Unknown fault-category detection with CGSA

Investigate whether and how the Condition-Guided Selective Adaptation framework can identify previously unknown fault categories, while evaluating its robustness in realistic industrial scenarios and more diverse operating environments.

Background

The paper develops the Condition-Guided Selective Adaptation framework for fault diagnosis under previously unknown operating conditions. Its experiments evaluate adaptation to distribution shifts involving known healthy and faulty categories, but they do not establish the framework’s ability to recognize fault categories absent from offline training.

The authors explicitly identify unknown fault detection as requiring further investigation and designate it as future work, together with robustness evaluation under more realistic industrial conditions and a wider range of operating environments.