CIVIL ENGINEERING 365 ALL ABOUT CIVIL ENGINEERING



AbstractRecycled aggregate concrete (RAC) technology is broadly adopted in the construction industry. However, such technology tends to promisingly be implemented only in countries with developed economies, leaving behind countries with emerging economies. To increase the utilization of RAC in these emerging economy countries, this research program aims to investigate the applicability of using the artificial neural network (ANN) technique to predict onsite construction activities of RAC. The construction activities are modeled for the 909 dataset, which includes costs, concrete volume for construction, and total construction time. The results indicate that the mean absolute percentage error values of the RAC cost, concrete volume for construction, and total construction time (including recycled aggregate production and RAC production processes) are 1.98, 28.21, and 2.96, respectively. The mean squared error values of RAC cost, concrete volume for construction, and total construction time are 56,979, 20.9, and 0.56, respectively. Moreover, the coefficient of determination (R2) of RAC cost, concrete volume, and construction time of concrete were calculated at 0.999, 0.976, and 0.968, respectively. Both statistical values indicate that the ANN modeling technique is well implemented for constructing RAC. The results also indicate that ANN modeling can be effectively used in time series for predicting the construction activities of making RAC products. The outcomes offer benefits to stakeholders in construction activities, including improved cost estimations, reduced waste from less concrete going unused, and more accurate project scheduling. ANN modeling represents a relatively simple prediction and can be adopted in preconstruction stages, such as project planning and investment decision making, leading to sustainable construction.IntroductionThe construction industry is one of the backbone industries that is very important to economic growth for every country. Construction activities develop various projects, including infrastructure, residential, industrial, and architectural projects. These projects use a significant amount of concrete mixture, from foundation to building members. Cement, gravel, and sand are the main raw materials for the production of any concrete product. However, the production of each ingredient emits significant greenhouse gases, contributing to global warming. Currently, the materials used from natural resources in many places seem to be insufficient to meet the growing demand. Since 2014, increasing annual greenhouse gas emissions from construction activities in Taiwan was reportedly approximately 2.7% with fewer natural resources (Hung et al. 2019). Where domestic natural resources are insufficient, their import is ubiquitous, leading to transporting construction materials that release a larger portion of greenhouse gas emissions (Lederer et al. 2020; Sezer and Fredriksson 2021). In India, one of the largest producers of construction materials, although the country has abundant resources, it began to import natural sand (Akhtar and Sarmah 2018). Urbanization in India has continued rapidly because millions of Indians have moved to urban cities, leading to concerns over supply shortages of natural resources (Jain et al. 2020). Recycling demolition waste is another way to obtain reused materials from aged concrete instead of materials from depleting natural resources. Demolition waste minimizing and recycling programs began in Europe approximately half a century ago. Theoretically, approximately 80% of the demolition waste can be reused. Although the initiative program of reusing the waste was established for a long time, significant progress in waste management is apparently present only in developed-economy countries. In particular, little progress has been made in today’s emerging economy countries. This issue still generates many environmental problems because emerging economy countries are currently the main producers and consumers of concrete material (Ma and Zhang 2020; Sereewatthanawut and Prasittisopin 2020), which is a significant concern because these emerging economy countries, such as China, India, and others in Southeast Asia, reportedly produce more than 60% of total global production, which is growing rapidly (Tam et al. 2018). Koc and Okudan (2021) assessed the life cycle risks of deconstruction and reported that the most impacted risks during deconstruction are related to inaccurate estimations, quality, health and safety issues, insufficient reuse and recycling of building components, unavailability of deconstruction contractors, and lack of infrastructure. Yang et al. (2017) reported that less than 10% of the demolition waste in China was recycled, and 84% was dumped into landfills. This situation represents the typical case of many developing and underdeveloped countries—most of the waste went through landfills. Some of these dumpings were illegal and created major soil and underground water pollution, including leaching of heavy metals, construction chemicals, polychlorinated biphenyls, and asbestos (Lu 2019). These chemicals are accumulating and cause major long-term health problems.The issue of disposal management of demolition waste is still a serious concern because of the population growth in urban cities. Technology for adding value to demolition waste has been widely developed and successfully implemented. Recycled aggregate concrete (RAC) is used to turn aged concrete waste from demolition sites into recycled aggregates. These recycled aggregates are then substituted with natural aggregates in fresh concrete mixtures. Numerous studies were done and apparently exhibited a full understanding of the impact of recycled