Characterization of Synthetic Non-Metallic Inclusions Consisting of TiN, Ti₂O₃, and Oxides of Al₂O₃ and MgO·Al₂O₃ Spinel Using Raman Spectroscopy

Abstract

Steel quality and properties can be affected by the formation of complex inclusions, including Ti-based inclusions such as TiN and Ti₂O₃ and oxides like Al₂O₃ and MgO·Al₂O₃ (MA). This study assessed the prospective use of Raman spectroscopy to characterize synthetic binary inclusion samples of TiN–Al₂O₃, TiN–MA, Ti₂O₃–MA, and Ti₂O₃–Al₂O₃ with varying phase fractions. The relative intensities of the Raman peaks were used for qualitative evaluation and linear regression calibration models were used for the quantitative prediction of individual phases. The model performance was evaluated with root mean square error of cross-validation (RMSECV) and root mean square error of prediction (RMSEP). For the raw Raman spectra data, R² values were between 0.48–0.98, the RMSECV values varied between 3.26–14.60 wt%, and the RMSEP ranged between 2.98–15.01 wt% for estimating the phases. The SNV Raman spectra data had estimated R² values within 0.94–0.99 and RMSECV and RMSEP values ranged between 2.50–3.26 wt% and 2.80–9.01 wt%, respectively, showing improved model performance. The study shows that the specific phases of TiN, Al₂O₃, MA, and Ti₂O₃ in synthetic inclusion mixtures of TiN–(Al₂O₃ or MA) and Ti₂O₃–(Al₂O₃ or MA) could be characterized by the Raman spectroscopy.

1. Introduction

The addition of titanium to steel plays several essential roles, such as acting as a deoxidizing agent or binding nitrogen to prevent BN formation in boron-alloyed steels [1]. In addition, titanium may also affect the mechanical properties of the final product [2,3]. However, earlier investigations have demonstrated the formation of Ti-based inclusions, such as TiN and Ti₂O₃, in steels and their potentially detrimental effect on the steel properties [4,5]. Furthermore, numerous studies [6,7,8] have shown that the presence of TiN or Ti₂O₃ inclusions generated during steelmaking or in final steel products could occur as a mixture with other detrimental oxide inclusion phases. For instance, studies [6,7,8] further suggested that the addition of Ti and the potential generation of TiN precipitates could lead to the possible formation of a complex inclusion of Al₂O₃-TiN [6,8] or TiN-MgO·Al₂O₃ spinel [7] during steel solidification and, thus, in the final steel products. Furthermore, inclusion phase mixtures, such as Ti₂O₃ with Al₂O₃ or MgO·Al₂O₃, could be generated during steel solidification depending on the composition of the steel melt. These binary or complex inclusion mixtures complicate the ability to achieve clean steel; therefore, it is essential to characterize them.

The application of various analytical techniques, such as scanning electronic microscope with energy dispersive spectroscopy (SEM-EDS) and others classified by Zhang et al. [9], has contributed immensely to characterizing inclusions commonly found in steel. However, the general expectations concerning the analytical technique used for identification and analysis of inclusions are to that it has features such as a relatively short duration of sample preparation, improved accuracy, and reliable analysis results, as these are of interest to researchers and steelmakers. Raman spectroscopy has been demonstrated to be a robust method and has most of these features. Consequently, Raman spectroscopy is a prospective analytical tool for inclusion characterization compared to other methods, such as SEM-EDS, in terms of sample preparation and analysis duration.

Previous studies have demonstrated the prospective application of Raman characterizing oxide mixtures from Al₂O₃, MgO·Al₂O₃ spinel (MA), and calcium aluminates from synthetic inclusions phase [10]. In addition, the study conducted by S. Li and Hihara (2017) [11] has shown that individual phases of calcium-aluminate-based inclusions, such as CaO∙Al₂O₃ (CA), in steel samples could be identified in steel samples using Raman spectroscopy. Furthermore, Matsuura et al. [12] and Pan et al. [13] have shown the possibility for Ti-based inclusion occurring as a mixture with oxides in steel samples, while TiN and Ti₂O₃ interactions with slag towards steel cleanliness was demonstrated by Michelic et al. [14]. However, fewer studies have been conducted concerning the use of Raman spectroscopy for detailed characterization for a mixture of phases, such as TiN and Ti₂O₃ with other oxide inclusions, like MgO·Al₂O₃ spinel and Al₂O₃, and the use of a calibration model to quantify the relative content of the phases present. For instance, studies conducted by [15,16] demonstrated the use of Raman spectroscopy to mainly identify Ti-based inclusions phases such as TiN, Ti₂O₃, and other phases from Ti-deoxidation process-based products [15] and under different experimental conditions [16], but not for predicting the individual phase content using calibration model as shown in this work. Consequently, there is a need to expand the scope concerning the potential use of Raman spectroscopy for inclusion studies since inclusions generated in steels could occur as a mixture of phases, as illustrated by previous research. Consequently, the purpose of this work was to demonstrate the prospect of the application of Raman spectroscopy as an analytical technique to identify and characterize synthetic inclusion mixtures consisting of titanium nitride (TiN), aluminum oxide (Al₂O₃), magnesium–aluminate spinel (MgO·Al₂O₃, MA), and Ti₂O₃. Here, binary phases of TiN–Al₂O₃, TiN–MA, Ti₂O₃–MA, and Ti₂O₃–Al₂O₃ were investigated.

