Search

• A new integrated hybrid modelling approach is proposed – Neural Network Programming. • NNP formulates autodifferentiable algorithmically structured neural networks (ASNN). • The performance of NNP-based models is superior to those of black-box models. • ASNNs are transferable between processes with similar structural features. • Hidden layers can be used to solve large sets of linear equations like mass balances. This work introduces Neural Network Programming (NNP) as an integrated hybrid modelling approach. NNP consists in formulating a set of first principles equations that is later decomposed and transcribed into an Algorithmically Structured artificial Neural Network (ASNN). NNP leverages the advantages of the universal approximation theorem and neural network optimization algorithms in order to generate physically coherent machine learning models. Since ASNNs are not mere approximations of physics equations, it is not necessary to modify either the gradient or performance function in order to account for errors with respect to the first principles equations. ASNNs are trained faster and more accurately than typical hybrid models because the gradient is computed through automatic differentiation instead of numeric differentiation. It is shown that the same ASNN architecture is transferable between processes with similar characteristics. In particular, a flash separator, distillation column, and a biogas upgrading process, were modelled using an identical architecture. [ABSTRACT FROM AUTHOR]

• A new formulation (PEF) of the governing equations for unit operation design. • In average, the PEF can reduce the computation time by one order of magnitude. • The Eulerian formulation and the PEF are mathematically equivalent. • A PEF-based model of a gas-liquid contactor is validated with experimental data. A computationally-efficient numerical method that uses a Pseudo-Eulerian formulation (PEF) for the design calculation of unit operations is presented and validated. This method is applicable to any unit operation that can be modelled using a system of ODEs. Performing the design of a unit operation in the PEF is tenfold faster than with the conventional Eulerian formulation (EF). The mathematical equivalence between the PEF and the EF is demonstrated by proving that the solution of different unit operation design problems provides the same numerical result independently of the formulation. It is shown that reducing the computation of the unit operation design problems also speeds the computation time of a process design or an optimization flowsheet. Additionally, as opposed to other computationally efficient methods for unit operation design, the PEF allows the accurate estimation of the concentration or temperature profiles of complex unit operations such as a multiphase multicomponent reactor system. [ABSTRACT FROM AUTHOR]

Abstract: The knowledge about pressure–volume–temperature–composition (PVTxy) properties plays an important role in the design and operation of many processes involved in CO2 capture and storage (CCS) systems. A literature survey was conducted on both the available experimental data and the theoretical models associated with the thermodynamic properties of CO2 mixtures within the operation window of CCS. Some gaps were identified between available experimental data and requirements of the system design and operation. The major concerns are: for the vapour–liquid equilibrium, there are no data about CO2/COS and few data about the CO2/N2O4 mixture. For the volume property, there are no published experimental data for CO2/O2, CO2/CO, CO2/N2O4, CO2/COS and CO2/NH3 and the liquid volume of CO2/H2. The experimental data available for multi-component CO2 mixtures are also scarce. Many equations of state are available for thermodynamic calculations of CO2 mixtures. The cubic equations of state have the simplest structure and are capable of giving reasonable results for the PVTxy properties. More complex equations of state such as Lee–Kesler, SAFT and GERG typically give better results for the volume property, but not necessarily for the vapour–liquid equilibrium. None of the equations of state evaluated in the literature show any clear advantage in CCS applications for the calculation of all PVTxy properties. A reference equation of state for CCS should, thus, be a future goal. [Copyright &y& Elsevier]

Abstract: 13C and 1H NMR spectra were obtained for AEEA (2-[(2-aminoethyl)amino]-ethanol)–H2O–CO2 systems and quantum mechanical calculations were carried out for the different AEEA species. The results suggest that the main AEEA species under the conditions studied are free amine, primary carbamate, and secondary carbamate. There is also some indication that a dicarbamate species is formed, this species does however only appear to be formed in small amounts. Comparison between experimental data and quantum mechanical calculations suggest that most AEEA species take on conforms with some degree of intramolecular hydrogen bonding. [Copyright &y& Elsevier]

An integration of Artificial Neural Networks (ANNs) and thermodynamics through the application of Neural Network Programming (NNP) is proposed. Thermodynamic consistency is achieved because the thermodynamic relationships and constraints are transcribed into a specially crafted ANN. Moreover, the developed models allow predicting and extrapolating the model outside the experimental data boundaries. The Wilson and NRTL models are used as case studies. Modifications to these models based on sigmoid functions are rigorously assessed in order to perform the simultaneous modelling of VLE and excess enthalpy. The automatic differentiation together with the ANN optimization algorithms can find sets of parameters that are better than the ones obtained with traditional gradient-based optimizers. The frequently disregarded concepts of thermodynamic modelling with ANNs are discussed in-depth. A mathematical analysis of the impossibility of typical fully connected ANNs to formulate thermodynamically consistent equilibrium models is discussed and their use is discouraged (e.g., VLE, LLE, or adsorption.). [ABSTRACT FROM AUTHOR]

