Digital Chemical Engineering最新文献

Effective water resource management is essential in large metropolitan cities. Digital Twins (DT), supported by IIoT and machine learning technologies, provide opportunities for real-time prediction and optimization for effective decision-making in water distribution systems. A framework for the digital twin of the Water Distribution Network (WDN) is developed in this paper to achieve higher operational efficiency using ‘WNTR’, the Python-based library of EPANET. All computational experiments and methods were validated on the benchmark hydraulic C-TOWN network (Ostfeld et al., 2011). The hydraulic parameters and quality parameters of the DT model for the water network were calibrated using the Differential Evolution (DE) algorithm. The calibrated DT served as a real-time proxy to generate simulation data, which is used for two different applications in large-scale water networks: (i) Disinfectant dosage regulation task using booster stations and (ii) pipe leakage localization task. The calibrated DT was utilized to estimate the optimal disinfectant dosing rates, ensuring water quality control within an acceptable range using optimization. The results highlight the effectiveness of the neural network and real-time optimization strategy to achieve the optimal dosing rate. For the leakage localization task, the Graph Convolution Networks (GCN) based neural network trained on the DT was found to predict leakage location very accurately.

The tight control of the process parameters through appropriate tuning of controllers is an art that imperatively employed to various process industries. Most of these industries are influenced by the nonlinearity that occurred due to the input parameter variation and presence of disturbances. The aim of this work is to investigate the nonlinear dynamics of a paper industry based energy intensive unit named Multiple Stage Evaporator (MSE) in presence of different Energy Reduction Schemes. MSE is used to concentrate the weak Black Liquor (BL), a biomass based byproduct. Hence, to extract the bioenergy from the BL, the quality of the product liquor needs to be appropriately controlled. The quality of BL is measured by two process parameters, product concentration and temperature. Hence, in this work, an intelligent controller Fraction Order Proportional-Integral-Derivative controller has been studied and employed to resolve the servo and the regulatory problem occurred during the process. A state-of-art metaheuristic approach, Black Widow Optimization Algorithm has been proposed here to tune the controller parameters and compared with another optimization approaches named Water Cycle Algorithm. The simulated result demonstrates the usefulness of the proposed strategy and confirm the performance improvement for the process parameters. To enlighten the advantages of the proposed control scheme, a comparative analysis have also been performed with conventional PID, 2-DOF-PID and FOPID controllers.

This work presents a cloud-based Virtual Flow Metering (VFM) system powered by a hybrid physics-data approach to estimate the water production per well in a gas field. This hybrid approach, which allows accurate calculations near real-time conditions, is based on the description of the flow through the wellbore using physics-based models pertaining to gas-liquid flows with high gas volume fraction. A data-driven approach is implemented to tune the flow model using well test data. This implementation accounts for changes in the well performance and increase in water production, resulting in a self-calibrating solution. This means that the model will remain accurate and relevant as production and well conditions change. Results from the VFM show good agreement with the well test data for steady-state conditions. The VFM calculations are performed remotely using a cloud-based DataOps platform where results are also stored. This allows continuous access to live sensor data to be used as input to other applications or visualized through a web interface. The VFM system uses a set of readily available sensors installed in the wells. Thus, it represents cost reduction in both capital and operating expenditures when compared to the installation of multiphase flow meters or separators.

The increasing demand for net zero solutions has prompted the exploration of electrochemical CO₂ reduction reaction (eCO₂RR) systems powered by renewable energy sources. Here, we present a comprehensive AI-enabled framework for the adaptive optimisation of the dynamic eCO₂RR processes in response to the intermittent renewable energy supply. The framework includes (1). a Bi-LSTM (bidirectional long-short-term memory) to predict the meteorological data for renewable energy input; (2). a deep learning surrogate model to predict the eCO₂RR process performance; and (3). a NSGA-II algorithm for multi-objective optimisation, targeting the trade-off of the single-pass Faraday efficiency (FE), product yield (PY) and conversion. The framework seamlessly integrates the three different AI modules, enabling adaptive optimisation of the eCO₂RR system composed of electrolyser stacks and renewable energy sources, and providing insights into system's performance and feasibility under real-world conditions.

This paper presents the utilization of the ESP32 microcontroller as a teaching tool for signal acquisition, processing, and control theory in biotechnological engineering. The ESP32 microcontroller, equipped with Bluetooth and WiFi capabilities, offers an affordable and versatile solution for educational purposes. By leveraging the Arduino© software, students can easily learn microcontroller programming and utilize various peripherals such as sensors and actuators. Several practical exercises related to process control have been conducted using this microcontroller. Additionally, remote process monitoring and control are enabled through integration with a database. Furthermore, concepts of artificial intelligence are explored using the Edge Impulse platform to obtain an artificial neural network that can be downloaded onto the ESP32. Positive feedback from students highlights the effectiveness and engagement of utilizing these microcontrollers, and the integration of internet connectivity enhances the overall learning experience.

