Articles
The lack of accurate recognition of the reaction kinetics and the catalyst deactivation are challenges in commercializing the methanol-to-propylene process (MTP). Accordingly, this research aims to develop reliable intrinsic kinetic models for MTP reactions and catalyst deactivation on an industrial ZSM-5 catalyst. An efficient reaction network was developed based on a combination of hydrocarbon pool and dual-cycle mechanisms considering individual pathways for producing olefins, paraffins, and aromatics. Six deactivating models were investigated based on the possible coke precursors of aromatics, olefins, and oxygenates. Since the deactivation rate of the catalyst at normal operating conditions is slow, the “acceleration deactivation” technique was employed to reduce the time and cost of deactivating experiments. The proposed kinetic models considered the combined effect of water on reducing the rate of progress of reactions and catalyst deactivation. The experiments were performed in a fixed-bed reactor under conditions relevant to industrial operations leading to a full conversion of oxygenates as follows: temperature of 733–763 K, feed WHSV of 5–14 h−1, and feed methanol content of 50–93 wt%. Therefore, the model is only valid for predicting the behavior of the reactors operating under full conversion conditions, making it useful for the simulation of industrial reactors. Oxygenates were found to be the main responsible for catalyst deactivation through coke formation by parallel decay reactions according to first-order kinetics. The detrimental effect of water in suppressing MTP reactions is overshadowed by its benefit in surviving the catalyst activity. Reducing the feed WHSV and increasing the reaction temperature and water content enhance feed conversion and propylene selectivity. A good agreement between the calculated results and experimental data was observed with average errors of less than 10 % and 3 % for kinetic models of reaction and catalyst deactivation, respectively. This confirms the accuracy of these kinetic models, making them reliable for designing and optimizing industrial reactors. © 2024 Elsevier Ltd
Chemical Engineering Research and Design (17443563)212pp. 121-133
Catalytic dehydrogenation of long-chain normal paraffins is the most attractive route for producing of linear alkyl benzene. To make this happen, the radial-flow packed-bed reactors are employed as one of the most efficient currently available technologies. Simplifying assumptions that are sometimes imposed on reactor models to reduce the computational cost may also significantly decrease the accuracy of simulations. Here, it is decided to shed light on this matter by assessing the effect of typical model-simplifying assumptions on simulation results. To this end, one- and two-dimensional semi-homogeneous models are used to simulate an industrial-scale radial-flow packed-bed dehydrogenation reactor under isothermal and adiabatic conditions. Simulations are designed in four 1D isothermal, 1D adiabatic, 2D isothermal, and 2D adiabatic modes to compare different modeling strategies and investigate the effect of flow distribution on the reactor performance. An appropriate LHHW kinetics model is considered for paraffin dehydrogenation and the main associated side reactions over a commercial Pt-Sn-K-Mg/γ-Al2O3 catalyst. The model equations are solved numerically using the finite element method by COMSOL Multiphysics CFD software. The results show a 1–3 % discrepancy between the predictions of one- and two-dimensional models for feed conversion under isothermal and adiabatic conditions. In contrast, the comparison of isothermal and adiabatic results for each one- and two-dimensional models indicate a discrepancy of 33–36 %. Furthermore, the two-dimensional model shows a low non-uniformity in flow distribution under reaction conditions (∼ 0.175), which has a trivial negative effect on paraffin conversion. © 2024 Institution of Chemical Engineers
Journal of Loss Prevention in the Process Industries (09504230)91
High-pressure pipelines that transport sour natural gas contain high levels of hydrogen sulfide, which is poisonous and has irreparable effects on human health even in low concentrations. These pipes are break valve-assisted and buried underground to minimize gas leakage and protect people nearby. This study examines their leakage through a series of time-dependent three-dimensional CFD simulations. In contradiction of previous works that only considered the above-ground environment, here, for more realism, the computational domain includes the pipeline, trench, covering soil, and above-ground environment. The impact of hole size, leak location on the pipe, wind velocity, atmospheric stability class, time of occurrence (day or night), and the presence of break valves on the dispersion of leaked gas are comprehensively investigated. Results indicate that the effect of hole diameter on hydrogen sulfide concentration in the above-ground environment is dominant to other factors. In addition, the probability of fatality due to gas release and the intensity of the gas leak exposure crisis are studied by combining the dose-response model and CFD simulation results. In this line, LT50, which measures how long it takes for 50% of people in different areas around the pipeline to die from exposure to hydrogen sulfide is calculated. © 2024 Elsevier Ltd
