基于深度信念网络的航空发动机维修等级决策

准确的航空发动机维修等级决策,能够避免过维修和欠维修,在保证航空发动机运行安全的前提下节约维修成本。结合航空发动机状态监控信息和维修等级特点,采用深度信念网络（DBN）算法,挖掘状态监测及维修等级决策之间的深层次对应关系,实现对维修等级的分类和预测。该模型通过DBN预训练和反向传播(BP)神经网络反向微调提取出样本特征,从而提高维修等级预测准确率。以某航空公司CF6航空型发动机的状态参数和维修等级数据作为实例进行验证,结果显示:该模型能够通过构建多层网络结构挖掘出样本的更深层次信息，在分类能力、决策准确性方面优于传统神经网络，有较强的特征提取能力，对维修等级分类有较高的正确率，能得出更准确的维修等级决策结果，避免因维修等级误判而带来不必要的损失。 

Maintenance level decision for aero-engine based on deep belief network

The accurate maintenance level decision can avoid the excessive maintenance and shortage of maintenance and save maintenance cost on the premise of ensuring the safe operation of aero-engine. To achieve the classification and prediction for maintenance level, monitor information and characteristics of maintenance level, using algorithm of deep belief network（DBN）, were combined, excavating deep relationship between the condition monitoring and maintenance level decision making. The model can extract sample feature from DBN pretreatment and back propagation (BP) neural network reverse fine-tuning and improve the forecast accuracy of maintenance level. Taking the state parameters and maintenance level data of an airline CF6 engine as example, the analysis results showed that the model could excavate the deeper information of the sample through the construction of multi-layer network structure, which was superior to the traditional neural network in the classification ability and the accuracy of decision-making. So the model had strong ability of feature extraction and higher classification accuracy for the maintenance level. The model was able to get more accurate results maintenance level decision and avoided unnecessary losses due to misclassification of maintenance level.