Prediction Of Product Yield Research Articles

A hybrid soft sensor for key product yield of FCC unit based on deep learning framework driven by data and process mechanism

The model for online prediction of product yield embedded in the optimization and control framework has the potential to improve the profits of fluid catalytic cracking (FCC) units. In this work, a hybrid soft sensor model is developed to predict key product yield in commercial FCC units. The deep learning framework based on semi-supervised learning is driven by data and process mechanism to construct the hybrid model with a double-layer structure. Moreover, process simulation strategy is utilized for key input variable selection of the hybrid model to ensure model reliability. The hybrid model is compared to a pure data-driven model in terms of accuracy and trend consistency of model predictions. The results demonstrate that the hybrid model exhibits a superior prediction performance. This work suggests that the hybrid model can drive innovation and industrial practice in the optimization and control of FCC units.

A multiscale adaptive framework based on convolutional neural network: Application to fluid catalytic cracking product yield prediction

Since chemical processes are highly non-linear and multiscale, it is vital to deeply mine the multiscale coupling relationships embedded in the massive process data for the prediction and anomaly tracing of crucial process parameters and production indicators. While the integrated method of adaptive signal decomposition combined with time series models could effectively predict process variables, it does have limitations in capturing the high-frequency detail of the operation state when applied to complex chemical processes. In light of this, a novel Multiscale Multi-radius Multi-step Convolutional Neural Network (MsrtNet) is proposed for mining spatiotemporal multiscale information. First, the industrial data from the Fluid Catalytic Cracking (FCC) process decomposition using Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) extract the multi-energy scale information of the feature subset. Then, convolution kernels with varying stride and padding structures are established to decouple the long-period operation process information encapsulated within the multi-energy scale data. Finally, a reconciliation network is trained to reconstruct the multiscale prediction results and obtain the final output. MsrtNet is initially assessed for its capability to untangle the spatiotemporal multiscale relationships among variables in the Tennessee Eastman Process (TEP). Subsequently, the performance of MsrtNet is evaluated in predicting product yield for a 2.80 × 106 t/a FCC unit, taking diesel and gasoline yield as examples. In conclusion, MsrtNet can decouple and effectively extract spatiotemporal multiscale information from chemical process data and achieve a approximately reduction of 30% in prediction error compared to other time-series models. Furthermore, its robustness and transferability underscore its promising potential for broader applications.

Prediction of product yields from lignocellulosic biomass pyrolysis based on gaussian process regression

The physicochemical characteristics of biomass, as well as the operational conditions of pyrolysis, are crucial factors that influence the production yield of tri-phase products during the pyrolysis of lignocellulosic biomass. Gaussian process regression (GPR) algorithms, being significant machine learning algorithms, are utilized for the prediction of pyrolytic product yields. These algorithms take into account the influence of biomass characteristics and pyrolysis conditions in a comprehensive manner. All of the prediction models exhibit strong performance, with R2 values exceeding 0.90 and RMSE values below 0.18. The study employed the Partial Dependence Plot (PDP) method to assess the influence patterns of individual features or interactions between two factors on the pyrolysis products. The findings indicate that the ultimate temperature reached during the pyrolysis process is a critical factor in influencing the generation of gaseous byproducts and solid residues. Specifically, higher temperatures are associated with increased production of gaseous byproducts and decreased production of solid residues. The optimal liquid phase yield was achieved at temperatures between 500 °C and 650 °C. In this research, the Particle Swarm Optimization (PSO) algorithm was employed to predict the optimal yields of pyrolysis products. This study offers a novel perspective on predicting product yields from the pyrolysis of lignocellulosic biomass with varying biomass characteristics and under different operational conditions.

