Prof. Fink talks about Domain Adaptation at Nanjing University

The virtual talk provided insights into potential solutions that enable to transfer models and operational experience between different units and between different operating conditions also in unsupervised setups where data on faulty conditions is not available. Moreover, a synthetic-to-real framework for domain adaptation was presented, where only knowledge of the healthy class is needed. The healthy class is augmented with synthetic faults to generate the rare fault examples based on expert understanding of how faults affect the measured signals. A domain adaptation approach is proposed to adapt the model from synthetic faults (source) to the unlabeled real data (target) overcoming the different imbalance levels in source and target datasets.