Online monitoring of the spatial-temporal distribution of harmful gases has always been a complex problem in the environmental field. This paper proposes a novel mathematical method for online monitoring of the spatial-temporal distribution of reactants by machine learning, which can help to remove harmful gases efficiently. In this model, we take the advanced oxidation of NO as an example to evaluate the model performance. The spatial features were extracted by CNN, and GRU extracted the temporal features in the sequence of spatial features. Five physical field variables (mass fraction of ozone, velocity, temperature, the wind direction of the horizontal plane, and the wind direction of the vertical plane) were put into the network to predict NO's spatial-temporal mass fraction distribution. Furthermore, the impact of sampling time interval on monitoring performance was also evaluated. The results show that both the instantaneous and continuous CFD (Computational Fluid Mechanics) and predicted values show high consistency, which indicates that the model can online monitor the spatial-temporal distribution of reactants successfully. In addition, the most suitable sampling time interval is 0.5 s, with low training error (RMSE=0.06 and nRMSE=0.3) and high relation coefficient (r=0.99), which shows the model has great perceived and predicted performance under this condition.

A recent trend in chemical synthesis is photo-catalysis, which uses photo-active catalyst materials that are semiconductor materials. A well-known electronic property of semiconducting materials is the band gap. A photo-catalyst’s desired band gap range is between 1.5 eV and 6.2 eV. A rational design and synthesis of photo-active catalysts require knowledge of the band gap as an initial screening parameter. Herein, we propose an integrated deep learning-based framework to classify the photo-active catalysts and predict their band gap using compositional features. To this extent, we have utilized the dataset extracted from the “catalyst hub” site by web scraping with the help of a Python script. Extensive data cleaning and pre-processing are done to make input data amenable for training the models. Also, more valuable features are made using two methods: (a) one hot-encoding and (b) calculating the mean of the embeddings of catalysts computed by Mat2Vec, a pre-trained transformer-based model. With the help of this generated feature set, we have proposed a two-stage deep-learning framework for classification and regression tasks. In the first stage, a 2D-Convolutional Neural Net (CNN)-based classifier is used to classify whether a catalyst belongs to the photo-active catalyst class. After the first stage screening, in the second stage, we use a 1D-VGG-based gradient boosting framework to predict the band gap of the photo-active catalyst only using compositional features as inputs. 2D-CNN for the classification task has an accuracy of 0.903 and 0.886 for the train and test datasets, respectively. Further, the proposed integrated model that uses 1D-Convolutional layers of VGG followed by the XGBoostRegressor has a test $R^2$ of 0.750, much higher than baseline models reported in the literature.

Fluidized beds are central to numerous applications such as drying, combustion, gasification, pyrolysis, CO₂ utilization, mixing, and separation. The design and development of fluidized beds are still evolving owing to the complex hydrodynamics. Various experimental investigations and CFD simulations have been carried out to understand its hydrodynamics. Whereas the experimental approaches are very costly and limited to small scale, CFD modeling on the other hand requires significant computational resources and time. Thus, in this contribution, we propose a hybrid CFD-based ML model for estimating the hydrodynamics of fluidized beds. The CFD simulations of Taghipour et al., 2005 were performed and validated with the experimental measurements for a wide range of inlet gas velocities encompassing multiple flow regimes. A time-averaged simulation data of the CFD model was used for developing a Deep Neural Network (DNN) model. The hydrodynamic parameters, such as solid velocity field, volume fraction, and bed pressure drop, are predicted using the CFD-based DNN model. The results demonstrate that DNN has superior spatial learning capabilities and that, when used with CFD, it can reduce the computational power required without sacrificing accuracy. To evaluate the versatility of the CFDbased DNN model with different operating conditions and hydrodynamic parameters, independent data from (Cloete et al., 2013) and (Li and Zhang, 2013) were used for satisfactory validation.

Next-generation etching processes for semiconductor manufacturing exploit the potential of a variety of operating conditions, including cryogenic conditions at which high etch rates of silicon and very low etch rates of the photoresist are achieved. Thus, tight control of wafer temperature must be maintained. However, large and fast changes in the operating conditions make the wafer temperature control very challenging to be performed using typical etch cooling systems. The selection and evaluation of control tunings, material, and operating costs must be considered for next-generation etching processes under different operating strategies. These evaluations can be performed using digital twin environments (which we define in this paper to be a model that captures the major characteristics expected of a typical industrial process). Motivated by this, this project discusses the development of a computational fluid dynamics (CFD) model of a wafer temperature control (WTC) system that we will refer to as a “digital twin” due to its ability to capture major characteristics of typical wafer temperature control processes. The steps to develop the digital twin using the fluid simulation software ANSYS Fluent are described. Mesh and time independence tests are performed with a subsequent benchmark of the proposed ANSYS model with etch cooling system responses that meet expectations of a typical industrial cooling system. In addition, to quickly test different operating strategies, we propose a reduced-order model in Python based on ANSYS simulation data that is much faster to simulate than the ANSYS model itself. The reduced-order model captures the major features of the WTC system demonstrated in the CFD simulation results. Once the operating strategy is selected, this could be implemented in the digital twin using ANSYS to view flow and temperature profiles in depth.