Prediction of concrete strength using Multilayer Perceptron Neural Network-based utilizing sustainable waste materials

doi:10.21203/rs.3.rs-6884648/v1

This paper presents a laboratory investigation on optimum level of Vitrified polish waste (VPW) and Ground Granulated Blast Furnace Slag (GGBS) as a partial replacement of cement to study the strength characteristics of concrete. Ordinary Portland cement was partially replaced by 5%, 10%, 15%, and 20% mix of Vitrified polish waste and GGBFS. The water to cementitious materials ratio was maintained at 0.38 for all mixes. The strength characteristics of the concrete were evaluated by conducting compressive test, strength test, splitting tensile strength test and flexural strength test. The compression strength test was conducted for 7 and 28 days of curing, while the split tensile strength test and flexural strength test were performed on an M30, M35 and M40 grade concrete. The mix proportion M30, M35 and M40 was found to be 1:1.615:3.427, 1:1.50:3.25, and 1:1.40:3.15 respectively. The test results proved that the compressive strength, split tensile strength and flexural strength of concrete mixtures containing GGBFS and VPW increases as the amount of GGBS and VPW increase. A Multilayer Perceptron (MLP) Neural Network was utilized to evaluate the concrete strength, with the predicted values aligning closely with the observed data. The results indicate that an optimum point, at 15% of GGBFS and VPW of the total binder content, the further addition of GGBFS and VPW does not improve the compressive strength, split tensile strength and flexural strength.

Multilayer Perceptron Neural Network

Sustainability

Concrete strength

Vitrified polish waste

Ground Granulated Blast Furnace Slag

There are two principal categories of pavements: flexible and stiff. Flexible pavements consist of several layers that allow for flexural movement, whereas stiff pavements are constructed using Plain Cement Concrete (PCC) surface courses (Aly et al. 2019). In recent decades, bitumen roads have gained increased prevalence. These roads consist of a composite of bitumen and aggregates that are heated and applied onto a substrate of granular material. The substitution of cement with waste powder is a trend attracting significant attention due to the increasing awareness of environmental conservation and sustainable construction (Tipu et al. 2024). The lifespan of bitumen roads often spans 20 to 30 years, with its degradation from weather and pollution being more pronounced than that of rigid pavements. Conversely, rigid pavements comprise a blend of aggregates, water, and cement. Their lifespan varies between 20 and 40 years, and they inflict less environmental damage than flexible pavements. Currently, concrete pavements are extensively utilized (Beladi et al. 2012). The present global situation is marked by numerous challenges stemming from rapid population expansion, technological breakthroughs, and changes in lifestyles. In both developed and developing countries, the effective management of waste and environmental protection is a key concern, since rising disposal prices exacerbate the issue, posing threats to public health and imposing economic burdens. To address this issue, waste materials produced by the construction sector are being assessed, including Ground Granulated Blast Furnace Slag (GGBFS) and ceramic waste. These two will serve as a partial substitute for cement in pavement construction. Ceramic solid waste is produced during production, installation, transportation, and demolition activities (Tokareva et al. 2024). This trash is gathered from building businesses post-demolition and utilized as recyclable material. In the majority of countries, solid waste is not effectively utilized (Kumar et al. 2024). The annual generation of ceramic waste from industry is approximately 100 million tons. It is estimated that 15% to 30% of waste in the tile business is produced during the complete manufacturing process (Fu and Lee 2024). Nevertheless, ceramic waste and GGBFS exhibit hardness, durability, and significant resistance to chemical, physical, and biological pressures. These are utilized in both structural and non-structural applications, particularly in rigid pavements. It minimizes pollution, mitigates weather-related damage, and is environmentally sustainable. Ground granulated blast furnace slag (GGBFS) is a by-product of iron production in blast furnaces. The principal constituents of GGBS are molten calcium silicate and aluminosilicate, which need frequent extraction from the blast furnace. In India, the annual output of GGBS is 15 million tons, with merely 55% being employed in the building sector. Consequently, it is essential to augment the utilization of industrial waste on a broader scale for environmental advantages within the concrete industry (Kumar et al. 2008). Often used as a supplemental cementitious material in blended cement, it reduces the need for Portland cement when creating mortar or concrete. The utilization of this compressive power enhances both the durability and strength of concrete. Vitrified Polished Waste: The ceramic industry is the primary producer of waste ceramic (WC) (Rahla et al. 2019). Waste ceramic powder (WCP), sanitary ware, and ceramic electrical insulator waste represent various categories of waste ceramic (WC). Ceramic tiles, produced by heating natural materials like clay, quartz, feldspar, and sand, are a category of WC (Vivek and Dhinakaran 2017). The annual output of ceramics in India approximates 100 million tons (Kora and Vyas 2016). The daily production of ceramic trash in Goa is 30 metric tons. The entire manufacturing process in the tile business generates an estimated 15% to 30% of waste. The objective of the current study is to mitigate the environmental hazards associated with the utilization of GGBFS and ceramic waste (Alhassan and Ballim 2017). Utilizing this industrial waste in roadway engineering beds reduces pollution compared to the use of plastic or disposable materials. The strength of concrete is enhanced by 80% to 90% with the incorporation of ceramic and GGBFS waste, with concern for human health (Arora et al. 2024). Not much research has looked into using ceramic powders in cement mixtures, highlighting terracotta and porcelain powders as useful extra materials for cement. The tests used included the Chapelle test and XRD analysis to check how well different types of ceramics reacted, showing that their reactivity varied. The inclusion of limestone filler with ceramic particles diminished strength relative to the use of ceramic powders only. Also, (Kannan et al. 2017; Adesanva 1996; Rawat et al. 2024; Yaswanth et al. 2024; Boddepalli et al. 2024; Kora and Vyas 2016; Cheng et al. 2014; and Beladi et al. 2012) studied high-performance concrete (HPC) that used ceramic waste powder to partially replace Portland cement. At 28 days of age, the control mixture had the greatest compressive strength value of 51.5 MPa. As the quantity of cement diminishes incrementally, the 28-day compressive strength declines by 15%, 17%, 18%, and 20% for HPC-10, HPC-20, HPC-30, and HPC-40, respectively. Post 90 days of age, the rate of compressive strength growth was marginally enhanced for all concrete mixtures, including CWP. Ceramic waste exhibits two distinct impacts; firstly, it serves as a filler, facilitating the formation of a dense packing system that enhances the performance of concrete. Second, as concrete ages, its pozzolanic impact becomes more significant compared to its early years. However, there needs to be enough CH in the hydrated concrete mix for the CWP to work effectively. The limited pozzolanic reactivity observed in the analyzed HPC mixtures results from insufficient water content in the concrete blend, which is caused by the relatively low water/cement ratio of the reference concrete mixture. Limited research (Nehring et al. 2023; Gopal and Reddy 2017; Keramatikerman et al. 2016; El‑Hassan and Kianmehr 2016) has concentrated on the application of GGBS in their experiments. Several studies (Bandi et al. 2024; Tipu et al. 2024; Kumar et al. 2024; Kevah 2024; Kaveh and Khavaninzadeh 2023; Bilim et al. 2009; Glasser 1994) have concentrated on machine learning algorithms to forecast concrete strength, particularly Bilim et al. (2009) conducted a study using an artificial neural network (ANN) to predict the compressive strength of concrete containing ground granulated blast furnace slag. The study encompassed laboratory experiments utilizing a dataset, demonstrating that slag replacement ratios influence concrete strength. The research underscores the advantages of incorporating ground granulated blast furnace slag in concrete manufacturing to improve durability and diminish environmental effects. The ANN model demonstrated efficacy in forecasting compressive strength based on concrete constituents. Additionally, only a few researchers have looked into how the amount of slag and the fineness of clinker affect the properties of Portland slag cement. The comparison of the environmental benefits of using a mix of concrete with Supplementary Cementitious Materials (SCMs) and Self-Compacting Concrete (SCC) has changed the construction industry with its innovative features. Previously, Shinde and Kumar (2023) examined the deflection, maximum stress, maximum strain, and failure of pavement joints in traditional and remix concrete pavement models. We fabricated nine pavement slab specimens and subjected them to static loading using a hydraulic jack and load cell. The study investigates the impact of remix concrete on inflexible pavements, taking into account the concrete's time lag and blend ratio. Results indicate that incorporating higher-grade fresh concrete into aged concrete enhances compressive strength and quality, in addition to the modulus of subgrade reaction. Kumar (2012) examined the efficient application of fly ash and extra cementitious materials within the construction sector. It examines the properties and features of fly ash and GGBS, as well as their effects on concrete mix proportions, workability, and compressive strength. The research seeks to optimize the ternary material combination to produce high-strength, durable concrete while minimizing cement usage.

