Optimal Compressive Strength of RHA Ultra-High-Performance Lightweight Concrete (UHPLC) and Its Environmental Performance Using Life Cycle Assessment

Frequent laboratory needs during the production of concrete for infrastructure development purposes are a factor of serious concern for sustainable development. In order to overcome this trend, an intelligent forecast of the concrete properties based on multiple data points collected from various concrete mixes produced and cured under different conditions is adopted. It is equally important to consider the impact of the concrete components in this attempt to take care of the environmental risks involved in this production. In this work, 192 mixes of an ultra-high-performance lightweight concrete (UHPLC) were collected from literature representing different mixes cured under different periods and laboratory conditions. These mix proportions constitute measured variables, which are curing age (A), cement content (C), fine aggregate (FAg), plasticizer (PL), and rice husk ash (RHA). The studied concrete property was the unconfined compressive strength (Fc). This exercise was necessary to reduce multiple dependence on laboratory examinations by proposing concrete strength equations. First, the life cycle assessment evaluation was conducted on the rice husk ash-based UHPLC, and the results from the 192 mixes show that the C-783 mix (87 kg/m 3 RHA) has the highest score on the environmental performance evaluation, while C-300 (75 kg/m 3 RHA) with life cycle indices of 289.85 kg CO 2eq . Global warming potential (GWP), 0.66 kg SO 2eq . Terrestrial acidification and 5.77 m 3 water consumption was selected to be the optimal choice due to its good profile in the LCA and the Fc associated with the mix. Second, intelligent predictions were conducted by using six algorithms (ANN-BP), (ANN-GRG), (ANN-GA), (GP), (EPR), and (GMDH-Combi). The results show that (ANN-BP) with performance indices of R; 0.989, R 2 ; 0.979, mean square error (MSE); 2252.55, root mean squared error (RMSE); 42.46 MPa and mean absolute percentage error (MAPE); 4.95% outclassed the other five techniques and is selected as the decisive model. However, it also compared well and outclassed previous models, which had used gene expression programming (GEP) and random forest regression (RFR) and achieved R 2 of 0.96 and 0.91, respectively.


