基于AGRU的航空发动机剩余使用寿命预测研究

The Remaining Useful Life Prediction of Aero-engine Based on AGRU

摘要:

故障预测与健康管理中的核心环节之一是剩余使用寿命预测,传统循环神经网络及其变体在解决剩余使用寿命预测的回归问题时性能较低。提出一种基于注意力机制的门控循环单元(GRU)用于剩余使用寿命预测。对数据集进行预处理,并采用分段线性退化方法对剩余使用寿命标签进行建模。提出基于编码器-解码器框架且融合注意力机制的门控循环单元网络,利用GRU网络从预处理数据中学习时间序列特征,基于注意力机制的编码器-解码器模型学习特征与时间步长的重要性,优化特征利用效率,实现剩余使用寿命的高效预测。在NASA的C-MAPSS基准数据集上验证了所提模型的有效性,结果表明,该方法性能优于GUR及其其他变体方法,为剩余使用寿命预测提供了一种新的解决方案。

Abstract:

One of the core links in prognostics and health management ( PHM ) is the prediction of remaining useful life ( RUL ). The traditional recurrent neural network ( RNN ) and its variants have low performance in solving the regression problem of remaining useful life ( RUL ) prediction. An attention-based gated recurrent unit ( attention GRU, AGRU ) is proposed for remaining useful life prediction. The data set is preprocessed, and the piecewise linear degradation method is used to model the remaining useful life label. A gated recurrent unit ( GRU ) network based on encoder-decoder framework and attention mechanism is proposed. The GRU network is used to learn time series features from preprocessed data. The encoder-decoder model based on attention mechanism learns the importance of features and time steps, optimizes feature utilization efficiency, and achieves efficient prediction of remaining useful life. The effectiveness of the proposed model is verified on the NASA C-MAPSS benchmark data set. The results show that the performance of this method is better than that of GUR and other variants, which provides a new solution for remaining useful life prediction.