As technology continues to advance, so do its abilities to improve the process. At Smart Industry 2018, which took place Sept. 24-24 at the Loews Chicago-O’Hare in Rosemont, Illinois, Tim Alexander, brewery operations technology manager at Deschutes Brewery, explained how the Bend, Ore.-based craft brewer is improving its process with predictive analytics technology in a breakout session titled “Deschutes Boosts Quality, Throughput with Predictive Analytics.”
Once a company that thrived on a single 50-barrel brewpub, Deschutes has grown, now distributing its beer to more than 30 states and shipped 336,000 barrels in 2017, explained Alexander. However, this growth brought challenges, and maintaining quality and consistency of its brands while optimizing the process are key among them.
Thus, Deschutes began working with OSIsoft and Microsoft Azure to create a combination of process automation, data historian and cloud-based analytics technology that can help predict the path of its fermentation processes.
The fermentation process includes several transitions, which must take place at the opportune times to ensure that the yeast and other elements have reached the desired levels. The brewers identify the transitions by measuring the specific gravity of the tank, Alexander explained. Thus far in the project Deschutes focused on two transitions, the transition from free rise to diacetyl rest and the transition from diacetyl rest to cooling.
Data for a transition is collected via manual measurements and entered into the OSIsoft PI System. That data is then sent to a PI Integrator for Microsoft Azure, then to the Azure Data Factory, where it is analyzed and sent to another SQL data warehouse for retrieval by Deschutes.
The analyzed data for each tank is compared to the cataloged data from previous batches of that brand of Deschutes beer (ie. Fresh Squeezed IPA, Mirror Pond Pale Ale, etc.). This data is then translated into a graph depicting curves for past and predicted future data of both the current and previous batches of that brand, including actual measurements and transition targets.
Using the data brewers can predict when fermentation has reached the desired set-point to move the process on potentially sooner than previously thought, Alexander explained.
“We use the actual batch prediction to calculate the hours to free rise. So, at that cross-point, we can say, OK we’ve got 15 hours until free rise. The key is we want to hit that free-rise point,” Alexander said. “This can help the brewers decide, oh, I’m going to take another gravity at this point, and as we move toward more and more confidence in this model, we’re starting to say OK, let’s just move the tank on at this time right here, at 15 hours, and then maybe we’ll check the gravity again and make sure it didn’t pass that point on the next check. Maybe we don’t even need a specific gravity check at that time before we move the tank over.”
The brewery has found the predictions to be fairly accurate, he added, noting that often the percent-error lies within the 1%-or-less range as more data is accrued to aid the predictions.
Key to the success of this implementation was building trust with the Deschutes brewers to help them buy in to the program. Alexander emphasized that building trust between the brewers and the predictive analytics was essential.
“We didn’t throw this in the first week and say, alright, we’re going to move all our tanks based on exactly what this says. At first, it was just kind of showing to the department and saying, hey, here’s another tool you guys can use, and also let us know if it is not making any sense, because we’d like to see those cases so we can fix it. And that’s the key, it’s just another tool to help make better beer more efficiently,” he said. “It’s been mentioned a couple times [at Smart Industry], but we’re not trying to eliminate brewer jobs with this, really, we’re just trying to help them make better decisions. And … when you’re taking specific gravity measurements on tanks, it’s not a job anybody wants to do, so if you can take fewer gravities and also make your transitions more accurate, everybody wins.”
Could all this be for nothing? Of course not. Alexander outlined several benefits the brewery has seen.
- Increased quality,
- Decreased process time,
- Fewer manual measurements,
- A 4% decrease in total fermentation time, and
- A 2% decrease in diacetyl rest time
Now two years in to the project, the brewery continues to review and improve the process, while also seeking new opportunities for implementation.
“One of the next steps is cooling and maturation,” Alexander explained. “So, there is an automatic transition, so we never miss it. But it’s also an important transition, because if cooling is not functioning correctly … you can lose a lot of time.”
Other areas where predictive analytics could prove beneficial are preventive maintenance, lautering logic, gas chromatograph mass spectrometer and even sales data, Alexander said.
The power of data analytics implemented in ways that significantly improve processes can certainly be amazing. Let’s raise a glass to better beer through data and predictive analytics.
