Research On The Combined Method Of Physical Model And Machine Learning Of Flow Around Circular Cylinder And Vortex-induced Vibration

Flow around a circular cylinder is one of the basic problems in fluid mechanics.The flow boundary is simple,but it is rich in nonlinear dynamic characteristics and complex flow phenomena.Therefore,many scholars have paid much attention to the problem of circular cylinder flow,carried out many numerical simulation and experimental studies,and accumulated many reliable numerical simulation and experimental data.On the other hand,vortex-induced vibration is a common problem in many engineering fields,among which the most famous is Karman Vortex Street phenomenon.Due to the alternating shedding of vortices,the cylinder is subject to the flow resistance and vertical lift,which are also periodic changes.The structural fatigue caused by this has become a common problem in engineering,such as the riser pipe,submarine pipeline and suspension cable of long-span bridge in ocean engineering,which are all faced with the fatigue breakage caused by vortex-induced vibration.In recent years,the fusion of machine learning and physical model has brought a new research paradigm for fluid mechanics and its related engineering fields.Through the machine learning modeling and prediction method,the accumulated data value of Computational fluid dynamics(CFD)and Experimental fluid mechanics(EFD)can be reused.By feeding data to the machine model,the model can predict the flow field and related problems under other working conditions.Based on the research paradigm of the combination of physical model and machine learning,this paper carries out the research on the prediction of Flow around a circular cylinder and vortex-induced vibration.The main contents include:Prediction of flow resistance and lift around a bluff body.Based on the Convolutional neural network(CNN)of the depth learning model,the data set is the CFD numerical results of the drag coefficient and Jift coefficient of 27 bluff bodies with different geometric shapes.The data are divided into 13 training sets and 14 prediction sets.The prediction results are compared with the CFD standard results.The results show that the deep learning method,CNN has obvious advantages in image processing and is especially suitable for dealing with the large dimension of boundary information around the blunt body.Using CNN to predict the drag coefficient and lift coefficient of the bluff body,it has the advantages of high precision and fast speed.Compared with CNN,the accuracy of BP method is higher than CNN.Machine learning model for prediction of cylindrical flow field.Based on the CNN model,two data-driven deep learning neural network models are proposed,namely the PC-C model for predicting the flow resistance coefficient and lift coefficient of the cylinder with different Reynolds number,different radius position and size,and the PC-F model for predicting the velocity field.Data set from open source OpenLB calculated the Reynolds number Re=10,20,30,...,150 CFD numerical results of cylindrical flow around 150 different position radius sizes.Through CNN modeling and prediction,the feasibility and accuracy of the two models in predicting the flow around a finite boundary cylinder are verified.The model successfully predicted the flow field around the cylinder,and compared with the calculation time of OpenLB's physical model,the efficiency of the deep learning method was improved by at least one order of magnitude within a small error range.3.The fusion of machine learning and physical model of vortex-induced vibration around circular cylinder.Cylinder motion equation using the spring oscillator model,flow field of the incompressible Navier-Stokes(NS)equation of the two physical model embedded neural network,in the form of error defined as the loss function of neural network,and then by gradient descent and back propagation algorithm of neural network,and the internal parameters of the neural network,achieve the goal of predict vortex-induced vibration problem.The author thinks this part of the content is the interesting part of the paper.Small data can verify knowledge,while big data can discover knowledge.The research paradigm of combining machine learning with physical model will bring a new world for the research of engineering problems based on physical model.