The aforementioned research indicates that comprehending the interactions between waste species and cement components is essential for forecasting the performance of cement-conditioned waste. The utilization of vitrified polish waste with GGBFS enhances strength and durability. The primary downside of the highway and installation is its fragility, since it is susceptible to damage upon collision with hard objects. The utilization of this industrial waste diminishes cement output and lowers construction expenses. The current study demonstrates a notable enhancement in strength, durability, and performance characteristics of concrete by the incorporation of GGBFS and ceramic waste. This study aims to address the deficiencies identified in prior research by pursuing the following objectives:

a) To examine the impact of partially substituting Ordinary Portland Cement (OPC) with Vitrified polish waste (VPW) and Ground Granulated Blast Furnace Slag (GGBS) on the strength properties of concrete.

b) To determine the ideal substitution levels of VPW and GGBS that enhance the compressive strength, split tensile strength, and flexural strength of concrete.

c) To evaluate the mechanical properties of concrete across several grades (M30, M35, and M40) with differing proportions of VPW and GGBS.

d) To evaluate the performance of concrete mixtures via laboratory testing at various curing ages and mix ratios.

e) To use a multilayer perceptron (MLP) neural network model to predict how strong the concrete will be and to assess the results from the experiments.

The experimental research is carried out in accordance with the technique that is described below. Materials that are required for the study have been obtained, and preliminary studies have been carried out. These investigations have included an evaluation of the mix design calculations and the outcomes achieved for the production of M30, M35 and M40 grade concrete. The mechanical properties of test specimens were evaluated after they were manufactured with different amounts of GGBFS and VPW in place of OPC. These proportions were 0%, 5%, 10%, 15%, and 20%, respectively. After conducting tests to establish the compressive and flexural strengths of the cast specimens at 7 and 28 days, it was determined that the optimal amounts of VPW and fly ash were for the substitution of OPC. A cost-benefit analysis has been carried out, and the design of the cement concrete pavement has been accomplished in accordance with the requirements that are outlined in IRC 44:2017. Further, the detailed procedure of present study is shown in Fig. 1.

Materials and Methods

The experimental inquiry utilizes Ordinary Portland Cement (OPC) of 53 grade, conforming to IS: 12269 − 2013. Table 1 presents the characteristics of cement evaluated in accordance with IS: 4031 − 1988.

