基于神经网络信息融合的发动机磨损磨粒识别

An Information Fusion Technique for Wear Particle Recognition Based on Neural Network

磨粒识别和分类是铁谱分析技术在发动机故障诊断和状态监测的关键环节。针对单一神经网络模型磨粒识别的局限性,提出了一种基于不同类型神经网络信息融合的磨粒智能识别方法。首先利用径向基函数(RBF)神经网络和反向传播(BP)神经网络对磨粒进行识别,得到2组初始识别结果,归一化后作为2组基本概率分配函数,然后利用D-S证据理论对其融合得到最终识别结果。实例计算表明,与单一神经网络模型相比,提出的信息融合方法提高了磨粒识别的区分度和准确率,并具有良好的通用性和容错性。

Wear debris recognition is a focus in ferrography technology, which is now widely applied to engine fault dianosis and condition monitoring. In order to overcome the disadvantages of the single neural network method for wear particle recognition, an information fusion technique based on different neural network was proposed. Two single methods, radial based function （RBF） neural network,back-propagation （BP） neural network were selected to recognize wear particles, and two initial results were got which can be used as multi-source evidence for fllsion. The effective data fusion technique Dempster-Shafer （D-S） evidence theory was employed to fuse the two evidences, and the wear particle recognition the resuh was got. According to the application in the wear particle recognition, this new method gets a more precise result compared with each single neural network and it is more adaptive and tolerant.