Artificial Intelligence has earned a real place at many modern ethanol plants thanks to engineers like Kelly Whittenberg. “Your plant is not only producing ethanol every day, but also a river of data,” says Whittenberg, a senior asset management engineer at Black & Veatch Corp. Whittenberg has worked in several large industries—including ethanol—to create AI-assisted business intelligence dashboards that enhance plant performance and efficiency while also helping prevent unplanned outages.
For the ethanol sector, Whittenberg has linked the capabilities of fast-acting AI with the engineering and monitoring ability of Black & Veatch. Pairing AI with the company’s core capabilities is enabling ethanol plants to detect, diagnose and resolve issues of all kinds faster than ever. The use of the AI-infused system lags only 15 minutes behind real-time operations and can monitor everything from bearing temps in decanter centrifuges to evaporator cleaning effectiveness.
“We are looking at a much broader spectrum of data than the plants have the capacity to look at,” Whittenberg says.
From the framework of a Microsoft Power BI (business intelligence) platform designed for biorefineries, Whittenberg and his team are now able to provide ethanol producers with insight captured on a dashboard customized in part with the help of ethanol plant managers. Anyone with access can read the information and get alerts from anywhere with an internet connection. Historically, solutions like this would be software purchased and installed on a PC at a facility, often with a “seat” license for each user. Because BI dashboards are applications running in a browser, Whittenberg says, there is no software to download and the dashboard can be shared across an organization, if desired.
The dashboards don’t replace primary control systems at a plant, but they do, however, monitor how the plant should be running versus how the plant is actually running. The combination of thousands of models and operational inputs allows the AI to create an expected outcome, which it displays via line graph or other means on the dashboard. Imagine two horizontal lines on your screen. One is red, the other blue. The red line is what the AI predicts for the plant. The blue line is how the plant is actually operating. When the two lines are matched, the plant is running as it should. When the blue line (actual operating conditions) is above or below the red line (AI prediction), the plant isn’t running as it should. The Black & Veatch system monitors what is actually happening at the plant, compares the actual to the AI-model and provides alerts or insight to the operators. Of course, as Whittenberg says, that is the simplified explanation of what is really happening. As his full explanation of the Power BI dashboard shows, there is much more to know about the system and its benefits.
Building AI-Driven BI Dashboards
To understand how the Power BI dashboards are built, you first must know how much information Whittenberg and his team look at. To monitor most ethanol plants, the team considers the balance of a biorefinery, which roughly includes 220 plant assets, 3,000 data points, more than 460 models (of how the plant could be operating given certain variables), 116 operating modes (full capacity, during winter, etc.) and overall performance metrics.
The dashboards for ethanol plants can be customized, but typically include mash and fermenter information, the beer column feed, total sieve feed rates, backsets, sieve efficiencies, CORT feed densities, 190 proof levels and more. The systems show ranges and data over the last seven days. Doing so helps the plant managers or engineers monitoring the system detect issues as trends unfold. If a section of the dashboard shows stability for five days straight, but then shows a change during the last two days, there is reason to believe something should be looked at, either through the data or physical inspection. The dashboards display information in three ways: normal, normal violation and limit violations. The normal range means the plant is operating within its expected range. The normal violation shows that a plant is still operating in its expected range but certain variables are just outside of that given range. In most cases, Whittenberg says, the normal violation display happens because the ranges an operating model are set at—temperatures, for example—are fairly tight. Then, there is the limit violation display. In those instances, the actual operation of the plant is operating well outside of what the AI predicted it should be.
To customize a dashboard, the team builds a hierarchy of the information a plant would want to know. Then, they backfill or search for data that they can collect linked to the main areas of operation that a plant wants to hone in on. After that, the team will create operating modes for a wide range of operating conditions and other factors that would constitute an operating model. And, finally, they use AI to “train” the models to create predictive modeling of how a plant should operate given particular operating modes.
“It takes about three months to put together,” Whittenberg says. “[But when it is complete], everything someone needs to know about an ethanol plant can be easily covered [via dashboard] in the time it takes to drink a cup of coffee.”
