Margins in the chemical industry are under increasing pressure, and the cheapest new capacity is often the plant you already own, provided it can be operated safely closer to its limits.

This talk illustrates how BASF Antwerp addresses the challenge of operating industrial distillation processes closer to their limits through the deployment of soft sensors. These machine-learning models continuously estimate laboratory quality parameters from real-time process data, providing operators with immediate insight instead of waiting for the next laboratory result. By integrating OT, data platforms, edge computing, and machine learning in a demanding industrial environment, the approach eliminates measurement delays and reduces sampling costs while providing the timely process insight needed for advanced process control and increasingly autonomous, value-driven plant operation—paving the way for scalable industrial AI applications.

While soft sensors are by no means new, the approach presented here introduces several important innovations.

First, it enables plant engineers to flexibly develop and deploy models using a broad range of machine-learning technologies, allowing them to adopt the most suitable or latest techniques without being constrained by the underlying infrastructure.

Second, a model developed on an engineer's laptop can be seamlessly deployed to Siemens Industrial Edge as Edge application. Through direct integration with the plant automation infrastructure, near-real-time process data can be acquired, processed, and converted into live quality estimates with minimal latency. 

Third, the current architecture could pave the way for scaling machine-learning applications across multiple production units and sites.