Topic Model for Maintenance Record Classification

Abstract

Product maintenance records contain a lot of useful information for improving design and maintenance of the product. However, it is cumbersome to extract useful information form them because many of them are written in natural language. In this study, we propose a method to estimate the topics in the maintenance records. In the maintenance records, it is common that more than one topic such as a type of failure, a cause of failure and a repair method are written. In addition, there are usually causal relationships between those topics. In this study, modeling these features as a topic model, we propose a method for estimating the topics of multiple viewpoints described in maintenance records. We applied the proposed method to maintenance records of analytical systems, and it was confirmed that, in about 80% of the maintenance records, correct topics are included in the top three items of the estimated topic.