aggregates’ substitution in RAC from small to large percentages. Although its fresh and hardened performance characteristics tend to be deficient when substituting the recycled aggregates, adjusting the concrete mix design, adding mineral and chemical additives, and processing various types of quality improvements can be done to minimize deficiencies (Behera et al. 2014; Makul 2021; Prasittisopin et al. 2017; Tam et al. 2021). The techniques to reuse and recycle the recycled aggregates into concrete mixture are promising and are reported to be proper solutions. Nonetheless, the real implementation still struggles in emerging and underdeveloped economy countries that lack rigid national regulation of disposal of waste streams. More in-field studies on the benefits of using RAC in the collection, transportation, and disposal in construction and waste management are required (Masood et al. 2001). Yu et al. (2021) studied the construction activities regarding RAC and, for regions in which the recycling process was at a lower level than the extraction of natural resources, suggested that an effective approach is to first implement a waste classification process to maintain the purity of the waste. This led to lower recycling costs. The cost component of waste management seems to be a key factor in policy and strategy decision making.From previous studies, machine learning methods, including XGBoost, Random forest, Adaboost, Naïve Bayes, logistic regression, K-nearest neighbor, Support vector machine, and artificial neuron network (ANN) techniques, have been introduced (Jan et al. 2019; Koc et al. 2021). The ANN technique was introduced later to predict concrete performance (Lai et al. 1997; Yeh 1998). Numerous studies reported that using ANN modeling methods offers a promising tool for procuring raw materials and optimizing a mix design of concrete and RAC given their fresh and mechanical properties (Chen et al. 2008; Ezeldin and Sharara 2006; Moon and Chowdhury 2021). Moon and Chowdhury (2021) used the ANN technique of the three-day concrete compressive strength to predict 28-day concrete compressive strength. The three-day compressive strength tested is usually attributed to the strength needed to lift a structural member, which is related to the design value of 28-day compressive strength. Chen et al. (2008) also studied the ANN technique for vendors’ evaluation and indicated that the technique could be a powerful construction tool that leads to more sustainable construction. Most studies on concrete and RAC using ANN modeling have been evaluated to predict the performance characteristics of the resulting products. The ANN modeling used to predict concrete performance characteristics include the slump of fly ash and slag concrete (Yeh 2006), setting time and compressive strength of concrete with aluminum waste and sawdust ash (George and Elvis 2019), compressive strength, split tensile strength, and flexural strength of lightweight aggregate concrete (Nagarajan et al. 2020), bond strength of reinforcement of concrete (Yartsev et al. 2019), demystified behavior of cement-based materials (Chabib and Nehdi 2005), and compressive strength of concrete containing waste rubber (Nyarko et al. 2019). From the RAC aspect, some ANN modeling studies have predicted its performance characteristics since 2013, including compressive strength (Deshpande et al. 2014; Duan et al. 2013b, 2018; Getahun et al. 2018; Kandiri et al. 2021; Khademi et al. 2016), tensile strength (Getahun et al. 2018), modulus of elasticity (Duan et al. 2013a, 2018; Seyedhamed et al. 2019), triaxial loading behavior (Xu et al. 2019), water permeability (Chen et al. 2020), and dynamic modulus and permanent strain (Ghorbani et al. 2021). All reported ANN modeling studies focused on material and performance aspects of RAC products. To the best of our knowledge, no relied on the use of the ANN modeling technique for the construction management activities to produce RAC products, especially for in-field implementations. Therefore, this study aims to predict the construction activities (i.e., RAC cost, RAC volume, and construction time) of RAC using the ANN modeling technique, especially for onsite implementations. The data input into the model are derived from current in-field practices, such as demolishing buildings, transport to industrial stationary crushers, mixing, and casting as precast members at precast factories. The total dataset input into the models calculated is the 909 dataset. The output from this study differs from previous studies because this study focuses on construction management activities and ANN modeling for in-field implementations. The results from this study indicate the benefits of using ANN modeling to precisely predict construction activities, thus identifying a proper and easy estimation tool during the preconstruction stage.The ANN modeling technique predicts that construction output positively impacts the construction industry when the accurate prediction of such a complex system is needed. Lam and Oshodi (2016) used ANN modeling to predict construction output in Hong Kong and indicated that the applicability of the ANN modeling was solid. The authors suggested exploring the investigation programs in other regions. The authors also emphasized that quantitative data needed to be validated to support the strategic planning of nations’ long-term construction activities (Lam and Oshodi 2016). Predicting construction output based on ANN modeling directly contributes to a nation’s GDP and future economic growth (Aryal and Wang 2003). In addition, Shen et al. (2020) also used the ANN modeling method to predict and optimize construction’s safety/risk factors. The authors found that the precontrol measures based on modeling could minimize the number of