Umetrics Suite Blog

Four uses of data analytics in the food and beverage industry

January 29, 2019

Consumers expect a certain consistency in quality and taste from the food and beverage brands they love. But many factors can influence the way a product tastes when it reaches the consumer – ranging from the manufacturing process to seasonality of ingredients to storage temperatures. Similarly, a number of other factors may influence the overall quality attributes that matter, such as alcohol content of beer or stability of the whiskey aging process.

Beer production quality attributes

Data analysis methods such as statistical process control can help predict the factors that will affect quality of beer and distilled products like whiskey.

One way food and beverage manufacturers can take more control of their quality parameters is through data analytics. Data analysis provides a way to understand which elements will have the greatest effect on a product during manufacturing and predict the impact of these factors on quality and taste.

These insights can be used in real-time to adjust manufacturing processes to account for deviations while maintaining a consistent quality output. Data analysis can also be used to evaluate the impact of factors such as storage and transportation on the quality of packaged foods and beverages.

Let’s take a look at four key ways that data analytics can be used in the food and beverage industry, focusing on brewing and distilling processes as examples.

1. Predictive statistical process control

The first way data analytics can be applied in the food and beverage industry is predictive statistical process control of a batch process, such as for a batch-based fermentation process like that used for brewing and distilling. Real-time data monitoring combined with a prediction engine (such as SIMCA-online) allows operators to make adjustments to batch productions as deviations occur.

How? Statistical process control uses a framework known as model predictive control (MPC) based on multivariate projection models created using two data analytics methods - Partial Least Squares (PLS) and Principal Component Analysis (PCA). The model creates predictions based on past behavior and process parameters of previous batches. Real-time visibility and prediction enables immediate action, either manually or automatically, directly at the point of deviation.

deviations

A prediction engine uses multivariate data to project the trajectory of a process so deviations can be forecasted.

2. Predicting shelf-life

Over time, quality characteristics of a product can change, or degrade. Understanding the parameters that may affect a product’s shelf life – in time to make adjustments to counteract them – can save food and beverage industry manufacturers a lot of time and money, as well as prevent product waste.

We can use an example from the pharma industry in which the shelf-life of tablets are predicted. This can be compared, for example, to the potency of whiskey as it matures over time. It’s possible to use several batch samples to understand which attributes have the most effect on potency (in this case, time) and predict the shelf life based on specific parameters. In the pharma industry, these predictions are required to meet stringent specifications for regulatory compliance and must be precisely accurate.

Using the same methods in the food and beverage industry, allows an accurate prediction of shelf life or prediction of the aging time needed for a specific beverage.

coefficients plot

The coefficient plot indicates that time (T) has the largest impact on potency or API degradation. Longer time increases API degradation.

 

3. Measuring critical quality attributes

In the brewing industry, the alcohol content of a beverage is a critical quality parameter that is routinely analyzed. But measuring this quality parameter can be time-consuming (and occurs far along in the production process). Beer manufacturers might consider using other quality measurements instead, which allow the production process to continue unimpeded. For instance, Near Infrared Spectroscopy (NIR)analyses are made on raw materials like malt and hops. Can this technique be applied to analysis of alcohol content in beer, offering a more rapid analysis method than the traditional wet-chemistry method?

Data analytics can be used to determine whether new (less expensive and less time consuming) methods of quality analysis are as effective as traditional methods. In one case, using multivariate Orthogonal Partial Least Squares (OPLS) regression models to check alcohol content and color, showed this method was able to obtain prediction errors comparable with those of reference methods. The implications are that an NIR sensor could be implemented online, and thereby facilitate real time quality control.

umetrics blog 3-food bev

Observed vs predicted plots for alcohol content (5 OPLS components) and color (1 OPLS component) by means of Orthogonal Partial Least Squares (OPLS) .

4. Counterfeit modeling

To assure quality and prevent false labeling, developing a method of “fingerprinting” product samples to indicate their original source is useful. Advanced data analytics can be used to discriminate between real products and fake replicates, such as verifying that a product comes from a certain region or has a certain claimed composition.

For example, it can be used to make a distinction between Cava and Champagne wines based upon data about the trace element composition of the wines. Champagne is a much stricter controlled variety of wine than Cava, with very specific demands to qualify as a Champagne, so this type of data analytics can be useful to determine counterfeits.

It involves creating a definition based on how closely specific parameters fall from the acceptable model domain, and determining what the critical limits are in order for a wine to fall within the acceptable range.

cava-chamagne-plot

The first score vector separates Cava (green) from Champagne (Blue). The first loading vector expresses information related to this distinction: High amounts of for instance Zn, P, Mn and small amounts of Al, NA, Sr indicate a Champagne, and vice versa for a Cava.

Prediction and control of processes

These four examples show some of the ways that data analytics and statistical process control can be used in the food and beverage industry to improve processes and better predict parameters that will affect key quality attributes.

Find out more

Want to know more about how data analytics can be used to improve food and beverage manufacturing?Download this paper highlighting four case examples. 

Get the article

 

 

Topics: Real Time Process Monitoring, Statistical Process Control, Manufacturing Quality Control, Batch processes, Food & Beverage

Robert Zuban

Written by Robert Zuban

Robert Zuban is a Senior Data Scientist at Sartorius Stedim Biotech in Sweden. He is also a member of the Faculty of Engineering at Lund University.