2. Materials and Methods

The synthetic binary phase samples prepared for this study were TiN–Al₂O₃, TiN–MA, Ti₂O₃–MA, and Ti₂O₃–Al₂O₃. The chemical powders used for this work were titanium nitride (TiN), magnesium aluminate spinel MgO·Al₂O₃ (MA), titanium (III) oxide (Ti₂O₃), and aluminum oxide (Al₂O₃). These chemicals were obtained from commercial suppliers; Sigma-Aldrich (Merck KGaA, Darmstadt, Germany) and Alfa Aesar (Erlenbachweg 2 76870, Kandel Germany), with purity in the range of 99.7% to 99.9%. The required proportions (weight percentage) of the phases (TiN, MgO·Al₂O₃, Al₂O₃, and Ti₂O₃) were measured by using an analytical balance. The weighed samples were milled to achieve a uniform size and sample homogeneity. The milling further ensured that the samples inside each spectrum acquired during the Raman spectroscopy measurements would represent the phases present in the binary sample prepared. For synthetic samples containing TiN–(Al₂O₃ or MA), the phase fraction of TiN was varied between 10 wt% and 90 wt% for each binary phase sample (TiN–Al₂O₃, TiN–MA). Similarly, the phase fraction of TiN for a sample matrix consisting of Ti₂O₃–(MA or Al₂O₃) was varied approximately from 10 wt% to 90 wt% to generate binary phase mixtures of Ti₂O₃–MA and Ti₂O₃–Al₂O₃.

Thorough and careful mixing was performed several times to ensure that the prepared synthetic binary phase samples were homogenous. The confirmation of the prepared samples’ homogeneity was carried by using X-ray fluorescence (XRF) and X-ray diffraction (XRD) analysis techniques based on measuring replicates of the same synthetic binary phase sample. Additionally, XRF and XRD were used as complementary analytical techniques to evaluate and verify the phases present in the samples. The chemical and phase composition of the starting chemicals (TiN, Al₂O₃, Ti₂O₃, and MgO·Al₂O₃) were evaluated by using XRD and XRF. Details concerning setup parameters for XRD and XRF instruments used for this study were explained in our previous work [10].

Table 1. Prepared sample composition and XRF analyses obtained from binary phases of Ti₂O₃–Al₂O₃ and Ti₂O₃–MA.

Initial Phase Composition (wt%)		Phases Calculated Based on XRF Analysis (wt%)		Initial Phase Composition (wt%)		Phases Calculated Based on XRF Analysis (wt%)
Ti2O3	Al2O3	Ti2O3	Al2O3	MgO·Al2O3	Ti2O3	MgO·Al2O3	Ti2O3
90	10	90.3	9.7	10	90	9.0	91.0
80	20	81.6	18.4	20	80	18.0	82.0
70	30	74.5	25.5	30	70	32.1	67.9
60	40	58.0	42.0	40	60	39.0	61.0
55	45	52.0	48.0	45	55	47.5	52.5
50	50	48.5	51.5	50	50	51.0	49.0
40	60	35.9	64.1	60	40	73.7	26.3
30	70	28.0	72.0	70	30	65.5	34.5
20	80	21.0	79.0	80	20	82.3	17.7
10	90	7.0	93.0	90	10	91.0	9.0

and

Table 2. Prepared sample composition and XRF analyses obtained from binary phases of TiN–Al₂O₃ and TiN–MA.

Initial Phase Composition (wt%)		Phases Calculated Based on XRF Analysis (wt%)		Initial Phase Composition (wt%)		Phases Calculated Based on XRF Analysis (wt%)
TiN	Al2O3	TiN	Al2O3	MgO·Al2O3	TiN	MgO·Al2O3	TiN
10	90	9.0	91.0	10	90	8.0	92.0
20	80	16.0	84.0	20	80	22.0	78.0
30	70	28.8	71.2	30	70	28.8	71.2
40	60	36.8	63.2	40	60	36.8	63.2
50	50	46.7	53.3	45	55	46.7	53.3
60	40	53.0	47.0	50	50	52.0	48.0
70	30	56.7	43.3	60	40	56.7	43.3
80	20	75.3	24.7	70	30	75.3	24.7
90	10	83.9	16.1	80	20	83.9	16.1
				90	10	91.2	8.8

illustrates a comparison between the weight percentages of the phases (TiN, Al₂O₃, MA, and Ti₂O₃) measured by using analytical balance for generating the sample mixtures and specific phases calculated based on the XRF elemental analysis.

