Cross-condition remaining useful life prediction based on cumulative features and composite adversarial domain adaptation

Abstract

Listen

Translate article

Accurate prediction of remaining useful life (RUL) in mechanical equipment is an essential part of ensuring the smooth operation of the equipment. However, the extraction of degradation feature and transfer application of prediction algorithms across different operating conditions is a critical issue that needs to be addressed. This article proposes a cross-condition RUL prediction algorithm based on cumulative features and a composite adversarial domain adaptation (CF-CADA) algorithm. Firstly, a degradation feature extraction method based on cumulative transformation (CT) and amplitude correction is proposed to extract cumulative features (CF) of the devices. Secondly, a composite adversarial domain adaptation model is constructed to achieve the RUL prediction across different working conditions. The model consists of two main parts: marginal distribution alignment based on the Wasserstein generative adversarial network (WGAN) and conditional distribution alignment based on the improved triplet loss and momentum network. Finally, an asynchronous training strategy is used to train the model. Comparative experimental results based on two datasets prove that the proposed algorithm can provide an effective reference for cross-condition RUL prediction.