+ Reply to Thread
Page 5 of 6 FirstFirst ... 3 4 5 6 LastLast
Results 41 to 50 of 56

Thread: Testing, Testing 1,2 ...

  1. #41
    Senior Member
    Join Date
    Mar 2008
    Location
    Brighton, MI
    Posts
    686

    Steer under or steer over ?

    A good question about the UG you probably ought to determine for the sake of beating up those design judges who are way out of their element.

    Just the facts, Ma'am:

    1) Constant speed procedure. Takes a LOT of real estate to get the job done. This includes getting it up to speed, collect some data and then hitting the binders without smoking the tires. The test can produce a complete profile of a vehicles capability: nonlinear steady state gains and transient response characteristics of lateral acceleration yaw velocity and roll. You would normally want to a assess steer removal responses, too, because getting into a corner ain't the same cigar as asking the tires to straighten you out. Transient responses can get interesting, too. For all the hype about rice cars, these ugly machines tend to produce traits that even Grandma would object to if her Honda acted this way. Either your simulation matches the results or it doesn't. You can't hide behind a "test Variation" clause because these cars ought to be super-duper, but many of them obviously aren't. (just look at some recent pictures posted on Faceplant). Testing at and airport using crossed runways is a Best Practises move. Plus you can run at any speed the car will go and maintain the speed during the cornering segment. Some people cheat by going in hot and scrubbing bubbles to get a constant average speed.

    2) Constant steer procedure: An easy one as long as the steering is nicely designed and assembled. Otherwise you will have some (usually) puzzling steer position effects. No transient metrics, just steady state gains. As some may recall, I ran this procedure on my Bass boat on Woodland Lake a few years ago to look at propeller blade effects. Still a good test to validate a nonlinear computer simulation. Yes there are/can be speed (aero, tire, tractive force) effects, but the results need to pass a laugh test. All you would really need is yawrate and speed to get you a plot of UG vs. Ay out to the "limit".

    3) Constant Radius Test: Known amount of real estate makes it very practical. Requires a high bandwidth driver 'cause its a closed loop test. (that means you need at least a learners permit to get to the test track, much less operate on it. One benefit of this procedure is the determination of the 'Tangent Speed'. I.e. the speed where body sideslip is ZERO. You don't need an expensive sideslip transducer to get a value of ZERO at some speed around the circle (where the vehicles axis is tangent to the test circle). The tangent speed is a great contest and tuning metric. Changes to the front or rear of the car that raises the tangent speed are 'good'. Things that lower it are 'bad'. There are two ways to run the test. One is to slowly increase speed while staying on track. This implies throttle and differential and aero blending but that's the way it goes. The other method uses segments of constant speed runs. Run it in both directions, please. Your simulation will naturally include all these complex interactions for which you will have no data to characterise them.

    4) The constant confusion test is a last resort used in the Design Event where you wave your hands and paw through you many notebooks and logs to explain why a naturally occurring oversteering yet stable vehicle is horribly understeering during the autocross. Say it ain't so even though you memorized RCVD in 3 languages.

    Don't forget about a Frequency Response test ! When you pipe your iPhone into a Bose, you don't compare the sound at one frequency or loudness level, do you ? All the steady state values and transient response stuff (bandwidth, get it ?) are right there in front of you with a few hand typed Matlab statements. AND, the evil effects of nonlinearities (tires and steering) are way overstated and hyped by the non-believers. (It's Fake Vehicle Dynamics theory).

    All these tests are intended to confirm that you have accounted for all the tire and chassis structural modes in the vehicle that you said you have. And what you said you have I'm sure you feel is the bestest, happiest, warmest, amazing, wonderful, awesome, and incredible university-speak that money can buy. But once you produce a test result with a measured UG, the lawyers will be all over you. Just make sure you have the sim that goes along with it.

    Make it so...

  2. #42
    Lets get excited about frequency response.

    So you've got your UG test, but now with the constant speed test we have transient data? Who cares about what the transient test procedure is, as long as there is enough frequency content - time domain, frequency domain, (z-domain? o_O).

    No matter how the system gets identified you hope you get correlation in both time and frequency domains. Simulating and parameter estimation in time domain is conceptually closer to watching the real thing. Frequency domain is ???. You have a preference?

  3. #43
    Senior Member
    Join Date
    Mar 2008
    Location
    Brighton, MI
    Posts
    686
    The 2 processes are essentially interchangeable. Each one promotes a set of constitutive describing functions based on gain, damping and natural frequency. What makes FSAE cars so special is their VERY low understeer (theoretically) and thus corresponding heavily damped (i.e. sluggish) responses. You can see this is recent videos posted. Actually, the mention and consideration of "damping" (as in 'yaw damping') is almost ridiculous in a FSAE car context because the tires and architecture of them eliminates the traits associated with such metrics. The fact that some cars may have a 'yaw damping' trait means that the car is hurting for some engineering.

