基于深度学习的滑油监测方法研究

针对传统的数据特征提取方法难以提取航空发动机滑油监测数据有效特征的缺陷,提出了一种基于多尺度卷积神经网络(Multi-scales convolutional neural network,MSCNN)、长短期记忆(Long short-term memory,LSTM)神经网络和BP网络的单通道网络模型MSCNN-LSTM-BP。将多尺度学习融入CNN,MSCNN和LSTM以串行方式提取数据在空间维度和时间维度的二维特征。实验结果表明:3尺度的MSCNN-LSTM-BP对数据样本的分类准确率达到98.2%,单组电容数据采集测试时间仅为2.1986ms,综合分类率F1达到98.57%,总体性能优于CNN,LSTM和传统的多尺度特征提取方法。MSCNN-LSTM-BP满足航空发动机滑油监测对于实时性和准确性的要求,具有良好的适用性。

Research on Monitoring Method of Lubricating Oil Based on Deep Learning

Aiming at the defect that the traditional data feature extraction method is difficult to extract the effective features of monitoring data of ECT aviation engine oil, a single-channel network model MSCNN-LSTM-BP based on multi-scales convolutional neural network (MSCNN), long short-term memory (LSTM) neural network and BP network is proposed. Integrating multi-scale learning into CNN, MSCNN and LSTM extract the two-dimensional features of data in spatial and temporal dimensions in a serial mode. The experimental results show that the classification accuracy of the 3 scales MSCNN-LSTM-BP for data samples reached 98.2%, single set of capacitor data acquisition test time is only 2.1986ms, the score of F1 reached 98.57%. Overall performance is superior to CNN, LSTM and traditional multi-scale feature extraction methods. MSCNN-LSTM-BP meets the requirements for real-time and accuracy of aero-engine oil monitoring and is provided with good applicability.