onsite accidents. Hence, studies of ANN modeling on construction activities are necessary, including for RAC, for which the current policy-making challenges that exist in nations with emerging economies are very concerning.As aforementioned, the importance of a quantitative evaluation of RAC on construction activities is highly influenced by nationwide policy-making. When the quantitative database in construction practices is more defined, it can stimulate either the promotion or regulation of government programs by adopting appropriate policies to manage RAC activities. Examples of nationwide policies to promote or regulate relative to RAC include the tax compensation from using RAC, extra charges for dumping construction waste, and the establishment of partnerships between the government and private companies to invest in RAC plants. However, as per the publications presented, their practical database knowledge is lacking. To offer a precise model for predicting these construction activities would be a cutting-edge finding that potentially benefits construction activities holistically and results in increasing sustainable development value to nations’ economies. Precise models, such as ANN modeling, are powerful tools for generating more defined factors from significant amounts of complex data retrieved from various construction sites. For these reasons, this research program aims to present the applicability of using ANN modeling to predict construction activities that produce RAC based on onsite implementations. Three studied factors on construction activities of RAC evaluated here include the cost of concrete, the volume of ready-mixed concrete for construction, and total construction time. The RAC volume produced from fresh concrete casting is noted as being different from the ready-mixed concrete volume from construction because, typically, the ready-mixed concrete volume is sufficiently in excess to be placed into the formwork only once. This leads to additional fresh concrete waste. The model prediction with accurate results creates a database to be used as a decision-making tool in development planning and forthcoming investments in the construction industry. The model is expected to offer significant benefits to construction industry organizations in emerging economy countries.MethodsModel Inputs Using ANN TechniqueStudying the model inputs using ANN followed the Circular Economy Management Model (CE-CR) recycled coarse aggregate management theory (POEDCR) (Kim et al. 2005). The model was developed for project management and adopted into the RAC model. Data were collected from a practice onsite construction and building demolition site and then land transported to an industrial recycled aggregate plant. The demolition sites are in Bang Khen district, Bangkok. The buildings were approximately 60-year-old two-story reinforced concrete buildings. The distance from the demolished sites to the On-nut recycle aggregate plant was approximately 35 km. To be noted is that the hauling distance could significantly affect RAC production costs. In general, the distance in this work is a normal distance within the area in an urban city for aggregate and concrete transport. The recycle aggregate plant consists of a feeder, a crusher to reduce the aggregate size, a magnetic separator to remove waste metal, a cooling water spray system, a dust filter system to reduce dust, and a screener to separate aggregate particles with a mesh size of 40 mm. This plant has a production capacity of 500 tons of recycled aggregates per day and is located in the Prawet district, Bangkok. Its manufacturing process is displayed in Fig. 1. After obtaining the recycled aggregates from the recycled aggregate plant, the recycled coarse aggregate was used for concrete mixing at the 20% weight replacement level. The concrete mixture consists of portland cement Type I: 295 kg, sand: 810 kg, natural coarse aggregate (3/4”-#4): 908 kg, recycled coarse aggregate: 227 kg, and water: 180 kg. The precast concrete mix design had a designed compressive strength of 45 MPa at 28 days. Its slump was fixed at 10 to 15 cm. Fresh concrete mixture was then cast as cubed specimens having a size of 150  mm×150  mm following BS EN 12390 (2019). The compressive strengths were tested from the cubed specimens after curing to ensure good quality resulting RAC. After mixing, the fresh concrete mixtures were placed into the steel molds having sizes 60  cm×300  cm×15  cm. These RAC members were used as structural precast wall panels. This preparation is based on the assumption of producing the exact dimension of the precast panels for housing and building where their formworks were normally used in the plant. Two panels were cast for each test because of the limited room size in the controlled environmental condition, and triplicate tests were performed in this study. After data collection, a total of 909 datasets were input into the ANN modeling. The ANN modeling preprocessing step was first conducted to minimize some issues regarding inconsistencies, outliers, duplicates, and missing values. The overview of the RAC production process is indicated in Fig. 2.Constructing a Forecasting Model Using a Ready-Made Computer ProgramAfter collecting the in-field implementation from the recycling process to produce RAC at the precast factory, a forecasting model was then performed using DTREG standard software (Sherrod 2014). DTREG software is based on ANN modeling. The concrete costs, ready-mixed concrete volume, and time of construction of RAC are the factors predicted in this study. To be noted is that the in-field implementation conducted in this study was well planned and in normal conditions; therefore, no interrupting circumstances occurred. In this case, unexpected risk factors