Elsevier

Talanta

Volume 89, 30 January 2012, Pages 286-291
Talanta

Screening analysis of beer ageing using near infrared spectroscopy and the Successive Projections Algorithm for variable selection

https://doi.org/10.1016/j.talanta.2011.12.030Get rights and content

Abstract

This work proposes a method for monitoring the ageing of beer using near-infrared (NIR) spectroscopy and chemometrics classification tools. For this purpose, the Successive Projections Algorithm (SPA) is used to select spectral variables for construction of Linear Discriminant Analysis (LDA) classification models. A total of 83 alcoholic and non-alcoholic beer samples packaged in bottles and cans were examined. To simulate a long storage period, some of the samples were stored in an oven at 40 °C, in the dark, during intervals of 10 and 20 days. The NIR spectrum of these samples in the range 12,500–5405 cm−1 was then compared against those of the fresh samples. The results of a Principal Component Analysis (PCA) indicated that the alcoholic beer samples could be clearly discriminated with respect to ageing stage (fresh, 10-day or 20-day forced ageing). However, such discrimination was not apparent for the non-alcoholic samples. These findings were corroborated by a classification study using Soft Independent Modelling of Class Analogy (SIMCA). In contrast, the use of SPA-LDA provided good results for both types of beer (only one misclassified sample) by using a single wavenumber in each case, namely 5550 cm−1 for non-alcoholic samples and 7228 cm−1 for alcoholic samples.

Highlights

Near infrared spectroscopy (NIR) is used for screening analysis of aged beers. ► Variable selection is accomplished by using the Successive Projections Algorithm (SPA). ► Only one wavenumber is selected for screening analysis of non-alcoholic or alcoholic beers. ► NIR and SPA-LDA can be used for quality control of beer samples.

Introduction

An important challenge in the brewing industry consists of minimizing changes in the quality of beer from production to consumption stages. In this context, beer ageing is considered to be a major quality problem since the ageing flavours are mostly experienced as unpleasant [1]. Therefore, flavour stability over the ageing process has become one of the most important topics in brewing research over the past few years [2], [3], [4].

Beer flavour is conventionally assessed through the combination of common analytical tools (e.g. gas chromatography) and organoleptic profiling panels known as human sensory panels. However, these expensive and time-consuming methods may not be appropriate for on-line monitoring in breweries. In addition, organoleptic panels are prone to assessor fatigue and subjectivity, which may compromise the accuracy and reproducibility of the results.

Within this scope, one of the most promising directions for the development of innovative solutions is the use of spectroscopic methods. In fact, the speed and on-line capabilities of such methods meet the trends of automation and continuous processing in the brewing industry. Nuclear magnetic resonance (NMR) spectroscopy, for example, can be used to monitor the change in chemical composition of beers during the period of storage [5]. However, the cost associated to this technique may be prohibitive for routine use in the production line. Alternatively, the application of vibrational spectroscopy, such as near infrared (NIR), may constitute a less costly approach to characterize changes in the organic compounds involved in beer ageing. The joint use of NIR spectroscopy and chemometrics techniques has been reported in the literature in several applications of quality control and analysis of fuel [6], vegetable oil [7], cigarettes [8], and biodiesel [9], for example.

The NIR range extends from 13,300 to 4000 cm−1 and comprises overtones and combinations of fundamental vibrational transitions that occur in the middle infrared region [10]. As a result, a NIR spectrum is formed by several overlapping bands, and thus the use of multivariate analysis tools is usually required. For classification purposes, methods based on Principal Component Analysis (PCA), such as Soft Independent Modelling of Class Analogy (SIMCA), can be employed to exploit the spectrum in the entire working range [11], [12], [13]. As an alternative, variable selection methods can be used to identify specific spectral variables that convey useful information for the analytical problem at hand.

Variable selection may have several advantages, such as removal of noise and nonlinearity, as compared to using the full spectrum [14]. Algorithms described in the literature for selection of variables in chemical data include the Genetic Algorithm [15], Simulated Annealing [16], Tabu Search [17] and Colony of Ants [18] among others. In this context, Araújo et al. proposed the Successive Projections Algorithm (SPA) for selection of variables in multiple linear regression (MLR) [19]. In a subsequent work, Pontes et al. proposed a modification to SPA so that it could also be applied to classification problems in conjunction with Linear Discriminant Analysis (LDA) models [20].

