6 ms·

> expanding the use of digital analysis over costlier physical testing Oh no. > For example, when vibrating a fuselage on an enormous platform to expose weakn

> expanding the use of digital analysis over costlier physical testing Oh no. > For example, when vibrating a fuselage on an enormous platform to expose weaknesses - known as fatigue testing - the vast majority of the time the tool itself breaks instead of the airframe, according to a person with knowledge of past tests. Such work is costly and has reliably confirmed engineers’ expectations, he added. Is this just hubris, or is this real? That's a lot of confidence to place in a software model.

ineedasername 7y ago
Yeah, and it seems that if the tools your using the test for fatique to destruction are themselves too weak for the task, then you upgrade the tools, not abandon physical testing. No computer model will tell you if there are internal microfractures to a component, and any number of other possibilities.
chroem- 7y ago
It's not software: it's mathematics. You don't build a bridge by successively creating bridges with more and more structural reinforcement until the bridges stop falling down. Instead, you calculate the appropriate size of the columns using a model and only build a single bridge. This is no different. Finite element methods are such a known quantity that we're able to design nuclear weapons without ever testing them in meatspace. If it's good enough for nuclear weapons, I am strongly convinced it's good enough for something as simple as stress analysis.
- tempguy9999 7y ago
  I'm neither an aero engineer nor a nuke weapons designer but I am pretty sure that nuclear weapons are a whole lot more complex than a plane. As for not testing nukes, I'm damn sure they test the relevant non-nuclear parts such as the lensing (conventional) explosives. > ...for something as simple as stress analysis Of entire wings? From the article "..such as using machines to bend the wings to extreme angles and shaking the fuselage until it cracks" Your job isn't in this area I guess. I'm sure of it. BTW the model is only as good as its inputs so if you have a defective batch of components (and wasn't there very recently a case where a bunch of deliberately below-spec aluminium parts were delivered to various aerospace companies which cost lots to rectify) then your model had better refelect that defect. But how are you going to find out? With pure maths?
  - chroem- 7y ago
    > Your job isn't in this area I guess. I'm sure of it. I'm a mechanical engineer. My point is that stress analysis is the oldest, simplest, and most well understood application of finite element analysis. I can't even begin to imagine the multiphysics that goes into simulating a nuclear weapon, but if we are willing to bet our entire nuclear deterrent on it, it seems reasonable to skip some additional rounds of physical tests that are mostly a formality at this point. Also, defective materials are caught much earlier in the manufacturing process through quality assurance methods, before they become airplanes. And then you apply a factor of safety on your calculations for added assurance.
    - samcday 7y ago
      > Also, defective materials are caught much earlier in the manufacturing process through quality assurance methods, before they become airplanes. Okay so maybe this is an apple's to oranges comparison, but I thought of the story how NASA was scammed by a supplier passing off shitty aluminium as space grade [1]. Is it reasonable to assume that if it can happen to NASA, it can happen to Boeing? https://weather.com/news/news/2019-05-02-nasa-launch-failures-caused-by-faulty-aluminum-scam https://weather.com/news/news/2019-05-02-nasa-launch-failure...
      - chroem- 7y ago
        They check the airframes with nondestructive ultrasonic testing that doesn't involve oblitering entire aircraft. That would be caught at one of several different QA points. We're mostly talking about design verification, which is different.
        tempguy9999 7y ago
        Parent post gave the example I was remembering. And the company, Sapa, didn't (AFAICT) manufacture the components, they were a third-party testing service! They provided exactly the tests you say should have picked up the problems.
    - 7y ago
- gameswithgo 7y ago
  people still spot check bridges with physical tests because we can’t perfectly simulate material properties or fluid flow
  - chroem- 7y ago
    Absolutely, but NDT is a far cry from testing an aircraft to failure.
- ridgeguy 7y ago
  > something as simple as stress analysis. Stress analysis isn’t simple. In aircraft applications, the analysis has to model not only metals, but composite materials that have anisotropic and nonlinear responses, account for acoustic energy distribution and local concentration effects, incorporate static- low- and high-cycle fatigue mechanisms (which differ in the materials imperfections that initiate them and in how the ultimate failures play out), incorporate time varying and differential spatial temperature distributions ranging from -55°C to ~1500°C (hot section, commercial engines) and on and on. It’s not simple. I really like physical testing to validate analytical models of complex physical systems.
- Bendingo 7y ago
  > we're able to design nuclear weapons without ever testing them in meatspace Perhaps you are forgetting about all the nuclear weapons testing that was carried out in the 40's 50's and 60's ? (The last tests by USA and USSR, took place in the early 90's.) [0] https://en.wikipedia.org/wiki/List_of_nuclear_weapons_tests https://en.wikipedia.org/wiki/List_of_nuclear_weapons_tests
rjsw 7y ago
Both Boeing and Airbus have been putting a lot of money and time into improving engineering software, and are working together to do it. Now probably isn't the best time to put out press releases on this though.

