جستجوی مقالات مرتبط با کلیدواژه « شبکه عصبی مصنوعی » در نشریات گروه « مهندسی شیمی، نفت و پلیمر »

Lozenge rug chaleshtor is in terms of artistic and aesthetic aspects, it is considered one of the consumer and capital goods that have international fame. From the past until now, this native art has been traditionally produced with the same design, pattern, and colour without paying attention to the tastes of its audience, which can be one of the reasons for the failure of this original art in the global rug markets shortly. Therefore, knowledge and awareness of consumers' taste, which is considered as one of the steps before production, and using modern science, can help to produce according to the taste of the audience and, accordingly, to be more successful in terms of sales and providing high export statistics. To achieve this, in this article, sub-branches of artificial intelligence were used to achieve the taste of Lozenge rug chaleshtor audiences. Three algorithms more related to the subject, 1. artificial neural networks, 2. decision tree learning and 3. support vector machine are compared and finally, the most suitable algorithm for this subject is the artificial neural networkss for receiving the taste of the audience of the Lozenge rug chaleshtor and providing suitable patterns in the field of design. , role, colour, dimensions, texture and price according to the audience's taste, and it was tried to answer the question of whether that artificial neural networkss can introduce the principles of the audience's taste of lozenge rug chaleshtor. in structural fields. For this purpose, primary data in the field of design and role, colour, raw materials, texture, dyeing, dimensions and price were collected through the contact questionnaire and then, using an artificial neural networkss, an algorithm was designed, the results showed that traditional lozenge designs with half motifs curved and quiet, with bright, limited, soft colours, natural dyeing, dimensions under 6 meters and preferably square with a price per meter of up to 10 million tomans is the final taste of the audience of this type of rugs.Keywords: Taste,Lozenge Rug, Chaleshtor, Artificial Neural Networks

Understanding the car tire market is very important for both producers and consumers due to its impact on the final price of the car, incurring fixed costs for the car owner and the high amount of financial transactions. The purpose of the current research is to explain and model the behavior of tire buyers based on four factors using a combined approach. In the first step, by reviewing the literature and the background of the research, the factors affecting the purchase of Iranian or foreign tires were identified according to the four factors of personal, market, store and product. In the second step, in order to analyze the behavior of Iranian or foreign rubber buyers, using non-hierarchical cluster analysis (mean k) and based on the 24 factors of the analyzed population and the number of optimal clusters according to the Davis-Bouldin index, Then, using artificial neural networks, the independent variables of the research (four factors of personal, market, store and product) were used to predict the dependent variable (purchase of Iranian or foreign tires). The statistical population of the research was Iranian or foreign rubber buyers in the west of Mazandaran province, and due to the unlimited population of the research, the unlimited population formula was used to determine the sample size, the minimum sample size was 384 and the analysis was done on 527 usable questionnaires. The results of the research showed that in the first cluster, in the order of price, safety standards, image and reputation of the store, and in the second cluster, in the order of proximity, friendly behavior of sellers, and the breadth of choice, they have obtained the most centers of the clusters. The use of neural networks showed that the independent variables of the research have a good predictive ability and by using them, 0.886 changes of the dependent variable can be predicted.

In this study, we developed artificial neural network-based model for prediction of equilibrium solubility of carbon dioxide in the amine solvent system of (triethanolamine + piperazine + water) for the purpose of carbon dioxide uptake. In the MLP model, the solubility data (CO2 loading in the amine solution) were investigated as functions of CO2 partial pressure, system temperature, and amine composition. The Levenberg–Marquardt back-propagation (LMP) algorithm was used to predict the partial pressure of carbon dioxide. The final ratio of training, validation, and test datasets was approximately 70:15:15. The optimum multilayer perceptron (MLP) structure in Levenberg-Marquardt algorithm for CO2 partial pressure is created with 20 neurons in the first hidden layer and 10 neurons in the second hidden layer. There was a 0.99546 correlation coefficient between the experimental results and the artificial neural network (ANN) calculations, demonstrating excellent compatibility between them. The best validation performance was 0.0043497 from epoch 13. In general, the results show that the applied model can provide an accurate prediction of partial pressure or solubility for different operating conditions.