    All this boils down to real estate (how much room have you got ?) and what kind of theory do you subscribe to. And if you have the background in solutions to differential equations, a time based formulation of a vehicle's step steer transient response contains 2 gain terms, a damping term, a frequency term and a phase term.

    Note that the interchangability of Time and Frequency domains is demonstrated by the chirp_test and VHSIM simulations posted on my Google drive.

  4. #44
    Hello everyone, the discussion on this thread has been extremely helpful. I was also wondering how useful can estimating TF from just Slalom or DLC data using lsqcurvefit or some nonlinear fitting routine in MATLAB be? I collected data on a non FS vehicle and fitted a TF using not-so-close initial guess for the TF coefficients (num,den) from the bicycle model to match the Ay signal of the VBOX. The results seem to be fairly decent IMO, to check this I fitted the data for a DLC run at a speed close to 65kmph which produced my TF. Further I compared the data for a Slalom at 65kmph with the output of the TF for the steer signal. I am skeptical about the validity of this approach and wanted to discuss it here hoping something useful could come out of it. I understand that the TF won't be as accurate when the tires enter the non-linear regime but can teams use this as a starting point to study metrics?
    Edit: A chirp or pulse with enough frequency content can be more accurate (using system ID) but in this case I did not have enough real estate and a trained driver for doing this on a passenger vehicle.
    Y axis - right in figure 2 is Speed in kmph
    Figure 3 is the USG from this TF, example for metric evaluation
    TF_est_fsaeforum_1.jpgTF_est_fsaeforum_2.jpgUSG_fsaeforum_3.jpg
    Last edited by saurabh_vyas97; 06-12-2019 at 05:26 AM.

  5. #45
    Senior Member
    Join Date
    Mar 2008
    Location
    Brighton, MI
    Posts
    686

    Nice job !

    For a first round experiment, this looks very good ! First suggestion would be to improve/reduce speed variation. With constant speed, you eliminate a very important state variable.

    Next suggestion would be to reduce the maximum/peak lateral acceleration. No need to push the tires that far into non-linear zone.

    Then, forget the DLC. There's really only about 1 frequency in this maneuver.

    A pulse steer input (approximating a step steer input) should work pretty well. I have a Matlab Demo program on my Google/FSAE drive which allows you to play with various types of inputs, filters/smoothing and processing techniques.

    Try to fit your data to a TF involving the num/den of an equivalent bicycle model. That will reveal quite a bit about the makeup of your car (steady state, damping, and bandwidth/response times.chirp_test.jpg

  6. #46
    Hey BillCobb, what's the impact of the relaxation length on the quality of fit? This of course means there are four states to track now.

  7. #47
    Senior Member
    Join Date
    Mar 2008
    Location
    Brighton, MI
    Posts
    686

    Fit Quality:

    Quote Originally Posted by turtle View Post
    Hey BillCobb, what's the impact of the relaxation length on the quality of fit? This of course means there are four states to track now.
    I'm not seeing a huge range of fit quality if the transfer function in Bode Form are used. Yes the resulting damping and frequency parameters that you would obtain will be different.

    2 Examples: a Pretty complete vehicle model with nonlinear tire FY, MX, MZ, etc. and nonlinear MZ chassis compliance. These are all measured vehicle parameter values.

    First case is No Relaxation followed by Slow Relaxation. 'No' is essentially really fast.

    Then I switched over to a 2DOF bicycle model with a tire having only FY generation properties. Pretty much the same results: you can see the relaxation effects progression. However no change in fit quality. shorter response times, less damping, higher damped frequency in all modes (Yaw velocity and sideslip).turtle1.jpgturtle2.jpgturtle3.jpgturtle4.jpg

  8. #48
    Hi Bill,

    Thanks for lending us your clever brains. As mentioned before, I think most teams struggle to get real data due to resource (also difficult to convince people as to why that data is useful, too, when instead they could be building a tangible car), and once they do have the data, they don't know what to do with it. A lot of stuff we learn academically is based on rigid models which are very simplified and are used to teach fundamental theory.

    I had a question about the different compliance terms that you mentioned. Our team is receiving KnC data for an older car. My understanding is as follows (and hoped you could correct me if I'm wrong): In terms of KnC tests, we have lateral non-trailed toe compliance, lateral trailed compliance, and pure aligning moment toe compliance. All in and out of phase. ENF would be obtained from the pure aligning moment (so no trailed Lateral force, just the pads moving underneath to provide an Mz)? And the lateral force compliance EYF would be pure lateral force applied at TCP (no Mz). Technically, the Lateral trailed then should give us a combination of the two and summaries the total effect, but since we have them as separate tests, we can isolate and decouple them better in our simulation, especially since the lateral force and moments from the tyres aren't synchronised?

