Aggregate fingerprints identification based on its compositions and machine learning algorithm

Wei Wang; Chenchen Wang; Junan Shen; Xinsheng Li

doi:10.1016/j.arabjc.2023.104810

View/Download PDF

Buy Reprints

PDF

Translate this page into:

Review article

07 2023

:16;

104810

doi:

10.1016/j.arabjc.2023.104810

Aggregate fingerprints identification based on its compositions and machine learning algorithm

Wei Wang^{^a,^b}, Chenchen Wang^{^a}, Junan Shen^{^c}, Xinsheng Li^{^a,⁎}

a Jiangsu Technology Industrialization and Research Center of Ecological Road Engineering, Suzhou University of Science and Technology, Suzhou 215011, China

b School of Civil Engineering, Chongqing Jiaotong University, Chongqing 400074, China

c Department of Civil Engineering and Construction, Georgia Southern University, Statesboro, GA 30458, USA

⁎Corresponding author. Lxs_sz@126.com (Xinsheng Li)

Received: 2022-9-18, Accepted: 2023-3-11,

Disclaimer:
This article was originally published by Elsevier and was migrated to Scientific Scholar after the change of Publisher.

Peer review under responsibility of King Saud University.

Abstract

The type and properties of an aggregate affect the properties of their mixtures with either Portland cements or asphalt binders. How to quickly identify the information on an aggregate, providing a reliable basis for the quality assurance and quality control of aggregates, i.e., guarantee the source of aggregates is vital important. The purpose of this study is to explore a new and rapid detective technology for aggregate fingerprint identification using Fourier Transform Infrared Spectroscopy (FTIR). Machine learning algorithm of statistical analysis software (SPSS) was performed for principal component analysis, cluster analysis and linear discriminant analysis on collected information of the aggregates. The results showed that the aggregates of the same origin can be aggregated well by principal component analysis, cluster analysis and linear discriminant analysis as well. The cross-validation accuracy is very high.

Keywords

Aggregates

Fingerprint identification

FTIR

Machine learning

Show Related Articles from PubMed

1 Introduction

As an asphalt pavement has the characteristics of comfortable driving and easy maintenance over concrete cement pavement, it is widely used in road construction. The combined effect of weathering and vehicle loading on the asphalt pavement will cause many failures such as cracks, rutting and water damage and the other early diseases, consequently seriously affecting its service life. The poor quality of the aggregates is one of the main reasons of those failures. The selection of high-quality aggregates and their quality assurance during the production of the mixtures are extremely important (Xie et al., 2014; Liao, 2018; Ren, 2020; Dai et al., 2018 ). Current evaluation of aggregates mainly uses their mechanical and physical properties (such as crushing value, abrasion value, angularity, etc.). Although the lithology and micro-features of aggregates have been studied in the field of engineering, very few has been in actual use (Wang, 2019; Wang, 2014). For quality control, testing for macro machinal properties requires many samples, and takes a long time to complete.

The compositions of aggregates are complex, which affected their cement and asphalt mixtures due to the different geological conditions of their quarries (Jia, 2012). For an example, the main chemical components of basalt are silicon dioxide, aluminum oxide, iron oxide, calcium oxide, and magnesium oxide (Yu and Chen, 2020). Each aggregate produced from different quarries holds its distinct composition of the various percent of the components, as unique as a human fingerprint, which is called aggregate fingerprint information. That is, aggregates of different origin have unique fingerprint information, so the determination of fingerprint information is helpful to the analysis of aggregate origin and composition.

Several research have been done on the fingerprint recognition of asphalt binders (Liu et al., 2013; Wang et al., 2018; He et al., 2016; Xu et al., 2020; Margaritism et al., 2020 ). Ren et al. used FTIR and principal component analysis (PCA)-linear discrimination to discriminate asphalt samples from three different origins, with an accuracy of 96.2%, provided a scientific basis for rational selection, quality control and guarantee of asphalt quarries (Ren et al., 2019). Cheng used Fourier transform infrared spectroscopy combined with chemometrics, infrared spectra of different brand asphalt recognition and quantitative analysis, the results showed that the characteristics of different brand, label asphalt absorption peak position was consistent, but the intensity of obvious differences, infrared spectra of different asphalt tiny difference effectively distinguish between (Cheng and Kong, 2020). Cui in the division of highway asphalt pavement maintenance sections based on principal component-clustering method, determined the priority of maintenance of corresponding sections according to the results of principal component-clustering analysis and the average value of four evaluation indexes of pavement performance after clustering, and the division results could be used for scientific maintenance decisions (Cui et al., 2019). In other aspects, fingerprint identification using Fourier transform infrared spectroscopy and principal component - linear discrimination method is also relatively mature. Kuhnen and Shirley used Fourier transform infrared spectroscopy and principal component analysis to identify the flour of 26 maize local varieties in southern Brazil. Fourier transform infrared spectroscopy and principal component analysis showed that Fourier transform infrared spectroscopy and principal component analysis are effective tools for rapid screening and identification of maize with different chemical components (Kuhnen et al., 2010).