The common methods that are usually used to identify the devoid rough rice from the healthy ones are often time-consuming and expensive. For this reason, in this research, a smart and fast method based on machine vision system coupled with artificial neural networks is presented in order to predict the percentage of devoid/healthy rough rice grains. Digital images of five varieties of paddy were prepared in three states: healthy, devoid, and mixed, in two states scattered and piled. After pre-processing and segmentation, 3 color features and 5 morphological features were extracted for each rice grain. Principal component analysis (PCA) method was then used in order to identify the most effective features in distinguishing devoid rough from healthy rice. In the next step, multilayer perceptron (MLP) algorithm based on the main components obtained by PCA method was used to create models for identifying and classifying the samples. Root Mean Square Error (RMSE), correlation coefficient (R2), specificity and sensitivity were used to evaluate the modeling capability and validation of each algorithm. The obtained results showed that the designed intelligent method can identify devoid rough rice seeds with acceptable accuracy in all cultivars (R2P>0.81, RMSEp<0.219, Sensitivity>0.8 & Specificity>0.98). Therefore, the machine vision system in combination with artificial neural networks can be used as an intelligent and fast method at the entrance of rice bleaching factories to evaluate the quality of harvested rough rice and predict the percentage of unhealthy rough rice.

Assessment of petroleum generation potential of the source rock as a function of total organic carbon content and kerogen type is of great importance in oil and gas exploration studies. The main aim of this research is to compare the performance of artificial neural networks trained by back propagation algorithm (ANN-BP) and metaheuristic methods including genetic algorithm (ANN-GA) and particle swarm optimization (ANN-PSO) for prediction of total organic carbon (TOC) content and remaining petroleum potential (S2) from the wireline data. For this purpose, Pabdeh Formation (Paleocene-Oligocene) in Mansuri oilfield is studied. Based on the results of linear regression on the test data, ANN-PSO method provides more accurate predictions of Rock-Eval derived TOC and S2 parameters with correlation coefficient (R2) values of 0.8548 and 0.9089, respectively. In addition, hydrogen index (HI) is appropriately predicted based on the relationship between TOC and S2 values obtained from the ANN-PSO method with R2 value of 0.6882, from which different types of kerogen can be distinguished with classification accuracy of 74 percent. Geochemical zonation of Pabdeh Formation based on organic richness and kerogen type reveals three distinctive parts, among which the middle part (Brown Shale Unit, BSU) demonstrates the greater petroleum generation potential with having the significant values of total organic carbon and hydrogen index. Therefore, the BSU can play an important role in hydrocarbon charging of the oilfield traps if it attains proper level of thermal maturity. Accordingly, precise determination of petroleum generation characteristics of Pabdeh Formation using the ANN-PSO model proposed in this study will lead to a reduction in uncertainty associated with petroleum system modeling, and therefore will considerably increase the exploration efficiency in the Mansuri oilfield.

Time and cost are of great important issues in each project. Therefore, in the drilling industry as well, which is one of the most costly industries, there should be taken appropriate measures to save time and cost. In drilling, the operation can be carried out at a lower cost and time, by choosing the appropriate tool, as well as predicting accurate and timely parameters and probable problems. In this regard, one of the most effective strategies is to investigate field data and subsequently development and improvement of tools for the achieved results. In the drilling industry, in order to identify the problem and improve the drilling performance, generally, either laboratory tests or empirical models are used in accordance with previous experience. In this research, we tried to use novel methods and intelligent modeling to predict the penetration rate in drilling operations. For this purpose, artificial neural network (ANN) approach was used. In this paper, using an artificial neural network and data from Shadegan oil field drilling data (400 data), a model was developed to predict the rate of penetration, and then the results were compared with the results of modeling with multivariate regression. To evaluate the main characteristics of the model, its accuracy was measured using adjacent wells data. The results of the research showed that the correlation coefficient for the neural network is 0.97, and determination coefficient is 0.94 and the error histogram rate is proportional to the zero error 0.005, while the corresponding value in the statistical analysis the correlation coefficient is 0.94 and the determination coefficient is equal to 0.89, which indicates a higher accuracy of neural network modeling.

