Roll angle sensor

[3rd_gear_Ted]

I would like to integrate one of these Bosch roll sensors to my AIM unit, the goal is to start to gather roll angle data from multiple cars that are set up in a similar fashion.

In the rosy world of the fuzzy future that data would lead to some anti roll bar performance optimization and hopefully some improved overall GRIP.

Please chime in as you all see fit.


http://www.bosch-motorsport.de/media/catalog_resources/Lean_Angle_Sensor_LAS-1_Datasheet_51_en_2778855307pdf.pdf

[jpnovak]

Doesn't your AIM have a 3-axis accelerometer? Just calculate the roll angle from the output g data in the top/bottom and side/side planes. It just takes a little geometry knowledge.

FYI, The Bosch sensor is only a 2 axis accelerometer putting out an angle like an inclinometer. Its the same measurement.

[3rd_gear_Ted]

From the dashboard mounted AIM I can use geometry to figure out the front or rear roll angles. Oh well, I tried

[Flieger]

Originally Posted by 3rd_gear_Ted

From the dashboard mounted AIM I can use geometry to figure out the front or rear roll angles. Oh well, I tried

Front and rear roll angles are the same in theory, which is why anti-roll bars can be used to alter the roll couple distribution. If you've got a coupe it should be reasonably valid. With a targa or cab you could have a significant amount of chassis twist. The softer the chassis the less effect your suspension changes have. I think the minimum stiffness you need is 10 times the suspension stiffness.

If you have an x, y, and z acceleration you can approximate which way is up but the data will be quite noisy due to vertical accelerations from bumps and pitch, so if you just tried to make there be a 1 g component in the z direction then you'd have errors in the lateral and forward acceleration at times. On a smooth skidpad you could probably get a good measurment. Just look at what the z component is before you start because it won't be exactly 1 g. And, for that matter, neither will the others. You will have offset error due to the unit not being perfectly level and gravity not being perfectly 1 g and the accelerometer not being perfectly accurate. You will want to make sure the skidpad is level too since the the banking of the track will be part of the roll angle you calculate.

Laser ride height sensors is probably the best way to do it but not really doable for the DIY'er.

[Flieger]

BTW, a radio-controlled helicopter gyro (also found in some smartphones) should be just about as good as the lasers but will need correcting for the slope/bank of the road surface. It will give you an angle(s) reading rather than linear accelerations, but that angle is relative to inertial space not the road surface.

Something that might be cool would be to do one lap at walking speed so that there is no roll or pitch and then the gyro data would be giving you the road slope and bank alone and you could subtract that data from the set taken at speed to give you the angles relative to the road. You could also then resolve your acceleration components into components relative to the road so you know the true lateral acceleration achieved and how much vertical acceleration was helping or hurting you.

Laser ride height sensors would help account for aerodynamic forces though.

[3rd_gear_Ted]

Max,
Thanks for the input . here is a couple questions and some more situation info

Does a Safety Devices Roll Cage with some sill gusseting front & rear welded to the sills answer the mail for the stiffness factor needed to get valid data?

I have access to some FFT (Fast Fourier Transformation) hardware that slices, dices, offsets and massages the sensor signal, I hope to be able to plot some kind of RMS signal value for the roll and then time relate it to overall lateral G's.

To address the precision level mounting of the sensor to the car, I have a "laser string" that lines up the spindle C/L's along with a machine tool grade precision level.

With the oncoming rain season, I hope to get some raw "smooth" data from the asphalt jungle here in SoCal and have some fun too

[Matt Romanowski]

I haven't seen G data to determine roll anywhere, probably because it would only be a very rough estimation. Anyone who uses roll data does so from suspension movement and/or laser ride height.

Also, while in theory front and rear roll gradients would be the same, they are never in practice. When measured, the ratio of roll front to rear is a very important setup tool to get the car back to a similar good setup (or away from a bad one).

Comparing the data from car to car could give you some idea of what is better or not, it won't give you much really good data as each chassis and driver preference will be very different.

If you want to learn more about it, check out A Practical Guide to Race Car Data Analysis
or Analysis Techniques for Racecar Data Aquisition

Also, I'm an AiM dealer and can help you out with shock pots or any other sensors you might want. Email me at matt@trailbrake.net if there is anything I can help with.