included a longer period of bad weather conditions, traffic jams, accidents during construction, unavailability of a transport truck or RAC production plant, and electricity power outages. Interrupting circumstances can cause delays and increase the cost of the construction process. The model is applicable for the “normal” condition of the expected risk factors. The model was developed by obtaining waste from the construction site, manufacturing the recycled aggregates, and casting the precast wall panels. The overview of the study procedure is provided in Fig. 3. The results of the DTREG software were reported, and their accuracy was calculated using MAPE, MSE, and R2.From Fig. 3, the prediction process was divided as functions of the system into the three following steps: 1.Input data: consisting of material quantity data, concrete quantity, transportation costs, labor costs, and related construction time,2.ANN modeling: uses a Gaussian function to convert into values and then uses a cross-validation method that divides the data into 10 groups (10-fold cross validation); data collected from the sites were analyzed to measure modeling performance, and3.Output data: reported were the cost of fresh RAC at a 20% replacement level by weight of coarse aggregate, the amount of ready-mixed concrete used in construction, and the total construction; the model was adopted for future construction projects for emerging income countries.Results and DiscussionInput Data for ANN ModelingAs previously mentioned, all input data for the ANN modeling were collected from the waste at a real building demolition site in Bang Khen district, Bangkok. The waste was land transported, and concrete was produced using an On-nut industrial recycling plant in Prawet district, Bangkok. The concrete mixtures were then cast. The average values from several in-field tests were input. ANN modeling was performed by manufacturing the amount of concrete as a small construction project at very large construction sites (from 0.3 to 100  m3). Eight parameters were input into the ANN modeling software to predict concrete quantity, concrete cost, and construction time. Details of the input data are provided in Table 1. The ranges of all eight input parameters are in between 88.5 and 29,500 kg for Cement, 243 and 81,000 kg for Sand, 68.1 and 23,00 kg for Recycle aggregate, 54 to 18,000 kg for Water, 0.3 to 100  m3 for Concrete volume, two to six days for Transport, four to nine people for Labor, and eight and 22 days for Time. The summary of input data for all models is also provided in Table 2.Table 1. Parameters of input data for ANN modelingTable 1. Parameters of input data for ANN modelingCodeInput parameterDescriptionNumber of datasetsCCementCement as an ingredient for producing RAC101SSandSand as an ingredient for producing RAC101RAGRecycle aggregateRecycle aggregate obtained from crushing plant used as an ingredient for producing RAC101WWaterWater as an ingredient for producing RAC101CVConcrete volumeConcrete volume produced from batching plant101TRATransportTransportation cost during production40LABLaborLabor cost during production2TTimeConstruction time from obtaining recycle aggregate and producing RAC3Table 2. Summary of input data for all modelsTable 2. Summary of input data for all modelsModelVariableNumber of data rows usedMinimumMaximumMeanStandard deviationRAC costConcrete amount101798.41265,570.00132,792.9177,412.78Cement101213.6070,800.0035,402.1220,637.95Sand101243.0081,000.0040,502.4123,611.24Aggregate101341.00113,500.0056,753.3833,084.88Water1010.81270.00135.0178.70RAC volumeTime1011.002.001.120.32Concrete volume1010.30100.0050.0026.15Construction timeConcrete volume1010.30100.0050.0029.15Time1018.0022.0012.164.20Fig. 4 provides a Gantt chart for manufacturing RAC with different volumes to predict construction activities using ANN modeling. The volume of preparation and the production of RAC directly affect the total time needed to construct the precast concrete panels. As indicated, the RAC batch sizes evaluated in this study’s concrete volume range from 0.3 to 46  m3—typical batch sizes produced. The results indicate that longer construction time is required for larger concrete volume produced. In every case, the time for collecting demolition waste is more than 50% of the total time. Additionally, in every case, the crushing and screening process takes only one day because the industrial plant capacity is large enough to process all of the waste within a day. To be noted here that is the bottleneck of this process is collecting the demolition waste, which requires sorting various kinds of construction material before the concrete waste can be loaded into the truck. As discussed, Yu et al. (2021) reported an effective approach of first implementing a waste classification process at the demolition site. Indeed, developing waste classification to manage the bottleneck at this stage is a priority.The internal structure of ANN modeling to predict concrete costs is indicated in Fig. 5. The structure contains three layers, including the input layer, the hidden layer, and the output layer. The input layer is connected to the hidden layer, and the hidden layer is connected to the output layer. The number of different layers constitutes the network structure. Prior to the application, the ANN modeling is trained; in other words, the connection weights and bias values are first fixed, and the computerized optimization algorithm processes until a very low error value is achieved. The ANN modeling is then analyzed using an unintroduced complex dataset to prove its accuracy. The ANN modeling is trained using different types of training algorithms to achieve the minimum error between the actual and ANN predicted values. The optimal structure of this model is