The purpose of this paper is to explore a new fast detection technology for aggregate fingerprint identification. The characteristic functional groups of aggregates from different producing areas were identified and quantitatively analyzed by Fourier transform infrared spectroscopy. Principal component analysis and cluster analysis of infrared spectral data were carried out by machine learning method. The discriminant function is established by linear discriminant analysis, and a fast, accurate, nondestructive and stable method for identifying aggregate origin was explored. The research results provide scientific basis for rational selection, quality monitoring and guarantee of aggregate source.

2 Materials and methods

2.1

2.1 Materials

Three types of aggregates including limestone, basalt and granite, which are commonly used in asphalt mixtures for road pavement, were selected. Each aggregate came from a different place of origin. Limestone came from Jiangxi, Hunan, Anhui, Hebei and Zhejiang provinces in China. Basalt came from Hainan, Shandong, Anhui, Inner Mongolia and Hebei provinces in China. Granite came from Jiangsu, Anhui and Hebei provinces in China, respectively.

2.2

2.2 Test methods

Thermo Fisher’s IS5 FTIR was used. FTIR uses the absorption characteristics of different wavelengths of infrared radiation to analyze molecular structure and chemical composition. The scanning range was 400–4000 cm⁻¹. The number of scanning for each sample were 16. The resolution was 4 cm⁻¹. Background scanning was carried out for each scan to eliminate the interference on the spectrum of the sample.

The procedure for FTIR test was as follows: 1) Grind the aggregate into a particle size of less than 2 µm with a ball mill to avoid the influence of scattered light. 2) Fill the probe slot with 50 mg samples should uniformly for scanning, as shown in Fig. 1. 3) After each experiment, the sample table was cleaned with alcohol to avoid affecting subsequent experiments. 4) A program written by OMNIC software performs baseline correction on the original spectral data to eliminate the baseline effect.

FTIR set used for this study: (a) FTIR test instrument; (b) FTIR sample.

2.3

2.3 Analysis method

All FTIR data were analyzed by the combination of principal component analysis (PCA), cluster analysis and linear discriminant analysis (LDA), and all the data were analyzed in SPSS. SPSS is a commonly used machine learning software for statistical analysis operations, data mining, predictive analysis and decision support tasks. Principal component analysis is a method to convert multiple variables into a few principal components through dimension reduction techniques. These principal components retain most of the information of the original variables and are usually expressed as linear combinations of the original variables (Wang, 2014). For example, there are n samples, each sample observes P indicators, and the original data is written as a matrix.

(1)

X = [\begin{matrix} X 11 & X 12 & \dots & X 1 p \\ X 21 & X 22 & \dots & X 2 p \\ ⋮ & ⋮ & ⋮ & ⋮ \\ X n 1 & X n 2 & \dots & Xnp \end{matrix}]

First, the original data is standardized and the correlation coefficient matrix R of variables is established.

(2)

R = [\begin{matrix} X 11 & X 12 & \dots & X 1 p \\ X 21 & X 22 & \dots & X 2 p \\ ⋮ & ⋮ & ⋮ & ⋮ \\ X p 1 & X p 2 & \dots & Xpp \end{matrix}]

(3)

r_{ij} = \frac{\sum_{k = 1}^{n} (x_{ki} - {\bar{x}}_{i}) (x_{ki} - {\bar{x}}_{j})}{\sqrt{\sum_{k = 1}^{n} {(x_{ki} - {\bar{x}}_{i})}^{2} \sum_{k = 1}^{n} {(x_{ij} - {\bar{x}}_{j})}^{2}}}

Then find the eigen roots of R and the corresponding unit eigenvectors.

