A single model to monitor multistep craft beer manufacturing using near infrared spectroscopy and chemometrics

Highlights

• NIR and MSPC to monitor and control the beer production.

• Simple multivariate control charts established for all the steps of the process.

• Variability within-batches is smaller than the variability within-steps.

• The complete procedure monitor and control with a single PCA model.

Abstract

This manuscript presents a comprehensive approach to monitoring the whole process of craft beer production (mashing, circulation, boiling, fermentation and carbonatation), using a simple, rapid and green methodology like Near Infrared spectroscopy combined with MSPC (Multivariate Statistics Process Control). A Principal Component Analysis model is calculated with near infrared spectra (range between 800–2500 nm) collected in all the steps of the process (i.e., using a batch-to-batch approach), and a multivariate control chart is generated in order to monitor the beer development. Each batch was composed of a variable number of samples (average of 55 samples per batch) depending on the sampling time of every step. Four batches working under normal operating conditions are used to construct the model. Three external batches are used to validate the proposal (two of them with induced disturbances and another one working under normal operating conditions). The results were compared to those obtained by monitoring the solid soluble content (SSC) by using Partial Least Squares regression to ascertain the richness of the information given by NIR. The results illustrate the versatility and simplicity of the proposal and its reliability towards a global monitor and control of the beer-making procedure.

Introduction

Beer is the largest alcoholic beverage consumed worldwide, accounting for 78.2% of the total consumption of alcoholic beverages (Gómez-Corona et al., 2016). Beer prevalence in the market is also linked to its suitability for large-scale industrial production. However, from the last decades, consumers are changing their attitude and perception of food and beverages, resulting in the increasing consumption of craft products (Gómez-Corona et al., 2016). Thus, the craft beer producers are called to learn, from the large-scale brewing industry, the strategies to control the process in order to keep constant the perceived quality and the safety standards.

Despite this trend, the quality parameters that must be monitored in the handcraft beer making process are the same as in large scale breweries. These parameters are temperature, pH (Lachenmeier, 2007), soluble solids content (SSC) or specific gravity, and, in some cases, optical density (OD) (Lachenmeier, 2007; McLeod et al., 2009). Whereas it is quite tricky that other relevant parameters, such as yeast viability, undesired microbial contamination, bitterness unit, colour and organic acids concentration, could be continuously controlled in handcraft brewing. Indeed, the vast number of parameters to be controlled calls for onerous and lagging techniques in terms of money, human resources and chemical handling. Moreover, each technique will evaluate a single parameter per time, neglecting the overall vision to maintain the process under control (Bamforth, 2013).

The reliability of NIR spectroscopy as a process analyser is strictly linked to the proper chemometric strategy to extract the relevant information from the spectra. Principal component analysis (PCA) is, by far, one of the most reliable methods that has been used for in the Multivariate Statistical Process Control (MSPC) scenario (Ferrer, 2007; MacGregor and Kourti, 1995; Nomikos and MacGregor, 1995; Wold et al., 1998). PCA is a variable reduction method that seeks for variance (changes) in process data by projecting the data into a principal components (PCs) space where each PC is independent of each other and account to explain the major sources of variance in the data. This property has been widely used in MSPC scenarios by setting conditions to define what is known as normal operating conditions (NOC), where all measurements are correlated to the quality of the product and controlled. This fact conveys the ability to establish certain statistical boundaries to detect which measurements are not fulfilling the requirements for being under NOC.

This approach is in agreement with the Process Analytical Technology (PAT) vision, firstly introduced by FDA for the pharmaceutical industry (Kourti, 2006), but effective also as modelling and control strategy for the food industry (Grassi et al., 2014a). In the beer manufacturing process, there are several steps where controlling the quality of the product by PAT approaches is essential to assure the quality in the proceeding steps, like boiling, circulation, fermentation, mashing, carbonatation, among others. The implementation of PAT approaches into craft brewing offers some apparent advantages which include (1) controlled and optimised utilisation of raw materials, leading to less variation in the final product quality, (2) waste reduction, (3) process cycle time reduction and (4) replacement of slow and costly laboratory testing with newer and reliable sensor technologies, like Near Infrared (NIR) spectroscopy (O’Donnell et al., 2014). The NIR success in the brewing industry has been largely proven. It has been used for authentication and ageing purposes (Ghasemi-Varnamkhasti et al., 2012; McLeod et al., 2009) and quality control (Duarte et al., 2004; Ghasemi-Varnamkhasti et al., 2012; Iñón et al., 2005; Lachenmeier, 2007). Few works dealt with small scale (Giovenzana et al., 2014) or handcraft beer production (Grassi et al., 2014b,), and they mainly focused on the fermentation step.

As defined earlier, beer is produced in a multi-step procedure where every step has its definition and specific parameters to assess quality. This could lead to having to define particular NOCs for each step, meaning that every step should be defined by a specific model with its statistical boundaries to consider a measurement being under NOC. The construction for such models could be cumbersome and may lead to a loss of final reliability of the global process and, hence, the final product. Instead, if the same measuring device is used in all the steps, and the variations within each step are demonstrated to be negligible compared to the variation of the whole process, an alternative would be to create a general model that involves all the steps, and, consequently, having the specific statistical parameters to control the process as a whole.

Therefore, this manuscript aims to develop a global strategy that encompasses the generation of a NOC, covering all the steps in beer making procedure by using NIR and PCA. This will demonstrate that with NIR, a simple PCA and a properly validated MSPC chart, it is possible to control the good development of the beer in every single stage of its manufacture. PCA will offer enough chemical understanding of the NIR spectra collected at different times and stages to check the development of the beer, while the MSPC chart will ensure that the changes in beer are under controlled conditions. As supplementary information and in order to demonstrate the linkage between our proposal and more analytical parameters, the soluble solid compounds (SSC) will be quantified and verified using the same NIR spectra and multivariate regression methods (Partial Least Squares Regression – PLS).