Table 1

Characteristics of Cement
Test	Result	Code used	Range
Specific gravity	3.15	IS 4031 Part-9:1988	3.0-3.20
% Fineness	5.33%	IS 4031 Part-1:1988	Residue < 10%
Normal consistency	32%	IS 4031 Part-4:1988	27–33% by wt.
Initial Setting time	60min	IS 4031 Part-4:1988	Should be > 30min
Final setting time	480min	IS 4031 Part-5:1988	Should be < 10hrs
Soundness	2mm	IS 4031 Part-3:1988	Should be < 2mm

The fine aggregate contains the following: (IS: 2386 Part-1, IS 383: 1970 Table 4, IRC 44 Table 2) (Vasudeva 2023). In this experimental program, the fine aggregate that is used as river sand is purchased from the Godavari River region in accordance with the specifications of IS: 383–1987. The sample is then sieved with a 4.75mm IS sieve in order to remove any harmful sand particles that are larger than the required size. The sample is then tested for its properties in accordance with IS: 2386 − 1963, and the results are presented in Table 2.

Table 2

Properties of Fine Aggregate
Test	Result	Code used	Range
Specific Gravity	2.62	IS 2386 Part-3:1963	2.5-3.0
Sieve Analysis	Zone-III	IS 2386 Part-1:1963	N. A
Fineness Modulus	2.29	IS 2386 Part-1:1963	2.2–2.6
Water Absorption	1%	IS 2386 Part-3:1963	0.1-2%

Sieve Analysis of Fine Aggregate: In order to determine the gradation of the fine aggregate, sieve analysis is carried out. In accordance with the International Standard 383–1987, the zoning of fine aggregate and the fineness modulus are derived. For the purpose of this investigation, one thousand grams of fine aggregate is collected and then sieved using the list of sieves that is presented below. The results are presented in Table 3.

Table 3

Gradation of Fine Aggregate
S. No	Sieve size (mm)	Particle size (mm)	Wt. retained (gm)	% of Wt. retained	Cumulative % retained	% of Passing	Zone- III
1	4.75	4.75	40	2	2	98	90–100
2	2.36	2.36	56	2.8	4.8	95.20	85–100
3	1.18	1.18	180	9	13.8	86.20	75–100
4	0.6	0.6	381	19.05	32.85	67.15	60–79
5	0.3	0.3	896	44.8	77.65	22.35	12–40
6	0.15	0.15	415	20.75	98.40	1.60	0–10
7	0	0	30	10	99..9	0.10

The table that is located above can be used to determine the grade of fine aggregate. In accordance with the International Standard 383–1987, the grading of fine aggregate is in accordance with zone third (Yaragal et al. 2019). It has been determined that the fineness modulus is 2.29, and the graphical depiction of the grain size analysis of fine aggregate can be found below (Fig. 2).

Coarse Aggregate: (International Standard 2386 Part-3), (IRC44:2017 Clause 3.4.1)

In accordance with the International Standard IS:383–1987, this investigation makes use of crushed stone aggregate with diameters of 20mm and 10mm. The specific gravity of both fractions has been achieved to be equal. The flakiness and elongation index of the coarse aggregate have been achieved to be less than 15%. The impact and crushing value of the coarse aggregate that was tested reaches less than 30% in accordance with the International Standard 2386 − 1963. The results of these tests are presented in Table 4, which can be found below.

Table 4

Properties of Coarse aggregate
Test	Results	Code used	Range
Specific gravity	2.74	IS 2386 part-3:1963	2.5-3.0
Crushing value	19.5%	IS 2386 Part-4:1963	Should be < 30%
Impact value	23.6%	IS 2386 Part-4:1963	Should be < 35%.
Flakiness index	8.42%	IS 2386 Part-1:1963	Should be < 15%.
Elongation index	14.28%	IS 2386 Part-1:1963	Should be < 15%.
Water absorption	0.6%	IS 2386 Part-3:1963	0.1-2%

Vitrified polish waste, it is a waste material that is generated from the tiles industry. For the purpose of this investigation, VPW was collected from R.A.K. Ceramics in Samalkot, which is located in the East Godavari district of Andhra Pradesh, India. The specific gravity of VPW was determined to be 2.45, and it achieved a fineness of 93.7% when it was passed through a 90-micron sieve. The results of the tests are presented below in Table 5.

Table 5

Properties of Vitrified polish waste
Tests	Results	Code used	Limits
Specific Gravity	2.45	IS 4031 Part-11:1988	-
% Fineness for 90 micron IS sieve	6.3%	IS 4031 Part-1:1988	Residue < 10%

A byproduct of the iron production process that takes place in blast furnaces is known as ground granulated blast-furnace slag, or GGBFS for short. This slag is frequently referred to as GGBFS. It is largely made up of a combination of silicate and aluminosilicate elements that are formed from molten calcium, which is frequently taken from the blast furnace. The ground granulated blast furnace slag that is being used for this inquiry was obtained from Astrra Chemicals in Tamil Nadu, which is located in India. Following the analysis, the results are presented in Table 6 (Nwankwojike et al. 2014). These results are based on the Indian norms.