Introduction
Concrete is the most commonly used building material in the construction sector due to its many advantages, including being readily available in all parts of the world, being inexpensive, and having good mechanical qualities [1][2][3][4]. The building sector consumes the majority of raw materials [5]. Cement is utilized as a binder ingredient in concrete and is responsible for the construction industry's carbon emissions [6,7]. In other words, concrete manufacturing has a significant environmental impact, and environmental issues in the building sector are well known [8][9][10][11].
Researchers all across the world are working to reduce the amount of cement used in concrete in order to enhance its engineering characteristics. To partially decrease the quantity of cement in concrete, the various available alternate binding materials are blast furnace slag (BFS) [12], fly ash [13,14], rice husk ash [15], plastic powder [16], bagasse ash [16], etc. The supplementary materials are sustainable for nature and also lower the overall cost of construction [17]. However, these novel materials must be assessed in terms of cost, quality, performance, and environmental implications. To attain this aim, a well-known method called life cycle assessment (LCA) method may be used to study the impact of a material or methodology on the environment [18,19]. LCA permits us to balance both the energy and material usage while quantifying total environmental implications [20,21]. LCA is a potentially viable method for evaluating the environmental performance of construction materials [22][23][24][25][26][27].
Due to developments in mechanical and durability qualities with compressive strengths greater than 150 MPa, Ultra High-Performance Concrete (UHPC) is now considered the most recent breakthrough in the world of concrete technology [28][29][30]. Mineral admixtures, such as rice husk ash, are natural pozzolanic materials with high amorphous silica content and highly fine texture, which can be a UHPC ingredient like silica fume, but in developing nations the cost of silica fume is very high [31][32][33][34][35][36]. It is now critical for researchers and construction workers to identify an alternate material with equivalent properties to use in the manufacturing of UHPC. Silica fume has comparatively large particle sizes that are beyond 10 µm and is also extremely tough to break due to its strong inter-atomic interactions (Van der Waal's forces), even in the existence of moisture. In terms of technical and environmental impact evaluation at a cheaper cost, RHA can be an excellent option, and it can significantly improve its mechanical and durability qualities [37][38][39]. RHA has almost similar chemical composition to SF [40], but the difference in particle size makes it particularly effective in the hydration development and improving the microstructure of the paste of cement [41,42].
Rice husk (RH) is the hard external layer of paddy rice that separates rice grains during the milling process. It accounts for roughly 20-30% of the weight of yearly rice output, which was anticipated to reach 760 million tons in 2017 and is probable to grow by 2050. Nearly 150 million tons (18-25%) of milled material are disposed of in landfills or waterways, posing environmental concerns like pollution [43,44]. During the burning process, it loses approximately 80% of its weight and changes to RHA since it comprises 20% inorganic and 80% organic components [45]. It has been proven to be a natural supply of silica as well as having faults that may be addressed with citric acid to do the purification before burning, so contributing to the production of sustainable concrete [46]. With a high silica content, high specific area, an amorphous silica crystallization phase, high pozzolanic activity, and a porous microstructure, the solid residual RHA has attracted a lot of interest due to its availability and affordability. As a result, it has the potential to be a useful supplemental cementitious material [47][48][49][50][51]. Using RHA instead of SF in concrete manufacturing has enhanced the concrete's characteristics while also reducing pollution and expense. This RHA can increase the hydration and mechanical characteristics of the concrete, in addition to its high durability, chemical attack resistance, and freeze-thaw resistance, according to a group of Chinese researchers [52].
Van Tuan et al. [53] investigated the feasibility of employing RHA to manufacture the UHPC. The consequences demonstrated that the mechanical strength of UHPC containing RHA may exceed 150 MPa under typical curing conditions. The fact that RHA had a greater influence on the enhancement of compressive strength in UHPC than SF is noteworthy. Furthermore, the sample containing the ternary cement mix with 10% SF and 10% RHA had a higher mechanical strength than the reference sample without SF or RHA. This combination proved to be the most effective solution for obtaining the maximum synergy. The goal of the Lu et al. study [54] was to use a mathematical technique to build a revolutionary lightweight UHPC (L-UHPC) with low density and great performance. The L-UHPC mix design included two lightweight ingredients (expanded shale aggregates and micro-sized glass microspheres). The L-UHPC was created using numerical optimization using a central composite design, which proved to be a successful strategy. The developed L-UHPC had functional and durability attributes that were equivalent to or even higher properties as compared to high-performance cement mortar (HPCM) and normal-weight UHPC.
Various studies on the LCA of cement and concrete have been conducted. Heede & Belie [55] shows the comparison between conventional concrete and sustainable concrete study the impact on the environmental. Authors used the "damage-oriented IMPACT 2002plus method" and "problem-oriented CML 2002" technique to assess the environmental consequences. Their findings showed that green concrete (concrete manufactured from furnace slag) pollutes the environment less than concrete created from Portland cement. According to the Liu et al. [56], environmental LCA of reactive powder concrete and conventional concrete on a dam building plan in China, RFC concrete had 64% and 55% lower greenhouse gas emissions and energy usage, respectively, than conventional concrete. The quantity of carbon dioxide was decreases mainly in four stages such as material production, transportation, construction, and operation and maintenance with values 72%, 25%, 51% and 15.6%, respectively.
Using SimaPro8 software, Cheung & Tait [57] investigated the cradle-to-gate (LCA) of concrete contains 100% cement (first case), 35% FA and 65% cement (2nd case), and 70% slag of the furnace and 30% cement (3rd case),and environmental impressions were evaluated using Eco-points 97 and Eco-indicator 99 methods. In comparison to standard concrete, their findings revealed that the 62% and 32% of carbon emission reduced in the second and third type of the concrete, respectively.
Anastasiou & Papayianni conducted a comparative LCA for six unique concrete pavements in their work [58]. Pozzolanic Portland cement (PPC), mixed-type binder with PC and FA and a novel hydraulic road binder were used binder materials in the road pavements. Adding to that, two distinct aggregates such as; steel slag and crushed limestone, were used in the comparative LCA. For the LCA, the construction, usage, and maintenance, as well as end-of-life recycling, were all taken into account during a 40-year period. The findings revealed that the concrete pavements with a large proportion of substitute materials can considerably lower CO2 emissions and so improve their environmental footprint when compared to standard concrete pavements.
The impact of recycled aggregates and FA in concrete was explored by Kurad et al. [59]. To examine the potential for global warming of concrete, the CML approach was used to analyze environmental implications. There were three mix groups: 100% fine RCA, 50% fine RCA and 0% fine RCA. The average global warming potential (GWP) was reduced by 0.91%, 0.02%, and 0.91% per kg, respectively, and it was further reduced by 2% by applying a 1% superplasticizer. Figure 1 presents the multiple benefits of RHA in concrete.

