Skip to Main Content
Most industrial products are developed based on their former products including software. Revising existing software according to new requirements is thus an important issue. However, innovative techniques for software revision cannot be easily introduced to projects where software is not a central part. In this paper, we report how to explore and apply software engineering techniques to such non-ideal projects to encourage technology transfer to industry. We first show our experiences with industrial partners to explore which tasks could be supported in such projects and which techniques could be applied to such tasks. As a result, we found change impact analysis could be technically supported, and traceability techniques using information retrieval seemed to be suitable for it. We second had preliminary experiences of a method using such techniques with data in industry and evaluated them with our industrial partners. Based on the evaluation, we third improved such a method by using following techniques, indexing of technical documents for characterizing requirements changes, machine learning on source codes for validating predicted traceability and static source code analysis for finding indirect impacts. Our industrial partners finally evaluated the improved method, and they confirmed the improved method worked better than ever.