Evaluation of acetic acid and ethanol concentration in a rice vinegar internal venturi injector bioreactor using Fourier transform near infrared spectroscopy

Abstract

In the process of fermenting rice vinegar, the concentration of acetic acid and ethanol concentration must be measured for monitoring of the total concentration. Near infrared spectroscopy has been used to rapidly monitor the concentration of acetic acid and ethanol concentration daily during 10 cycles of the fermentation process. The model was developed using partial least squares regression. For predicting concentration of acetic acid with near infrared spectroscopy, the coefficient of determination (R²), root mean square error of calibration, root mean square error of cross validation, ratio of standard error of validation to standard deviation, and bias was 0.96, 2.30 g L⁻¹, 2.44 g L⁻¹, 1.11 g L⁻¹, and 5.56, respectively. For ethanol concentration, the value of R², root mean square error of calibration, root mean square error of cross validation, bias and ratio of prediction to deviation were predicted to be 0.94, 3.15 g L⁻¹, 2.73 g L⁻¹, −0.40 g L⁻¹, and 4.04, respectively. However, both models provided fair performance when tested with an external set of samples, indicating that the models could be applied for rough screening.

Keywords

Rice vinegar Fourier transform near infrared spectroscopy acetic acid ethanol internal Venturi injector bioreactor

Introduction

Vinegar, a popular flavor enhancer and condiment, can be produced from the fermentation of various sources such as rice, corn, fruit, or sugar using microorganisms (or yeasts). The first step of the process yields alcohol, which is later mixed with bacteria to produce the final product of vinegar. The same process occurs in rice vinegar production. The process needs regular monitoring of the acetic acid concentration (AAC) and ethanol concentration (ETC) to optimize the total concentration (TC) needed which is 8.0. Normally, AAC is determined by titration, while ETC is analyzed by Ebulliometer as a standard method. These methods, however, are time-consuming and have limited practicality for use in a continuous processing plant. Therefore, rapid and accurate techniques are required which leads to the potential use of near infrared (NIR) spectroscopy. NIR spectroscopy is also an environmentally friendly method, as no chemicals are needed. Ji-yong et al. reported the rapid detection of total acid content and classified different types of vinegar based on NIR spectroscopy and least-squares support vector machine classification.¹ Ouyang et al. showed the real-time monitoring of process parameters in rice wine fermentation by a portable spectral analytical system combined with multivariate analysis.² Liu et al. reported the comparison of calibrations for the determination of the soluble solids content and the pH of rice vinegars using visible and short-wave NIR spectroscopy.³ Sáiz-Abajo et al. reported the prediction of organic acids and other quality parameters of wine vinegar by NIR spectroscopy,⁴ while Yano et al. reported the prediction of the concentrations of ethanol and acetic acid in the culture broth of a rice vinegar fermentation using a monochromator NIR system accompanied with multiple linear regression (MLR).⁵ Although literature exists on the prediction of acetic acid and ETC of a rice vinegar fermentation using NIR spectroscopy by MLR, no information is available on the daily monitoring of acetic acid and ETC of rice vinegar fermentation using NIR spectroscopy by partial least squares (PLS) regression. The PLS regression calibrations should provide better prediction performance because all spectral wavelengths would be used for determination and would allow for less multicollinearity among the latent variables. The aim of this study was to develop PLS regression calibrations using Fourier-transform (FT) NIR spectroscopy for the daily monitoring of acetic acid and ETC in the internal Venturi injector bioreactor of rice vinegar. Models were developed through 10 cycles of the fermentation process in this experiment. The off-line protocol developed could save time to measure acetic acid and ETC and reduce sample loss.

Materials and methods

Sample preparation and fermentation

Rice vinegar was prepared from sticky rice in the Laboratory of Fermentation Technology, Faculty of Agro-Industry, King Mongkut’s Institute of Technology Ladkrabang (KMITL), Bangkok, Thailand, and used as the raw material for acetification in the internal Venturi injector bioreactor.⁶ The inoculums preparation was as followed⁶: The high-acid tolerant strain of Acetobacter aceti WK was used in this study. The acetic acid bacteria (AAB) starter was cultured under aeration of 4.5 L min⁻¹ for seven days at 30 ± 1℃ in a complex medium (Merck, Darmstadt, Germany) composed of (per liter of water) 50 g glucose, 5 g yeast extract, 0.2 g Mg₂SO₄.7H₂O and 0.5 g (NH₄)₂HPO₄. The ETC and AAC in rice wine were 110 ± 1.5 g L⁻¹ and 2.5 ± 0.4 g L⁻¹ respectively, while AAC in the vinegar was 80 ± 0.8 g L⁻¹. This was standardized by adjusting the ETC and AAC to give a target TC of 35 ± 1 g L⁻¹ ETC and 45 ± 1 g L⁻¹ AAC for a TC of 80 g L⁻¹. Every morning at around 9 a.m., the two samples were collected from two bioreactors. The fermentation was followed for 75 days. It took about 7–10 days for a single fermented rice to vinegar cycle process with a total of 10 cycles of the fermentation being performed.

