Remaining Useful Life Prediction of Aircraft Engines based on FA-SDAE-BiGRU

Aiming at the problems of insufficient feature extraction and inaccurate prediction in existing aircraft engines remaining useful life prediction algorithms, a fusion model based on feature attention mechanism, stacked noise reduction autoencoder and bidirectional gated cycle units (BiGRU) was proposed. Firstly, weight is assigned to each sensor data of aircraft engine by the feature attention mechanism, and new weight data is obtained. Next, the stack noise reduction autoencoder is used to extract the weighted data. Then, bidirectional gated recurrent units are employed to predict the time series data and obtain the remaining useful life of the aircraft engine. To accurately select the hyperparameters of BiGRU, the Bayesian optimization algorithm is utilized for hyperparameter optimization. Finally, ablation experiments were conducted on the CMPASS dataset to validate the effectiveness of the proposed method, resulting in a root mean square error and score of 13.51 and 252.61, respectively.