Modelling of a pressure swing adsorption unit by deep learning and artificial Intelligence tools

Oliveira, Luis M. C.; Koivisto, Hannu; Iwakiri, Igor G.I.; Loureiro, José M.; Ribeiro, Ana M.; Nogueira, Idelfonso B. R.

doi:10.1016/j.ces.2020.115801

Cited by 31 publications

(20 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this work, this procedure will be used with the NOE and NARX structures to determine the best number of input and output delays to be used in each predictor. The value p = 0.01 N was chosen following the literature recommendation [18]. Figure 5 presents the results obtained after applying the Lipschitz analysis on the synthetic data obtained from the virtual plant.…”

Section: Embedding Dimensions Optimal Selectionmentioning

confidence: 99%

“…On the other hand, it presents the identification of machine learning (ML) models to simulate the dynamic behavior PSA unit. The machine learning strategies employed here were the traditional feedforward neural network (FNN), a recurrent neural network (RNN), and a deep neural network (DNN) [18]. Therefore, three main contributions of this work are highlighted.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Machine Learning-Based Dynamic Modeling for Process Engineering Applications: A Guideline for Simulation and Prediction from Perceptron to Deep Learning

et al. 2022

Self Cite

View full text Add to dashboard Cite

A misusage of machine learning (ML) strategies is usually observed in the process systems engineering literature. This issue is even more evident when dynamic identification is performed. The root of this problem is the gradient explode and vanishing issue related to the recurrent neural networks training. However, after the advent of deep learning, these issues were mitigated. Furthermore, the problem of data structuration is often overlooked during the machine learning model identification in this field. In this scenario, this work proposes a guideline for identifying ML models for the different applications in process systems engineering, which are usually for simulation or prediction purposes. While using the proposed guideline, the work also identifies a virtual analyzer for a pressure swing adsorption unit. In these types of adsorption separations, it is usual that the measurement of the main properties is not done online. Therefore, the virtual analyzer is another contribution of this manuscript. The overall results demonstrate that even though the test provides good performance during the ML model identification, its quality might degenerate over the application domain if the model application is overlooked.

show abstract

Section: Embedding Dimensions Optimal Selectionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Machine Learning-Based Dynamic Modeling for Process Engineering Applications: A Guideline for Simulation and Prediction from Perceptron to Deep Learning

et al. 2022

Self Cite

View full text Add to dashboard Cite

show abstract

“…Some optimization methods have been used in PSA, such as sequential quadratic programming and the single discretization method; however, these methods are complex and require time-consuming calculations [26,27]. Multiple artificial intelligence models have been applied to predict the performance of PSA cycles, and research results show that the deep learning model has the best predictive effect [28]. Dual-and tri-objective optimizations were applied to a four-bed, eight-step PSA model to produce hydrogen from an SMR gas mixture [29].…”

Section: Introductionmentioning

confidence: 99%

Hydrogen Purification Performance Optimization of Vacuum Pressure Swing Adsorption on Different Activated Carbons

et al. 2021

View full text Add to dashboard Cite

Hydrogen purification is an important part of hydrogen energy utilization. This study aimed to perform hydrogen purification of multi-component gas (H2/CO2/CH4/CO/N2 = 0.79/0.17/0.021/0.012/0.007) by one-column vacuum pressure swing adsorption (VPSA) and pressure swing adsorption (PSA). AC5-KS was selected as the adsorbent for hydrogen purification due to its greater adsorption capacity compared to R2030. Furthermore, VPSA and PSA 10-step cycle models were established to simulate the hydrogen purification process using the Aspen Adsorption platform. The simulation results showed that the hydrogen purification performance of VPSA is better than that of PSA on AC5-KS adsorbent. The effects of feeding time and purging time on hydrogen purity and recovery were also discussed. Results showed that feeding time has a negative effect on hydrogen purity and a positive effect on hydrogen recovery, while purging time has a positive effect on hydrogen purity and a negative effect on hydrogen recovery. By using an artificial neural network (ANN), the relationship between the inputs (feeding time and purging time) and outputs (hydrogen purity and recovery) was established. Based on the ANN, the interior point method was applied to optimize hydrogen purification performance. Considering two optimization cases, the optimized feeding time and purging time were obtained. The optimization results showed that the maximum hydrogen recovery reached 88.65% when the feeding time was 223 s and the purging time was 96 s. The maximum hydrogen purity reached 99.33% when the feeding time was 100 s and the purging time was 45 s.

show abstract

“…Pai et al 15 extended the use of feed-forward ANN models to predict the axial profiles of the intensive variables for a four-step VSA process at CSS, and the models were experimentally validated. Furthermore, Oliveira et al 16 developed a real-time soft sensor for a PSA unit based on neural network models. Three types of ANN architectures, namely, feed-forward, recurrent, and long short-term memory (LSTM) based on multi-input and a single output, are used to predict the PSA process performance over the number of cycles.…”

Section: Introductionmentioning

confidence: 99%

Physics-based neural networks for simulation and synthesis of cyclic adsorption processes

Subraveti

Prasad

et al. 2021

Preprint

View full text Add to dashboard Cite

A computationally faster and reliable modelling approach called a physics-based artificial neural network framework for adsorption and chromatography emulation (PANACHE) is developed. PANACHE uses deep neural networks for cycle synthesis and simulation of cyclic adsorption processes. The proposed approach focuses on learning the underlying governing partial differential equations in the form of a physics-constrained loss function to simulate adsorption processes accurately. The methodology developed herein does not require any system-specific inputs such as isotherm parameters. Accordingly, unique neural network models were built to fully predict the column dynamics of different constituent steps based on unique boundary conditions that are typically encountered in adsorption processes. The trained neural network model for each constituent step aims to predict the entire spatiotemporal solutions of different state variables by obeying the underlying physical laws. The proposed approach is tested by constructing and simulating four different vacuum swing adsorption cycles for post-combustion CO2 capture without retraining the neural network models. For each cycle, 50 simulations, each corresponding to a unique set of operating conditions, are carried out until the cyclic-steady state. The results demonstrated that the purity and recovery calculated from the neural network-based simulations are within 2.5% of the detailed model's predictions. PANACHE reduced computational times by 100 times while maintaining similar accuracy of the detailed model simulations.

show abstract

Modelling of a pressure swing adsorption unit by deep learning and artificial Intelligence tools

Cited by 31 publications

References 31 publications

Machine Learning-Based Dynamic Modeling for Process Engineering Applications: A Guideline for Simulation and Prediction from Perceptron to Deep Learning

Machine Learning-Based Dynamic Modeling for Process Engineering Applications: A Guideline for Simulation and Prediction from Perceptron to Deep Learning

Hydrogen Purification Performance Optimization of Vacuum Pressure Swing Adsorption on Different Activated Carbons

Physics-based neural networks for simulation and synthesis of cyclic adsorption processes

Contact Info

Product

Resources

About