2.1. Raman Spectroscopy

The Raman spectroscopy used for this study was TimeGated^® Raman spectrometer (TG532 M1) from Timegate instruments limited, Oulu, Finland. Details concerning the features for the Raman spectroscopy used to conduct the measurements have been presented in previous studies by the authors of this study [10]. The same parameter settings associated with the Raman spectroscopy, such as a resolution of 10 cm⁻¹ laser power (545) and duration for spectra acquisition, were used during the prepared sample measurements. The Raman spectra obtained from the Raman spectroscopy measurements were within the Raman shift region between 100–1100 cm⁻¹ and under ambient conditions. During the Raman measurements, some procedures were used to achieve relatively enhanced repeatability and decrease the challenges associated with sample inhomogeneity. These included total spectra with at least three replicates collected at different spots for each sample, and the spectra acquisition duration was between 1–3 min.

Additionally, the Raman spectroscopy setup was equipped with a rotating sample holder stage, making it possible to obtain an average Raman spectrum at the different spots of each batch of the sample measured. The pre-processing software (from Timegate instrument Limited) was used for the initial Raman spectra data treatment, such as basic background subtraction. Other statistical software programs, such as OriginPro, 2020 and an unscrambler (version 11.0, by Camo Software, Oslo, Norway), were used for subsequent Raman spectra pre-processing, peak fitting, and multivariate analysis. A Gaussian peak shape was used for the Raman peak fitting since it was generally demonstrated to be suitable for the solid samples prepared and measured with the Raman spectroscopy. Further explanation concerning the selection of this peak shape can be found in [17].

2.2. Calibration Model

A calibration model and multivariate analysis were used to establish a relationship between the specific phase content present in the samples and the identified Raman peak ratios. The variation in the phase content (wt%) present in the sample was evaluated by using the summed Raman spectrum to explain the relative integrated areas of the identified peaks present in the Raman spectrum. The calibration candidate is expressed as:

$${\mathit{x}}_{\mathit{c}}=\frac{{\mathit{A}}_{\mathit{m}}}{({\mathit{A}}_{\mathit{n}}+{\mathit{A}}_{\mathit{m}})},$$

(1)

where ${\mathit{x}}_{\mathit{c}}$ is the calibration feature candidate, ${\mathit{A}}_{\mathit{m}}$ is the integrated area for the Raman peak n, and ${\mathit{A}}_{\mathit{n}}$ is the area corresponding to the Raman peak n.

Evaluation of the model performance for each calibration feature candidate was carried out using partial least squares (PLS) regression for this study [18]. The PLS regression method was first carried out with the raw Raman spectral data from the Raman measurement and, subsequently, on pre-processed Raman spectra data by using the standard normal variate (SNV) technique. Studies have shown that the application of SNV to raw spectral data provides some benefits for subsequent analysis of the data [19]. These include minimizing the potential baseline shifting and reducing some possible sources of non-linearity linked with the phase content of the analyses associated with the measured Raman spectra [19]. This can help evaluate the performance of a PLS regression method when the raw Raman spectra data are used, comparing these to the pre-processed Raman spectra data measured from the multiphase samples.

In performing the PLS analysis, the obtained Raman spectra data were randomly split into calibration (training) and the prediction (external validation) sets. Approximately 75% of the samples were used as the calibration set to obtain the quantification model. The remaining randomly split samples of 25% were used as the prediction set for testing the model. In this study, Raman spectra from 100 to 900 cm⁻¹ were identified as being suitable to use for the PLS model since most of the dominant Raman peaks for the phases used in preparing the samples were within this region. The accuracy of the model was carried out using the cross-validation process with the leave-one-out cross-validation technique. More details concerning the PLS analysis procedure for this study can be found in [20,21].

In evaluating the performance of the calibration model used for estimating the phase content in the prepared samples, three performance indicators were used for the calibration set [21,22]. These were the coefficient of determination (R²), the root mean standard error in calibration (RMSEC), and the root mean standard error in cross-validation (RMSECV). Similarly, the coefficient of determination and root mean square error of prediction (RMSEP) were used to assess the external validation (prediction) or test set performance. According to Kachrimanis et al. [22], the root mean standard error (RMSE) can be given as:

$$\mathrm{RMSE}=\sqrt{\frac{{{\displaystyle \sum }}_{\mathit{i}=1}^{\mathit{n}}{\left({\mathit{y}}_{\mathit{i}}+{\hat{\mathbf{y}}}_{\mathit{i}}\right)}^2}{\mathit{n}}}$$

(2)

where ${\mathit{y}}_{\mathit{i}}$ represents the reference value, and the number of samples marked as $\mathit{n}$ and $\hat{\mathbf{y}}$ denote the calculated value.

