A Study on Knowledge Extraction Methods from Equipment Maintenance Documents

In enterprises operating large-scale equipment, such as plant enterprises, maintenance workers must quickly and accurately find and understand the information required in the equipment maintenance documents to perform maintenance tasks effectively. If the equipment maintenance documents exist in each file for each equipment and the sentence expression constituting each document is ambiguous, it will interfere with the effective performance of the maintenance, and it leads to loss of the company. In order to solve these problems, attempts have been made to efficiently manage equipment maintenance documents and fault documents and extract key information or maintenance knowledge. However, they have the limitations of not quantitatively presenting the effectiveness of the proposed method or considering the relationship between the entities. Therefore, in this paper, we propose a method for effective maintenance knowledge extraction by extracting entities for equipment, failures, and solutions from equipment maintenance documents through named entity recognition, and further building a set of relationships between individual entities using dependency parsing. Equipment maintenance documents used in domestic plant enterprises were used to show validation of the proposed approach, and 74.3% of correct relations were found for the test sentences.