Constrained Thermochemistry for Process Digitalization

The report summaries a set of computational studies applying process digitalization in terms of thermodynamics, biological modelling and artificial intelligence. The approach is based on computational chemical thermodynamics with special emphasis on advanced applications of VTT's unique Constrained Free Energy (CFE) method, the use of which in VTT’s ChemSheet and KilnSimu software is shortly reviewed.
The present work introduces systematic and automated method to add several reaction constraints to thermodynamic multicomponent system and thus widens the applicability of CFE in e.g. biochemical systems, which often have multiple reactions to be constrained. The developed methodology can be applied also inversely to the addition of equilibrium reactions to kinetic reaction systems. In addition, the tools of modelling bioprocesses combined with CFE modelling have been tested using Sulphate Reducing Bacteria (SRB) reactor as an example process. Moreover, a CFE related constraint deduced from the saturation index is used to calculate supersaturation conditions in aqueous solutions. The reaction constraint is further applied for computing oxidation-reduction potentials in non-equilibrium redox systems.
An additional new development of CFE method is its usage to produce phase diagrams for time-dependent reactive systems using extents of reaction as diagram axis. Applying such diagrams, the dynamic reaction conditions can be analysed graphically without doing elaborate and time-consuming kinetic experiments. Moreover, the simulation of time dependent features related to VTT’s KilnSimu has been advanced with an application example for lime kiln in kraft recovery process.
Several new thermodynamic mixing models have been implemented to VTT’s ChemSheet software. These include LIQUAQ and LIFAC electrolyte models to allow for coupling of electrolyte modelling with various vapour-liquid-liquid and solid phase systems. In addition, a new approach for computing highly concentrated aqueous solutions is presented by connecting the Pitzer activity model with the adsorption theory to enable modelling of aqueous multiphase multicomponent systems covering the range from dilute solutions to the hydrated salt instead of earlier applicability on diluted or medium concentrated solutions.
Machine learning (two-layer feed-forward neural network model) are applied for predicting mean activity coefficients for ion pairs in concentrated aqueous solutions. Further, the methods applying neural networks in process modelling are investigated and different approaches to replace the first principle based process models partially or as a whole, are tested targeting at faster calculation and better predictions in systems with scarce data.

The developed models and methods are targeted to be used in chemical and process engineering tools and work. In the wider perspective the deep-tech digitalization is facilitating more efficient process operation and design and thus enable better product quality with lower consumption of energy and pristine raw materials in an environmentally friendly and more economic processes.