Digital Chemical Engineering最新文献

This study aims to predict the solubility of salicylic acid in 13 different solvents, such as methanol, water, ethanol, ethyl acetate, PEG 300, 1,4-dioxane, 1-propanol, and others, given the significance of salicylic acid in the pharmaceutical industry. based on machine learning has been studied. In this study, 6 machine learning algorithms including neural network, linear regression, logistic regression, decision tree, random forest and kNN (k- Nearest Neighbors) were used. The comparison between the predictions of these algorithms and experimental data highlights the accuracy of predicting the solubility of salicylic acid for 217 samples based on 15 variables (13 solvents, temperature, and pressure). Based on the results of this study, the lowest total error (difference between experimental and predicted values) was 0.00016835 related to the random forest algorithm, and the highest value was 0.024768 related to k-Nearest Neighbors.

The demand for engineering graduates with technical skills in data science, machine learning (ML), and artificial intelligence (AI) is now growing. Chemical engineering (ChemE) departments around the world are currently addressing this skills gap by instituting AI or ML elective courses in their program. However, designing such a course is difficult since the issue of which ML models to teach and the depth of theory to be discussed remains unclear. In this paper, we present a graduate-level ML course particularly designed such that students will be able to apply ML for research in ChemE. To achieve this, the course intends to cover a wide selection of ML models with emphasis on their motivations, derivations, and training algorithms, followed by their applications to ChemE-related data sets. We argue that this algorithmic approach to teaching ML can help broaden the capabilities of students since they can judge for themselves which tool to use when, even for problems outside the process industries, or they can modify the methods to test novel ideas. We found that students remain engaged in the mathematical details as long as every topic is properly motivated and the gaps in the required statistical and computer science concepts are filled. Hence, this paper also presents a roadmap of ML topics, their motivations, and bridging topics that can be followed by instructors. Lastly, we report anonymized student feedback on this course which is being offered at the Department of Chemical Engineering, University of the Philippines, Diliman.

A powerful fault detection and diagnosis (FDD) system plays a pivotal role in achieving operational excellence by maximizing system performance, optimizing maintenance strategies, and ensuring the longevity and resilience of process plants. In the context of FDD for multivariate sensor data, this study presents an improved FDD approach using graph-based neural networks. This graph neural network uses an adjacency matrix developed by extracting the expert domain knowledge and topological information of the multi-sensor system. This additional graph representation of the system is incorporated along with multivariate sensor data to capture the spatial and temporal information in neural networks efficiently. In this regard, we propose and evaluate: 1) A Graph Auto Encoder (GAE) based fault detection strategy and 2) An Attention-based Spatial Temporal Graph Convolution Network (ASTGCN) based fault diagnosis methodology. By leveraging the additional knowledge in the form of graphs, the GAE captures the complex relationships and dependencies among sensors, enabling effective anomaly detection, which identifies abnormal patterns and deviations from normal behavior, thus indicating potential faults in the system. The ASTGCN incorporates attention mechanisms to selectively focus on relevant sensor nodes and capture their spatial and temporal dependencies for fault diagnosis. The effectiveness of the proposed FDD approach is demonstrated using the benchmark Tennessee Eastman Process (TEP) problem. The results show that the proposed approaches outperform traditional methods and highlight the importance of leveraging graph-based knowledge for FDD in complex systems.

