自组织神经网络定量分析内燃机油粘度指数

QUANTITATIVE ANALYSIS OF VISCOSITY INDEX OF ENGINE OIL BY SELF-ORGANIZING NERVE NETWORK

对自组织神经网络仅具有的定性分析功能进行了发展,将自组织神经网络从无监督聚类方法改为有监督聚类方法,建立了近红外光谱-内燃机油粘度指数定性分析模型;从自组织神经网络连接神经元权重中提取定量信息,建立了近红外光谱-内燃机油粘度指数定量分析模型。该方法不仅实现了用近红外光谱技术对内燃机油粘度指数同时进行定性和定量分析,而且能够优化自组织神经网络的训练。使用了不同生产厂家、不同牌号的内燃机油,用20个样品作为训练集训练该模型,用10个样品作为测试集检验该模型。研究结果表明,内燃机油的近红外光谱中含有与粘度指数相关的信息,用该模型能够实现对内燃机油粘度指数的定性和定量分析。

The self-organizing map was used to set up the qualitative analysis model of N1R spectra-viscosity index. Quantitative information was drawn from the weight of the neuron in the self-organizing map, which was used to set up the quantitative analysis model of NIR spectra viscosity index. The method can not only realize the qualitative analysis and the quantitative analysis for the viscosity index of engine oil simultaneously, but also can optimize the training for the self-organizing nerve network. In this study, the engine oils from different manufacturer with different brand were used to test the method. 20 samples and 10 samples were set up the training set and testing set respectively. Research result showed that the information of the near-infrared spectroscopy of engine oil had relation with the viscosity index, and the calibration model could realize the qualitative analysis and the quantitative analysis for the viscosity index of engine oil simultaneously.