Figure 1. Benefit of rice husk ash to the structural and environmental behavior of UHPLC concrete
In this paper, the life cycle assessment of RHA in UHPLC has been evaluated to determine its potential contribution to global warming potential under carbon footprint considerations, terrestrial acidification and water potential. Also, the compressive strength of the studied concrete has been predicted by using intelligent techniques, which include three hybrids of ANN, genetic programming, evolutionary polynomial regression and combinatorial group method of data handling. In the LCA evaluation and prediction exercises, concrete age, cement content, rice husk ash, water, super plasticizer, and fine aggregates were used as the input's variables from a data of 192 mixes collected from literature.

Ultra-High-Performance Concrete Mix Data Collection and Statistical Analysis
Extensive literature search and data curation were the methods of data collection adopted in this research work. A concrete mix database of 192 mixes were collected from a previous research study by Iftikhar et al. [60]. The records were collected from different mixes of experimentally tested concrete mixtures with different components' ratios, cured under different conditions. Each record contains the following data: concrete age (A) days, cement content (C) kg/m 3 , rice husk ash content (RHA) kg/m 3 , water content (W) kg/m 3 , super plasticizer content (PL) kg/m 3 , fine aggregates content (FAg) kg/m 3 , cylinder compressive strength of concrete (Fc) MPa. The collected records were divided into training set (150 records) and validation set (42 records). In Tables 1 and 2, the statistical characteristics and the Pearson correlation matrix of the collected data for the studied variables are summarized and Figure 2 shows the histograms for both inputs and outputs.

Research Program Plan
Life cycle impact assessment evaluation was carried out on the rice husk ash (RHA) ultra-high performance lightweight concrete (UHPLC) to determine the global warming potential (GWP) (carbon footprint), terrestrial acidification (acid potential), and the water potential of the multiple concrete mixes produced under different laboratory conditions and cured at different days. Furthermore, five different soft computing techniques were deployed to forecast the compressive strength (Fc) of ultra-high-performance lightweight concrete using the gathered dataset. The implemented techniques are the hybrids of artificial neural network, which include ANN-BP, ANN-GRG and ANN-GA, genetic programming (GP) and the evolutionary polynomial regression (EPR). All these intelligent techniques were used to predict the compressive strength of concrete at certain age (Fc, MPa) using concrete curing age (A) days, cement content (C) kg/m3, rice husk ash content (RHA) kg/m 3 , water content (W) kg/m 3 , super plasticizer content (PL) kg/m 3 and fine aggregates content (FAg) kg/m 3.
Each deployed intelligent method is based on different approaches, which mimic the human brain for the working of the ANN, optimization of mathematical regression for EPR and simulating evolution of natural creatures for GP. However, for all techniques, their accuracies were estimated on basis of mean absolute percentage error (MAPE), mean squared error (MSE), root of mean of squared errors (RMSE), correlation coefficient (R) and determination coefficient (R 2 ). The next sections present the results of the environmental impact assessment and each intelligent method utilized and its accuracy metrics.

