However, after two weeks of contact, I found that what I mastered was just an ordinary model. If it really involves aerodynamics, it will be difficult.
At present, summarize the learning results of these two weeks:
simplified model
The simplification of the model is very important, which is related to the grid division and solution time, and then affects the accuracy of calculation. I think it should be simplified according to my own practical experience. For example, if a person who doesn't understand airplane design gives him a model airplane, he may simplify the wing into a flat plate. Of course, this is a joke. As far as the author is concerned, all the drones currently exposed have similar fuselage contours. So I think it is necessary to simplify the fuselage, but the joint between the fuselage and the wing is still more important. Keep it safe. However, simplification cannot be too much. It is necessary to appropriately increase the windward area of some external stores. Due to the interference and parasitic resistance in the actual flight of UAV, the author can't guarantee that he can calculate it so accurately.
Selection of solver
At present, I don't know much about this. Let's record a few points first.
Reynolds number is calculated according to the velocity and size characteristics of the object, and whether the airflow flowing through it is turbulent or laminar is judged. The viscosity of air must be considered. I read in a book that the difference between the calculated results and the actual results is about 8% because of ignoring the existence of air viscosity. There is also the problem of attack angle of incoming flow. In FLUENT, the problem of attack angle of incoming flow is transformed into velocity vector, that is, passive is transformed into active, which is not difficult to understand. However, on previous occasions, the resistance was negative. After Baidu, they said it was because the vector of force was set incorrectly. You can draw a vector triangle yourself and think about it.
Grid division
Before, the author has been using the divider that comes with workbench to divide it. As you can imagine, for simple objects, this is enough. However, for UAV, due to the grid problem, the author has divergent residual value, which leads to the positive and negative changes of lift-drag coefficient with the change of iteration steps, and the value has increased by more than 10 thousand times. You can imagine. At present, I am studying icem (the more I learn, the more I feel ignorant), and I hope I can draw a grid at will soon.
Solution time
In fact, the longer the solution, the more accurate it is, because everyone knows convergence and divergence. Some calculations can be used when they reach a certain error value, and only need to observe their convergence. Some results are not accurate with the increase of solution time, which is caused by computer accuracy problems.
5. Flow field volume
The author also has a deep understanding of this. First of all, the volume of the flow field is roughly 10 times the size of the object to be calculated. But after reading the book, I found that the size of the flow field region also affects the accuracy of the solution. You can think about how big the airflow around a 787 changes with it when it flies in the air, so in the case of computer NB, the bigger the better, and the grid can change from dense to sparse with the size.
After studying the icem grid division for a week this week, I can finally draw a better boundary layer grid for the airfoil. However, there is still a small problem, that is, the leading edge grid mapping of airfoil is not ideal, and importing grid files into fluent always fails. (It was later found that the leading edge of the airfoil was independent of the line. )
Icem's grid division method is really abstract, and mapping and topology have never been touched. But now I almost understand that for three-dimensional entities, block represents the grid distribution of a certain area.
File save:
I don't have a good habit of saving money when I divide it. If the computer crashes, you will also have a certain chance of crashing when you divide the grid format, so you must form this habit, which will also make the following analysis faster.
Grid quality:
Good grid quality means that the results of this analysis are highly reliable (after all, it is analog analysis, and there will be errors), so it is necessary to carry out standard test on the grid divided by yourself.
Grid number control
After two weeks of intermittent speculation, the author can finally draw the boundary layer grid of a simple wing. However, another problem is found, that is, the distribution of grid nodes. Generally speaking, the boundary layer grid is controlled by the E #%% function. Firstly, the number of nodes on the whole target line is determined, then the growth rate of boundary layer is determined, and then the minimum node size is determined. The irrelevant part is generally controlled by the b #%% function, that is, the average distribution.
