 108 experiments were used in the training and testing processes of these models. Then, the prediction performances of the models were examined in terms of the MAPE error measure and it was determined that the ANN model with six neurons in its hidden layer had the best prediction performance. In addition, regression analysis was performed for this model. Thus, the regression equation and R 2 were obtained. The R 2 value calculated for this ANN model also confirms the high prediction consistency of the relevant model (R 2 = 0.986).

  1. Funding information: This research received no external funding.

  2. Conflict of interest: The author declares no conflict of interest.

  3. Data availability statement: Not applicable.

  4. Ethical approval: The conducted research is not related to either human or animal use.

References

[1] Ashby MF, Jones DRH. Engineering materials 2: an introduction to microstructures and processing. 4th edn. London: Butterworth Heinemann; 2013. p. 3–12.Search in Google Scholar

[2] Han CF, Sun YF, Wu Y, Ma YH. Effects of vanadium and austempering temperature on microstructure and properties of CADI. Metallogr Microstruct Anal. 2015;4:135–45. 10.1007/s13632-015-0197-1.Search in Google Scholar

[3] Campbell J. Casting. 2nd edn. London: Butterworth Heinemann; 2003. p. 70–231.Search in Google Scholar

[4] Vicente ADA, Sartori Moreno JR, Santos TFDA, Espinosa DCR, Tenório JAS. Nucleation and growth of graphite particles in ductile cast iron. J Alloy Compd. 2019;775:1230–4. 10.1016/j.jallcom.2018.10.136.Search in Google Scholar

[5] Di Cocco V, Iacoviello F. Ductile cast irons: microstructure influence on the damaging micromechanisms in overloaded fatigue cracks. Eng Fail Anal. 2017;82:340–9. 10.1016/j.engfailanal.2017.06.039.Search in Google Scholar

[6] Yan H, Wang A, Xiong Z, Xu K, Huang Z. Microstructure and wear resistance of composite layers on a ductile iron with multicarbide by laser surface alloying. Appl Surf Sci. 2010;256(23):7001–9. 10.1016/j.apsusc.2010.05.015.Search in Google Scholar

[7] Kamińska J, Angrecki M, Stefański Z, Palma A. Effect of wall thickness on the microstructure of ductile iron castings manufactured by the inmold process using a reaction chamber. Arch Foundry Eng. 2018;18(4):50–4. 10.24425/123632.Search in Google Scholar

[8] Theuwissen K, Lacaze J, Laffont-Dantras L. Structure of graphite precipitates in cast iron. Carbon. 2016;96:1120–8. 10.1016/j.carbon.2015.10.066.Search in Google Scholar

[9] Gouveia RM, Silva FJG, Paiva OC, Andrade MF, Silva L, Moselli PC, et al. Study of the heat-treatments effect on high strength ductile cast iron welded joints. Metals. 2017;7:382. 10.3390/met7090382.Search in Google Scholar

[10] Fraś E, Górny M, Stachurski W. Problem of super-thin wall nodular cast iron castings. Found Rev. 2006;56:230.Search in Google Scholar

[11] Torrance JW, Stefanescu DM. An investigation on the effect of surface roughness on the static mechanical properties of thin-wall ductile iron castings. AFS Trans. 2004;112:757–72.Search in Google Scholar

[12] Dix LP, Ruxanda R, Torrance J, Fukumoto M, Stefanescu DM. Static mechanical properties of ferritic and pearlitic lightweight ductile iron castings. AFS Trans. 2003;111:1149–64.Search in Google Scholar

[13] Ruxanda R, Stefanescu DM, Piwonka TS. Microstructure characterization of ductile thin wall iron castings. AFS Trans. 2002;110:1131–47.Search in Google Scholar

[14] Shepel SV, Paolucci S. Numerical simulation of filling and solidification of permanent mold castings. Appl Therm Eng. 2002;22(2):229–48. 10.1016/S1359-4311(01)00068-0.Search in Google Scholar

[15] Pathak N, Kumar A, Yadav A, Dutta P. Effects of mould filling on evolution of the solid–liquid interface during solidification. Appl Therm Eng. 2009;29(17–18):3669–78. 10.1016/j.applthermaleng.2009.06.026.Search in Google Scholar

[16] Ravi KR, Pillai RM, Amaranathan KR, Pai BC, Chakraborty M. Fluidity of aluminum alloys and composites: a review. J Alloy Compd. 2008;456(1–2):201–10. 10.1016/j.jallcom.2007.02.038.Search in Google Scholar

[17] Chelladurai C, Mohan NS, Hariharashayee D, Manikandan S, Sivaperumal P. Analyzing the casting defects in small scale casting industry. Mater Today. 2021;37(2):386–94. 10.1016/j.matpr.2020.05.382.Search in Google Scholar

