|Mustafa Alas||Civil Engineering Department, Near East University, Near East Boulevard, Nicosia, 99138 North Cyprus, Turkey|
|shaban Ismael Albrka Ali||Civil Engineering Department, Near East University, Near East Boulevard, Nicosia, 99138 North Cyprus, Turkey|
Complexity in the behaviour of an asphalt binder is further escalated with geopolymer (fly ash and alkali liquid) modification, thus making it difficult to accurately predict the performance of the binder. This study employs artificial neural network modelling to predict the complex shear modulus, storage modulus, loss modulus and phase angle outcomes of experimental results from dynamic shear rheometer (DSR) oscillation tests under four separate scenarios. The proposed artificial neural network models received test conditions (temperature and frequency) and three different geopolymer concentrations (3%, 5% and 7% by the weight of bitumen) as the predictor parameters. The variants of the optimal algorithms were Levenberg-Marquardt (LM), Scaled conjugate gradient and Polak-Ribiere conjugate gradient (CPG) training algorithms with different combinations of network structures and tan-sig and log-sig as activation functions. The coefficient of determination, covariance and root mean square error (RMSE) were used as statistical measures of model prediction performance. Based on the statistical performance indicators, the LM algorithm with a 3-5-1 network architecture and tan-sig as the activation function was the best performing model for predicting the complex modulus with R2 values of 0.996 for the training dataset and 0.971 for the testing dataset and RMSE values of 0.118 and 0.139 for the training and testing datasets, respectively. Furthermore, it was observed that the least efficient model was the phase angle prediction model developed with the CPG training algorithm, which had a 3-8-1 network architecture and log-sig as the activation function. The model yielded R2 values of 0.909 and 0.829 for the training and testing datasets, respectively. Poor prediction performance for the testing dataset indicated that the model was unable to learn complexity in the data and would perform below a significance level of 0.90 in predicting using untrained data.
Artificial neural networks; Complex shear modulus; Geopolymer modified asphalt binder; Loss modulus; Phase angle; Storage modulus
Asphalt binder is a complex viscoelastic material that is predominantly used in highway construction (Ali et al., 2015; Ali et al., 2018). Asphalt binders are highly temperature susceptible, which means that they behave like elastic solids at cold temperatures and under low dynamic loading and behave like a Newtonian fluid at high temperatures and under heavy dynamic loading. The viscoelastic property of asphalt binders influences the high-temperature rutting and
low-temperature thermal cracking failures of asphalt pavements (Tapk?n et al., 2009). Asphalt pavements undergo repeated dynamic loading due to vehicular traffic (Du & Huang, 2012). A dynamic shear rheometer (DSR) was used in the experimental investigation of the rheological properties of binders at medium to high temperatures to simulate the dynamic loading effect and to evaluate the viscoelastic properties of asphalt binders, which significant for predicting the durability and service life of asphalt pavements (Al-Mansob et al., 2016). Complex modulus (G*) and phase angle (?) are the two parameters revealed by the DSR oscillation test and are used to evaluate the viscoelastic behaviour of asphalt binders, as in the Superpave specification for the evaluation of asphalt cement for fatigue and rutting (Clyne et al., 2003). G* is the binder’s resistance to deformation under repeated shear loading, and ? is the lag between the applied shear stress and the resulting shear strain. Larger ? values indicate more viscosity, while larger G* values are related to more elastic binders. The definition of ? and the relationship between the ? and G* are demonstrated in Figures 1 and 2. High elastic and low viscous properties are desired at low temperatures, whereas the high viscous and low elastic behaviour of asphalt binders are desired at high temperatures. On this basis, binder modification is a common practice and is referred to as modified asphalt cement (MAC).
