Linking process-property relationships for multicomponent agglomerates using DEM-ANN-PBM coupling

Technological parameter optimization for walnut shell-kernel winnowing device based on neural network

The detection method for technological parameter is outdates as the traditional test cycle is long as well as the measurement error and the test amount are huge. Moreover, it is difficult to disclose the operation mechanism of devices as the operation is time-consuming and laborious. Therefore, numerical simulation was used in this study to reveal the mechanism of the walnut shell-kernel winnowing device. Moreover, the influence of baffle opening combinations, inlet wind velocity and inlet angle on cleaning rate and loss rate was predicted by the neural network model. The results demonstrated that inlet wind velocity was the primary influencing factor of cleaning rate, followed by baffle opening and inlet angle. Besides, inlet wind velocity was the primary influencing factor of loss rate, followed by inlet angle and baffle opening. The winnowing device performed best (79.91% cleaning rate, 14.37% loss rate) when the baffle opening, inlet wind velocity and inlet angle were 7.01 cm, 24.36 m/s, and 9.47°. In addition, 1/8 walnut shells and 1/4 walnut kernels were incorrectly classified due to the increase in inlet wind velocity. The inlet wind velocity was considered the major cause behind the deteriorating winnowing performance of the device. Finally, the bench test and simulation optimization results were compared. The cleaning rate and loss rate relative error during the simulation test was lower than 1.06%, which ascertained the feasibility and validity of the neural network as well as the combined numerical simulation method. This study could be useful for future research and development of shell-kernel winnowing devices for hard nuts.

Use of Multiscale Data-Driven Surrogate Models for Flowsheet Simulation of an Industrial Zeolite Production Process

The production of catalysts such as zeolites is a complex multiscale and multi-step process. Various material properties, such as particle size or moisture content, as well as operating parameters—e.g., temperature or amount and composition of input material flows—significantly affect the outcome of each process step, and hence determine the properties of the final product. Therefore, the design and optimization of such processes is a complex task, which can be greatly facilitated with the help of numerical simulations. This contribution presents a modeling framework for the dynamic flowsheet simulation of a zeolite production sequence consisting of four stages: precipitation in a batch reactor; concentration and washing in a block of centrifuges; formation of droplets and drying in a spray dryer; and burning organic residues in a chain of rotary kilns. Various techniques and methods were used to develop the applied models. For the synthesis in the reactor, a multistage strategy was used, comprising discrete element method simulations, data-driven surrogate modeling, and population balance modeling. The concentration and washing stage consisted of several multicompartment decanter centrifuges alternating with water mixers. The drying is described by a co–current spray dryer model developed by applying a two-dimensional population balance approach. For the rotary kilns, a multi-compartment model was used, which describes the gas–solid reaction in the counter–current solids and gas flows.

An efficient multiscale bi-directional PBM-DEM coupling framework to simulate one-dimensional aggregation mechanisms

The mesoscale population balance modelling (PBM) technique is widely used in predicting aggregation processes. The accuracy and efficiency of PBM depend on the formulation of its kernels. A model of the volume- and time-dependent one-dimensional aggregation kernel is developed for predicting the temporal evolution of the considered particulate system. To make the developed model physically relevant, the PBM model needs three unknown parameters as input: volume-dependency in collisions, collision frequency per particle and aggregation probability. For this, the microscale discrete element model (DEM) is used. The system’s collision frequency is extracted periodically using a novel collision detection algorithm that detects and ignores duplicate collisions.

Finally, a multiscale bi-directional PBM–DEM coupling framework is presented to simulate the aggregation mechanism. PBM and DEM simulations take place periodically to update the particle size distribution (PSD) and extract the collision-frequency, respectively. The coupling framework successfully explains the dependence between the PSD and the collision frequency. Additionally, computational cost of the algorithm is optimized while maintaining the accuracy of the results. Lastly, the accuracy and efficiency of the developed framework are verified using two different test cases. In one of the examples, a simple aggregation is simulated directly inside the DEM for the first time.