CPACT
Webinar on
AI-Enhanced
Process Intensification:
From
Molecular Design to Unit Operations
Ulderico
Di Caprio, Delft University of Technology
2nd
April 2026 at 3pm (UK time)
Process intensification is widely recognised as a key
strategy for improving the efficiency and sustainability of chemical and
pharmaceutical manufacturing. Despite its potential, its implementation remains
challenging due to the strongly multiscale nature of intensified processes,
where decisions made at the level of materials and solvents propagate through
unit operations and ultimately affect overall process design and control. In
this context, modelling and optimisation tools that can consistently operate
across these scales are required. This presentation discusses how machine
learning, when combined with physical insight, can support process
intensification at various scales, from material selection to flowsheet-level
decision making, with examples in the domains of pharmaceutical processing and
CO2 capture.
At the material and solvent level, hybrid physics-informed
models are used to describe complex absorption and solubility phenomena that
are difficult to capture with conventional correlations alone. Using CO2
capture as a case study, it is shown how machine learning can be embedded
within mechanistic frameworks to predict the performance of novel solvent
systems and intensified contacting devices, enabling faster screening while
maintaining physical consistency.
At the unit operation scale, data-driven and hybrid
approaches are applied to improve both process understanding and monitoring.
Examples include the prediction of mass transfer performance in non-standard
absorption equipment and the use of machine learning models to interpret Raman
spectroscopy data for concentration estimation and anomaly detection. These
approaches provide more accurate and flexible models than traditional methods,
supporting improved equipment design and real-time decision making.
At the process scale, reinforcement learning is explored as
a tool for optimising dynamic and constrained operations. By training agents on
digital twins, operating policies can be learned directly from process
behaviour rather than predefined heuristics. Applications in pharmaceutical
processing demonstrate how such approaches can identify unconventional yet
practical operating strategies, which have been validated experimentally and
shown to reduce processing time and cost.
Overall, the presentation highlights the role of machine
learning as an enabling tool for multiscale process intensification,
particularly when used in combination with first-principles knowledge. Rather
than replacing existing modelling approaches, these methods complement them,
offering new ways to explore design and operation spaces that are otherwise
difficult to access. The work illustrates how such tools can contribute to more
efficient, flexible, and sustainable manufacturing processes.
This webinar will take
no longer than one hour.
The webinar is for
CPACT members only.
Please register directly at https://universityofstrathclyde.webex.com/weblink/register/rf99aa1063a806a27659bb289a5701115