BASF, the world’s largest chemical producer, has worked with chemical engineers at LSU to better understand and predict its own production ebbs and flows using artificial intelligence, or AI.
The project adds to an ongoing partnership between LSU and BASF to develop emerging STEM talent across all disciplines in Louisiana.
BASF’s chemical manufacturing facility at Geismar in Ascension Parish is one of the company’s six largest integrated production sites in 80 countries. It supplies products to a wide variety of industries, including agriculture, construction, energy and healthcare. Chemicals such as solvents, amines, resins, glues, electronic grade chemicals, industrial gases, basic petrochemicals and inorganic chemicals are produced at Geismar in approximately 30 interconnected production units, each containing its own subunits.
“Chemical manufacturing is complex,” said Kerr Wall, digitization manager in BASF’s monomers division and alumnus of Cain’s chemical engineering department (LSU 1999). “Operating conditions can change from minute to minute and there is a lot of data to process. Big data offers a great opportunity to optimize our processes and become more predictive to improve our yields and our use of public services. This will make us more energy efficient and support our global value of producing chemicals for a sustainable future.
BASF is working with LSU to develop better data mining processes to organize its data and more easily compare current operating conditions with historical data.
Wall and his colleague Eric Dixon, also a digitization and production engineer in BASF’s intermediates division since graduating from LSU in 2008, had read research done by LSU professor José Romagnoli on optimization and control of complex systems, in particular using AI and machine learning to derive new knowledge from heterogeneous data.
“It just made sense for us to collaborate with LSU…José [Romagnoli] has over 500 publications and a lifetime of experience with process control, machine learning, and other areas that would be beneficial for us to understand. We try to automatically determine the most important process and quality parameters to help us relate current data to historical data and then use that data as predictive tools to enable process optimization,” said Wall, who is also visiting professor at LSU with a Ph.D. in bioinformatics.
Part of the project’s goal is to deliver optimized and automated workflows, and to develop what are called soft sensors, a machine learning term often used in manufacturing.
“Soft sensors are when data alone can tell you the real-time quality parameters of your material, for example, without you having to analyze lab samples throughout the day,” Wall said. “Soft sensors help estimate a particular variable, at any time. It can also help predict the ultimate quality of what we produce.
Physical sensors are often impractical or impossible to use in the extreme operating conditions of a chemical manufacturing plant.
“Instead of waiting 12 hours for a lab sample or for the next shift to take a new sample, LSU can help us find a way to predict what’s happening inside our units, only on database and AI,” Wall said.
LSU researchers are using an unsupervised clustering approach to help BASF categorize and label their production data. Time is a key parameter, as one of the main objectives of the project is to discover how a change in one production unit can force different operating conditions in other connected units.
“We can use throughput; the material that goes in and out of a factory,” Dixon said. “If a plant is operating at 50% capacity and a sister plant closes, the feed plant may have to cut rates by 20% until everything is sorted out. Understanding how and when one event leads to another allows us to make better decisions as events occur and evolve.
“Unsupervised machine learning allows us to capture the intrinsic behavior of processes and discover things we weren’t even necessarily looking for,” said Jose Romagnoli, LSU Cain endowed chair and professor of chemical engineering. “With machine learning and AI, we have a better opportunity to optimize production.”
His colleague, assistant professor Xun Tang of LSU Cain’s chemical engineering department, is also involved in the project.
An unsupervised learning approach allows LSU’s machine learning to “think outside the box” and uncover production patterns and correlations that neither BASF nor LSU’s chemical engineers would think to look for. Knowing how a change in one unit can force a change in other connected units helps optimize operations in large integrated plants, like BASF’s plant in Geismar, south of Baton Rouge.
“BASF has plenty of production data, but understanding the underlying dynamics can be difficult,” Tang said. “Thanks to this collaborative project, we can learn directly from the data in order to identify patterns. Then we can use what we have learned to predict new conditions to optimize and maintain system operation. In this way, we can help BASF automate and optimize their plant, and also improve their product yield and quality while reducing costs. »
Prior to this collaboration with BASF, Romagnoli and his team had completed a similar project with oil and gas company ExxonMobil. The first LSU student to earn a Ph.D. and work on this project was Gregory Robertson, who now leads the Applications Engineers of ExxonMobil’s Automation and Innovation Section in Baton Rouge.
“It takes a significant amount of knowledge to develop data-driven techniques for fault detection and diagnosis,” Robertson said. “In most cases, you have sensors on key variables to protect you from abnormal events. But when a faulty condition is difficult to define, the data-driven techniques LSU develops are a useful tool in your toolbox.
BASF’s longstanding partnership with LSU encompasses a multitude of areas, including career and recruitment programs for students through scholarships, internships, mentoring, job shadowing, senior projects, etc. and inclusion programs in STEM fields to engage women, minorities and veterans, and more.
#BASF #enlists #LSU #optimize #operations #artificial #intelligence