基于1DCNN-BiLSTM的航空发动机故障分类研究

Research on aero-engine fault classification based on 1DCNN-BiLSTM

随着航空发动机运行状态的变化,其故障模式也会发生变化。针对航空发动机的运行退化趋势,提出一种基于1DCNN-BiLSTM的航空发动机故障分类模型。该模型可以直接用于原始监测数据,不需要其他算法提取故障退化特征,并且能充分利用1DCNN提取时间维度局部特征的优势,以及BiLSTM处理非线性时间序列及利用双向上下文信息的特点,最后连接全连接层来学习双向时序依赖的特征信息,并使用softmax函数来诊断故障类别。在美国航空航天局公开的CMAPSS数据集上进行验证,将故障模式分为无故障、HPC故障(单一故障)、HPC&Fan故障(混合故障)三种类型。实验结果表明,与其他模型对比,所提模型具有较高的分类精度,这对提高航空发动机运行可靠性和进一步进行剩余使用寿命预测有一定的实用价值。

As the operating status of aircraft engines changes,their failure modes will also change.A fault classification model for aircraft engines based on 1DCNN(1 dimensional convolutional neural network)BiLSTM is proposed to address the operational degradation trend of aircraft engines.This model can be directly used for raw monitoring data without the need for other algorithms to extract fault degradation features.In this model,the advantages of 1DCNN in extracting local features in the time dimension,and the ability of BiLSTM to handle nonlinear time series well and the characteristics of bidirectional contextual information are utilzied.A fully connected layer is connected to learn the feature information of bidirectional temporal dependencies,and the softmax function is used to diagnose fault categories.The verification is performed on the publicly available CMAPSS dataset by NASA,and the fault modes are classified into three categories:faultless,HPC fault(single fault),and HPC&Fan fault(mixed fault).The experimental results show that,in comparison with other models,the proposed model has higher classification accuracy,which is of practical value for improving the operational reliability of aero-engines and further remaining useful life(RUL)prediction.