Publicação
Application of Artificial Intelligence techniques: Machine Learning for airport pavement condition index (PCI) assessment
| Resumo: | The evaluation of pavement quality is an essential component of Airport Pavement Management Systems as it supports maintenance strategies aimed at ensuring service levels and the safety of operations and users. In this context, diverse methodologies contribute to assess the overall condition of pavements. However, these methodologies often require a significant amount of time, from conducting in-situ inspections to performing complex calculations. Thus, with the recent 4.0 industry revolution, new digital technologies, namely Artificial Intelligence (AI), has led the concept of Machine Learning to emerge as a highly potential tool for analyzing and processing data obtained from in-situ inspections. This contributes to optimize current methodologies for pavement condition analysis, modernizing Transportation Infrastructure Engineering. Therefore, the application of this technique in airport pavement evaluation helps reduce the time and complexity involved in calculating the Pavement Condition Index (PCI). The reliability degree mainly depends on the size and characteristics of the database used in the modeling. The present work aims to identify the most suitable machine learning algorithms for modeling the PCI index considering surface pavement distress, based on ASTM D5340- 23 standard. To achieve this objective, a pavement distress density and severity database of 261 airport runway sample units was used, considering the PCI in three distinct ways – numerical PCI (ranging from 0-100), categorical PCI with 3 classes (Good, Fair, and Poor), and categorical PCI with 7 classes (Excellent, Good, Satisfactory, Fair, Poor, Very Poor, Failed). Linear regression, decision tree, random forest, artificial neural network, and support vector machine algorithms were tested in WEKA software for three learning processes – use training set, 10-fold cross-validation, and 80% training and 20% testing split. Found results confirmed the models´ capability to output the PCI value based on the density and severity level of pavement surface distress, indicating the feasibility of using Machine Learning-based algorithms for the PCI calculation process, highlighting the random forest algorithm with 10-fold cross-validation as the better performance model. |
|---|---|
| Autores principais: | Studart, André de Paula Pessoa |
| Assunto: | Degradações de Pavimentos Machine Learning Pavement Condition Index (Pci) Pavimentos Aeroportuários |
| Ano: | 2024 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade da Beira Interior |
| Idioma: | inglês |
| Origem: | uBibliorum |
| Resumo: | The evaluation of pavement quality is an essential component of Airport Pavement Management Systems as it supports maintenance strategies aimed at ensuring service levels and the safety of operations and users. In this context, diverse methodologies contribute to assess the overall condition of pavements. However, these methodologies often require a significant amount of time, from conducting in-situ inspections to performing complex calculations. Thus, with the recent 4.0 industry revolution, new digital technologies, namely Artificial Intelligence (AI), has led the concept of Machine Learning to emerge as a highly potential tool for analyzing and processing data obtained from in-situ inspections. This contributes to optimize current methodologies for pavement condition analysis, modernizing Transportation Infrastructure Engineering. Therefore, the application of this technique in airport pavement evaluation helps reduce the time and complexity involved in calculating the Pavement Condition Index (PCI). The reliability degree mainly depends on the size and characteristics of the database used in the modeling. The present work aims to identify the most suitable machine learning algorithms for modeling the PCI index considering surface pavement distress, based on ASTM D5340- 23 standard. To achieve this objective, a pavement distress density and severity database of 261 airport runway sample units was used, considering the PCI in three distinct ways – numerical PCI (ranging from 0-100), categorical PCI with 3 classes (Good, Fair, and Poor), and categorical PCI with 7 classes (Excellent, Good, Satisfactory, Fair, Poor, Very Poor, Failed). Linear regression, decision tree, random forest, artificial neural network, and support vector machine algorithms were tested in WEKA software for three learning processes – use training set, 10-fold cross-validation, and 80% training and 20% testing split. Found results confirmed the models´ capability to output the PCI value based on the density and severity level of pavement surface distress, indicating the feasibility of using Machine Learning-based algorithms for the PCI calculation process, highlighting the random forest algorithm with 10-fold cross-validation as the better performance model. |
|---|