Predicting Marshal Flow and Marshal Stability of Asphalt Pavements using Multi Expresion Programming

Abstract

The most widely used method for mix design of asphalt pavements in Pakistan is Marshall Mix Design Method (MMDM) which is based on Asphalt Institute MS-2 in accordance with National Highways Authority’s general specifications. The type and amount of bitumen, as well as the grading characteristics of aggregates, dictates the MMDM. The traditional method of obtaining optimum bitumen content and the relevant parameters entails time-consuming, complicated and expensive laboratory procedures and require skilled personnel. Likewise, it is becoming increasingly vital to use new and advanced methodologies for the design and quality control of Marshall parameters. Therefore, this research study uses innovative and advance machine learning technique named Multi Expression Programming (MEP) to develop empirical predictive models for the Marshall parameters i.e. Marshall Flow (MF) and Marshall Stability (MS) for Asphalt Base Course (ABC) and Asphalt Wearing Course (AWC) of flexible pavements. The comprehensive, reliable and wide range of datasets from various road projects of Pakistan were produced for MMDM. The collected datasets contain the 253, and 343 results of MMDM for ABC and AWC, respectively. Eight input parameters were considered for modeling the output parameters i.e. MS and MF. The overall performance of the models was evaluated using statistical measures such as mean absolute error (MAE), root mean square error (RMSE), correlation coefficient (R), relative root mean square error (RRMSE), performance index (ρ), root square error (RSE), and objective function (OF). The relationship between input and output parameters was determined by performing parametric analysis, and the results of trends were found to be consistent with earlier research findings stating that the developed predicted models are well trained. The results revealed that developed models are superior and efficient with respect to prediction and generalization capability for output parameters of MMDM as evident by R (in this case>0.90) for both ABC and AWC.