Table 6

Properties of GGBFS
Tests	Results	Code used	Limits
Specific Gravity	2.45	IS 4031 Part-11:1988	-
% Fineness for 90 micron IS sieve	3.6%	IS 4031 Part-1:1988	Residue < 10%

Chemical Admixture: Haksons Ultra Clear Epoxy Resin is a resin that is of superior quality, possesses a crystal clear appearance, and is ideal for a wide variety of applications. When this resin is allowed to cure, it forms a finish that is not only scratch-resistant but also resistant to yellowing, fading, and fading. A professional and smooth finish may be achieved on a wide range of surfaces using this product, which is simple to use (refer Table 7 for properties)

Table 7

Properties of Chemical Admixture
Property	Value
Colour	Milky White
Density	1100kg/m³
Ph	7
Specific gravity	1.12
Chloride content	1,000 g/min at 27°C
Air entrainment	4%< 7%

Deign and Mix Proportions

The batch mixer requires that testing samples be prepared in a specific order: first coarse aggregate, then fine aggregate, followed by cement, VPW, and GGBFS, and finally, superplasticizer mixed with water. The complete procedure for preparing the total mix proportions is finalized within 5 minutes, and to guarantee color consistency, adequate mixing of VPW and GGBFS is essential. The composite mixture is subsequently poured into three distinct molds: a cube measuring 150mm x 150mm x 150mm, a prism measuring 500mm x 100mm x 100mm, and cylindrical molds with a diameter of 150mm and a height of 300mm. Prior to casting the molds, waste oil is applied to their surfaces. We layer the mixture in three strata within the molds and then crush it with a tamping rod to remove any air voids. The table is subsequently vibrated, and the upper surface is smoothed using a trowel. The samples are removed from the molds after 24 hours and immersed in a curing tank for 7 days and 28 days.

Table 8

Details of Mix Proportions
Ingredient	Weight
Cement	388.042 kg/m³
Fine Aggregate	626.83 kg/m³
Coarse aggregate	1330.185 kg/m³
Water	147.456 kg/m³
Super Plasticizer	3.880 kg/m³
Water/Cement	0.38

Mix designation of replacement of VPW and GGBFS

The total quantities of VPW and GGBFS replacement in the cement along with different proportions used in concrete shown in Table 9.

Table 9

Mix designations of cementitious materials replacement
Name of the Mix	Constituents
C.C	100%C + 0% of (VPW + GGBFS)
(VPW + GGBFS)-1	95%C + 5%of (VPW + GGBFS)
(VPW + GGBFS)-2	90%C + 10 % of (VPW + GGBFS)
(VPW + GGBFS)-3	85%C + 15% of (VPW + GGBFS)
(VPW + GGBFS)-4	80%C + 20% of (VPW + GGBFS)

Table 10

Quantity of ingredients used in mix proportion
Type of mix proportion	Cement (kg/m³)	VPW (kg/m³)	GGBFS (kg/m³)	F.A (kg/m³)	C.A (kg/m³)	Water content (kg/m³)	Admixture (kg/m³)
C.C	388.042	0	0	626.83	1330.185	147.456	3.88
(VPW + GGBFS)-1	368.639	9.70	9.70	626.83	1330.185	147.456	3.88
(VPW + GGBFS)-2	349.237	19.40	19.40	626.83	1330.185	147.456	3.88
(VPW + GGBFS)-3	329.835	29.10	29.10	626.83	1330.185	147.456	3.88
(VPW + GGBFS)-4	310.433	38.80	38.80	626.83	1330.185	147.456	3.88

MLP model

In the present study, ANN model named Multilayer Perceptron (MLP) Neural Network has been employed to predict the compressive strengths of concrete. % VPW + GGBS, grade of concrete, mix proportions and water cement ratios are input by the input layer of an MLP, a form of artificial neural network. The hidden layers learn complex correlations between input and output. Neurons in every layer use activation functions like linear activation for regression outputs and Rectified Linear Unit (ReLU) for hidden layers to make changes to math. The detailed methodology of MLP model can be seen in the flow chart Fig. 4.

Table 11

Input and output parameters used in MLP model
Input parameters	Output parameters
% VPW + GGBS replacement	Compressive strength
Curing time	Compressive strength
Grade of Concrete (M30, M35, M40)	Split tensile strength
Mix proportion values	Split tensile strength
Water to cementitious	Flexural strength

Table 12

Predicted strengths from MLP model
Grade	VPW-GGBS (%)	Cement (%)	Fine aggregates (%)	Course aggregates (%)	Water-cement ratio (%)	Curing time (days)	Predicted compressive strength (MPa)	Predicted Tensile strength (MPa)	Predicted Flexural strength (MPa)
M30	0	100	1.615	3.427	0.38	28	40.5	3.56	3.56
M30	5	95	1.615	3.427	0.38	28	41.22	3.95	4.12
M30	10	90	1.615	3.427	0.38	28	40.65	3.95	4.12
M30	15	85	1.615	3.427	0.38	28	43.65	4.25	4.98
M30	20	80	1.615	3.427	0.38	28	39.6	4.02	3.98
M35	0	100	1.50	3.25	0.38	28	45.52	3.99	3.68
M35	5	95	1.50	3.25	0.38	28	46.65	4.56	4.02
M35	10	90	1.50	3.25	0.38	28	45.88	5.25	4.23
M35	15	85	1.50	3.25	0.38	28	48.55	5.01	4.99
M35	20	80	1.50	3.25	0.38	28	40.65	5.02	3.87
M40	0	100	1.40	3.15	0.38	28	50.67	4.12	4.56
M40	5	95	1.40	3.15	0.38	28	51.23	5.12	4.56
M40	10	90	1.40	3.15	0.38	28	52.36	5.68	5.1
M40	15	85	1.40	3.15	0.38	28	52.12	5.36	5.01
M40	20	80	1.40	3.15	0.38	28	44.56	5.35	4.1