New experimental viscosity data of three binary CO 2 -rich mixtures with approximate mole fractions of 0.93 CO 2 + 0.07 H 2 , 0.80 CO 2 + 0.20 H 2 , and 0.90 CO 2 + 0.10 N 2 are reported. The measurements were performed at four different isotherms between 273 K and 473 K. Two independent rotating body viscometers with different pressure range were utilized, suitable for measurement of pressures up to 2 MPa and 20 MPa, respectively. The maximum expanded combined relative uncertainty (k = 2) in viscosity was 0.25% for the data obtained from the low-pressure viscometer and 0.6% for the data measured by the high-pressure viscometer. The experimental data were compared to the data estimated by the extended corresponding states (ECS) model implemented in the NIST REFPROP 10.0 database. The maximum deviation between the model and the new data is 1.86%. The data measured with the two apparatuses agree within the estimated uncertainty. [ABSTRACT FROM AUTHOR]

Abstract: The behavior of seven binary particle drag closures from the literatures which have been derived based on the kinetic theory of granular flows (KTGF) for the application to fluidized beds are investigated using a binary KTGF model. The dynamic size segregation experimental data from the literature is used to validate the simulation results. The validations show that all of the seven binary particle drags under-predict the binary particle coupling in the dense fluidized bed, satisfactory results could be obtained if an extra semi-empirical frictional binary particle drag is included. The semi-empirical frictional binary particle drag considering the long term particle–particle contact effects is a correction of the short term collisional frictional binary particle drag from Syamlal. Furthermore, a group of comprehensive calculations shows that the model using the semi-empirical frictional binary particle drag can fairly well predict the particle segregation rates and the bed heights of the individual particles with a change of gas fluidization velocity, the initial bed height, and the small particle ratio with the same fixed correction coefficient. Therefore, it is proposed that the frictional binary particle drag which represents the binary particle momentum exchange due to the long term particle contacts (sliding/rolling) must be included in order to model dense binary fluidized beds. [Copyright &y& Elsevier]

Abstract: The paper presents a multi-fluid Eulerian model derived from binary kinetic theory of granular flows, free path theory and an empirical friction theory. The effects of the inter- and inner-particle collisions, particle translational motions and particle–particle friction are included. As the effects due to fluiddynamic particle velocity differences and particle–particle friction are considered, some unconventional terms are produced compared with the previous models. Model validation using the data from shows that the coupling terms give a stronger and more realistic particle–particle coupling because the effects due to the fluiddynamic velocity differences are considered. The model gives reasonable predictions of the particle volume fraction, particle velocities and velocity fluctuations. The model analysis reveals that the basic particle velocity fluctuations constitute 2 terms: the velocity fluctuations of the discrete particles, and the velocity fluctuations of the continuous fluid flow. Furthermore, the simulation results show that the velocity fluctuations of the continuous fluid flow are dominant in a binary riser flow. [Copyright &y& Elsevier]

Carbon dioxide/water (CO 2 /H 2 O) mixtures are of much interest in carbon capture and storage, atmospheric science, in the description of human lungs and in the processing of food and beverages. We present a comprehensive comparison of thermodynamic models for describing their P V T x y behavior, i.e. densities and phase compositions. The most accurate experimental data in the temperature range 273–478 K and at pressures below 61 MPa are selected after a critical data evaluation. The most reliable phase equilibrium data are used to fit the binary interaction parameters of a wide range of thermodynamic models: cubic equations of state (EoS) with quadratic/Wong–Sandler/Huron–Vidal mixing rules, CPA, PC-SAFT and PCP-SAFT with different association schemes, and corresponding states models with various reference fluids. We test the predictive ability of the models by comparing to data outside of the region used in the parameter-fit. All of the thermodynamic models are fitted with the same experimental data and compared on the same basis, facilitating a general discussion about their strengths and weaknesses. As a benchmark for the performance of the models, we compare with the performance of two multiparameter EoS: GERG-2008 and EoS-CG. At least three fitting parameters are needed to represent the P V T x y behavior of CO 2 /H 2 O mixtures within an accuracy of 10%. By including a fourth parameter, it is possible to significantly improve the accuracy for phase compositions, where the Peng–Robinson cubic EoS with the Huron–Vidal mixing rule and volume shift gives the best results with an average accuracy of 4.5% and 2.8% for phase compositions and densities respectively. In comparison, the most accurate multiparameter EoS, EoS-CG, exhibits an average accuracy of 8.0% and 0.6% for phase compositions and densities respectively. [ABSTRACT FROM AUTHOR]