Research On The Verification Of Grade Asphalt Based On Infrared Spectrum

As a basic material for pavement construction,asphalt quality directly affects the performance and service life of the entire road.In order to ensure the effective construction of roads,it is essential to control the quality of asphalt at the road construction site.Among them,ensuring that the grade of asphalt transported meets the specifications is the primary problem facing road engineers.At present,the identification of grade asphalt mainly uses some complicated physical and chemical methods,such as: dynamic shear rheology experiment,gel permeation chromatography,etc.,but these methods have the disadvantages of time-consuming,low accuracy,and inconvenience,which are difficult to meet Road construction site identification requirements for rapid and accurate asphalt materials.As a potentially convenient,efficient,fast,and low-cost analysis technique,ATR-FTIR(Attenuated Total Reflection Fourier Transform Infrared Spectroscopy)is gradually being used in the fields of food quality analysis,industrial component detection,and so on.Under this background,this paper studies a new method for rapid identification of graded asphalt.This method uses ATR-FTIR spectroscopy as the data carrier of asphalt and machine learning algorithm as the core to construct the final graded asphalt identification model.Affected by the collection conditions,the collected original spectrum usually has noise,baseline shift,light scattering and other noise.Usually,S-G smoothing is used to eliminate noise,scatter correction and standard normalization are used to mitigate the effect of spectral scattering,and derivative correction is used to eliminate baseline shift.Since different spectral pretreatment has its own focus on the ability to improve the quality of the spectrum.It is necessary to explore the effect of different pretreatments and their combination methods on the ATR-FTIR spectrum of grade asphalt.In order to obtain the optimal spectral pretreatment method,this paper constructs different spectral pretreatment strategies.According to the spectral principal component analysis of different pretreatment strategies,the optimal graded asphalt pretreatment strategy is obtained.The experimental results show that the graded asphalt ATR-FTIR In terms of spectrum,S-G smoothing + scattering correction + standard normalization is the conclusion of the best spectral preprocessing method.In order to make up for the defect that the single feature extraction method is difficult to fully extract the effective features of the graded asphalt spectrum,this paper proposes to extract the effective features of the spectrum based on the multidimensional feature fusion reconstruction method.This method mainly extracts spectral features from two aspects: feature selection and feature fusion.Finally,the multi-angle features extracted by multiple methods are subjected to secondary fusion to construct graded asphalt spectral effective features.The experimental results show that the multiangle fusion features proposed in this paper can characterize the differences between grades of asphalt more than the features extracted by a single algorithm.In addition,in view of the problem of the effectiveness of partial least squares coefficients in competitive adaptive reweighting algorithms,this paper proposes an improved algorithm that improves partial least squares multiple times on the original basis and obtains the importance of variables Evaluation indicators are more effective.After extracting effective features,considering the complex and changeable environment of the construction site,establishing an identification model with good performance and strong generalization ability is the most critical step.Stacking integrated learning has the advantages of good smoothness,highlighting the performance of the optimal base model and diluting the poor base model,and the established model is effective and stable.Therefore,this paper uses Stacking integration method to model fusion of three typical classifiers(Support Vector Machine,Random Forest,Extreme Gradient Boosting)to establish the final grade asphalt identification model.The comparison experiment results show that the multi-classifier fusion model achieves 100% and 99.21% accuracy in the cross-validation and test sets respectively,its performance is higher than that of any single-classification model before fusion,and its generalization ability is stronger.Above all,the asphalt grade verification method proposed in this paper can be used in China's common petroleum asphalt grade verification,providing important means and basis for asphalt quality supervision.