Input and output parameters

The Multi-Layer Perceptron (MLP) model employs designated input and output parameters for its functionality. The input parameters for the MLP model include key concrete mix design variables: the percentage of VPW + GGBS replacement, curing time, concrete grade (M30, M35, M40), mix proportion values, and the water-to-cementitious ratio. The model is designed to predict three output parameters viz., compressive strength, tensile strength and flexural strength. The detailed input, hidden and output parameters of MLP architecture has been given in Tables 11 & 12. Table 12 represents, various combinations of input parameters to predict the output strength. M30 grade exhibits variations in VPW-GGBS replacements (0, 5, 10, 15, and 20) percentages, while cement value remains constant at 100% or adjusted according to VPW-GGBS. Further, fine aggregates and course aggregates are fixed at 1.615 and 3.427 respectively. The water-cement ratio is set at 0.38, and the curing time established at 28 days. Similarly, the same procedure is adopted for remaining grades (M35, M40). Wherein, fine aggregates and course aggregates are fixed at 1.5 and 3.25 respectively for M35. On the other hand, fine aggregates and course aggregates are fixed at 1.4 and 3.15 respectively for M40. The water-cement ratio is set at 0.38 in both cases. Figure 5 illustrates the MLP design and its comprehensive structure, comprising 15 nodes in the Input layer, 3 nodes in the Hidden layers, and 3 nodes in the Output layer. MLP proposes a prospective arrangement of a multilayer perceptron, consistent with the mathematical foundation outlined below. Training multilayer perceptron networks necessitates the implementation of a backpropagation method (Taud and Mas 2017). A multilayer perceptron is a form of supervised feedforward neural network consisting of at least three layers: an input layer, one or more hidden layers, and an output layer. The hidden layer and the output layer utilize a nonlinear activation function. The total input $\:{x}_{j}^{k+1}$ received by a neuron j in layer $\:k+1\:$can be expressed as follows in Eq. (1):

$$\:{x}_{j}^{k+1}=\sum\:_{i}{y}_{i}^{k}{z}_{i,j}^{k}-{\alpha\:}_{j}^{k+1}$$

1

In this equation, yi represents the state of the ith neurone within the kth layer, while zij denotes the weight connecting the ith neurone in layer k to the jth neurone in layer k + 1. θ represents the threshold of the jth neurone located in the k + 1 hidden layer (Raheli et al. 2017). Additionally, the output of a neurone in any layer, excluding the input layer, can be expressed according to Eq. (2):

$$\:{y}_{j}^{k}=\frac{1}{1+{e}^{{-x}_{j}^{k}}}$$

2

Throughout the training period, the network demonstrated high R² values for compressive strength (CS), tensile strength (TS), and flexural strength (FS), with values of 98.29, 95.34 and 78.45 respectively. Conversely, the compressive strength, water absorption, and hardness exhibit low error values of 0.72, 0.42, and 0.58, respectively. The MLP's performance is shown in Table 13. It is the sum of the differences in measured and calculated output parameters (CS, TS, and FS) during the training, testing, and validation phases. This performance is indicative of the goodness of fit between experimental measurements and model-calculated outputs. The random function was employed to select samples for these stages. We divided the collected database into three groups: 70% for training, 15% for testing, and 15% for validation.

The performance of the MLP model in predicting the strengths of compressive, tensile, and flexural was evaluated using training, testing, and validation for M30, M35, and M40 grades, respectively (see Table 13). It can be observed that the model demonstrated high accuracy across all strengths. For compressive strengths, validation results show the highest predictive accuracy, with 98.29% for M30, 97.53% for M35, and 97.51% for M40, indicating the model has strong generalization capability. Similarly, for tensile strength, the validation accuracy was slightly higher in comparison with training and testing: 95.34% (M30), 98.34% (M35), and 97.61% (M40). But, in the flexural strength case, although the accuracies were comparatively lower, the model still achieved valid results, with validation accuracies of 78.45% (M30), 86.90% (M35), and 91.63% (M40). Overall, the constant increase in accuracy from M30 to M40 across strength parameters suggests the model performs slightly better with higher-grade concrete.

Table 13
R² between observed and predicted by MLP during training, testing, and validation process
	Compressive strength			Tensile strength
	M30	M35	M40	M30	M35	M40	M30	M35	M40
Training	97.21	95.35	96.21	93.89	96.54	96.12	77.56	84.74	90.21
Testing	97.35	96.54	96.47	94.66	97.32	96.54	77.33	85.47	90.32
Validation	98.29	97.53	97.51	95.34	98.34	97.61	78.45	86.90	91.63

The analysis of how well the MLP models predict the compressive, tensile, and flexural strengths for the three grades, M30, M35, and M40, shows that they are reliable for making predictions. Table 14 indicates that R² was high across all grades for compressive strength, with values of 98.29% (M30), 97.53% (M35), and 97.51% (M40). However, the RMSE for M30 and M35 remained low at 1.484 and 1.631, respectively, despite being demonstrated to be higher for the M40 grade. Additionally, SSE adhered to a comparable trajectory and exhibited a minor error in the M40 grade. The model showed strong reliability in measuring tensile strength, with R² values of 95.34% for M30, 98.34% for M35, and 97.61% for M40, and RMSE values below 0.25, which means it is very accurate. Further, SSE values are 0.360, 0.317, and 0.495 for M30, M35, and M40, respectively. However, in the flexural case, the model shows slightly lower strengths but reliable R². The flexural case demonstrates RMSE and SSE within acceptable limits, indicating a gradual improvement with concrete grade.

