Remaining useful life(RUL)is a significant challenge in prognostics and health management.Existing methods suffer from a severe performance drop,as testing data from engine sensors exhibits high nonlinearity and compl...Remaining useful life(RUL)is a significant challenge in prognostics and health management.Existing methods suffer from a severe performance drop,as testing data from engine sensors exhibits high nonlinearity and complicated fault modes.In this paper,the authors introduce a reinforcement neural architecture search technique based on upper confidence bound(UCB)to optimize an efficient model.UCB explores the combinatorial parameter space of a multi-head convolutional layers concatenate with recurrent layers to search for a suitable architecture.To address the highly nonlinear dataset in complicated working conditions,rainflow counting algorithm is applied to extract features.Experiments are conducted on C-MAPSS dataset.Compared with state-of-the-art,the proposed approach yields better results in both RMSE and scoring function for all the sub-datasets.In multiple working conditions,the authors achieve lower RMSE with significant superiority.The experimental results confirm that the proposed method is an efficient approach for obtaining highly precise RUL predictions.展开更多
基金supported by the National Natural Science Foundation of China under Grant Nos.62073197and 61933006。
文摘Remaining useful life(RUL)is a significant challenge in prognostics and health management.Existing methods suffer from a severe performance drop,as testing data from engine sensors exhibits high nonlinearity and complicated fault modes.In this paper,the authors introduce a reinforcement neural architecture search technique based on upper confidence bound(UCB)to optimize an efficient model.UCB explores the combinatorial parameter space of a multi-head convolutional layers concatenate with recurrent layers to search for a suitable architecture.To address the highly nonlinear dataset in complicated working conditions,rainflow counting algorithm is applied to extract features.Experiments are conducted on C-MAPSS dataset.Compared with state-of-the-art,the proposed approach yields better results in both RMSE and scoring function for all the sub-datasets.In multiple working conditions,the authors achieve lower RMSE with significant superiority.The experimental results confirm that the proposed method is an efficient approach for obtaining highly precise RUL predictions.
