基于粗集-神经网络的磨粒模式识别

Wear Debris Identification Based on Rough Set Artificial Neural Networks

应用粗集理论中最小约简的近似算法 ,对磨粒的形状参数进行约简 ,找出判断磨粒模式的形状参数长短轴比 Rt和圆度 Rd,同时采用这些参数训练神经网络以进行磨粒模式识别 .结果表明 :应用此算法 ,对 63个已知样本和历时 2年多对柴油机 1 4 5个润滑油油样进行制谱分析判断 ,准确率在 90 %以上 ,比原来用模糊识别的准确率提高了约 1 0个百分点 ;使用 BP网络减化了网络结构 ,使网络的训练速度加快 ,整个系统变得简单、可靠。

The min value reduction algorithm in the rough set was applied to reduce the number of shape values of wear debris,such as the aspect R t , roundness R d , liner fitness L n , concavity C v , disperse fitness D ρ , and integrated fitness C d . This is aimed at finding the most important shape values R t and R d for identifying the mode of wear debris with artificial neural network. In field test, the identification procession of the wear modes of 63 known samples and spectrum analysis of 145 oil samples in a diesel engine by the algorithm is above 90%, which is about 10% higher than that of fuzzy identification method. Because only two parameters R t and R d are used in BP networks, the structure is reduced and the speed of discipline accelerated considerably. Thus the identification system becomes simple, reliable and effective.