PSI - Issue 17

Danial J. Armaghani et al. / Procedia Structural Integrity 17 (2019) 924–933 Danial J. Armaghaniet al. / Structural Integrity Procedia 00 (2019) 000 – 000

925

2

1. Introduction

The evaluation of shear capacity comprises one of the basic steps toward the design of structural members of reinforced concrete structures. In the past, numerous researchers as well as structural design codes have proposed various mechanistic models to evaluate the shear capacity of reinforced concrete beams. A typical approach to the problem includes the structural code or a research study which takes into account various (not common) parameters; however, these approaches lead to quite different results for the shear strength. The lack of adequate and reliable empirical or analytical relations for the evaluation of shear resistance of beams made of reinforced concrete has resulted in the last two decades to attract the interest of researchers dealing with non-deterministic techniques. Detailed and in-depth review and critical literature examination can be found in the works of Flood and Kartam (1994), Adeli (2001), Asteris and Plevris (2013, 2017) and Asteris et al. (2016a,b). Among non-deterministic methods, the method of Artificial Neural Networks (ANN) appears to be very attractive and reliable. Artificial Neural Networks have materialized as an innovative simulation technique, with wide spectrum of applications in variety levels of technological disciplines. Over the last two decades, there has been extensive use of ANNs in predicting the behavior or evaluate the mechanical properties of structural materials and especially of concrete (Waszczyszyn and Ziemiański 2001, Asteris and Kolovos 201 9, Asteris et al. 2017). The pertinent literature includes publications on the application of the ANNs to calculate the compressive strength and elasticity modulus of concrete (Dias and Pooliyadda 2001, Lee 2003, Topçu and Saridemir 2008, Trtnik et al. 2009). Furthermore, for the evaluation of compressive strength of concrete materials, various other methods of artificial intelligence, such as the fuzzy logic and genetic algorithms have been examined (Baykasoǧlu et al. 2004, Akkurt et al. 2004, Özcan et al. 2009). In addition, the use of soft computing techniques has been highlighted in many studies in field of civil engineering (Armaghani et al 2014, 2017, Abad et al 2018, Koopialipoor et al 2018, Pham et al 2017a,b,c&d, and Pham et al 2018a,b). Moreover, in the last decades, artificial neural networks have been proposed to estimate the shear strength of reinforced concrete structural elements (Sanad and Saka 2001, Mansour et al. 2004, Seleemah 2005 & 2012, El-Chabib et al. 2006, Amani and Moeini 2012, Kotsovou et al. 2017). This paper examines the adoption of Artificial Neural Networks for the estimation of shear strength of reinforced concrete. In particular, a heuristic algorithm is proposed to determine the optimal Artificial Neural Network architecture for estimating shear resistance of reinforced concrete members in terms of mean square error. For the training of the network, a research database is used, which is available in literature and has to do with the shear resistance of reinforced concrete beams specimens with various dimensions, material and geometric properties. The backward propagation method was examined in the procedure of neural network design and development, while the well-known Levenberg-Marquardt algorithm (Lourakis 2005) was used as the training algorithm. In the present study, a preliminary attempt is made to reveal the influence of mechanical and geometrical parameters of concrete beams on the value of their shear strength. Namely, using ANN techniques a set of different architectures based on different numbers of the input parameters have been investigated and the optimum one for each one case is presented and compared to the available in the literature proposals. Artificial Neural Networks (ANNs) are information-processing models that are configured to learn and perform several tasks such as classification, prediction, and decision-making. A trained ANN maps a given input onto a specific output, and therefore it is considered to be similar to a response surface method. The main advantage of a trained ANN over conventional numerical analysis procedures (e.g., regression analysis) is that the results are more reliable and can be produced with much less computational effort (Asteris et al. 2016a, Hornik et al. 1989, Plevris and Asteris 2014a&b, Plevris and Asteris 2015, Giovanis and Papadopoulos 2015). The concept of an artificial neural network is based on the concept of the biological neural network of the human brain (Fig. 1). The basic building block of the ANN is the artificial neuron, which is a mathematical model trying to mimic the behaviour of the biological neuron. Information is passed into the artificial neuron as input and processed with a mathematical function leading to an output that determines the behaviour of the neuron (similar to fire-or-not 2. Materials and Methods 2.1. Artificial Neural Networks