Life Cycle Assessment: Goal, Scope and Functional Unit
Environmental performance of materials and circular economy are considered two critical factors when developing any kind of new material. The normative ISO 14040 [61] analyzes the effect on the environment of energy use and materials, among other parameters taking a life cycle perspective into account. A life cycle assessment (LCA) allows for the evaluation of the environmental burden and optimization opportunities without shifting it from one step of the life cycle to another. The main goal of the LCA is to evaluate and analyze the environmental impact of adding rice husk ash to a cement matrix. A more complex life cycle assessment study has been performed based on varying a large number of data compositions from the 192 mixes of UHPLC. The functional unit established is 1 m 3 of concrete mix using rice husk ash (RHA). The scope of this study covers the extraction and use of raw materials such as cement, superplasticizer, fine aggregates, and RHA. The system boundary represents the scheme of the scope of the LCA study as presented in Figure 3.
The concrete mixes considered for the LCA in this study were identified using cement composition. If two or more concrete mixtures have the same amount of Ordinary Portland cement (OPC), then the superplasticizer is used as an identifier. For instance, C-425-SP-5 identifies a concrete mix that uses 425 kg of Portland cement and 5 kg of superplasticizer.

Life Cycle Assessment: Inventory and Method
The data was modelled using the EcoInvent Database 3.2 [62] for materials such as Portland cement, fine aggregates, superplasticizer, electricity, water, and creating the inventory for the rice husk ash (RHA) as a waste from the cogeneration of electricity. The inventory accounts for materials, energy, and transportation that can be converted directly to impact indicators. In this study, ReCiPe Midpoint H [63] method was used for data conversion to impact categories. This method allows for a global approach while maintaining a hierarchy perspective at the midpoint level. This means a balanced environmental performance assessment for the cradle-to-gate evaluation of the cementitious material.

. Summary Remarks
The life cycle analysis was performed on the different mixes with varying RHA as an alternative cementitious material. Three main categories were evaluated in the assessment: global warming/carbon footprint, terrestrial acidification/acid potential, and water consumption/water footprint. Table 3 presents the results from the different concrete mixes analysed for the specified categories. Results show a direct relation in the use of cement in the concrete with the increase of certain environmental indicators such as global warming. The importance of including rice husk ash as a silicate source will result in the enhancement of the concrete matrixes' environmental profile.

Carbon Footprint
Normalizing data in an LCA study helps the reader understand and compare the results appropriately. For this purpose, the normalized LCA results are also given for comparison between evaluated concrete mixes in the current study. Figure 4 shows the normalized environmental impacts for the concrete mixes analysed. The carbon footprint increases with the addition of cement to the concrete mix. The C-783 mix is the most pollutant one in terms of carbon dioxide emissions. On the other side, C-249 and C-300 concrete samples emit lower CO2. The line of tendency shows an increase because of the use of cement, but variations can be seen because of the inclusion of superplasticizer without RHA addition. The OPC concrete has a global warming potential of 386 kg CO2 eq., which is 33% more carbon dioxide emitted than the C-300 mix [64]. Asadollahfardi et al. [64] analysed geopolymer concrete, micro, and nano-silica concrete. The results showed that the geopolymer concrete has roughly the same carbon footprint as C-300 but a higher impact on the silica mixes. Different studies in the literature report an average impact of 350-400 kg of CO2 emissions for the production of similar concrete samples [64][65][66].

Acid Potential
The acid potential can be seen in Figure 5. As seen in the figure, normalized results were given for this impact indicator for an easier comparison. Once again, the C-783 concrete mix shows a higher impact in this category. This reinforces the significance of the Portland cement on the overall impact of the concretes to be studied. The lowest impact in this category was found to be the mix C-375. However, the concrete C-300 has performed well in the terrestrial acidification category. In the same sense, literature [64,66,67] shows that pure OPC has an acidification potential 27% higher than the C-300 mix. Even though it is not the optimal mix for this indicator, it has a better environmental performance than pure OPC.

