Research On Expressway Asphalt Pavement Maintenance Management And Optimization Decision-making

As China’s highway mileage continues to increase,the amount of pavement maintenance work is also increasing year by year.However,maintenance funds are limited.Hence,the rational use of maintenance funds has become a key issue for pavement maintenance.The current practice at home and abroad is to propose a set of pavement maintenance optimization decision approaches to develop pavement management systems.However,the current pavement maintenance optimization decision-makig approaches involve a large number of empirical results,insufficient data analysis and utilization,and other limitations,such as the lack of holistic consideration of the pavement management data cleaning process and the neglect of construction length requirements for preventive maintenance projects in the maintenance decision process.To address these issues,this study proposed a data analysisbased approach to pavement maintenance management and optimization decision-making.The main research content and conclusions are as follows.Firstly,five pavement-related data were collected and combined into a single pavement data.Analysis of this data revealed the presence of missing pavement maintenance project’s data,and anomalous pavement condition detection data.As these two types of data affect each other,the abnormalities of the pavement condition data were ignored and a neural networkbased model for filling in the maintenance project’s data was proposed.Then a pavement condition anomaly cleaning model was established based on the interquartile range method and reliability analysis.Based on the above models,the cleaned pavement management data were obtained.Secondly,information on unmaintained pavement sections was extracted from the cleaned pavement management data,and the explanatory variables affecting performance deterioration were selected using the correlation coefficient method.As there is uncertainty in pavement performance deterioration,Bayesian theory was combined with artificial neural networks to construct a probabilistic prediction model of pavement performance deterioration based on Bayesian neural network(BNN).The probabilistic models are divided into two types according to the approximate inference methods,namely probabilistic weights and structures.A deterministic prediction model based on artificial neural networks was constructed to provide the prior probability distribution of weights and the type of variational distribution for the BNN model with probabilistic weights.Comparing and analyzing the three deterioration prediction models and the Markov-based probabilistic model,the optimal model based on the BNN with probabilistic structures was concluded.Thirdly,information on the maintained pavement sections was extracted from the cleaned pavement management data to establish a post-maintenance road age change and pavement condition jump method.A post-maintenance road age adjustment of 0 means that the maintained road section is equivalent to a new road.Based on this idea and the concept of reliability,a rule was proposed to determine whether the road age is adjusted to 0 or remains constant.The pavement condition jump also involves uncertainty,so a pavement condition probabilistic jump model was constructed based on the theory of BNN with probabilistic model structures.Fourthly,based on the above pavement performance deterioration prediction model,postmaintenance road age change and pavement condition jump methods,probabilistic and deterministic optimization decision models were constructed for pavement maintenance at section level and network level respectively with reference to the two-stage bottom-up decisionmaking approach.When an initial maintenance action is fixed to a road section,the sectionlevel optimization model is able to search for the best treatment path and its corresponding effectiveness-cost ratio(E/C).The network-level pavement maintenance optimization model selects an initial maintenance action for each pavement section.Finally,this study incorporated the construction length requirement as a constraint or objective into the network-level maintenance optimization model,which constitutes the constrained and bi-objective network-level decision models respectively.The constrained network-level decision model is more difficult to solve,and this study proposed a sliding window random repair method,which can be combined with a genetic algorithm to solve the constrained network-level decision model.Comparing the constrained and bi-objective network-level decision models,the former model is more efficient and provides a maintenance plan for any construction length constraint value,but the recommended maintenance plans are inferior to the latter.The latter model is less efficient and may not yield a maintenance plan for a given construction length constraint.By way of examples,comparing the probabilistic and deterministic network-level road maintenance plans,the former has a lower E/C than the latter;comparing the two maintenance plans that considers or ignores the construction length requirement,the former also has a lower E/C than the latter.These results suggest that if one of the practical engineering issues – the uncertainty of pavement condition deterioration and jump,and the construction length requirement – is ignored,the network-level road maintenance plans will mislead engineers to overestimate the expected effectiveness.The probabilistic network-level plan incorporating the construction length requirement is also compared to actual maintenance projects,with the former expected to achieve higher levels of performance at a lower cost.These results demonstrate the superiority of the probabilistic optimization decision models for network-level road maintenance incorporating construction length requirements(the main model).This study addressed practical problems in pavement maintenance management systems,such as data cleaning and ignoring construction length requirements,and proposed models for pavement management data cleaning,pavement condition deterioration,road age change and pavement condition jump,and then incorporated the construction length requirement into network-level road maintenance decision model based on a two-stage bottom-up decisionmaking approach,forming a set of pavement maintenance management and optimization decision-making system.The relevant research results can provide a theoretical level reference for constructing or improving pavement maintenance management systems,and have a positive effect on the rational use of limited maintenance funds and the formulation of practical maintenance plans.