NIR spectroscopy

Acquisition of NIR spectra was performed using a FT-NIR spectrometer (MPA, Bruker Optics, Ettlingen, Germany). The range used for scanning was between 12,500 and 4000 cm⁻¹ (800–2500 nm) with a resolution of 16 cm⁻¹. Samples were scanned 32 times to obtain one average spectrum. Before every single scanning the samples, a reference material made of gold was scanned for background. Each sample was transferred into a quartz-glass vial of 22.0 mm diameter and 48.0 mm height, as a measurement cell. Then the sample was covered with a transflection plate with standoff gap 0.1 mm made from aluminum which provided a full optical path length of 0.2 mm and scanned. All scanning was conducted in an air-conditioned room at a temperature of 25.0 ± 2.0℃ while maintaining the temperature of sample storage at 0 ± 1℃ to prevent bacteria. This is because bacteria in vinegar can change ethanol to acetic acid very rapidly. The bioreactor was at the Laboratory of Fermentation Technology, Faculty of Agro-Industry, KMITL, and the FT-NIR spectrometer was at the Near Infrared Spectroscopy Research Center for Agricultural Product and Food, Faculty of Engineering, KMITL. It took 10 min to move the sample. For this reason, the sample was kept at 0℃ to stop bacterial growth.

Determination of AAC and ETC

AAC was determined by acid-base titration with 0.1 mol⁻¹ NaOH using phenolphthalein as an indicator. ETC was analyzed by Ebulliometer (Dujardin Salleron, Paris, France).⁶

Determination of outlier of the reference data

Before developing the calibration model, potential outliers of in the reference data were identified using the following the Z-score equation (1)

\frac{X_{i} - \bar{X}}{SD} \geq \pm 3

(1)

where X_i is the value of sample,

\bar{X}

is the average of the value of all samples and SD is the standard deviation of all samples. If equation (1) is satisfied, the corresponding sample was considered as an outlier.⁷

Spectrum preprocessing and NIR spectroscopic modeling

Calibrations for the prediction of acetic acid and ETC in rice vinegar fermented bioreactor were developed by PLS regression using QUANT-2 (Bruker Optics, Ettlingen, Germany). The software provided different spectral data preprocessing techniques including constant offset elimination, straight line subtraction, vector normalization, min–max normalization, multiplicative scattering correction (MSC), first derivative, second derivative, first derivative + straight line subtraction, first derivative + vector normalization, and first derivative + MSC. However, the vector normalization and MSC were selected for calibration because the spectra had effected by the multiple scattered which caused the tilting and baseline shift. The spectra used were either raw spectra or preprocessed spectra using the above-mentioned techniques before making the model. The sample data were run in ascending order and the samples were randomly separated into calibration set and test set by the ratio of 7:3 by the software where 30% of the data were assigned for the validation set and 70% for the calibration set. The data range of the calibration set covered those of the validation set. The model performance was determined by the coefficient of determination (R²), root mean square error of calibration (RMSEC), root mean squared error of prediction (RMSEP), ratio of prediction to deviation (RPD), and bias. The optimum model was then validated using the unknown set of 45 samples by predicting the samples with the models developed and comparing the r², RMSEP, bias, and RPD.

Results and discussion

Spectrum analysis

Figure 1 shows the spectra of rice vinegar, 99.85% acetic acid and 99.5% ethanol in the spectral range 12,500–3600 cm⁻¹ (800–2778 nm). Table 1 shows the dominant absorption bands and the corresponding vibration of the spectra. However, some wavenumber ranges could not be used because the spectrum was saturated, particularly in the region of the water absorbance due to the path length being too long. Subsequently, the full range of spectra of samples was subjected for PLS modeling and the saturated regions were not selected by the software for modeling. The absorbance was higher in the 3000–4000 wavenumber region and contained noise. The software did not select the noisy area or the water band of combination vibration.

