基于贝叶斯网络的润滑油发射光谱数据挖掘研究

Research on Data Mining of Lubricating Oil Atomic Emission Spectroscopy Based on Bayesian Network

针对船舶柴油机活塞-气缸套磨损工况的监测问题,对某型柴油机69个工作油样的发射光谱数据进行了分析。油样来自7种不同活塞-气缸套间隙工况,光谱数据包括21种典型元素浓度值,构建了NB贝叶斯网络分类模型和K2贝叶斯网络分类模型对油样光谱数据进行了挖掘研究。研究结果表明,两种分类模型的工况识别正确率均大于90%,K2贝叶斯网络模型的分类规则与实际装备的工作机理更加吻合。

Aiming at monitoring the wear condition of the clearance between piston and cylinder for marine diesel engine, the atomic emission spectroscopy data of 69 working oil samples from a diesel engine were analyzed. The oil samples were from 7 different clearances between piston and cylinder and the spectral data included 21 typical element concentrations. Naive Bayes（NB） network and K2 Bayesian network classification models were constructed to study the spectral data of oil samples. The results show that the recognition accuracy of the two classification models are more than 90%, and the classification rules of the K2 Bayesian network model is nearly consistent with the working mechanism of the actual equipment.