cam profiles

[costonjs]

i did a search on this and found only one thread from 2006 where a guy did some dial indicator readings from an engle 110 cam and produced intake/exhaust profiles. he also did a stock cam for comparison purposes.

i don't want to do it that way. it's easy enough to take a photo of a cam lobe and then plot the lobe, spin the lobe in software while recording lifts @ duration. i only have 1 mild nearly stock cam which i used to build the software. here's a pic of what i'm looking for. the photo is taken straight down the centerline of the cam from about 18" away with the zoom on. i have a piece of white paper behind the lobe. good contrast helps. my camera just doesn't want to focus but it's focused enough.

i'd really like to plot the following: engle fk41, fk42, fk7
the engle W series might be nice to play with too given their popularity.
i'll look at others if anyone wants to participate.

_________________
73 sport beetle blue 1745cc
74 sunbug grey 2016cc

[Alexander_Monday]

I don't have those cams, but here is the raw cam lift data in 0.4 crank degree increments in excel for an FK8 & and FK10/87 that I gathered with my DIY cam doctor.
To look at it in fine detail requires running it through a smoothing algorithm due to dust specks that are unavoidable.

https://drive.google.com/file/d/16DXIar_ZQDQ8NdqR1mZR9g9MWKkdVYX-/view?usp=sharing

https://drive.google.com/file/d/1VkaRNzNsi9SBFEAUxtCzZgLPMVrhIvT9/view?usp=sharing

_________________
Danth’s or Parker’s Law:
“If you have to insist that you've won an internet argument, you've probably lost badly.”

Alexander_Monday->What were the rings gapped at?
bedlamite->Almost enough.

[costonjs]

[costonjs]

[Alexander_Monday]

I was thinking about the noise since I posted the data.
Went to my machine and had an epiphany.
I ran a ground wire from the frame near the gear motor to the rs232 ground from the digital indicator and eliminated a lot of it.
I need to delve deeper into it with my scope.
I think I have a or some ground loops that are picking up noise from the PWM signal to the gear motor.
_________________
Danth’s or Parker’s Law:
“If you have to insist that you've won an internet argument, you've probably lost badly.”

Alexander_Monday->What were the rings gapped at?
bedlamite->Almost enough.

[costonjs]

[modok]

The idea of using optics to measure a cam is a really cool idea!
And a cam doctor is a really cool idea!

But don't forget, the way I did it, worked just fine, and was probably faster too, tho it's just technical difficulties to blame and I'm sure you guys can get that worked out eventually

Alexander.... Turn the thing 90 degrees, (gravity is a handy constant, don't fight it make it work for you) and put it in vee blocks not round bores.

[mikedjames]

Do you have to spin the cam at speed to get measurements? Looks to me like a stepper motor drive might be more appropriate.. Move the cam to position, hold it, average measurements, move. Acceleration can be derived from differences in sufficiently accurate positions.

Or if it has to be moving, synchronise the ADC doing the sampling with a brushless motors driver pulses.


After all it is not like a distributor where it has to be moving at the correct speed to get the cam lobes in the correct centrifugal advance/friction pad dragging derived relative positions.
_________________
Ancient vehicles and vessels

1974 VW T2 : Devon Eurovette camper with 1641 DP T1 engine, Progressive carb, full flow oil cooler, EDIS crank timed ignition.
Engine 1: 40k miles (rocker shaft clip fell off), Engine 2: 30k miles (rebuild, dropped valve). Engine 3: a JK Preservation Parts "new" engine, aluminium case: 26k miles: new top end.
Gearbox rebuild 2021 by Bears.

1979 Westerly GK24 24 foot racer/cruiser yacht Forethought of Gosport.
1973 wooden Pacer sailing dinghy

[Alexander_Monday]

I have some stuff I can cobble together and use a closed loop stepper motor.
And I will redesign to measure from above instead of the side.
Biggest drawbacks to what I have available are that the dial indicator readings via RS232 are not very fast (relative to what it could be if a different communication standard was used) and the older microcontroller I am using has limited RAM to store the results from the 6000p encoder I am currently using.
I am very familiar with programming in the 4D Systems micro but might need to use one of the spare PI's I have and learn more Python the hard way or use a computer and brush up on Visual Basic in Visual Studio.
Having initially learned programming in Fortran, PL1, Cobol, and IBM370 assembler, object-oriented programming and classes just don't come naturally for me.
_________________
Danth’s or Parker’s Law:
“If you have to insist that you've won an internet argument, you've probably lost badly.”

Alexander_Monday->What were the rings gapped at?
bedlamite->Almost enough.