In this study, the application of bio-acid leaching method based on the use of lemon juice to extract copper and zinc metals from mobile printed circuit boards has been investigated. Three important factors were investigated include lemon juice concentration, Solid / Liquid (S/L) ratio, and hydrogen peroxide (H2O2) concentration. Response surface methodology (RSM) was used to optimize the effective factors. The results showed that for particles with a size of 150 to 180 μm at a constant temperature of 20 ° C and time 4 h under optimal conditions including 1.41% (w/v) S/L ratio, 12.2% (v/v) H2O2 and 74% (v/v) lemon juice, copper and zinc recovery efficiencies are 89% and 73%, respectively. Moreover, the artificial neural network was used to predict the extraction of copper and zinc metals as a function of the studied factors. To validate the model, laboratory results were considered as evaluation data. The results of neural network modeling showed high accuracy to predict the target variable. The values of MRE, MSE, and R2 were 9.485, 15.254, and 0.9356% for the copper extraction model and 1.854%, 1.094, and 0.9963% for the zinc extraction model, respectively.

In recent decades, hybrid optimizations methods based on natural phenomenon have placed special position according to their capabilities in finding optimal solutions without expensive computational loads and disassociation on choosing initial points. Artificial Neural Network is used as one of the powerful tools of Artificial Intelligence for process simulation. The employment of the neural network in the modeling of m-Cresol alkylation process of with isopropanol as well as meta-heuristic methods in obtaining the optimal conditions for the catalyst and the reaction can prepare an effective step towards a high efficiency process.

 In the present study, the artificial neural network is applied to model alkylation of m‐Cresol with isopropanol process. In addition, the bee colony is employed in order to optimize the process yield. To verify its performance, the proposed method is used in prediction of the m‐Cresol conversion and Thymol selectivity of the alkylation process with isopropanol 120 data. In this process, the input variables are Weight Hourly Space Velocity (WHSV), pressure and temperature; m-cresol conversion and thymol selectivity are considered as the output variables of the neural network. Five hidden neurons are considered for the proposed neural network. 120 data is used to train the neural network. The meta-heuristic approach based on bee colony (BC) is applied to maximize the yield of the process.

The results confirm that the proposed method develops the accurate model with an R2 value of greater than 97.5%. The maximum yield is obtained 28.9% by bee colony algorithm with adjustable variables that are WHSV of 0.062 hr-1, the pressure of 1.5 bar and the temperature of 300oC. In addition, in order to achieve the better performance of the optimization algorithm, the appropriate values of acceleration coefficient and population size are chosen 100 and 10 during the trial-and-error phase.

Static mixers are applied for increasing the mixing in chemical reactors as well as for increasing the heat transfer coefficient in heat exchangers. In the study, fluid flow characteristics in tubes equipped with modified static mixers with different geometric parameters were investigated by computational fluid dynamics. The classic twisted tape, perforated twisted tape, and V-Cut twisted tape were evaluated for Reynolds numbers between of 3000 to 19000. The fluid flow and pressure drop for the mixers were investigated. The validated simulation results were employed to train the artificial neural network model. Reynolds number and geometric parameters of the mixers were used as input variables of the neural network for predicting the friction factor. The model accuracy for estimating the friction factor was investigated and a relative error of less than 1% was obtained. The main relative errors for all data in classical, V-Cut, and perforated twisted tape were 0.75%, 0.57%, and 0.52%, respectively, and for validation data were 1.1%, 0.92%, and 0.62%. 30% of the data were randomly selected for the neural network to prove the model validity.