[Dave at Pelican Parts]

I've only seen G data used as roll data once or twice, and it was ludicrously erroneous. (Sorry, I wanted to use polysyllabic words to make myself feel more edumucated.) The driver was seeing readouts of 45 degrees of roll or more. Yeah, not gonna happen unless you finish up on your roof...

The inertial measurement sounds like it may have some promise. It would be relatively low-impact in terms of equipment (i.e., you don't have to instrument each shock) and would certainly be more accurate than G readings.

And there's something to be said for just using video to estimate an overall roll angle. If you can pick a reasonable horizontal or vertical reference in frame, you can measure from a known line on the car to there... It is very low-cost, but exceedingly tedious and only gives you an average for the whole car.

--DD

[Matt Romanowski]

Most times you want the suspension movement along with the calculated body roll (done through shocks pots) to figure out everything. Having body roll by itself doesn't help too much because you don't know the suspension movement. It's the suspension movement that will tell you where in the camber and toe curves the wheel is and that has to be added to the body roll to find the end wheel camber.

[Flieger]

Originally Posted by Matt Romanowski

I haven't seen G data to determine roll anywhere, probably because it would only be a very rough estimation. Anyone who uses roll data does so from suspension movement and/or laser ride height.

Also, while in theory front and rear roll gradients would be the same, they are never in practice. When measured, the ratio of roll front to rear is a very important setup tool to get the car back to a similar good setup (or away from a bad one).

Comparing the data from car to car could give you some idea of what is better or not, it won't give you much really good data as each chassis and driver preference will be very different.

If you want to learn more about it, check out A Practical Guide to Race Car Data Analysis
or Analysis Techniques for Racecar Data Aquisition

Also, I'm an AiM dealer and can help you out with shock pots or any other sensors you might want. Email me at matt@trailbrake.net if there is anything I can help with.

To be clear, I am talking about roll angle being the same front and rear, not roll stiffness (which is what I think you mean by roll gradient). The front and rear are "rigidly" connected and so constrained to have the same roll angle, that of the chassis. The difference between theory and the real world is that chassis are not infinitely stiff.

[Matt Romanowski]

The chassis may be stiff, but the suspension is not. The suspension for each corner moves independently, so the roll is not the same front to rear. When you graph this, you will see that the front and rear do not roll the same amount (well, they could, but nearly never do).

Here is a graph of front and rear roll angle vs lateral G from my car. I've also included the calculation so you can see where the info comes from.

[Flieger]

I stand by my statements and your data supports what I said about chassis stiffness. In this case, the green is rear data and the red the front. The rear is rolling less than the front which is consistent with chassis twist with a rear heavy car with front-biased roll couple distribution.

Here are a few ways roll can be measured to be different front to rear, only #1 is "real" in my book.

1. You have laser ride height sensors mounted to the chassis front and rear. Your roll couple distribution does not match the weight distribution and you have a torsionally soft chassis. If you have a very stiff front end on a 911 (which would also cause a wheel lift) then you are transmitting a torque through the chassis, it twists and so the front ride height sensors will give you a smaller roll angle than the rear. This is true with the lasers whether the front wheel lifts or not.

2. You are measuring damper travel. The front inside tire lifts off the ground. Now the car squats down on the outside rear tire while the front has already transferred all the weight it can. You see the front roll staying the same while the rear increases. You would see the roll angle level off with lateral acceleration. What is really happening is you have a combination of roll and warp and operating in this regime will not help you at all to tune anti-roll bars to change roll couple distribution because now the front cannot contribute any more roll resistance so your roll couple distribution is now just a function of the lateral acceleration since all additional roll couple is taken by the rear.

3. You are measuring damper travel. Your front tires have a significantly different spring rate compared to the rear. The tire spring rate factors into the roll stiffness (so your actual roll stiffness is less than you'd calculate and the distribution is now different as well because of the front to rear difference). You are only measuring part of the total deflection of each corner, that taken by the suspension. You are missing the tire deflection and so there can be differences in the front to rear "damper roll" but when factoring in the tire deflection they would be equal if the chassis was "rigid".

4. Either lasers or damper travel. Car drives around a banked turn and heads onto a level straight section. Road twists therefore roll angles are different front to rear in a transient. This is also a case of warp superimposed on roll.

5. Trigonometric or other installation ratio errors.

In the end, the principle of using anti-roll bars to tune the roll couple distribution is based on having the same roll angle front to rear, otherwise if they were totally independent there would be no way to change the handling with anti-roll bars. There'd be no way for the front to talk to the back.

