Suspension strain gauge loads vs. time

[apexspeed]

Can anyone post a sample of SLA control arm and pushrod strain gauge data from the front or rear end of a race car, hopefully with a few other traces (steering angle, pedal travel, accelerometers?) to illustrate a cause/effect picture? I'm hoping that data will help me fill in a few blanks in my understanding of how forces on the suspension build up and relax for corner entry, exit, and sudden/transient direction change.

[Jersey Tom]

Should be able to answer these questions without strain gauge data (which in itself I imagine would be quite hard to come by). If you want to be able to do internal force resolution you first have to know the external forces on the body. If you want to know the external forces, start with external accelerations.

So, first step - determine accelerations vs. time. Covered in Tune To Win and a variety of driving books... your generic road course corner would be brake in straight line, reduction of braking with some cornering, pure cornering, reduction of cornering with some forward acceleration, pure forward acceleration.

In any of those cases you can come up with the net Fy and Fx that must be acting on the car, and then proportion that to each tire's footprint forces under some assumption of yaw acceleration (let's say we're trimmed at zero in some quasi-steady state case). Once you have the forces at the footprint you can work out control arm loads through statics / geometry.

From the front view, the vertical and lateral loads can be resolved into the pushrod, UCA an LCA quite simply through statics. Suffice to say that generally the majority of lateral load is taken up by the LCA (closest to the ground) and the majority of vertical loads tend to be taken up by the push- or pull-rod (closest to vertical). From the side view, longitudinal forces can be proportioned similarly. From the top view, those forces established at the wheel-side control arm points can then be resolved into chassis-side pickup point forces again through simple geometry, visualizing each control arm as a truss.

[apexspeed]

Thanks, JT. Your reply made me think a little harder about why I thought the data would be helpful. My emphasis was on "vs. time". My understanding is at the level you just described, which I feel is rather quasi-static. I can resolve forces at the contact patch into compression/tension in the upper and lower links, pushrod, and tie rod. I can sum forces and moments iteratively from a car running in a straight line to a car reaching maximum roll angle at maximum steady state lateral acceleration.

But as far as I know, this isn't real vehicle dynamics. This is Gillespie, or Puhn, or Staniforth, etc. It is sufficient to roll a new race car off the trailer for the first time, but that's about it, and only if the competition isn't any smarter.

If I take that narrow quasi-static view and switch it to the plan view, I start thinking about the yaw moment required to enter and exit a corner. The simple, 2-D front view taught me that rise of roll angle was always going to lag the rise of lateral acceleration, if for no reason other than moment of inertia of the sprung mass and damper forces. The plan view implies that the lateral acceleration will lag the force build up. Since the steering axle has to generate force at the contact patch/in the links first, or else there is no yaw moment, no yaw acceleration, then there will be tension/compression in all the front end links, including the pushrods, before significant lateral acceleration has been developed, which means not much rolling moment.

So let's just say all of the above amounts to my first question: how do roll angle and lateral acceleration lag yaw moment?

Then my next two questions would be:

How can I apply this knowledge to transient cornering? (My expectation is that the time constants associated with initial yaw acceleration are too small to be important for anything other than a transient scenario.)

How can I apply this knowledge to roll angle overall? (The quasi static front view iteration didn't reveal that when and how rapidly a control moment is applied would determine the roll rate.)

Please tell me if I'm on the wrong track!

[Greg Locock]

"Can anyone post a sample of SLA control arm and pushrod strain gauge data from the front or rear end of a race car"

Oddly enough the excellent Millikens have done that for you, it is towards the end of Race Car Vehicle Dynamics.

I've done the same thing but in more detail, in ADAMS and it is very interesting.

[Jersey Tom]

But as far as I know, this isn't real vehicle dynamics.

Define "real." Are quasi-steady state simulations and predictions the end-all, be-all of ultimate accuracy and precision? No. For that matter, F = -kx is not the ultimate model of coil spring load vs. deflection, yet it is "real" and works quite well for most applications.