(4)

a_{1} = [\begin{matrix} a_{11} \\ a_{21} \\ \cdot \cdot \cdot \\ a_{p 1} \end{matrix}], a_{2} = [\begin{matrix} a_{12} \\ a_{22} \\ \cdot \cdot \cdot \\ a_{p 2} \end{matrix}], . . ., a_{p} = [\begin{matrix} a_{1 p} \\ a_{2 p} \\ \cdot \cdot \cdot \\ a_{pp} \end{matrix}]

Principal component expression: i = 1,…,p.

The principal component contribution rate and its cumulative contribution rate are calculated.

Contribution:

(5)

\frac{λ i}{\sum_{k = 1}^{p} λ k} (i = 1, 2, \cdot \cdot \cdot, p)

Cumulative contribution rate:

(6)

\frac{\sum_{k = 1}^{i} λ k}{\sum_{i = 1}^{p} λ k} (i = 1, 2, \cdot \cdot \cdot, p)

Generally, the component whose principal component eigenvalue is greater than or equal to 1 is selected as its principal component.

Systematic clustering analysis is one of the most widely used clustering methods in various research fields at present. Its basic idea is as follows: the closest samples are collected into clusters first, and the distant samples are clustered into clusters later. The process goes on all the time, and each sample can finally be clustered into the appropriate class. Sample spacing is used to measure their similarity. There are many methods to calculate sample distance, and Euclidean distance is used in this paper (Dong, 2016).

Euclidean distance is:

(7)

d (x_{i}, x_{j}) = {[\sum_{k = 1}^{p} {(x_{ik} - x_{jk})}^{2}]}^{\frac{1}{2}}

Let d_ij = d(x_i,x_j), D = (d_{i j}) _p×p to form a distance matrix:

(8)

[\begin{matrix} 0 & d_{12} & \dots & d_{1 n} \\ d_{21} & 0 & \dots & d_{2 n} \\ ⋮ & ⋮ & 0 & ⋮ \\ d_{1 n} & d_{2 n} & \dots & 0 \end{matrix}]

Where: d_ij = d_ji; d_ij is the distance between variables i and j.

According to the nearest distance matrix, the two samples with the nearest distance are combined into one class. According to the requirements of the Euclidean distance method, the sum of square deviation method is used for clustering. When G_p and G_q are merged into G_r, the distance from other G_k is the recurrence formula is:

(9)

D_{rk}^{2} = \frac{n_{k} + n_{p}}{n_{r} + n_{k}} D_{pk}^{2} + \frac{n_{k} + n_{q}}{n_{r} + n_{k}} D_{qk}^{2} - \frac{n_{k}}{n_{r} + n_{k}} D_{pq}^{2}

Where: np, nk, nr, nq are the number of samples of G_p, Gk, Gr, and Gq, respectively.

Linear discriminant analysis firstly establishes functions (linear combinations of independent variables) based on the properties of things of known categories (independent variables), and then judges new things of unknown categories to put them into known categories (Zhang and Li, 2015). The discriminant function group automatically established by SPSS through discriminant analysis is:

(10)

\{\begin{matrix} d_{i 1} = b_{01} + b_{11} x_{i 1} + \cdot \cdot \cdot + b_{p 1} x_{i p} \\ d_{i 2} = b_{02} + b_{12} x_{i 1} + \cdot \cdot \cdot + b_{p 2} x_{i p} \\ \cdot \cdot \cdot \\ d_{i k} = b_{0 k} + b_{1 k} x_{i 1} + \cdot \cdot \cdot + b_{p k} x_{i p} \end{matrix}

Where, k is the number of discriminant functions in the discriminant function group, and is the value of function min (number of categories −1, number of prediction variables);

d_ik is the value of the i^th case obtained by the KTH discriminant function;

p is the number of prediction variables;

b_jk is the JTH coefficient of the KTH discriminant function;

X_ij is the value of the JTH predictor variable in the ith case.

