1 Introduction

Developing new materials in the construction sector and trying to find more economical and environmental materials are among the current studies. Mineral additives are used in concrete in order to improve some properties of concrete or to give concrete special qualities [17].

These mineral additives used in concrete are called supplementary cementitious materials (SCMs) and increase the durability and strength of concrete [27]. The widespread use of SCMs and their use in concrete mixtures have shown great improvement. SCMs are fine-grained mineral additives with pozzolanic properties. The most widely used among these materials are silica fume and fly ash [59]. These materials are economical as they are locally sourced.

In the world, the most greenhouse gas formation is caused by energy-related emissions. Many studies are carried out to minimize the effect of these emissions, resulting from human-induced activities [25, 28]. One of the factors causing the highest greenhouse gas emissions is cement production. Saving cement is undoubtedly an important parameter in terms of economy, as well as an important detail in terms of the environment, as less carbon will be released to nature in parallel with less cement production [21, 24]. A reduction in the amount of cement by ignoring all the possibilities that will create a negative situation may create a physically more void concrete structure, contrary to the situation we want. Undoubtedly, a void concrete is not only unprotected against harmful outdoor conditions, but it can also create negative situations in terms of durability and strength [51, 52]. Therefore, the reduced cement must be replaced by an additive by volume or by weight [36, 37, 49].

1.1 Prior art

Mineral additives are used together with portland cement clinker or portland cement in the production of long-lasting reinforced concrete structures that are more resistant to chemical, electro-chemical and physical external effects in our country and in the world [45]. Fly ash and silica fume are commonly used mineral additives used by replacing cement [7, 11]. In addition to those additive, metakaolin is also used. Metakaolin is a white and soft clayey material formed as a result of the separation of clay minerals and feldspar by natural methods [43]. This substance, which is frequently seen in nature, is mostly used in porcelain production, jean and textile manufacturing. The most important mineral forming the essence of kaolin is kaolinite (Al2Si2O5(OH)4). Kaolinite is a clay mineral composed of aluminum hydrosilicate composition [14, 15]. Metakaolin is a highly fine reactive alumino-silicate pozzolan obtained by firing and then grinding purified kaolin or kaolinite clays at a certain temperature range [26]. The kaolinite used in making ceramics releases its hygroscopic water when heated below 200 °C. It turns into metakaolin by leaving the bond water in its chemical formula between 500 and 600 °C. The new substance formed in this way constitutes the basic element of metakaolin [33, 43]. Metakaolin can react with calcium hydroxide to form hydrates. This contributes to the improvement of the physical and chemical properties of both mortar and concrete. How much of the metakaolin will react depends on the mineral it contains, the raw kaolin source, and increases of production [51].

In 1962, for the first time in the construction of Jupia Dam in Brazil, concretes obtained as a result of substitution of metakaolin and portland cement were used to increase durability. After this study, it was observed that there was a significant increase in the use of metakaolin in the formation of cement and concrete. Today, in addition to these uses, it is used in mortar and high-performance concrete. Most of the researches on this subject so far have focused on determining the optimum amount of metakaolin suitable for use in concrete. Studies on the use of metakaolin as a substitute with portland cement and the effects of metakaolin added to concrete in different proportions on the properties of concrete continue today. In addition to these, the reactivity level of the use of metakaolin with C3S is also experimentally investigated. Such studies have provided us with general information about the importance of the use of metakaolin in mortar and concrete. It has been observed that metakaolin used in mortar and concrete has an effect on many of the properties of fresh and hardened mortar and concrete [18, 57, 58]

One of the most important environmental factors affecting concrete negatively is the elevated temperature effect [31, 32, 35]. Exposure of buildings to elevated temperatures poses a great risk. Exposure of concrete to elevated temperature causes physical and chemical deterioration such as dehydration of cement paste, loss of strength and mass and modulus of elasticity [10, 38]. Thus, undesirable structural damages occur in the structure by negatively affecting the mechanical properties of the concrete. The degree to which concrete is affected by temperature varies according to the components it contains [14, 15, 60]. According to the thermal conductivity of the components used in concrete at elevated temperatures, deterioration and fragmentation occur. In order to overcome this situation with minimum damage, concrete mixtures with high strength and durability and low degree of exposure to heat should be developed [2, 20].