The oxidative coupling of methane (OCM) to produce ethane and ethylene (C2 compounds) as platform chemicals involves complex chemistry with reactions both in the gas phase and on the catalyst surface, resulting in a distribution of products at the expense of C2 selectivity. This work uses experimental data from a variety of mixed metal oxides on supports at different reaction conditions (temperature, contact time, and reactant flow rates) to train a random forest regressor that predicts methane conversion and C2 selectivity (key performance indicators (KPIs)). The kinetically validated random forest models are deployed to locate optimal conditions that maximize C2 yield for each of the catalysts. Investigating the regressor interpretability via feature importance reveals that the choice of metals and support are crucial to C2 selectivity predictions in addition to the reaction conditions, while the predictions of methane conversion are largely governed by the reaction conditions. The machine learning (ML) regressor is used as a kinetic surrogate to find a locus of optimal reaction conditions that maximize both selectivity-conversion for each of the catalysts via a multi-objective optimization routine. The maximum C2 yields for catalysts are projected to be improved by 15% on average. Analyzing the catalysts with respect to a popular OCM catalyst, Mn-Na${}_2$WO${}_4$/SiO${}_2$, using the optimal locus eliminates variability in the process conditions to reveal distinct patterns based on intrinsic properties of metals and supports. Further, the decision space with catalyst descriptors and reaction conditions is optimized for high C2 yields using the ML surrogate, in a static multi-objective optimization routine, and an adaptive Bayesian routine, where the latter was found to have a wider field focus in proposing catalyst formulations and reaction conditions. Transition metal oxides on a variety of supports were proposed but not their lanthanide oxide counterparts. The framework has the potential to lend itself to materials acceleration platforms where it is crucial to consider multi-scale phenomena that impact downstream KPIs.

Fault detection constitutes a fundamental task for predictive maintenance, requiring mathematical models that can be conveniently provided by data-driven techniques. Autoencoders are a particular type of unsupervised Artificial Neural Networks that can be suitable for fault detection applications. Diverse architectures might be used for autoencoders, resulting in different fault detection performances, which are usually compared by means of Fault Detection Rates for a fixed threshold of the False Alarm Rate, limiting the conclusions to particular cases. To improve the comparability, the present work uses the area under the receiver operating characteristic curve to compare autoencoder architectures for a range of false alarm rates using the Tennessee Eastman Process benchmark. Performances obtained for shallow and deep autoencoders were compared with those of the denoising and variational autoencoders for undercomplete and sparse structures. Overall, the results indicate better performances for sparse structures, especially for the variational autoencoder and the deep denoising autoencoder, with area under the curve of 98.35%.

Unexpected accidents and events in industrial chemical processes have resulted in a considerable number of causalities and property damage. Safety process management in industrial chemical processes is critical to avoid and ensure casualties and property damage. However, due to the immense scope and high complexity of current industrial chemical processes, the traditional safety process management approaches cannot address these challenges to attain adequate fault detection accuracy. To address this issue, an innovative machine learning-based distributed canonical correlation analysis-dynamic independent component analysis (DICA-DCCA) approach is needed to improve the fault detection effectiveness of complicated systems. The (DICA-DCCA) model could potentially detect anomalies and faults in industrial chemical data by utilizing three essential statistics:$I_d^2$,$I_e^2$and squared prediction error (SPE). The practical effectiveness of the proposed frameworks is evaluated and compared using a continuous stirred tank reactor (CSTR) framework as a standard benchmark study. The research findings present that the suggested (DICA-DCCA) approach is more resilient and effective in detecting abnormalities and faults than the ICA and DICA approaches with FDR 100 % and FAR 0 %. The implied research approach is robust, operational, and productive.

This paper presents a novel approach to address computational challenges in predicting flow features by employing a Graph Neural Network (GNN), which is proficient in predicting flow domain values. Traditional Computational Fluid Dynamics (CFD) simulations, although effective, often require substantial computational resources and time, limiting their applicability in time-sensitive scenarios and optimization studies necessitating extensive case studies. The main objective was to evaluate the feasibility of employing node classification on a graph generated from a 2D baffle flow system to segment the domain based on relative fluid age. A second objective was to compare the computational time required for CFD simulations with the inference time of the network to quantify the efficiency gains achieved by utilizing the network. Results demonstrate the potential of utilizing graph convolutional networks for domain segmentation to predict regions of holdup and bypass. The GNN achieved 97% and 92% accuracy in predicting recirculation regions in single and double baffle cases, particularly excelling in high Reynolds number scenarios. Importantly, the proposed GNN-based approach reduces computation time by over 2100%, showcasing significant efficiency gains. Results here highlight the promise of employing graph convolutional networks for flow feature prediction, offering substantial computational efficiency improvements over traditional CFD simulations.