3. Results

3.1. Raman Spectra Analysis Based on Phase Content

The identifiable Raman bands (peaks) assigned to each phase (TiN, MgO·Al₂O₃, Al₂O₃, and Ti₂O₃) were used to prepare the synthetic inclusion samples and some referenced studies are presented in

Table 3. Measurement Raman peaks (bands) for the synthetic phase used for this study (where s is strong, m is medium, and w is weak).

Phase	Measured Peaks [Raman Shift, cm−1]	Reference Peaks [Raman Shift, cm−1]	Reference
TiN	157 s, 260 m, 620 m	152 s, 340 m, 410 m, 620 m,	[23]
Al2O3	422 s, 383 m, 760 m,	413–420 s, 375–380 m, 758 m,	[24]
MgO·Al2O3	416 s, 674 m, 773 m	409–412 s, 666–674 m, 767–772 m	[25]
Ti2O3	247 s, 335 m,	247 s, 336 m, 423 w	[26]

. Figure 1 shows the Raman spectra for synthetic inclusion phases of TiN, MgO·Al₂O₃, Al₂O₃, and Ti₂O₃ used to prepare the binary samples consisting of TiN–Al₂O₃, TiN–MA, Ti₂O₃–MA, and Ti₂O₃–Al₂O₃. The Raman spectra for starting phases are arranged from the top in Figure 1 as follows: spectrum for TiN (Figure 1i), spectrum for MgO·Al₂O₃ (Figure 1ii), spectrum for Al₂O₃ (Figure 1iii), and spectrum for Ti₂O₃ (Figure 1iv).

3.2. Qualitative Analysis of Phase Content Based on Raman Spectra

Figure 2 and Figure 3 show some examples from the relationship between the change in the phase content in the samples measured using Raman spectroscopy and the observable effects on the Raman peak intensities. This phenomenon, where a variation in the phase content in the binary samples has a corresponding impact on relative Raman peak intensities, provides an opportunity to quickly conduct qualitative analysis for the phases present in the samples.

3.2.1. TiN–Al₂O₃ and TiN– MgO·Al₂O₃ (MA)

Figure 2 consists of a sample’s matrix with a varying phase fraction of TiN, Al₂O₃, and MA. Figure 2a contains selected binary samples for TiN–MA with phase fractions for MA varying between 20, 40, and 80 wt% in each sample. Similarly, Figure 2b shows examples of TiN–Al₂O₃ with the phase content for Al₂O₃ varying between 20, 40, and 80 wt%. It is observable from Figure 2a that variation (increase or decrease) in MA content for samples containing TiN–MA had a responding effect on the Raman peak intensity at 773 cm⁻¹. Based on the Raman spectra observation presented in Figure 2b, a Raman peak at 157 cm⁻¹ is attributable to TiN changes in relative peak intensity as the phase content (wt%) in TiN–Al₂O₃ samples varied. Additionally, for the TiN–Al₂O₃ samples, an increase in the phase fraction of Al₂O₃ exhibited an increase of a Raman peak shoulder at the region of 383 cm⁻¹. Furthermore, it was noticed that Al₂O₃ and MA phases had the most intense Raman peak (422 cm⁻¹ for Al₂O₃ and 416 cm⁻¹ for MA), found within a similar region as some Raman peaks associated with TiN (between 375–450 cm⁻¹), with a high potential for peak overlap. Consequently, the medium Raman peaks could be used to relate the specific phase content in the sample as a function of its relative intensity.

3.2.2. Ti₂O₃–Al₂O₃, Ti₂O₃–MgO·Al₂O₃ (MA)

Figure 3a illustrates Raman spectra for binary samples containing a mixture of Ti₂O₃ and MA, with the phase content varying between 20, 60, and 80 wt%. It was observed that a rise in the MA phase content (wt%) in the Ti₂O₃–MA samples also led to an increase in the Raman peak intensity within the region of 410–420 cm⁻¹. Figure 3b presents examples of Ti₂O₃–Al₂O₃ prepared with the phase fraction for Al₂O₃ varying between 20, 40, and 80 wt%. It was noticed from the Raman spectra presented in Figure 3b that with an incremental increase in the phase content of Al₂O₃ in the binary samples (Ti₂O₃–Al₂O₃), the Raman peak intensity associated with Al₂O₃ at 382 and 422 cm⁻¹ rose relatively to correspond to the increasing content of Al₂O₃. Similarly, in Figure 3a,b, a change in relative Raman peak intensity at approximately 247 cm⁻¹ due to a variation in the phase fraction of Ti₂O₃ in Ti₂O₃–Al₂O₃ and Ti₂O₃–MgO·Al₂O₃ samples was observed. Therefore, for synthetic inclusion phases containing Ti₂O₃, Al₂O₃, and MgO·Al₂O₃, the relative intense Raman peaks associated with the specific phases could be used for qualitative estimation.