Structures can be exposed to elevated temperatures. The use of the concrete after this exposure, the degree of damage and its serviceability depend on the performance remaining in the concrete. Therefore, knowing the performance of concrete before and after elevated temperatures is of great importance [41]. The performance of concrete before and after elevated temperature is also determined by various tests [40]. These tests are performed in a destructive–non-destructive way [2, 18]. As described, measurement results are obtained by various experiments. Tests take a long time. In addition, materials with different properties and ratios are used for each test, and at the same time in each experimental study, the test equipment used wears out. Intelligent systems can be used to prevent time loss, to eliminate material waste and to prevent wear that may occur in equipment.

Estimation of concrete compressive strength is made by using the results obtained with different test methods in determining concrete compressive strength [5, 13]. Regarding prediction, some studies in the literature have also used artificial intelligence methods to determine concrete compressive strength. In one of the studies, non-destructive testing methods were used and the results obtained from the test methods were used using artificial neural network, support vector machine and adaptive neuro-fuzzy inference systems [16]. Similarly, the data obtained using ultrasound transmission rate and Schmidt hammer test were also used in the artificial neural network and concrete compressive strength was estimated [6]. In a different concrete compressive strength prediction study using compressive strength data taken from concrete samples cured for 7 and 28 days, comparison was made using artificial neural network, decision tree, support vector machine and linear regression [29]. The artificial neural network method was also used to estimate the compressive strength of concrete exposed to high temperatures [5].

For this goal, many intelligent approaches such as artificial neural network (ANN), fuzzy logic, genetic algorithm, particle swarm optimization, etc., are used for the evaluation of concrete performance [47, 48].

1.2 Contribution of the study

In this paper, the performance of concrete exposed to high temperatures was evaluated using ANN with destructive and non-destructive tests. In addition, unlike other studies, more than one SCM was used in concrete production and the effects of these admixtures on concrete were investigated. Therefore, this study aims to ensure that the SCMs used within the scope of the experimental study are recycled and reused in line with the goals of sustainable development and circular economy by considering various SCMs, reducing the consumption of natural resources and evaluating alternative uses of SCMs in a holistic approach. Thus, the use of SCMs in concrete production will provide significant gains in terms of sustainable development and circular economy.

In the experimental study phase, as a result of the literature research, 22 concrete series were prepared, one of which was a control series, containing 10% silica fume in its main matrix. In these series, 5–10–15–20–25–30–35% fly ash and 10–15–20% metakaolin were used. The proportions of mineral additives were determined as a result of the pozzolanic activity experiments. The mechanical properties of the prepared concrete series before and after elevated temperature were investigated. Before the elevated temperature, pressure and tensile strength, porosity and ultrasonic pulse velocity tests were applied to the samples at the end of the 7th-, 28th- and 90th-day curing periods. In order to determine the strength losses after elevated temperature and to observe the change in the internal structure of the concrete, the 28-day samples were exposed to 400, 600, and 800 °C temperatures in a high-temperature furnace. In parallel with the results obtained from the tests, it was observed that the combined effect of 10% fly ash additive and 20% metakaolin additive increased the strength. It was seen that ultrasonic pulse velocity and porosity experiments also supported this result. When the compressive strength values after elevated temperature were examined, strength loss was observed in all series depending on the temperature increase, while the series showing the highest strength differed depending on the temperature.

By using waste materials in concrete mixtures, environmental pollution was reduced, thus providing significant gains in terms of both environment and cost. The amount of cement has been reduced by the use of waste materials in concrete production. Therefore, more environmentally friendly and economical concrete production is ensured by providing energy savings.

After experimental study, using the data get from the experimental study, compressive strength results for the 7th, 28th and 90th days were obtained using ANN. Thus, it is aimed to estimate the compressive strength with ANN for different materials and different ratios without the need for experimental work. By using ANN, the time required for experimental work will be eliminated, that is, it will not be necessary to wait 7, 28, and 90 days for compressive strength determination. Moreover, by using ANN, the compressive strengths of concretes with known mixing ratios at different ages can be quickly estimated without prior experimentation. Thus, advantages are gained in terms of time, cost and labor.

2 Materials and methods

In this study, CEM I 42.5 N portland cement, which was supplied from the Elazig cement factory and complied with the TS EN 197–1 standard, was used as the binding material [53]. Metakaolin supplied from Mikrons Kimya Ltd. Şti., as mineral additives and fly ash from Çelikler A.Ş., and silica fume obtained from Antalya Eti Elektrometallurgy were used by replacing the cement in the specified proportions by volume [56]. Chemical and physical properties of cement and mineral additives used are given in Table 1.

