Adeli and Yeh  proposed a new model for engineering design, which employed machine learning technique (Block, 1962). Perception is the sensor that receives an input signal. The perceptron devoid of hidden layers was used to formulate the problem of the structural design. Vanluchene and Sun  applied the BP algorithm to the structural engineering problems such as the selection of beam cross section. Design decisions and showed the efficiency of neural network approaches .
Hajela and Berke  advocated the use of neural networks in the optimization of structures and observed its efficient performance. Hung and Adeli  based on perceptron theory developed a new model called as PERHID for machine learning. Yu and Adeli  and Hung and Adeli (1994a) utilized object-oriented programming for machine learning and engineering design. Some researchers have introduced an iteration process, which triggered itself in the original perception learning model to inhibit the error at the end of each iteration cycle . The authors observed an enhanced performance of new theory in the design of steel beams. Theocaris and Panagiotopoulos  applied the BP learning algorithm for parameter identification in fracture mechanics. Gunaratnam and Gero  observed better efficiency of neural networks in solving problems of structural engineering. Messner et al.  used neural networks for selection of most efficient cross sections for structural members for a given building data like space, site location, budget, and height of the structure.
Yeh  used neural networks for the identification of transverse cracking and spalling of concrete joints prestressed concrete piles. Kang and Yoon  used neural networks for efficient designing of roof trusses, while others used neural networks for the improvement of wavefront and profile characteristics. Rogers  used neural networks in conjunction with the approximate method of structural analysis for structural optimization. Mukherjee and Deshpande  used the BP algorithm for the preliminary design of the structures. Alsugaired and Sharar (1995) used neural networks to determine the characteristics of semirigid connections with single-angle plate beam configuration.
Turkkan and Sluivastava  conducted wind analysis using neural networks for membranes of different shapes. Mukherjee and Deshpande  determined the behavior of axially loaded columns subjected to buckling based using ANNs. Anderson et al.  conducted the experimental study on beam-column connections and determined load-carrying capacity of different types of connections. Noor  conducted nonlinear and sensitivity analysis using the BP algorithm. Mirmiran and Shahawy  used ANNs to accurately predict the behavior of concrete columns with fiber-reinforced composites. Yang et al.  determined mechanical properties of lightweight concrete using ANNs. Analysis results elucidated a close agreement between the experimental results and results obtained by ANNs. Konin et al.  employed ANNs to determine the penetration of chlorides due to presence of microcracking in the high-performance concrete. Hegazy and Ayed  identified selection characteristics, strain in reinforcement bars, distribution of stress in concrete, and crack patterns in concrete using ANNs. Chuang et al.  determined the load-carrying capacity in the ultimate range for reinforced concrete columns with pinned supports. Stavroulakis and Antes  identified steady-state cracks in the structural members. Cao et al.  modeled a cantilever beam subjected to the pah of concentrated loads as an analogy to model the aircraft wings. Mathew et al.  employed the ANN to assess the behavior of masonry panels under the influence of different types of bending. Kishi et al.  developed an ANN model to assess the alignment of structural members subjected to different end conditions. The authors observed the overestimation of the size of structural members, which yielded a conservative design. Hung and Jan  adopted ANNs for determination of the performance of braced and unbraced fl ame using AISC alignment charts. The global stiffness matrix represents a relation between the load and deflection, but they consume a lot of tune. However, the ANN can be effectively used for this purpose with more efficiency and lesser time consumption. Adeli and Kumar  adopted the BP algorithm for determining the approximate displacements at the end of each iteration. The analysis results showed that the combination of neural network and BP algorithm improved efficiency. Consolazio  presented different combinations of the neural networks in conjunction with iterative equation-solving techniques. William and Shoukry  used finite-element analysis in conjunction with neural networks for design and analysis of flat slab highway. The authors observed that adoption of neural networks leads to better convergence. This was later confirmed by Li . Oztas et al.  observed the better performance of the ANN in the prediction of 28 days compressive strength of concrete in comparison to traditional methods. Similar observations were reported by Kewalramani and Gupta , Nataraja et al. , Adhikary and Mutsuyoshi , Narendra et