Brewery Optimizes Predictive CO2 Model Built in Ignition

Project Summary:

Carlton and United Breweries redeveloped an Excel macro- and VBA-driven predictive CO2 model from beer production in Ignition’s Perspective Module. The model shows the predicted amount of liquid CO2 in storage hour by hour over two weeks. The system also monitors key quality and performance indicators in the liquid CO2 system and provides historical capabilities.

Problem:

Carlton and United Breweries (CUB) operates as the biggest beer manufacturer in the Australian state of Victoria, boasting a production capacity of approximately 6,000 kiloliters (kL) of beer per week. Among the crucial processes involved in beer production is carbonation, where CO2 is injected into liquid beer lines to achieve the desired level of carbonation. Typically, breweries can sustainably source CO2, as it is generated during the fermentation process and subsequently collected from fermenter headers. This captured CO2 is then processed and reintroduced into the beer before packaging, ensuring consistency and quality in the final product.

To anticipate and manage future CO2 levels within its storage facility, CUB employs a comprehensive CO2 modeling system. This model serves the dual purpose of presenting current CO2 quantities in storage (measured in metric tons) and extrapolating future levels on an hourly basis for up to two weeks. By providing this foresight, the model enables CUB to assess whether it is necessary to import or export CO2 to maintain optimal storage levels. Additionally, the model furnishes real-time key indicators essential for monitoring the CO2 system, fostering transparency, and facilitating proactive interventions in the event of any irregularities or anomalies.

Despite its utility, CUB’s previous CO2 modeling system was on an Excel spreadsheet integrated with VBA macros. These macros executed SQL queries to retrieve desired inputs, which were placed into the spreadsheet, where it performed calculations and produced a graph depicting current and forecasted CO2 levels.

However, several shortcomings plagued this approach. The model's algorithm, rooted in outdated methodologies of beer production and CO2 generation, introduced a notable margin of error, often deviating by 20-30 metric tons (equivalent to a 10% inaccuracy based on CUB’s storage capacity of 230 tons). Such deviations, compounded over a single day of production, could result in a substantial inaccuracy. Furthermore, this unreliability could be critical given a single day of production can draw down 40T of CO2, rendering the model unreliable. The reliance on Excel macros compromised operational efficiency, as the system operated sluggishly. The model also had "key indicators" that would be displayed, but they required manual inputs from its operators, negating the proactive capabilities of the model.

Overall, the previous CO2 model, which was maintained on an Excel spreadsheet, required constant operator intervention, undermining its effectiveness in preempting potential issues and often proved unreliable.

Solution:

Compared to Excel, Ignition allowed CUB to access a whole range of data not just on the SQL server but also straight from the PLC. Ignition’s memory tags and scripting system allowed for easy manipulation of this data and ultimately allowed CUB to achieve expectations of creating a CO2 model that had real-time key indicators, which is best shown by the Brew Plan page. The brew plan page combines real-time data of the CO2 collection system comparatively with what the CO2 algorithm predicts will happen in the CO2 collection system, allowing CUB to gauge abnormalities at a moment’s notice.

Using the Python scripting built into Ignition’s memory tags, CUB was able to improve the accuracy of the CO2 model. Utilizing memory datasets in Ignition, CUB can suit a dataset to each different recipe while using other sets of memory tags for global parameters all of these features, allowing for the CO2 model to be dynamic. The ability for memory tags to historically store data was also a massive feature that allowed the model to run for any given timeframe, displaying a more historical view and providing overall transparency of the model over time.

There are other features, such as events, that can be triggered from a tag change or “on change” event that can trigger a Python script to run, which opens up many capabilities for developing the model. Another handy feature during development was the ease of rolling back project versions and how an update from the Ignition designer could be pushed into the live environment almost instantly. These features fast-tracked development, allowing developers to quickly make changes and test for bugs and, if need be, rollback the project.

Results:

Accuracy in storage capacity (Hourly Basis) with Ignition CO2 model (Approx answer): 95%

Accuracy in storage capacity (Hourly Basis) with Excel CO2 model (Approx answer): 70%

Accuracy in predicting Brews with Ignition CO2 model (Approx answer): ± one hour

Accuracy in predicting Brews Excel CO2 model (Approx answer): ± 10 hours

Time to load current week with Ignition CO2 model (Approx answer): 15 seconds

Time to load current week with Excel CO2 model (Approx answer): 60 seconds

Live Features with Ignition CO2 model (Approx answer): 10+

Live Features with Excel CO2 model (Approx answer): None

Max Sessions with Ignition CO2 model (Approx answer): Unlimited

Max Sessions with Excel CO2 model (Approx answer): One

Historical Data with Ignition CO2 model (Approx answer): Unlimited

Historical Data with Excel CO2 model (Approx answer): 2 weeks

Debugging Tool with Ignition CO2 model (Approx answer): Yes

Debugging Tool with Excel CO2 model (Approx answer): No

Algorithm Variability with Ignition CO2 model (Approx answer): Yes

Algorithm Variability with Excel CO2 model (Approx answer): No

The Ignition CO2 model was better in every way than the Excel model. It had an accuracy of ~95%, the algorithm was variable instead of static, it updated faster, had many features that did not require operator input, and could be accessed on a webpage by anyone instead of an Excel spreadsheet that only one person can operate at a time.

There were many unforeseen advantages that came from the development of the Ignition CO2 model. CUB is able to utilize the CO2 model not only as a tool to predict if it needs to import or export CO2 but also as a tool to optimize CO2 collection efficiency. As the CO2 algorithm is stored as a Jython script that is easily accessed by components in Ignition, CUB can leverage this to provide insight into where the CO2 in the system is being lost and what amount. The CO2 model is also able to expose areas of weakness, as the model is only as good as the inputs. The CO2 model started to emphasize inaccurate meters affecting CO2 readings, allowing CUB to further improve its CO2 collection process.

Start Date: November 2022

Deploy Date: August 2023