    The difference between in and out of phase tests will also show the corners (suspension parts) only, and remove the effects of everything in-between such as steering. And then there's the camber compliance side of it too.

  9. #49
    Senior Member
    Join Date
    Mar 2008
    Location
    Brighton, MI
    Posts
    686

    Part-I of the rest of the story

    Have you seen this presentation I had Pat post on the Facebook FSAE site ?

    https://drive.google.com/drive/folde...zlziZZN_7az3Yo

    I have a rear compliance effects deal waiting in the wings, too.

    I'm not sure of the question you are asking. Snow over again, I didn't get the drift...

    The trailed lateral force (Fy) inputs plus the Mz loads are what the tierods will see. Thus the steering mechanism, its mounts, it's transmissibility up the steering column is the bad actor for the most part. As you add caster, the Fy component blooms.

    There's also this piece I posted for Greg Locock (Ford Australia). If you know the tire properties all the cornering compliance allocations are solveable. My own observations (and from others) tells me that the FSAE cars are quite flimsy.

    All the major players can be measured for sufficient accuracy on a wheel alignment machine. Put in some forces and moments, write down the toe and camber readings. Get used to the idea that wheel bearings do have a sizable contribution to the the ENF compliance.
    Sometimes the wheel, too. It's incorrect that they test tires on wheels made from Unobtainium. The wheel may look pretty but can actually be pretty ugly from a compliance standppoint. It all depends on the reference sensors on the K&C machinery.

    https://www.eng-tips.com/viewthread.cfm?qid=463747

    I would like to stress that this ought to strike you as common sense because NOTHING in a factory designed suspension or chassis is left to natural causes.

    Its ALL designed with a certain amount of elasticity in mind for each of the handling related compliance terms. In ALL cases, the tires by themselves can NOT do the entire job.

    If left to the imagination, they will crap all over the car's handling quality.

  10. #50
    Senior Member
    Join Date
    Mar 2008
    Location
    Buffalo, NY USA
    Posts
    333
    Quote Originally Posted by BillCobb View Post
    Have you seen this presentation I had Pat post on the Facebook FSAE site ?
    Thanks Bill, those are great slides. In my FSAE lecture I try to cover some of the same ground. After an overview of tire force and moment properties, applying the tire data to a vehicle model requires operating conditions--what is the load on each tire and where are the tires pointing. That is where K&C comes in because it's not at all obvious how a suspension is going to deflect in operation.

    I strongly recommend that anyone planning to use K&C data witness a test in person. There is a lot going on during the test and there are many detail differences from one rig to the next--some machines (ABD, MTS, original Chevrolet VHF) are comprehensive, others test piece-wise and may not cover every combination of loads and deflections.

    For example, John Ellis* built an early rig which applied horizontal force to move a vertical axis pivot laterally, this pivot supported a pad under the tire contact patch. To apply varying amounts of aligning torque resulting from the applied force, he would move the pivot point (and wheel pad) away from the contact patch center. Of course this meant that he wasn't applying a pure moment (the way it's commonly done now). From memory he developed a clever way to separate the two effects(?) and his machine was much lower cost to build. *https://www.worldcat.org/title/vehic...oclc/655191086

    At the Chrysler Proving Ground, they used to have two separate axle testing rigs, one capable of heave & roll for kinematics, the other for compliances.

    Before designing their SPMM, ABD used their 4-post ride rig with additional fixturing/actuation to measure partial K&C properties for their customers.

    Lotus built their own low cost rig in house, it took multiple setups to apply forces in different directions. They may have sold some copies (not sure).

    The history goes back a good ways. Maurice Olley at GM had a clue. Here's a short quote from our book, "Chassis Design" which is based on his notes from the early 1960s,
    --------------
    3.7 Lateral Deflection Steer

    In addition to the changes in the pointing of the wheels (relative to the median plane) due to the roll of the car, there are sometimes appreciable changes due to deflections in the suspension mechanism under lateral forces. These have been less thoroughly studied than roll steer effects, partly because they are more difficult to measure accurately.

    Probably the only realistic way of measuring them is by instrumentation on the road. However static shop tests give clear indications that lateral deflection steering effects exist. Their importance became apparent with increasing use of rubber bushings.
    --------------

    -- Doug Milliken
    Last edited by DougMilliken; 03-05-2020 at 11:13 AM.

+ Reply to Thread
Page 5 of 6 FirstFirst ... 3 4 5 6 LastLast

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts