Data-Driven Process Optimisation Syngenta
CTE and its partner Learning Machines used the Data Science Framework to demonstrate how production data can be leveraged to optimize energy-intensive processes. The goal of the pilot project at Syngenta in Kaisten was to identify additional opportunities for optimization in a dewatering column and to operate the process more efficiently without compromising the required quality standards.
Mandate
The dewatering column is a quality-critical process. Even minor changes to the parameters can have a direct impact on the efficiency of subsequent process steps. Traditional optimization approaches reach their limits in this context: the process space is complex, interactions between parameters are not fully transparent, and existing models provide only a simplified representation of the actual plant behavior.
Syngenta therefore wanted to determine whether additional opportunities for optimization could be identified based on real operational data—without compromising existing quality standards. In collaboration with CTE, a data-driven approach was adopted to better understand the plant’s actual behavior and to derive well-founded optimization recommendations that could be tested in operation.
Procedure
The project was developed iteratively and closely aligned with actual operating conditions from the outset. The goal was to identify and validate opportunities for optimization not in theory, but directly during ongoing operations. Relevant process data from the existing PI system was used as a basis and prepared for analysis, while quality data was manually transferred from the LIMS.
Based on this, data-driven models were developed to identify correlations within the process and provide specific recommendations for optimized parameter sets. These recommendations were then tested under real operating conditions and gradually refined. This iterative approach ensured that new insights were continuously incorporated into the models and that the optimization was continually improved. This was done in alignment with Syngenta’s process knowledge and requirements.
Achievements
The parameter sets developed and tested in the project resulted in measurable energy savings of approximately 3% without compromising the defined quality limits. Crucially, the optimization recommendations were not validated under idealized conditions, but rather during the plant’s ongoing operation. The results are therefore based on the actual behavior of the column and not on simplified model assumptions.
At the same time, the project demonstrated that the iterative approach works: by continuously feeding in new operational data, the model was able to map the relationships within the process with increasing precision and further improve optimization. The entire project was implemented in just 2.5 months, demonstrating that such a project can be carried out with manageable effort. Based on the results achieved, the approach is currently being applied to other columns and implemented at additional sites.
The pilot project thus demonstrates that data-driven models, process knowledge, and real-world operational data can be combined to form a robust optimization approach—one that is directly relevant for industrial applications.