I'd make the argument that no racecar is ever "fully optimized." If you're at a different race track every week (or every other week), if the tires frequently change, or as the rules get shaken up, you constantly have a moving target that you can never totally nail. As such, winning may not be so much finding "that last 1%" as it is finding the first 99%. How much weight do quasi-steady state (QSS) performance carry vs truly dynamic response? Is it 1:1? Is it 10:1? Is it 100:1? That's something you'll have to decide for yourself. May be specific to your application or just built on your own opinions and experiences. As a corollary of sorts I'd say you can shoot yourself in the foot focusing too much on the smaller, ancillary stuff rather than core fundamentals.

But in a purist sense, yes, there are response lags between yaw rate and lateral acceleration, lateral acceleration and roll angle, steer angle and body slip angle - whatever. There have been a number of SAE papers written over the past several decades on transient handling - you can purchase and download these or get them from a University library, etc. How do you use knowledge of transient handling to competitive advantage? Again, that's a discovery in and of itself.

For the time being you can "sandbox" these things, play around and see what they do. To Greg's point, ADAMS is one tool for that. Alternatively you could write a little 2-DOF bicycle model dynamic sim and play around with tire properties, wheelbase, mass distribution, radius of gyration, etc and see what they do. Can add roll inertia and effects, or really make it whatever you want. Or as an even simpler first step you could parameterize a generic corner with some driver line and start by determining the steady-state yaw rate at each step of the maneuver. From there you can back out what the yaw accelerations must be to complete the maneuver, then the yaw moments, and the forces required to achieve them. How do those compare to the steady state forces? Are they big? Or are they insignificant?

You'll find some schools of thought (perhaps in RCVD? or Tune to Win?) making the case that racecar handling really isn't that dynamic. Drivers tend to try to be smooth rather than exercising the car in J-turn maneuvers. As such, the untrimmed moments may not be as large as you think.

[apexspeed]

JT, I take it that your underlying message is run a sim/wrote some code and figure it out for myself. I don't have a problem with that. But it sounds like you're downplaying the value of what I would learn from that exercise (which is not quite the same as implying that I'll calculate insignificant relative magnitudes). Maybe I'm placing too much importance on this study of transient handling, but here's why I didn't think I was doing that:

Let's say I choose a roll stiffness from a target maximum roll angle. That roll stiffness is presumably somewhere between a 1950s Buick and a kart. If I want to keep going stiffer, toward kart territory, to reach the "fastest" response time between turn in and some other measurable event (such as reaching a lateral acceleration target, or peak roll angle, or the build up of lateral forces at the opposite end of the car), there is probably a point of diminishing returns. The plot in my head is that the stiffness vs. response rate curve starts to asymptote, and I wouldn't gain much by going stiffer, and would probably lose something in terms of ride rate, bump compliance, etc.

Is that not the kind of thing you'd expect to fall out of this study of linkage (contact patch) forces vs sprung mass response?

Greg, once upon a time, I was taught to never run FE analysis unless I could get most of the way to the solution via a closed form equation. And I saw enough students proudly presenting completely implausible structural and thermal analyses to know that this was good advice. I do have access to ADAMS. I have no proficiency in it. I guess I have to convince myself that I understand enough pieces of the closed form puzzle before I start trusting myself to apply the boundary conditions to someone else's math.

[Greg Locock]

As I said, it is already done for you in THE book. A thorough understanding of the graphs in RCVD will get you most of what you need I suspect, or at least give you a great insight into where you can go with this.

If you are particulalrly concerned with load transfer, forces build up in the following order initially as the car enters a turn

1) arms (RCH)
2) sta bar
3) springs
4) shocks

But the contribution of each changes during the roll event, so the lines cross over. Also item 1 is rather more complicated as they are also reacting the lateral loads.

