DANNTe: a case study of a turbo-machinery sensor virtualization under domain shift
This work addresses sensor virtualization for gas turbines, an incremental extension of existing domain adaptation methods to time series applications.
The authors tackled the problem of virtualizing a gas turbine sensor under domain shift by proposing an adversarial learning method for domain adaptation in time series regression, achieving a significant improvement in regression performance compared to a baseline trained only on the source domain.
We propose an adversarial learning method to tackle a Domain Adaptation (DA) time series regression task (DANNTe). The regression aims at building a virtual copy of a sensor installed on a gas turbine, to be used in place of the physical sensor which can be missing in certain situations. Our DA approach is to search for a domain-invariant representation of the features. The learner has access to both a labelled source dataset and an unlabeled target dataset (unsupervised DA) and is trained on both, exploiting the minmax game between a task regressor and a domain classifier Neural Networks. Both models share the same feature representation, learnt by a feature extractor. This work is based on the results published by Ganin et al. arXiv:1505.07818; indeed, we present an extension suitable to time series applications. We report a significant improvement in regression performance, compared to the baseline model trained on the source domain only.