1. UKM Learning and Research Repository
2. PHD Thesis / Tesis PHD
3. Faculty of Engineering and Built Environment / Fakulti Kejuruteraan dan Alam Bina

Title:	Integrated finite element and artificial neural network methods in determining asphalt concrete dynamic modulus |E∗| master curve coefficients
Authors:	Asmah Hamim (P73020)
Supervisor:	Nur Izzi Md. Yusoff, Ir. Dr.
Keywords:	Pavements Asphalt concrete Neural networks (Computer science) Finite element method Universiti Kebangsaan Malaysia -- Dissertations Dissertations, Academic -- Malaysia
Issue Date:	23-Sep-2019
Description:	The dynamic modulus |E*| is among the essential material property inputs in the American Association of State Highway and Transportation Officials (AASHTO) Mechanistic-Empirical Pavement Design Guide (MEPDG). It is also one of the parameters for asphalt concrete stiffness that measure the strains and displacements of flexible pavement structure in loading or unloading condition. An asphalt concrete (AC) dynamic modulus |E*| master curve is employed to characterise the modulus of asphalt concrete in terms of temperature and frequency rates. However, the standard laboratory test procedures for establishing asphalt concrete dynamic modulus |E*| and the plotting the AC dynamic modulus |E*| master curve are time consuming and require considerable resources. Therefore, this research aims to develop a framework for determining the AC dynamic modulus |E*| master curve coefficients by utilising the falling weight deflectometer (FWD) deflection-time history data. Firstly, the Simple Performance Test (SPT) of the dynamic modulus |E*| was conducted in laboratory on the five core specimens to obtain the dynamic modulus |E∗ | data at several test temperatures and load frequencies. Following this, ANSYS R18.0, which is general-purpose finite element (FE) software, was used to develop the FE models for 132 locations of the FWD test to generate deflection-time history data. A two dimensional (2D) axisymmetric with 5000mm × 5000mm model geometry was selected for developing the FE models by considering static and dynamic FWD loading conditions, viscoelastic properties of asphalt concrete layer, and by following the field FWD configuration. The results of the FE models show that the difference percentage root mean squared error (RMSE) for peak deflection basin value between field measurement and FE model are small and range between 0.45 and 7.45%. Finally, artificial neural network (ANN) models were designed employing the FWD deflection-time history data generated by the finite element method (FEM) to determine the AC dynamic modulus |E*| master curve (C1, C2, C3 and C4) coefficients. The two types of ANN models employed in this research are the multilayer feedforward neural network (MLFN) and the radial basis function network (RBFN), and the following three case studies are taken into considering: 1) case study I considers only D1 deflection-time history, 2) case study II considers D1, D2 and D3 deflectiontime-history, and 3) case study III considers the difference between the magnitudes of D1 and D2 deflection-time history as the main input data of the ANN model. D1, D2 and D3 are the registered FWD deflection-time history data recorded at 0, 300 and 600mm offsets from the center of the FWD loading plate, respectively. A total of 132 databases were used to develop the ANN models. The results of the ANN show that both MLFN and RBFN models have a good potential in determining the AC dynamic modulus |E*| master curve coefficients with coefficient of determination (R 2 ) values greater than 0.9. A comparison of the two types of ANNs reveals that, for case study I in particular, RBFN has a higher of R 2 of 0.957 and lower MAE, MSE and RMSE values, and is therefore more accurate than MLFN.,Ph.D.,PENYEPADUAN KAEDAH UNSUR TERHINGGA DAN RANGKAIAN SARAF TIRUAN DALAM MENENTUKAN PEKALI-PEKALI LENGKUNG INDUK MODULUS DINAMIK |E∗ | ASFALT KONKRI
Pages:	206
Call Number:	TE278.A836 2019 3 tesis
Publisher:	UKM, Bangi
Appears in Collections:	Faculty of Engineering and Built Environment / Fakulti Kejuruteraan dan Alam Bina

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.