摘要
机械故障与润滑油的性状具有紧密关系。因此,研究一种能够快速、无损对润滑油品牌识别方法至关重要。该研究应用近红外光谱分析法结合偏最小二乘判别分析(Partial Least Squares-Discriminant Analysis,PLS-DA)模式识别方法对7种润滑油品牌进行识别。研究结果表明,采用近红外光谱结合PLS-DA方法对校正样本建立判别模型,模型的校正相关系数均大于0.980,校正集均方根误差(RMSEC)和预测集均方根误差(RMSEP)都小于0.062,对7种润滑油品牌识别率均为100%。结合遗传算法对变量进行筛选,选出62个波数点代替全波段进行建模,模型对未知样本的识别率均为98.1%,大大缩减建模的计算量,为在润滑油判别分析仪器开发方面提供一定的理论指导。
There were lots of relations between mechanical failure and the character of lubricant oil.Therefore it's essential to find a way that could fast nondestructive identify lubricant oil.In this study,near infrared spectroscopy(NIRS) coupled with Partial Least Squares-Discriminant Analysis(PLS-DA) pattern recognition methods were used to identify the seven kinds lubricant oil.The results showed that seven kinds lubricant oil were identified by NIR spectra coupled with PLS-DA models,the identification rates were all 100%,the correlations between the predicted category variables of calibration or validation and the measured category variables were all remarkable with a correlation coefficient(r) over 0.980 and low RMSEC and RMSEP(<0.062).Screening for variables with genetic algorithm,selecting the 62 variable points instead of the whole wave-band to set a model,the identification rates of the model for the unknown samples were 100%.This work could reduce the computational modeling;provide some theoretical guidance for the development of instrument.
出处
《中国农机化学报》
北大核心
2013年第1期181-185,共5页
Journal of Chinese Agricultural Mechanization
基金
华东交通大学科研基金项目(09JD100)