Research on Product Yield Prediction and Benefit of Tuning Diesel Hydrogenation Conversion Device Based on Data-Driven System

In the refining process, a large amount of data are generated in daily production, and how to make full use of these data to improve the accuracy of simulation is the key to improving the operation level of refineries. At the same time, with the increasing environmental regulations and the improvement of gasoline and diesel quality standards, the ratio of diesel to gasoline is also changing with people’s demand for fuel consumption. Catalytic cracking light cycle oil (LCO) hydrogenation conversion technology (react LCO into gasoline, RLG) can produce modified diesel with high-octane gasoline, a high cetane number, and a low sulfur content, which improves the added value of the product. In this article, based on the production and operation data of a 1 million tons/year RLG device, a device yield prediction model was established using a deep neural network (DNN) algorithm, and the model was further optimized using a genetic algorithm (GA) to maximize the economic benefits of the device. As a result, the gasoline production yield increased by more than 3%. The experimental results show that the established model has a good reference value for improving the economic benefits of the RLG device.

A novel strategy for rapid identification of pyrolytic synergy and prediction of product yield: Insight into co-pyrolysis of xylan and polyethylene

The distributions of biomass and plastic co-pyrolysis products are complicated by abundant component combinations, pyrolysis conditions, and synergies. Herein, the hierarchical clustering analysis (HCA) and response surface methodology (RSM) were used to rapidly determine synergies and predict product yields of xylan and polyethylene (PE) co-pyrolysis at 500–700 °C. The results showed that co-pyrolysis promoted liquid production and suppressed solid and char formation. Pyrolytic interactions improved the decomposition of the PE-derived wax, resulting in 1.5–1.9- and 1.7–2.1-fold higher yields of heavy gas oil and C≥26 hydrocarbons compared to the theoretical values. HCA classified pyrolyzates with similar synergy into the same cluster, which reflected the suppressed carbonyl compound production, enhanced furfural and phenols yields at 700 °C, and greater C17–C30 hydrocarbon production. The quadratic model of RSM predicted the yields of gas, CO, C3Hn, C4Hn, liquid, ethanol, acetaldehyde, hydrocarbon oil, gasoline, solid, and char influenced by synergies. Owing to complex interactions, the cubic model fitted the CH4 and C2H4 yields. The linear model described the CO2 yield without synergy. This work illustrates the utility of combining RSM and HCA to predict product distributions of various waste co-treatment processes.

The modeling and products prediction for biomass oxidative pyrolysis based on PSO-ANN method: An artificial intelligence algorithm approach

In this work, the oxidative pyrolysis of two kinds of biomass was conducted on a fixed bed reactor. The pyrolysis product yields (water, tar, gas and char) as well as two kinds of gaseous species (CO and CO2) were analyzed. Based on the experiments, the artificial neural network (ANN) model was built and trained for product yield prediction. Experimental results show that the reaction mechanism of oxidative pyrolysis was much complicated than inert pyrolysis. With the addition of oxygen, homogenous and heterogenous oxidation of volatiles or nascent char significantly influenced the product distribution. Modeling results showed that the combination transfer function of logsig and purelin for hidden layer was suitable for modeling of biomass oxidative pyrolysis. Individual ANN model had weak prediction performance for the data points located around the boundary of database. This weakness as well as the tendency of local convergence leaded to the decline of prediction accuracy for individual ANN model. Particle swarm optimization (PSO) was further used to optimize the initial weight and threshold value of the ANN model, which significantly increased the prediction accuracy. The relative error of the PSO-ANN model decreased to below 10%, which proved to be a effective tool for modeling of biomass oxidative pyrolysis.

Improving Hydrothermal Carbonization (htc) Processes by Hydrochar Gasification

Hydrochar is produced by means of hydrothermal carbonization at relatively low temperatures (180-260°C) in sub-critical water. While, in many respects similar to biochar, its physical and chemical properties differ significantly (Basu, 2018). Of particular interest for this work are the higher energy density and lower ash content, that make hydrochar a possible feedstock for gasification plants with energy generation purposes. A gasification step is used instead of direct combustion to convert solid fuels in energy to achieve cleaner combustion and higher efficiency (Wang and Stiegel, 2016). In this work a 400 kg/h hydrochar gasification plant was modelled to identify optimal conditions and energetic yield of hydrochar obtained from municipal sewage. The fixed bed, updraft, gasification reactor was modeled in detail using a multi-scale, multiphase methodology already widely tested on biomass (Corbetta et al., 2015; Ranzi et al., 2014). The gas solid kinetic model was coupled with a detailed gas-phase kinetic scheme with over 200 species, including reaction intermediates, and 2000 reactions for reliable product yield prediction. Using Visual Basic Application as an interface, the predictions from the detailed simulation package were delivered to a commercial simulation package to model the energy generation section of the plant. Aspen HYSYS V10 was used for this purpose for the simplicity of integration and its widespread use in similar industrial plants. The gasification was carried out with air, air enriched in oxygen to 28%, air enriched in oxygen to 35% and pure oxygen with different amounts of steam to control the temperature in the chamber and at different values of equivalence ratio. The gasification performance was evaluated in terms of lower heating value of the generated fuel gas while the H2S formation was accounted for only in a superficial manner using rules of thumb derived from previous experimental experience.

