Diagnosis-Based Domain-Adaptive Design Using Data Augmentation and Transfer Learning

Recently, research on PHM diagnosis technology has been actively conducted. However, existing diagnostic techniques focus on qualitative analysis such as anomaly detection and cause classification for maintenance of equipment. Therefore, quantified diagnostic solutions are required to enable users to take action in advance. In this study, we propose a diagnostic design solution, from a system design perspective, using domain adaptation and designable data augmentation. If a system is newly designed similar to the previously developed system, the uncertainty of diagnostic knowledge of the existing system is quantified and adapted to the new system through transfer learning. Furthermore, deep learning design algorithms allow us to estimate system design solutions with a small number of data. To verify the proposed method, a case study was conducted using mathematical models and Modelica-based physical system models. Consequently, we validated that the estimated design solution obtained a higher system normal probability than that by the initial design.