基于可见-近红外光谱技术的润滑油酸值无损检测方法研究

Non-invasive Measurement of Acid Value of Lubricant Using Visible and Near Infrared Spectroscopy

研究了基于可见-近红外光谱技术的润滑油酸值无损检测方法。获得了润滑油在475～975 nm范围内的可见-近红外光谱。采用最小二乘支持向量机(LS-SVM)建立可见-近红外光谱检测模型,并通过将无信息变量消除算法(UVE)与连续投影算法(SPA)相结合进行光谱有效波长选取。通过UVE-SPA法进行变量选择计算,最终将原始光谱所包含的500个光谱变量减少到了8个(分别为489 nm、553 nm、591 nm、874nm、893 nm、910 nm、935 nm和951 nm)。基于这8个变量建立的LS-SVM模型获得了预测集确定系数为0.9546、误差均方根为0.0081和剩余预测残差为4.5663的预测结果,说明可见-近红外光谱技术可以用于润滑油酸值无损检测。与酸值测定标准方法相比,该方法具有快速、无损和成本低等优点。同时,UVE-SPA法是一种有效的光谱变量选择方法。

The non-invasive lubricant acid value measurement method based on a visible and nearinfrared spectroscopy is studied. The visible and near-infrared spectra of the lubricant in the region from 475 μm to 975 μm are obtained. A visible and near-infrared spectral detection model is established by using a least square support vector machine. By combining the uninformative variable elimination with the successive projection algorithm, effective spectral wavelengths are selected. Among the 500 variables, only eight wavelength variables namely 498 nm, 553 nm, 591 nm, 874 nm, 893 nm, 910 nm, 935 nm and 951 nm are selected. The least square support vector machine model based on those eight wavelength variables has obtained a prediction set of 0.9546, a root mean square error of 0.0081 and a residual predictive deviation of 4.5663. This prediction result shows that the visible and near-infrared spectroscopy can be used to measure the acid value of lubricant non-invasively. Compared with the standard acid value measurement method, it has the advantages of fastness, non-invasion and low cost. Meanwhile, it is also an effective algorithm for spectral variable selection.