Search results

Current practices for inspecting highway construction predominantly rely on manual processes, which result in subjective assessments, errors and time inefficiencies. The purpose of this study is to address the inefficiencies and potential inaccuracies inherent in manual highway construction inspections. By leveraging computer vision and ontology reasoning, the study seeks an automated and efficient approach to generate structured construction inspection knowledge in the format of checklists for construction activities on highway construction job sites.

This study proposes a four-module framework based on computer vision and ontology reasoning to enable the automatic generation of checklists for quality inspection. The framework includes: (1) the interpretation of construction scenes based on computer vision, (2) the representation of inspection knowledge into structured checklists through specification processing, (3) the connection of construction scenes and inspection knowledge via ontology reasoning and (4) the development of a prototype for the automatic generation of checklists for highway construction.

The proposed framework is implemented across four distinct highway construction scenarios. The case demonstrations show that the framework can interpret construction scenes and link them with relevant inspection knowledge automatically, resulting in the efficient generation of structured checklists. Therefore, the proposed framework indicates considerable potential for application in the automatic generation of inspection knowledge for the quality inspection of highway construction.

The scientific and practical values of this study are: (1) the establishment of a new method that promotes the automated generation of structured inspection knowledge for highway construction by integrating computer vision and ontology reasoning and (2) the development of a novel framework that provides efficient and immediate access to inspection knowledge related to what needs to be inspected at highway construction job sites.

To develop an automated system for identifying and repairing cracks in asphalt pavements, addressing the urgent need for efficient pavement maintenance solutions amidst increasing workloads and decreasing budgets.

The research was conducted in two main stages: Crack identification: Utilizing the U-Net deep learning model for pixel-level segmentation to identify pavement cracks, followed by morphological operations such as thinning and spur removal to refine the crack trajectories. Automated crack repair path planning: Developing an enhanced hybrid ant colony greedy algorithm (EAC-GA), which integrates the ant colony (AC) algorithm, greedy algorithm (GA) and three local enhancement strategies – PointsExchange, Cracks2OPT and Nearby Cracks 2OPT – to plan the most efficient repair paths with minimal redundant distance.

The EAC-GA demonstrated significant advantages in solution quality compared to the GA, the traditional AC and the AC-GA. Experimental validation on repair areas with varying numbers of cracks (16, 26 and 36) confirmed the effectiveness and scalability of the proposed method.

The originality of this research lies in the application of advanced deep learning and optimization algorithms to the specific problem of pavement crack repair. The value is twofold: Technological innovation in the field of pavement maintenance, offering a more efficient and automated approach to a common and costly issue. The potential for significant economic and operational benefits, particularly in the context of reduced maintenance budgets and increasing maintenance demands.