Known classes aware and emerging unknown classes rejection based on adversarial training for open set fault diagnosis

Area of Science:

• Machine Learning
• Artificial Intelligence
• Fault Diagnosis

Background:

• Traditional domain adaptation methods assume identical label spaces, failing with novel fault states.
• Open-set domain adaptation is challenging due to partially overlapped label spaces between source and target domains.

Purpose of the Study:

• To propose an approach for open-set domain adaptation in fault diagnosis that handles emerging unknown fault states.
• To improve the accuracy and robustness of diagnostic systems in real-world applications.

Main Methods:

• Introduced an adaptive weighted learning scheme based on entropy to the maximum classifier discrepancy method.
• Employed interactive adversarial training to extract domain-invariant features.
• Developed binary cross-entropy schemes and entropy modules to differentiate known and unknown classes.
• Established an integrated criterion for rejecting target unknown classes.

Main Results:

• Demonstrated superior performance on three machinery datasets.
• Effectively aligned target known-type samples with source known-type samples.
• Successfully suppressed the influence of unknown-type samples during feature alignment.

Conclusions:

• The proposed Known Classes Aware and Emerging Unknown Classes Rejection (KAEUR) approach effectively addresses open-set domain adaptation in fault diagnosis.
• KAEUR enhances diagnostic accuracy by distinguishing between known and novel fault states.
• The method shows significant potential for real-world industrial applications requiring robust fault detection.