Remaining useful life of the rolling bearings migration prediction model based on EADA

A new prediction model for the migration of the remaining useful life (RUL) of rolling bearings based on enhanced adversarial domain adaptation (EADA) is proposed, aiming at handling the low prediction accuracy and poor model generalization facing RUL estimation of rolling bearings under variable operating conditions. First, in the feature extraction stage, the improved convolutional neural network (CNN), as a feature extractor, serves for the autonomous mining of the bearing vibration signal’s deep features. Then, in the adversarial training stage, a domain discriminator is used to distinguish the source domain and target domain bearing features to achieve edge distribution alignment, and a state discriminator is constructed to distinguish the target domain bearing state and retain target domain-specific information. Finally, the migration prediction scenarios under six variable operating conditions are evaluated by the accelerated degradation dataset of rolling bearings, and according to the evaluation results, the model excels in realizing the RUL prediction of rolling bearings under these conditions, as well as exhibits a superior prediction performance compared with other models.