Water Footprint
Another critical category to be evaluated is the water footprint because water supply shortages occur worldwide. Thus, the demand for water from the cement and concrete industry can be significant. Figure 6 shows variation in the results for water consumption. As seen in the figure, the amount of water required in all the processes to achieve the preparation of the C-391 concrete sample is the highest. This is related to the higher content of water added to the mixture plus the consumed water by the elaboration of the other materials like the plasticizer and cement. The concrete C-300 has a balanced impact towards the middle of the data evaluated.

Process Contribution
The C-300 matrix was analysed to assess the contribution of each of the materials needed for the concrete because it performed well under the LCA methodology. The first parameter is that this concrete mix does not require a superplasticizer, which would add significant value to the environmental categories discussed earlier. In addition, the OPC is the major contributor (~56%) in almost every category of the 18-impact indicators analysed with the ReCiPe Midpoint H [63] method. The rice husk ash has the majority of the contribution to ozone depletion (67%), marine eutrophication (88%), and freshwater eco-toxicity (55%). Fine aggregates have lesser contributions except for terrestrial eco-toxicity and water consumption, with 48 and 45%, respectively as shown in Figure 7. One study showed similar behaviour for the contribution of concrete towards environmental impact indicators, reporting the OPC as one of the significant contributors [68].

ANN Prediction
Three models were developed using (ANN) technique. All the models have the same layout (6:5:1), standardization method (Var/S.D) and activation function (Hyper Tan). However, each model utilized different training algorithm as follows: the traditional "back propagation (BP)" algorithm, the well-known mathematical algorithm "gradually reduced gradient (GRG)", the famous (AI) optimization technique "genetic algorithm (GA)". These three developed models were used to predict (Fc) values. The used networks layout is illustrated in Figure 8 while the weight matrixes of each model are showed in Tables 4 to 6. The average errors of total dataset are 5.3MPa, 5.8MPa and 7.3MPa and the R 2 values are 0.977, 0.973 and 0.957 respectively. These outcomes compare well with a previous study [60] which utilized GEP and RFR and achieved R2 of 0.96 and 0.91 with ANN-BP and -GRG showing better outcome and ANN-GA at par with GEP of the previous study.   H(1:1) H(1:2) H(1:3) H(1:4)

GP Prediction
The developed GP model has six levels of complexity. The population size, survivor size and number of generations were 200 000, 30 000 and 300 respectively. Equation 1 present the output formulas for Fc. The average error of total dataset is 13.8MPa, while the R 2 value is 0.818. The close-form equation shows the possible application of the GP model in the manual and computer computation of the compressive strength of the lightweight concrete during design and the concrete mechanical strength behavior during use.  (1)

EPR Prediction
Finally, the developed EPR model was limited to cubic level for (Fc), for 6 inputs, there are 84 possible terms (56+21+6+1=84) as follows: (2) GA technique was applied on these polynomials to select the most effective 32 terms. The outputs are illustrated in Equation 3. The average error and R 2 values were 7.9MPa and 0.951 respectively. This outcome compares well with and better than the outcome of RFR model of a previous study [60] which showed an R 2 of 0.91. (3)

