基于D-S证据理论和集成神经网络的磨粒识别

Wear Particles Classification Based on Dempster-shafter Evidential Reasoning and Integrated Neural Network

针对磨粒的识别问题,利用数字磨粒图像分析方法,结合D-S证据理论和BP神经网络,建立了基于D-S证据理论的集成神经网络磨粒融合诊断方法。首先对磨粒图像进行处理,并利用统计分析方法和傅立叶分析方法对处理好的磨粒图片进行分析得到磨粒特征;然后基于统计分析方法和傅立叶分析方法建立对应的两个BP分类子神经网络,利用典型的磨粒样本对BP子神经网络进行训练,得到初步的诊断结果;最后用D-S法对子神经网络诊断结果进行融合,得到最终的诊断结果。算例分析结果表明,基于D-S证据法和集成神经网络的磨粒融合诊断方法比单个诊断方法具有更高的准确性。

A wear particles classification method based on dempster-shafter evidential reasoning and integrated neural network was put forward. Firstly, digital wear debris images were converted to the images needed, and the wear particles characters were obtained. Then two sub-neural networks based on statistical analysis and Fourier analysis was established, and many typical wear particles features as training samples were provided. After each sub-neural network was trained successfully,the preliminary diagnosis of each sub-neural network was achieved. By using of the dempster-shafter evidential reasoning,the finial fusion diagnosis results were obtained. A practical example shows that the fusion identification method based on dempster-shafter evidential reasoning and integrated neural network is more accurate than the single identification method.