Table 1 Physical and chemical properties of cement and mineral additives (%)
Full size table

2.1 Preparation and properties of concrete samples

In the first stage of the study, many trial mixtures were prepared in the laboratory using studies in the literature [26]. In the literature, various mixtures consisting of only metakaolin, methacholine and certain proportions of silica fume or metakaolin and fly ash have been widely used [1, 33]. In this study, unlike the mixtures used in the literature, two SCMs were used together with metakaolin. Tests were carried out on the trial samples obtained as a result of the prepared mixtures. Reference mixtures were determined based on the appropriate data received as a result of these tests. The parameters taken into consideration during the analysis phase of the study included water/cement (W/C), aggregate/cement (A/C), metakaolin/cement (M/C) and fly ash/cement (FA/C). The purpose of determining these parameters was to investigate in detail the effects of these factors on the early age and final age strength of concrete.

In the second stage of the study, ten concrete series were prepared, one of which was the control series, with a wide range of fly ash ratios of 10–20–30% and metakaolin ratios of 10–15–20%. Then, the series were prepared by determining the fly ash ratios in a narrower range in order to determine the change between the ratios exactly and to get higher accuracy results in the ANN program. In this way, a total of 22 concrete series, one of which is a control series, were prepared. The data of the mixtures are given in Table 2. Mixture names were determined, for example, CSF10, control series containing 10% silica fume, FA10M10 series containing 10% fly ash and 10% metakaolin.

Table 2 Proportions of the concrete mixtures (kg/m3)
Full size table

Cement amount was taken as 370 kg/m3 in all mixtures. The produced samples were placed in molds of 100 × 100x100 mm and kept in the laboratory for 24 h. Afterward, the cured samples were taken out of the molds and left in the curing pool. In order to determine the hardened concrete properties, the samples were removed from the curing pool, and standard pressure and tensile strength, porosity and ultrasonic pulse velocity tests were applied to the samples on the 7th, 28th and 90th days before the elevated temperature. Compressive strength and porosity tests were performed at 400, 600, and 800 °C temperatures on the 28th day after elevated temperature. The flow chart of the study is presented in Fig. 1.

Fig. 1
figure 1

Experimental program of present study

Full size image

2.2 Artificial neural network (ANN)

Artificial intelligence is a field of science that aims to enable computers to perform functions such as learning, problem solving and decision making better by imitating human intelligence [30, 42]. Artificial intelligence is used in many areas such as health, military, engineering, etc. There are different artificial intelligence algorithms according to their technical features and purposes of use. ANN, linear regression (LR), support vector machines (SVM), and gaussian process regression (GPR) are among the commonly used artificial intelligence models. In this study, the data obtained as a result of the experimental study were used in various artificial intelligence algorithms, and the study was continued with artificial neural networks that gave the best results.

ANN is a smart technique which simulates biological nerves and works similar to biological nerves [8]. ANN method reads the input data and provide optimal decisions using the predefined conditions. ANN is used for solving nonlinear problems to provide a practical, easy and fast solutions [22]. Because of these properties, the ANN is becoming more popular in all engineering applications including civil engineering [44, 46, 50].

3 Experiments and analysis

3.1 Pozzolanic activity experiment

In order for the pozzolanic material to show sufficient activity, it must have a fine-grained amorphous structure and contain a sufficient amount of "silica + alumina + ferroxide." When the pozzolans are ground to a finer particle size and mixed with slaked lime and water, they enter into a chemical reaction [12]. As a result of this chemical reaction, it reacts with the calcium-hydrate (C–H) released during the hydration of Portland cement and forms additional calcium-silica-hydrate (C–S–H) gels [9, 34]. In the study, pozzolanic activity test was carried out using Eq. (1) in accordance with ASTM C 311 standard [9]. When Fig. 2 is examined, it is seen that the results of the mortar samples with mineral additives are in accordance with the standard. When the pozzolanic activity results of all three mineral additives used in the experiments are examined, it is concluded that all three can be used as pozzolans.

$$Strength\;Activity\;Index = (A/B) \times 100$$
(1)
Fig. 2
figure 2

Strength activity ındex of mineral additives

Full size image

A: Average compressive strength of mortar samples with additives.

B: Average compressive strength of control mortar samples.

3.2 Compressive and tensile strength tests

The compressive strength test was applied at a loading speed of 3 kN/sec, as specified in the TS EN 12390–3 standard [55]. The series, which completed the curing periods of 7, 28, 90 days, were tested in the Autotest 3000 hydraulic load-controlled pressure strength press (Fig. 3a).