GMDH-Combinatorial Prediction
The combinatorial group method of data handling algorithm (GMDH-Combi) is a single-layer parametric algorithm that is self-organizing and can carry out a full exploration of all possible models [69]. In the GMDH-Combi method, models are generated from all probable combinations of input variables, and the best final model would be chosen according to a selection criterion, which is usually an error criterion such as root mean squared error (RMSE).
The value of the single-layer self-organizing algorithms comprises straightforwardness and their ability to categorize model structure [70,71]. It should be noted that in the GMDH-Combi approach, different combinations of input variable with user-defined-power-orders is considered and apparently a time-consuming calculation process would be formed [71][72][73]. For instance, considering to model a dataset with two inputs, (x1 and x2) and a target variable (y), a function of quadratic polynomial that the optimization process needs to be performed would be as y=c0+c1•x1+c2•x2+c3•x1•x2+c4•x1²+c5•x2² (where ci, i = 0, 1, …, n is the model constants). GMDH-Combi would lead to choose an optimized and compound model in which y=c0+c3•x1•x2 is a subdivision of searching relationships. The aim would be finding as smallest as possible model in terms and error. As full combinatorial exploration of model with entire terms takes too much time, the search domain is limited to n terms and the user controls the number of terms in the polynomial. Many trial and error were conducted in the present database to achieve the best orders along with the least error and the best robustness. This process gave rise to the closed-form equation in Equation 4 with a performance index of 0.899 for the R 2 and 9.39% as the MAPE. This shows that GMDH-Combi performed better than GP.
Overall, Table 7 shows the summary of the performance accuracy of the models predicted by using ANN (-BP, -GRG and -GA), GP and EPR and GMDH-Combi. In Figures 9 to12, the Taylor diagram, variance diagram, best fir scatter diagram and residual sketch, respectively, which shows the standard deviation and root mean square errors comparison of the models with respect to measured data and comparison between the models and the measured data and the plot of the residuals. Although, the prediction accuracy of EPR model is lower than ANN models, but the output is a closed form equation, which could be used manually or as software unlike the ANN output which can't be used manually. The results indicated that the accuracy of the ANN model is slightly affected by the training algorithm. The back-propagation (BP) showed the best level of accuracy (94.7%), gradually reduced gradient (GRG) came in the second order with accuracy of 97.9% and genetic algorithm (GA) showed the lower level of accuracy (96%).

Sensitivity Analysis of the Studied UHPLC Constituents
A useful concept is provided to identify the importance of each of the "cause" (input) factors on the "impact factor" (output). This allows us to identify the most sensitive factors affecting Fc hierarchically. To achieve this goal, sensitivity analysis was performed to identify the relative impact of each parameter on output in the mode using the cosine domain method [74]. To apply this method, all of the data pairs were expressed in common X-space. To undertake this technique, all data pairs should be utilized to build a data array X as follows [75][76][77][78]: Each of the elements, xi, in the data array X is a vector of lengths of m, that is: The strength of the relation between the dataset, xi and xj, is presented as follows: Results from the sensitivity analysis are shown in Figure 13. This figure shows that the PL, C, FA, W and A are the most significant parameters determining the Fc, respectively. Besides, RHA has the least effect on Fc.

Conclusions
The life cycle assessment was applied to several concrete mixes incorporating OPC, rice husk ash, fine aggregates, and superplasticizer. The environmental performance of the mixtures was related mostly to the percentage of cement in the mix. Thus, the concrete structure with the highest score in the environmental performance evaluation was C-783. In contrast, C-300 was selected as the optimal choice because of the good environmental profile found in the LCA and the compressive strength associated with the concrete mix.
This research presents six models using (AI) techniques (GP, ANN-BP, ANN-GRG, ANN-GA, EPR, and GMDH-Combi) to predict the compressive strength of ultra-high-performance concrete at a certain age (Fc) using concrete age (A), cement content (C), rice husk ash content (RHA), water content (W), super plasticizer content (PL), and fine aggregate content (FAg). The results of comparing the accuracies of the developed models could be concluded in the following points:  GP model is the simplest and the less accurate one (86.0%). Then EPR and ANN-GA with accuracy of 95.3% and 96% respectively, finally ANN-BP and ANN-GRG models have almost the same accuracy of 97.9% and 97.4%, respectively while the GMDH-Combi has a performance of 89.9%;  The sensitivity analysis of the variables indicates that the plasticizer (PL) is the most important factor, then cement and fine aggregates, while rice husk ash is the least influential;  Both the GP and EPR models didn't include RHA, which indicated its minor impact on the concrete strength. However, GMDH-Combi showed better performance than GP and had a less complex closed-form equation;  The GA technique successfully reduced the 84 terms of the conventional polynomial regression quadrilateral formula to only 32 terms without significant impact on its accuracy;  The GA technique successfully reduced the 84 terms of the conventional polynomial regression quadrilateral formula to only 32 terms without significant impact on its accuracy.

Data Availability Statement
The data presented in this study are available in the article.

Funding
The authors received no financial support for the research, authorship, and/or publication of this article.

Conflicts of Interest
The authors declare no conflict of interest.