When drawing the grid of an analysis model, it is necessary to determine how much the grid needs to be controlled, which is determined by the computer configuration. Because the author is not a local tyrant, the grid nodes are generally controlled below 500 thousand. Then determine whether boundary layer grids are needed and how many layers are needed. Because this part of the grid is not easy to calculate. Then the distribution of grid nodes is determined by addition, subtraction, multiplication and division.
Tetrahedral grid
The author soon came into contact with the drawing method of tetrahedral grid, which originated from the paper on aerodynamic analysis of UAV that I saw before. Among them, students used tetrahedral mesh and 5 layers of boundary layer thickness. Today, after calculation, it is found that the grid quality is difficult to control and the generation speed is slow. But it can be directly imported into fluent without conversion.
Distortion of fluent calculation model results at angle of attack
During this time, we have been trying to solve the problem of attack angle of incoming flow with fluent, but the result is not much different from that without attack angle, and there is no credibility at all. The current analysis is related to these factors:
Grid quality
Boundary layer number
Solver setup error (that's it)
Selection of aircraft model solution model
At present, all the problems about grid have finally come to an end, and I have been fluent in learning for more than two months. Among them, I reviewed {Introduction to Air and Gas Dynamics} and read many books about fluent and icem. Now it is possible to draw a plane grid with certain difficulty, but there is another problem in solving the model selection.
Because the courses are all aerodynamic calculation of subsonic aircraft, the pressure stability solver, the turbulence model of spalart-allmaras equation and the far-field pressure inlet are selected. The turbulence model of k-ε equation, velocity inlet and pressure outlet are adopted.
Summary of the first stage
The author has been in contact with icem fluent for almost 4 months, so far he can finally solve the aerodynamic analysis of low Reynolds number UAV and some simple post-processing. But I always think it shouldn't take so long. The aerodynamic analysis of UAV is summarized as follows:
Note: It is best to keep a backup after each step is completed.
simplified model
List reliable algorithms (with wind tunnel numerical comparison) and have similarities with them.
Verification algorithms, including grid independence verification, Y+ calculation experiment, moment coefficient of lift and drag, and comparison of pressure distribution in a certain section, allow errors within a certain range.
Take the above experience, do grid comparison and Y+ iteration, divide the grid (if the geometry is simple, divide the structural grid, if it is complex, it is unstructured grid), calculate the size of the first layer grid, and try to control the number of grids within a certain critical value according to the hardware configuration.
At the initial stage of meshing, the computational domain is generally at least 10 times the characteristic length, and the incompressible inlet can be slightly close to the model.
Check that the grid quality 2*2*2 standard is greater than 0.2 as much as possible, and the angle is greater than 18 degrees as much as possible. It is necessary to smooth the mesh.
Generally, spalart-allmaras turbulence equation, ideal viscous gas and relaxation factor are adopted and tried to be adjusted.
At present, the continuous equation does not converge, but if the lift-drag coefficient converges, the result can be considered credible.
Post-processing, compared with the actual experience. Import ansys post-processing software and extract the required parameters. (End of the first phase)
Icem use skills summary
During this time, the author has a lot of tasks, so the models I came into contact with are only some simple outline models, but I found that if you want to draw the grid efficiently, you still have to use your head sometimes (it feels like nonsense). Generally speaking, don't stick to a routine, such as a strange water pipe. Non-strange places can be generated by grid stretching, and it is much easier to divide strange places into blocks.
A flying wing model is being tried at present. The author first draws it with a structural grid. After many adjustments, when the number of grids is kept at about 200w, the grid quality is greater than 0.35, which can be said to be quite good. However, the author found it difficult to converge in the process of solving. After trying to solve the problem overnight, the number of convergence steps is about 4000. Analysis may be due to the following reasons:
The number of grids exceeds the computer load, which leads to the slow solution process.
This model includes the fuselage, and the process is complicated.
The volume gap between adjacent grids is too large.
The distribution of grid nodes on the wall is unreasonable.
The above reasons cannot be verified for the time being, so the author tries to draw unstructured field grids and control the number of grids (unstructured solution is slow under the same number of grids).