U of I helps Idaho company with employee safety

For nearly 30 years, the Amalgamated Sugar management team searched for a workable way to automate their steam dryer.

After listening to their team describe what they hoped to see in a solution, U of I graduate student Hunter Hawkins came up with an idea while waiting for a delayed flight.

Using a combination of AI algorithms to monitor the plant’s steam dryer, Hawkins and John Shovic, director of the Center for Intelligent Industrial Robotics at U of I Coeur d’Alene, developed a system to warn the company of potential problems, which will allow the Amalgamated team to improve employee safety while keeping their production line moving.

“When we have a plugging event, we have to shut down the dryer and send employees in to unplug it,” said Trent Holcomb, plant manager at Amalgamated Sugar. “Because of the heat, pressure and moisture content inside the dryer, it can become very flammable when we open it up. Not having to send our people in there to unplug it is one of the biggest safety opportunities we have.”

Sweet success

For more than a century, Amalgamated Sugar has operated in Idaho, becoming a staple of the Treasure Valley after building its Nampa facility in 1942. The company processes more than 7 million tons of sugar beets each year at its facilities in Nampa and Paul.

During peak season, the Nampa facility produces 4.4 million pounds of granulated sugar a day and 87,000 tons of dried-pulp cattle feed a year. With local and regional farmers bringing in sugar beets around the clock, the company doesn't want to have to halt production because of a plugged steam dryer.

The process of producing granulated sugar from sugar beets is relatively straightforward — as truckloads of beets arrive at the factory, the root vegetables are cleaned and sliced into large strips that resemble French fries. The strips are then immersed into hot water, which allows sugar to diffuse from the pulp.

After diffusion, the pulp — still at about 72% moisture — moves into the steam dryer. Ninety seconds in the dryer at nearly 450 degrees reduces the moisture to around 10%, leaving a dry product that’s sold as shredded or pelletized cattle feed.

The process of drying pulp often leads to plugging issues. Excess pulp, incorrect temperature or steam pressure, or a mass that sticks and begins to burn can all cause problems. For the Amalgamated team, pinpointing the specific issue and exact location of a problem in the dryer required shutting it down and sending employees inside to investigate — until Hawkins and Shovic devised a system that helps the Amalgamated team anticipate issues before they arise.

“The steam dryer has always been on my radar as something we needed to automate,” Holcomb said. “It’s a fairly self-contained system with a lot of data points. Looking at an AI solution seemed like a great fit because there are about 100 different data points in the dryer and we have years’ worth of data we’ve already collected on it.”

Intelligent solutions

As a busy doctoral candidate, Hawkins had no time to waste. When he and Shovic faced a long wait for their flight after meeting the Amalgamated team in Nampa last year, he pulled out his computer and began working on the problem.

Specializing in computer science and industrial automation, Hawkins agreed with the Amalgamated team that an AI-driven solution was feasible. The challenge lay in choosing the best AI algorithm to use. The idea Hawkins pitched to Shovic was novel to them both.

“I actually combined two models for this project,” Hawkins said. “I wanted to combine the prediction power of a deep-learning AI model with the explainability of a traditional machine-learning model. The result is a hybrid model that drives a dashboard, giving the Amalgamated team predictions and alerts to potential issues.”

The deep learning model, referred to as an LSTM (Long Short-Term Memory), takes in different data points throughout the steam dryer on a per minute historical basis, such as the steam input from two boilers, motor amps and other supporting systems, to predict what the data points will be five to 20 minutes in the future. The future predicted data points are then given to a machine learning decision tree ensemble to determine the probability of a problem.

The AIPA (Artificial Intelligence Predictive Appliance), built by Hawkins, is a dashboard that contains the hybrid model, as well as other predictive models. It is designed to detect patterns and anomalies in the data that indicate potential problems. This technology also allows the data to be displayed in a way that notifies the end user when, and more importantly where, an issue might take place.

“Our operators will be sitting at their control stations and this PC will be sitting next to the station and it will tell us very quickly and very clearly when and where it detects a problem,” said Scott Hyer ‘10, operations manager at Amalgamated Sugar. “We can have the steam dryer restarted in less than an hour if we need to stop it prior to a plugging event.”

This project has sparked interest at Amalgamated Sugar in expanding AI modeling to make future processes more efficient. As they look ahead, Hyer and the rest of the management team are grateful for their partnership with U of I.

“We feel very lucky to be involved with U of I,” he said. “They’ve come down to us multiple times and that has helped us gain a better understanding of what they’ve put together and how the process works. Having a local collaboration is very beneficial for both Amalgamated and U of I.”