Clustering-Based Similarity Detection of Pavement Segments Considering Multiple Contributors to Deterioration: Fid-Bau Portal

Clustering-Based Similarity Detection of Pavement Segments Considering Multiple Contributors to Deterioration

Karimzadeh, Arash / Sabeti, Sepehr / Shoghli, Omidreza

Transportation asset managers leverage predictive analytics to optimize highway maintenance by lowering maintenance expenses and enhancing performance over the life span of assets. Deterioration prediction approach—a major element of life cycle planning—can highly impact the accuracy and effectiveness of prioritizing maintenance tasks. Forecasting future conditions is often a data-driven process based on historical condition measurements, the history of maintenance activities, and contributing factors to degradation. Therefore, inadequate data on all road segments leads to some shortcomings in predictive analytics and limits scalability and extensibility of the models. To resolve these problems, most research studies used family groups of assets with similar deterioration characteristics to investigate degradation. Subjective classifications are the most commonly used techniques to find similarities between asset items at different locations and to categorize them into family groups. The subjective categorization becomes a challenge when a large variety of contributing factors are considered. To tackle this challenge, we used a clustering approach. By minimal human intervention, clustering technique uses fewer assumptions about inter/intra correlations in dataset. We used k-means and agglomerative hierarchical algorithms to extract similarities over a large set of road segments. We implemented the algorithms on a case study in Virginia. We considered a range of factors, including weather and traffic, to identify family groups of flexible pavement segments. We additionally determined the optimal number of clusters by using the elbow and average silhouette score methods. Since the data clustering was accomplished by minimal subjective assumptions, the outcomes of the study provide insights into reliable condition predictions and can be used to develop prediction models for each family group. Consequently, the results of this research have potential benefits to improve the maintenance planning practices for states’ Department of Transportation (DOT).