Fig. 3
figure 3

Compressive and Tensile Strength Tests

Full size image

In the tensile strength test, as specified in the TS EN 12390–6 standard, plates were placed above and below the concrete samples and compressive loading was applied to the samples in the vertical direction [54]. Three samples from each series were subjected to tensile tests, the read values were written in Eq. (2), the tensile stresses were found and their averages were recorded (Fig. 3b).

$${\text{f}}_{{{\text{ctd}}}} { } = \frac{{2{\text{F}}}}{{{\uppi }.{\text{L}}.{\text{d}}}}$$
(2)

In the formula, fctd, indicates split tensile strength, F denotes maximum load (N), L is the length of the contact line of the sample to the loading part (m) and d is the selected cross-sectional dimension of the sample (m).

3.3 Ultrasonic pulse velocity tests

Ultrasonic pulse velocity test, as specified in the TS EN 12504–4 standard, was calculated according to Eq. (3) by measuring the passage time of ultrasonic waves sent into the concrete from one surface of the concrete to the other.

$$V = \frac{S}{t}*10^{6}$$
(3)

Here, V (m/s) is the ultrasonic wave velocity, S (m) is the distance between the surface of the concrete block where the ultrasonic wave is sent and the surface where the wave is received and t (μs) is the distance from the concrete surface where the ultrasonic wave is sent to the surface where it is received time (TSE, 2004).

3.4 Elevated Temperature Test

In elevated temperature test, 1200 °C capacity, heating rate 2.5 °C/min. Protherm HLF 150 branded laboratory oven was used (Fig. 4). Before the samples were placed in the oven, they were brought to a constant weight at 100 ± 5 °C in the oven. Then, after the furnace reached temperatures of 400, 600 and 800 °C, the samples were exposed to these temperatures for 1 h. After 1 h, the oven was automatically turned off, and when it came to room temperature, they were taken out of the oven and subjected to a compressive strength test.

Fig. 4
figure 4

Protherm HLF 150 Laboratory-Type Oven

Full size image

3.5 ANN

The architecture of ANN has inputs, outputs and hidden layers. The connection between the input and output layers is provided by hidden layers. The information in the input layer, where external data is entered, is processed by the hidden neurons in the hidden layer and sent to the output layer. In other words, the hidden layer has an important place on the information sent to the output layer. If the network structure is complex, then the number of hidden layers can be increased. However, increasing the number of hidden layers will increase the learning capacity, complexity and running time of the network [30]. Online modulation recognition of analog communication signals using neural network. Expert Systems with Applications, 33(1), 206–214. Therefore, the number of hidden layers is determined depending on the network structure. In this study, the desired results were obtained with a single hidden layer due to the network structure.

A percentage of the samples needed to train the network are taken from input data for training, testing and, validation. The samples are trained using different techniques. The ANN system provides the outputs for corresponding inputs depending on the conditions [3, 39].

The input data obtained from the experimental tests are stored in the look-up tables for use in the ANN system. W/C, A/C, M/C and FA/C are used as inputs. Two set of outputs are obtained from the ANN. One set is for concrete performance before elevated temperature exposion on the 7th, 28th and 90th day. The other set of output is for concrete performance after exposed to elevated temperatures of 400, 600 and 800 °C on the 28th day. Consequently, depending on given input values, corresponding concrete strength values are calculated by using the ANN method. In the both ANN system, four inputs, 20 hidden neurons and three outputs are used as given in Fig. 5.

Fig. 5
figure 5

ANN diagram used in system

Full size image

4 Results and discussion

4.1 Evaluation of tests before and after elevated temperature

All the test results before and after elevated temperature are reported in Table 3. In this table, UPV, fctd and porosity are results at 28th days. Furthermore, compressive strength are results at 7th, 28th and 90th days, respectively. Compressive strength test was applied at 400, 600 and 800 °C temperatures on the 28th day after elevated temperature.

Table 3 Results of tested specimens before and after elevated temperature
Full size table

4.1.1 Evaluation of UPV-porosity tests results

In UPV and porosity test results, the speed of sound waves passing through the sample varies according to the usage rates of M and FA. In order to make a complete inference, when the FA ratio is kept constant and the increase–decrease in the velocity and porosity value with the change of the M ratio, in the FA10 series, the speed value increases with the increase of the M ratio. Therefore, the amount of void decreases. However, M15 and M20, the use of M ratio after 15% caused a decrease in the velocity value (Fig. 6). Metakaolin has positive properties such as accelerating hydration in concrete and filling the voids due to its thin material [51]. However, in order for the pozzolanic material M to be hydrated with cement, sufficient cement and water must be present in the environment. Therefore, as seen in Fig. 6, the use of M ratio more than 15% cannot form sufficient C–S–H gels [1]. In this case, void concrete is obtained. Therefore, the passage time of the sound waves through the sample is prolonged. Likewise, since the FA ratio is a pozzolanic material, this is also true if the FA ratio increases.

