Application of Adaptive Iteratively Reweighted Penalized Least Squares Baseline Correction in Oil Spectrometer

Academic Journal of Materials & ChemistryISSN 2616-5880 Vol.2， Issue 1： 8-12， DOI： 10.25236/AJMC.2021.020102 Academic Journal of Materials & Chemistry Application of Adaptive Iteratively Reweighted Penalized Least Squares Baseline Correction in Oil Spectrometer Rongwang XuLa，* Kunshan Soohow Instrument Technology Co.， Ltd.， Suzhou， China “Email：xrw@soohow.com *Corresponding author Abstract： In order to solve the problem of baseline drift caused by spectral l ines ofoil spectrometer； an adaptive niteratively reweighted penalized least squares method is proposed to correct the baseline.The oil spectrometer of Kunshan Soohow Instrument Technology Co. Ltd. was used to measure t he contents of various elements in the standard oil with four concentrations of 0ppm， 10ppm， 30ppm and 50ppm. The spectral line data collected by the first CCD was processed， and a good baseline correction effect was obtained. Keywords： A daptive iterative weighting， Least square method， Baseline correction， Oil spectrometer 1. Introduction In order to reduce the friction loss of mechanical equipment， l i fting equipment use safety and service life of the oil is widely used i n mechanical equipment lubricat i on protection， but with the increase of using time or mechanical design or i nstallation i tself on the tiny mismatch defects， such as mechanical equipment running process st i ll produces a small amount of di f ferent degree of wear and tear. These wear wil l produce metal or non-metal abrasive particles i n the lubricat i on system of the equipment， which will affect the lubrication effect of t he oil in t he process of repeated use. More serious， i t will lead to accidents or even complete damage of the equipment ， and the economic l oss will be inestimablell. Therefore， the routine component detection of multi-element abrasive particles in lubricat i ng oil i s of great significance. Such routine and accurate detection can not only r outinely detect the health condition of mechanical equipment， but also provide early warning before the occurrence of equipment wear fault， effectively guaranteeing t he normal operation of equipment!21. Lubricating oil is composed of fewer elements， but its structural state is more complex and exists i n the form of organic matter. The elements measured are mainly the lubricating oil in the operation process due t o equipment wear and tear into the oi l. In the specified time， as long as the wear quantity does not exceed the allowable value， i t i s considered normal wear， normal wear rate can be used to predict and determine the life expectancy of the machine， performance ratio and other indicators； The abnormal wear r ate indicates a potential trouble with the machine. Through the analysis of the lubricat i ng oil sample of check out in the lubrication system， according to the result of the metal elements concentration data can infer the metal working parts wear degree， so as to detect t he cause of equipment failure or avoid device and a fatal fai l ure due to abnormal wear， oil spectrum analysis technology is based on the fact.By analyzing the contents of t hese wear elements (Fe， Al， Cu， Ni， Cr， Sn， Pb， Ag， Cd， Mn， T i， v) B31. Penalty least squares is a flexible smoothing method， whic h mainly includes two aspects： t he smoothing of the signal by penal least squares and the penal l east squares algorithm t hat t ransforms t he penal process i nto a baseline estimation by adaptive i teration[4]. This method i s easy to operate without any user intervention and initial i nformation. Based on the penalized least squares， the sum of squares for error (SSE) weight between the fitting baseline and the original signal i s adjusted adaptively i n t he iterative process. The SSE weight is obtained by the difference between the f itting baseline and t he original signal before the adaptive use. This method can quickly and f lexibly deduct t he baseline of irregular changes[s]. Thi s baseline processing method has l ow data requirements， convenient and fast processing process， and i s very suitable for l arge-scale sample sets. ISSN 2616-5880 Vol.2， Issue 1： 8-12， DOI： 10.25236/AJMC.2021.020102 2. Algorithm principle The experiment uses MATLAB software to adopt adaptive i terat i vely Reweighted Penalized least-squares (airPLS) can eliminate the spectral background in batches f rom t he oil sample spectra[6-8]adaptive iteratively Reweighted Penalized least-squares i s a basel i ne correction method recently proposed by Liang Yizeng et a1.[91. The method is i n t he punishment ， on t he basis of l east square by adaptively adjusting i n the process of i terative fitting between baseline and original signal SSE the weight of residual sum of squares)，quickly and f lexibly to find irregular change of basel i ne and deduction[10]. 2.1 Punish the least square method Penal i zed least-squares algorithm was first proposed to fi l ter. But in r ecent years， i t has been extended to the background subtraction of spectral signals. Here’s how it works. Suppose x is the vector of the analysi s S S si I g 2n T al and z is the fitting vector with l ength m. The f idelity of z against x can be expressed by t he population variance between them： The roughness of f itted data z can be expressed by the sum of squares of i ts difference： The balance between precision and roughness can be expressed by Q： 2.2 Adaptive iteration weighting The adaptive i terat i ve reweighting method is similar t o the Weighted Least Square and Iterative Penalty Least Square processes， but t he di f ference l ies in the weight calculation method and the addition of penalty terms to control the smoothness of t he fitting background. The optimized objective function i s： Each iteration w can be obtained as follows： The vector d' contains all elements whose di f ference between the original