The surge in demand for hydrogen (H₂) across diverse sectors, including clean energy transportation and chemical synthesis, underscores the need for a thorough investigation into H₂ production dynamics and the development of effective controllers for industrial applications. This paper focuses on an electrically heated steam methane reforming (SMR) process for H₂ production, offering advantages such as enhanced environmental sustainability, compactness, efficiency, and controllability compared to conventional reforming methods. Electric heating of the entire system allows for adjustments in current to control reactor temperature, thereby impacting hydrogen production rates. However, accurately modeling hydrogen production dynamics presents a formidable challenge, as complex models with high precision are computationally unsuitable for real-time control integration. Considering these factors, an accurate and efficient first-principles-based lumped-parameter model is developed to provide a dependable estimation of hydrogen production in an electrically-heated steam methane reformer. This model is validated experimentally and then utilized in a model predictive controller (MPC). To obtain the necessary state estimate information for the MPC, an extended Luenberger observer (ELO) method is employed to estimate state variables from limited, infrequent and delayed measurements of gas-phase reactor outlet stream and frequent measurements of the reactor temperature. Simulation comparisons with a proportional-integral (PI) controller reveal a much faster response in achieving the desired H₂ production rate under the estimation-based MPC. Additionally, the simulations demonstrate the robustness of the controller to process variability such as a decrease in catalyst activation energy, commonly encountered in the SMR process, highlighting its effectiveness in maintaining stable operation under varying process conditions.

Machine learning methodologies are gaining significant recognition as an effective approach for tackling and modelling challenges related to membranes. This study delves into the utilization of machine learning algorithms to forecast the quality of reverse osmosis (RO) water. Specifically, we conduct a comparative analysis of four popular algorithms: decision tree, random forest, support vector machine (SVM), and K-nearest neighbours (KNN). Our dataset comprises essential water quality evaluation features such as temperature, pH, and conductivity. Using these features, we train and test our models, evaluating their performance with metrics like accuracy and root-mean-squared error (RMSE). The outcomes indicate that all four algorithms perform admirably in predicting RO water quality, achieving accuracy scores ranging from 80 % to 95 %. Notably, KNN stands out with the highest accuracy score of 95 %, establishing it as the most effective algorithm for this task. Besides its performance, KNN's simplicity of implementation and interpretability make it a pragmatic choice for real-world applications. This study serves as compelling evidence of the potential of machine learning algorithms for forecasting RO water quality, particularly highlighting KNN's effectiveness in this context. To further enhance the accuracy of RO water quality prediction, future research could explore the inclusion of other features or alternative algorithms.

This perspective article highlights our opinions on the imperative of incorporating Process Intensification (PI) into undergraduate chemical engineering education, recognizing its pivotal role in preparing future engineers for contemporary industrial challenges. The trajectory of PI, from historical milestones to its significance in advancing the United Nations’ Sustainable Development Goals (SDGs), reflects its intrinsic alignment with sustainability, resource efficiency, and environmental stewardship. Despite its critical relevance, the absence of dedicated PI courses in numerous undergraduate chemical engineering programs presents an opportunity for educational enhancement. An exploration of global PI-related courses reveals the potential of educational platforms to fill this void. To address this gap, we advocate for the introduction of a standalone PI course as a minor elective, minimizing disruptions to established curricula while acknowledging the scarcity of PI expertise. The challenges associated with PI integration encompass faculty workload, specialized expertise, curriculum content standardization, and industry alignment. Surmounting these challenges necessitates collaborative efforts among academia, industry stakeholders, and policymakers, emphasizing the manifold benefits of PI, faculty development initiatives, and the establishment of continuous improvement mechanisms. The incorporation of PI into curricula signifies a transformative approach, cultivating a cadre of innovative engineers poised to meet the demands of the evolving industrial landscape.