Implementation of ensemble Artificial Neural Network and MEMS wireless sensors for In-Situ asphalt mixture dynamic modulus prediction

Abstract

Dynamic modulus is an essential mechanical property for pavement structural condition assessment yet obtaining the in-situ dynamic modulus of asphalt mixture efficiently and accurately has never been easy. With the development of sensing technology, a type of Micro-Electromechanical Sensor (MEMS), SmartRock, has been applied in pavement monitoring because of its convenience and high efficiency; but interpreting the collected data as useful engineering information becomes a new challenge. This paper aims to present a data processing algorithm for sensing data in pavement engineering and develop a dynamic modulus predictive model based on the ensemble artificial neural network (ANN) model using SmartRock sensing data. Five asphalt mixtures were tested by the uniaxial dynamic modulus test, and one of the 5 mixes was also used in the third-scale model mobile load simulator (MMLS3) test. SmartRock sensors were installed in the specimens to collect the mechanical response during the tests. The empirical mode decomposition (EMD) method was used for data preprocessing. Dynamic modulus test data, in combination with a small portion of early-stage MMLS3 sensing data, were used as the training dataset, while the remaining MMLS3 data were used as the testing dataset. The results show that the ensemble ANN model is feasible and robust in predicting the dynamic modulus of the asphalt mixture in the MMLS3 test. The parametric analysis shows that the ensemble ANN model has reliable stability. Future studies are recommended to expand the database with more material variety and loading conditions for model verification.