al. , El- et al. , and Ji et al. . Jeyasehar and Sumangala  observed that the ANN yielded close results in comparison to experimental results obtained from nondestructive testing of prestressed concrete. Saini et al.  observed the superior performance of the ANN in the analysis of doubly reinforced beams. Neocleous (2006) accurately determined the flexural capacity of concrete with steel fibers using the ANN. Altun et al.  adopted BP networks to forecast the characteristic mechanical properties of lightweight concrete. The authors manufactured lightweight concrete with a density ranging from 350 to 450 kg/m3. The steel fibers (Dramix type) were added in different dosages up to 60%. The input layers for the ANN were the quantity of steel fiber, w/c ratio, superplasticizer, etc. The BP ANN showed better performance as compared to multiple linear regression (MLR) analysis in the estimation of compressive strength of concrete. Topcu and Sandemir  adopted both ANN and fuzzy logic models for the prediction of mechanical properties associated with the concrete. The authors chose a large amount of experimental data from previous research works for ANN modeling. The ANN model developed adopted raw materials (used in the manufacture of concrete) as input parameters. The compressive strength was chosen as the output parameter. The analysis results showed that the amalgamated system exhibited better performance than the ANN. Demir  accurately determined the elastic modulus of concrete using the ANN model. Jung and Kim  accurately predicted the mechanical properties of concrete beams without shear reinforcement. Karthikeyan et al.  observed the better performance of ANN as compared to conventional methods in predicting durability characteristics such as creep and shrinkage of concrete.
Prasad et al.  determined the compressive strength of concrete with high workability employing ANN. The authors used a large number of experimental results to train the developed ANN model. The analysis results showed the superior performance of the ANN as compared to traditional methods. Naderpour et al.  observed the efficiency of the ANN in predicting the compressive strength of concrete with fiber-reinforced plastics as reinforcement. Bilgehan and Turgut  conducted ultrasonic pulse velocity test in determination of the strength of concrete. The authors used the experimental dataset in developing ANN. The authors observed a close corr elation between ANN and experimental results.
Furthermore, the ANN yielded better results as compared to classical methods. Sobharri et al.  confirmed the superior performance of the ANN over the ANFIS model in prediction of concrete strength with low workability. Ashrafi et al.  employed the ANN to determine the load- displacement behavior of concrete reinforced with composite fibers and observed its efficient performance. Erdern  formulated an ANN model to predict the moment-resisting capacity of flat slabs and found close agreement between experimental and ANN results. Basyigit et al.  obtained similar results for high-performance concrete using neural network and fuzzy logic models. Uysal and Tarryildizi  determined the strength parameters of concrete at 28 days using the ANN. The concrete was manufactured, incorporating different minerals to increase the strength of concrete. The manufactured concrete employed two combinations. The first combination included both limestone and fly ash (FA) in ratios ranging from 15% to 30%. The analysis results showed a good correlation between the results of the experimental study and the ANN. Atici  predicted the characteristic compressive strength of blast furnace slag concrete and observed superior performance of ANN as compared to MLR analysis. Hakim et al.  found the ANN to predict the mechanical properties of concrete with the superplasticizers accurately. Similar results were observed by Siddique et al.  for self-compacting concrete. Nazari and Riahi  predicted the split tensile strength and water permeability using ANN. The authors compared the genetic programming technique and the ANN for predicting mechanical and durability properties. The authors conducted the experimental study and collected a dataset of 144 specimens with a different period of curing. The authors adopted 16 different mix proportions with varying proportions of cement, coarse aggregate, fine aggregate, and water. The ANN was developed considering eight different input parameters such as type of aggregate, cement content, water content, nanoparticle content, type of superplasticizer, etc. The analysis results showed comparable performance between neural network and genetic algorithm models.
Uysal  used an ANN model for estimation of characteristic compressive strength of the ANN model with polypropylene (PP) fiber under high temperature. In their research study, the authors adopted mineral additives such as FA, granulated blast furnace slag, zeolite, basalt powder, and PP fiber of density 2 kg/m3. The analysis results showed a loss of compressive strength as the temperature exceeds 600 °C, but the presence of PP fiber prevented the risk of concrete spalling.