Multivariate statistical data analysis of cell‐free protein synthesis toward monitoring and control

AbstractThe optimization and control of cell free protein synthesis (CFPS) presents an ongoing challenge due to the complex synergies and nonlinearities that cannot be fully explained in first principle models. This article explores the use of multivariate statistical tools for analyzing data sets collected from the CFPS of Cereulide monoclonal antibodies. During the collection of these data sets, several of the process parameters were modified to investigate their effect on the end‐point product (yield). Through the application of principal component analysis and partial least squares (PLS), important correlations in the process could be identified. For example, yield had a positive correlation with pH and NH3 and a negative correlation with CO2 and dissolved oxygen. It was also found that PLS was able to provide a long‐term prediction of product yield. The presented work illustrates that multivariate statistical techniques provide important insights that can help support the operation and control of CFPS processes.

Adaptive Modeling Strategy Integrating Feature Selection and Random Forest for Fluid Catalytic Cracking Processes

This study proposes a hybrid approach for the modeling of the fluid catalytic cracking (FCC) process, with the aim to establish an adaptive and accurate product yield prediction model. Because of t...

Ethane Steam Cracking Inferential Model Development using Artificial Neural Network

Abstract During normal operation, the product yield distribution of ethane steam cracking is obtained using analyzers and lab sampling to measure the cracking severity. Since analyzers take time to produce result, primarily depending on analyzers and lab testing to determine main product yield will hinder immediate control action to the process. In this study, data generated from ASPEN Plus Dynamics is used as feed data to develop the inferential model for product yield prediction based on real time operating parameters. The finalized model is a 4x2 ANN model, with 10 hidden nodes. Mean Squared Error (MSE) analysis shows that the model can accurately predict ethane and ethylene yields of the process, with MSE of 2.78x10-5 and 3.41x10-6, respectively. On top of that, R-squared analysis shows goodness-of-fit of 99.4% and 98.4%, respectively.

Advanced models for the prediction of product yield in hydrothermal liquefaction via a mixture design of biomass model components coupled with process variables

Hydrothermal liquefaction (HTL) has recently attracted great interest as a thermochemical conversion technique for biofuels production, however, suffers a lack of broadly applicable models for the prediction of product yield. This study developed a unique model for the prediction of HTL products yield via a mixture design of biomass model components coupled with process variables. The model compounds used in this study were soya protein for a protein representative, a mixture of cellulose and xylan for a saccharide representative, alkaline lignin for a lignin representative and soybean oil for a lipid representative. Reaction temperature (270–320 °C), time (5–20 min) and mass ratio of water/feedstocks (6:1–12:1) were chosen as the process variables of interest. The developed predictive models for biocrude yield and solid residue yield showed accuracy of (R2adj 94.6% and 93.2%, respectively), and were further validated using modelled feedstock and actual feedstock. These models can be used either to optimize HTL conditions when feedstock is known, or to optimize the composition of feedstock when reaction conditions are given. It was also observed that within the experimental design range, relatively mild HTL conditions eliminated alkaline lignin-lipid interaction and protein-lipid interaction, and thus enhanced biocrude formation; while more severe HTL conditions were preferred to reduce solid residue formation through promoting protein-saccharide interaction and saccharide-alkaline lignin interaction.

