j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 1 9 9 ( 2 0 0 8 ) 437–439 journal homepage: www.elsevier.com/locate/jmatprotec Short technical note Artificial Neural Network approach to predict mechanical properties of hot rolled, nonresulfurized, AISI 10xx series carbon steel bars Mehmet Sirac Ozerdem a , Sedat Kolukisa b,∗ a b Dicle University, Department of Electrical and Electronics Engineering, Diyarbakır 21680, Turkey Dicle University, Department of Mechanical Engineering, Diyarbakır 21680, Turkey a r t i c l e i n f o a b s t r a c t Article history: In this study, Artificial Neural Network approach to predict mechanical properties of, Received 6 April 2006 hot rolled, nonresulfurized, AISI 10xx series carbon steel bars were obtained using a Received in revised form back-propagation neural network that uses gradient descent learning algorithm. In Arti- 21 June 2007 ficial Neural Network training module, C%, Si%, Mn% contents were employed as input Accepted 21 June 2007 and tensile strength, yield strength, elongation, reduction in area, hardness were used as outputs. ANN system was trained using the prepared training set (also known as learning set). After training process, the test data were used to check system accuracy. Keywords: As a result the neural network was found successful for the prediction of mechani- Artificial Neural Network cal properties of, hot rolled, nonresulfurized, AISI 10xx series carbon steels under given Prediction of mechanical properties conditions. © 2007 Elsevier B.V. All rights reserved. 10xx series steel bars 1. Introduction The performance of steels depend on the properties associated with their microstructures, that is, on the arrangements, volume fractions, sizes, and morphologies of the various phases constituting a macroscopic section of steel with a given composition (alloying elements) in a given processed condition (George, 2007). Effects of alloying elements, alloying elements mechanical property relationships have always been studied for decades for each alloying element. Repeated and relevant distractive, nondistractive tests are applied to achieve the effect of alloying elements on the mechanical properties ∗ of alloys, since these properties are essential for design engineers as and where necessary. Recently, with the developments in artificial intelligence; researchers have a great deal of attention to the solution of non-linear problems in mechanical properties of alloys (Altinkok and Koker, 2004). Features of multi layer perceptron architecture with back-propagation learning algorithm (Haykin, 1994) were employed to predict the tensile strength, yield strength, elongation, reduction in area, hardness of hot rolled, nonresulfurized, AISI 10xx series carbon steel bars as a function of alloying elements other than Fe. Corresponding author. Tel.: +90 412 248 84 03; fax: +90 412 248 84 05. E-mail addresses: sozerdem@dicle.edu.tr (M.S. Ozerdem), kolukisa@dicle.edu.tr (S. Kolukisa). 0924-0136/$ – see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.jmatprotec.2007.06.071 438 j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 1 9 9 ( 2 0 0 8 ) 437–439 2. Artificial Neural Network (ANN) approach and modeling data with network An ANN is a mathematical model consisting of a number of highly interconnected processing elements organized into layers, the geometry and functionality of which have been likened to that of the human brain. The ANNs are parallel process elements which has characteristic such as: (a) ANN is a mathematical model of a biological neuron; (b) ANN has various process elements which are related to each other; (c) ANN keeps knowledge with connection weights. The network has one input layer, one hidden layer and one output layer. The input layer consists of all the input factors. Information from the input layer is then processed in the course of one hidden layer, following output vector is computed in the final (output) layer. A schematic description of the layers is given in Fig. 1. In developing an ANN model, the available data set is divided into two sets, one to be used for training of the network, and the remaining is to be used to verify the generalization capability of the network (Haykin, 1999). Input–output pairs are presented to the network and weights are adjusted to minimize the error between the network output and actual value. Among the various kinds of ANN approaches that exit, the multi layer perceptron (MLP) architecture with backpropagation learning algorithm has become the most popular in engineering applications (Okuyucu et al., 2007; Song et al., 2005). Back-propagation algorithm, which is common in literature, has been used to update the forward path parameters in ANNs. This method is based on minimization of the quadratic cost function by tuning the network parameters. The mean square error is considered as a measurement criterion for a training set. Fig. 1 – The structure of three-layered neural network in present study. 3. Experimental procedures 3.1. Collecting the experimental data The chemical compositions and the mechanical properties of hot rolled, nonresulfurized, AISI 10xx series carbon steel bars were collected from related standards and CASTI metals black book—North American Ferrous Data, steel manufacturers (CASTI). 3.2. Working platform MATLAB platform was used to train and test the ANN. In the training, increased number of neurons (8–12) in the hidden layer has been used in order to define the output accurately. After training the network successfully, it has been tested by using the known data. Statistical methods were used to compare the results produced by the network. Errors occurring at the learning and testing stages are called the root-mean squared (RMS), absolute fraction of variance (R2 ), and mean percentage error (MPE) values. 4. Evaluation of results and discussions The aim of employing an ANN model is to predict the mechanical properties of hot rolled, nonresulfurized, AISI 10xx series, carbon steel bars. The network has three input parameters: C, Mn and Si contents and five output parameters: tensile strength, yield strength, elongation, reduction in area and hardness. So, the architecture of ANN becomes 3-10-5, 3 corresponding to the input values, 10 to the number of hidden layer neurons and 5 to the outputs. The experimental data set includes 44 patterns, of which 33 patterns were used for training the network and 11 patterns were selected randomly to test the performance of the trained network. All the input and output values were normalized between 0.1 and 0.9 by using linear scaling. The log-sigmoid transfer function was used in the hidden and output layer. During the training period, the averaged square error decreased with increasing number of iteration. After 5000 training cycles, significant effect on error reduction has not been traced. The performance changing of ANN in training stage is given in Fig. 2. With the increasing number of reliable input data, improves the integrity of the training session and target outputs. Comparison of measured and predicted mechanical properties (tensile strength, yield strength, elongation, reduction in area and hardness) at training and testing stages indicated that there is a high correlation between them. The decision as to the number of neurons used in the hidden layer usually depends on the arithmetical mean of the number of inputs and outputs. In this application 8–12 hidden layers were employed to test. The algorithm with 10 hidden layer neurons is suggested to be used in present application. Statistical values of tensile strength, yield strength, elongation, reduction in area and hardness were presented in Table 1. As a result of statistical values namely, root-mean squared (RMS), absolute fraction of variance (R2 ) and mean percent- ID 791782 Title ArtificialNeuralNetworkapproachtopredictmechanicalpropertiesofhotrolled,nonresulfurized,AISI 10xxseriescarbonsteelbars http://fulltext.study/article/791782 http://FullText.Study Pages 3