Predictive study on Mechanical strength of Lightweight concrete using MRA and ANN

: The lightweight concrete is preferred over regular density concrete as it which reduces the dead load of the structure due to its lower density. The reduction in dead load of the structure, resulting in a considerable decrease in the size of structural elements and reinforcements; thereby, the building's cost can be reduced. The lightweight concrete is achieved through natural lightweight aggregates, artificial lightweight aggregates, coconut shells, oil palm shells, aeration in concrete, etc. The mechanical properties like compressive strength, tensile strength, density depend upon lightweight aggregate, fine aggregate, super-plasticizer, cement content, water-cement ratio, etc. The mechanical properties can also be predicted using artificial intelligence from the existing data. This research aims to predict lightweight concrete's mechanical properties using MRA and ANN accurately.

Lightweight aggregate concrete (LWAC) is a kind of concrete which has a low unit weight when balanced to that of normal weight aggregate concrete (NWAC).The low mass density of it has one of the big favors correlated with truncated self-weight of structures & is also enacted in long-span bridges and high-rise buildings .Also, the, structural LWAC, with a strength that is akin to NWAC, enables the limiting of construction outlay as it entails less reinforcement, minuscule assisting deck members, beams, & piers, & less earth tremor ruinous, the viable ease of LWAC is the haulage cost stockpile achieved by outstrip the upheave skillfulness in the construction field and lowering shipping cost, compared to conventional NWAC products.
LWAC has the same concrete components as conventional NWAC with a partial or complete substitute of normal weight aggregate (NWA) with lightweight aggregate (LWA).The LWAs have an inherently great porosity, contributing in low density, low strength, and deformable particles.LWAs generally has a density lower than 1920 kg/m 3 .A lower density of LWAC can be achieved by using a heftier lump of porous LWA, trickle-down abject mechanical performance.Compressive strength of LWAC relay on not only the content of LWAs, but also on other factors.Hence, these experimental studies shows that the properties & amount of LWAs influenced the mechanical behavior of LWAC. the mix proportions of LWAC are also the key parameters incite the capacity of LWAC, such as water-tocement ratio (w/c) & mass of aggregate, water, & binders including cement, fly ash, & silica fume.The intricate relationship between concrete constituents & properties of cement-based construction materials, researchers have employed artificial neural networks (ANN).In the field of construction materials, ANN methods were applied for creating concrete properties, including mechanical, fluidity, & durability in concrete components-related information as input parameters.This study gives a prediction model created on ANN and MRA based on mechanical characteristics of LWAC, which enable us to produce high-quality LWAC, satisfying the target performance.
Detailed & extensive data on the mix proportions & the mechanical behavior of LWAC are taken from literature.The vast amount of data allows to enhance the reliability and accuracy of the prediction model.The prediction model is evaluated and compared to the results obtained from the commonly used statistical models.

Multiple Regression Analysis (MRA):
In this study, the linear-type MRA modeling is done using MS excel.Coefficients of regression are evaluated by considering 95% confidence level, the error tolerance level is restricted to utmost of 5%.For a given input variable, the probability value is considered to be significant, only if it is less than 0.05.*From MRA, the backing coefficients presented in Table () were found and substituted in linear multiple regression equation (equation ( 1)):

Statistical Test:
The prediction model is done with MRA and ANN and the analysis is done regression analysis where the coefficient of determination (R 2 ) where the accuracy is checked with the values which gives us the validation of the model which is being created by various prediction modeling.This coefficient generally checks the difference or the amount of deviation from one value to the other value.Here the coefficient of determination is used for checking the deviation of the predicted value from the original value.The range of the R 2 varies from 0 to 1 (i.e., 0 to 100 %).(R 2 ) determination is give in equation ( 2), precision of the predictions of a network was appraised by RMSE difference, between the experimented and the predicted values.
Sum The prediction of ANN and MRA for compressive strength of 3 days is shown in  , the compressive strength of 7 days is used for the prediction which has the MRA and ANN analysis respectively here it also shows that the ANN model is better for the prediction as its error limit is less and it will give a proper prediction.For Fig. 10,11 and 12, compressive strength of 28 days is used for the prediction which has the MRA and ANN analysis respectively here it also shows that the ANN model is better for the prediction as its error limit is less and it will give a proper prediction.

Fig 1 , 2
and 3 where the R 2 predictions are shown.It has been found out that prediction for MRA is 0.5474, ANN (10 neurons) is 0.854 whereas on the other side for ANN (15 neurons) it is 0.8698.On the basis of these results, we can easily say that ANN (15 neurons) has more accuracy and can be used for prediction model.The efficiency of prediction model is totally depending on the accuracy of the output.In MRA the lower value of coefficient of regression only depicts that there will be more errors occur as compared to ANN model.So, we cannot use MRA model here for prediction of compressive strength of light weight concrete.Only ANN model can be taken into consideration for output.

2. Prediction Modeling and Testing:
Depending on the input parameter & target values, the output was effectuated through MRA and ANN, output values were equated with target (actual) values.Types of fibers and its respective literature source are presented in Table 1.Active compressive strength (3 days) data set has 64 columns and 3916 rows (64 × 3916) of input data and 1 column and 3916 rows (1 × 3916) of target data.Active compressive strength (7 days) data set has 64 columns and 3916 rows (64 × 3916) of input data and 1 column and 3916 rows (1 × 3916) of target data.Active compressive strength (14 days) data set has 64 columns and 3916 rows (64 × 3916) of input data and 1 column and 3916 rows (1 × 3916) of target data.Active compressive strength (28 days) data set has 64 columns and 3916 rows (64 × 3916) of input data and 1 column and 3916 rows (1 × 3916) of target data.Active split tensile strength data set has 5 columns and 1328 rows (64 ×1322) of input data and 1 column and 119 rows (1 × 1322) of target data.Active Density data set has 64 columns and 2872 rows (64 ×2872) of input data and 1 column and 2872 rows (1 × 2872) of target data.Target data for density, compressive strength and split tensile strength were used in both the MRA and ANN model as separate target in this study.

Table 1 : Range of parameters in data base for prediction model S. No. Type Type of Material Material Unit Content Range
2.1 Artificial Neural Network (ANN):Prediction model done is through MATLAB with two hidden layers, (10 and 15 neurons) in every hidden layer & one output layer with dependent variable as density, compressive strength and split tensile strength.Along with all the data, approximately 70%, 15%, &15% has been scrutinized for training, testing, &validation.The Levenberg-Marquardt (LM) algorithm is utilized for training due to its robustness & speed.Layered feed-forward networks have been practiced in this algorithm, in which the neurons are grouped in layers.Here, signals are sent forward & errors are propagated backwards.

Table 2 : R 2 values
To determine compressive strength of various days based on the parameter having various types of inputs, by using ANN and MRA.The validation of the model is made with coefficient of regression (R 2 ) shown in table 1.
In this study, the models were prepared to predict the mechanical behavior (mechanical strength) of LWAC based on input parameters, & four methods were used, ANN, MRA, Orange & Anaconda, prediction models are validated R 2 & RMSE & are consolidated in Table.