Figure 1 Definition of phase angle (Abedali, 2015)
Figure 2 Viscoelastic behavior of bitumen (Abedali, 2015)
Bitumen modification with polymers and nanomaterials is a traditional and effective way to enhance the viscoelastic properties of asphalt binders. However, modification introduces further complexity to the behaviour of binders; hence, in order to determine optimum solutions, extensive laboratory investigations are essential before field application (Fang et al., 2013). In recent years, some studies have presented mathematical and computational methods to model the behaviour of modified asphalt binders to eliminate or provide assistance for the experimental procedures. Some of the successful modelling techniques include numerical modelling using finite element analysis, mathematical modelling using regression models, statistical modelling using response surface methodology (RSM) and heuristic prediction techniques (Huang et al., 2015; Venudharan & Biligiri, 2017; Ziari et al., 2018). Artificial neural networks (ANN) are heuristic prediction techniques that have been gaining the attention of researchers in the field of material science as acknowledged in the literature (Tasdemir, 2009). Baldo et al. (2018) utilized the ANN modelling technique to analyse the mechanical behaviour of asphalt concretes using base bitumen and polymer modified bitumen data observed through experimental investigations. The models developed in their study included single outputs of four mechanical parameters, namely Marshall stability, flow, quotient and stiffness modulus and input parameters, including the bitumen type, the bitumen content, the filler-bitumen ratio, air voids, voids in the mineral aggregates, voids filled with bitumen and the type of production process. Their study produced satisfactory results of correlation coefficient values in the range of 0.98798–0.91024 with the testing dataset. Furthermore, closed-form equations for all four models were developed for repeatability of the results with different materials within the framework provided in their study. The feasibility of the application of ANN in pavement engineering is discussed in Section 2. ANN is a modelling technique used for the classification, regression and prediction of non-linear datasets by learning from supplementary data and predicting new data based on the learned pattern of the data (Zuna et al., 2016; DeRousseau et al., 2018). ANN consists of three layers, including the input layer, in which the predictor variables are fed to the network; the hidden layer, which is the middle layer; and the output layer, in which the network targeted variables are predicted. The performance of ANN strongly depends on the type of data as well as dataset congruence with network features such as the network topology, the training algorithm and the activation function. Readers are referred to (Baldo et al., 2018) for a detailed description of the mathematical theory behind the modelling of neural networks.
The objective of this study was to evaluate the prediction performance of ANN models developed under four separate scenarios with different combinations of ANN architectures, training algorithms and activation functions to predict complex shear modulus (G*), phase angle (?), storage modulus (G’) and loss modulus (G’’) by using test conditions (temperature and frequency) and different geopolymer concentrations (3%, 5% and 7% by the weight of binder) in order to eliminate the drawbacks of the experimental procedures.
The objective of this study was to develop and evaluate the performance capacity of ANN models to predict experimental results from DSR oscillation tests. G*, ?, G’ and G” were attempted to be predicted from mechanical test conditions for modified asphalt binders with the addition of 3%, 5% and 7% geopolymer composed of fly ash and alkali liquid.
The following conclusions can be drawn from this study: (1) The best performing model was developed for predicting G*. The features of the model included an LM training algorithm, tan-sig activation function and 1-5-1 network structure. The model performance evaluated by R2, RMSE and COV produced metrics of 0.996, 0.117 and 22.41, respectively; (2) Models developed for predicting G’ and G” performed satisfactorily regarding performance indicator metrics. However, the variation observed in the R2 values between the training and testing data was an indication that the performance of the models may not have been as precise for predicting untrained datasets; (3) Based on the R2 results, the model developed for predicting ? was observed as the least efficient model regarding prediction capacity.
The value of 0.823 with the testing dataset revealed that the model was unable to learn complexity in the data and that it would perform poorly with untrained new datasets.
Abedali, A.H., 2015. Predicting Complex Shear Modulus using Artificial Neural Networks. Journal of Civil Engineering and Construction Technology, Volume 6(3), pp. 15–26