Hydrothermal liquefaction of biomass model components for product yield prediction and reaction pathways exploration

Hydrothermal liquefaction (HTL) is a promising technology for crude bio-oil production from a variety of biomass, however, there is a lack of prediction models for the yield of products and the reaction pathways is not well understood. Prediction models for biocrude yield and solid residue (SR) yield were developed by using a mixture design of five model components, including xylan (hemicellulose), crystalline cellulose, alkaline lignin, soya protein and soybean oil in this study. The model predictability was verified by using actual feedstock as well as a mixture of model components based on the chemical composition of the feedstock of concern. The biocrude yield, solid residue yield and quantitative chemical yields obtained from bio-oil were used to explore the reaction pathways as well as possibly existing synergistic and/or antagonistic interactions between two studied model components. It was found that both hemicellulose and lipid (H∗Lip) and cellulose and lipid (C∗Lip) interactions had synergistic effect on the biocrude yield, while SR yield was antagonistically decreased by the cellulose and lignin (C∗Lig) interaction. Maillard reactions between protein and carbohydrates and amide formation between protein and lipid were observed. The carbohydrates and lipid interactions had effects on the acid yield (in H∗Lip), hydrocarbon yield and ketone yield (in C∗Lip), but lignin and lipid (in Lig∗Lip) behaved independently in the HTL processes. The findings of this research can be used to assess the potential of various kinds of biomass, provide guidance for using mixed biomass (co-liquefaction) and tailor the chemical composition of feedstock for a desirable product distribution in HTL processes.

Slow pyrolysis polygeneration of bamboo (Phyllostachys pubescens): Product yield prediction and biochar formation mechanism

Slow pyrolysis of bamboo was conducted at 400–600 °C and pyrolysis products were characterized with FTIR, BET, XRD, SEM, EDS and GC to establish a pyrolysis product yield prediction model and biochar formation mechanism. Pyrolysis biochar yield was predicted based on content of cellulose, hemicellulose and lignin in biomass with their carbonization index of 0.20, 0.35 and 0.45. The formation mechanism of porous structure in pyrolysis biochar was established based on its physicochemical property evolution and emission characteristics of pyrolysis gas. The main components (cellulose, hemicellulose and lignin) had different pyrolysis or chemical reaction pathways to biochar. Lignin had higher aromatic structure, which resulted higher biochar yield. It was the main biochar precursor during biomass pyrolysis. Cellulose was likely to improve porous structure of pyrolysis biochar due to its high mass loss percentage. Higher pyrolysis temperatures (600 °C) promoted inter- and intra-molecular condensation reactions and aromaticity in biochar.

Prediction of Product Yields for Different Reactor Types in Catalytic Pyrolysis

AbstractFür den Wandel hin zu einer postfossilen Gesellschaft werden rohstoffvariable Verfahren wie die katalytische Pyrolyse von festen Kohlenwasserstoffen benötigt. Zur Vorhersage der Produktausbeute aus unterschiedlichen Einsatzstoffen wurden Modellansätze entwickelt, die Rohstoffqualität, Katalysatoreigenschaften und Prozessbedingungen berücksichtigen. Eine gute Beschreibung für die stoffliche Nutzung von Braunkohle ist mit wenigen Einflussgrößen und signifikanten Modellparametern möglich. Perspektivisch sollen solche Modelle auf die Spaltung polyolefinischer Rezyklate und fettreiche Biomassen übertragen werden.

Analysis of the production of salmon fillet – Prediction of production yield

The aim was to investigate the influence of raw material variation in Atlantic salmon from aquaculture on filleting yield, and to develop a decision tool for choosing the appropriate raw material for optimized yield. This was achieved by tracking salmon on an individual level (n = 60) through a primary production site. The majority of the salmon exhibited a heavier right fillet compared to the left fillet after filleting. No explicit explanation was found for this observation although the heading procedure was shown to have a large impact. A Partial Least Square model was built to predict the yield after filleting. The model was based on six pre-processing variables and allowed an acceptable prediction of the filleting yield with a root mean square error cross validation of 0.68. The presented model can estimate the slaughter yield for a certain batch before ordering from the slaughterhouse. This may facilitate optimal planning of the production of salmon fillets by ordering and assigning the right batch to the right product category to obtain an optimal yield and quality.