Fig. 6
figure 6

Comparison of UPV and porosity test results

Full size image

When Fig. 7 is examined, as the FA ratio increased, the porosity values of the series increased in the majority. The porosity change of FA before elevated temperature was also observed after elevated temperature too. On the other hand, as the M ratio increased, the porosity values of the series decreased.

Fig. 7
figure 7

Comparison of porosity test results before and after elevated temperature

Full size image

4.1.2 Evaluation of compressive strength tests results

Samples exposed to high temperatures decreases strength under increasing temperatures [2]. When Fig. 8 is examined, except for the control series without FA and M, It is seen that the compressive strength of the series applied at 400 °C is generally higher than the series at 20 °C. Similarly, in previous studies, an increase in strength was observed in the strength of mineral-added series up to 350 °C [19]. Dvorkin et al. have investigated the strength development of concrete up to 40 °C [23]. This increase in strength has been explained in the literature as follows. With the increasing temperature, the free water evaporated and the cement gel layers started to move closer to each other. Increasing Van der Waal's forces between these gel layers caused an increase in strength [36].

Fig. 8
figure 8

Comparison of compressive strength test results before and after elevated temperature

Full size image

While the FA5M20 series had the highest compressive strength before elevated temperature, the FA5M15 series gave the highest compressive strength value regardless of the temperature value in the series exposed to elevated temperature. In order to obtain higher strengths in structures that will be exposed to elevated temperatures, it has been concluded that the use of 15% M is sufficient in mineral additive concrete mixtures to be prepared.

4.2 Structures of ANN

In this paper, 22 experimental data were used for each of four inputs. Sixteen samples of the data (70%) were used for training, three samples (15%) were used for validation and three samples (15%) were used for testing. The Bayesian regularization algorithm was chosen for training ANN.

The architecture of ANN system for concrete performance before elevated temperature exposion on the 7th, 28th and 90th days used is given in Fig. 9.

Fig. 9
figure 9

The ANN architecture for concrete performance before elevated temperature

Full size image

Figure 10 presents the Simulink block diagram for ANN results before exposion of elevated temperature for given input sample values.

Fig. 10
figure 10

ANN outputs before elevated temperature exposion for sample input values

Full size image

The performance of ANN is evaluated using mean squared error metric. The regression (R) values represent the similarity ratio of the actual data and system outputs. The R value is expected to be 1 for better performance. Because of the performance evaluated from the ANN outputs is very similar to that of the experimental data, R values tend to be around 1.

Figure 11 provides R values for training, test and all data sets. As the results of ANN and the training data obtained with experimental results are analyzed, it is seen that the data obtained by ANN are quite similar with experimental results, where 0.999, 0.995 and 0.998 are obtained for R values for training, test and all data, respectively.

Fig. 11
figure 11

The R values for training, test and all datasets before elevated temperature

Full size image

The architecture of ANN system for concrete performance after elevated temperature exposion on the 28th day is used presented in Fig. 12.

Fig. 12
figure 12

The ANN architecture for concrete performance after exposed to elevated temperatures

Full size image

Figure 13 shows the Simulink block diagram for ANN results after exposion of elevated temperatures of 400, 600 and 800 °C on the 28th day for given input sample values.

Fig. 13
figure 13

ANN outputs after elevated temperature exposion for sample input values

Full size image

Figure 14 indicates R values for training, test and all data sets after elevated temperature exposion. It has been observed that the data taken from the output of ANN are very similar with the experimental results, where 0.999, 0.997 and 0.999 are obtained for R values for training, test and all data, respectively.

Fig. 14
figure 14

The R values for training, test and all datasets after exposed to elevated temperatures

Full size image

To see the effectiveness of ANN, sample experimental input values are taken and the corresponding ANN output values are obtained and tabulated for comparison purpose in Table 4. As Table 4 is analyzed, it can be observed that the ANN gives the closest results to the experimental results both before and after exposed to elevated temperatures.

