CHAI

Machine learning (ML) to help process engineers

Process engineers interpret a high volume of data from sensors to steer chemical processes. However, before acquiring the necessary experience, they need to supervise actual production environments for years. This makes finding the right people a real bottleneck when scaling up production plants.

Machine learning (ML) could drastically increase efficiency by supporting less experienced process operators and more experienced engineers. The former could get actionable insights and expertise, and the latter would have more time to focus on crucial tasks. This will be done by creating smart hybrid AI solutions that learn from data (i.e., historical examples) and knowledge elicited from highly experienced engineers.

Challenges for ML in chemical process engineering

These are some of the challenges that should be solved before ML can be adopted to optimize chemical processes:

• Today’s ML models are purely data-driven; they cannot take into account the expertise and know-how of the chemical process engineers.
• Black-box models are hard to understand whereas process engineers need to be able to trust the predictions. Only by understanding why these predictions were made can they take the appropriate control actions knowing they are safe.
• The control of a chemical process needs to be improved by co-optimizing productivity and safety rather than giving precedence to one over the other.
• Feedback is needed to continuously tune the ML to process engineers’ ever-changing experience and knowledge.
• The ML needs to be able to deal with conflicting feedback.

Towards explainable hybrid AI

With these challenges in mind, the CHAI consortium will design explainable hybrid AI algorithms. These will incorporate expert knowledge into the machine learning. As a result, data streams will automatically be translated into contextualized insights, outcome predictions, and suggested control actions.

The insights, predictions, and suggested actions will be brought to both operators and experienced process engineers through contextualized and dynamic visualization in responsive dashboards. These will be enriched with hybrid AI insights and explanations of how the AI arrived at its conclusions.

A feedback loop will allow the operators and experts to optimize the hybrid AI decision-making intuitively and continuously. This way, improvements are made in the standardization of processes, the amount of expert attention needed for low-level routine monitoring, and the speed and quality of the process outcome.

“The CHAI consortium will leverage ML that incorporates expert knowledge for better outcome prediction and control actions. This will allow scaling the expertise of process operators and engineers, whatever their experience.”