Variable indexing method in rule documents for ship design using extraction of portable document format elements

Abstract

Design rules for ships have become more extensive and detailed due to an increase in the sizes of ships. Several variables and equations used in the rules are complex, thereby impeding their review by reviewers due to their voluminosity. In addition, because these rules are constantly revised, professional investigators may miss these changes. To prevent such confusion, a shipping register, which approves ship drawings, constantly automates the search and review processes of the rules. Consequently, this study proposes a method for recognizing variables in documents to review the rules and build relationships between variables. Each component of a document must be accurately identified. The document containing these rules includes different components such as equations, figures, and strings. Because these rules are mainly converted to a portable document format (PDF) for compatibility, it is challenging to extract each component as raw data. This study used a public library to extract elements from the PDF and utilized the positional relationship between the elements to identify the variables. By applying the Levenshtein distance algorithm, which compares the differences between two strings, the document was partitioned following to the table of contents. Hence, the identified variables were indexed into sections of the table of content. Additionally, based on the indexed information, a data structure was proposed to show the equations, definition of variables, and relationships. This study applied it to common structural rules, which are widely used in the shipbuilding industry. The effectiveness of the proposed method was confirmed by achieving the F1 score = 0.93 in variable recognition and intuitively visualizing the relationship between the variables.