基于URP-ANCNN的变转速齿轮箱智能故障诊断方法
Intelligence Fault Diagnosis Method for Gearboxes under Variable Rotational Speed Based on URP-ANCNN
陈向民 1舒文伊 1韩梦茹 1张亢 1李博1
作者信息
- 1. 长沙理工大学 能源与动力工程学院,长沙 410114
- 折叠
摘要
由于齿轮箱振动信号在变转速工况下出现的调频、调幅等现象,使得信号征兆与故障模式之间的映射关系变得复杂,导致齿轮箱故障难以精确诊断.鉴于深度神经网络在自动提取数据特征和分类上的优势,提出一种基于无阈值递归图编码(Un-threshold Recurrence Plot,URP)和自适应归一化卷积神经网络(Adaptive Normalized Convolutional Neural Network,ANCNN)的变转速工况齿轮箱故障诊断方法.该方法先使用快速傅里叶变换(Fast Fourier Transform,FFT)将时域信号转化为频域信号,再利用URP编码将得到的频域信号转化为二维递归图,并提取图像特征输入到ANCNN模型.在ANCNN模型中,采用批量归一化算法消除因转速变化引起的特征分布差异,同时处理转速波动下产生的频移和调制特性,并使用遗传算法自动调整该网络模型的超参数,以提高该网络的整体性能.采用转速波动的齿轮箱试验数据对该方法进行验证,实验结果表明,该方法能够克服转速波动的影响,成功实现对不同齿轮故障的准确识别.
Abstract
Due to the frequency modulation and amplitude modulation of the vibration signal of the gearboxes under variable speed conditions,the mapping relationship between the signal symptoms and the failure mode becomes complicated,which leads to the difficulty to accurately diagnose the gearbox faults.In virtue of the advantages of deep neural network in data features extraction and classification,a gearbox fault diagnosis method based on Un-threshold recurrence plot(URP)and adaptive normalized convolutional neural network(ANCNN)is proposed.In this method,a time-domain signal is converted to a frequency-domain signal by using Fast Fourier transform(FFT)and then the frequency-domain signal is transferred into a 2D recursive diagram with the URP encoding.Then,the image features are extracted and input into the ANCNN model.In the ANCNN model,the batch normalization(BN)algorithm is used to eliminate the characteristic distribution differences caused by rotation speed changing,and to deal with the frequency shift and modulation characteristics caused by rotation speed fluctuation.To improve the global performance of the network,the optimal hyper parameters of the network model are determined by genetic algorithm(GA).The gearbox test data sets with fluctuating rotational speed are used to verify the effectiveness of the proposed method,and the results show that the method can overcome the effect of speed fluctuation and achieve accurate identification of different gear fault types.
关键词
故障诊断/卷积神经网络/无阈值递归图/批量归一化/变转速工况/齿轮箱Key words
fault diagnosis/CNN/URP/batch normalization/variable speed working condition/gearbox引用本文复制引用
基金项目
湖南省自然科学基金(2018JJ3541)
湖南省教育厅项目(20B019)
湖南省教育厅项目(21B0347)
出版年
2024
国家科技期刊平台
NSTL
万方数据