TVAR在非平稳工况转子故障诊断中的应用

研究了基于BP网络（ANN）的时变参数自回归模型（TVAR）及其在非平稳工况旋转机械故障诊断中的应用。首先提出了一种基于BP算法的TVAR参数辨识方法，然后利用TVAR方法对一非线性调频仿真信号进行时频分析，并与典型时频分析方法短时傅里叶变换（STFT）及Choi—Williams分布（CWD）的分析结果进行比较。结果表明，TVAR方法具有时频分辨率高、无交叉干扰项及计算速度快等优点。最后利用TVAR方法分析了转子启动过程正常及故障工况下转子实验台的非平稳振动信号。研究表明，TVAR不但能够有效地分析非平稳振动信号，而且具有较强的故障特征提取和抗噪声能力，是在时频域上进行故障诊断的有效方法。

ANN-Based TVAR and Its Application to Fault Diagnosis of Rotors Under Nonstationary Conditions

A BP-ANN(back propagation-artificial neural network)based time-varying autoregressive model and its application to the fault diagnosis of rotation machine under nonstationary conditions were studied.Firstly,a BP-based method for the estimation of TVAR coefficients was presented and validated.Secondly,by the analysis of a nonlinear frequency-modulated synthetic signal,the time-frequency analysis performance of TVAR was compared with some traditional methods as Short Time Fourier Transform(STFT)and Choi-Williams Distribution(CWD).The results show some merits of TVAR such as high re-solutions,no cross terms and rapid computation.Finally,some nonstationary vibration signals collected from a rotation machine test rig under normal and fault conditions during the speed-up period were analyzed.It is shown TVAR excels at the disposal of nonstationary signals and has a superior feature extracting ability,and is less sensitive to noises.In conclusion,TVAR is an excellent method for time-frequency analysis and fault diagnosis of rotation machine under nonstationary conditions.