Fast and Actionable Machine Analytics: A Chemical Processor Case Study

Brian Wolf has several years of manufacturing experience setting up data analytics for digital transformation for multiple companies across hundreds of facilities. Wolf is currently director of operations excellence at LSB Industries, a leading North American producer of industrial and agricultural chemicals including ammonia, urea ammonium nitrate (UAN), ammonium nitrate (AN), and nitric and sulfuric acid. The company has production facilities in El Dorado, Arkansas; Cherokee, Alabama; Prior Oklahoma; and Baytown, Texas. At the 27th Annual ARC Industry Forum, Wolf described how the rapid application of machine analytics got results in a short time for LSB Industries.

Digital transformation goals

Wolf explained, looking at the classic P (Predicted Failure) to F (Failure) curve, that the goal is to predict future failures with the application of analytics in a timely manner so as to take action before this impacts production.  He emphasized, “If we do all this work but don’t stop the failure, what’s the point? Getting the right information to the right person at the right time is absolutely key.”  

LSB Industries partnered with Atonix to capture information and “make plant data smart.” The goal is to alert people who can mitigate potential problems early, benefiting from predictive maintenance with higher production efficiency.

Each of LSB’s plants has 5,000 to 6,000 data points coming in real time. The data is used to improve operations by predicting, “What could potentially take down a facility. Where do you look in the data to find that needle in the haystack?” This process of prediction also requires getting data out of their silos, and facilitating collaborative teamwork for problem solving between departments.

Use existing data first

The company is first using data from installed systems at the site without adding new instruments.  Wolf made an important point: “If you have enough data to run your plant, you likely have enough data to do analytics and monitor processes.” Part of this process involved adding a company historian to aggregate data and getting it in one place, such as a PI historian. That gathering included what Wolf terms “stranded data”—data  stuck in PLCs, logbooks, file cabinets, and as part of tribal knowledge, including data stuck in people’s heads.

Making use of the existing process data first and foremost yielded results with a rapid deployment that required little effort, Wolf said.

Using methods including advanced pattern recognition and neural net math to predict failures, the system reports specific information so appropriate people can act on it. Wolf summarized the process:

1. Make use of existing process data
2. Practice rapid deployment with minimal effort and expertise
3. Use software that drives the process, not just analytics.

Thousands of measured data points come in in near real time and filtering all the noise out of data to get to actions happens for a couple of things a week. LSB applied Atonix to build algorithms in the models that could  drive success with just existing data at each one of the company’s facilities. 

No-Code algorithm development

The Atonix method of no-code algorithm development and model configuration makes complex math simple for asset experts.  The tool has thousands of templates that are built for real-world industry applications. Process engineers  can build a template without hiring data scientists.  The tool utilizes the intelligence of engineers coupled with existing model templates and machine learning functions to create applications with significant ROI.

Workflow

The Atonix tool drives the process of resolving, prioritizing, and managing issues through their life cycle.  Working on the right issues at the right time is important. “Just because you discover something doesn’t mean the process needs to be shut down right now. But mediation needs to be added to the planned workflow,” said Wolf.

Getting results

Wolf described various examples based on implementation throughout the company. “We were able to roll this out across our entire facility, fleet wide, in three months across everything,” he said. “Each site took about a month get online. The  big part of that that takes longer than even the deployment is the change of management philosophy at the facilities. What do we do with the information now changes and improves the way we work.”   

Wolf also described the importance of engaging site people and integrating with software that drives the process, as opposed to just doing analytics.