Integrating rate based models into multi-objective optimisation of process designs using surrogate models

Multi-objective optimisation (MOO) of super-structured process designs are expensive in CPU- time because of the high number of potential configurations and operation conditions to be calculated. Thus single process units are generally represented by simple models like equilibrium based (chemical or phase equilibrium) or specific short cut models. In the development of new processes, kinetic effects or mass transport limitations in certain process units may play an important role, especially in multiphase chemical reactors. Therefore, it is desirable to represent such process unit by experimentally derived RBMs (rate based models) in the process flowsheet simulators used for the extensive MOO. This increases the trust engineers have in the results and allows enriching the process simulations with newest experimental findings. As most RBMs are iteratively solved, a direct incorporation would cause higher CPU-time that penalises the use of MOO. A global surrogate model (SUMO) of a RBM was successfully generated to allow its incorporation into a process design & optimisation (PDO) tool which makes use of an evolutionary MOO. The methodology was applied to a fluidised bed methanation reactor in the process chain from wood to Synthetic Natural Gas (SNG). Two types of surrogate model, an ordinary Kriging interpolation and an artificial neural network, were generated and compared to its underlying rate based model and the chemical equilibrium model. The analysis showed that kinetic limitations have significant influence on the result already for standard bulk gas chemical components. From experimental experience it can be stated that a significant amount of chemical compounds, which are near to complete conversion according the thermodynamic equilibrium, are measured in the gas phase after the reactor. Thus including RBMs in PDO improves the quality of results. This approach allows a significant improvement of information exchange between process design & optimisation workflow and experimental development of PUs in early stages of process development.

Integrating Rate Based Models into a Multi-Objective Process Design & Optimisation Framework using Surrogate Models

Levent Sinan Teske

In the development of energy and chemical processes, the process engineers extensively apply computer aided methods to design & optimise these processes and corresponding process units. Such applications are multi-scale modelling and multi-objective optimisation methods. Multi-objective optimisation of super-structured process designs are expensive in CPU-time due to the high number of potential configurations and operation conditions to be calculated. Thus single process units are generally represented by simple models like equilibrium based (chemical or phase equilibrium) or specific short cut models. In the development of new processes, kinetic effects or mass transport limitations in certain process units may play an important role, especially in multiphase chemical reactors. Therefore, it is desirable to represent such process units by experimentally derived rate based models (i.e. reaction rates and mass transport rates) in the process flowsheet simulators used for the extensive multi-objective optimisation. This increases the trust engineers have in the results and allows enriching the process simulations with newest experimental findings. As most rate based models are iteratively solved, a direct incorporation would cause higher CPU-time that penalises the use of multi-objective optimisation. A global surrogate model (SUMO) of a rate based model was successfully generated to allow its incorporation into a process design & optimisation tool which makes use of an evolutionary multi-objective optimisation. The methodology was applied to a fluidised bed methanation reactor in the process chain from wood to Synthetic Natural Gas (SNG). Two types of surrogate model, an ordinary Kriging interpolation and an artificial neural network, were generated and compared to its underlying rate based model and the chemical equilibrium model. The analysis showed that kinetic limitations have significant influence on the result already for standard bulk gas chemical components. A case study applying the previous version of the process design model and the revised version (with rate based model introduced as a set of five surrogate models) will demonstrate that the prediction uncertainties of the process design & optimisation methodology are reduced due to the integration of the rate based model of the fluidised bed methanation reactor. It will be shown that the different process design models predict considerably different optimal operating conditions of the Wood-to-SNG process. This emphasises the importance of the integration of rate based models into the process design models. The presented approach has been developed for the fluidised bed methanation reactor, however, it is a generic approach which can be applied to other process unit technologies as well. Future investigations will target other technologies to further improve the process design & optimisation predictions and support project development.