the [4-12-1] ANN structure, which consists of four input layers, 12 hidden layers, and one output layer and uses 10,000 cycles. After cycling, a suitable model for predicting RAC costs is well-established.Fig. 6 indicates the relative importance variables used as input data for the RAC cost model, as detailed in data 1. The scores of the relative importance variable range between 0 and 100. Higher input data scores indicate a stronger impact on the output data (herein, RAC cost). The important variables for predicting concrete costs are cement>water>sand>aggregate. The relative importance calculated from ANN modeling represents the influential rankings of such variables to the predicted parameter. The cement indicated a relative importance variable of 100. Hence, the cement variable has the most influence for RAC cost prediction, followed by the cost of water, sand, and coarse aggregate, respectively. Cement typically costs more than other ingredients. To be noted is that the importance of water is ranked second, which means that, regarding cost, water can be a more influential parameter than sand and coarse aggregate in emerging-economy countries.Fig. 7 indicates the optimal structure of the ANN modeling to predict concrete costs [140-4-1]. The structure consists of 140 input layers, four hidden layers, and one output layer and uses 28,532 cycles. A suitable model is developed to cycle the data.Fig. 8 indicates the relative importance variables used as inputs data for ready-mixed concrete volume from construction modeled by the ANN. After cycling the variables, the results indicate that time is the most important variable affecting the volume of RAC produced. This variable is followed by the transportation and aggregate variables. As previously discussed, time herein represents the time to sort and gather the demolition waste from demolition sites that can be bottlenecked, as also reported by Yu et al. (2021). Therefore, more efficiently sorting out the demolition sites to reduce this time factor is critical.The internal structure of the ANN model is used to predict the total construction time of an RAC precast wall panel as shown in Fig. 9. The structure of the modeling input consists of 140 layers, four hidden layers, and one output layer. The model is performed 10,000 cycles. After performing the modeling, a suitable model is established to predict the total construction time of RAC precast wall panels. The results of the relative importance variable of this prediction modeling are indicated in Fig. 10. The relative importance of the volume variable is exhibited to be the highest, followed by the weight, transportation, water, cement, sand, and aggregate variables.ANN Modeling Application for RAC Cost, Construction Volume, and Total Construction TimeThe plot comparing ANN modeling to predict concrete costs with the values collected from actual construction sites is indicated in Fig. 11. The results indicate that when the RAC volume increases, the cost increases linearly. Table 3 indicates the relationship between RAC volume with its actual cost and its predicted cost by ANN modeling. The value indicates the relatively different values between the actual and the predicted costs. The results exhibit that all predicting percent error values are less than 1%. The statistical analysis also exhibits a MAPE result of 1.98 and MSE result of 56,979. The calculated R2 value equals 0.999, which is close to 1. The predicted cost values of RAC are only a little different from the actual cost values. The ANN modeling method can be adopted to accurately predict the RAC cost in-field.Table 3. Internal structure of ANN model for predicting RAC costTable 3. Internal structure of ANN model for predicting RAC costRAC volume (m3)Actual cost (THB)Predicted cost (THB)Error (%)RAC volume (m3)Actual cost (THB)Predicted cost (THB)Error (%)0.31,798.411,785.01−0.55272669,048.2069,079.32−0.000512,655.703,549.97−0.25192771,703.9071,746.80−0.000625,311.406,058.77−0.12342874,359.6074,449.22−0.001237,967.108,583.49−0.07182977,015.3077,125.80−0.0014410,622.8011,136.61−0.04613079,671.0079,796.79−0.0016513,278.5013,676.09−0.02913182,326.7082,446.71−0.0015615,934.2016,252.99−0.01963284,982.4085,104.45−0.0014718,589.9018,833.00−0.01293387,638.1087,795.96−0.0018821,245.6021,412.49−0.00783490,293.8090,468.47−0.0019923,901.3024,005.64−0.00433592,949.5093,131.29−0.00201026,557.0026,618.45−0.00233695,605.2095,776.55−0.00181129,212.7029,235.90−0.00083798,260.9098,428.20−0.00171231,868.4031,843.920.000838100,916.60101,099.33−0.00181334,524.1034,482.030.001239103,572.30103,754.17−0.00181437,179.8037,127.830.001440106,228.00106,417.38−0.00181539,835.5039,767.660.001741108,883.70109,057.61−0.00161642,491.2042,410.610.001942111,539.40111,689.60−0.00131745,146.9045,062.910.001943114,195.10114,354.59−0.00141847,802.6047,726.470.001644116,850.80117,012.31−0.00141950,458.3050,389.130.001445119,506.50119,651.87−0.00122053,114.0053,050.770.001246122,162.20122,287.08−0.00102155,769.7055,705.770.001147124,817.90124,928.69−0.00092258,425.4058,371.070.000948127,473.60127,565.04−0.00072361,081.1061,075.890.000149130,129.30130,208.87−0.00062463,736.8063,750.13−0.000250132,785.00132,855.84−0.00052566,392.5066,425.35−0.000551135,440.70135,475.62−0.0003The results comparing the ANN modeling for predicting the RAC volume with the values collected from actual construction sites are provided in Fig. 12. The results indicate that when the volume of concrete increases, the cost increases and exhibit that all percent error values are less than 1%. The statistical analysis also exhibits that the MAPE result is 28.21. The MSE result is 20.7, and the calculated R2 value is 0.976, which is close to 1. The predicted cost values of RAC are only slightly different from the actual cost values. The