| Asphalt concrete pavement is the basic pavement structure of highways in China,with high driving comfort,easy repair and other functional characteristics.However,with the increase of service time,asphalt pavement will produce a variety of diseases,including longitudinal and transverse cracks,which will not only affect the comfort of driving,but also further cause a variety of other diseases,and even lead to the overall destruction of the pavement structure.However,the crack evolution model based on physical calculation and analysis cannot meet the needs of engineering practice,and the relevant influencing factors that can be considered are limited.In view of this,this paper predicts the change of length and number of longitudinal and transverse cracks in asphalt pavement during service.On the basis of investigation and comprehensive analysis of the main factors affecting the development of longitudinal and transverse cracks,the corresponding data are selected from the long-term pavement performance database,and the input characteristics are subject to correlation analysis,gray correlation analysis and random forest importance ranking.An artificial neural network prediction model for the length of longitudinal and transverse cracks and the number of transverse cracks is established.Finally,the model is optimized according to the ranking results of importance.The prediction model of longitudinal and transverse cracks established in this paper can provide some reference for road operation management and scientific maintenance.The main contents of this paper are as follows:(1)Investigate the basic conditions of expressway pavement in Zhejiang Province and the causes of cracks.The basic construction conditions of the investigated pavement are basically the same.The cracks account for the largest proportion of the diseases,and the longitudinal and transverse cracks are the main part of the cracks.The generation and development of cracks are closely related to the material structure,traffic volume,temperature and rainfall of asphalt pavement.The specific variable data are selected from the long-term pavement performance database,and the processed data are statistically analyzed.The analysis results show that the sample data range includes the expressway data of Zhejiang Province.(2)Analyze the relationship between transverse and longitudinal cracks and their influencing factors.The results of correlation analysis show that there is a moderate correlation between the historical disease information and the existing cracks on the road,and a weak correlation with age,traffic and climate information.The gray correlation analysis shows that the historical disease information has the highest gray correlation with the current state of the cracks,the gradation parameter is the lowest.The random forest model has great prediction effect,reliable feature importance ranking,among which historical disease information is the most important for modeling.(3)The relationship between longitudinal and transverse cracks and their influencing factors is analyzed.The results of correlation analysis show that historical diseases have the greatest impact on disease development,followed by time,traffic and climate information;The results of grey correlation analysis show that the grey correlation degree between historical diseases and the current state of cracks is the highest,followed by the external environmental information,and the grading parameters are the lowest;The prediction effect of random forest model is good.The ranking result of feature importance is that the historical disease information has the highest importance to the model.The external environment,time and surface layer thickness have a great impact on the disease development.(4)The prediction model is optimized by the results of feature importance ranking.According to the random forest importance ranking,the input parameters of different models are successively deleted.After the prediction models of transverse crack length,number and longitudinal crack length are respectively deleted by 2,3 and 5 low importance degrees,the characteristic model has the best effect;After deleting 9,6 and 8 low importance features respectively,the model can be simplified while keeping R2 higher than 0.80.The expressway data of Zhejiang Province is applied to the optimization model,and the overall prediction results are good. |
