Detection Of Sulfur Content In Diesel Oil Based On Near-Infrared Spectroscopy

The primary source of power for diesel engines is diesel.After burning,the sulfur and sulfides in diesel will not only shorten the life of the engine’s engine by causing corrosion on the mechanical surface and deterioration of the lubricating oil.Fast,real-time,and non-destructive testing requirements cannot be reached by the conventional method of detecting the sulfur level in diesel because it is either difficult to analyze or has low reproducibility.The field of diesel detection research has quickly advanced near-infrared spectroscopic detection technology due to its low cost,excellent repeatability,and sensitivity.In this study,samples with varying levels of diesel sulfur concentration were generated,yielding a total of 161 samples.A Fourier transform near-infrared spectrometer was used to gather the near-infrared spectra of each sample of diesel with sulfur content.The influence of various pretreatment techniques on spectral modeling was compared using the near-infrared near-infrared spectrum of the diesel sulfur content.Different characteristic wavelength screening algorithms were then used to further optimize the model,and a quantitative detection model of SSA-BP neural network was established.To determine the range of diesel sulfur concentration,the near-infrared spectrum is converted into a two-dimensional image,and the method of integrating MTF image coding and convolutional neural network is investigated.The SSA-SVM model and the MTF-CNN diesel sulfur concentration interval identification model have both been developed.Ultimately,a MATLAB-based diesel sulfur content detecting software is created.The following constitutes the paper’s primary research findings:（1）The models for the quantitative determination of diesel sulfur content with PLS,BP,and SSA-BP were developed using near-infrared spectroscopy.The original spectrum is preprocessed using techniques like decentralization,normalizing,multivariate scattering correction,15-point quadratic smoothing,first-order differential,and second-order differential in the whole spectrum range of 12493～4458 cm^-1.The most effective preprocessing technique is second-order differential+15-point quadratic smoothing.Amongst them,the R² value is 0.972,the RMSEP value is 50.07,and the RPD value is 5.57.The whole spectrum quantitative analysis model was optimized using three wavelength selection techniques:BOSS,CARS,and MCU-VE.Following comparison,the BOSS algorithm outperformed the others,simplified the model,and enhanced its predictive power.The BOSS method extracts a total of 38wavelength points,and creates a PLS model using those 38 wavelength points.Among these,the R² value is 0.977,the RMSEP value is 39.79,and the RPD value is 7.01.As the quantitative detection model of diesel sulfur content,the BP neural network model optimized by the SSA method was chosen.The SSA-BP algorithm’s R² value is 0.982,its RMSEP value is 29.0,and its RPD value is 7.52.（2）A two-dimensional picture of the near-infrared spectrum was created,and the SSA-SVM model and the MTF-CNN diesel sulfur concentration interval identification model were evaluated.According to the concentration interval,the sulfur content of various diesel oils was categorized,and the limit of 350 mg/kg was used to separate the sulfur content into two categories:qualified and unqualified,of which 99 were qualified and 62 were unqualified.To create an SSA-SVM identification model,the spectrum is preprocessed using techniques such decentralization,normalization,multivariate scattering correction,15-point quadratic smoothing,first-order differential,and second-order differential.The established SSA-SVM achieved the best identification accuracy rate of 96.97%among them using the first-order differential+15-point quadratic smoothing preprocessing.MTF image encoding is used to transform1D spectral sequence data into 2D images.Create a CNN structure based on the Le Net-5 scheme design,and create a model for identifying intervals of diesel sulfur content concentration based on MTF angle field image coding and CNN.The accuracy of the MTF-CNN classification model in identifying diesel sulfur content in various concentration intervals is 96.97%following the preprocessing of the second order differential+15 points quadratic smoothing.The outcomes demonstrate that employing MTF mathematical image coding technique,one-dimensional near-infrared spectral data may be combined with deep learning image processing.（3）The software for detecting the sulfur content of fuel was created using MATLAB.The software can analyze and identify the gathered near-infrared spectrum of diesel sulfur content in a clearer and more succinct way.It also contains near-infrared data reading of diesel sulfur content,model training,and detection based on diesel sulfur content near-infrared spectrum.