Figure 1.

The average spectra of rice vinegar, 99.85% acetic acid and 99.5% ethanol.

Table 1.

The dominant absorption bands and the corresponding vibration of the spectrum of rice vinegar, pure acetic acid and pure ethanol.

Samples	Wavenumber (cm⁻¹)	Wavelength (nm)	Wavelength (nm) referred from reference⁹	Bond vibration	Structure	Absorbance
Rice vinegar	6881	1453	1450	2 × C–H str. + 2 × C–H def.	aromatic	1.116043
Rice vinegar	5176	1932	1940	O–H str. + O–H def.	H₂O	3.329828
Pure acetic acid	5949	1681				0.540129
	5832	1715				0.541751
	4950	2020				0.610438
	4435	2255				1.553230
	3965	2522				1.433914
Pure ethanol	5909	1692				0.466187
	5801	1724				0.448305
	5785	1729				0.475663
	4752	2104				0.636312
	4327	2311				1.668432

Data of the models

Table 2 shows the statistical data of AAC and ETC. A total of 150 samples were used for model development of AAC and ETC. The data for the AAC model had minimum value of 42 g L⁻¹, maximum of 86 g L⁻¹, mean of 51.7 g L⁻¹ and standard deviation (SD) of 11.0 g L⁻¹, and for ETC model the minimum was 0.0 g L⁻¹, maximum was 38.3 g L⁻¹, mean was 22.57 g L⁻¹ and SD was 11.14 g L⁻¹.

Table 2.

The statistical data of AAC and ETC for PLS regression modeling.

Parameter	Data set	Sample	Min (g L⁻¹)	Max (g L⁻¹)	Mean (g L⁻¹)	SD (g L⁻¹)
AAC	Total samples	150	42	86	51.7	11.0
	Calibration set	105	42	86	51	10
	Prediction set	45	42	85	54	12
ETC	Total samples	150	0.0	38.3	22.1	10.9
	Calibration set	105	0.0	38.3	23.5	10.9
	Prediction set	45	0.0	38.0	21.1	10.9

AAC: acetic acid concentration; ETC: ethanol concentration.

Model for AAC

Figure 2 shows preprocessing spectra by MSC of rice vinegar for AAC modeling. Figure 3(a) shows the comparison of AAC predicted values and measured values of the calibration set and the test set. Table 3 shows the PLS analysis of the model for AAC. The model used wavenumber at 9403.8–6094.3 and 4605.4–4242.9 cm⁻¹ with five factors being required. The calibration set had a value of R² of 0.95 and RMSEC of 2.30 g L⁻¹. The validation set had a value of r² of 0.96, RMSEP of 2.44 g L⁻¹, bias of 1.11 g L⁻¹ and RPD of 5.6. Williams⁸ recommended that r² values between 0.92 and 0.96 could be used for most applications including quality assurance and process monitoring and that the RPD between 5.0 and 6.4 could be used for good screening.

Figure 2.

Multiplicative scattering correction (MSC) preprocessing spectra of one fermented cycle samples of rice vinegar for AAC model.

Figure 3.

graph of AAC model (a) Scatter plots (b) X-loading weight plot, and (c) the expanded region from 5000 to 4000 cm⁻¹.

Table 3.

Result of PLS regression modeling for AAC and ETC.

Parameter	Regions (cm⁻¹)	Preprocessing	PLS factor	Model						Unknown
				Calibration		Validation			RPD	Unknown
				R²	RMSEC (g L⁻¹)	r²	RMSECV (g L⁻¹)	Bias (g L⁻¹)	RPD	R²	RMSEP (g L⁻¹)	SEP (g L⁻¹)	Bias (g L⁻¹)	RPD
AAC	9403.8–6094.3 4605.4–4242.9	MSC	5	0.95	2.30	0.96	2.44	1.11	5.6	0.91	2.83	2.90	−0.05	2.9
ETC	8454.9–7498.3 4605.4–4242.9	Vector normalization	3	0.92	3.15	0.94	2.73	−0.40	4.0	0.98	2.22	1.40	−1.75	6.3

AAC: acetic acid concentration; ETC: ethanol concentration; PLS: partial least squares; RMSECL: root mean square error of calibration; RMSECV: root mean square error of cross validation; RPD: ratio of prediction to deviation.

