View Single Post
      06-19-2008, 08:24 AM   #40
Major General
lucid's Avatar

Drives: E30 M3; Expedition
Join Date: Apr 2007
Location: USA

iTrader: (0)

Originally Posted by jm1234 View Post
Note: I have a masters and part of a doctorate in control system design. For several years I designed aircraft model control systems for wind tunnel usage and then control systems for the wind tunnels themselves. I have no specific knowledge of suspensions, just feedback loops.

I doubt if you had access to the source code that there is a variable that's called spring rate. The point is that the control algorithm is designed to operate in conjunction with the as designed system parameters (including the stock springs). Any control algorithm if connected to a system with different parameters will not operate as designed. The source code could have a variable in it called spring rate and you could adjust it for the new system. I doubt it does because I doubt BMW built this system to be used with any spring. It's a mathematical point, functionally the system may operate exactly the same with different springs.

Lookup tables are not inherently robust or stable. Step changes in any control surface are inherently destabilizing because they introduce a frequency spread of noise into the system. Positive feedback may exist at any frequency. The larger the step, the greater the noise and the greater the likelihood of exciting a positive feedback loop. If a lookup table artificially limits the quantity of control values then this will increase the size of the step change. They could indeed use lookup tables but it would be because the system is inherently stable, not because the lookup table is.

I would doubt the behaviors are too complex to model practically. NASA, Boeing, etc... model airplanes in great detail (many flexing components, high stress, fluid dynamics to get the airflow and vortexes, multiple control surfaces, etc...). They still spend many hours testing but it's generally to validate the models as much as to fine tune the systems. Bridge builders/skyscaper architects build models with many more flexing components than a car suspension. It's easier to test ideas on a car and make mistakes than to do so in other applications so there may be less reliance on fine tuning the models. NASA will spend decades testing models in wind tunnels and then in the air, it doesn't mean there is no model. BMW could build a model, they probably have. Testing is to validate the model (which is not 100%, no model is) and fine tune the design, not replace the model. Having said this, there is no system that you can model "comprehensively" if you are anal enough about defining the word comprehensive.

FYI - I've seen a million dollar model rip itself to pieces in seconds and fly into a wind tunnel turbine because the programmer got a number wrong in a large control matrix. Small changes in feedback loops can (not will) have a dramatic impact.

I saw this video of an out of control feedback loop on a car suspension. I don't think changing the springs would do this...
NASA and Boeing are in a different industry. They don't have any options other than to spend hundreds of millions of dollars to model whatever they can. But please don't tell me they can model everything. There is a guy next door to my office who does CFD models all day long to model very specific flow issues. It takes him months to come up with an "acceptable" model for a constrained case. Modeling buildings is also a different ball game. Yes, there is dynamic loading, but the loads are much more predictable. A car manufacturer will not invest the kind of resources NASA or Boeing invests into modeling the dynamics of a car. If testing was simply to verify the model, they would not need to spend thousands of hours driving the car around acquiring data. I’ve watched some clips on the development of the M3. Their test driver is an engineer. He explicitly said that the design engineers rely on him to assess handling and make suggestions to improve the behavior. Why would that be necessary if they had an accurate model which simply needed verification? Why can’t BMW engineers arrive at a comprehensive accurate model after some iteration so that they can take the test driver off-line? Cars have been around for a long time. Although I am not a dynamics expert, I think you are oversimplifying the variation in potential road and driver inputs and their interactions--the sheer number of scenarios that can entail even if you have an accurate model. And, I doubt that the ECU is doing any kind of deterministic calculation in real-time that includes a general system model of the car, but of course, neither of us knows that. (The reason is that a car is not an unstable aircraft that needs to be actively managed to stay in the air). Computational models are the place to start and they have clearly transformed the way engineers work, but reality is complex and one discovers and learns a lot about the phenomenon while testing; testing is not just to validate the model. Models can and do change during testing. There is nothing wrong with that. (I am not advocating the type of approach the F1 "guru" made to solve the suspension problem although it makes a case about how complex a car suspension system can be if you want every bit of control).

P.S. The video was cool. But, without crunching the numbers, how can you say swapping springs will not do that!?

Last edited by lucid; 06-19-2008 at 09:30 AM.