In the 50s I think the aim was for the driver to transition smoothly between a series of steady accelerations. This made analysis easy but ignores rather a lot of current driving practice, where the transitions are the major part of the circuit (trail braking into the corner, accelerating away from the apex). this makes analysis much more difficult, and detailed analysis and optimisation of the car almost futile. To give an example from the production car world there are 2 double lane change maneuvers specified by ISO, lane change A and B. A doesn't matter. B is akin to the moose test, roughly one lane width change at 70 kph, with a total event length of around 30m, from memory. You hit about 1g. ADAMS takes an hour or two to simulate this 6 second event while optimising the path through it. That's ONE corner effectively, with no gear changes or braking, and with an endlessly patient driver who is concentrating solely on maximum exit speed while not hitting the poles. Even then, I'm sure Tom will delight in explaining why the results are indicative at best.

Incidentally we do measure this stuff on cars, using wheel force transducers. it is then a relatively quick exercise to work from those 6 forces at each wheel back to a set of dynamic suspension forces. These are used for durability modelling, but a cunning engineer could use them to create or modify tire models.

here's a plot of some of the parameters you mentioned in a step turn, you can see from this that 'lag' is a problematic descriptor.

http://files.engineering.com/getfile.as ... psteer.png

[gixxer_drew]

Most cars now days are equipped with strain gauges, especially on pushrod suspensions, but the problem you will have for trying to derive something like that is the signal to noise ratio on a car in motion on track is pretty bad. Of course you can filter it and do all the normal tricks and you can try and make something out of it, but IMO it wont be the right way to do the job. Don't get me wrong though, it can be very useful for something more aimed at trends, downforce, total load transfers and model validation... but seeing something as fine as system responses I would look elsewhere. Try to get on Dave W's rig my $.02

[apexspeed]

Hmm, there seems to be a post missing since my last look at this thread.

Greg, I do of course have a copy of RCVD, and I believe I know exactly which figure you're referring to (front and rear lateral forces, steering angle, roll angle, and load transfer plotted vs time, yes?). If I'm right, then I assure you that the significance of it was not lost on me when I first looked at it over a decade ago. I'm sure that an answer like that would lead you to conclude that either I'm fibbing to you or to myself about what I really know, or that if I really understand that plot then I had no reason to start this thread. But it really was as simple as I stated in my first post: I was trying to fill in some blanks in my understanding of how forces on the suspension build up and relax for corner entry, exit, and sudden/transient direction change. I can't be more specific than that, because I don't know what I don't know.

Drew, your post about the noise is well taken. I asked for data because I know that when someone hands me a "standard" set of traces for steering angle, throttle plate opening percentage, brake line pressures, vehicle speed, and two or three axis accelerations, I can pick out particulars of that car's or that driver's behavior. If you replaced everything but the speed trace or graphical track position with strain gauge data... I don't know what I'd be able to pick out.

Thanks to all three of you for your replies.

[Jersey Tom]

The underlying message... been trying to think of how best to sum it up, hence scrapping the other reply. I've come up with, "zoom out."

We started by talking about control arm load evolution vs. time - why, though? What does that teach us about handling and performance? Certainly important to the design engineer making a new widget, but otherwise.. drilling down to the component load level is very low level stuff. I would make the argument that control arm load measurement isn't nearly as common as an IMU and capturing several driver inputs along with - and for good reason.

Then if we say the intent is really to capture truly transient handling - why again? How or why do we say that that is truly differentiating performance, more important than or mutually exclusive to steady state handling?

In any event, my feeling is the best way to learn these things are with a "sandbox" of simple simulation tools where you you can see the influence of many things. Whereas with "real" data - which is nice in that it is, well, real - you are at a disadvantage in that it effectively boils down to "Well once upon a time, we gave a car X input and it gave us Y output" without quite as much conceptual understanding of what's going on.

Other underlying message - a very strong one - is that "real vehicle dynamics work" does not mean drilling down to the n-th level of detail in a multibody simulation package. I can do "real" vehicle dynamics work and learn quite a bit about how I would want to design or setup an arbitrary car by using a simulation tool which approximates the car as nothing more than a single point mass without even any suspension to speak of. Might even learn more just that way than diving into ADAMS.

[Greg Locock]

I can't actually find the plots in RCVD now I look.

As to ADAMS vs anything else, the bicycle model works just as well for most linear range stuff, but obviously if I want to evaluate two different damper tunes and two jounce bumpers it is much easier in ADAMS than in some advanced bicycle model.