3.3. Quantitative Estimation of the Phase Fractions in the Samples

For quantitative estimation of the individual phases in the binary samples of TiN–Al₂O₃, TiN–MgO·Al₂O₃, Ti₂O₃–Al₂O₃, and Ti₂O₃–MgO·Al₂O₃, a linear calibration model was utilized. The calibration model was used to establish the relative Raman peak area and phase content in the samples. The relative Raman peak areas were estimated using Equation 1 for each of the peaks attributable to specific phases in the Raman spectra measured, and phase content based on XRF analysis was used as the dependent variable. Raman peaks at approximately 157, 260, and 620 cm⁻¹ were used for TiN; 247 and 335 cm⁻¹ peaks were associated with Ti₂O₃; 382, 422, and 760 cm⁻¹ peaks were used for Al₂O₃; and 416, 674, and 773 cm⁻¹ peaks were used for MgO·Al₂O₃ for relative Raman peak area calculations and analyses. The Raman peaks presented for calculations and analysis for TiN, Ti₂O₃, MA, and Al₂O₃ are listed in

Table 4. Estimation of the coefficient of determination (R²) between the relative Raman peak area and phase content from XRF analysis for (a) TiN-Al₂O₃, (b) Ti₂O₃-Al₂O₃, (c) TiN-MgO·Al₂O₃, and (d) Ti₂O₃-MgO·Al₂O₃.

System	Relative Intensity	R2	System	Relative Intensity	R2
(i) TiN-Al2O3	382/157	0.99	(iii) TiN-MgO·Al2O3	674/157	0.91
	382/260	0.74		674/260	0.97
	422/157	0.94		416/157	0.98
	422/260	0.72		422/260	0.96
	760/157	0.98		773/157	0.98
	760/335	0.87		773/335	0.95
	382/620	0.88		674/620	0.93
	760/620	0.95		773/620	0.97
(ii) Ti2O3-Al2O3	382/247	0.86	(iv) Ti2O3-MgO·Al2O3	416/247	0.98
	382/335	0.95		416/335	0.95
	422/247	0.97		674/247	0.81
	422/335	0.93		674/335	0.89
	760/247	0.90		774/247	0.79
	760/335	0.95		774/335	0.87

. These Raman peak areas were used to estimate a linear regression between the phase content in the sample and the relative area Raman peak. The range of coefficient of determination (R²) values obtained from the linear regression model by using relative integrated Raman peak area, and phase content in the samples were 0.72–0.99 for (i) TiN–Al₂O₃, 0.86–0.97 for (ii) Ti₂O₃–Al₂O₃, 0.93–0.98 for (iii) Ti₂O₃–MgO·Al₂O₃, and (iv) 0.79–0.98 for Ti₂O₃–MgO·Al₂O₃, which are presented in Table 4.

Figure 4 shows the best possible Raman peaks ratios for estimating the phase contents based on the relative integrated Raman peak area for MgO·Al₂O₃ (MA) as a function of the phase content of MA in Ti₂O₃–MgO·Al₂O₃ (Figure 4a) and TiN–MgO·Al₂O₃ (Figure 4b) samples. In addition, Figure 5 illustrates the best possible Raman peaks ratios for estimating the phase contents based on the relative integrated Raman peak area for Al₂O₃ (wt%) as a function of the phase content of Al₂O₃ in Ti₂O₃–Al₂O₃ (Figure 5a) and TiN–Al₂O₃ (Figure 5b) samples. Further estimation of the individual phases in the samples prepared was performed with a PLS regression model to establish the relationship between the Raman signals and the phase concentration (wt%) in the samples prepared.

3.4. Further Phase Estimation Assessment

The PLS regression model was first applied to the raw Raman spectral data and pre-processed Raman spectra to evaluate the phases present in the binary samples.

Table 5. Estimated figures of merit for calibration and validation using the PLS models for (i) TiN–Al₂O₃ and (ii) TiN–MgO·Al₂O₃.

System	Phase	Method	R2CAL	R2CV	R2PRED.	RMSEC, (wt%)	RMSECV, (wt%)	RMSEP, (wt%)
(i) TiN–Al2O3	TiN	Raw Data	0.86	0.79	0.58	11.56	15.14	15.01
		SNV Data	0.95	0.94	0.84	6.68	7.79	9.08
	Al2O3	Raw Data	0.86	0.79	0.58	11.56	15.14	15.11
		SNV Data	0.96	0.95	0.94	6.11	7.71	9.07
(ii) TiN–MgO·Al2O3	TiN	Raw Data	0.61	0.48	0.61	14.63	18.10	14.63
		SNV Data	0.99	0.99	0.98	2.60	3.10	3.01
	MgO·Al2O3	Raw Data	0.86	0.79	0.81	11.83	13.99	14.14
		SNV Data	0.98	0.99	0.98	2.94	2.97	3.12

and

Table 6. Estimated figures of merit for calibration and validation using the PLS models for (i) Ti₂O₃–Al₂O₃ and (ii) Ti₂O₃–MgO·Al₂O₃.

