materials Article Predicting Performance of Lightweight Concrete with Granulated Expanded Glass and Ash Aggregate by Means of Using Artificial Neural Networks Marzena Kurpinska 1,* and Leszek Kułak 2 1 Faculty of Civil and Environmental Engineering, Gdansk University of Technology, 80-233 Gdansk, Poland 2 Faculty Applied Physics and Mathematics, Gdansk University of Technology, 80-233 Gdansk, Poland; [email protected] * Correspondence: [email protected] Received: 3 May 2019; Accepted: 21 June 2019; Published: 22 June 2019 Abstract: Lightweight concrete (LWC) is a group of cement composites of the defined physical, mechanical, and chemical performance. The methods of designing the composition of LWC with the assumed density and compressive strength are used most commonly. The purpose of using LWC is the reduction of the structure’s weight, as well as the reduction of thermal conductivity index. The highest possible strength, durability and low thermal conductivity of construction materials are important factors and reasons for this field’s development, which lies largely in modification of materials’ composition. Higher requirements for construction materials are related to activities aiming at environment protection. The purpose of the restrictions is the reduction of energy consumption and, as a result, the reduction of CO2 emission. To limit the scope of time-consuming and often high-cost laboratory works necessary to calibrate models used in the test methods, it is possible to apply Artificial Neural Networks (ANN) to predict any of the concrete properties. The aim of this study is to demonstrate the applicability of this tool for solving the problems, related to establishing the relation between the choice of type and quantity of lightweight aggregates and the porosity, bulk density and compressive strength of LWC. For the tests porous lightweight Granulated Expanded Glass Aggregate (GEGA) and Granulated Ash Aggregate (GAA) have been used. Keywords: building; lightweight concrete; granulated expanded glass aggregate; artificial neural networks; prediction properties; granulated ash aggregate; artificial neural networks 1. Introduction For several decades, we have been observing steadily growing popularity of cement composites. This trend depends most of all on continuous research, aimed at environment protection and improving basic properties of the materials, as well as on giving them new, innovative features. One of the many examples of that is lightweight concrete (LWC) containing lightweight artificial aggregates. Its enormous advantage is that the components for the production of artificial aggregates are recycled or are waste materials [1–6]. Lightweight cement composites, in particular, LWC is one of the most important groups of materials, used in the construction industry. This is due to its main advantages, such as: possibility to obtain any shape of the element, good physical, mechanical and chemical properties, as well as good insulating properties [7–10]. In order to determine the impact of individual components on the properties of the LWC, various design methods are used, namely, experimental and analytical-experimental. The suitability of these methods for computer applications is also relevant [11–15]. It should be, however, emphasised that each method of designing of the composition of concrete mixture must be verified experimentally. Materials 2019, 12, 2002; doi:10.3390/ma12122002 www.mdpi.com/journal/materials Materials 2019, 12, 2002 2 of 16 Necessary tests, aiming at calibration of the proposed models, are being carried out, taking into account quality and quantity parameters of the components [16–20]. The tests are usually time-consumingMaterials 2019, 12 and, x FOR expensive. PEER REVIEW What is more, they must be repeated in case of each significant2 of change16 of the type or quality of even one concrete component. Therefore, it is reasonable to use such models, Necessary tests, aiming at calibration of the proposed models, are being carried out, taking into which correlate components and concrete properties so as to minimise laboratory works. account quality and quantity parameters of the components [16–20]. The tests are usually time- The application of artificial neural networks (ANN) to predict important parameters of concrete, consuming and expensive. What is more, they must be repeated in case of each significant change of includingthe type LWC, or quality can be of an even example one concrete here [21 component.