Although studies on spectroscopic analysis of beer have been reported in the literature [21], [22], [23], [24], [25], [26], [27], this approach has not yet been pursued for the characterization of ageing. The objective of the present investigation consists of using NIR spectroscopy together with SPA-LDA for screening analysis of ageing in alcoholic and non-alcoholic beers. For comparison, full-spectrum SIMCA models are also employed.

Section snippets

The Successive Projections Algorithm for Linear Discriminant Analysis

The SPA-LDA algorithm is aimed at selecting a subset of variables with small collinearity and suitable discriminating power for use in classification problems involving C  2 different classes. For this purpose, it is assumed that a training set of N objects with known class labels is available to guide the variable selection process. In the case of spectroscopic data, for example, each object consists of a spectrum recorded over K wavenumbers (or wavelengths).

The SPA-LDA algorithm can be divided

Samples

A total of 83 samples of regular beer (42 alcoholic and 41 non-alcoholic) packed in bottles and cans were used in this study. To simulate a long storage period, 55 samples were stored in an oven at 40 °C, in the dark, in the original closed containers (bottles and cans). This procedure is known as forced ageing in the literature [32], [33], [34]. After 10 days, 28 samples (termed class F1) were removed from the oven. The remaining 27 samples (termed class F2) were removed after an additional

NIR spectra

As seen in Fig. 1a, the raw spectra of the 83 beer samples with water background subtracted display high noise levels, as well as systematic variations of baseline. Such problems were corrected by using the first derivative Savitzky-Golay filter, which resulted in the derivative spectra shown in Fig. 1b. These spectra were employed throughout the study.

Principal Component Analysis

Fig. 2a presents the PCA score plot for the overall data set, with the percentage explained variance indicated at each axis. As can be seen, the

Conclusions

This paper proposed a methodology for screening analysis of beer ageing employing NIR spectrometry and Linear Discriminant Analysis coupled with the Successive Projections Algorithm for wavenumber selection. PCA results indicated that the NIR spectrum in the adopted working range (12,500–5405 cm−1) provides a clear discrimination of alcoholic beer samples with respect to the ageing stage (fresh, 10-day or 20-day forced ageing). However, such a discrimination was not apparent for the

Acknowledgments

The support of the Iran National Science Foundation (INSF) is gratefully appreciated regarding the finance of this research. Financial support by the Spanish Ministry of Science (grant AGL2009-12660/ALI) is gratefully acknowledged as well. Also, M. C. U. Araujo, R. H. K. Galvão and A. A. Gomes acknowledge the support of CNPq (research fellowships) and CAPES (scholarships).

References (37)

  • B. Vanderhaegen et al.

    Food Chem.

    (2007)
  • B. Vanderhaegen et al.

    Food Chem.

    (2006)
  • M. Ghasemi-Varnamkhasti et al.

    Trends Food Sci. Technol.

    (2011)
  • J.A. Rodrigues et al.

    Anal. Chim. Acta

    (2011)
  • A.F.C. Pereira et al.

    Food Res. Int.

    (2008)
  • E.D.T. Moreira et al.

    Talanta

    (2009)
  • G. Veras et al.

    Talanta

    (2010)
  • S. Wold

    Pattern Recogn.

    (1976)
  • M. Daszykowski et al.

    Chemom. Intell. Lab. Syst.

    (2011)
  • M.C.U. Araújo et al.

    Chemom. Intell. Lab. Syst.

    (2001)
  • M.J.C. Pontes et al.

    Chemom. Intell. Lab. Syst.

    (2005)
  • A. Nordon et al.

    Anal. Chim. Acta

    (2005)
  • F. Inon et al.

    Anal. Chim. Acta

    (2006)
  • B.M. Nicolai et al.

    Postharvest Biol. Technol.

    (2007)
  • G. McLeod et al.

    J. Food Eng.

    (2009)
  • H. Li et al.

    J. Near Infrared Spectrosc.

    (2009)
  • F. Liu et al.

    Anal. Chim. Acta

    (2009)
  • R. de Maesschalck et al.

    Chemom. Intell. Lab. Syst.

    (2000)
  • Cited by (52)

    • Biophotonics in food technology: Quo vadis?

      2023, Journal of Agriculture and Food Research
      Citation Excerpt :

      The alcohol content, soluble solid content, fat content, protein contents, pH, adulterants, other nutrients and minerals in liquid food have been studied and analyzed using NIR spectroscopy. The feasibility of NIR spectroscopy to detect the alcohol content in wine was studied [51–55,58]. pH is an important factor as it is directly related to food spoilage and acidity.

    View all citing articles on Scopus
    View full text