红外光谱法结合PLS-BP网络对在用油中燃油稀释的测定

Determination of fuel dilution content in used oil by the method of infrared spectrum combined with PLS-BP network

采用红外光谱法建立PLS-BP模型检测在用润滑油燃油稀释的含量。在新润滑油中加入柴油配制柴油含量在0~8%的标准样品,采集样品光谱。经光谱特征分析,选择1500~680 cm^(-1)为建模区间,对光谱进行一阶微分和SG平滑处理,净化谱图信息。基于PLS回归系数选择10个特征波数建立BP神经网络模型,经反复试验寻优,隐含层神经元个数为10,最小均方误差为2.8×10^(-8),学习速率为0.04,模型Rp达到0.9642,SEP为0.5716%。对在用油样品进行加标回收试验,回收率在92.6%~107.2%。结果表明:PLS-BP模型可以用于在用油柴油稀释的检测。

Infrared spectrum method was adopted to establish PLS-BP（ Partial Least Squares-Back Propagation） model for testing the content of fuel dilution. Diesel was added in new lubricating oil to make the content of standard samples content in 0 - 8% and then the sample spectra were collected. By analyzing the spectrum characteristics,the range 1500 - 680 cm-1 was selected for modeling and 1 st order differential combined with SG（ Savitzky-Golay） smoothing for purifying spectrum information. 10 characteristics were chosen based on PLS regression coefficient for establishing BP neural network model.Through error optimization,the number of hidden layer neurons,minimum mean square error,the learning rate,the model ＇s Rp,and SEP were 10,2. 8 × 10-8,0. 04,0. 9642,and 0. 5716%respectively. The addition recovery test for used sample was carried out and recoveries were in 92. 6% to107. 2%. The results demonstrated that PLS-BP model was suitable for the determination of diesel oil dilution test.