[Jersey Tom]

Oh certainly, right tool for the right job. Point is that sometimes the best tool for a given project is a full ADAMS model.. sometimes the best tool for the job is a bicycle model... sometimes it's a point model. Sometimes you need full in- and out-of plane dynamics, sometimes QSS sims are the best approach, and no one thing vs another is any less "real" or applicable as far as vehicle dynamics work.

[silente]

apexspeed,

i know it´s not the optimal solution and that to have real data or to use a comprehensive package like ADAMS would be better...But have you thought to use something like rFactor or other driving simulations to try to understand how load is changing with time in a corner maneuver?

Of course, it´s not the final answer and i cannot bet it is 100% accurate. But it allows to monitor many many variables with data acquisition, to change your model quite easily (if you know rFactor modeling "rules" and "language") and to also see what is the effect of different driving styles.

As i said, probably it is not the holy grail and the bad thing is that the basic code is not accesible, but this driving simulations basically work as a multibody code (where you can also model in a quite accurate way suspension geometry, masses, inertias, aero loads etc), although they probably simulate only a fraction of the things that ADAMS simulate. But i guess to have an approximate idea they could be an interesting tool. And they are very cheap...and funny to use!

In the last months i have spent some time modeling some cars into rFactor and building up my tools to do so and i can tell you the differences between the behaviour of a well and accurately modeled rFactor car and the real counterparts are really small.

I wrote something about that here (although i have gone much more forward after my last post):

http://drracing.wordpress.com/

[DaveW]

I asked for data because I know that when someone hands me a "standard" set of traces for steering angle, throttle plate opening percentage, brake line pressures, vehicle speed, and two or three axis accelerations, I can pick out particulars of that car's or that driver's behavior.

The car's behavior, yes, although I would add yaw rate to your list. I'm not sure about driver's behavior, however.

If you replaced everything but the speed trace or graphical track position with strain gauge data... I don't know what I'd be able to pick out.

I'm not sure anybody else would be able to help you.... I noticed that you included all the control arms on your wish list. That is quite right (plus position measurements to define the geometry, being pedantic). The problem is that (in my experience), teams usually confine themselves to just push rod loads, and that is part of the reason for gixxer_drew's comment about S/N ratio.

Try to get on Dave W's rig my $.02

That's kind, but not completely true. In my biased view, it can help but it is certainly not sufficient. Rig data combined with K&C data would be required to validate Greg's models, assuming the vehicle is available to test.

Overall, I like JT's approach. At least it would force you think about the problem. I know the some teams combine a mathematical model with track measurements, using the model as an elegant filter. That approach can be misleading if the model is not accurate.

[gixxer_drew]

I asked for data because I know that when someone hands me a "standard" set of traces for steering angle, throttle plate opening percentage, brake line pressures, vehicle speed, and two or three axis accelerations, I can pick out particulars of that car's or that driver's behavior.

The car's behavior, yes, although I would add yaw rate to your list. I'm not sure about driver's behavior, however.

If you replaced everything but the speed trace or graphical track position with strain gauge data... I don't know what I'd be able to pick out.

I'm not sure anybody else would be able to help you.... I noticed that you included all the control arms on your wish list. That is quite right (plus position measurements to define the geometry, being pedantic). The problem is that (in my experience), teams usually confine themselves to just push rod loads, and that is part of the reason for gixxer_drew's comment about S/N ratio.

Try to get on Dave W's rig my $.02

That's kind, but not completely true. In my biased view, it can help but it is certainly not sufficient. Rig data combined with K&C data would be required to validate Greg's models, assuming the vehicle is available to test.

Overall, I like JT's approach. At least it would force you think about the problem. I know the some teams combine a mathematical model with track measurements, using the model as an elegant filter. That approach can be misleading if the model is not accurate.

. Other than s/n, I also tend to think it would be a huge amount of work to maintain all of those sensors track side. So intended to push toward JTs model or lab. Push rod strains alone are enough pain, budget solves everything.