“…However, for the output parameter, the flow characteristics such as velocity distribution is used in the training method. Detailed descriotion of ANFIS method can be found in our previous publications 23 , 24 , 27 – 29 , 46 , 48 – 50 . Figure 2 ANFIS structure with three inputs, number of inputs MFs = 4, type of MFs = dsigmf .…”
Section: Anfis Methodsmentioning
confidence: 99%
“…In about one decade, some researchers 23 , 24 have combined the CFD method with machine learning algorithms to predict the fluid flow, particularly multiphase flow and heat transfer problems 25 , 26 . They examined different tuning parameters to achieve the best accuracy of the model in predicting the flow 27 – 29 .…”
In this study, a square cavity is modeled using Computational Fluid Dynamics (CFD) as well as artificial intelligence (AI) approach. In the square cavity, copper (Cu) nanoparticle is the nanofluid and the flow velocity characteristics in the x-direction and y-direction, and the fluid temperature inside the cavity at different times are considered as CFD outputs. CFD outputs have been assessed using one of the artificial intelligence algorithms, such as a combination of neural network and fuzzy logic (ANFIS). As in the ANFIS method, we have a non-dimension procedure in the learning step, and there is no issue in combining other characteristics of the flow and thermal distribution beside the x and y coordinates, we combine two coordinate parameters and one flow parameter. This ability of method can be considered as a meshless learning step that there is no instability of the numerical method or limitation of boundary conditions. The data were classified using the grid partition method and the MF (membership function) type was dsigmf (difference between two sigmoidal membership functions). By achieving the appropriate intelligence in the ANFIS method, output prediction was performed at the points of cavity which were not included in the learning process and were compared to the existing data (the results of the CFD method) and were validated by them. This new combination of CFD and the ANFIS method enables us to learn flow and temperature distribution throughout the domain thoroughly, and eventually predict the flow characteristics in short computational time. The results from AI in the ANFIS method were compared to the ant colony and fuzzy logic methods. The data from CFD results were inserted into the ant colony system for the training process, and we predicted the data in the fuzzy logic system. Then, we compare the data with the ANFIS method. The results indicate that the ANFIS method has a high potentiality compared to the ant colony method because the amount of R in the ANIFS system is higher than R in the ant colony method. In the ANFIS method, R is equal to 0.99, and in the ant colony method, R is equal to 0.91. This shows that the ant colony needs more time for both the prediction and training of the system. Also, comparing the pattern recognition in the two systems, we can obviously see that by using the ANFIS method, the predictions completely match the target points. But the other method cannot match the flow pattern and velocity distribution with the CFD method.
“…However, for the output parameter, the flow characteristics such as velocity distribution is used in the training method. Detailed descriotion of ANFIS method can be found in our previous publications 23 , 24 , 27 – 29 , 46 , 48 – 50 . Figure 2 ANFIS structure with three inputs, number of inputs MFs = 4, type of MFs = dsigmf .…”
Section: Anfis Methodsmentioning
confidence: 99%
“…In about one decade, some researchers 23 , 24 have combined the CFD method with machine learning algorithms to predict the fluid flow, particularly multiphase flow and heat transfer problems 25 , 26 . They examined different tuning parameters to achieve the best accuracy of the model in predicting the flow 27 – 29 .…”
In this study, a square cavity is modeled using Computational Fluid Dynamics (CFD) as well as artificial intelligence (AI) approach. In the square cavity, copper (Cu) nanoparticle is the nanofluid and the flow velocity characteristics in the x-direction and y-direction, and the fluid temperature inside the cavity at different times are considered as CFD outputs. CFD outputs have been assessed using one of the artificial intelligence algorithms, such as a combination of neural network and fuzzy logic (ANFIS). As in the ANFIS method, we have a non-dimension procedure in the learning step, and there is no issue in combining other characteristics of the flow and thermal distribution beside the x and y coordinates, we combine two coordinate parameters and one flow parameter. This ability of method can be considered as a meshless learning step that there is no instability of the numerical method or limitation of boundary conditions. The data were classified using the grid partition method and the MF (membership function) type was dsigmf (difference between two sigmoidal membership functions). By achieving the appropriate intelligence in the ANFIS method, output prediction was performed at the points of cavity which were not included in the learning process and were compared to the existing data (the results of the CFD method) and were validated by them. This new combination of CFD and the ANFIS method enables us to learn flow and temperature distribution throughout the domain thoroughly, and eventually predict the flow characteristics in short computational time. The results from AI in the ANFIS method were compared to the ant colony and fuzzy logic methods. The data from CFD results were inserted into the ant colony system for the training process, and we predicted the data in the fuzzy logic system. Then, we compare the data with the ANFIS method. The results indicate that the ANFIS method has a high potentiality compared to the ant colony method because the amount of R in the ANIFS system is higher than R in the ant colony method. In the ANFIS method, R is equal to 0.99, and in the ant colony method, R is equal to 0.91. This shows that the ant colony needs more time for both the prediction and training of the system. Also, comparing the pattern recognition in the two systems, we can obviously see that by using the ANFIS method, the predictions completely match the target points. But the other method cannot match the flow pattern and velocity distribution with the CFD method.
