OptiAIM – Integrated Optimization and Machine Learning for Reliable Modelling, Monitoring, and Control aims to enable autonomous, resilient, and resource- and energy-efficient chemical and energy systems through a tight integration of optimization and machine learning (ML). The project addresses a key limitation of current digitalization approaches: purely data-driven methods often lack physical reliability and interpretability, while first-principles models alone struggle with uncertainty, complexity, and scalability. OptiAIM introduces a bidirectional integration paradigm in which optimization is used to train physically consistent ML models and decision algorithms, and ML is used to enhance optimization-based decision making, improving robustness, versatility, and transferability across operating conditions and production units.

The methods are developed and applied consistently across three pillars: modelling, monitoring, and control. In modelling, the project develops hybrid, multi-fidelity, and multi-model approaches that combine first-principles knowledge with data-driven learning to achieve both physical consistency and practical accuracy. In monitoring, advanced algorithms for state estimation and anomaly detection are designed by combining optimization, ML, and explainable first-principles structure, including the use of graph neural networks to capture complex nonlinear dependencies. In control, the project develops advanced methods based on optimal control, Koopman operator frameworks, and model predictive control (MPC), including ML-assisted and uncertainty-aware MPC for improved real-world performance. The approach is validated on three industrially relevant case studies: plastics pyrolysis for chemical recycling, crystallization processes, and microgrid energy management. Together, these demonstrators showcase the broad applicability and high impact potential of OptiAIM for sustainable, reliable, and efficient process operation.