–24]. The main Therefore, purpose it is ofreasonable the work to is use to presentsuch models, computer techniqueswhich incorrelate the form components of ANNs, and which concrete will properti be usedes to so predict as to minimise compressive laboratory strength, works. porosity and bulk density ofThe LWC, application depending of artificial on the selectionneural networks of the quantitative (ANN) to predict and qualitative important parameters composition of ofconcrete, lightweight aggregatesincluding (LWA). LWC, ANN can be can an be example used for here predicting [21–24]. The output main data, purpose based of the on work a defined is to present input dataset computer [25 ,26]. Thetechniques advantage in the ofform using of ANNs, computer which techniques, will be used involving to predict compressive ANN, to solve strength, the problem, porosity and defined above,bulk is thatdensity there of isLWC, no needdepending to derive on the explicit selection mathematical of the quantitative relationships, and qualitative because composition networks, of during the machinelightweight learning aggregates process, (LWA). ascribe ANNadequate can be used values for predicting to the subsequent output data, variables,based on a defined pursuing input a given dataset [25,26]. model, obtained in the process of an experimental research [27–30]. In the case, presented in the The advantage of using computer techniques, involving ANN, to solve the problem, defined study, the reference values derived from the results of laboratory tests of LWC, containing granulated above, is that there is no need to derive explicit mathematical relationships, because networks, during expandedthe machine glass aggregatelearning process, (GEGA) ascribe and adequate granulated valu ashes to aggregate the subsequent (GAA) vari [ables,31–34 pursuing]. The present a given study revealsmodel, the possibility obtained in of the using process artificial of an neural experimental networks research to predict [27–30]. the parametersIn the case, ofpresented LWC e.g., in strength,the porosity,study, and the apparent reference values density. derived from the results of laboratory tests of LWC, containing granulated Anexpanded additional glass aggregate purpose of(GEGA) the work and granulated was to check ash aggregate existing computer(GAA) [31–34]. techniques The present in the study form of ANNs,reveals which the can possibility be used of to using design artificial the LWC neural composition networks withto predict the desired the parameters properties. of LWC e.g., strength, porosity, and apparent density. 2. MaterialsAn additional and Methods purpose of the work was to check existing computer techniques in the form of ANNs, which can be used to design the LWC composition with the desired properties. Portland cement CEM I 42,5R according to [35] was used to perform the tests. Chemical composition2. Materials and and physical Methods properties of the cement CEM I 42,5R were shown in Table1. Portland cement CEM I 42,5R according to [35] was used to perform the tests. Chemical Table 1. Chemical composition and physical properties of cement CEM I 42,5R [15]. composition and physical properties of the cement CEM I 42,5R were shown in Table 1. Compressive Setting Setting Blaine Loss of Water Table 1. Chemical compositionStrength and [MPa] physical properties of cement CEM I 42,5R [15]. Start Time End Time Fineness Roasting [%] Demand [min] [min] 2d 28d [cm2/g] [%] Setting Setting Compressive Blaine Water Start155 End 195Strength 30.2 [MPa] 57.3 3504Loss of 3.4 27.5 Fineness Demand Time Time Roasting [%] 2d 28d Content[cm [%]2/g] [%] [min] [min] SiO2 Al2O3 Fe2O3 CaO MgO SO3 Na2O K2O TiO2 Cl 155 195 30,2 57,3 3504 3,4 27,5 21.7 6.2 3.1 63.4 1.0 3.9 0.16 0.64 0.25 0.06 Content [%] SiO2 Al2O3 Fe2O3 MineralogicalCaO MgO composition,SO3 contentNa2O [%] K2O TiO2 Cl 21Na,7 2Oeq6,2 3,1 C633SC,4 1,0 2SC3,9 0,163AC 0,64 0,254 AF 0,06 Mineralogical composition, content [%] 0.7 63.1 7.6 6.1 8.9 Na2Oeq C3S C2S C3A C4AF 0,7 63,1 7,6 6,1 8,9 GEGA of 2 and 4 mm grain size (Figure1a,b) and GAA of 8 mm grain size (Figure1c) were used as the mainGEGA component of 2 and 4 of mm LWC. grain Chemical size (Figure composition 1a,b) and GAA of aggregates of 8 mm grain was size given (Figure in Table1c) were2. used as the main component of LWC. Chemical composition of aggregates was given in Table 2. (a) GEGA 2 mm (b) GEGA 4 mm (c) GAA 8 mm FigureFigure 1. Aggregates:1. Aggregates: (a(,ab,b)) granulatedgranulated expanded glass glass aggregate aggregate (GEGA), (GEGA), (c) (granulatedc) granulated ash ash aggregateaggregate (GAA). (GAA). Materials 2019, 12, 2002 3 of 16 Table 2. The chemical composition of the aggregate [15]. Content [%] Aggregate Loss of Type SiO Al O Fe O CaO MgO SO Na OK O 2 2 3 2 3 3 2 2 Roasting GEGA 63.33 0.74 - 14.19 2.98 0.32 13.35 0.57 4.53 GAA 52.82 24.28 7.5 4.5 3.19 0.43 - 0.2 7.1 The main component of cement is CaO, and its content is from 4% to 63%, while the content of SiO2 silica is approximately 21.7%. It is shown in Table1. The main component of LWA (GEGA and GAA) is SiO2 silica, and its content in GEGA is from 33% to 63% and in GAA is 52.82%, as shown in Table2.