基于改进的CNN-SVM的机载燃油泵故障诊断  

Fault Diagnosis of Airborne Fuel Pumps Based on Improved CNN-SVM

作　　者：章余 景博[1] 焦晓璇 何宇廷[1] 孙宏达 鲍杰 ZHANG Yu;JING Bo;JIAO Xiaoxuan;HE Yuting;SUN Hongda;BAO Jie(School of Aeronautical Engineering,Air Force Engineering University,Xi'an 710038,China)

机构地区：[1]空军工程大学航空工程学院,西安710038

出　　处：《噪声与振动控制》2022年第6期137-142,共6页Noise and Vibration Control

基　　金：武器装备预先研究资助项目(50902060401);装备发展部基础研究资助项目(514010504-304);陕西省自然科学基金资助项目(2022JQ-586)。

摘　　要：针对机载燃油泵故障诊断过程中存在需要先验知识、专家经验、特征解释及信号特征提取困难等问题,提出基于模拟退火遗传优化CNN-SVM的诊断方法。该方法利用CNN具有自动提取故障特征的特性实现特征快速提取。并针对传统CNN诊断方法所存在网络结构不确定、计算效率低等问题,采用模拟退火遗传算法对其模型结构和参数进行优化,使用SVM取代全连接层的Softmax分类器优化CNN分类效果,最后应用t分布随机近邻嵌入使其故障特征学习过程可视化,评估其特征提取能力。试验结果表明:与SA-GA-CNN、1DCNN、ANN、GA-SVM及GA-BP诊断方法相比,基于改进CNN-SVM的方法能更有效实现机载燃油泵的故障诊断。Aiming at the problems of requiring prior knowledge,expert experience,feature interpretation and signal feature extraction in the process of fault diagnosis of airborne fuel pumps,a diagnosis method based on simulated annealing genetic algorithm is proposed to optimize CNN-SVM.The method utilizes the characteristic of CNN having automatic extraction of fault features to realize fast feature extraction.Aiming at the problems of uncertain network structure and low computational efficiency in the traditional CNN diagnosis method,the simulated genetic annealing algorithm is used to optimize the model structure and parameters,and SVM is used to replace the Softmax classifier in the full connection layer to optimize the CNN classification effect.Finally,t-distributed Stochastic Neighbor Embedding is used to visualize the feature learning process of fault diagnosis to evaluate the ability of feature learning.The test results show that compared with SA-GA-CNN,1DCNN,ANN,GA-SVM and GA-BP,the improved CNN-SVM can realize the fault diagnosis of airborne fuel pumps more effectively.

关 键 词：故障诊断 CNN-SVM 模拟遗传退火算法 机载燃油泵 故障特征可视化 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置]