基于堆栈自编码器和DeepAR的航空发动机剩余寿命预测

Prediction of Remaining Useful Life of Aero-Engine Based on Stacked Autoencoder and DeepAR

关键词:  航空发动机  寿命预测  预测模型  深度学习  数据融合

Key words:  Aero-engine  Life prediction  Autoencoder  Prediction model  Deep learning  Data fusion

摘要:

针对现有航空发动机剩余寿命预测大多基于单点预测模式，不能准确给出预测结果置信区间的问题，提出了一种基于堆栈自编码器结合DeepAR模型的概率分布预测模型。首先，堆栈自编码器通过无监督式深度学习对发动机监测数据进行特征提取，构建反映性能退化的健康指标（HI），基于双向长短期记忆（BiLSTM）网络构建DeepAR预测模型，将提取后的HI序列输入到DeepAR模型中，预测模型对HI序列与使用时间的隐含关系进行全局学习，并输出发动机剩余寿命的概率分布参数。利用C-MPASS涡扇发动机退化数据集进行实验，验证所提方法的有效性。结果表明，本文所提预测方法同其他方法相比，对监测数据融合的效果更好，预测模型性能提高6.4%，实际剩余寿命基本在95%置信区间内。

Abstract:

Aiming at the problem that most of the existing aero-engine remaining useful life （RUL） predictions are based on single-point prediction models， and the confidence interval of the prediction results cannot be accurately given， a probability distribution prediction model based on stacked autoencoder （SAE） and DeepAR model is proposed. First， the SAE extracts the features of engine monitoring data through unsupervised deep learning， and constructs health index （HI） reflecting performance degradation， the DeepAR prediction model is constructed based on bi-directional long-short term memory （BiLSTM） network， and the extracted HI sequence is input into the DeepAR model， the prediction model learns the hidden relationship between HI sequence and service time globally， and outputs the probability distribution parameters of engine RUL. The effectiveness of the proposed method is verified by experiments on C-MPASS turbofan engine degradation dataset. The results show that compared with other methods， the prediction method proposed in this paper has better monitoring data fusion effect， the performance of the model is improved by 6.4%， and the actual RUL is basically within the 95% confidence interval.