基于改进DRSN的航空发动机故障风险预警模型

Aeroengine fault risk early warning model based on improved DRSN

航空发动机属于多发性故障机械,运用先进的计算训练方法可有效地实现准确的风险预警分析,为发动机的运维指导提供参考。在发动机故障风险预警征兆数据集中提取多变量时间序列样本,将样本矩阵化,转换为灰度图样本。预处理并增强图像数据样本,热编码化序列样本标签。深度残差收缩网络(deep residual shrinkage network,DRSN)中融入深度注意力机制与带有阈值的残差收缩块,获取高判别性特征,实现软阈值化。结合长短时记忆神经网络层与多个隐层,改进DRSN模型,使用主成分分析重构特征与主元提取,累积可解释方差贡献率为93.7%。对潜在20种故障征兆识别、分类并预警,训练精确度为96.1%。提出了改进DRSN航空发动机故障风险预警模型,与其他算法相比有较强的鲁棒性,预警正确率至少提高4.4%。

Aero-engine is a kind of mechanical equipment with possible multi-fault risk.The application of advanced computing training method can effectively realize accurate risk early warning analysis,and provide reference for the guidance of engine operation and maintenance.Multivariable time series samples were extracted from the early warning symptom data set of engine failure risk,and the samples were matrix-transformed into gray scale samples.Image samples were preprocessed and enhanced,and sequence sample tags were thermally encoded.Deep attention mechanism and residual shrinkage block with threshold were integrated into the deep residual shrinkage network(DRSN),so as to obtain high discriminant features and realize soft thresholding.Combining long short term memory layers with multiple hidden layers,DRSN model was improved,and principal component analysis was made to reconstruct features and extract principal components.The cumulative interpretable variance contribution rate was 93.7%.The training accuracy for identifying,classifying,and warning 20 potential fault symptoms was 96.1%.An improved early warning DRSN model of engine fault risk was proposed.Compared with other algorithms,this model of strong robustness improved the accuracy by at least 4.4%.