Click here to load reader
Click here to load reader
Apr 10, 2015
Scientific Research and Essay Vol. 5 (1), pp. 081-092, 4 January, 2010 Available online at http://www.academicjournals.org/SRE ISSN 1992-2248 2010 Academic Journals
Full Length Research Paper
An experimental study of steel fibre reinforced concrete columns under axial load and modeling by ANNUlku Sultan Yilmaz1, Ismail Saritas2*, Mehmet Kamanli1 and Mevlut Yasar Kaltakci1Department Of Civil Engineering, Engineering And Architectural Faculty, Selcuk University, Campus, 42031, Konya, Turkey. 2 Department Of Electronic And Computer Education, Technical Educational Faculty, Selcuk University, Campus, 42031, Konya, Turkey.Accepted 7 December, 20091
Concrete is a construction (building) material of which its usage in different fields has become widely spread by growing due to some of its effectiveness such as being easily shaped, resistance against physical and chemical outer effects, economical and having convenience in production. As a result of being widespread, it has been understood that concrete will serve more effective than the expected classical quality of it if it is consolidated with new techniques and new materials against outer physical and chemical effects. Different techniques are being developed to meet the requirement of various effects which exist in places where they are used. One of these techniques is to use steel fibre that has high technical properties. In addition to this, fibres which are produced from different materials may also be used with the concrete. As the day passes, the usage fields of the concrete that is produced by consolidating with different amount of steel fibre are increasing. In this study, the behaviours of ferroconcretes with steel fibre and without steel fibre were investigated under the axial load as experimentally. At the experimental stage, axial force-unit shortening ratios were obtained by loading 4 items of prismatic column samples with 160 160 840 mm dimension as axially in the mechanism that has load control in it. ANN model was done by data obtained from experimental study. Backpropagation algorithm was used in this study. ANN was designed as one input, one hidden layer and two output layers. 75 of 112 obtained data were used as training data whereas the rest was used as test data. Data was normalized and modelled by Matlab NNToolbox and obtained data were compared with experimental results by SPSS statistical programme. When the comparison was made between the results of the experiments, it was determined that there was no significant increase in the carrying power of the elements. The same results were obtained by ANN model. Since p > 0.05 as the result of the statistical analysis done in the 95% confidence interval between data obtained from experiments and ANN model, the reliability of the ANN model was proven. Key words: Artificial Neural Network, steel wire/fibre, ferroconcrete column, axial load effect, column with fibre. INTRODUCTION Concrete with fibre that is produced by substituting different ratios and certain properties of steel fibre into normal concrete is increasing performance of traditional concrete by compensating the most of the drawbacks of it. The most important positive subject for behaviour of the ferroconcrete may be improvement of crispy property concrete that forms the ferroconcrete. Various researches (Sukontasukkul et al., 2005; Ayers and Van, 2003) showed that steel fibre increases ductile; first split resistance, pull resistance, bending carry power resistance, fatigue resistance, cutting resistance and elasticity module of normal concrete in significant amount. Today, researches (Ramesh et al., 2003; Sheikh, 1982) about this topic concentrate on the effect of using steel fibre to the behaviour regarding to detrital and split development. Especially, the limitation effect of fibre on the splits of axial loaded elements creates a wound effect for an element
*Corresponding author. E-mail: [email protected] Tel: +90 332 2233354. Fax: +90 332 2412179.