Methanol steam reforming (MSR) is one the most interesting routes for production of fuel cell grade hydrogen. As this reaction is endothermic, its energy supply is one of the most important problems. In this line, recycling and burning the unconverted hydrogen exits from the exhaust of the fuel cells on the shell side of a shell-and-tube reactor (HR) has been suggested in the literature as one the energy supply solutions. In this work, the performance of a membrane-assisted shell-and-tube reactor (MR-HCO), in which part of the hydrogen produced by the MSR is continuously transferred to the shell side through the membrane layer and burned to supply energy, is investigated and compared with HR reactor. To this end, a set of heterogeneous 1D and 2D models are employed to model the shell and the tube sides of the reactors, respectively. The influence of air and steam-methanol feed hourly space velocities on maximum combustion temperature and methanol conversion are examined. It is observed that the MR-HCO has higher methanol conversion, hydrogen flow, and thermal efficiency, while lower CO concentration compared with the HR reactor. Furthermore, the MR-HCO showed a good potential to control the shell temperature rise and prevent hot spots. © 2023 Elsevier Ltd
Neural Computing and Applications (09410643)35(11)pp. 8517-8541
Using a reliable predictive model is important for modeling, controlling, and optimization of the isomerization process. This process has a significant impact on the gasoline quality, which can reduce greenhouse gases by improving the octane number. On the other hand, the accuracy of the predicted results of a data-driven model depends on the quality of input data; this is while the measured variables of industrial units are inevitably contaminated by various errors. Hence, the present work proposes an improved adaptive machine learning model and a new hybrid multiscale filter to predict the gasoline research octane number reliably from error-contaminated data of a light naphtha isomerization reactor. The proposed machine learning model is based on the integration of the feature selection algorithm of the double-level similarity with the support vector regression model (named DLS-SVR model) for adaptive prediction. The new hybrid filter is based on a combination of the wavelet transform and median absolute deviation, named multiscale median absolute deviation (MSMAD). MSMAD filter is proposed with the aim to establish an accurate method to identify and eliminate outliers and gross errors from the measured process variables. A pilot-scale reactor is employed to provide the required experimental validating dataset to evaluate the predictive performance of the proposed filter–model combination. Inputs of the DLS-SVR model are operating conditions (temperature: 115–150 °C, pressure: 28–42 bar, space velocity: 0.38–3 h−1) and feed composition (benzene: 0–3.5 wt%, cyclohexane: 0.8–23.2 wt%, methylcyclopentane: 1–29 wt%, H2/naphtha ratio: 0.03–0.3). The performance of the DLS-SVR model is compared with the response surface methodology, support vector regression, and double-level locally weighted extreme learning machine through the fivefold cross-validation technique. The particle swarm optimization–sequential quadratic programming algorithm is used to optimize the hyper-parameters of these models. The results prove that the generalized DLS-SVR model outperforms the other generalized models. Furthermore, the performance of the MSMAD filter is compared with the multiscale median, finite impulse response–median hybrid, median, and median absolute deviation filters by rectifying the error-contaminated temperature signal. Findings reveal that the DLS-SVR model utilizing the rectified signal by the MSMAD filter has a maximum coefficient of determination, R2 = 0.91, and minimum root mean square error, RMSE = 0.0562, among the other filter's rectified temperature signals. These values for error-free data are R2 = 0.945 and RMSE = 0.0439. © 2022, The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature.