[modok]

COOL!
Keep us updated

[Alexander_Monday]

Got the frame converted with the dial indicator vertical and the stepper motor.
Still using cam bearings in 3D printed stands since I am using a belt drive and it would pull the cam out of alaignment with v blocks.
Why belt drive?
Because stepper motors are usually either 200 or 400 steps per revolution.
Even with microstepping there is not a way to land on exactly 0.1 degree increments which is what I want.
But with a 16 tooth pulley on the stepper and 36 tooth on the cam and 1/8 microstepping you can.
200 step per 360 degrees = 1.8 degrees per step.
(1.8 degrees × (16 / 36)) / 8 = 0.1 degrees.
At the end of the video watch what happens when I remove the ground jumper between the dial indicator RS232 ground and the frame.
That is the noise I was getting.
Now to do a little hardware modification since I don't need to read an encoder or control a DC motor, just either send UART commands to the closed loop stepper or simple step pulses and direction.


Link

_________________
Danth’s or Parker’s Law:
“If you have to insist that you've won an internet argument, you've probably lost badly.”

Alexander_Monday->What were the rings gapped at?
bedlamite->Almost enough.

[Alexander_Monday]

Software is in beta and I am adding features constantly.
I greatly improved stability of the readings with the latest version.
I also spent many hours researching digital data filtering and had to brush up on calculus and matrix math.


Link
[/youtube]
_________________
Danth’s or Parker’s Law:
“If you have to insist that you've won an internet argument, you've probably lost badly.”

Alexander_Monday->What were the rings gapped at?
bedlamite->Almost enough.

[costonjs]

[Alexander_Monday]

If you are using Visual Studio NET Framework or Core you can NuGet a free package called MathNet.Numerics.
It has a function called Fit.Polynomial in the form of Double[] Polynomial(Double[] x, Double[] y, int order, DirectRegressionMethod method).
It does a Least-Squares fitting of the points in an x data array and a y data array to a k-order polynomial y : x -> p0 + p1*x + p2*x^2 +... + pk*x^k, returning its best fitting parameters as [p0, p1, p2,..., pk] array.
That part does the heavy lifting of finding the coefficients like an excel polynomial trend line does, but does not apply them back to the input.

So pick an odd number of data points as a window at least 1 data point higher than the number of coefficients you want and it will return the coefficients.
Then apply the x data point in the middle of the odd numbered window to the coefficients to get the y point that best fits the coefficients curve and put that in the result array.
Rinse and repeat for each data point moving the window 1 data point forward.
I added data points at the beginning from the end and the end from the beginning in the amount of (number of window points - 1) / 2 so I would have a valid set of points for each end of the original data set.
(basically making it circular for the amount I needed)

I am still experimenting with the window size and number of coefficients.
A larger window acts like a low pass filter and a narrower window acts like a high pass filter.
So far doing 3 consecutive passes with windows = 29 => 11 => 5 using 6 => 4 => 3 polynomial coefficients seems to work pretty good.
There is a weighted version Fit.PolynomialWeighted(xData, yData, weighted array, degree) that you can use Savitsky-Golay weights (or any that you make into an array) that I am tinkering with.

There is also a MathNet.Filtering package that includes a Butterworth filter which might be useful but I haven't played with it yet.

If you are using NET Core the package SignalSharp has a Savitsky-Golay filter, but I either am using it wrong or it does not give very good results.
_________________
Danth’s or Parker’s Law:
“If you have to insist that you've won an internet argument, you've probably lost badly.”

Alexander_Monday->What were the rings gapped at?
bedlamite->Almost enough.

[Alexander_Monday]

Well then, there is a much simpler way to get a Least-Squares fitting in Math.Net.Numberics than I was doing

This function will return the fitted point without having to know the coefficients or do the math with them:

[Alexander_Monday]

_________________
Danth’s or Parker’s Law:
“If you have to insist that you've won an internet argument, you've probably lost badly.”

Alexander_Monday->What were the rings gapped at?
bedlamite->Almost enough.

[costonjs]

[Alexander_Monday]

If it helps any, here is the function I use as a filter.
I run the lift through it 3 times, 1st with 29, 2nd with 11, and 3rd with 5.
Then I calculate the velocity by subtracting the lift 2 crank (1 cam) degree before the lift point from the lift 2 crank (1 cam) degree ahead and divide that result by 2 to get the average velocity at the lift point.
Then I run the velocity through the filter the same as the lift.
Rinse and repeat for acceleration and jerk.
I went to this filter which has the matrix coefficients and divisor for the Savitzky-Golay filter for 29, 11, and 5 points to avoid the computational power required to implement the filter as a polynomial for any number of points.