Quality Evaluation on Compaction of Asphalt Pavements Based on Machine Learning
To improve accuracy of asphalt pavement compaction prediction,quality evaluation on compaction asphalt pavements was conducted on the basis of machine learning.By comparing vibration patterns of different rollers on asphalt pavements in practical engineering,a dynamical model of vibration roller-asphalt pavement system was constructed.The constructed model was simulated by using Simulink numerical simulation software,and eight working conditions were designed to compare applicability differences of six quality evaluation indicators of compaction under different working conditions including compaction meter value,compaction control value,machine drive power,roller-integrated stiffness,filling energy,and unit volume compaction power.Five traditional machine learning methods,namely support vector machine,logistic regression,k-nearest neighbor,decision tree,and naive Bayes,were used to train quality evaluation indicator samples of compaction under different working conditions,and prediction accuracy of compaction was compared.Rolling routes of the roller with different rolling times were designed to compare predict accuracies of compaction using the optimal quality evaluation indicators of compaction and the single quality evaluation indicators of compaction.The results show that the applicability of different quality evaluation indicators of compaction is different under each working condi-tion.Even under the same working condition,the applicability with different rolling times is also different.Using the op-timal quality evaluation indicators of compaction to evaluate indicators instead of the single quality evaluation indicators of compaction,the prediction accuracy of compaction is increased by 5.8%.Among the five traditional machine learning methods,naive Bayes method has the highest accuracy of predicting types of the optimal compaction quality evaluation indicators,which is 96.22%.
asphalt pavementquality evaluation on compactionmachine learningcompactiondynamical model
NETL
NSTL
万方数据
