Post by Dr. Oskar Schneider

[Mini Interview] Crafting AI Solutions to Run Better Chemical Plants   Meet Frank Mollard, he serves as Chief Data Scientist and Head of Business Intelligence at Infraserv GmbH & Co. Höchst KG, a company running major chemical and pharma facilities.   Tell us your journey with Operations Research (OR)?   I began my career as a statistician at the Deutsche Bundesbank, where I used AI to improve data quality.   Later, as a consultant at FTI-Andersch, I carried out retrospective A/B testing of marketing measures and conformal sales predictions, eventually ending up in the chemical industry.    I find the chemical industry fascinating because of the enormous amounts of data that are constantly being generated.   Here, I quickly realized that the goal should be to systematize the creation of added value, meaning that the success of a project should become more predictable.    This led to a process model based on the credo “keep it as simple as possible, but no simpler than necessary.”    This means gradually increasing complexity in order to ultimately find a solution whose complexity is in balance with the complexity of the problem itself.   Tell us about a recent OR project?   Large technical systems consisting of complex individual components can often be represented as convex polyhedrons.    This makes it possible, for example, to numerically optimize recooling systems. The optimization relates directly to the power consumption of the physical systems and indirectly to wear and tear.    The goal is to propose an optimal operating mode. The challenges lie particularly in sustainable modeling, as the algorithms are intended to be long-lasting. This is countered by wear and tear and changing environmental condition.    To stay informed about the physical system's condition, continuous monitoring using IIoT sensors combined with control system data is essential.    The optimization procedure must also reflect non-linear dynamics and possible machine outages to stay effective and up to date. We have implemented real-time algorithms that meet all these requirements.    This allows us to optimize the targets but also simulate what would be the best structure for the plant, leading to lower energy consumption and longer maintenance intervals.   Where is the biggest potential for OR at chemical companies?   In my experience, general-purpose generative AI - such as agent-based models - often falls short of delivering the intended outcomes, particularly when compared to customized, non-gen AI approaches.    The more specialized a solution is to a specific problem, the more effective and valuable it becomes. Nevertheless, there is clear value in designing solutions with enough generality to be transferable to similar applications.    This principle holds true for the optimization of recooling systems.    In Germany alone, optimizing such plants in industrial parks could lead to annual energy savings estimate of 36 GWh, which equals about 11,500 tons less CO₂.