الفهرس 
Chapter 1: IntroductionOnly 14 pages are availabe for public view
 |  | from | 252 |  |  |
| | | from | 252 | | |
Abstract
Recently, all efforts have been directed not only at the construction of new roads but also at maintaining the existing road network’s high performance by implementing the pavement maintenance management system (PMMS) in order to evaluate the pavement condition and select the most cost-effective maintenance decisions after conducting an economic analysis. The implementation of PMMS is very crucial for determining the appropriate treatment strategies for flexible pavements by maintenance or rehabilitation (M&R), and although it is often considered a key process in the PMMS system, it is also considered a complex process. Previously, many highway agencies relied on the pavement condition index (PCI) to evaluate the pavement condition and determine the M&R strategies through some assumptions, which differed from one agency to another, called a ”decision tree.” The PCI is considered a good indicator for evaluating pavement condition but not for determining appropriate maintenance or rehabilitation (M&R).
A simplified, innovative decision tree system called the Maintenance Unit (MU) system was developed through a research programme conducted at Cairo University, Egypt. In this system, M&R activities are directly determined by the overlapping of the density of localised maintenance for each distress (rather than the density of a single distress). Also, through research published previously in the International Journal of Pavement Engineering, a maintenance decision model (MDM) using the MU system was developed to forecast future maintenance decisions. The Egyptian Code of Practice has recently relied on the maintenance unit (MU) concept for maintenance decision prediction. Because this kind of decision system requires a lot of mental work, it must be set up automatically using artificial intelligence (AI).
Therefore, this research aims first to apply the latest machine learning technique for forecasting the current and future pavement maintenance decisions based on the MU and MDM systems according to the Egyptian code considerations to develop a one-step enhanced decision-making tool. Secondly, to perform an economic analysis for the decisions that were predicted from the artificial neuronal networks (ANN) models and then determine the most cost-effective maintenance decision to achieve the goal, a pattern recognition algorithm (a neural network) was applied to 54.3 km of surveyed roads in Minia Governorate, Egypt. The selected roads belong to two institutions. The first is the General Authority for Roads and Bridges and Land Transport (GARBLT), and the second institution is the Roads and Transportation Directorate, Minia Governorate. Five roads have been selected for this study, with a total length of 54.3 km. The selected roads are as follows:
• Minia - Western desert highway, with a specified length of 3400 m. (GARBLT)
• El Mohit Road 53, with a specified length of 20200 m. (Roads and Transportation Directorate)
• El Gomhorya road, with a specified length of 1900 m. (Roads and Transportation Directorate)
• Minia Ring Road, with a specified length of 20300 m in two directions. (GARBLT)
• Cairo - Aswan agriculture highway, with a specified length of 8500 m in two directions. (GARBLT)
Based on the MU system and the MDM model, three ANN models have been developed to recommend current maintenance decisions (CMDM), future maintenance decisions for the roads of the General Authority for Roads, Bridges, and Land Transport (FMDMG), and future maintenance decisions for local roads (FMDML). The forecasted maintenance decisions determined using the CMDM model were compared to the actual current maintenance decisions calculated using the MU system for 44 different samples in the design cases database, which were not taken in the training process. The average accuracy of all samples in the testing set was determined to be 97.1%. Also, the average accuracy rate for the testing set is 96%, 94%, 93%, 93%, and 100% when FMDMG or FMDML is used to predict maintenance decisions after 2, 5, 10, 15, and 25 years.
The CMDM, FMDMG, and FMDML models were used to determine each section’s preliminary maintenance decisions, and then feasible maintenance alternatives were generated for each preliminary maintenance alternative. All of this was used to perform an economic analysis to determine the most cost-effective maintenance decision, considering the rate of inflation in prices and the rate of discount in the value of the currency over time, which were 9.60% and 13.25%, respectively, according to the statistics of the Central Bank of Egypt for the current year 2022.
For the current state, it was the most cost-effective decision: reconstruction of AC layers for sections A3, A5, A6, A10, A12, and A15; and slurry seal for sections A1, A4, A8, A9, and A11; and thin overlay without milling for sections A2, A14, and A17; and distress-by-distress for sections A13 and A7; and sand seal for section A16. While the reconstruction of AC layers for sections A2, A3, A5, A6, A11, A14, A15, and A17 will be the most cost-effective maintenance decision when deferring maintenance for two years; and a thin overlay without milling for sections A1, A4, A8, and A9; sand sealing for section A13; slurry sealing for sections A7 and A16; and reconstruction up to subgrade for sections A10 and A12.Also, when maintenance was put off for five years, the most cost-effective decisions for maintenance were to rebuild the AC layers in sections A1, A2, A4, and A14, use a thin overlay without milling in sections A7 and 16, use a slurry seal in section A13, and rebuild other sections up to the subgrade.
The results of this study indicated that using artificial neural networks to recommend current and future maintenance decisions for flexible pavements is a helpful and essential tool. These applications can simplify the decision-making process, and the economic analysis of the predicted decisions can save a lot of money.