System	Phase	Method	R2CAL	R2CV	R2PRED.	RMSEC, (wt%)	RMSECV, (wt%)	RMSEP, (wt%)
(i) Ti2O3–Al2O3	Ti2O3	Raw Data	0.98	0.98	0.98	3.26	3.57	2.98
		SNV Data	0.99	0.98	0.98	3.26	3.64	2.98
	Al2O3	Raw Data	0.99	0.99	0.98	3.11	3.68	2.94
		SNV Data	0.98	0.99	0.98	3.35	3.70	2.95
(ii) Ti2O3–MgO·Al2O3	Ti2O3	Raw Data	0.94	0.97	0.95	5.14	6.09	5.75
		SNV Data	0.98	0.98	0.97	4.81	3.22	3.02
	MgO·Al2O3	Raw Data	0.93	0.96	0.94	5.23	6.46	7.17
		SNV Data	0.98	0.97	0.97	4.81	5.28	4.79

present the statistical parameters and evaluated values obtained from the coefficient of determination for calibration (R²_CAL), validation (R²_CV), prediction (R²_PRED), the root mean square of calibration (RMSEC), the root mean square of cross-validation (RMSECV), and the root mean square error of prediction (RMSEP). Table 5 illustrates the model performance values from a PLS regression analysis to evaluate the phases (TiN, MA, and Al₂O₃) used to prepare TiN–MA and TiN–Al₂O₃ samples. The raw Raman data used to evaluate the MA phase content in TiN–MA and Ti₂O₃–MA had the following range of values for the figure of merits: R²_CAL (0.61–0.94), R²_CV (0.48–0.97), R²_PRED (0.61–0.95), RMSEC (5.0–14.7 wt%), RMSECV (6.0–18.1 wt%), and RMSEP (5–14.7 wt%). For pre-processed (SNV) Raman data obtained from the samples, MA phase content for R²_CAL, R²_CV, and R²_PRED values ranged between 0.97–0.99 and the RMSEC, RMSECV, and RMSEP ranged between 2.5–5.5 wt%. Similarly, for the raw Raman data, Al₂O₃ phase fraction analysis in TiN–Al₂O₃ and Ti₂O₃–Al₂O₃ had estimated figures of merit values varying between R²_CAL (0.86–0.99), R²_CV (0.79–0.99), R²_PRED (0.58–0.98), RMSEC (3.0–11.60 wt%), RMSECV (3–15.5 wt%), and RMSEP (2.9–15.11 wt%). Using SNV Raman data for the analysis, Al₂O₃ phase content estimation also had R²_CAL, R²_CV, and R²_PRED values varying between 0.94–0.98 and 2.98–9.20 wt% for RMSEC, RMSECV, and RMSEP values.

In estimating the performance of TiN contained in TiN–MA and TiN–Al₂O₃, the coefficient of determination for the calibration (R²_CAL), cross-validation (R²_CV), and prediction (R²_PRED) values ranged between 0.48–0.86 wt% and those of RMSEC, RMSECV, and RMSEP were between 11–18.10 wt% for the raw Raman spectra data. Additionally, Raman spectra data pre-processed using SNV provided estimated values for R²_CAL, R²_CV, and R²_PRED varied between 0.84–0.99 and values for RMSEC, RMSECV, and RMSEP were between 2.60–9.07 wt% based on the model for evaluating the phase content of TiN binary samples TiN–(Al₂O₃, MA), as presented in Table 5. A similar trend could be observed for the assessment parameter in predicting the phase content for Ti₂O₃ in binary samples prepared from Ti₂O₃–(Al₂O₃, MA), as presented in Table 6. The PLS model for the raw Raman data had R²_CAL, R²_CV, and R²_PRED values ranging between 0.93–0.98, with RMSEC, RMSECV, and RMSEP values varying between 2.98–6.50 wt% for Ti₂O₃ analyses in these samples. Estimating the individual phases by using SNV, the coefficient of determination (R²_CAL, R²_CV, and R²_PRED) values varied between 0.97–0.99 and RMSEC, RMSECV, and RMSEP values were between 2.95–5.30 wt% for estimation of the Ti₂O₃ phases present in the samples for this study.

4. Discussion

4.1. Observation of Specific Raman Peaks Relative to the Change in Phase Content

As demonstrated in Figure 2 and Figure 3, the Raman spectra can be used to qualitatively predict the phase content present in the binary samples based on the relative intensity of the Raman peaks associated with a specific phase. For TiN–Al₂O₃ and TiN–MA, medium Raman peaks at 157 cm⁻¹ for TiN, at 383 cm⁻¹ for Al₂O₃, and approximately 773 cm⁻¹ for MA were used to explain the phenomenon of a change in phase content having a corresponding effect on the relative Raman peaks. In Figure 3, observation based on variation in the phase fraction and the related impact on the relative Raman peak was examined for binary samples of Ti₂O₃–MgO·Al₂O₃, and Ti₂O₃–Al₂O₃. In these binary samples, and illustrated in Figure 3, Raman peaks at 247 and 335 cm⁻¹ associated with Ti₂O₃, at 416 and 773 cm⁻¹ attributable to MA, and 383, 422, and 760 cm⁻¹ associated with Al₂O₃ can be used in predicting the most dominating phase present in the sample mixtures. For sample mixtures of TiN–Al₂O₃ and TiN–MA, as presented in Figure 2, the phases (TiN, Al₂O₃, and MA) had their most intense peaks quite close to each other, within the Raman band region between 410–460 cm⁻¹. Therefore, due to the effect of overlap of Raman peaks, the medium peaks were used for qualitative analysis. However, with the binary samples of Ti₂O₃–MgO·Al₂O₃ and Ti₂O₃–Al₂O₃, the most intense Raman peaks were used to observe the effect of a change in phase content and the relative Raman peak intensity since the potential peak overlap was not a challenge.

