About the research
To extract the necessary data and features from the existing inspection reports, a manual process is currently in place. This means that, even for a simple query, the engineer/staff member in charge needs to browse several pages, locate appropriate details, determine quantities, and note down the required information. This task becomes further demanding if the scope of the query is extended to several bridge elements, involving a post-processing of various sources of inspection information. However, with the advancements made in machine learning and artificial intelligence (ML/AI), new opportunities have emerged to move from a manual to an automated process for data and feature extraction from bridge plan sets and inspection reports. Thus, this research project aims to develop the first-known computational platform to automate the process of extracting defects from available inspection records with the ultimate goal of quick delivery of high-quality information about the condition state of bridges. The main features of this platform include distinguishing different physical objects, determining their boundaries and dimensions, detecting various signs of defect, and finally providing qualitative and quantitative assessments of defect for the bridge elements of interest in desired output formats. Further to the listed features, an important capability of this platform is that, after training and quality control, it can work with no immediate supervision, minimizing the time and effort required to plan preventive and corrective actions for bridge structures.