ANN modeling method can be adopted to accurately predict the in-field RAC volume. Moreover, the correlation is 0.988, which is close to 1, meaning that ANN modeling is suitable for predicting RAC volume. To be noted is that the predicted model is quite simple. Simply put, the linear regression model of these output data can be model fit in Excel. Boussabine (2018) reviewed the use of ANN in construction management and indicated that one of the dominant advantages of using ANN modeling is its adaptability. ANN modeling can train or automatically adjust its weights to optimize its output. Such automated optimization allows the neural to design itself. When large amounts of data with different patterns (e.g., from other construction sites or in different countries) are input, the ANN can adjust and report the output itself as a clustered group or as each separate data source. In contrast, to use Excel, data need to be added, and the importance of variables changing every time needs to be observed. Table 4 exhibits the relationships between RAC volumes with their actual and predicted costs using ANN modeling, as well as their predicting percent error values. The values indicate the relatively different values between actual and predicted costs. The predicting percent error values range from −1.22 to 0.37, which are large at very small volumes of concrete mix, namely, 1.3  m3 and 2.5  m3. When the RAC volume is sufficiently large, the predicting percent error values can be almost negligible.Table 4. Internal structure of ANN model for predicting RAC volumeTable 4. Internal structure of ANN model for predicting RAC volumeActual volume (m3)Predicted volume (m3)Error (%)Actual volume (m3)Predicted volume (m3)Error (%)Actual volume (m3)Predicted volume (m3)Error (%)1.30−1.39−1.2235.5035.270.0167.5067.44−0.002.501.820.3737.5037.000.0173.0068.250.074.504.160.0839.5039.010.0170.5070.55−0.007.507.480.0031.5041.27−0.2472.5072.490.0010.009.880.0143.5043.76−0.0174.5074.040.0111.0011.04−0.0034.5045.20−0.2481.5076.550.067.5014.07−0.4746.5047.32−0.0278.0078.40−0.0115.5015.80−0.0245.0048.44−0.0780.0080.77−0.0117.5017.56−0.0053.5049.80−1.2285.0082.800.0320.0019.650.0256.0052.160.0783.5083.64−0.0013.0020.85−0.3852.0054.030.0786.5085.890.0124.0023.690.0154.5054.66−0.0495.5095.72−0.0026.5026.470.0056.5056.69−0.0095.5095.75−0.0020.5027.62−0.2654.5058.94−0.0095.0095.76−0.0130.0030.10−0.0061.5061.59−0.0896.0095.760.0031.0032.57−0.0563.5063.06−0.0092.5095.77−0.0333.5033.87−0.0165.5065.600.0192.5095.80−0.03Fig. 13 provides the results comparing ANN modeling for predicting RAC volume to the values collected from actual construction sites, as well as the predicting percent errors. The results indicate that construction time increases when producing larger RAC volumes. The calculated MAPE value is 2.96, the MSE value is 0.56, the R2 value is 0.968, and the correlation is 0.984. These values are close to 1, indicating that this ANN modeling can be properly adopted to predict the RAC construction time. The relationships between RAC volumes with their actual and predicted costs by ANN modeling, as well as their predicting percent error values, are given in Table 5. The results indicate that all predicting percent error values range from −0.04 to 0.07. These values also ensure that the ANN modeling technique is effectively used to predict RAC construction time. A summary of the output data for all models calculated using ANN modeling is provided in Table 6.Table 5. Internal structure of ANN modeling for predicting construction timeTable 5. Internal structure of ANN modeling for predicting construction timeActual time (days)Predicted time (days)Error (%)Actual time (days)Predicted time (days)Error (%)Actual time (days)Predicted time (days)Error (%)87.4540.0788.122−0.021514.9020.0187.5930.0588.162−0.021514.9130.0187.6980.0488.227−0.031514.9540.0087.7880.0388.254−0.031515.022−0.0087.8220.0288.322−0.041515.083−0.0187.8410.0288.613−0.071515.118−0.0187.8730.0288.883−0.101515.243−0.0287.9030.011511.4820.311515.280−0.0287.9190.011513.8260.081515.379−0.0287.9380.011514.1200.061515.496−0.0387.9510.011514.4690.041515.580−0.0487.9770.001514.5560.031515.671−0.0487.9990.001514.6670.021516.747−0.1088.024−0.001514.7030.022221.4770.0288.035−0.001514.7540.022221.9120.0088.049−0.011514.8030.012222.520−0.0288.089−0.011514.8410.011514.9020.01Table 6. Summary of output data for all modelsTable 6. Summary of output data for all modelsOutput data for ANN modelingRAC costRAC volumeConstruction timeProportion of variance explained by model (R2)0.9990.9760.968Coefficient of variation (CV)0.00180.090.062Normalized mean square error (NMSE)0.000010.020.032Correlation between actual and predicted0.999990.990.984Root mean squared error (RMSE)238.7104.550.748Mean squared error (MSE)56979.920.70.560Mean absolute error (MAE)160.942.110.370Mean absolute percentage error (MAPE)1.9828.212.960The results of the ANN modeling used in this study indicate a promising study. The level of confidence is sufficiently high, which conforms to other studies (Heravi and Eslamdossst 2015; El-Gohary et al. 2017; Maya et al. 2021). For example, predicting construction project performance in Syria using ANN modeling is reported to have an accuracy of 96.1%. The high level of confidence values from using ANN modeling in this study allow ANN modeling to be efficiently adopted as a prediction tool for manufacturing RAC activities. Furthermore, concrete volume, cost, and construction time are known to be among the frequently used estimations for the practical measurement of construction efficiency (Wu et al. 2014). For instance, Poon et al. (2004) estimated construction waste by recording concrete volume from each truck. The collected data were used to minimize the waste from public housing projects in Hong