10/06/2020 | Research meets practice

Well advised with Artificial Intelligence

The KEEN research projects aims to develop AI-based applications for the process industry

Can plant operators, engineers and process developers of chemical production plants be supported by self-learning systems? Within the research project KEEN, various AI-based applications for the process industry are developed.

Planning and operating chemical and biotechnological processes is becoming increasingly complex. Humans need support for their decision making, and they get it with a system based on artificial intelligence (AI). Currently, AI methods are implemented in the process industry within the project KEEN – "Artificial Intelligence Incubator Laboratories in the Process Industry". The project is funded by the German Federal Ministry for Economic Affairs and Energy; it connects 20 industrial and scientific institutions and is divided into three thematic areas: AI-based optimization, AI-based engineering of plants and AI-based modelling. Among other things, the use of machine learning, which is a subtopic of artificial intelligence, will be investigated. The functionality of machine learning is in principle comparable to human learning. Humans use examples from their experience to differentiate and cope with complex requirements. A self-learning machine learns from data and can make decisions and predictions based on this data. These predictions can support the plant operator, the engineer as well as the process developer in their activities. Artificial intelligence acts as a cognitive amplifier.

Operating plants safer and more efficiently

A concrete application in the process industry is to support operating personnel of chemical plants with AI-based control and assistance systems. These are developed by technical universities (TU) in Dresden, Dortmund and Berlin, Germany, and others. They use models that have been developed from historical sensor
data and simulations of complex models. These models recognize critical plant conditions and predict the future behaviour of a plant. The learning process can be supervised or unsupervised. With an unsupervised learning process, a model is developed from data without a specific target via pattern recognition. Based on the models, the performance and safety of the plants is to be improved by the use of control algorithms as well as by the support of the plant operator being able to intervene on time and efficiently.

Recognizing hazards faster

The Artificial Intelligence is also to be used in the planning of conventional and modular process plants. For example, it is used for the Hazard & Operability (HAZOP) analysis, which must be provided for the approval and operation of a plant. The AI-based system learns from existing HAZOP data, heuristics and literature data. This accelerates the preparation of risk assessments, as is currently demonstrated for existing laboratory plants at the Laboratory of Equipment Design of TU Dortmund, Germany. The results will contribute to the acceleration of approval planning, the introduction of new processes and the training of the operators. Thus, the engineer is supported with an intelligent tool for the plant development.

Predicting substance data

For the process developer, knowledge of material data for process simulation is fundamental. Today's material databases are extensive, yet incomplete. Artificial intelligence methods predicting substance data open up a whole new perspective. This works similarly to portals that suggest movies to their users based on movies they have already watched: You watched “Harry Potter” – you might also like “The Lord of the Rings”. For modelling, physical knowledge, such as substance data and physical laws, is integrated into machine learning methods. The corresponding algorithm is developed by KEEN partners in Kaiserslautern, Germany (Fraunhofer Institute for Industrial Mathematics ITWM and TU Kaiserslautern).

The focus is on people

When scientists and developers consider various use cases, they always ask themselves how well end users can understand AI-based solution and decision proposals. The new AI methods should be comprehensible and contribute to better understanding by the responsible persons. Otherwise, there is a risk that users will not accept the methods or the use of AI-methods does not do justice to operator responsibility. Artificial intelligence will expand the scope of action of humans when used correctly as a cognitive amplifier.


Prof. Leon Urbas

Leon Urbas directs the Chair of Process Control Systems Engineering at Technische Universität Dresden. His research focuses on the digital transformation of the process industries.



Always up to date

With our newsletter you will receive current information on ACHEMA on a regular basis. You are guaranteed not to miss any important dates.

Subscribe now