3 Results and discussions

3.1

3.1 FTIR

Fig. 2 showed FTIR results of the limestone, basalt, and granite. The information of the peak positions for the four types of aggregates was provided as the following. For granite aggregates, 4 main absorption peaks were selected as variables, as shown in Fig. 3, including the absorption peak around 1000 cm⁻¹ caused by the stretching vibration of SI-O. It is related to the participation of AL ions. The absorption peak about 780 cm⁻¹ is the extensional vibration absorption of SI-SI. The absorption peak around 725 cm⁻¹ is absorbed by the extensional vibration of SI-AL, but is related to the participation of AI ions. The absorption peak around 649 cm⁻¹ is O-SI-O stretching vibration absorption (Li et al., 2015). For limestone, four main absorption peaks are also selected as variables. The absorption peak around 1420 cm⁻¹ is composed of [CO3]^2- internal stretching vibration. The absorption peak at about 1000 cm⁻¹ is caused by the stretching vibration of Si-O. The absorption peak around 875 cm⁻¹ is caused by the out-of-plane bending vibration of [CO3]^2-. The absorption peak around 712 cm⁻¹ is caused by [CO3]^2- in-plane bending vibration. For basalt aggregates, the peak is near 1000 cm⁻¹, which is formed by Si-O stretching vibration (Yang et al., 2015).

Typical infrared atlas of three different aggregates.

Fig. 3 showed spectra of the 10 limestone from Jiangxi province. FTIR of the same kind of aggregates are basically the same in the trend of absorbance, and the subtle differences in the spectra indicate the difference in their origins. The same type aggregates were obtained in different quarries, while the peak positions and intensities of the same functional groups appeared slightly different from the quarries. This means that the spectra and composition of the same aggregate from different provinces are more different.

3.2

3.2 Analyses and discussions -----PCA

3.2.1

3.2.1 Granite

FTIR tests were performed on granite from three different areas, 10 samples of granite from each area were tested. The absorbance values of 4 characteristic absorption peaks in the infrared spectrum of 30 samples were introduced into machine learning based SPSS. The results are shown in Table 1 below.

Table 1 FTIR PCA results of granite.

Component	Initial eigenvalue
Component	Total	Percentage variance	Cumulative percentage%
1	3.71	92.52	92.52
2	0.17	4.24	96.76
3	0.11	2.75	99.51
4	0.02	0.49	100.00

There was a principal component: PCA1 whose eigenvalue was greater than or equal to 1, and the variance contribution rate of the first principal component is 92.52%. The principal component value of each aggregate sample according to the component score coefficient matrix was calculated. The expression of PCA1 by SPSS software was calculated and the output its expression is as follows:

(11)

PCA = 0.51 \times X_{1} + 0.52 \times X_{2} + 0.49 \times X_{3} + 0.49 \times X_{4}

Where, X₁-X₄ are the absorbance values around 1000, 780, 725, and 649 wave numbers in the infrared spectrum, respectively.

According to the principal component value of each sample, a scatter figure is plotted as shown in Fig. 4. The aggregates of the three origins were distributed differently as can be seen by the PCA1 scores by FTIR. The PCA score of the aggregates from Jiangsu was the largest, greater than 0.06. That from Hebei was the smallest, less than 0.036. The score in Anhui was between 0.036 and 0.06. This means that the method can effectively distinguish the same aggregate from different origin.

3.2.2

3.2.2 Limestone

The calculation results of 50 limestone samples (from five different regions, 10 samples per region) were shown in Table 2. There were two principal components: PCA1 and PCA2. The variance contribution rate of the first two principal components is 78.87%. The first two principal components represented most of the original data.

Table 2 FTIR PCA results of limestone.

Component	Initial eigenvalue
Component	Total	Percentage variance	Cumulative percentage/%
1	3.13	52.17	52.17
2	1.62	26.70	78.87
3	0.93	15.53	94.40
4	0.17	2.76	97.16
5	0.12	2.06	99.22
6	0.05	0.78	100.00

The principal component value of each aggregate sample according to the component score coefficient matrix is calculated. The expression of PCA1 and PCA2 by SPSS software is calculated and the output its expression is as follows:

(12)

P C A 1 = - 0.42 X 1 - 0.20 X 2 - 0.47 X 3 + 0.53 X 4 + 0.22 X 5 + 0.49 X 6

(13)

P C A 2 = 0.48 X 1 + 0.15 X 2 + 0.59 X 3 + 0.19 X 4 + 0.70 X 5 + 0.31 X 6

Where, X₁-X₆ are the wave numbers and absorbance values of the infrared spectrum around 1420, 1000 and 875, respectively.

According to the principal component score of each sample, a scatter figure is plotted, as shown in Fig. 5. The limestone aggregates of the five origins were again distributed independently, which can be effectively distinguished by FTIR. As a whole, FTIR classified two PCA that grouped the aggregates clearly from the quarries of the aggregates again.

