ABSTRACT

The permanent deformation of asphalt pavement under similar traffic conditions depends on a number of factors. The factors considered in this study are bitumen source, aggregate source, aggregate gradation, bulk specific gravity of aggregates (G_sb), percentage of aggregates passing #4 sieve, air voids (Va), optimum bitumen content, binder grade, load repetitions, temperature, and Marshall stability. Asphalt pavement analyzer (APA) test, Cooper wheel tracking test (CWTT), and repeated load axial test (RLAT) were performed on thirteen different types of hot mixed asphalt (HMA) mixtures. Three artificial neural network (ANN) algorithms, namely Backpropagation (BP), Conjugate gradient (CG), and Broyden-Fletcher Goldfarb-Shanno (BFGS) were used to analyse the data. The best fit ANN algorithm for each of the laboratory tests (APA, CWTT, RLAT) was selected, based on the coefficient of determination (R-squared), root-mean-square error (RMSE), mean bias error (MBE) and the mean square error (MSE) closest to the gamma statistic Г. The results showed no single ANN algorithm is suitable for predicting all HMA rutting susceptibility tests data. The BP algorithm most appropriately predicts APA test data, the BFGS algorithm precisely fits CWTT results, and the CG algorithm seems most suitable to predict RLAT data. However, further, differentiating testing is required for a more precise comparison of rutting predicting ability of various ANN algorithms.

摘要

类似交通条件下沥青路面的永久变形取决于多种因素。本研究考虑的因素有沥青来源、骨料来源、骨料级配、骨料体积比重（G _sb)、通过#4 筛子的骨料百分比、气孔 (Va)、最佳沥青含量、粘合剂等级、负载重复、温度和马歇尔稳定性。对 13 种不同类型的热拌沥青 (HMA) 混合物进行了沥青路面分析仪 (APA) 测试、库珀车轮跟踪测试 (CWTT) 和重复载荷轴向测试 (RLAT)。三种人工神经网络 (ANN) 算法，即反向传播 (BP)、共轭梯度 (CG) 和 Broyden-Fletcher Goldfarb-Shanno (BFGS) 用于分析数据。根据决定系数 (R-squared)、均方根误差 (RMSE)、平均偏差误差 (MBE) 和最接近 gamma 统计量 Г 的均方误差 (MSE)。结果表明，没有单一的人工神经网络算法适用于预测所有 HMA 车辙敏感性测试数据。BP 算法最适合预测 APA 测试数据，BFGS 算法精确拟合 CWTT 结果，而 CG 算法似乎最适合预测 RLAT 数据。然而，进一步地，需要差异化测试来更精确地比较各种人工神经网络算法的车辙预测能力。