ineedasername 7y ago

Yeah, and it seems that if the tools your using the test for fatique to destruction are themselves too weak for the task, then you upgrade the tools, not abandon physical testing. No computer model will tell you if there are internal microfractures to a component, and any number of other possibilities.

chroem- 7y ago

It's not software: it's mathematics. You don't build a bridge by successively creating bridges with more and more structural reinforcement until the bridges stop falling down. Instead, you calculate the appropriate size of the columns using a model and only build a single bridge. This is no different. Finite element methods are such a known quantity that we're able to design nuclear weapons without ever testing them in meatspace. If it's good enough for nuclear weapons, I am strongly convinced it's good enough for something as simple as stress analysis.

tempguy9999 7y ago
I'm neither an aero engineer nor a nuke weapons designer but I am pretty sure that nuclear weapons are a whole lot more complex than a plane. As for not testing nukes, I'm damn sure they test the relevant non-nuclear parts such as the lensing (conventional) explosives. > ...for something as simple as stress analysis Of entire wings? From the article "..such as using machines to bend the wings to extreme angles and shaking the fuselage until it cracks" Your job isn't in this area I guess. I'm sure of it. BTW the model is only as good as its inputs so if you have a defective batch of components (and wasn't there very recently a case where a bunch of deliberately below-spec aluminium parts were delivered to various aerospace companies which cost lots to rectify) then your model had better refelect that defect. But how are you going to find out? With pure maths?
- chroem- 7y ago
  > Your job isn't in this area I guess. I'm sure of it. I'm a mechanical engineer. My point is that stress analysis is the oldest, simplest, and most well understood application of finite element analysis. I can't even begin to imagine the multiphysics that goes into simulating a nuclear weapon, but if we are willing to bet our entire nuclear deterrent on it, it seems reasonable to skip some additional rounds of physical tests that are mostly a formality at this point. Also, defective materials are caught much earlier in the manufacturing process through quality assurance methods, before they become airplanes. And then you apply a factor of safety on your calculations for added assurance.
  - samcday 7y ago
    > Also, defective materials are caught much earlier in the manufacturing process through quality assurance methods, before they become airplanes. Okay so maybe this is an apple's to oranges comparison, but I thought of the story how NASA was scammed by a supplier passing off shitty aluminium as space grade [1]. Is it reasonable to assume that if it can happen to NASA, it can happen to Boeing? https://weather.com/news/news/2019-05-02-nasa-launch-failures-caused-by-faulty-aluminum-scam https://weather.com/news/news/2019-05-02-nasa-launch-failure...
    - chroem- 7y ago
      They check the airframes with nondestructive ultrasonic testing that doesn't involve oblitering entire aircraft. That would be caught at one of several different QA points. We're mostly talking about design verification, which is different.
      - tempguy9999 7y ago
        Parent post gave the example I was remembering. And the company, Sapa, didn't (AFAICT) manufacture the components, they were a third-party testing service! They provided exactly the tests you say should have picked up the problems.
  - 7y ago
gameswithgo 7y ago
people still spot check bridges with physical tests because we can’t perfectly simulate material properties or fluid flow
- chroem- 7y ago
  Absolutely, but NDT is a far cry from testing an aircraft to failure.
ridgeguy 7y ago
> something as simple as stress analysis. Stress analysis isn’t simple. In aircraft applications, the analysis has to model not only metals, but composite materials that have anisotropic and nonlinear responses, account for acoustic energy distribution and local concentration effects, incorporate static- low- and high-cycle fatigue mechanisms (which differ in the materials imperfections that initiate them and in how the ultimate failures play out), incorporate time varying and differential spatial temperature distributions ranging from -55°C to ~1500°C (hot section, commercial engines) and on and on. It’s not simple. I really like physical testing to validate analytical models of complex physical systems.
Bendingo 7y ago
> we're able to design nuclear weapons without ever testing them in meatspace Perhaps you are forgetting about all the nuclear weapons testing that was carried out in the 40's 50's and 60's ? (The last tests by USA and USSR, took place in the early 90's.) [0] https://en.wikipedia.org/wiki/List_of_nuclear_weapons_tests https://en.wikipedia.org/wiki/List_of_nuclear_weapons_tests

rjsw 7y ago

Both Boeing and Airbus have been putting a lot of money and time into improving engineering software, and are working together to do it. Now probably isn't the best time to put out press releases on this though.