PREDICTION OF MECHANICAL STRENGTH OF POLYPROPYLENE FIBRE REINFORCED CONCRETE USING ARTIFICIAL NEURAL NETWORK PREDVIĐANJE MEHANIČKE ČVRSTOĆE BETONA OJAČANOG POLIPROPILENSKIM VLAKNIMA KORIŠĆENJEM VEŠTAČKE NEURONSKE MREŽE

The Polypropylene fibre reinforced concrete (PFRC) contains randomly distributed concrete and it acts as internal reinforcement so as to enhance the properties of the cementitious composite (concrete). The principal reason for incorporating the Polypropylene fibres into a cement matrix is to reduce cracking in the elastic range, and to increase the tensile strength and deformation capacity and increase the flexural strength of the resultant concrete. These properties of PFRC primarily depend upon the length and volume of propylene fibres (PPF) used in the concrete mixture. The strength properties of the concrete reinforced with polypropylene fibre and alkali resistant glass fibres was predicted by Regression modelling and found near to the experimental results of all the specimens [1]. The predicted compressive strength of concrete form the different batching plants for fresh concrete and early strength using artificial neural network [2]. The Intelligent Prediction system of concrete strength was developed, to provide strength information for removal of form work and scheduling the construction [3]. The split tensile strength and percentage of water absorption of concrete containing TiO2 nanoparticles were predicted by using ANN and genetic programming and also concluded that ANN prediction is better than genetic programming [4]. The application of the Artificial neural network for predicting drying shrinkage of concrete was performed and found that prediction was in good agreement with the experimental strength [5].


INTRODUCTION
The Polypropylene fibre reinforced concrete (PFRC) contains randomly distributed concrete and it acts as internal reinforcement so as to enhance the properties of the cementitious composite (concrete). The principal reason for incorporating the Polypropylene fibres into a cement matrix is to reduce cracking in the elastic range, and to increase the tensile strength and deformation capacity and increase the flexural strength of the resultant concrete. These properties of PFRC primarily depend upon the length and volume of propylene fibres (PPF) used in the concrete mixture. The strength properties of the concrete reinforced with polypropylene fibre and alkali resistant glass fibres was predicted by Regression modelling and found near to the experimental results of all the specimens [1]. The predicted compressive strength of concrete form the different batching plants for fresh concrete and early strength using artificial neural network [2]. The Intelligent Prediction system of concrete strength was developed, to provide strength information for removal of form work and scheduling the construction [3]. The split tensile strength and percentage of water absorption of concrete containing TiO2 nanoparticles were predicted by using ANN and genetic programming and also concluded that ANN prediction is better than genetic programming [4]. The application of the Artificial neural network for predicting drying shrinkage of concrete was performed and found that prediction was in good agreement with the experimental strength [5]. The strength predictions of high performance steel fibre concrete with different volume of fibres were established using regression models and its second order regression model gives better predictions when compared to actual strength [6]. In the recent years an Artificial Neural Network (ANN) has been used to predict the behaviour of the mechanical properties of the concrete and various civil engineering structures. The compressive stress of the FRP confined and unconfined concrete column was predicted from the experimental and analytical results using ANN [7]. Previous studies [8][9][10] have developed the concrete compressive strength prediction model using ANN and Genetic Programming. This paper discusses the experimental study on the polypropylene fibre reinforced concrete to find the mechanical properties for different volume of fibre at 7 days and 28 days of curing. The prediction of strength using ANN approach was also discussed and compared.

EXPERIMENTAL STUDY
The ingredients of concrete were studied properly to identify its properties of the ingredients. Table 1 presents the properties of the polypropylene fibres. Concrete preparation was done by mixing constitute materials with different percentage of fibre thoroughly to avoid the segregation of fibres. Figure 1 shows the batching of ingredients of the polypropylene concrete. All the specimens casted with different proportion were cured under room temperature in the water tank. Figure 2 and 3 shows the cube, cylinder and prism specimens in order to find the compressive strength, split tensile strength and flexural strength of the specimen respectively. After 7 and 28 days of curing, the specimens were tested under compression testing machine of capacity 2000 kN and universal testing machine of capacity 600 kN to measure compressive strength, split tensile strength and flexural strength of the specimens respectively. Figure 4 shows the loading position of the specimens.   The increase in the percentage of fibres in the concrete increases the compressive strength, flexural strength and split tensile strength by 8 %, 20 % and 40 % respectively for change in the percentage of fibre from 0.5 % to 2.0 % by volume of cement. Figure 5 shows the comparison between the strength and % of fibre content for two curing period 7 days and 28 days.

ARTIFICIAL NEURAL NETWORK
The ANN prediction model is performed using MATLAB with one hidden layer of twenty-five hidden  (1) where , and denote the experimental value, predicted value and the number of data points, respectively.       Figure 10 shows convergence characteristics of the ANN model during the training, validation and testing phases respectively and the Table 3 gives the comparison of the test conducted on the predicted models. In addition, it shows the number of epochs (4, 4 & 6 iterations) and the best validation performance, which is 0.71201 at 2 iterations, 0.01369 at 4 iterations and 0.00404 at 5 iterations for compressive strength, flexural strength and split tensile strength respectively.

CONCLUSIONS
This study is aimed to find the best model to predict the compressive strength, flexural strength and split tensile strength at 7 days and 28 days curing using ANN. A total of 90 values were trained to predict the mechanical strength of polypropylene fibre Reinforced Concrete by 3 ANN models. All the outputs are compared using the coefficients of determination calculated for the highest R 2 values. The following conclusions are obtained from the study.
• From the predicted results of mechanical properties of the polypropylene fibre reinforced concrete, it is concluded that ANN models are more suitable and accurate for prediction of 28 days strength based on the R 2 values obtained from the ANN model.
• The Mean Squared Error (MSE) for all three models for predicting the compressive, split tensile and flexural strength are very low and nearly equal to zero. The correction coefficient (R) for all three models are greater than 0.85 and it is acceptable.
• ANN model with R 2 = 0.9844, MSE = 0.0136 and SSE = 0.03844 are found to be capable of predicting the 28 days split tensile strength.
• The increase in the number of independent variables leads to the increase in the accuracy of the ANN model.