Mathematical modeling of unsteady-state gasification of petroleum residue

ABSTRACTGasification is a thermochemical process which converts organic fuels into a high caloric value syngas and other chemicals in the presence of a gasification agent. Tar generation represents the strongest barrier for the use of fixed-bed reactors for liquid fuel gasification, whereas sufficing is only possible with catalytic activities and expensive physical processing. In this work, a kinetic model of waste oil gasification was proposed which can be used a flexible model to provide a quantitative prediction of product yield of other fuels. Results were validated against the experimental measurements and showed a good agreement. In accordance with the modeling results, it was found that the greenhouse gas emission (GGE) of waste oil is higher than that of biomass materials, but the caloric value of the syngas generated is higher. The residence time was also recognized as an important design parameter to improve the syngas volume fraction.

Estimation of Biodiesel Yield using Fast Decorrelated Neural Network Ensemble

The accurate and reliable online measurement of the product yield is very essential for the control and optimization of the biodiesel process. A biodiesel yield prediction model based on the fast decorrelated neural network ensembles (FDNNE) was established to enhance the estimated performance. The random vector functional link (RVFL) networks were inserted into the fast decorrelated neural network ensemble frame as the base model since it could provide better generalized performance and faster speed. The FDNNE product yield prediction model initializes the hidden layer parameters of base models randomly, and calculates the output layer parameters using the least square method with negative correlation learning. Simulation results show that the proposed method has relatively higher accuracy and reliability compared with the single RVFL model.

Pattern Recognition Technology Application in Intelligent Processing of Heavy Oil

Reliable and efficient product yield estimation for unknown oils after the fluid catalytic cracking (FCC) reaction is one of the key components in heavy oil intelligent processing. This paper describes the use of two chemometric pattern recognition methods, k-nearest neighbor (k-NN) classification and supervised self-organizing maps (SSOMs), for building classification models to determine the most similar oil sample to an unknown sample in a given data set and to use the FCC yields record of the correspondent oil as the product yield prediction for the unknown sample under the same reaction conditions. Two-sided t test, correlation analysis, and hierarchical cluster heat map analysis were performed to assess the quality of the models. The work provides laboratory evidence that k-NN or SSOMs techniques could all be employed for FCC product yield estimation, while the k-NN model would be more suitable for industrial application in terms of stability and efficiency.

Scratch-concerned yield modeling for IC manufacturing involved with a chemical mechanical polishing process

In existing integrated circuit (IC) fabrication methods, the yield is typically limited by defects generated in the manufacturing process. In fact, the yield often shows a good correlation with the type and density of the defect. As a result, an accurate defect limited yield model is essential for accurate correlation analysis and yield prediction. Since real defects exhibit a great variety of shapes, to ensure the accuracy of yield prediction, it is necessary to select the most appropriate defect model and to extract the critical area based on the defect model. Considering the realistic outline of scratches introduced by the chemical mechanical polishing (CMP) process, we propose a novel scratch-concerned yield model. A linear model is introduced to model scratches. Based on the linear model, the related critical area extraction algorithm and defect density distribution are discussed. Owing to higher correspondence with the realistic outline of scratches, the linear defect model enables a more accurate yield prediction caused by scratches and results in a more accurate total product yield prediction as compared to the traditional circular model.

A theoretical study of the OH-initiated gas-phase oxidation mechanism of β-pinene (C10H16): first generation products

An extensive mechanism for the OH-initiated oxidation of β-pinene up to the first-generation products was derived based on quantum chemical calculations, theoretical kinetics, and structure-activity relationships. The resulting mechanism deviates from earlier explicit mechanisms in several key areas, leading to a different product yield prediction. Under oxidative conditions, the inclusion of ring closure reactions of unsaturated alkoxy radicals brings the predicted nopinone and acetone yields to an agreement with the experimental data. Routes to the formation of other observed products, either speciated or observed as peaks in mass spectrometric studies, are also discussed. In pristine conditions, we predict significant acetone formation following ring closure reactions in alkylperoxy radicals; in addition, we predict some direct OH recycling in subsequent H-migration reactions in alkylperoxy radicals. The uncertainties on the key reactions are discussed. Overall, the OH-initiated oxidation of β-pinene is characterized by the formation of a few main products, and a very large number of products in minor to very small yields.