基于深度学习的短波发信机故障诊断应用  

Short-Wave Transmitter Fault Diagnosis Application based on Deep Learning

作　　者：刘程 陈斌[1] 瞿珊瑚 张海天 LIU Cheng;CHEN Bin;QU Shan-hu;ZHANG Hai-tian(College of Electronic Engineering,Naval University of Engineering,Wuhan Hubei 430033,China)

机构地区：[1]海军工程大学电子工程学院,湖北武汉430033

出　　处：《通信技术》2018年第12期3033-3037,共5页Communications Technology

摘　　要：传统的发信机故障诊断多基于故障树、专家系统或机器学习的方法,而这些方法都需要花费大量的人力和精力来提取数据特征,对工作人员的专业性要求很高,且容易出现诊断误差。针对这一问题,提出一种基于深度学习的短波发信机故障诊断方法。首先,采集发信机的原始故障信号,对故障信号进行归一化处理后,将数据样本分为训练集和测试集。训练过程中,用无标签数据逐层训练深度神经网络,用有标签数据精调网络参数,完成对故障信号数据的特征提取。最后,用训练好的网络对5种状态的信号进行故障诊断。仿真分析表明,该方法有效减小了诊断误差,诊断性能优于传统的诊断方法。Traditional transmitter fault diagnosis is mostly based on fault trees,expert systems or machine learning methods.These methods all require a lot of manpower and effort to extract data features.This is highly demanding for the professionalism of the staff and is prone to diagnostic errors.Aiming at this problem,a short-wavelength transmitter fault diagnosis method based on deep learning is proposed.Firstly,the original fault signal of the transmitter is collected.After the fault signal is normalized,the data samples are divided into a training set and a test set.During the training process,the deep neural network is trained layer by layer with non-label data,and the network parameters are finely adjusted with the labeled data to complete the feature extraction of the fault signal data.Finally,the trained network is used to diagnose the signals of the five states.The simulation analysis proves that the method effectively reduces the diagnostic error and its diagnostic performance is superior to the traditional diagnostic method.

关 键 词：短波发信机 深度学习 堆栈自编码 故障诊断 特征提取 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置]