3.2.3

3.2.3 Basalt

The calculation results of 50 basalt samples (from five different regions, 10 samples per region) were shown in Table 3. There were two principal components: PCA1 and PCA2. The variance contribution rate of the first two principal components is 75.92%. Therefore, the first two principal components are selected as the principal components.

Table 3 FTIR PCA results of Basalt.

Component	Initial eigenvalue
Component	Total	Percentage variance	Cumulative percentage
1	1.97	49.16	49.16
2	1.07	26.76	75.92
3	0.71	17.70	93.62
4	0.26	6.38	100.00

(14)

P C A 1 = - 0.44 X 1 + 0.15 X 2 + 0.64 X 3 + 0.64 X 4

(15)

P C A 2 = 0.39 X 1 - 0.86 X 2 + 0.23 X 3 + 0.24 X 4

Where, X₁-X₄ are the wave numbers and absorbance values of the infrared spectrum around 1000 and 630 respectively.

According to the principal component score of each sample, a scatter figure is plotted, as shown in Fig. 6. There were two PCA identified by FTIR Basal. The PCA scores of each aggregate took their area that can differ their aggregate sources.

3.3

3.3 Cluster analysis

The main idea of cluster analysis is to group and classify all data, so that similar data together, and different data distributed in different groups. According to this idea, the test results of FTIR of different aggregates from different regions can be used for classification and statistics of aggregates.

The transmittance values of 4 characteristic absorption peaks in FTIR spectra of the 30 granite samples were further introduced into the SPSS software for cluster analysis and the pedigree diagram is shown in Fig. 7. The horizontal axis in the figure represents the Euclidean distance and vertical axis represents the sample number. When Euclidean distance value was about 6–11, we can divide the pedigree diagram into three main clusters. It can be seen from the figure below that the samples numbered 1 ∼ 10, 11 ∼ 20 and 21 ∼ 30 can be grouped together when Euclidean distance value was 6–11. For example, the granite aggregate in Jiangsu, samples 3,6,7, and 10 were grouped into group A, samples 2,4,5 and 8 were grouped into group B, samples 1 and 9 were grouped into group C. Then groups a and b were gathered together first, and group c was gathered together last. Nos. 1 ∼ 10 were the samples from Jiangsu, Nos. 11 ∼ 20 were the samples from Anhui, and Nos. 21 ∼ 30 were samples from Hebei Formation. This is completely consistent with actual situation and the results of principal component analysis. The analysis results of the other two types of aggregates were similar, and their pedigree diagrams were not given due to space constraints.

Cluster analysis pedigree chart of different aggregates.

3.4

3.4 Linear discriminant analysis

In principal component analysis and cluster analysis, aggregates from the same origin can gather together well, indicating that there are differences in aggregates in different regions, which provides the possibility to distinguish aggregates from different origins. According to the principal component method, the information of the infrared spectrum of each aggregate sample was further introduced into SPSS software for linear discriminant analysis. The information of aggregate infrared spectrum was taken as independent variable, and the classification number of origin obtained by principal component analysis and cluster analysis was introduced into SPSS software as grouping variable. Through linear discriminant analysis, Fisher discriminant function eigenvalue was obtained. For example, the results of limestone discriminant analysis are shown in Table 4. The cumulative contribution value of the first two discriminant functions is 100%, and the corresponding discriminant functions 3.6 and 3.7 are obtained. Fig. 8 below presented the results of three aggregates, indicating that the origin grouping of each type pf aggregate was very good.

(16)

Y_{1} = - 2459.29 + 2.19 \times X_{1} + 0.08 \times X_{2} - 0.81 \times X_{3} - 13.98 \times X_{4} \dots . + 109.25 X_{5} + 17.59 X_{6}

(17)

Y_{2} = 581.47 - 0.36 \times X_{1} + 0.30 \times X_{2} - 0.44 \times X_{3 - 53.71} \times X_{4} + 129.90 X_{5} + 44.95 X_{6}

Table 4 Characteristic values of limestone discriminant function.

Function	Eigenvalue	Variance/%	Cumulative/%	Canonical correlation
1	35.18	85.14	85.40	0.99
2	5.59	14.60	100.00	0.93

Discriminant function scatter plot of different aggregates.

Where, Y1-Y2 are discriminant functions, and X1-X6 are the wave numbers and absorbance values of the infrared spectrum around 1420, 1000 and 875, respectively.

