Deschutes Brewery: better data for better beer
- ChallengeFrequent changes to production lines were creating unnecessary downtime, cash flow issues, and excess inventory problems.
- SolutionUse the PI System as the backbone of a phased approach to gaining deeper insight into assets, inventory, and production processes.
- BenefitsCost savings from identifying $10 million in excess inventory and $500,000 per month in process improvements; growing predictive ability.
At OSIsoft’s 2016 Users Conferences in San Francisco and Berlin, Brian Faivre (Brewmaster of Operations) and Tim Alexander (Assistant Brewmaster) explained how the PI System has helped Deschutes:
• Reduce production time by up to 72 hours for each batch
• Maximize capacity and postpone $8 million in capital upgrades
• Understand the brewing personality of each of its beers through machine learning
Maximizing production with better temperature control
In 2015, Deschutes Brewery, the seventh-largest craft brewery in the United States, had a problem. During the cooling phase of the brewing process, temperatures were spiking in a new class of fermenters, an anomaly that could potentially reduce capacity potential and affect the quality and flavor of their beers. Luckily, Deschutes had a secret weapon: the PI System.
The new class of fermenters was large, holding 31,000 gallons and taking seven brew batches to fill. Loading just one of these batches into the fermenter is a two-hour process, and the fermenters had three temperature gauges—one just above the bottom cone, one in the middle, and one at the top—to control three independent temperature zones.
During the cooling phase of brewing, the brewers have to decrease the temperature from 60°F to just below 30°. As the temperature drops, yeast cakes at the bottom of the fermenter. Brewers often pull this yeast to reuse in future batches. However, when Deschutes started to pull yeast from the new fermenters, the temperature spiked in the top temperature zone. This spike increased the cooling time for each fermenter, which increased overall brewing time and reduced capacity. Deschutes contemplated capital upgrades. First, however, the company decided to see if it could use the operational data collected by the PI System to shorten the cooling phase while maintaining beer quality.
Deschutes took an iterative approach to the problem, making a change and then reading the data to see how it affected cooling. The brewers started by adding an air valve to the top of the tank and adjusting the piping that the cooling agent (glycol) passed through. These mechanical changes improved the cooling time, but the topzone temperature was still spiking. Deschutes needed more insight. The brewers added a fourth temperature sensor to the bottom cone of the fermenter, an area that had previously not been monitored. Deschutes used this fourth sensor to control the cooling in the bottom cone of the fermenters. With this increased insight and control, the company eliminated the unwanted temperature spike. As Faivre explained, the company was “seeing consistent and repeatable fermentation cooling times with a time savings of about 60% in some cases… We [we]re able to put off the unnecessary $8 million capital project… while assuring the highest quality in all our brands.”
Machine learning and predictive analytics for fermentation
Craft breweries release new beers frequently to maintain customer interest. The fermentation process for each of these beers is different. That process can be separated into nine distinct phases, and different beers transition from one phase to the next at different times. Typically, at Deschutes, it required regular manual readings and analysis to know when to move a particular beer from one phase to the next.
Deschutes wondered if it could use PI System data and machine learning to predict when transitions occurred to minimize the need for manual readings. In 2016, Deschutes joined the exclusive, invitation-only OSIsoft | Microsoft Red Carpet Incubation Program (RCIP) to explore advanced analytics to optimize operations.
Within a few weeks Deschutes built out a PI System Asset Framework for all 50 fermenters, which gave context to the sensor data. The company then implemented the PI Integrator for Microsoft Azure to automate the preparation of PI System data, context, and events into a format that could be fed into Microsoft’s Cortana Intelligence Suite.
Saving Time by Smoothing Transitions
Deschutes focused on predicting one phase transition – from fermentation to free rise – for its different beers. For the transition from fermentation to free rise, the company tracked the apparent degree of fermentation (ADF), or the percentage of the beer that had fermented. Using machine learning to update predictions after each manual reading, the company was able to effectively predict the shift from fermentation to free rise for any beer. The predicted trends were then sent back to the PI System, which allowed users to consume the information within their existing PI Visualization tools.
As a result, Alexander said, “within 24 hours of the start of fermentation, we can have a pretty accurate prediction of where it’s going to end up.” He added that these predictions “not only can save time in just moving to the next step in fermentation. You can actually save time in the future steps of fermentation because those steps go more smoothly if you transition out of this step at the right time.”
The result: operationalizing the predictions for when different beers move from fermentation to free rise saved Deschutes up to 72 hours of production time for each fermenter.
In the future, Deschutes wants to fully automate predictions for every beer and every phase transition, or as Alexander put it, “get to the point where the system, is just saying, ‘Let’s move to the next step. It’s time.’”
For more information about Deschutes Brewery and the PI System, watch the full presentation here.