Table 4 The comparison of experimental and ANN outputs
Full size table

To prove the predictive performance of ANN, a comparative analysis was made about various artificial intelligent models, including ANN, GPR, SVM and LR. RMSE values were used in the analysis to determine the algorithm that gave the best results. The closer the RMSE value is to zero, the smaller the difference between the actual and predicted values. When the Table 5 is examined, it is seen that the RMSE value of the result obtained with the ANN model is the smallest value.

Table 5 Comparison of RMSE values of different artificial intelligent models
Full size table

The issue of how generalizable the model can be for real-world data and how reliable it is important. Therefore, cross-validation strategy is used [4]. With cross-validation, in addition to preventing a model from being trained in a data-dependent manner, it is also possible to evaluate the generalizability of the model.

In the cross-validation method, the data set is divided into small subsets and each subset of the data set is used as the test set in turn, while the other subsets are used as the training set. The model is trained on the training set and then evaluated on the test set. This step is performed at each iteration of cross-validation. The performance parameters obtained at the end of each cross-validation iteration are collected and averaged to evaluate the overall performance of the model.

In this study, a fivefold cross-validation strategy was used to objectively evaluate the performance of the model. Accordingly, RMSE and R values of the model were obtained and tabulated in Table 6.

Table 6 RMSE and R values of model outputs using fivefold cross-validation
Full size table

Additionally, the response plot showing the best error rate between the prediction data obtained at the model output using the fivefold cross-validation method and the actual values is given in Fig. 15.

Fig. 15
figure 15

Response plot of fivefold cross-validation

Full size image

5 Conclusion

In this study, the effect of the use of SF, FA and M in different proportions in concrete mixtures with fixed proportions on the mechanical properties of concrete before and after elevated temperature was investigated. In addition, after the experimental study, the performance evaluation of the concrete was made using the ANN. In the literature, different studies have been found to examine the mechanical properties of concrete under high temperature in which the mineral additives are replaced by cement alone, but there have been no study on the use of more than one mineral admixture. In current studies, it has been shown that the use of FA in concrete mixes greatly affects the mechanical properties of concrete. However, it has been observed that M is more effective than FA additive in the use of fixed ratio of SF and FA and M mineral additives together. In addition, it has been observed that the use of M up to a certain amount in the concrete mixture under the effect of high temperature has a positive effect on the strength of the concrete.

In the use of FA at different rates, the best UPV value was seen in the series with a FA ratio of 5%. It was concluded that the amount of FA should be increased in a controlled manner when the addition of M and its ratio are increased. In the use of M, the best UPV result was obtained at 20% M.

When evaluated in terms of compressive strength results, the highest compressive strength value at each cure age was obtained in the FA5M20 series. However, in the FA30 series, the M10-added series showed the highest strength, especially on the 28th and 90th days. It was concluded that when the amount of FA is used at the rate of 30%, it is not necessary to use the amount of M at the rate of 20%.

It is seen that there is a decrease in the strengths with the increase in the amount of FA at every temperature. Based on this judgment, regardless of the M ratio, it is seen that the optimum amount of FA in the prepared series will be 5%. In order to obtain higher strengths in structures that will be exposed to elevated temperatures, it has been concluded that the use of 15% M is sufficient in mineral additive concrete mixtures to be prepared.

In the study instead of making many experimental test, ANN is proposed for the evaluation of concrete performance to prevent material waste and time consume. The concrete performance obtained from the ANN outputs is very similar to the experimental values both before and after elevated temperatures exposion. The ANN and experimental results prove the effectiveness of the ANN-based concrete performance evaluation.

Finally, the aim of the study is to apply destructive and non-destructive tests to determine concrete performance, evaluate the data obtained as a result of these tests in an artificial neural network and develop a mathematical model. For this purpose, different proportions of SCM were used in concrete mixtures. While determining SCM rates, the literature studies were examined in the first stage. Afterward, many trial mixtures were poured. Finally, the reference mixture was determined.

Although this study does not have any limitations, it can be enriched with other studies with intermediate mineral contribution rates and intermediate temperature values. Similarly, the study can be expanded by increasing the number of concrete series or varying the concrete mixture ratios. Expanding the study means more data. Since data entry is important in predictions made using ANN, the more data input is made, the accuracy of the prediction will increase at the same rate. Additionally, deep network architectures can be used to increase the prediction performance obtained as a result of different mixtures used. The holistic approach realized within the scope of this study has shown that the use of SCMs in concrete production will provide significant gains, considering the increasing use of concrete in the construction sector with the developing material technology, the increase in cement and aggregate consumption, sustainable development and circular economy goals.