[18] Vignesh R, Sanjay Gandhi M, Vignesh A, Rajarajan P. Effect of squeeze cast process parameters on fluidity of aluminium LM6 alloy. Int J Adv Technol. 2016;7(2):157. 10.4172/0976-4860.1000157.Search in Google Scholar

[19] Sabatino MD, Arnberg L. A review on the fluidity of Al based alloys. Metall Sci Technol. 2013;22:9–15.Search in Google Scholar

[20] Saxena S, Sharma PK. Casting fluidity of metals and alloys. Int J Innov Res Sci Eng Technol. 2017;6(2):3018–31. 10.15680/IJIRSET.2017.0602171.Search in Google Scholar

[21] Borowiecki B. Conventional flow curves of liquid cast iron put on spheroidization. Arch Foundry Eng. 2008;8(1):23–6.Search in Google Scholar

[22] Górny M. Structure of ductile iron in thin walled castings. Arch Foundry Eng. 2007;7(4):73–8.Search in Google Scholar

[23] Górny M. Fluidity and temperature profile of ductile iron in thin sections. J Iron Steel Res Int. 2012;19(8):52–9. 10.1016/S1006-706X(12)60139-3.Search in Google Scholar

[24] Han Q, Xu H. Fluidity of alloys under high pressure die casting conditions. Scr Mater. 2005;53(1):7–10. 10.1016/j.scriptamat.2005.03.025.Search in Google Scholar

[25] Haque MM. Investigation on properties and microstructures of spheroidal graphite Fe–C–2Si and Fe–C–2Al cast irons. J Mater Process Technol. 2007;191(1–3):360–3. 10.1016/j.jmatprotec.2007.03.030.Search in Google Scholar

[26] Aslandoğan R. Dökümde akıcılık ve akıcılığı etkileyen faktörlerin araştırılması [dissertation]. Yıldız Technical University; 2009.Search in Google Scholar

[27] Yang L, Li W, Du J, Wang K, Tang P. Effect of Si and Ni contents on the fluidity of Al-Ni-Si alloys evaluated by using thermal analysis. Thermochim Acta. 2016;645:7–15. 10.1016/j.tca.2016.10.013.Search in Google Scholar

[28] Hu C, Wang Y. An efficient convolutional neural network model based on object-level attention mechanism for casting defect detection on radiography ımages. IEEE Trans Ind Electron. 2020;67(12):10922–30. 10.1109/TIE.2019.2962437.Search in Google Scholar

[29] Ding S, Shi Q, Chen G. Flow stress of 6061 aluminum alloy at typical temperatures during friction stir welding based on hot compression tests. Metals. 2021;11(5):804. 10.3390/met11050804.Search in Google Scholar

[30] Li YR, Zhang CN. A neural network prediction analysis of breakout continuous casting based on differential evolution (de). Metalurgija. 2020;59(3):291–4.Search in Google Scholar

[31] Soundararajan R, Ramesh A, Sivasankaran S, Vignesh M. Modeling and analysis of mechanical properties of aluminium alloy (A413) reinforced with boron carbide (B4C) processed through squeeze casting process using artificial neural network model and statistical technique. Mater Today. 2017;4(2):2008–30. 10.1016/j.matpr.2017.02.047.Search in Google Scholar

[32] Moon IY, Jeong HW, Lee HW, Kim SJ, Oh YS, Jung J, et al. Predicting high temperature flow stress of nickel alloy A230 based on an artificial neural network. Metals. 2022;12(2):223. 10.3390/met12020223.Search in Google Scholar

[33] Pham QT, Phan TKD. Apply neural network for improving production planning at Samarang petrol mine. Int J Intell Comput Cyb. 2016;9(2):126–43. 10.1108/IJICC-09-2015-0032.Search in Google Scholar

[34] Şenyiğit E, Atici U. Artificial neural network models for lot-sizing problem: a case study. Neural Comput Appl. 2013;22:1039–47. 10.1007/s00521-012-0863-z.Search in Google Scholar

[35] Simeunovic N, Kamenko I, Bugarski V, Jovanovic M, Lalic B. Improving workforce scheduling using artificial neural networks model. Adv Prod Eng Manag. 2017;12(4):337–52. 10.14743/apem2017.4.262.Search in Google Scholar

[36] Kasgari AA, Divsalar M, Javid MR, Ebrahimian SJ. Prediction of bankruptcy Iranian corporations through artificial neural network and probit-based analyses. Neural Comput Appl. 2013;23:927–36. 10.1007/s00521-012-1017-z.Search in Google Scholar

[37] Ko PC, Lin PC. Resource allocation neural network in portfolio selection. Expert Syst Appl. 2008;35(1–2):330–7. 10.1016/j.eswa.2007.07.031.Search in Google Scholar