In order to verify the accuracy of the discriminant function, SPSS software was used to cross-validate the discriminant function, the results were shown in Table 5. The correct classification rate of the original group of granite shown was 100% (that is, the total correct judgment rate of the discriminant function and the actual classification). Cross-validation was performed on 30 aggregate samples. The total discrimination accuracy was 100%. The correct classification rate of the original limestone group was 100.00%. The total discrimination accuracy of 50 aggregate samples was 98.00%. The original group of basalt was correct the classification rate was 100%, and 50 aggregate samples were cross-validated, and the total discrimination accuracy was 96.00%. This showed that the discriminant function in the experiment was stable and can be used to discriminate the source of a set of samples.

Table 5 The original and cross-validation of the FTIR Fisher discriminant function.

Aggregate type	Original result	Cross-validation result
Granite	100.00%	100.00%
Basalt	100.00%	96.00%
Limestone	100.00%	98.00%

4 Validations of the method in engineering project

In order to further verify the fingerprint identification technology, relevant experiments were carried out by sampling from actual projects. Sampling section is Chunshen Road, Xiangcheng District, Suzhou. The aggregate used for asphalt pavement is basalt, which comes from Inner Mongolia.

The 5 samples were extracted by on-site drilling core sampling, and the aggregate in the sample was extracted by crushing, and then the infrared spectrum was tested. After pretreatment, the spectral data were substituted into the model of Formula 14 and 15 for verification, and the results were shown in Fig. 9.

FTRI tests were carried out on 5 basalt samples from the engineering site and calculated by substituting them into Formula 14 and 15. Combined with the results shown in Fig. 6, PCA1 and PCA2 of these 5 samples were in the Inner Mongolia basalt group. So it can be determined that the aggregate samples were from Inner Mongolia, which was consistent with the actual situation. The results provided the feasibility for the application of aggregate infrared spectrum fingerprint identification in road engineering.

5 Summaries and conclusions

In this paper, attenuation total reflection Fourier transform infrared spectrometer combined with machine learning method is used to identify the origin of aggregates. The main conclusions are as follows:

Principal component analysis method is used to reduce dimension, and the obtained principal component score map can realize the identification and differentiation of limestone, basalt and granite origin. The origin of unknown aggregates can be determined by substituting FTIR spectra data of unknown aggregate samples into corresponding principal component functions.
Cluster analysis results: According to the Euclidean distance on the abscess of the pedigree chart, limestone, basalt and granite can complete the clustering and classification corresponding to different producing areas under the corresponding Euclidean distance, with high accuracy.
With infrared spectral information as independent variable, and principal component and clustering analysis to obtain the corresponding aggregate origin classification number as the classification variable, linear discriminant analysis can be carried out, and the discriminant function can be established to draw scatter diagram, which can determine and classify the origin of each aggregate well. At the same time, the classification accuracy of original group and cross validation is above 90%, which has reference value.

The monitoring of aggregate producing area can be realized by substituting the materials used in engineering into the corresponding model.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References

