基于EMD-PNN网络的刚性罐道故障诊断方法

Fault diagnosis method of rigid cage guide based on EMD-PNN network

针对刚性罐道故障种类识别困难,提出了一种基于经验模态分解-概率神经网络(EMD-PNN)的刚性罐道故障诊断方法。首先,搭建立井提升实验平台,使用北京东方振动和噪声技术研究所的INV3062T0设备采集罐道的振动信号,然后对含噪的振动信号进行EMD降噪;其次,提取降噪后振动信号的能量参数、偏度参数、峰度参数、波形参数、峰值参数、峭度参数、脉冲参数、裕度参数构成特征向量,作为PNN网络输入层的训练样本和测试样本;最后,利用训练样本建立PNN网络模型,选取测试样本检测概率神经网络的模式识别效果。实验证明,本方法对台阶凸起故障、接头错位故障和正常状态3种模式的识别率达到100%,为立井提升等非线性非平稳复杂系统的故障诊断提供一种通用可行的解决方案。

Aiming at the difficulty in identifying fault types of rigid cage guide,a fault diagnosis method for rigid cage guide based on EMD-PNN network is proposed.Firstly,the vertical shaft lifting experiment platform is set up,the INV3062 T0 equipment of China Orient Institute of Noise&Vibration is used to collect the vibration signals of the rigid cage guide,and then EMD noise reduction on the noisy vibration signals is performed.Secondly,the energy parameter,skewness parameter,kurtosis parameter,waveform parameter,peak value parameter,kurtosis parameter,pulse parameter and margin parameter of the vibration signal after noise reduction are extracted to form the feature vector,which will be used as the training sample and test sample of PNN network input layer.Finally,the training samples are used to establish the PNN network model,and the test samples are selected to detect the patterns recognition effect of the probabilistic neural network.The experiment proves that the recognition rate of the three patterns of step protrusion fault,joint misalignment fault and normal state can reach 100%,which provides a universal and feasible solution for fault diagnosis of nonlinear non-stationary complex systems such as vertical shaft lifting.