基于神经网络的机车车轮故障诊断

The faults identification of locomotive wheels with ultrasonic testing based on neural network

针对现有机车车轮超声检测系统无法准确区分其故障类型,探讨了一种基于小波包变换与BP神经网络相结合的方法来识别基于超声检测机车车轮的故障类型。该方法对机车车轮的超声检测的回波信号进行小波包分解,并提取了频谱能量特征值,由此构造不同类型机车车轮内部伤损对应的特征向量,然后利用改进的BP神经网络算法训练随机抽取的样本空间,并对测试样本数据进行了识别验证。通过实验验证了该方法能够快速准确地识别出机车车轮不同类型的故障对应的超声检测信号。

To resolve the problem that ultrasonic testing system cannot identify the faults in locomotive wheels, this paper proposed a method based on wavelet packet analysis and back propagation neural network to identify the faults in locomotive wheels. Firstly, groups of ultrasonic echo signals were filtered and compressed by ultrasonic testing system, and then signals were decomposed with wavelet packet analysis, the frequency spectrum energy characteristics of the signals were extracted from the reconstructed wavelet packet nodes coefficients, and characteristics were normalized to compose the eigenvector. A modified BP neural network （BPN） trained by samples was used to identify faults in locomotive wheels. Then the eigenvectors of testing samples were sent to the trained BPN. The result of testing shows that the trained BP neural network can identify the ultrasonic faults in locomotive wheels accurately.