6 / 2021-05-24 17:16:02

Remaining Useful Life Estimation Under Variable Failure Behaviors via Transferable Metric Learning

Accurate estimation of remaining useful life (RUL) can effectively reduce the maintenance cost and prevent mechanical accident. However, the failure modes of different bearings caused by variable failure behaviors may lead to a domain shift. Traditional methods to solve the domain shift problem try to derive domain invariant features, but fail to consider the domain relations of unknown samples, resulting in limited performance. To solve above challenges, a transferable cross-domain approach based on deep transferable metric learning for RUL estimation is proposed. The hidden features are extracted adaptively by temporal convolution network. To minimize the domain discrepancy, a new cross-domain adaptation architecture is designed to learn the domain invariant features in which the contrastive loss of metric learning is used to improve the complex transformation invariance of the multi-kernel maximum mean discrepancy to make contribution to the common scenarios in RUL estimation. The experimental results show that our method is improved by more than 17% and 21% in two different evaluation metrics, which is obviously superior to the other methods compared.