Table 4 shows the absorption bands of regression coefficient spectra and X-loading weight spectra and their bond vibration of the model for AAC. Figure 3(b) and (c) show the X-loading weight plots of first three PLS factors and the expansion of the plots at 5000-4000 cm-1 where the dominant peaks were, respectively.

Model of ETC

Figure 4 shows the vector normalization preprocessing spectrum of a rice vinegar for ETC modeling. Figure 5(a) shows the comparison of ETC-predicted values and measured values of the calibration set and the test set. Table 3 shows the PLS analysis of the model for ETC. The model used wavenumber at 8454.9–7498.3 and 4605.4–4242.9 cm⁻¹ with three PLS factors. The calibration had a value of R² of 0.92 and an RMSEC of 3.15 g L⁻¹. The validation model provided r² of 0.94, RMSEP of 2.73 g L⁻¹, bias of −0.40 g L⁻¹ and RPD was 4.0. Williams⁸ recommended that RPD values between 3.1 and 4.9 could be used for screening.

Figure 4.

Vector normalization preprocessing spectra of one fermented cycle samples of rice vinegar sample of ETC model.

Figure 5.

Graph of ETC model (a) Scatter plots, (b) X-loading weight plot, and (c) the expanded region from 5000 to 4000 cm⁻¹.

Table 4 also shows the absorption bands of regression coefficient spectra and X-loading weight spectra and their bond vibration of the model for ETC. Figure 5(b) and (c) show the X-loading weight plots of first three PLS factors of the ETC model and the expansion of the plots at 5000–4000 cm⁻¹ where the dominant peaks were, respectively.

Table 4.

The absorption bands of regression coefficient spectra and X-loading weight spectra and their bond vibration of the model for AAC and ETC.

AAC
Wavenumber (cm⁻¹)	Wavelength (nm)	Wavelength (nm) referred form reference⁹	Bond vibration	Structure	Absorbance
Regression coefficient
6981	1432	1440	O–H str. first overtone 2 x C–H str. + 2 x C–H def.	Starch CH	46.338932
4590	2179	2280	C–H str. + C–H def.	CH₃	−89.767281
4451	2247	2252	O–H str. + O–H def.	starch	341.017517
4404	2271	2280	C–H str. + C–H def.	CH₃	−96.277000
4343	2271	2310	C–H str. + C–H def.	CH₂	−155.831085
Wavenumber (cm⁻¹)	Wavelength (nm)	Wavelength (nm) referred form reference⁹	PLS latent variable	Bond vibration	Structure	Absorbance
X-loading Weight
4474	2235	2200	3	C–H str. + C=O str.	–CHO	0.206202
4466	2239	2200	1	C–H str. + C=O str.	–CHO	0.110993
4451	2247	2252	5	O–H str. + O–H def.	Starch	0.236384
4443	2250	2252	2	O–H str. + O–H def.	Starch	−0.184861
4412	2267	2280	4	C–H str. + C–H def.	CH₃	−0.275287
ETC
Wavenumber (cm⁻¹)	Wavelength (nm)	Wavelength (nm) referred form reference⁹	Bond vibration	Structure	Absorbance
Regression coefficient
4458	2243	2252	O–H str. + O–H def.	Starch	−665.659302
4412	2267	2252	O–H str. + O–H def.	Starch	691.231018
4343	2303	2310	C–H str. + C–H def.	CH₂	673.218689
4289	2332	2336	C–H str. + C–H def.	Cellulose	−246.244705
4258	2348	2352	C–H def. second overtone	Cellulose	58.500179
Wavenumber (cm⁻¹)	Wavelength (nm)	Wavelength (nm) referred form reference⁹	PLS latent variable	Bond vibration	Structure	Absorbance
X-loading weight
4474	2235	2252	3	O–H str. + O–H def.	Starch	−0.340247
4412	2266	2252	1	O–H str. + O–H def.	Starch	0.296166
4242	2357	2352	2	C–H def. second overtone	Cellulose	0.182185

AAC: acetic acid concentration; ETC: ethanol concentration; PLS: partial least squares.

Validation by unknown samples

Figure 6(a) shows the scatter plot of prediction of an unknown by model for AAC and Figure 6(b) by model for ETC. Table 2 shows the PLS regression analysis result of predicted unknown for AAC and ETC. For AAC, the r² was 0.91, RMSEP was 2.83 g L⁻¹, SEP was 2.90 g L⁻¹, bias was −0.051 g L⁻¹, and RPD was 2.9. For the model of ETC, r² was 0.98, RMSEP was 2.22 g L⁻¹, SEP was 1.49 g L⁻¹, bias was −1.75 g L⁻¹ and RPD was 6.3. The r² of the ETC of the unknown sample was higher than the model, because the ethanol data of unknown samples were within the range of the data of the calibration set and the data of the calibration set was a good representative sample of future data from the fermentation cycles. The RPD of the AAC of the unknown sample however was lower than the model, since the SEP was higher due to larger variation of the samples.