EPFL2014

Développement de réacteurs microstructurés pour la conversion du propane en produits valorisés

Olivier Wolfrath

Industry requires continuously new chemical processes to convert raw materials to valuable products limiting the formation of by-products. Since a few years, an increasing interest has been given to the intensification of chemical processes. This represents a new approach in chemical reaction engineering and calls for a new type of chemical reactors: the microstructured reactors. They are characterized by miniature structures whose components have a sub-millimeter size. The flow channels, for example, have a diameter from ten to several hundred microns. The distinctive feature of these structures is a very large surface/volume ratio, namely between 10'000 and 50'000 m2/m3. In these conditions, heat transfer coefficient is very high, up to 25 kW · m-2 · K-1. It is accepted that the residence time distribution in a reactor strongly influences the product selectivity and yield. In these thin channels, the RTD is particularly narrow due to a short radial diffusion time (ms), allowing a precise control of the residence time. Another advantage of these microstructures is an increase of the inherent safety due to a small amount of the compounds used. However, fabrication of microreactors, especially the introduction of active catalyst in microchannels remains a difficult operation. And integration of microsystems in conventional chemical processes is still a challenge. The aim of this work is first to develop a novel membrane reactor based on a new microstructured catalytic packing in the filamentous form and to apply it to the nonoxydative dehydrogenation of propane. A second objective is to develop a reactor with micro heat exchangers allowing an optimized control of homogeneous reactions. This reactor was tested with the partial oxidation of propene to propylene oxide. The microstructured catalyst above is made of thin and long catalytic filaments placed in parallel with a tubular reactor. This new kind of catalyst is based on silica fibres covered by a γ-alumina layer. The position in parallel of these filaments of 7 μm diameter forces the gases to flow in the fibres direction between the filaments. The hydraulic diameter is about 70 μm. Due to this laminar and regular flow, the pressure drop in the bed was divided by 5 compared to a conventional bed (100-160 μm spheres) and the residence time distribution was particularly narrower. Such novel packing brings new opportunities to the heterogeneous catalysis domain because it can be used for many other reactions improving highly the hydrodynamics and keeping an excellent contact with the reactants. The microstructured catalyst was installed in two concentric zones of a tubular reactor separated by a palladium-silver membrane permeable to hydrogen. This product was removed from the gas phase to shift the thermodynamic equilibrium, enhancing the propane conversion from 22% up to 30%. Another significant advantage in this reactor is the increase of propene selectivity up to 97% in comparison to conventional industrial processes reaching 80-90%. This difference is due to the diminution of hydrogenolysis and hydroisomerization reactions. The dehydrogenation is known to be endothermic. Oxidation of the hydrogen passing through the membrane produced heat for this reaction along the catalytic zone. Moreover, a periodic regeneration by air allowed recovering the activity of the catalyst deactivated by coke formed during dehydrogenation. Periodic reaction operations allowed producing continuously propene with a high selectivity, and regenerating simultaneously the catalyst. The second microstructured reactor was developed integrating two micro heat exchangers at the inlet and outlet of a macroscopic tube. The aim is to heat up very fast the gas and to cool it down as well. Without these exchangers, heating and cooling duration corresponded each to about 20% of the total residence time. The use of the microstructures allowed diminishing significantly these durations (4-5% each). Applied to partial oxidation of propene, this reactor reached a more precise control of reaction temperature and duration. The propylene oxide (PO) selectivity was not increased with this system but, due to the microstructures, it was possible to work at higher temperatures where the reaction becomes very fast, without reaching an oxygen conversion of 100%. Indeed, it is essential to avoid this situation where oligomers and coke form dividing by 2 the PO selectivity. So another new kind of reactor was developed suitable for any fast exothermic reaction whose progression has to be limited because of potential consecutive reactions reducing, for example, the selectivity of an intermediate desired compound.

