(327d) Artificial Neural Network Aided Scale-up of a Multi-Tubular Reactor for the Oxidative Dehydrogenation of Butylene to Butadiene

Butadiene, a key raw material for the production of synthetic rubbers, and intermediate chemicals is conventionally manufactured via an energy intensive endothermic conversion of Butane. This process results in catalyst deactivation and low selectivity. Hence, an alternative, exothermic butadiene production pathway was proposed and is being studied extensively. In this study, an Artificial Neural Network (ANN) aided scale-up of a multi-tubular shell and tube reactor for the synthesis of Butadiene via the oxidative dehydrogenation of butylene was executed. A three-step solution strategy was adopted for the study. The step 1 comprised rigorous mathematical modelling of a lab-scale multi-tubular reactor using OpenFOAM. The developed model was used to generate data under varying process conditions (Feed composition and flowrates, coolant velocity, and temperature) for training an ANN surrogate model. The CFD results were validated against experimental data with minimal errors of less than 5%. In step 2, the generated data was fed to the ANN model for training and optimization. The optimized operating conditions were then used as inputs in the CFD model for simulation and cross-validation. Subsequently, the third and final step was implemented. In this step, the lab-scale multi-tubular reactor was scaled-up using the theory of dimensionality (Ï€-theory), conventional strategies, and domain expert knowledge. Several parametric studies were carried-out by varying the baffle height, number and spacing, tube arrangement, number, length and diameter as well as the shell side length and diameter. Each of these case studies were simulated for data generation using the CFD model by leveraging expertise in advanced meshing techniques to reduce the computational mesh density without compromising the accuracy and numerical stability of the simulation results. The generated data was again fed to an ANN model for optimization. The optimized reactor configuration was then designed and simulated with the CFD model for comparison and validation.