Predicting corporate financial distress is one of the main issues facing managers, which can significantly contribute to companies' success and survival because corporate managers can be informed of financial distress and bankruptcy by providing warning and timely signals. Therefore, it is possible to avoid wasting resources and damages due to bankruptcy through proper management. This study's main purpose was to select the financial variables affecting the prediction of financial distress of companies listed on the Tehran Stock Exchange and then to predict financial distress using the ANN method. Therefore, first, 106 companies were selected by the simple random sampling method. Their financial data were extracted during 2007-2020. The relationship between variables was examined through the Pearson correlation test. Out of 34 financial ratios, 24 ratios with a significant relationship were selected. Finally, the ANN method was used to predict corporate financial distress, in which the financial distressfulness or non-distressfulness of 103 companies was correctly predicted. Comparing the prediction made by the ANN method with the actual values of the dependent variable in 2020, it was found that this method correctly predicted the corporate financial distress with a confidence level of 97%.

Leaks in oil and gas pipelines could cause serious problems such as explosions, environmental pollution, and the loss of energy and financial resources. Early detection of leaks in pipelines is critical to prevent or reduce the occurrence of these losses. For this purpose, a leak detection module located on the infrastructure of a Scada system can be used. In this paper, first, Olga simulates leaks of different sizes and distances on oil pipeline. The output of the Olga, which includes the pressure and flow of different parts of the pipeline, was prepared for analysis using Power Query and Dax Studio tools. The data was entered into MATLAB and the artificial neural network was designed and trained to identify the size and location of the leak. Eventually, this module will be placed on the Scada system as a digital twin of that pipeline and will receive the necessary online data to monitor the condition of the pipeline using the industrial protocols.

Dispersed water-in-oil as a stable emulsion causes numerous problems in extraction, transportation and refining of the crude oil. In the most desalting units, high voltage electrical field is utilized to separate water and ionic components from the crude oil. The efficiency of desalting units depends on operational conditions and hence in this study the result of several parameters on salt content of output crude oil in a desalting unit was considered for both theoretical and experimental studies. For this goal, optimized artificial neural network (ANN) using cuckoo optimization algorithm was applied to simulate the process. The optimum temperature, water  injection rate, retention time, differential pressure of mixing valves and injection rate of demulsifier were predicted by the consequences of simulation as the optimum value for each of the parameters was respectively equal to 79 ppm, 3.25%, 8.5 bar and 90 ppm. Then, because of the significant effect of the demulsifiers, the variation of each parameter was evaluated in the presence of four types of demulsifier separately. The results showed that an increase in the basic sediment and water content (BS&W) and specific gravity of crude oil has adverse effects on desalting process efficiency.

One of the most important processes for avoiding global warming and increasing the heating value of natural gas is adsorption and separation of carbon dioxide and nitrogen using zeolite. In this study, experimental results of adsorption of methane, carbon dioxide and nitrogen by zeolite 13X was assessed using artificial neural network. The temperature and pressure was considered as inputs, and adsorption capacity was considered as the output of system. In all models, Levenberg-Marquardt back-propogation was used for training of the network. To find the optimum transfer function in hidden and output layers and optimum number of neurons, coefficient of determination, sum of squared errors, mean square error were calculated. Optimized number of neurons for methane, carbon dioxide and nitrogen was obtained 10, 10, and 15 respectively. Moreover, the best transfer functions were Logsig and Tansig for methane, Logsing and Pureline for carbon dioxide and nitrogen for hidden and output layers. In the end, average deviation percentage for predicted results with neural network was compared with the results obtained by Langmuir and dependent on temperature sip models. It indicates that neural network has high accuracy in comparison with other two models.

Addition of nanoparticles to a polymeric matrix leads to enhance the performance of membrane gas separation. In this study, the aim is to find the optimum operative point of polymeric membrane modified by adding nanoparticles in gas separation. The assessed factors are type of nanoparticle, percentage of added nanoparticle, and cross membrane pressure.  Nanoparticles of AL2O3, ZnO, and TiO2 were used. Further, the ranges of nanoparticle concentration and operative cross membrane pressure were 2.5 to 15% and 2 to 25 bar respectively. To optimize a process, developing a robust model is necessary. Therefore, first, a powerful model based on artificial neural network was developed, which it was able to predict the values of permeability of oxygen, nitrogen, methane, and carbon dioxide. Neural network models were developed that had R2 greater than 0.9. Next, the optimum operative conditions for assessed gases were found using methodology based on genetic algorithm and considering four strategies. The results of optimization show that the maximum values of permeability for oxygen, nitrogen, methane, and carbon dioxide are 334.7, 779.9, 902.7, and 270.4 respectively.