Table 14
Uncertainty analysis across the grades in compressive tensile and flexural strengths
		M30	M35	M40
Compressive strength	Coefficient of determination (R²)	98.29	97.53	97.51
	Root mean square error (RMSE)	1.484	1.631	21.62
	Sum of squares error (SSE)	11.23	21.29	21.62
Tensile strength	Coefficient of determination (R²)	95.34	98.34	97.61
	Root mean square error (RMSE)	0.212	0.199	0.248
	Sum of squares error (SSE)	0.360	0.317	0.495
Flexural strength	Coefficient of determination (R²)	78.45	86.90	91.63
	Root mean square error (RMSE)	0.408	0.322	0.308
	Sum of squares error (SSE)	1.336	0.829	0.759

The predictive performance of the MLP model was evaluated for different concrete grades across three mechanical properties, viz., compressive, tensile, and flexural strengths. From Fig. 6, it can be observed that the model demonstrates good calibration for compressive strength, with the predictive values closely aligning with observed values. Overall, the MLP model exhibits robust generalization for compressive and tensile strength predictions, with moderate performance for flexural strength, indicating potential for refinement using enriched feature sets or hybrid modeling approaches.

In the present study, an experimental investigation on fresh concrete mix using a slump cone test in accordance with IS: 516–1959. Furthermore, the hardened concrete assessments, including compressive, flexural strength, and split tensile strength tests for 7 and 28 days, were conducted in accordance with the same specifications. Fresh concrete is a newly mixed substance. The workability of the concrete mix, further, inquiry involved conducting a slump cone test to assess the workability of a fresh concrete mix. The slump cone test, or concrete slump test, is a technique for assessing the consistency of fresh concrete prior to its setting. It is utilized to assess the workability of the concrete, or its fluidity. The test is conducted utilizing a metal cone, referred to as a slump cone or Abrams cone, in seven phases. The cone is lifted vertically, and the distance the concrete has settled is measured to assess its uniformity (see Fig. 7). The slump values are obtained at the different mix proportions of VPW and GGBFS presented in the Table 14.

Table 14
Slump Values for (VPW + GGBFS) replacement
Mix Proportion	Slump (mm)
C.C	40
(VPW + GGBFS)-2	36
(VPW + GGBFS)-3	32
(VPW + GGBFS)-4	27
(VPW + GGBFS)-5	24

A conventional evaluation for hardened concrete involves measuring its mechanical strength and durability, confirming that the on-site concrete has attained the requisite strength under external loads. This study included tests on hardened concrete, specifically measuring compressive strength, flexural strength, and split tensile strength, following the rules set by IRC: 516–1959. The results were also looked at to find the right amount of VPW and GGBFS to replace in OPC, with the goal of increasing the compressive and flexural strength of the mixtures being studied.

The compressive strength test aims to assess the crushing strength of hardened concrete. We fabricate cubes measuring 150mm x 150mm x 150mm using VPW and GGBFS as cement substitutes for this test. Following a curing duration of 7 and 28 days, the samples are air-dried prior to testing. We determine the average compressive strength by evaluating three cubes in accordance with IS: 516–1959 and listed in Table 15 also, refer Fig. 8. The evaluation is conducted using a compressive testing machine (C.T.M.) with a capacity of 2000 kN and a loading rate of 5.20 KN/sec.

Table 15
Compressive strength testing Values
S. No	Name of The Mix	Mean Compressive Strength
		(MPa)
		7 Days	28 Days
1	C.C	26.77	39.42
2	(VPW + GGBFS)-1	27.13	40.95
3	(VPW + GGBFS)-2	28.69	41.67
4	(VPW + GGBFS)-3	31.45	44.09
5	(VPW + GGBFS)-4	24.12	38.59

The information regarding the compressive strength testing cube values is presented in the table 5.2 that can be found above. After seven and twenty-eight days of curing, these values increased up to fifteen percent of VPW and GGBFS. After twenty percent, the values gradually decreased, and the maximum value was reached at fifteen percent of VPW and GGBFS. These values are displayed in Fig. 7.

Flexural testing assesses the bending characteristics of a material. This procedure, occasionally termed a transverse beam test, involves placing a sample between two supports and applying a load via a third point or two points, commonly referred to as 3-point bend and 4-point bend testing, respectively. We employ a 2-point load testing machine on a prism measuring 500mm x 100mm x 100mm. The Flexural Testing Machine (F.T.M.) complies with IS: 516–1959 standards. This machine operates only through manual intervention. The tensile strength of concrete can be ascertained by evaluating its flexural strength. It signifies the concrete beam or slab's resistance to bending. The test assesses the concrete's resistance to bending failure, with the result expressed as a modulus of rupture (MR) in MPa. The details of the prism values used for flexural strength testing are shown in the Table 16 that can be seen above. Up to 15% of VPW and GGBFS, these values climbed, and after that, they steadily reduced until they reached 20%, at which point they reached their highest possible value. It is seen in Fig. 9 that the values for 15% (VPW + GGBFS) after 7 days and 28 days of cure are, respectively.