Raman spectrum (X) and the fitting vector Z/ i n the iteration process is negative. In t he first t-1 i teration， the f itting vector i s a candidate for background est i mation. If the current calculation is greater t han one candidate of this background est i mate， i t is considered to be i n t he position of t he peak. At this point， t he weight is r eset to 0 so that it does not affect the next i teration. In the process of air-PLS algorithm， i teration and reweighting are carried out automat i cally continuously， so that t he data points i n the peak position can be gradually eliminated and the background points can be retained in t he weight vector. The number of iterations or some convergence condi t ion can be set to t erminate t he iteration. ISSN 2616-5880 Vol.2， Issue 1： 8-12， DOI： 10.25236/AJMC.2021.020102 3. Experiment 3.1 Experimental Instruments The experiment was carried out using OIL8000 oil spectrometer (Kunshan Soohow) as shown in Figure 1. Optical system： high-performance holographic diffraction grating， grating lines 2700 /mm， focal length 500 mm； Multi-block high performance CCD detector system； Each CCD has 3，648 pixels， andt here are eight CCDs in t otal. Sspark power supply： oscillating arc discharge control， electronic acquisition and data readout system； The electronic system has t he function of multi-channel integration and data collection system controlled by microprocessor. High speed 16-bit analog-to-digital converter. Fig.1.OIL8000 oil spectrometer 3.2 Experimental samples Petroleum ether： analytical pure； Standard oi l ： 0×10，10×106，30×106，50×10-6PPM，respectively. The standard oi l i s shown in Figure 2. Fig.2.Standard oil 3.3 Experimental process In the best working condition of the instrument and t he selected working condit i ons， a series of standard oil samples are excited， and the working curve is drawn， and t hen the analysis samples are excited under the same working conditions. The spectra of t he first CCD at four concentrations were collected， and the spectra of the four concentrations were used as adaptive i terative weighted penalized least square method to correct the baseline. 4. Experimental results and analysis The experiment uses MATLAB software to adopt adaptive iterat i vely penalized l east square method Reweighted Penalized least -squares (airPLS) is used t o deduct the baseline batch from t he spect r a collected by t he f irst CDD of oi l samples with concentrations of 0×106， 10×106，30×106， and 50×106PPM. The effect before and after baseline correction i s shown i n Fig. 3-6. c (b) Spectral lines after baseline correction Fig.3. Standard oi l spectral l ine with a concentration of 0ppm (a) Spectral lines before baseline correction (b) Spectral lines after baseline correction Fig.4. Standard oi l spectral line with a concentration of 10ppm (a) Spectral lines before baseline correction (b) Spectral lines after baseline correction Fig.5. Standard oil spectral line with a concentration of 30ppm ISSN 2616-5880 Vol.2， Issue 1： 8-12， DOI： 10.25236/AJMC.2021.020102 (a) Spectral lines before baseline correction (b) Spectral lines after baseline correc t ion Fig.6. Standard oil spectral line with a concentration of 40ppm Through careful observation of Fig. 3-6， i t can be seen that t he baseline of t he spectral lines of each concentration is corrected to the vicinity of 0， reaching a relat i vely stable state. Such prominent baseline correction effect plays a role in improving the measurement accuracy in t he subsequent transition from spectral peak intensit y t o element content. 5. Conclusions In this paper， the adaptive iterative weighted penal i zed least square method is used to calibrate the spectra obtained by oil spectrometer. The experimental results show t hat t he adaptive i t erat i ve weighted penalized l east square method， as an advanced and simpl i fied basel i ne correction algorithm， h as a good application prospect for the spectral processing of oil spectrometer (i.e. r otating disc atomic emission spectrometer). References [1] Wu Xiaobo. Research on Preventive Measures of Large-scale Transformer Sudden Damage [D].North China Electric Power University， 2014. [2] Zhang Liuliu. Design and Research of Oil Spectral Excitation System Based on Constant Voltage Variable Frequency DC Spark Light Source [D]. China Jiliang University，2017. [3] Zhao Lihua， Ma Yue， Wei Miao. Analysis of 19 Elements in Lubricating Oil by Oil Spectrometer [].Baotou Steel Science and Technology， 2012，38(06)：40-42. [4] LI Shuifang， ZHANG Xin， LI Jiaojuan， et al. Nondestructive Determination of Fructose and Glucose Content in Honey by Raman Spectroscopy []. Transactions of the Chinese Society of Agricultural Engineering， 2014， 30(6)：249-255. [5] Wen Jianhui， Zhong Kejun， Jiang Jianhui， et al. An adaptive iterative weighted penalized least square method for baseline correction of flue gas chromatography. Beijing： Chinese Chemical Society，2011. [6] Yang Xin， Guo Pengcheng， Xu Chuanwei， Huang Wenjuan， Chen Xiangbai. Water quality analysis by portable Raman spectrometer combined with chemometric method []. Journal of Wuhan Institute of Technology， 2020，42(01)：28-32. [7] BURROWS C J ，STAPELFELDT K R，WATSON AM，et al. Hubble space telescope observations of the disk and jet ofHH 30 [J]. The Astrophysical Journal，2009，473(1)：436-437. [8] Meneghini C，Caron S，Poulin A C J.Determination ofethanol concentration by Raman spec t roscopy in liquid-core microstructured optical fiber [J]. IEEE SENSORS JOURNAL，2008，8( 7)： 15201525. [9] Zhiming Zhang，Shan Chen， Yizeng Liang. Baseline correction using adaptive interatively reweighted penalized least squares[J].Analyst，2010，135(5)：1138-1146. [10] Jianhui Wen， Kejun Zhong， Jianhui Jiang， Lijuan Tang. Study on adaptive iterative weighted penalized least squares for baseline correction of flue gas chromatography [A]. Chinese chemical society. Chinese Chemical Society： Chinese Chemical Society，2011：2.