Al-Mansob, R.A., Ismail, A., Yusoff, N.I.M., Albrka, S.I., Azhari, C.H., Karim, M.R., 2016. Rheological Characteristics of Unaged and Aged Epoxidised Natural Rubber Modified Asphalt. Construction and Building Materials, Volume 102(1), pp. 190–199
Ali, S.I.A., Abdulwahid, R., Eidan, M.L., Yusoff, M., Izzi, N., 2018. Evaluation of Moisture and Ageing Effects on Calcium Carbonite Nanoparticles Modified Asphalt Mixtures. International Journal of Engineering Research in Africa, Volume 34, pp. 40–47
Ali, S.I.A., Yahia, H.A.M., Ibrahim, A.N.H., Al Mansob, R.A, 2017. High Temperatures Performance Investigation of Geopolymer Modified Bitumen Binders. In: Tenth International Conference on the Bearing Capacity of Roads, Railways and Airfields, BCRRA 2017, At Athens, Greece
Ali, S.I.A., Ismail, A., Yusoff, N.I.M., Karim, M.R., Al-Mansob, R.A., Alhamali, D.I., 2015. Physical and Rheological Properties of Acrylate–styrene–acrylonitrile Modified Asphalt Cement. Construction and Building Materials, Volume 93, pp. 326–334
Baldo, N., Manthos, E., Pasetto, M., 2018. Analysis of the Mechanical Behaviour of Asphalt Concretes using Artificial Neural Networks. Advances in Civil Engineering, Volume 2018, Article ID 1650945
Chandwani, V., Agrawal, V., Nagar, R., Singh, S., 2015. Modelling Slump of Ready Mix Concrete using Artificial Neural Network. International Journal of Technology, Volume 6(2), pp. 207–216
Clyne, T.R., Li, X., Marasteanu, M.O., Skok, E.L., 2003. Dynamic and Resilient Modulus of Mn/DOT Asphalt Mixtures. Final Report MN/RC –2003-09, Minnesota Department of Transportation
DeRousseau, M., Kasprzyk, J., Srubar, W., 2018. Computational Design Optimization of Concrete Mixtures: A Review. Cement and Concrete research, Volume 109, pp. 42–53
Du, S.W., Huang, D.F., 2012. The Effect of Cement on Performance Properties of Foam Asphalt Cold Recycled Mixture. Applied Mechanics and Material, Volume 178-181, pp. 1379–1382
El-Badawy, S., Abd El-Hakim, R., Awed, A., 2018. Comparing Artificial Neural Networks with Regression Models for Hot-mix Asphalt Dynamic Modulus Prediction. Journal of Materials in Civil Engineering, Volume 30(7), pp. 04018128
Esen, H., Inalli, M., Sengur, A., Esen, M., 2008. Artificial Neural Networks and Adaptive Neuro-fuzzy Assessments for Ground-coupled Heat Pump System. Energy and Buildings, Volume 40(6), pp. 1074–1083
Fang, C., Yu, R., Liu, S., Li, Y., 2013. Nanomaterials Applied in Asphalt Modification: A Review. Journal of Materials Science & Technology, Volume 29(7), pp. 589–594
Firouzinia, M., Shafabakhsh, G., 2018. Investigation of the Effect of Nano-silica on Thermal Sensitivity of HMA using Artificial Neural Network. Construction and Building Materials, Volume 170, pp. 527–536
Huang, J., Pan, X., Dai, S., Cai, Y., 2015. FEM Analysis of Dynamic Behavior of Asphalt Pavement Structure Weakened by Grassroots with Account of Hydraulic and Vehicle Load Coupling Effects. In: IOP Conference Series: Materials Science and Engineering
Kok, B.V., Yilmaz, M., Sengoz, B., Sengur, A., Avci, E., 2010. Investigation of Complex Modulus of Base and SBS Modified Bitumen with Artificial Neural Networks. Expert Systems with Applications, Volume 37(12), pp. 7775–7780
Liu, J., Yan, K., Liu, J., Zhao, X., 2018. Using Artificial Neural Networks to Predict the Dynamic Modulus of Asphalt Mixtures Containing Recycled Asphalt Shingles. Journal of Materials in Civil Engineering, Volume 30(4), pp. 04018051
Naderpour, H., Mirrashid, M., 2018. An Innovative Approach for Compressive Strength Estimation of Mortars Having Calcium Inosilicate Minerals. Journal of Building Engineering, Volume 19, pp. 205–215
Sebaaly, H., Varma, S., Maina, J.W., 2018. Optimizing Asphalt Mix Design Process using Artificial Neural Network and Genetic Algorithm. Construction and Building Materials, Volume 168, pp. 660–670
Tapkin, S., Çevik, A., U?ar, Ü., 2009. Accumulated Strain Prediction of Polypropylene Modified Marshall Specimens in Repeated Creep Test using Artificial Neural Networks. Expert Systems with Applications, Volume 36(8), pp. 11186–11197
Tasdemir, Y., 2009. Artificial Neural Networks for Predicting Low Temperature Performances of Modified Asphalt Mixtures. Indian Journal of Engineering and Materials Sciences (IJEMS), Volume 16(4), pp. 237–244
Venudharan, V., Biligiri, K.P., 2017. Heuristic Principles to Predict the Effect of Crumb Rubber Gradation on Asphalt Binder Rutting Performance. Journal of Materials in Civil Engineering, Volume 29(8), pp. 04017050
Ziari, H., Amini, A., Goli, A., Mirzaeiyan, D., 2018. Predicting Rutting Performance of Carbon Nano Tube (CNT) Asphalt Binders using Regression Models and Neural Networks. Construction and Building Materials, Volume 160, pp. 415–426
Zuna, H.T., Hadiwardoyo, S.P., Rahadian, H., 2016. Developing a Model of Toll Road Service Quality using an Artificial Neural Network Aproach. International Journal of Technology, Volume 7(4), pp. 562–570