Cheng Z.Q., Kong F.S., . Identification and analysis of asphalt of different brands based on infrared spectroscopy and stoichiometry. J. Wuhan Univ. Technol.. 2020;44:909-913.
[Google Scholar]
Cui Y.J., Li B., Wang J., . Maintenance section division of expressway asphalt pavement based on principal component clustering method. Highway. 2019;64:290-296.
[Google Scholar]
Dai Z., Shen J.N., Shi P.C., . Nano-sized morphology of asphalt components separated from weathered asphalt binders. Constr. Build. Mater.. 2018;128:588-596.
[Google Scholar]
Dong H.S., . A comparative study of principal component analysis and linear discriminant analysis for dimensionality reduction. Modern Computer. 2016;29:36-40.
[Google Scholar]
He S.J., Wang C.Y., Li Y., . Evaluation of diagnostic ratios of medium and serious weathered oils from five different oil sources. Acta Oceanol. Sin.. 2016;35:1-8.
[Google Scholar]
Jia L., . Study on the effect of aggregate mineral composition on road performance of asphalt mixture. Xi’an, China: Chang 'an university; 2012. Master’s Thesis
Kuhnen S., Ogliari J.B., Dias P.F., . ATR-FTIR spectroscopy and chemometric analysis applied to discrimination of landrace maize flours produced in southern Brazil. Int. J. Food Sci. Technol.. 2010;45:1673-1681.
[Google Scholar]
Li J., Cai T., Wang N., . A qualitative identification method of basalt fiber. Synth. Fiber Industry. 2015;38:75-77.
[Google Scholar]
Liao N.W., . Effects of lean manufacturing aggregate characteristics on performance of asphalt mixtures. Chongqing, China: Chongqing Jiaotong University; 2018. Master’s Thesis
Liu X., Wang Z., Ma X.D., . Distinguishing crude oils from heavy fuel oils by polycyclic aromatic hydrocarbon fingerprints. Environ. Forensics. 2013;14:20-24.
[Google Scholar]
Margaritism A., Soenen H., Fransen E., . Identification of ageing state clusters of reclaimed asphalt binders using principal component analysis (PCA) and hierarchical cluster analysis (HCA) based on chemo-rheological parameters. Constr. Build. Mater.. 2020;244:118276
[Google Scholar]
Ren R., Han K., Zhao P., . Identification of asphalt fingerprints based on ATR-FTIR spectroscopy and principal component-linear discriminant analysis. Constr. Build. Mater.. 2019;198:662-668.
[Google Scholar]
Ren, Q.G. Study on the influence of fine aggregate of different lithologies on road performance of asphalt mixture. SUBGRADE ENGINEERING 2020, 48, 65-68+78.
Wang S., . Review of application of principal component analysis in comprehensive evaluation of academic journals in China. Inf. Res.. 2014;10:30-32.
[Google Scholar]
Wang F.F., . Test detection and quality control of cement concrete raw materials for Highway Engineering. Sci. Technol.. 2019;16:203-204.
[Google Scholar]
Wang K., Yuan Y., Han S., . Application of attenuated total reflectance Fourier transform infrared (ATR-FTIR) and principal component analysis (PCA) for quick identifying of the bitumen produced by different manufacturers. Road Mater. Pavement Des.. 2018;19
[Google Scholar]
Wang, X.D. Study on the influence of aggregate lithology and microscopic characteristics on road performance of asphalt mixture. Master’s Thesis, Jilin University, Changchun, China, 2014.
Xie Z.X., Shen J.N., Fan W.Z., . Laboratory investigation of the effect of warm mix asphalt (WMA) additives on the properties of WMA used in China. J. Test. Eval.. 2014;42(5):1165-1172.
[Google Scholar]
Xu M., Zhang Y.Z., Zhao P.H., . Study on aging behavior and prediction of SBS modified asphalt with various contents based on PCA and PLS analysis. Constr. Build. Mater.. 2020;265:120732
[Google Scholar]
Yang N., Kuang S.Y., Yue Y.H., . Analysis of infrared absorption spectrum characteristics of several common anhydrous carbonate minerals. Miner. Petrol.. 2015;35:37-42.
[Google Scholar]
Yu X., Chen L.H., . Composition changes and genesis of basalt in back arc basin. Acta Petrol. Sin.. 2020;36:1953-1972.
[Google Scholar]
Zhang Y.L., Li X.Y., . XRD and IR characteristics of feldspar in Huangling granitic body and their significance. Acta Sci. Nat. Univ. Sunyatseni. 2015;61:46-56.
[Google Scholar]

Introduction

Materials and methods

Materials
Test methods
Analysis method

Results and discussions

FTIR
Analyses and discussions -----PCA
Cluster analysis
Linear discriminant analysis

Validations of the method in engineering project

Summaries and conclusions

Fulltext Views
1,327

PDF downloads
1,163

View/Download PDF
Download Citations

BibTeX
RIS

Show Sections

[1] Cheng Z.Q., Kong F.S., . Identification and analysis of asphalt of different brands based on infrared spectroscopy and stoichiometry. J. Wuhan Univ. Technol.. 2020;44:909-913.
[Google Scholar]

[2] Cui Y.J., Li B., Wang J., . Maintenance section division of expressway asphalt pavement based on principal component clustering method. Highway. 2019;64:290-296.
[Google Scholar]

[3] Dai Z., Shen J.N., Shi P.C., . Nano-sized morphology of asphalt components separated from weathered asphalt binders. Constr. Build. Mater.. 2018;128:588-596.
[Google Scholar]

[4] Dong H.S., . A comparative study of principal component analysis and linear discriminant analysis for dimensionality reduction. Modern Computer. 2016;29:36-40.
[Google Scholar]

[5] He S.J., Wang C.Y., Li Y., . Evaluation of diagnostic ratios of medium and serious weathered oils from five different oil sources. Acta Oceanol. Sin.. 2016;35:1-8.
[Google Scholar]

