Online Remaining Useful Life Prediction Method Based on Unsupervised Deep Domain-adversarial Adaptation
To solve the problems of high computational cost and error accumulation in the field of online remaining useful life (RUL) prediction of rotating machinery with unknown working conditions,a new online RUL prediction method was proposed based on unsupervised deep domain-adversarial adaptation.Firstly,by employing offline degradation data and online early fault data,a deep domain-adversarial net-work was constructed as a pre-trained model.Secondly,the pseudo-labels of online sequential data blocks were obtained by feeding them into the regression predictor of the pre-trained model that was re-domain adaptation.Finally,the structure and parameters of the pre-trained model were transferred to a target model,and by freezing some parameters of the target model,the remaining parameters were fine-tuned with the online data block and its pseudo-labels to achieve the online RUL prediction.Experimen-tal results on the IEEE PHM Challenge 2012 bearing dataset demonstrated that the proposed method could sequentially and accurately predict the RUL values of online bearing,which provided a practical solution for bearing RUL prediction in online scenarios.
remaining useful life predictiontransfer learningdomain adaptationunsupervised learningonline learning
万方数据
维普