Kong. Therefore, using ANN modeling to predict RAC cost, volume, and construction time can be subsequently adopted to optimize these values.RecommendationsThis study applied ANN modeling techniques to make predictions of construction activity parameters. Other computerization techniques could be applied, such as support techniques, vectors, machines, and genetic networks. The basic advantages of using the ANN modeling technique enable it to automatically simulate the behavior of very complex systems of data, which is very useful for civil engineering activities because, fundamentally, such practices are highly uncertain and complicated. Therefore, the ANN modeling technique used herein can be an effective tool for implementation into civil engineering work. This modeling technique can offer more precise predictions for larger-scale projects. The ANN modeling theory was applied to predict RAC costs, ready-mixed concrete volume during construction, and total time from construction activities only in Thailand. The research limitation of the current study is that the data collection process did not include many construction sites, which could result in the ANN modeling not being applicable in real applications elsewhere. To further validate this ANN modeling technique in larger areas, additional dataset input from different sources, different timeframes, and different countries should be investigated. However, the outcomes can be offered as a referenced database for countries with emerging economies. Other parameters related to onsite construction, such as the future depletion of natural resources, the cost of materials in various categories of construction, related construction activities of various new construction technologies offered, construction project quality, and safety in construction sites should be predicted in advance.References Akhtar, A., and A. K. Sarmah. 2018. “Construction and demolition waste generation and properties of recycled aggregate concrete: A global perspective.” J. Cleaner Prod. 186 (Jun): 262–281. https://doi.org/10.1016/j.jclepro.2018.03.085. Aryal, D. R., and Y. W. Wang. 2003. “Neural network forecasting of the production level of Chinese construction industry.” J. Comp. Int. Manage. 6 (2): 45–64. Basakın, E. E., O. Ekmekcioglu, H. Cıtakoglu, and O. Mehmet. 2021. “A new insight to the wind speed forecasting: Robust multi-stage ensemble soft computing approach based on pre-processing uncertainty assessment.” Neural Comput. Appl. 2021 (Aug): 1–30. https://doi.org/10.1007/s00521-021-06424-6. Behera, M., S. K. Bhattacharyya, A. K. Minocha, R. Deoliya, and S. Maiti. 2014. “Recycled aggregate from C&D waste & its use in concrete—A breakthrough towards sustainability in construction sector: A review.” Constr. Build. Mater. 68 (Jun): 501–516. https://doi.org/10.1016/j.conbuildmat.2014.07.003. Chen, S., Y. Zhao, and Y. Bie. 2020. “The prediction analysis of properties of recycled aggregate permeable concrete based on back-propagation neural network.” J. Cleaner Prod. 276 (Jun): 124187. https://doi.org/10.1016/j.jclepro.2020.124187. Deshpande, N., S. Londhe, and S. Kulkarni. 2014. “Modeling compressive strength of recycled aggregate concrete by artificial neural network, model tree and non-linear regression.” Int. J. Sustainable Built Environ. 3 (2): 187–198. https://doi.org/10.1016/j.ijsbe.2014.12.002. Duan, Z., S. Hou, C. S. Poon, J. Xiao, and Y. Liu. 2018. “Using neural networks to determine the significance of aggregate characteristics affecting the mechanical properties of recycled aggregate concrete.” Appl. Sci. 8 (11): 2171. https://doi.org/10.3390/app8112171. Duan, Z. H., S. C. Kou, and C. S. Poon. 2013a. “Prediction of compressive strength of recycled aggregate concrete using artificial neural networks.” Constr. Build. Mater. 40 (Mar): 1200–1206. https://doi.org/10.1016/j.conbuildmat.2012.04.063. Duan, Z. H., S. C. Kou, and C. S. Poon. 2013b. “Using artificial neural networks for predicting the elastic modulus of recycled aggregate concrete.” Constr. Build. Mater. 44 (Jul): 524–532. https://doi.org/10.1016/j.conbuildmat.2013.02.064. El-Gohary, K. M., R. F. Aziz, and H. A. Abdel-Khalek. 2017. “Engineering approach using ANN to improve and predict construction labor productivity under different influences.” J. Constr. Eng. Manage. 143 (8): 04017045. https://doi.org/10.1061/(ASCE)CO.1943-7862.0001340. Flood, I., and N. Kartam. 1998. Artificial neural networks for civil engineers: Advanced features and applications. Reston, VA: ASCE. George, U. A., and M. M. Elvis. 2019. “Modelling of the mechanical properties of concrete with cement ratio partially replaced by aluminium waste and sawdust ash using artificial neural network.” SN Appl. Sci. 1 (1514): 1–18. https://doi.org/10.1007/s42452-019-1504-2. Getahun, M. A., S. M. Shitote, and Z. C. A. Gariy. 2018. “Artificial neural network based modelling approach for strength prediction of concrete incorporating agricultural and construction wastes.” Constr. Build. Mater. 190 (Nov): 517–525. https://doi.org/10.1016/j.conbuildmat.2018.09.097. Ghorbani, B., A. Arulrajah, G. Narsilio, S. Horpibulsuk, and M. W. Bo. 2021. “Dynamic characterization of recycled glass-recycled concrete blends using experimental analysis and artificial neural network modeling.” Soil Dyn. Earthquake Eng. 142 (Mar): 106544. https://doi.org/10.1016/j.soildyn.2020.106544. Hung, C., S.-C. Hsu, and K.-L. Cheng. 2019. “Quantifying city-scale carbon emissions of the construction sector based on multi-regional input-output analysis.” Resour. Conserv. Recycl. 149 (Oct): 75–85. https://doi.org/10.1016/j.resconrec.2019.05.013. Jain, S., S. Singhal, and S. Pandey. 2020. “Environmental life cycle assessment of construction and demolition waste recycling: A case of urban India.” Resour. Conserv. Recycl. 