What happens when you combine a 123-year-old company, a 172-year-old product, and today's bleeding-edge cloud technology? For The Hershey Company, the recipe was nothing short of delicious. The Pennsylvania-based maker of Twizzlers licorice was struggling to produce a consistent flow of red candy as it poured from a giant vat through an extruder (where it would eventually be cut into its spiral design).
Continue Reading Below
If the machine ran too cold, then not enough licorice was produced. If the machine ran too hot, then too much licorice was produced. When too much licorice was produced, Hershey had to give that candy away at a loss. We're talking tons and tons of free candy. For reference: When the extruder ran too hot, it would extrude 100 extra grams of licorice per minute. Great for us! Bad for Hershey. To maintain a steady outpour, Hershey assigned operators to make manual adjustments throughout the day to ensure the machine never ran too hot or too cold. This proved to be a minimally sufficient method for maintaining stable output; Hershey was determined to do even better.
The company implemented Microsoft Azure to calculate, measure, and monitor the temperature of the extruder. For those of you who aren't familiar with Microsoft Azure, it's a collection of IT applications housed on the Microsoft cloud. Microsoft Azure offers tools for e-commerce, business intelligence (BI), and even fun stuff such as game development. By adding sensors to its industrial equipment and feeding the data back to Microsoft, Hershey could develop a better understanding of when temperature deviations would occur.
Hershey wanted to predict what the weight of the licorice was going to be without having to stop the machine and weigh it, George S. Lenhart III, Senior Manager of IS Disruptive Solutions and IoT at The Hershey Company, told me. By using sensors built into the extruder, which fed data back to Microsoft Azure, Hershey was able to monitor exactly how long it took for the machine to cool down to an exact temperature—and how long it took for the machine to heat up again.
Once the standard deviation had been assessed, and solid predictions for when the machine would exceed satisfactory highs and lows had been made, Hershey programmed Microsoft Azure to control the extruder directly. Using Microsoft Azure's machine learning (ML) algorithms, which gave the manufacturing equipment autonomous control over raising and lowering the machine's temperature, the extruder was adjusted 240 times a day—without the help of a human operator.
Continue Reading Below
After the Microsoft Azure implementation, Hershey lowered unwanted extrusions from 100 grams per minute to about 25 grams per minute. Plus, the overall minutes were dramatically reduced (although Lenhart didn't provide specifics on how much time was reduced and how much money was actually saved).
The first experiment—monitoring the temperature highs and lows—took Hershey and Microsoft about a week to complete. Perfecting the experiment to automatically control the temperature adjustments took multiple iterations over the course of six months. Testing the cloud's security, which allowed the extruder to send data back and forth to Microsoft Azure, took about 10 months.
Hershey is now two years into the project. Lenhart said newer implementations for other Microsoft Azure clients will likely be much faster thanks to improvements to Microsoft Azure technology, as well as to newer back-end smart machinery, which has matured since 2015. "I would imagine we'd have anybody up and running in three to four months," Lenhart said, "and then take probably four to five more months to perfect it. And when I say perfect, I mean really have it all hammered out."
When asked if he thought Hershey and Azure could ever get to the point where no licorice is wasted during the extruding process, Lenhart shook his head. "We're just never going to get to zero," he said.
"So, there is a game in manufacturing called variability," he continued. "We want to reduce the variability. That game is always played. But, if you take the difference there, multiply it by the number of minutes by the number of hours by the number of extruders, you're talking about [saving] tons of licorice." Darn!