[Matt Romanowski]

This data is from a 914, but if you check out the literature, you'll see this is true for all cars, including F1 and DTM cars. Have you worked with this data before?

[Flieger]

Originally Posted by Matt Romanowski

This data is from a 914, but if you check out the literature, you'll see this is true for all cars, including F1 and DTM cars. Have you worked with this data before?

I will admit I have not worked with these sensors before but if you are giving me good data then that is irrelevant.

In the case of F1 the tires are a huge proportion of the suspension and they historically had very rear biased weight distribution and wildly different tire sizes. I doubt you'd have any accurate recent F1 data as that would surely be classified or else they would give disinformation.

All cars have compliance in the chassis and tires.

When talking about roll I would use the roll angle of the chassis relative to the road measured at an y-z plane through the center of mass. Then the differences in that roll angle at the front and rear I would add in as additional suspension compliance.

I am not specifically arguing about the roll angle being different front to rear, it was how you implied the cause of it was the "normal" suspension when really it is compliances in the chassis and tires.

[3rd_gear_Ted]

Hey guys, I really appreciate this discussion, but I would like to explain further what I'm after.
To be simplistic and do stuff I can understand, I model a car going around a 90 degree corner with the same dynamics as a CNC machine controlling a PID loop; The P (proportional) factor is the DRIVER making a steering wheel input, The I (integral)is that DRIVERS steering input being "integrated" at a some kind of acceleration rate (G's) into the chassis and suspension, the D (derivative) is the stuff that Max & Matt know very well.
When Dave mentioned the G reading of 45 degrees of G's, I translated that to mean the car integrated the 90 degree steering input into 45 degrees of instantaneous G's, as the derivative effect of the suspension stuff slopes into effect, the car is at max lateral G's of overall. In simple terms I want to measure the Gain or response rate of the DRIVERS inputs. Compare this data with a good driver in your car to yourself. Hope this makes sense,

[Matt Romanowski]

The 45G number was wrong. The car did not experience that at any point in Dave's example.

So are you asking for a way to compare steering inputs from driver to driver? I think I understand what you want and you can do this through corner radius, curvature (inverse radius), and steering smoothness average. This will tell you who is turning the steering when and how much. It will also give you an idea of who is sawing at the wheel and not. Just be careful, overall smoothness is not always the fastest at some tracks.

[Dave at Pelican Parts]

Originally Posted by Matt Romanowski

The 45G number was wrong. The car did not experience that at any point in Dave's example.

Correct--mine was just an example of how data can be very badly mis-interpreted by someone who doesn't really know what is going on.

--DD

[Flieger]

Originally Posted by 3rd_gear_Ted

Hey guys, I really appreciate this discussion, but I would like to explain further what I'm after.
To be simplistic and do stuff I can understand, I model a car going around a 90 degree corner with the same dynamics as a CNC machine controlling a PID loop; The P (proportional) factor is the DRIVER making a steering wheel input, The I (integral)is that DRIVERS steering input being "integrated" at a some kind of acceleration rate (G's) into the chassis and suspension, the D (derivative) is the stuff that Max & Matt know very well.
When Dave mentioned the G reading of 45 degrees of G's, I translated that to mean the car integrated the 90 degree steering input into 45 degrees of instantaneous G's, as the derivative effect of the suspension stuff slopes into effect, the car is at max lateral G's of overall. In simple terms I want to measure the Gain or response rate of the DRIVERS inputs. Compare this data with a good driver in your car to yourself. Hope this makes sense,

I am familiar with PID controls but I don't follow what you are talking about.

My understanding was that you wanted to measure roll angle and lateral acceleration to get an idea for how your sway bar changes were affecting roll. Is that correct? The best way to do this is a big open expanse of asphalt or concrete for a skidpad with a circle marked by cones as the inside radius.

Either that or you just want a roll vs. time response to a step steering input.

[3rd_gear_Ted]

To all,
Sorry for the confusion. I left out one important word

The goal is to measure the roll angle RATE of acceleration in G's.

Idea being is measure how responsive (how much gain) the car has as the steering inputs get translated into the car as body roll.

[Walt Fricke]

Isn't acceleration a rate to begin with? What's the term for the differential (tangent) at any given point on an acceleration curve, though?