The estimation for MA phase content analysis in Ti₂O₃–MgO·Al₂O₃ and TiN–MgO·Al₂O₃ based on the relative Raman peak area is shown in Table 4 (i and ii). The most suitable Raman peak area ratio with the best coefficient of determination value (R² = 0.98) for Ti₂O₃–MgO·Al₂O₃ was obtained from 416 cm⁻¹ for MA and 247 cm⁻¹ for Ti₂O₃ from a linear regression between the phase content and the relative Raman peak area. Similarly, to estimate the MA phase content for TiN–MgO·Al₂O₃ samples, Table 4 (ii) shows that the relative Raman peak area ratios of 416/157 and 773/157 (157 cm⁻¹ for TiN and 416 773 cm⁻¹ for MA) had an excellent coefficient of determination value of 0.98 linear regression.

For binary sample systems of Ti₂O₃–Al₂O₃, the linear regression plotted between the phase content and the relative Raman peak area, the phase content for Al₂O₃ was estimated as shown in Table 4 (ii), with the identified Raman peak ratio at 422 cm⁻¹ for Al₂O₃ and 247 cm⁻¹ for Al₂O₃ having an R² value of 0.97 as the best. An estimation for Al₂O₃ content in Ti₂O₃–Al₂O₃ also showed the best candidate peaks based on linear regression analysis to be 382, 760 cm⁻¹ for Al₂O₃ at approximately 157 cm⁻¹ for TiN, with coefficient of determination values ranging between 0.98–0.99. Thus, Table 4 demonstrates Raman peak areas at approximately 157 cm⁻¹ attributable to TiN, 382, 422, and 760 cm⁻¹ associated with Al₂O₃ and at 416 and 773 cm⁻¹ for MA as excellent indicators for the specific phases in the samples. Therefore, it can be concluded that integrated relative Raman peak areas as a function of the phase content present was a suitable method for the quantitative analysis in this study.

4.2. Further Phase Identification Evaluation

The analysis results obtained for each specific phase of TiN, Al₂O₃, Ti₂O₃, and MgO·Al₂O₃ (MA), contained in the TiN–Al₂O₃, TiN–MA, Ti₂O₃–MA, and Ti₂O₃–Al₂O₃ samples, showed excellent figures of merit, as illustrated in Table 5 and Table 6. For the Raman data used in conducting the PLS regression model, all of the coefficients of determination (R²_CAL, R²_CV and R²_PRED) values for estimating the specific phases ranged between 0.48–0.98. It should be noted that the low R² (R²_CAL, R²_CV and R²_PRED) values, such as 0.48, were obtained from estimating individual phases of TiN and MgO·Al₂O₃ from the binary system of TiN–MgO·Al₂O₃, as shown in Table 5. In addition, for the SNV-treated Raman spectra data, the PLS regression model estimated (R²_CAL, R²_CV, and R²_PRED) values to be between 0.94–0.99. It was observed from the R² values obtained for both the Raman data and the pre-processed data that Raman difference had a relatively broader range in predicting the individual phases present in the samples.

Additionally, as shown Table 5 and Table 6, the PLS regression model was carried out on the raw Raman data and pre-processed (SNV) data. The figures of merit demonstrated that the estimated respective values for the root mean standard error features, such as RMSEC, RMSECV, and RMSEP, were generally improved when the raw Raman data were processed. For instance, from Table 5, the average values for predicting individual phases in TiN–(Al₂O₃ or MA) based on raw Raman data gave a coefficient of determination (R²_CAL, R²_CV and R²_PRED) = 0.72 and root mean standard error features (RMSEC, RMSECV, and RMSEP) = 14.24 wt%. However, SNV Raman data displayed the average values of R²_CAL, R²_CV, and R²_PRED = 0.96 and (RMSEC, RMSECV, and RMSEP) = 5.35 wt%. Based on the figures of merit from the PLS regression model, individual phases present in the binary synthetic inclusion system can be estimated from the prepared samples measured with Raman spectroscopy. Consequently, based on performance parameters (figure of merits), the analysis results obtained from pre-processed Raman spectra data (using SNV) improved compared with the raw Raman data results illustrated in Table 5 and Table 6.

