基于卷积神经网络风力机轴承混沌空间故障分析与诊断

Fault Diagnosis and Analysis of Wind Turbine Bearing Chaotic Phase based on Convolutional Neural Network

针对风力机轴承振动信号的非线性特点,对轴承不同工作状态振动信号进行相空间重构,还原信号非线性动力学特性;通过计算获取嵌入维数和延迟时间以构建二维混沌相图,并以不同工作状态的混沌相图作为样本,输入二维卷积神经网络开展学习建模,构建相空间-卷积神经网络故障诊断模型。结果表明:轴承不同状态振动信号具有明显的混沌特性,二维混沌相图具有不同的非线性表征;二维相图随故障程度变化而变化,但保持原有的"相形";以相同尺寸二维相图,结合卷积神经网络构建故障诊断系统,不仅对相同故障程度的工作状态准确度高,且当同一故障其程度不同时,也非常精准,表明所提出故障诊断模型具有良好的泛化能力。

Chaotic theory is used to reconstruct the phase space of bearing vibration signal because that vibration signal of wind turbine bearing is nonlinear.Nonlinear dynamic feature of signal is restored by the method of phase space reconstruction,and the two dimension phase diagrams are built with the optimal parameter of embedding dimension and delay time.With the chaotic phase diagram as input samples,the intelligent fault diagnosis system is built by two-dimension convocation neural network.The results show that there are different chaotic characters when bearing working states are different.Two-dimension chaotic diagram has different nonlinear representations.The diagrams change with fault degree,but they keep the original"phase form";PS-CNN fault diagram system has high accuracy in bearing states classification not only in same degree of fault,but also in hybrid fault degree.It also shows that the proposed PS-CNN fault diagnosis system has great generalization.