车轮踏面擦伤的集成粗糙神经网络预示诊断

Prognostics for surface scrapes on wheels based on integrated rough neural network

车轮踏面擦伤的预示诊断对列车安全运行具有重要意义。在分析粗糙集和神经网络特点的基础上,结合预示诊断中多传感器、多特征的要求,提出了一种粗糙集与多个神经网络相结合的车轮踏面擦伤预示诊断方法。该方法采用时频域都具有高分辨率的小波分析从车轮振动信号中提取擦伤特征,利用粗糙集的数据约简确定神经网络的初始拓扑结构,通过网络训练建立故障特征与故障之间的映射关系,从而实现踏面擦伤的多传感器融合诊断。实验结果表明该方法具有良好的预示诊断性能。

The prognostics of surface scrape on wheels is important for train safety. A prognostics system using an integrated rough neural network with multiple sensors was developed to detect surface scrape on wheels. Neural networks are best for solving non-linear problems while the rough set is good for data reduction. The integrated rough neural network provides reliable prognostics. Wavelet analysis, which can be used to analyze noisy signals in both the time domain and the frequency domain, was used to analyze the multiple features from the multiple signals. The rough set data reduction algorithm based on a discernibility function was used to select the key features. The network was trained to reflect the non-linear mapping between inputs and outputs. Experimental results with train wheel surface scrapes show that the approach efficiently differentiates the degrees of surface scrape.