4.3. Potential Limitations in the Measurements and Prospect for Future Application

Some potential limitations attributable to the application of Raman spectroscopy for characterizing the synthetic samples in this study could include a slight instrumental variation, the penetration depth of laser, and the potentially heterogeneous nature of the sample mixtures prepared. In addition, some of the generally expected instrumental operational challenges could be associated with variations in incident laser power, such as potential sample heating or influence of fluorescence, which could adversely impact the quality of the measured Raman spectra. In addition, it can be safely assumed that inclusion found in an actual steel samples matrix could be more heterogeneous compared to the synthetic inclusion sample prepared for this study. The limitations associated with instrumental variation and the possible heterogeneous state of samples were mitigated in this study by using a sample holder with a rotating stage and measuring the same sample, on average, three times under a relatively wide laser spot size. Furthermore, the potential effect of fluorescence on the Raman spectra was suppressed based on the type of Raman spectroscopy (time-gated) used for the measurement, as demonstrated by Lipiäinen et al. [27]. It should be noted that Raman spectroscopy characterization is generally considered to be a surface analysis technique; therefore, the bulk composition of phases present in the samples could be obtained from extrapolation estimated from the planar distribution of phase analysis results.

This study demonstrated the prospect of using Raman spectroscopy for characterizing Ti-based (TiN and Ti₂O₃) inclusions combined with other detrimental oxide phases of Al₂O₃ and MgO·Al₂O₃ spinel inclusion types in steel samples, assuming the steel matrix to be Raman-inactive. The calibration models used in this work have been demonstrated to be a promising method for quantifying inclusion phases. Furthermore, combined with the previous studies [10,20], future inclusion characterization of steel samples could be enhanced by applying the Raman spectroscopy, thereby offering an opportunity for a potential reference database of inclusion characteristics of steel samples.

This study presents the prospect of contributing to the numerous studies [6,7,8,12,13,28,29,30] already conducted concerning the formation process and generation of data characterized in this work by using Raman spectroscopy to assist in identifying and predicting phase content, including that of Al₂O₃, Ti₂O₃, and MgO·Al₂O₃, in a multiphase system. For instance, this technique could be used to estimate the phase content of Ti₂O₃ and MgO·Al₂O₃ from a MgO–Al₂O₃–Ti₂O₃ system, as demonstrated by Ren et al. [28], who established equilibrium phases in their study. In addition, this could be used in the future to predict the phase content of MgO·Al₂O₃, Ti₂O₃, or Al₂O₃ from the formation of MgO·Al₂O₃–Ti₂O₃-based inclusions obtained from the reaction associated with soluble Ti and MgO·Al₂O₃ [29] and for the characterization of complex Al-Ti oxide formations, with Al₂O₃ in the core and Ti₂O₃ on the surface [30].

The utilization of Raman spectroscopy as an analytical method presents some advantages over known conventional techniques such as scanning electron microscopy equipped with energy-dispersive spectroscopy [6,8,29], cathodoluminescence (CL) [31], and others outlined by Zhang et al. (2003) [9] used for the identification and classification of inclusion in steel. Some of the merits of using Raman spectroscopy include the relatively shorter duration needed for sample preparation, acquisition and analysis of the results, and a prospect for both the qualitative and quantitative characterization for Raman-active materials [32]. Additionally, most non-metallic inclusions, such as calcium aluminates and Ti-based phases, are Raman active, and steel is considered Raman inactive. Consequently, the Raman spectra measured from steel samples are regarded as originating from the inclusions found within the steel matrix and identified based on the fingerprint of the specific component present in the steel sample.

5. Conclusions

Raman spectroscopy, combined with calibration models such as PLS regression in this study, has demonstrated its suitability for qualitative and quantitative analysis to estimate the relative phase content for different sets of binary samples consisting of synthetic phases Al₂O₃, TiN, Ti₂O₃, and MgO·Al₂O₃. The main findings obtained from this work are summarized as follows:

• The most suitable Raman peaks (bands) identified for estimating the phases content present in the samples were:
  • For Ti₂O₃–MA and Ti₂O₃–Al₂O₃ samples: at 247 cm⁻¹ for Ti₂O₃, at 416 and 773 cm⁻¹ for MgO·Al₂O₃, and 383, 422, and 760 cm⁻¹ for Al₂O₃.
  • For TiN–Al₂O₃ and TiN–MA samples: Raman peaks at 157 cm⁻¹ for TiN, at 773 cm⁻¹ for MgO·Al₂O₃, and 383 cm⁻¹ for Al₂O₃.
• The Raman spectra data pre-processed by using the SNV had improved statistical performance, such as R², RMSECV, and RMSEP, in the evaluation of the phase contents compared to the analysis made using the raw Raman spectra data.
• This work and previous studies combined present the prospect of using Raman spectroscopy to characterize a complex multiphase system that may contain oxide- and nitride-based inclusions in steel samples.