DOI：10.3969/j.issn.1672-3872.2025.18.035
作 者：姜锦涛 ，李文强 ，吕淮淼（沈阳航空航天大学机电工程学院，辽宁 沈阳 110136）
摘 要：【目的】解决传统航空发电机故障诊断方法缺乏时空建模能力以及准确率低的问题。【方法】提出了一种结合卷积神经网络（CNN）、双向长短时记忆网络（BiLSTM）和注意力机制（AM）的深度神经网络故障诊断组合模型。在MATLAB环境下通过故障仿真获取短路故障输出电流数据，利用CNN-BiLSTM-AM模型进行故障识别分类，并将该模型与其他4种模型进行效率对比。【结果】CNN-BiLSTM-AM模型对发电机故障的识别效果良好，诊断精度较高，平均诊断准确率达到98.71%，与其他方法相比具有最高的平均训练准确率和平均诊断准确率。【结论】CNN-BiLSTM-AM模型可以准确地识别不同的短路故障，极大提高了故障的平均诊断准确率，在故障诊断中的有效性和优越性较高。
关键词：航空发电机；短路故障；深度学习；注意力机制
Fault Diagnosis Research of Aviation Generator Based on CNN-BiLSTM-AM
Author: Jiang Jintao, Li Wenqiang, Lyu Huaimiao (School of Mechanical and Electrical Engineering, Shenyang Aerospace University, Shenyang 110136, China)
Abstract: [Objective] To solve the problems of lack of spatiotemporal modeling ability and low accuracy in traditional aviation generator fault diagnosis methods. [Method] A deep neural network fault diagnosis combination model combining convolutional neural network (CNN), bidirectional long short-term memory network (BiLSTM), and attention mechanism (AM) is proposed. Obtain short-circuit fault output current data through fault simulation in MATLAB environment, use CNN-BiLSTM-AM model for fault identification and classification, and compare the efficiency of this model with the other four models. [Result] The CNN-BiLSTM-AM model has a good recognition effect on generator faults, with high diagnostic accuracy and an average diagnostic accuracy of 98.71%. Compared with other methods, it has the highest average training accuracy and average diagnostic accuracy. [Conclusion] The CNN-BiLSTM-AM model can accurately identify different short-circuit faults, greatly improving the average accuracy of fault diagnosis and demonstrating high effectiveness and superiority in fault diagnosis.
Keywords: aviation generator; short-circuit fault; deep learning; attention mechanism
引文信息：[1]姜锦涛，李文强，吕淮淼.基于CNN-BiLSTM-AM的航空发电机故障诊断研究[J].南方农机，2025，56(18)：136-139+152.
查看全文请下载PDF文件↓