The team has to wade through 10,000-plus measured parameters in a plant control system during the initial configuration process of AI setup. The more information the AI gets, the more it can give.
“There is an actual and an expected mode of operation at a plant. Our system can call out when the plant is not actually operating as expected,” Whittenberg says.
Measured inputs might include temps on bearings, oil pressure readings or airflow sensors for nearly any part of the plant. The AI prediction modeling will pump out hundreds of models, each with expected values or trend lines of how certain actions should stay on in the future. But Whittenberg says plant managers shouldn’t fear that such a robust system will flood them with data and insights they might not know how to act on or interpret. “In our experience,” he says of working with several industries that run various types of plants, “they have neither the tools or bandwidth to monitor the balance of plant data in an efficient and consistent manner [outside the structure of the AI-powered system].”
After the AI prediction modeling is complete, an alert screening system is set up to monitor roughly 10 to 50 alerts per day that are generated by the AI system when the actual operating expectations of the plant fall outside of an operating value predetermined by AI. A team of engineers can look at the alerts, flag any that are truly worth looking at and then contact the plants directly with one to two plant escalation events per day, week or as frequently as necessary. Sometimes the escalation calls will be low-stress—“you might want to just check this,” sort of conversations. Others will be more urgent. And sometimes, the escalation calls will reveal an issue the plant is having, or may soon have, that operators at the facility would have never been able to see without the power of the AI system.
The Reality of AI Monitoring: Case Study Takeaways
Whittenberg knows it’s hard to see what AI-powered monitoring can actually bring to a plant. That’s why he’s quick to provide several real-life examples of how the system has benefited his ethanol clients. “The case studies are intended to show how AI tools can bring attention to concerning changes that a plant may not be aware of,” he says. “With AI tools, the plant can detect issues early and act before there is loss of production or loss of equipment.”
In one instance, he explains, the dashboard and AI-monitoring system showed a quick rise in a bearing temp. That was an easy one that required the plant to be notified and a bearing gasket to be fixed. From that, he says, “No trip or loss of equipment happened due to a failed bearing.”
In another instance, the dashboard’s expected-versus-actual operating lines revealed an issue with the lube oil cooling controls for a corn oil centrifuge. After noticing a difference in lube oil temperature ranges in its system and alerting the plant, the facility was able to reduce the amount of cooling water needed for its centrifuge. A lube oil temp control solenoid had failed and was causing an irregular amount of water to be used.
Sometimes the dashboard can use fermenter tank level changes to help show how different yeasts impact fermentations. In one example, Whittenberg says they were able to see how cream yeasts were performing differently than box yeasts.
“A lot of plants want help with their evaporators,” he says. “They have to clean the equipment but they don’t always know which one to clean except maybe which evaporator was cleaned the last time.”
The Power BI dashboard Whittenberg created for a client asking that very question was able to show four evaporators over the course of a month. By monitoring condensate flow and predicting what the condensate flow should be, they were able to see trends in how the flow dropped, or could be expected to, in each evaporator. That knowledge informed the plant on which evaporator should be cleaned.
Adding AI To Ethanol Production
Some of the plants Whittenberg’s team works with like to see numbers. They want to see the value brought in by AI in terms of dollars. Others are more relaxed and see the value without monetary correlation, he says. To explain the cost of the system, Whittenberg offers this: The ROI is many times greater than the cost of service.
In most cases, the team will rely on the existing monitoring and data feeds already available at a plant. The team will offer suggestions to plants if they believe more monitoring capabilities are warranted or needed. Most plants are monitored five days a week, 24-hours-a-day. If a plant has the bandwidth, the in-house team will be able to monitor the dashboard and take any necessary action. Most successful organizations assign people to the monitoring and dashboard tasks, Whittenberg says. Certain facilities will outsource that responsibility to a third party.
The number one reason these types of systems fail, he says, is because the original project champion gets promoted or moves on. “That’s why it’s important to have a succession plan in place and [commit to] broad training on the tools,” he says.
AI and dashboards hold a lot of promise for the ethanol industry, Whittenberg also says. But achieving success will take a combination of people, processes and the right tools to get there.
“In the end,” he says, “AI dashboards help users get to the right answer more quickly and accurately.”