Figure 6.

Scatter plot of prediction of unknown by model (a) for AAC and (b) for ETC.

The range of acetic acid of each cycle was 45–80 g L⁻¹ and that of ethanol was 0–35 g L⁻¹. For AAC, the error was 3.5% of the maximum value of total samples and for the ETC, it was 6.3% which was acceptable for rough screening.

Comparison with other research

Yano et al.⁵ reported the MLR prediction of the concentrations of ethanol and acetic acid in the culture broth of a rice vinegar fermentation using NIR spectrophotometer. They found that the r² was 0.998 and that SEC was 0.37 g L⁻¹ for ETC, and that r² was 0.88 and SEC was 0.39 g L⁻¹ for AAC, however the SEP was not reported. For the model for ETC in this research, the r² was 0.94 and SEP was 1.40 g L⁻¹, while for the model for AAC, the r² was 0.96 and SEP was 2.90 g L⁻¹. The results are different for the two studies because the range of concentrations is totally different. In this study the range of AAC was 45.0–80.0 g L⁻¹ while that of Yano et al. was 66.9–109 g L⁻¹. The range of ETC was 0.0–35.0 g L⁻¹ while that of Yano was 5.7–34.8 g L⁻¹. In addition the experiment conducted here used data which was monitored daily throughout 10 cycles of the fermentation process of two bioreactors. It was indicated that the models were for AAC and ETC daily monitoring process.

Conclusions

The model for daily monitoring of AAC and ETC developed using FT-NIR spectroscopy accompanied by PLS regression gave fair results from the r², which indicated that the models could be applied for rough screening and process monitoring. NIR spectroscopy could be a simple and time saving method for the monitoring of AAC and ETC in rice vinegar.

Footnotes

Acknowledgements

The authors appreciate the Near Infrared Spectroscopy Research Center for Agricultural Product and Food () at King Mongkut's Institute of Technology Ladkrabang, Bangkok, Thailand, for instrument support, and the Laboratory of Fermentation Technology, Faculty of Agro-Industry, King Mongkut’s Institute of Technology Ladkrabang. Bangkok, Thailand, for samples and reference test measurements.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

References

Ji-yong

Xiao-bo

Xiao-wei

, et al. Rapid detecting total acid content and classifying different types of vinegar based on near infrared spectroscopy and least-squares support vector machine. Food Chem 2013; 138: 192–199.

Ouyang

Zhao

Pan

, et al. Real-time monitoring of process parameters in rice wine fermentation by a portable spectral analytical system combined with multivariate analysis. Food Chem 2016; 190: 135–141.

Liu

Wang

. Comparison of calibrations for the determination of soluble solids content and pH of rice vinegars using visible and short-wave near infrared spectroscopy. Analytica Chimica Acta 2008; 610: 196–204.

Sáiz-Abajo

González-Sáiz

Pizarro

. Prediction of organic acids and other quality parameters of wine vinegar by near-infrared spectroscopy. A feasibility study. Food Chem 2006; 99: 615–621.

Yano

Aimi

Nakano

, et al. Prediction of the concentrations of ethanol and acetic acid in the culture broth of a rice vinegar fermentation using near-infrared spectroscopy. J Ferment Bioeng 1997; 84: 461–465.

Krusong

Yaiyen

Pornpukdeewatana

. Impact of high initial concentrations of acetic acid and ethanol on acetification rate in an internal Venturi injector bioreactor. J Appl Microbiol 2014; 118: 629–640.

Posom

Shrestha

Saechua

, et al. Rapid non-destructive evaluation of moisture content and higher heating value of Leucaena leucocephala pellets using near infrared spectroscopy. Energy 2016; 107: 464–472.

Williams P. Near-infrared technology—getting the best out of light. Nanaimo, British Columbia, and Winnipeg, Manitoba, Canada, 2007.

Osborne

Fearn

Hindle

. Advantages and limitations of NIR. Practical NIR spectroscopy with applications in food and beverage analysis, England: Longman Group, 1993.