In this research, the process of drying the quince fruit and the effect of various parameters such as the drying air speed, time, temperature and thickness on moisture ratio were studied. 7 mathematical models were fitted to the data obtained from 27 series of experiments and the best model was selected. Modeling was also performed by artificial neural network. In this modeling, the effect of all input parameters on the drying process was investigated simultaneously. The selective network structure was considered multi-layer perceptron with the back-propagation algorithm. By researching the number of different hidden layer neurons and different transfer functions, 9 neurons and "logsig” transfer function were used for the hidden layer and "purelin” transfer function for the output layer. Modeling by artificial neural network predicted the simultaneous effect of the four input parameters with very high accuracy. The results showed that ANN modeling had better accuracy than the best mathematical model.

In this research, investigation drying of four rice cultivars (Fajr, Taromi , Dom-siah and Neda) were done by microwave dryer and using ohmic pre-treatment. Three voltages of 125, 150, 175 volts were selected for the pre- treatment ohmic in three times 5, 10 and 15 min and the weight changing were measured and recorded. Finally, the samples placed in the microwave by power 600 watt and every two minutes until dry weight changes were measured. The results of the experiment were analyzed using response surface methodology and artificial neural network. Investigation on rice varieties in the ohmic process showed that Fajr variety had the highest water extraction and weight loss compared to other varieties (Taromi , Dom-siah and Neda) and during the pre-treatment process of ohmic and also during drying by the microwave, those had more moisture extracted. And after that, Tarom and Dom-Siah had the highest amount of moisture removal, Also the lowest moisture removal during the Ohmic heating and drying process were in Neda cultivar. In addition to weight loss in drying with a microwave dryer, Increasing voltage and ohmic time had higher moisture content than low voltage and low ohmic time and rice cultivar Fajr had more weight loss than other cultivars. Also, the predicted values of the response surface method (RSM) models and artificial neural network showed that the accuracy of the artificial neural network was 0.30 more than the predicted RSM numbers.

Surface quality is a key point for expressing the specifications of tools that these specifications are improvable by coating methods. In this study, seed layers was deposited on Al - substrate by dip coating process and chemical bath deposition was used for tin oxide coating over deposited seed layers. Taguchi L18 was utilized for designing the experiments and preparation of required samples for studying the effective parameters (surfactant type and concentration in seed layer, seed layer deposition cycles, and seed layer nature and concentration of precursors). Furthermore, surfactant concentration showed the highest effect on surface wettability. In addition, artificial neural network (ANN) was used for modelling and optimization of process. A multilayer perceptron (MLP) ANN with 4 and 5 neurons in the first hidden layer and second hidden layer, respectively, was obtained as the best network. Optimum point prediction of 142° and 132° was calculated by Taguchi design and ANN modelling, respectively. However, prediction error of 3.5% was calculated comparing with experimental results (137°).

Artificial Neural Networks (ANNs) are powerful modeling techniques that work in brief with arrays of neurons in memory and biological learning. In this research, the classification of dynamic and quasi-static loading type (broad and thin edge) was investigated using input data of phenol, antioxidant, vitamin C content and stiffness with artificial neural network. In this experiment for the classification of two Radial basic function and Multilayer perceptron networks were used with two hyperbolic tangent and sigmoid activation functions in one layer. According to the obtained results, the best value for R and Percent Correct for dynamic loading was (Percent Correct = 100 - R = 9999997), loading the thin edge (Percent Correct = 100 - R = 9999993) and loading the wide edge (Percent Correct = 100 - R = 9999992), which was created in the RBF network with a sigmoid function activation and 8 neurons in the one hidden layer. Also, the most accurate data found for the dynamic loading type, the wide edge and the thin edge was observed in the networks created for the RBF network, and this network has been able to 100% accurately classify the data rate for all loads. In sum, the neural network with the input of general data has the desirable capability in the stacking of dynamic loading and quasi-static data.