[38] Haider A, Hanif MN. Inflation forecasting in Pakistan using artificial neural networks. Pak Econ Soc Rev. 2009;47(1):123–38.Search in Google Scholar

[39] Etebari F, Najafi AA. Intelligent choice-based network revenue management. Sci Iran Trans E. 2016;23(2):747–56. 10.24200/SCI.2016.3860.Search in Google Scholar

[40] Tsai CF, Wu JW. Using neural network ensembles for bankruptcy prediction and credit scoring. Expert Syst Appl. 2008;34(4):2639–49. 10.1016/j.eswa.2007.05.019.Search in Google Scholar

[41] Wang N, Chen J, Xiao H, Wu L, Jiang H, Zhou Y. Application of artificial neural network model in diagnosis of Alzheimer’s disease. BMC Neurol. 2019;19:154. 10.1186/s12883-019-1377-4.Search in Google Scholar PubMed PubMed Central

[42] Chang RI, Chiu YH, Lin JW. Two‑stage classification of tuberculosis culture diagnosis using convolutional neural network with transfer learning. J Supercomput. 2020;76:8641–56. 10.1007/s11227-020-03152-x.Search in Google Scholar

[43] Şensoy AT, Çolak M, Kaymaz I, Dispinar D. Investigating the optimum model parameters for casting process of A356 alloy: a cross-validation using response surface method and particle swarm optimization. Arab J Sci Eng. 2020;45:9759–68. 10.1007/s13369-020-04922-8.Search in Google Scholar

[44] Öztemel E. Yapay sinir ağları. 3rd edn. Istanbul: Papatya Publishing; 2012.Search in Google Scholar

[45] Okkan U. Application of Levenberg-Marquardt optimization algorithm based multilayer neural networks for hydrological time series modeling. An Int J Optim Control Theor Appl. 2011;1(1):53–63. 10.11121/ijocta.01.2011.0038.Search in Google Scholar

[46] Malidarre RB, Akkurt I, Malidarreh PB, Arslankaya S. Investigation and ANN-based prediction of the radiation shielding, structural and mechanical properties of the hydroxyapatite (HAP) bio-composite as artificial bone. Radiat Phys Chem. 2022;197:110208. 10.1016/j.radphyschem.2022.110208.Search in Google Scholar

[47] Akkurt I, Malidarreh PB, Malidarre RB. Simulation and prediction the attenuation behavior of the KNN-LMN based lead free ceramics by fluka code and artificial neural network (ANN)-based algorithm. Environ Technol. 2021;1–15. 10.1080/09593330.2021.2008017.Search in Google Scholar PubMed

[48] Basyigit C, Akkurt I, Kilincarslan S, Beycioglu A. Prediction of compressive strength of heavyweight concrete by ANN and FL models. Neural Comput Appl. 2010;19(4):507–13. 10.1007/s00521-009-0292-9.Search in Google Scholar

[49] Arslankaya S. Estimation of hanging and removal times in eloxal with artificial neural networks. Emerg Mater Res. 2020;9(2):366–74. 10.1680/jemmr.19.00191.Search in Google Scholar

[50] Arslankaya S. Estimating the effects of heat treatment on aluminum alloy with artificial neural networks. Emerg Mater Res. 2020;9(2):540–9. 10.1680/jemmr.20.00059.Search in Google Scholar

[51] Polat TK. Forecasting of production and scrap amounts using artificial neural networks. Emerg Mater Res. 2022;11(3):1–11. 10.1680/jemmr.22.00036.Search in Google Scholar

[52] Nar M, Arslankaya S. Passenger demand forecasting for railway systems. Open Chem. 2022;20(1):105–19. 10.1515/chem-2022-0124.Search in Google Scholar

[53] Malidarre RB, Arslankaya S, Nar M, Kirelli Y, Erdamar IYD, Karpuz N, et al. Deep learning prediction of gamma-ray-attenuation behavior of KNN–LMN ceramics. Emerg Mater Res. 2022;11(2):276–82. 10.1680/jemmr.22.00012.Search in Google Scholar

[54] Teke Ç, Çolak M, Kiraz A, İpek M. Prediction of critical fraction of solid in low-pressure die casting of aluminum alloys using artificial neural network. Sci Iran Trans B. 2019;26(6):3304–12. 10.24200/sci.2019.50819.1881.Search in Google Scholar

Received: 2022-08-16
Revised: 2022-08-25
Accepted: 2022-08-29
Published Online: 2022-10-07

© 2022 Çağatay Teke, published by De Gruyter

This work is licensed under the Creative Commons Attribution 4.0 International License.

Downloaded on 5.2.2023 from https://www.degruyter.com/document/doi/10.1515/chem-2022-0210/html
Scroll Up Arrow