Table 16
Flexural strength testing values
S. No	Name of The Mix	Mean Flexural Strength
		(MPa)
		7 Days	28 Days
1	C.C	3.75	4.67
2	(VPW + GGBFS)-1	3.79	4.74
3	(VPW + GGBFS)-2	3.87	4.85
4	(VPW + GGBFS)-3	4.06	5.12
5	(VPW + GGBFS)-4	3.68	4.25

The split tensile test provides a roundabout way to evaluate the tensile test of concrete. For the purpose of this test, a conventional cylindrical specimen is positioned in a horizontal position. The force is then delivered to the cylinder in a radial way on the surface, which results in the production of a vertical crack in the specimen along its diameter. As validated by IS: 5816 − 1999, the diameter of the cylinder measures 150 mm, and its height measures 300 mm. The information regarding the compressive strength testing cube values is presented in Table 5.6 that can be seen above. The values increased to a maximum of 15% for both VPW and GGBFS, after which they gradually decreased over the seven-day and twenty-eight-day curing periods as shown in Table 17 and Fig. 10. We indicate these values at the point where they reach their maximum value.

Table 17
Split tensile strength values
S. No	Name of The Mix	Mean Split Tensile Strength
		(MPa)
		7 Days	28 Days
1	C.C	2.52	3.92
2	(VPW + GGBFS)-1	2.69	4.15
3	(VPW + GGBFS)-2	2.74	4.31
4	(VPW + GGBFS)-3	2.95	4.47
5	(VPW + GGBFS)-4	2.54	4.21

The present study investigated the feasibility of partially replacing OPC with VPW and GGBS in concrete, focusing on mechanical performance across grades, viz., M30, M35, and M40.The following conclusions were drawn from the study:

The experimental results revealed that incorporating VPW and GGBS improved the compressive, tensile, and flexural strengths of concrete, with optimal replacement of 15% of total binder content. We observed a decline in strength characteristics beyond this level, suggesting a threshold for effective substitution.

The slump cone test verified acceptable workability across mixes. In particular, the compressive strength showed a notable increase up to the 15% replacement level, after which diminishing returns were evident.

The MLP model was effectively employed to predict the mechanical properties of concrete. The model demonstrates high prediction accuracy with R² values exceeding 97% for compressive strength and over 95% for tensile case across the grades.

The model slightly underperformed in the flexural case but still produced reliable results of R² values above 78%. RMSE and SSE remained within acceptable limits, reinforcing the models' robustness, particularly for higher-grade concrete.

From overall observations, this study establishes that a 15% substitution of OPC with VPW and GGBS is optimal for enhancing strength properties without compromising workability. The integration of the MLP model provides an efficient and reliable method for predicting concrete behaviour, promoting the use of industrial waste in sustainable construction practices.

Author contributions

Laxmi Narayana Pasupuleti: Writing, original draft preparation, conceptualization, methodology, formal analysis, review, and final version.

Bhasker Rao Nalli: Computer programing, literature review, data analysis, review, and final version.

Ajay kumar Danikonda: Computer programing, data analysis, review, and final version.

Ragubabu uppara^: Experimental, data acquisition, writing, editing, methodology, Investigation, review, and final version.

Ramakrishna Mallidi: Writing, editing, methodology, Investigation, review, and final version.

Conflict of Interest

On behalf of all authors, the corresponding author states that there is no conflict of interest.

Data Availability Statement

No associated data, all the data mentioned in this paper are in the form of figures and tables.

Competing Interests

The authors have no relevant financial or non-financial interests to disclose.