Sadrmomtazi et al.  predicted mechanical properties of concrete manufactured by use of lightweight expanded polystyrene beads. The authors used a combination of adaptive network-based fuzzy inference system (ANFIS) and ANN models for predicting the compressive strength. The analysis results showed that the ANN is useful in prediction as compared to the prediction capacity of the ANFIS model. In contradiction. Yuan et al. [ 101 ] observed close agr eement with the BP algorithm and the ANN in evaluating the characteristic compressive strength of high-performance concrete. Chandwani et al.  developed a hybrid method by fusion of genetic algorithm and ANN for prediction of mechanical properties of concrete. The authors adopted parameters such as coarse aggregates, FA, fine aggregate, water binder, cement, etc., as input parameters. The analysis results showed the poor performance of linear regression analysis as compared to the ANN. Chithra et al.  determined mechanical properties of concrete using ANN and MLR analysis. The authors used copper slag and nanosilica for partial replacement of cement and fine aggregate. The cement was replaced with nanosilica up to 3% at intervals of 0.5%, and fine aggregate was replaced with copper slag up to 50%. In agreement with previous studies, MLR analysis exhibited poor performance as compared to the ANN.
Khaderni et al.  compared three different techniques, namely, (a) ANN, (b) MLR model, and (c) ANFIS for estimation of characteristic compressive strength of concrete. The analysis results showed superior performance of both ANFIS and ANN models in the determination of mechanical properties of concrete as compared to MLR analysis.
Eskandari-Naddaf and Kazemi  developed an ANN model for prediction of characteristic compressive strength of concrete up to 52.5 MPa. The ANN model was trained using an experimental dataset with a different water-cement ratio of up to 0.5%. The analysis results showed that ANN models yielded accurate results. Khademi et al.  confirmed the accuracy of the ANN in estimating the compressive strength for concrete of different grades. Naderpour et al.  developed an ANN model trained based on 139 experimental datasets for prediction of compressive strength of recycled aggregate concrete. The proposed ANN model consisted of six input parameters such as fine aggregate, coarse aggregate, water-cement ratio, and percentage replacement of recycled coarse aggregate. Bui et al.  used a hybrid system with a combination of modified firefly algorithm and ANN for prediction of the tensile strength of concrete. The ANN model was trained from the experimental data, and MFA was used to initialize the weights. The analysis results showed better performance of a hybrid system as compared to ANN in predicting the relationship between compressive and tensile strength. Belmood and Golafshani  replaced silica fume by cement to manufacture a silica fume concrete. The authors aimed at investigating the effect of silica fume as a partial substitute of cement to reduce the consumption of cement and inhibit the generation of CO,. The authors used the ANN and a new methodology called multiobjective gray wolf optimization. The partial replacement of silica fume in concrete yielded good compressive strength, and the ANN predicted the compressive strength accurately. Prasad et al. (2019) determined the efficiency of the ANN model to predict the compressive strength of self-compacting concrete. The authors used 99 datasets from the experimental study to train the neural network. The authors used five input nodes and a three-layered feedforward BP network with 10 hidden layers. The analysis results showed the accurate prediction of compressive strength by the ANN. Rajeshwari and Mandal  replaced cement with FA as the binding material to investigate the mechanical properties of concrete. The authors collected 270 datasets from the literature study and 12 from the experimental results to propose an ANN model to predict the compressive strength of concrete. The ANN was developed utilizing eight input parameters associated with raw materials used in the manufacture of concrete. The analysis results showed that proposed ANN accurately predicted the compressive strength with a high coefficient of correlation. It was observed that the compressive strength of concrete decreased with an increase in the percentage of replacement of cement with FA. Hammoudi et al.  predicted the compressive strength of concrete at the ciuing period ranging from 7 to 56 days. The concrete was manufactured by using the industrial byproduct of recycled aggregate and replacing it with coarse aggregate. The authors used response surface methodology and neural networks to predict the compressive strength. The experimental results indicated that strength decreases with the replacement of recycled aggregates was increased from 0% to 100%. The statistical analysis showed an excellent correlation between the response surface method and the ANN method, and both ways accurately predicted the strength of concrete in compression. Hadzima-Nyarko et al.  replaced the aggregate in concrete with waste rubber and determined change in the properties of concrete. The authors, based on their experimental results, prepared a dataset of 457 mixes and proposed an ANN model for determination of strength properties. The results of the analytical study showed that strength was strongly influenced by waste rubber percentage. The proposed ANN model exhibited efficiency in the estimation of compressive strength. Similar observations were reported by Yaseen et al.  and Mansouri et al. .