Super Smart Team Intelligence

 

Thanks to the intensive collaboration between data scientists and process experts in the research and production environment, we are increasingly succeeding in leveraging our existing wealth of data. This has reduced our energy consumption and sped up the pace of product innovations. It is made possible by artificial intelligence methods in conjunction with data platforms such as DIAP and visual tools such as Power BI..

 

Project 1: Faster Results

 

For the data scientists, it’s all about the perfect formulation. And because they aren’t poets formulating verse, for them this means finding a formulation for a product that will ultimately delight the customer. In this context, the customer can be from MPP, PLA, LAB, or another BU. “We’re the right ones to get in touch with if you’re looking to improve the features of a product,” says Lynn Ferres from the Data Science & AI team, or DSAI for short, which is part of the IT Group Function.

The DSAI team has developed an approach that makes it possible to generate promising suggestions for optimized product variants faster using machine learning support. They have successfully applied this approach for various business units. For example, the PLA BU had already developed an excellent flame retardant. Now they wanted to test this with different ingredients in different quantities in order to find an optimized version of a plastic. With the help of an AI model, they analyzed the complex multivariable inputs and outputs. This involved a continuous, iterative process, known as sequential learning, which supplied various formulations. “Our flame retardant experts He Qingliang and Christopher Simpson then decided which of these suggestions should be tested further,” explains Ferres. Thanks to their expertise, they were able to easily evaluate the AI suggestions. “This resulted in two candidates that exhibited excellent characteristics in all of the target variables,” says Ferres. The “chosen ones” then got to leave the microcosm of the laboratory and were produced on the scale of a few kilograms. The material is now being tested for its tensile and bending strength and elongation, as well as its flame retardant properties. But even if the candidates perform extremely well in the tests, the customer may still reject them.  “When they are too good, they are often also too expensive. Our job is to match the product with the customer’s wishes and applications,” Ferres explains matter-of-factly. After all, the company wants to generate revenue with its innovations – that is the primary goal. 

Hanna Kahlfeld, also a data scientist on the DSAI team, recently completed a project with the MPP BU. “Colleagues come to us when they can’t make any further progress in the lab. The cases are always tricky,” she says. With the help of the sequential learning approach, however, they quickly found a formulation that fulfilled the specified requirements. “This was further confirmation for us that our approach works,” she says. Others were similarly convinced, and so AI-supported formulation development was made a fixed part of Innovation Excellence. This is because it has ultimately shown that with this approach, LANXESS is capable of developing new and improved products at a faster rate. Kahlfeld now invites everyone to contact her and her team if they are facing similar challenges.

 

Projekt 2: Thanks to AI and Power BI: Standard Met and Production Knowledge Increased

 

Data scientist Marcel Dembek approaches his job differently, but also with the help of AI. Four years ago, when the requirements for the ISO 50001 standard were increased, Dembek was faced with a new challenge. The standard stipulates that companies must carry out an energy assessment that proves that energy use has continuously improved compared to a baseline level. This assessment must be verifiable and is reviewed by certification bodies.

Together with Stefan Geißler, Senior Officer Energy Management, PTSE GF, and the sites in the field of energy efficiency monitoring, Dembek developed models for this. The sticking point was the initial data. “It is influenced by many different factors,” says Dembek.  “With the help of DIAP, Power BI, and the use of artificial intelligence, we ultimately succeeded in defining it. We have now created over 3,000 baseline data points,” he explains, pointing to a poster. “Here we see the modeled curve for the energy consumption of a system that our AI has calculated from the baseline. And here we can see the system’s actual energy consumption. Actual energy consumption was lower than the calculated estimate.” That’s a win. But is this credible proof that energy consumption is actually lower? “If we assume that, together with the process experts, we were able to identify all the variables relevant to energy consumption, such as temperature or production quantities, then yes.” But what if the baseline year was mild and the year being calculated was extremely cold? Is the data still comparable in this case? “Yes,” says Dembek. “Our models take precisely this into account and adapt the calculation to the different conditions. The result is put into context to enable reliable comparisons.” Particularly in the current challenging times with reduced production quantities, the AI provides a fair assessment of energy performance. What employees previously had to tediously compile and prepare in Excel spreadsheets now comes together automatically in a Power BI dashboard via DIAP. “In the meantime, we have created models for 20 sites together with the process experts,” says Dembek. Some of the process experts were amazed at how significantly the outside temperature, for example, influences the amount of energy used to heat buildings and run production processes. This influence can now be precisely quantified. The tool therefore not only helps us meet the ISO standard and pass its audits, it also gives the experts deeper insights into the processes. And this is precisely what is often the first step towards implementing further energy-saving measures ...

 

Project 3: Efficient Product Yield Thanks to AI

 

Another project is being led by data scientists Hessam Ramezani and Rasit Faller under the title “AI-Driven Production Optimization.” As part of FORWARD!, the production process for the raw material NaMBT in Kallo was also put to the test. This raw material is required for the production of intermediates by the Rhein Chemie BU. In an initial analysis, the data scientists, together with the process experts, determined that a significant percentage of the expensive raw material aniline doesn’t end up in the target product during production. “It is bound in the form of resins in an unwanted byproduct and ultimately burned up,” explains Faller.In light of the fact that the raw material aniline is the most expensive component in the manufacturing process, this was simply unacceptable. “With an AI-assisted model that we built using DIAP, we have now been able to optimize the production process so that we lose a third less aniline,” says Ramezani. But that’s not all: “We’re now trying to find further process parameters with the help of AI in order to get even more aniline into the end product,” Ramezani explains.

 

Project 4: Cockpit is the Central Access

 

A further pilot project resulted from the production optimization project in Kallo. The data scientists Faller and Ramezani developed an innovative production dashboard together with process experts in Kallo and colleagues from PTSE GF. They implemented the cockpit in Microsoft Power BI. It enables users to dynamically and precisely monitor a system’s production status and energy consumption. “As a result, we provide a comprehensive and seamless overview of all key production and energy figures via a central access point,” explains Faller.

The production cockpit is already being actively used at the Kallo site and is being continuously expanded to further improve efficiency and decision-making. 

 

New Terms, Simply Explained

Machine learning: A model is trained with known input data to generate predictions for new, unknown data points, and the model learns to recognize correlations and patterns in the input data.

Sequential learning: In an iterative (i.e., repetitive) process, a machine learning model is trained and used to predict unknown data. The most promising predictions are then tested. In the next sequential learning step, the model is then additionally trained using the test results from the previous iteration. This helps the model better understand relationships and patterns so that the model’s predictive power improves with each iteration.

Data Integration & Analytics Platform (DIAP): This platform developed by the DSAI team makes it possible to combine, analyze, and visualize data from all LANXESS systems as well as from a variety of external systems. Within the framework of projects in the production environment, the platform is used to improve processes, identify bottlenecks, predict maintenance requirements, increase production output, and improve overall efficiency.

Power BI: This term refers to a collection of software services, apps, and interfaces that make it possible to link, import, and visualize data from different sources. Power BI can be used to access reports not only locally on an individual PC, but also via a browser from anywhere in the world. Using Mobile Power BI apps, reports and dashboards can also be accessed via smartphones. This applies to Windows, iOS, and Android devices.