“…One way that we can use during training is to randomly shuffle big data set and then let the training method learn the data set. In this case, we can stop learning of machine learning from pattern of data, and it is only based on location of each node 22,23 . In this study, we compute the interaction between dispersed and continuous phase and velocity of each component in a single size Eulerian framework beside RANS turbulence model.…”
Computing gas and liquid interactions based on interfacial force models require a proper turbulence model that accurately resolve the turbulent scales such as turbulence kinetic energy and turbulence dissipation rate with cheap computational resources. The k − ε turbulence model can be a good turbulence predictive tool to simulate velocity components in different phases and approximately picture the turbulence eddy structure. However, even this average turbulence method can be expensive for very large domains of calculation, particularly when the number of phases and spices increases in the multi-size structure Eulerian approach. In this study, with the ability of artificial learning, we accelerate the simulation of gas and liquid interaction in the bubble column reactor. The artificial learning method is based on adaptive neuro-fuzzy inference system (ANFIS) method, which is a combination of neural cells and fuzzy structure for making decision or prediction. The learning method is specifically used in a cartesian coordinate such as Eulerian approach, while for the prediction process, the polar coordinate is applied on a fully meshless domain of calculations. During learning process all information at computing nodes is randomly chosen to remove natural pattern learning behavior of neural network cells. In addition, different r and θ are used to test the ability of the learning stage during prediction. The results indicate that there is great agreement between ANFIS and turbulence modeling of bubbly flow within the Eulerian framework. ANFIS method shows that neural cells can grow in the domain to provide high-resolution results and they are not limited to the movement or deformation of source points such as Eulerian method. In addition, this study shows that mapping between two different geometrical structures is possible with the ANFIS method due to the meshless behavior of this algorithm. The meshless behavior causes the stability of the machine learning method, which is independent of CFD boundary conditions. Bubble column reactors are multiphase domains to provide physical and chemical interaction between gas, liquid, and solid phases 1-5. This physical interaction between two distinct phases can be observed in bubbly flow in the bubble column reactor. The sparging of gas into the continuous phase results in forming gas bubbles in continuous phase (e.g. liquid phase). This formation of dispersed into the continuous phase can be defined as a bubbly flow. However, in very low flow rates, bubbles are uniform and spherical shape, and by increasing the gas flow, bubbles start merging or breaking into a number of small bubbles. To design and optimize reactors and bubbly flow in the domain, understanding bubble column reactor hydrodynamics such as gas holdup , gas and liquid flow pattern, bubble size and shape specification, and turbulence parameters are important factors in different
“…Machine learning methods are recently widely used to link to computational fluid dynamics to better predict the turbulent kinetic spectrums, gas hold-up, the liquid flow pattern, and other flow features in the BCR 20 . By this combination, a great framework is created to map the results or provide multiple input/ output parameters 21,22 .…”
In the current study, Artificial Intelligence (AI) approach was used for the learning of a physical system. We applied four inputs and one output in the learning process of AI. In the learning process, the inputs are space locations of a BCR (bubble column reactor), which are x, y, and z coordinate as well as the amount of gas fraction in BCR. The liquid velocity is also considered as output. A variety of functions were used in learning, such as gbellmf and gaussmf functions, to examine which functions can give the best learning. At the end of the study, all of the results were compared to CFD (computational fluid dynamics). A three-dimensional (3D) BCR was used in this research, and we studied simulation by CFD as well as AI. The data from CFD in a 3D BCR was studied in the AI domain. In AI, we tuned for various parameters to achieve the best intelligence in the system. For instance, different inputs, different membership functions, different numbers of membership functions were used in the learning process. Moreover, the meshless prediction was used, meaning that some data in the BCR have not participated in the learning, and they were predicted in the prediction process, which gives us a special capability to compare the results with the CFD outcomes. The findings showed us that AI can predict the CFD results, and a great agreement was achieved between CFD computing nodes and AI elements. This novel methodology can suggest a meshless and multifunctional AI model to simulate the turbulence flow in the BCR. For further evaluation, the ANFIS method is compared with ACOFIS and PSOFIS methods with regards to model’s accuracy. The results show that ANFIS method contains higher accuracy and prediction capability compared with ACOFIS and PSOFIS methods.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