EPFL2001

Fast temperature cycling of the catalytic CO oxidation using microstructure reactors

Martin Luther

The possibility to increase the performance (productivity or selectivity) of a chemical reactor by using periodic variations of reaction parameters (e.g. reactant concentration or temperature) has been theoretically envisaged since the beginning of the 70th. The experimental validation of the predicted positive effects was successful in the case of concentration variation but failed for temperature variation. This was mainly due to the high thermal inertia of the conventional chemical reactors used for the measurements which prevented to create variations having a sufficiently high frequency. Microstructure reactors own, at the contrary to conventional reactors, a very low thermal inertia and allow to generate temperature oscillations with an amplitude of about ten to hundred Kelvin at a frequency in the order of magnitude of 10-1 to 4 Hz. These properties, coupled with the possibility to introduce a catalytic active material within the devices, seem to make them well suited for the study of the effects of fast periodic temperature variations of a catalytic reaction. The objective of this work was to demonstrate that non-stationary temperature conditions may increase the reaction rate of a heterogeneously catalyzed reaction up to values not predicted by the classical Arrhenius dependency towards the temperature. Two different types of microstructure reactors have been used. The reaction was taking place in the first one (FTC-type 2) in microstructured channels on whose walls a catalytic layer was deposed. In the second one (FTC-type 3), the reaction was taking place on a piece of sintered metal fibres (SMF) plate placed in a reaction chamber. The catalytic active material was deposed on the SMF plate filaments. Both devices where permanently heated by electrical resistors and periodically cooled down with water flowing through cooling channels incorporated in the devices. The test reaction chosen for the experimental measurements was the CO oxidation reaction, heterogeneously catalyzed by platinum supported on alumina (Pt/Al2O3). The reaction behaviour under stationary thermal conditions was consistent with the one predicted by the Arrhenius law. The dependency of the reaction rate with the temperature was exponential with an apparent activation energy of 104 kJ·mol-1, a negative partial reaction order for CO and a positive one for O2. The measurements effectuated under quasi-stationary thermal conditions (slow temperature ramps) have shown that a temperature change rate between 7 and 14 K·min-1 was not sufficient to observe any non-trivial effect of the temperature. The reaction behaviour was always predicted by the Arrhenius law. The surface coverage of the reactive species is, in this case, always able to follow the slow temperature changes and the reaction behaves as being always under steady-state conditions. The experiments realized under non-stationary thermal conditions using the FTC-type 2 reactor have also failed to demonstrate any non-trivial effects of the temperature oscillations. This device allows indeed to generate temperature oscillations with an amplitude of up to 120 K with a frequency of 0.1 Hz but unfortunately correlated with a very high thermal inhomogeneity. A temperature difference of up to 80 K was measured between a cold spot and a hot spot inside the device. Due to their very low local reaction rate the colder areas of the reactor have attenuated the product concentration (CO2) oscillations which should have been created by the temperature variations. This attenuation prevented the temperature oscillations to have any positive effect compared to the stationary thermal conditions. The FTC-type 3 reactor allowed temperature changes of lower amplitude but the thermal homogeneity was much better. The maximal temperature difference measured between two points within the reactor was only 15 K. Under temperature oscillation conditions, the measured instantaneous CO2 concentration was higher for any temperature within the oscillation range compared to the one recorded under stationary thermal conditions. The increase obtained in the mean CO2 concentration was ranging from 34% for a frequency of the oscillations of 0.035 Hz to 85% for a frequency of 0.052 Hz. The amplitude of the oscillations was kept constant at approximately 40 K and the mean temperature value at 437 K. The simulations effectuated using a theoretical model for the catalytic CO oxidation including a feed-back step in the form of the oxidation-reduction of the catalyst have shown that the experimentally observed increase may be qualitatively explained. Above a certain frequency, the temperature oscillations are fast enough compared to the characteristic time of the feedback step and are able to perturb the stationary established reactive species surface coverage. The formation of a transient surface coverage more favourable for the reaction than the surface coverage at high temperature allows during the transient period to reach an instantaneous surface reaction rate values higher than the one at high temperature. This reaction rate peak is then responsible for the increase of the mean reaction rate under non-stationary thermal conditions and, thus, for an increased yield.

EPFL2008