155 (Apr): 104642. https://doi.org/10.1016/j.resconrec.2019.104642. Jang, Y., I.-B. Jeong, Y. K. Cho, and Y. Ahn. 2019. “Predicting business failure of construction contractors using long short-term memory recurrent neural network.” J. Constr. Eng. Manage. 145 (11): 04019067. https://doi.org/10.1061/(ASCE)CO.1943-7862.0001709. Kandiri, A., F. Sartipi, and M. Kioumarsi. 2021. “Predicting compressive strength of concrete containing recycled aggregate using modified ANN with different optimization algorithms.” Appl. Sci. 11 (2): 485. https://doi.org/10.3390/app11020485. Khademi, F., S. M. Jamal, N. Deshpande, and S. Londhe. 2016. “Predicting strength of recycled aggregate concrete using artificial neural network, adaptive neuro-fuzzy inference system and multiple linear regression.” Int. J. Sustainable Built Environ. 5 (2): 355–369. https://doi.org/10.1016/j.ijsbe.2016.09.003. Kilinc, I., and K. Cigizoglu. 2012. “Reservoir management using artificial neural networks.” In Proc., 14th Regional Directorate of DSI (State Hydraulic Works). State College, PA: Pennsylvania State Univ. Koc, K., O. Ekmekcioglu, and A. P. Gurgun. 2021. “Integrating feature engineering, genetic algorithm and tree-based machine learning methods to predict the post-accident disability status of construction workers.” Autom. Constr. 131 (Nov): 109896. https://doi.org/10.1016/j.autcon.2021.103896. Lam, K. C., and O. S. Oshodi. 2016a. “Forecasting construction output: A comparison of artificial neural network and Box-Jenkins model.” Eng. Constr. Archit. Manage. 23 (3): 302–322. https://doi.org/10.1108/ECAM-05-2015-0080. Lam, K. C., and O. S. Oshodi. 2016b. “Using univariate models for construction output forecasting: Comparing artificial intelligence and econometric techniques.” J. Manage. Eng. 32 (6): 04016021. https://doi.org/10.1061/(ASCE)ME.1943-5479.0000462. Lederer, J., A. G. Kleemann, and J. Fellner. 2020. “Potentials for a circular economy of mineral construction materials and demolition waste in urban areas: A case study from Vienna.” Resour. Conserv. Recycl. 161 (Oct): 104942. https://doi.org/10.1016/j.resconrec.2020.104942. Makul, N. 2021. “Modified cost-benefit analysis of the production of ready-mixed self-consolidating concrete prepared with a recycled concrete aggregate.” J. Constr. Eng. Manage. 147 (4): 04021021. https://doi.org/10.1061/(ASCE)CO.1943-7862.0002019. Maya, R., B. Hassan, and A. Hassan. 2021. “Develop an artificial neural network (ANN) model to predict construction projects performance in Syria.” J. King Saud Univ. Eng. Sci. 2021 (May): 11. https://doi.org/10.1016/j.jksues.2021.05.002. Nagarajan, D., T. Rajagopal, and N. Meyappan. 2020. “A comparative study on prediction models for strength properties of LWA concrete using artificial neural network.” Rev. Constr. 19 (1): 103–111. https://doi.org/10.7764/rdlc.19.1.103-111. Nyarko, M. H., E. K. Nyarko, N. Ademovic, I. Milicevic, and T. K. Sipos. 2019. “Modelling the influence of waste rubber on compressive strength of concrete by artificial neural networks.” Materials 12 (4): 561. https://doi.org/10.3390/ma12040561. Poon, C. S., A. T. W. Yu, S. C. See, and E. Cheung. 2004. “Minimizing demolition wastes in Hong Kong public housing projects.” Construct. Manage. Econ. 22 (8): 799–805. https://doi.org/10.1080/0144619042000213283. Prasittisopin, L., P. Chaiyapoom, C. Thongyothee, and C. Snguanyat. 2017. “Using recycle concrete aggregate coating agent for improving concrete microstructure and hardened characteristics.” In Proc., HISER Int. Conf. 2017: Advances in Recycling and Management of Construction and Demolition Waste. Delft, Netherlands: TU Delft Library. Sereewatthanawut, I., and L. Prasittisopin. 2020. “Environmental evaluation of pavement system incorporating recycled concrete aggregate.” Int. J. Pavement Res. Technol. 13 (5): 455–465. https://doi.org/10.1007/s42947-020-0002-7. Seyedhamed, S., L. E. B. da Silva, D. C. Wunsch, and K. H. Khayat. 2019. “Artificial intelligence to investigate modulus of elasticity of recycled aggregate concrete.” ACI Mater. J. 116 (1): 51–62. https://doi.org/10.14359/51706948. Sezer, A. A., and A. Fredriksson. 2021. “Environmental impact of construction transport and the effects of building certification schemes.” Resour. Conserv. Recycl. 172 (May): 105688. https://doi.org/10.1016/j.resconrec.2021.105688. Shen, T., Y. Nagai, and C. Gao. 2020. “Design of building construction safety prediction model based on optimized BP neural network algorithm.” Soft Comput. 24 (2): 7839–7850. https://doi.org/10.1007/s00500-019-03917-4. Sherrod, P. H. 2014. “DTREG: Predictive modeling software.” Accessed January 16, 2021. https://www.dtreg.com/. Tam, V. W. Y., M. Soomro, and A. C. J. Evangelista. 2021. “Quality improvement of recycled concrete aggregate by removal of residual mortar: A comprehensive review of approaches adopted.” Constr. Build. Mater. 288 (Jun): 123066. https://doi.org/10.1016/j.conbuildmat.2021.123066. Xu, J., Y. Chen, T. Xie, X. Zhao, B. Xiong, and Z. Chen. 2019. “Prediction of triaxial behavior of recycled aggregate concrete using multivariable regression and artificial neural network techniques.” Constr. Build. Mater. 226 (Nov): 534–554. https://doi.org/10.1016/j.conbuildmat.2019.07.155. Yang, H., J. Q. Xia, J. R. Thompson, and R. J. Flower. 2017. “Urban construction and demolition waste and landfill failure in Shenzhen, China.” Waste Manage. 63 (May): 393–396. https://doi.org/10.1016/j.wasman.2017.01.026. Yu, Y., D. M. Yazan, S. Bhochhibhoya, and L. Volker. 2021. “Towards circular economy through industrial symbiosis in the Dutch construction industry: A case of recycled concrete aggregates.” J. Cleaner Prod. 293 (Apr): 126083. https://doi.org/10.1016/j.jclepro.2021.126083.



Source link

Leave a Reply

Your email address will not be published.