Sci. Res. Essays
Artificial Neural Network Artificial Neural Network is a kind of information processing technology which is constructed as the result of imitation of the thinking and working abilities of the human brain (Oztemel, 2003; Cogurcu et al., 2008). What is intended for artificial neural network is a model of biological neural network. So, an artificial system will be brought about which imitates the functionality of the biological neural network. Three components were included in artificial neural network structure as neuron, connections and weights. In Figure 1, structure and components of the ANN was shown. Artificial neural networks utilises data and results related with real life problem area or samples during learning process. Variables regarding to real life problem area constitute input sequence of artificial neural network whereas results regarding to real life obtained from these variables constitute target outputs sequence that artificial neural network must reach. The pattern that is required to be learned by ANN determines the relation between input and output in this training set and the weights of the ANN project in this pattern. In order to train ANN, lots of numbers of input and related output sequences are needed. The whole data that consists of the pairs of input and output sequence and used in training of ANN is known as training set (Cogurcu et al., 2008). The basic operation done in learning process of ANN is to change the values of weights (Cogurcu et al., 2008). The aim is to adjust the weights of the ANN to produce output sequence related with all input sequences correctly (Celik and Arcaklioglu, 2004). It is possible to think this as an arrangement of the coefficients of the input that comes to neuron. So, ANN becomes a presenter of real life pattern according to the utilised input and output. The mechanism which enables ANN to adjust the weights in network for producing required outputs is known as learning algorithm or learning rule (Figure 2) (Cogurcu et al., 2008). In a simple expression, an ANN learns by doing error. Three main steps exist in the learning process of ANN. These are (Rumelhart et al., 1986): a. Calculation of outputs; b. Comparison of these outputs with target outputs and calculation of error; c. Changing weights and repeating the process At the beginning of the learning process, the weights of the ANN are randomly assigned. Inputs are transferred to the hidden and output layer starting from input layer by being processed. So, ANN produces an output sequence under the effect of weights, total and transfer functions. The calculated difference between these outputs and target outputs is known as error. This error is used in network to compensate the difference between the weights weights of ANN and required outputs (Figure 3) (Caudill, 1987). There are lots of actively used learning algorithms. These learning algorithms may vary with the ANN archi-
Figure 1. Artificial Neural Network Model.
under the pressure. With this formed effect, ductile of element and henceforth ductile of system increases. This result is supported by research done by Shah and Rangan (1970). The studies about ferroconcrete columns with steel fibre (Craig et al., 1984) generally shows that cutting and moment capacities, pulling resistance, ductile and bonds of the elements increase and henceforth improve split control. Ferroconcrete columns are the most important carrier elements of frame systems that is made up of column and tendinous in the ferroconcrete buildings. Columns have important roles in earthquake and wind load apart from axial load carrying. If ferroconcrete columns present the ductile behaviour it will be very important for absorbing and consuming the energy that appeared at the time of the effect of the earthquake. In the present study, behaviour and deformation differences formed on ferroconcrete columns produced by having various steel fibre ratios according to TS 500 were investigated. For different fibre percentages, normal force-deformation graphics obtained from experiments were drawn including comparisons. Experimental values of tested columns were compared with the calculated values. So, the validity of ferroconcrete calculation basis was investigated for column with steel fibre stirrup. Axial load carrying capacity of manufactured concrete, axial load-deformation curves and time durations belong to these criterions and were recorded as the result of the experiments. By comparing the results of the experiments with theoretical values, appropriateness of the results were inquired with SPPS statistical packet program version 13 according to Variance analysis and T-test. It has been seen that reliability was found in 95% reliance interval.
Yilmaz et al.
Figure 2. Learning process and learning algorithm. Xij=Inpus, Yij=Outputs, Dij= Targets.
Figure 3. Calculation of error at the learning process.
tecture according to quantity of the problems. Hebb, Delta, Backwards Chaining (Generalized Delta), Kohonen, Hopfield and Energy function are mostly used learning algorithms among more than 100 types. Since backward chaining is the most commonly used in optimisation and evaluation problems, we used it as learning algorithm in the present study. ANN has been applied successfully in lots of areas starting from the guess of electrical charge and river flow, wind energy control, automotive sector to construction sector. Especially in construction sector, its usage in modelling and classification of the experimental studies increases day by day (Korres et al., 2002; Yuanwang et al., 2002; Jurado and Saenz, 2001; Ba bu , 1994; Elmandooh and Ghobarah, 2003; Hadi and Li, 2004). ElMandooh and Ghobarah (2003) investigated the applicability of the non-axial and nonlinear model of reinforced concrete column under periodical and dyn