基于双树复小波变换—最大熵谱估计和参数优化概率神经网络的齿轮箱故障诊断研究

齿轮振动信号具有非平稳性和非线性的特点。为了准确提取其故障特征并进行故障诊断，提出一种基于双树复小波变换(DTCWT)-最大熵谱估计(MESE)和惯性权重线性递减粒子群优化(LDWPSO)算法-参数优化概率神经网络(PNN)的齿轮故障诊断方法。首先，利用DTCWT把状态已知的齿轮振动信号分解为不同频带的模态分量。其次，采用MESE得到每个分量的最小偏差频谱估计，计算出不同频段的能量熵作为故障特征矩阵。然后利用LDWPSO算法寻找出最优神经网络参数——平滑因子。最后，将故障特征矩阵输入优化后的PNN模型，建立起故障特征和齿轮运行状况之间的数值化映射关系，进而完成齿轮故障诊断模型。经试验数据分析表明，采用提出的DTCWT处理齿轮的振动信号，并引入MESE处理关键分量，可以提取稳定的信号特征并降低噪声干扰。另外，相比于传统的PNN,基于改进的PNN的齿轮故障状态的数值化判别具有更高的诊断精度和稳定性。

Gearbox Fault Diagnosis Based on Dual Tree Complex Wavelet Transform-Maximum Entropy Spectrum Estimation and Parameter-Optimization Probabilistic Neural Network

Gear vibration signal has the characteristics of non-stationary and nonlinear.In order to accurately extract its fault features and carry out fault diagnosis,a gear fault diagnosis method based on dual tree complex wavelet transform (DTCWT)-maximum entropy spectrum estimation (MESE) and the linear decreasing weight particle swarm optimization (LDWPSO)-parameter-optimization probabilistic neural network (PNN) is proposed.Firstly,the gear vibration signal with known state is decomposed into modal components of different frequency bands by using DTCWT.Secondly,MESE is used to obtain the minimum deviation spectrum estimation of each component,and the energy entropy of different frequency bands is calculated as the fault characteristic matrix.Then LDWPSO is used to find the optimal neural network parameter-smoothing factor.Finally,the fault characteristic matrix is input into the optimized PNN model,the numerical mapping relationship between fault characteristics and gear operation is established,and then the gear fault diagnosis model is completed.The experimental data analysis shows that the DTCWT proposed can process the vibration signal of the gear and introduce MESE to process the key components,which can extract stable signal features and reduce noise interference.In addition,compared with the traditional PNN,the numerical discrimination of gear fault state based on the improved PNN has higher diagnostic accuracy and stability.