PAVEMENT ASSET MANAGEMENT INCORPORATING MACHINE LEARNING MODEL

Abstract

The authors have developed a pavement Asset Management System (PAMS) based on life cycle cost analysis and optimisation methods. The project was analysis based, which required data. This data was acquired from NHA, including SN (Structural Number), IRI (International Roughness Index), and AADT (Annual Average Daily Traffic). The project can be divided into three modules. Module number 1 covers vehicle operating cost (VOC), which was calculated with the help of Hep-Burn method (1991). Traffic data was required for module 1, which was acquired from NHA. Calculating the VOCs gave us an idea about the overall mileage and wear and tear cost currently incurred on vehicles due to present pavement conditions. So, the current VOC and the VOC after maintenance can be compared after any future maintenance. Module number 2 covers the strategies required to improve the road conditions. The conventional PAMS can be extended to the extraction and management of raw material and end-of-life materials by integrating life cycle assessment and cost analysis. The authors used a lifecycle optimisation model to determine the near-optimal pavement preservation strategy. In this thesis, we propose a network-level PAMS to help decision-makers maintain a healthy pavement network while minimising cost and lifecycle energy consumption within budgetary and other agency constraints. Module Number 3 encompasses the use of machine learning to minimise equipment usage. Road asset management requires an understanding of the deterioration of roads. A machine-learning algorithm was used to predict the deterioration of the International Roughness Index (IRI) based on long-term pavement performance data. For this analysis, we looked at readily available attributes. Municipalities with limited budgets or expertise may benefit from this approach. A large company or transportation agency could save significant money by eliminating the costs of collecting data in the field and the related safety hazards and risks. This approach can be helpful for smaller communities that may not have the money or expertise. Moreover, for larger ones and transportation departments, it could save the ever- increasing cost of collecting field data and any related safety risks. Further, we investigated the importance of data analytics in the management of assets by using this x attribute category. Without considering a casual model, can the trend of the data be used to predict IRI deterioration? The algorithms used were the simplest in machine learning – the linear regression model and the accuracy we saw was high provided with the limited data we had, with some reaching above 85%, which shows that the model can be relied on furthermore, due to the division of data into different regions, the importance of variables and the accuracy of the model changes. Reducing the number and size of prediction classes (level of deterioration) improved accuracy while increasing the span of prediction decreased accuracy. The model automatically calculates and predicts IRI and presents informative or important features for prediction.