摘要
通过对齿轮系统在不同的运转状态下不同的故障类型进行试验测试分析,获取了有关的测试信号,对振动特征信号进行了小波阈值去噪,采用离散小波变换(DWT)对去噪后的信号进行8层分解处理,对各层的小波系数进行了小波重构,得到8层细节信号和1层近似信号,并计算了各层信号的能量,得到了信号的能量分布特征.在此基础上把各层信号特征作为神经网络的输入,进行了网络的研究、分析处理和故障分类,并对小波神经网络方法与单独采用神经网络方法的故障诊断结果进行了比较评价.研究表明,去噪处理后的效果比没有去噪的信号特征更加明显,而采用小波神经网络诊断方法,对于齿轮无故障、齿根裂纹故障、分度圆裂纹故障和齿面磨损故障能够进行很好地区分与诊断,其诊断成功率均在95%以上,可对实际工程工作的齿轮系统进行故障诊断.
By measuring experimentally the vibration signals of the gear system at different rotating speeds for different faults, the testing signals were obtained. The feature signals of system were analyzed using wavelet de-noising by threshold in different running conditions. Using discrete wavelet transform （DWT）, the signals can be decomposed into eight detail components and an approximated component. After reconstruction of each level, their energy distributions were computed in order to extract the feature of the fault signals, these features were used for fault recognition using a neural network. The neural network was researched and analyzed. The results of using wavelet neural network and neural network were also compared. Based on the research, it indicates that the feature of the denoised signal is superior to the original one. When dealing with various situations, such as the crack at the gear root, the crack at the gear＇s reference circle and the wear abrasion fault of tooth surface, the performance rates are over 95%. The proposed method can be effectively used in engineering diagnosis of different faults of gear system.
出处
《航空动力学报》
EI
CAS
CSCD
北大核心
2010年第1期234-240,共7页
Journal of Aerospace Power
基金
国家自然科学基金(50575187)
航空科学基金(01I53073)
陕西省自然科学基金(2004E219)
西北工业大学研究生创业种子基金(Z200524)
关键词
离散小波变换
神经网络
特征提取
小波阈值去噪
故障诊断
齿轮
discrete wavelet transform
neural network
feature extraction
wavelet de-noising by threshold
fault diagnosis
gear