[6] Jia L., . Study on the effect of aggregate mineral composition on road performance of asphalt mixture. Xi’an, China: Chang 'an university; 2012. Master’s Thesis

[7] Kuhnen S., Ogliari J.B., Dias P.F., . ATR-FTIR spectroscopy and chemometric analysis applied to discrimination of landrace maize flours produced in southern Brazil. Int. J. Food Sci. Technol.. 2010;45:1673-1681.
[Google Scholar]

[8] Li J., Cai T., Wang N., . A qualitative identification method of basalt fiber. Synth. Fiber Industry. 2015;38:75-77.
[Google Scholar]

[9] Liao N.W., . Effects of lean manufacturing aggregate characteristics on performance of asphalt mixtures. Chongqing, China: Chongqing Jiaotong University; 2018. Master’s Thesis

[10] Liu X., Wang Z., Ma X.D., . Distinguishing crude oils from heavy fuel oils by polycyclic aromatic hydrocarbon fingerprints. Environ. Forensics. 2013;14:20-24.
[Google Scholar]

[11] Margaritism A., Soenen H., Fransen E., . Identification of ageing state clusters of reclaimed asphalt binders using principal component analysis (PCA) and hierarchical cluster analysis (HCA) based on chemo-rheological parameters. Constr. Build. Mater.. 2020;244:118276
[Google Scholar]

[12] Ren R., Han K., Zhao P., . Identification of asphalt fingerprints based on ATR-FTIR spectroscopy and principal component-linear discriminant analysis. Constr. Build. Mater.. 2019;198:662-668.
[Google Scholar]

[13] Ren, Q.G. Study on the influence of fine aggregate of different lithologies on road performance of asphalt mixture. SUBGRADE ENGINEERING 2020, 48, 65-68+78.

[14] Wang S., . Review of application of principal component analysis in comprehensive evaluation of academic journals in China. Inf. Res.. 2014;10:30-32.
[Google Scholar]

[15] Wang F.F., . Test detection and quality control of cement concrete raw materials for Highway Engineering. Sci. Technol.. 2019;16:203-204.
[Google Scholar]

[16] Wang K., Yuan Y., Han S., . Application of attenuated total reflectance Fourier transform infrared (ATR-FTIR) and principal component analysis (PCA) for quick identifying of the bitumen produced by different manufacturers. Road Mater. Pavement Des.. 2018;19
[Google Scholar]

[17] Wang, X.D. Study on the influence of aggregate lithology and microscopic characteristics on road performance of asphalt mixture. Master’s Thesis, Jilin University, Changchun, China, 2014.

[18] Xie Z.X., Shen J.N., Fan W.Z., . Laboratory investigation of the effect of warm mix asphalt (WMA) additives on the properties of WMA used in China. J. Test. Eval.. 2014;42(5):1165-1172.
[Google Scholar]

[19] Xu M., Zhang Y.Z., Zhao P.H., . Study on aging behavior and prediction of SBS modified asphalt with various contents based on PCA and PLS analysis. Constr. Build. Mater.. 2020;265:120732
[Google Scholar]

[20] Yang N., Kuang S.Y., Yue Y.H., . Analysis of infrared absorption spectrum characteristics of several common anhydrous carbonate minerals. Miner. Petrol.. 2015;35:37-42.
[Google Scholar]

[21] Yu X., Chen L.H., . Composition changes and genesis of basalt in back arc basin. Acta Petrol. Sin.. 2020;36:1953-1972.
[Google Scholar]

[22] Zhang Y.L., Li X.Y., . XRD and IR characteristics of feldspar in Huangling granitic body and their significance. Acta Sci. Nat. Univ. Sunyatseni. 2015;61:46-56.
[Google Scholar]

Aggregate fingerprints identification based on its compositions and machine learning algorithm

Abstract

Keywords

Aggregates

Fingerprint identification

FTIR

Machine learning

1 Introduction

2 Materials and methods

2.1 Materials

2.2 Test methods

2.3 Analysis method

3 Results and discussions

3.1 FTIR

3.2 Analyses and discussions -----PCA

3.2.1 Granite

3.2.2 Limestone

3.2.3 Basalt

3.3 Cluster analysis

3.4 Linear discriminant analysis

4 Validations of the method in engineering project

5 Summaries and conclusions

Declaration of Competing Interest

References

Suggested read for related articles:

We value your privacy