摘要
基于泵机组故障信号处理的需要,介绍了不变矩原理,同时对神经网络建模,包括其样本获取进行了详细讨论;由于泵机组的多种故障与表征其运行状态的轴心轨迹形状有关,根据不变矩的平移、伸缩和旋转不变性特征,对实时检测的轴心摆度信号进行不变矩处理,利用BP型神经网络对其进行模式识别,进而判断出轴心轨迹的形状.为了弥补泵机组用于神经网络训练样本的不足,采用数值模拟与现场测试相结合的方法,将获取的所有样本进行求不变矩处理,并连同样本对应的实际形状作为神经网络的训练样本.网络训练完成后,将其输出结果与轴心轨迹图形进行比较验证.以山西大禹渡泵站水泵机组故障检测及诊断为例,在样本中选取其中的10组数据,比较的结果表明神经网络自动识别的结果准确.该方法为泵机组轴心轨迹自动识别和实现泵机组故障诊断智能化提供了依据.
To meet the needs of signal processing on pump unit fault diagnosis, the principle of invariant moment theory was introduced. In addition, the neural network modeling as well as the sample acquisition in detail was discussed. As the shape of axis orbit responded the pump unit operation is related to a variety of fault, the real-time detection swing signals of axis on invariant moment were processed according to the invariant features of translation, scaling and rotation of invariant moment. And then the shape of axis orbit was determined by using BP neural network on pattern recognize. The combination of numerical simulation and on-site test were used to compensate the shortage of neural network training samples. All samples of both processed on invariant moment and the corresponding actual shape of the samples are of the neural network training ones. After network training completed, the output was compared with the actual shape of axis loci to validate this method. Taken the fauh detection and diagnosis of Dayudu Pump Station in Shanxi for example, 10 sets of data of the sample were selectd to be compared, and the resuhs show that the neural network recognition of the resuhs are accurate. The method can provide the basis tor orbit shape automatic identification and realizing fauh diagnosis system intellectualization of pump unit.
出处
《排灌机械工程学报》
EI
2011年第1期67-71,共5页
Journal of Drainage and Irrigation Machinery Engineering
基金
'十一五'国家科技支撑计划重点项目(2006BAD11B07)
关键词
轴心轨迹
自动识别
不变矩
BP神经网络
泵机组
axis orbit ： automatic identification
invariant moments
BP neural network
pump unit
