The Process Simulation Working Group sees itself as an open committee for engineers interested in simulation. At the biannual meetings, the focus is on an open–to–technology exchange on the diverse application areas of process simulation and its challenges in practice. This time, some options for data–driven modeling were examined on the basis of various practical examples and subsequently further developed in the joint workshop.
On September, 26, 2024, the 13th meeting of the Process Simulation Working Group took place in Würzburg, Germany, as part of the Smart Process Manufacturing Kongress on September, 24 and 25, 2024.
Expert Meeting Incl. Get-Together
This time around twenty interested experts from the fields of chemistry, pharmaceuticals, food and engineering gathered for the meeting of the working group on Thursday, 26 September 2024. Already on Wednesday evening, the participants met for a cosy get–together in the Alter Kranen, a culinary “institution” in Würzburg.
The topic of Data-Driven Modeling had been chosen for the actual meeting of the Process Simulation Working Group on Thursday. In the morning, three hands-on presentations provided an overview of current challenges and innovative solutions as well as initial opportunities to start the discussion. The afternoon was dominated by a workshop in World Café format on the three main topics of Data Preparation, Automatic Model Generation and Cross-Platform Synchronization.
Presentation 1: A Digital Twin At CSL
Alexander Krez (CSL), Dr. Christian Sonntag (INOSIM)
CSL Behring, a biopharmaceutical company with the core business of plasma products, planned a new basic fractionation system at the Marburg, Germany site, which is by 80% automated. The development partner was the Austrian ZETA Group, an end-to-end solutions provider, which is driving forward the end-to-end digitization of projects with its SES platform. INOSIM Simulation Software was used in the project throughout the entire planning phase right through to operation.
In the concept developed together with the customer, questions such as “Do we use the existing building or do we need a new building?” were regarded. In Detail Engineering, detailed process sequences and ancillary processes such as CIP were then mapped. In this way, batchsizing, staff requirements and the robustness of the process could be determined. Since there are long periods of automated steps in the plant, it was necessary to obtain a prediction of the manual steps in the plant. Therefore, the simulation model developed in engineering is used with INOSIM Foresight for predictive production support for the operation of the plant.
INOSIM Foresight can be described as the Google Maps of the process industry: just like paper maps were used for decades for planning trips and getting around, but were replaced by modern, always-up-to-date, real-time web-services. In the process industry, production is still mostly planned manually. INOSIM Foresight now enables accurate predictions with real-time data and automatic planning with high-precision models. A digital twin is initialized using live data from a real plant. In this way, specific process steps can be predicted. For accurate predictions, the model must correspond as precisely as possible to the real plant. To achieve this, INOSIM Foresight relies on the connection to existing systems. The presented project was the MES system, as it combines all the required data.
In Foresight projects, the simulation works behind a platform, which can then be used for a wide variety of use cases. For this purpose, CSL uses a departure board that predicts the manual steps in the system. Other use cases, such as decentralised maintenance planning or optimization of schedules, could be imagined here. Work is currently underway on the challenge of updating model parameters.
Presentation 2: Hybrid Simulation Models
Michael Schüler (Siemens)
Hybrid models consist of two components: First-Principle models are rigorous mathematical models that require little data, whereas Machine Learning requires a lot of data. Use cases for using hybrid models include heat exchanger design, chemical reactions, or model predictive control, which can lead to long computation times. So it can be beneficial to train a model with data.
In described use case, a simulation model for the reaction for methanol production in the CSTR (continuous stirred-tank reactor) is used, in which the reaction rate function is determined not from the Arrhenius approach, but from data machine learning. In a first approach, the reaction rate is determined as a function of a few quantities. If the activation energy is added to the existing approach, the reaction rate function obtained from the data accurately reflects the function from the Arrhenius approach.
Data-driven models are made available as a no-code method via a web interface (internal). Users can combine rigorous models with data-driven models, for example by incorporating quantities from data-driven models into their modeling tools. In this way, these models can later be made available to non-data users.
Discussion: Standards for co–simulation with different simulation tools
For the presented application, FMU (Functional Mock-up Interface) was used as an interface, which is not yet available for all simulation software. A second topic of discussion was the influence of strongly fluctuating raw material qualities on the presented method.
Presentation 3: Reduction of Plant-Model Mismatch
Carina Bisping (INOSIM)
When creating simulation models, a certain plant-model mismatch is inevitable. A model is usually created in a level of detail that makes sense for the application. However, when it comes to linking a model to a real plant, the model must correspond exactly to the plant. To do this, on the one hand you can derive the recipe structure from historical data (data-driven modeling), on the other hand you can compare model parameters with historical data (parameter estimators).
To simplify recipe creation in INOSIM, it will soon be possible to import production recipes from already existing recipe structures. Such existing recipe structures may derive from automation software or other simulation tools. An additional advantage is that this allows even non-INOSIM users to create a recipe, which is then imported into INOSIM.
The following discussion was about creating INOSIM recipes in an Excel format and then exporting them as recipes to INOSIM.
Workshop
In the following workshop, the participants addressed the following questions:
Preparation of historical data for the determination of simulation parameters – Is it always the mean value?
The processing of historical data strongly depends on the quality and type of data. Depending on the objective of the model, the statistical evaluation of real data can therefore be very different.
What are the requirements for (partially) automated model creation? What does the future of simulation users look like in this context?
The workshop participants agreed that the automatic generation of models can make repetitive tasks such as naming objects easier for users. However, a simulation expert is still needed to model special cases and control the model.
Simplified model maintenance through “synchronization” with engineering platforms, and vice versa?
Synchronization with other tools requires functioning interfaces in import and export. A single-source-of-truth solution is challenging due to the differences in the level of detail.
The next meeting
The next Process Simulation Working Group is planned for spring 2025 at SimPlan AG in Hanau, Germany. More detailed information will be available soon here on the INOSIM homepage and on the Working Group’s homepage. We look forward to your registration and will be pleased to invite you to our next meeting!
Do you have any questions or would like to know more about this topic? Please contact us.
General Inquiries 