Funding

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript

Adesanya, D. A. (1996). Evaluation of blended cement mortar, concrete and stabilized earth made from ordinary Portland cement and corn cob ash. Construction and Building Materials, 10(6), 451–456.
Alhassan, Y., & Ballim, Y. (2017). Experimental study on carbonation of plain and blended cement concrete. International Journal of Scientific & Technology Research, 6(08).
Aly, S. T., El-Dieb, A. S., & Taha, M. R. (2019). Effect of high-volume ceramic waste powder as partial cement replacement on fresh and compressive strength of self-compacting concrete. Journal of Materials in Civil Engineering.
Arora, R., Kumar, K., & Dixit, S. (2024). Comparative analysis of the influence of partial replacement of cement with supplementing cementitious materials in sustainable concrete using machine learning approach. Asian Journal of Civil Engineering, 25, 1517–1530.
Beladi, A. S. E., Azzouz, L., Kadri, E., & Kenai, S. (2012). Effect of natural pozzolana and marble powder on the properties of self-compacting concrete. Construction and Building Materials, 31, 251–257.
Bilim, C., Atiş, C. D., Tanyildizi, H., & Karahan, O. (2009). Predicting the compressive strength of ground granulated blast furnace slag concrete using artificial neural network. Advances in Engineering Software, 40(5), 334–340.
Boddepalli, M., Pasupuleti, L. N., & Nalli, B. R. (2024). A comprehensive study of floodplain analysis utilising HEC-HMS, HEC-RAS, and GIS on the Kosasthalaiyar River sub-basin. Water Practice & Technology, 19(11), 4612–4628.
Cheng, Y., Huang, F., Li, G.-l., Xu, L., & Hou, J. (2014). Test research on effects of ceramic polishing powder on carbonation and sulphate-corrosion resistance of concrete. Construction and Building Materials, 55, 440–446.
El-Hassan, H., & Kianmehr, P. (2016). Pervious concrete pavement incorporating GGBS to alleviate pavement runoff and improve urban sustainability. Road Materials and Pavement Design, 19(1), 1–15.
Fu, S., & Lee, J. (2024). Recycling of ceramic tile waste into construction materials. Developments in the Built Environment, 18, Article 100431.
Glasser, F. P. (1994). Immobilisation potential of cementitious materials. In Studies in Environmental Science, 60, 77–86.
Gopal, P., & Reddy, G. M. (2017). An investigational study on strength characteristics of concrete in rigid pavements by partial replacement of cement with GGBS. SSRG International Journal of Civil Engineering (SSRG-IJCE), 4.
Kannan, D. M., Aboubakr, S. H., El-Dieb, A. S., & Taha, M. M. R. (2017). High performance concrete incorporating ceramic waste powder as large partial replacement of Portland cement. Construction and Building Materials, 144, 35–41.
Kaveh, A. (2024). Applications of artificial neural networks and machine learning in civil engineering (Studies in Computational Intelligence, Vol. 1168). Springer.
Kaveh, A., & Khavaninzadeh, N. (2023). Efficient training of two ANNs using four meta-heuristic algorithms for predicting the FRP strength. Structures, 52, 256–272.
Keramatikerman, M., Chegenizadeh, A., & Nikraz, H. (2016). Effect of GGBFS and lime binders on the engineering properties of clay. Applied Clay Science.
Kore, S. D., & Vyas, A. (2016). Durability of concrete using marble mining waste. Journal of Building Materials and Structures.
Kumar, A. K. B. G. (2012). Effective utilization of fly ash and supplementary cementitious materials in construction industry. International Journal of Civil Engineering and Technology.
Kumar, K., Arora, R., & Tipu, R. K. (2024). Influence of machine learning approaches for partial replacement of cement content through waste in construction sector. Asian Journal of Civil Engineering, 25, 3201–3213.
Kumar, S., Kumar, R., Bandopadhyay, A., Alex, T. C., Kumar, B. R., Das, S. K., & Mehrotra, S. P. (2008). Mechanical activation of granulated blast furnace slag and its effect on the properties and structure of portland slag cement. Cement & Concrete Composites, 30, 679–685.
Nehring, L. H. P. V., Paiva, F. F. G. de, Tamashiro, J. R., Galvín, A. P., Kinoshita, A., & López-Uceda, A. (2023). Use of blast furnace slag in cementitious materials for pavements: Systematic literature review and eco-efficiency. Sustainable Chemistry and Pharmacy, 33, 101030.
Nwankwojike, B. N., Onwuka, O. S., & Ndukwe, E. C. (2014). An appraisal of different brands of Portland cement in Umuahia industrial market, Nigeria. Journal of Research Information in Civil Engineering, 11(2), 577–589.
Rahla, K. M., Mateus, R., & Bragança, L. (2019). Comparative sustainability assessment of binary blended concretes using supplementary cementitious materials (SCMs) and ordinary Portland cement (OPC). Journal of Cleaner Production, 220, 445–459.
Raheli, B., Aalami, M. T., El-Shafie, A., Ghorbani, M. A., & Deo, R. C. (2017). Uncertainty assessment of the multilayer perceptron (MLP) neural network model with implementation of the novel hybrid MLP-FFA method for prediction of biochemical oxygen demand and dissolved oxygen: A case study of Langat River. Environmental Earth Sciences, 76, Article 1–16.
Rawat, S., Zhang, Y. X., Fanna, J., & Lee, C. K. (2024). Development of sustainable engineered cementitious composite with enhanced compressive performance at elevated temperatures using high-volume GGBFS. Journal of Cleaner Production, 451, Article 142011.
Shinde, V. B., & Kumar, R. (2023). Experimental and analytical study of results of blended pavement concrete models tested in laboratory and finite element method. Proceedings on Engineering Sciences, 5(2), 259–266.
Taud, H., & Mas, J. F. (2017). Multilayer perceptron (MLP). In Geomatic approaches for modeling land change scenarios (pp. 451–455). Springer International Publishing.
Tipu, R. K., Panchal, V. R., & Pandya, K. S. (2024). Machine learning-based prediction of concrete strengths with coconut shell as partial coarse aggregate replacement: A comprehensive analysis and sensitivity study. Asian Journal of Civil Engineering, 25, 3183–3200.
Tokareva, A., Kaassamani, S., & Waldmann, D. (2024). Using ceramic demolition wastes for CO₂-reduced cement production. Construction and Building Materials, 426, Article 135980.
Vasudeva, R. (2023). Influence of reclaimed asphalt pavement aggregates on strength and durability properties of concrete mixes in rigid pavements. The Journal of Engineering Research, 20(1), 1–11.
Vivek, S. S., & Dhinakaran, G. (2017). Fresh and hardened properties of binary blend high strength self-compacting concrete. Engineering Science and Technology, an International Journal, 20, 1173–1179.
Yaragal, S. C., Gowda, S. B., & Rajasekaran, C. (2019). Characterization and performance of processed lateritic fine aggregates in cement mortars and concretes. Construction and Building Materials, 200, 10–25.
Yaswanth, K. K., Reddy, K. H. K., Anusha, N., Praveen, B., Chippymol, J., Revathy, J., & Isleem, H. F. (2024). Engineered geopolymer composites: A comprehensive state-of-the-art review on materials’ perspective. Archives of Civil and Mechanical Engineering, 24(3), 193.

No competing interests reported.

Prediction of concrete strength using Multilayer Perceptron Neural Network-based utilizing sustainable waste materials

Status:

Version 1

Abstract

Figures

Introduction

Methodology

Materials and Methods

Deign and Mix Proportions

Mix designation of replacement of VPW and GGBFS

MLP model

Input and output parameters

Results and discussion

Conclusions

Declarations

References

Additional Declarations

Status:

Version 1