The Military-industrial Scientific Research System of the Academic Master Chapter 491

Chapter 491: Won't you change it yourself?

For large tasks such as whole-machine aerodynamic simulation or structural simulation, the real calculation process must be completed on a supercomputer.

  But the saved log files can still be seen on the PC used as the input and output terminal.

 These days, many terminals and servers still use batch FTP to connect to each other, which only captures data once a day, rather than real-time linkage APIs.

Of course, because the computing speed of the server is relatively limited, it does not appear to be very inefficient.

“We have encountered basically the same troubles mainly in two directions of research.”

Yao Mengna put the two computers on the table, straightened her messy hair a little and began to explain:

“One is what I just said, the computational structural mechanics problem of structural analysis of the body part.”

As she spoke, she tapped a few times on the keyboard of one of the computers, entered an account, and called up a series of calculation results.

 Then he turned to the second computer and performed a similar operation:

“Another thing is that we are also studying the problem of airflow interference when the aircraft drops large-sized suspended objects.”

"In previous wind tunnel tests, it was found that some ammunition, especially some cruise ammunition with relatively strong lift, will not immediately free-fall after the whole body formed by them and the wing is destroyed, but will move upward and damage the wing. , we hope to solve this hidden danger before the seventh batch of aircraft is put into production.”

 The early Su-27 basically did not consider ground attack, did not need to carry any large ammunition, and did not have external auxiliary fuel tanks, so there was no need to consider such issues.

 Hence, when developing multi-purpose aircraft on the same platform, some strange problems will arise that are easily overlooked.

 “Let me take a look…”

Chang Haonan came to the first computer and bent down to carefully check the program's running log.

He Mingze, who was following behind, moved a chair very appropriately and placed it behind Chang Haonan.

“Mr. Chang, please sit down first.”

 A few seconds later, when he realized something, he brought a second picture to Yao Mengna and placed it next to Chang Haonan.

“You used hyperboloid structural mesh and volume structural mesh... Speaking of which, your modeling level is very high. The geometric modeling process only took a few days to complete?”

 Geometric modeling processing is the prerequisite step in the mesh generation process.

In 1997, the main problem faced by most computational fluid dynamics and computational structural mechanics simulations around the world was actually the lack of a mature application tool from CAD models to surface structure mesh generation.

 For existing mesh generation tools, for some models with slightly higher accuracy, it often takes about 6 months or even longer to complete the entire generation work.

For example, the simplified model of Boeing 747LCF took 4 months.

 The V22 Osprey took more than 9 months.

  And from a later perspective, it was clear that there were design flaws.

  In comparison, geometric modeling that takes several weeks to a month is not a big problem and is completely acceptable.

  But for TORCMultiphysics, the mesh generation process is generally shortened to 1-2 months, so the speed of geometric modeling processing seems a bit slow.

 It's just that Chang Haonan never had time to focus on modeling, and always used Dassault's software packages directly.

“The engineers from Sukhoi brought a new modeling software called KOMPAS. I slightly adjusted the interface compatibility of TORCHMultiphysics so that it can directly connect to the Kom format project files.”

“Originally, it was just to save trouble, but I found that this modeling method is very suitable for our software, so I have been using it ever since.”

 Although this is not the main problem to be solved today, Chang Haonan still paid a little attention.

Mao's electronics industry has stagnated for nearly 10 years and has gradually fallen behind. However, when it comes to software and algorithms, it has not fallen far behind.

Although the market is not big enough, there are few hits (except for telegram), but there are definitely some unique skills.

 After a short interlude, Chang Haonan once again focused on the grid generation log behind.

“The number of sub-area grid blocks is 253…”

“The number of grid cells is 133 million…it's really a lot, the complexity is very high…”

 “Grid assembly…”

 When he saw this last step, he quickly discovered something was wrong:

  “Huh?”

 According to the normal process, TORCMultiphysics should automatically "assemble" these 253 grid blocks together through inter-grid interpolation in the overlapping areas.

 But the log shows that the software got stuck at this step.

  To be precise, it got stuck during the data capture process before this assembly step.

 “Too much data?”

 This is the first possibility that Chang Haonan thought of. However, the logic of the grid generation algorithm he designed specifically reduces the computational complexity for large models.

  In other words, no matter how complex the model is, the calculation time will be extended indefinitely, and it will not lead to problems such as memory usage.

 In short, it is impossible to get stuck here.

 “Have you tried manual assembly?”

Chang Haonan asked after thinking for a moment.

“I tried it, but the other parts are easy to say. When it comes to the central lifting body, even if there are only minor changes in the sub-region grid, the assembly input file must be modified accordingly. If it is all done manually, it will probably take..."

Yao Mengna raised her head slightly and calculated:

“It takes about 60,000 lines of code to be entered manually…”

 60,000 lines of code may not sound like much, but considering that TORCH Multiphysics cannot be automated, parameters must be adjusted manually, which definitely involves a lot of calculation and conversion issues.

What's even more troublesome is that, as a new software, there are relatively few people who can complete this process proficiently.

 In short, it is obviously not something that can be completed in a short time.

When Chang Haonan tested the software before, it was impossible to use this model with hundreds of sub-areas and hundreds of millions of grid cells to test due to tight time and heavy tasks.

 In fact, since TORC Multiphysics was released for such a long time, judging from the feedback from various users, no one has used it on a model of this size.

 That's why I never found the problem before.

 “If manual assembly cannot continue…”

Chang Haonan frowned slightly and tapped his fingers lightly on the computer case next to the keyboard.

“Have you ever considered giving up the structural restrictions of the grid nodes and considering dynamically adjusting the model rules according to the changes in the object boundary when the surface deforms?”

There was a moment of silence.

 Yao Mengna and He Ming looked at each other.

　Judging from the expressions of the two of them, they didn't even fully understand the meaning of Chang Haonan's words.

  “For example.”

Looking at the two people with clear eyes, Chang Haonan had no choice but to explain further.

 But after searching around for a long time, I couldn't find any paper or pen.

Yao Mengna, who knew Chang Haonan's habits, quickly took it out of her bag and put it in Chang Haonan's hand.

“For example, we can compare the calculation area to a linear elastic body and determine the displacement of the grid nodes by solving the elastic mechanics equations..."

“The algebraic model means that the displacement of the grid is obtained by multiplying the moving boundary displacement by a coefficient. The coefficient takes 1 on the moving boundary and 0 on the far field boundary. The inner field is interpolated according to a certain function..."

 After some calculations and explanations, Chang Haonan drew a circle on the already filled paper:

"Of course, this is not possible everywhere. For too complex parts, each side of the grid unit can be regarded as a spring. The spring coefficient is related to the side length. When the boundary moves, the force balance of the nodes of the spring system can be solved by solving the problem. Problem determining new point location…”

“In this way, the size, shape and position of the grid points can be adjusted and controlled during the calculation process, and the requirements for the grid boundaries will be much reduced.”

 There was another moment of silence.

“There is this function in TORCH Multiphysics?”

　After quite a while, Yao Mengna asked in a tone that was somewhat doubtful about life.

 She was obviously personally involved in the development of this software, but after thinking about it for a long time, she really didn't remember that there was such a way to play it.

"of course not."

Chang Haonan clicked the tip of the ballpoint pen back:

 “But you can change it yourself!”

 (End of this chapter)

Chapter end