Pavement deterioration prediction models in the absence of archived data

Abstract

Strategic and long-term planning in Pavement Management System (PMS) relies mainly on deterioration prediction models, to ensure efficient and forward-looking management and for setting present and future budget requirements. PMS consist of many essential network and project levels activities. Modeling of pavement deterioration is one of the most important PMS network level activities that must be coordinately executed with other activities to have a functional PMS. Obviously, preventive maintenance of a pavement is less expensive than reconstructing it after being deteriorated. In many developing countries, roads face increasing damage because of the lack of regular maintenance. This reinforces the need to develop a system to predict the deterioration of roads in order to determine the Optimal Intervention Strategies (OIS) for the road network. Additionally, analyzing the progression of pavement deterioration over time enables better understanding for the pavement functional behavior to efficiently support a PMS. In general, pavement deterioration models can be developed deterministically or probabilistically. Under normal circumstances, pavement deterministic deterioration modelling requires regular measurements of the pavement condition over time. However, in the absence of such information and records in many cases such as in developing countries, such method cannot be used, and alternative is to use probabilistic modeling. This research presents three methodologies to predict and analyze pavement condition and its progression when archived pavement indices data is not available. First suggested method is a probabilistic approach of Bayesian linear regression to develop a deterioration model when archived data about pavement history is not available. Instead, the model uses expert knowledge as a prior distribution. As such, experts who have worked for a long time with the road and transportation agencies have been interviewed to develop a portion of the input data to feed the Bayesian model. The posterior distribution was calculated using the likelihood estimation function based on road condition inspections according to a predefined protocol. In this study, model parameters were estimated, and 95% confidence intervals established around these estimated parameters. The results are forecasting models of pavement deterioration prediction model based on a mixture of few on-site inspections interacting with expert knowledge. Second method is a probabilistic technique to analyze the progression of pavement deterioration over time to enable better understanding for the pavement functional behavior to efficiently support a PMS. The aim of this method is to investigate and forecast the trends of the pavement deterioration using Autoregressive Integrated Moving Average (ARIMA) time series analysis. Data used in this study is the International Roughness Index (IRI) which was estimated using visual pavement inspections. Models’ appropriateness are evaluated by plotting the residuals of Auto Correlation Function ACF and Partial Auto Correlation Function PACF. All residuals were within the bands which meant that all models are appropriate. Ljung–Box (Q) statistics was applied for all models and the results showed that all model coefficients are not significantly different from zero. The third method was to develop a deterioration prediction model to anticipate future pavement conditions using a Hidden Markov Model (HMM). This study explained how to estimate the most likely sequence of pavement condition states that a specific pavement goes through to failure, using 10 transitioning steps to run the pavement HMM. Viterbi algorithm was used to compute this sequence of pavement condition states. An initial database representing the pavement condition for a given period of time is used in the development process. Transition probabilities and emission probabilities are also calculated. The third method was to develop a deterioration prediction model to anticipate future pavement conditions using a Hidden Markov Model (HMM). This study explained how to estimate the most likely sequence of pavement condition states that a specific pavement goes through to failure, using 10 transitioning steps to run the pavement HMM. Viterbi algorithm was used to compute this sequence of pavement condition states. An initial database representing the pavement condition for a given period of time is used in the development process. Transition probabilities and emission probabilities are also calculated.

Date	29 Jun 2022
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Gabriel J. Assaf (Supervisor)