Originally Posted by NoDOHC (Post 123311)

Correct me if I am completely mistaken, I will try to explain the figure.

The diagonal lines indicate the position of each component as the rotor rotates.

The angle indications are in rotor degrees, not eccentric shaft degrees.
I believe they are in shaft degrees.

The distance from the left indicates degrees of rotation when that region of the chamber burned.

The different lines are the different ignition timing settings that clearly cause significantly different behavior.
X is observed through a quartz window, O is taken from an ion plug, and the other two are calculated.

The trailing side (due to squish influence ) seems to be the weird one, the flame front displacement seems to shift nicely for all points below the leading plug.

It looks like the flame front is traveling forward just fine, it is the backward part of the curve that confuses me. This would seem to indicate the the flame front follows the rotor rather than the housing (which makes sense, the rotor has the dish).
We should think of the mixture flowing like a river and ignition spacing like dropping two rocks into it. Each wave travels downstream easily, upstream not so well. But what happens when the two waves collide? This interaction slows the flame speed (notice the dogleg in the LL at the bottom of the graph).

I think I see why it knocks right above the trailing plug, the flame front actually reverses direction there (although not relative to the rotor). This must be right at the quench boundary at the edge of the rotor dish (probably when it meets the cusp on the housing).

Do you know what the engine speed was for this test? Do you know the manifold pressure? (I would guess NA).
The X Measurments were done through Quarts windows NA at 1000rpm.

It appears that at 20 BTDC (ATDC?) , the leading front has basically dissipated by 45 degrees (Eccentric shaft, 15 rotor) (which makes sense why you observed the highest pressures when the peak occured at 45 degrees)


Here is a theory about what is causing your knock on too much timing advance. Knock is typically caused by some shock wave colliding with the flame front (it can be a second flame front). My thought is that the leading and trailing sparks both touch off the mix in the chamber if the timing is advanced too far, this results in the two flame fronts colliding while they are stil moving very quickly. To test this, you could try unplugging the trailing plugs or adjusting the ignition split and observing what difference it makes.

This research is awesome. Where did you find that diagram?
This is from a Mazda paper Rotary86v6a4, Fig. 14, showing flame propagation.

Originally Posted by Barry Bordes (Post 123263)

The sensor specs are ±1% for Combustion.
The timing trigger that I fabricated adjacent to Mazda’s timing wheel would be the area for greatest possible error.

To check this a test run is then made where the engine ignition is cut at 6000 rpm and the throttle is opened fully. This double-checks TDC in relation to the logged actual compression hump.

To my knowledge the rest are calculations.

Originally Posted by vex (Post 123449)

By sensor specs for combustion do you mean the transducers? Out of curiousity how did you calibrate your transducers? Did they calibrate them for you and ship them with a cert of it (sorry for the questions, it's the engineer in me again--we've been repeatedly told "never trust some one elses calibration unless you absolutely have to"). Would I be correct in assuming the sensors are linear (yes linear) in nature until a certain point, or are they non-linear from min-read to max? (out of curiosity do you know what the resolution of your timing sensor is? ie: can it read only 1 degree or can it read minutes, or seconds?)

You realize at 360º X 8000 rpm it is taking at almost 3 million samples /min. My laptop is the restriction right now and I would like to add 2 more sensors ( for intake and exhaust ports).

If you know the calculations they're running for any given reading you can easily perturb the uncertainties to get a culmulative percentage of error. That way you can at least know the accuracy of your results. From my glances you could be on the mark or you could be slightly off. Without the error it's hard to gage application to different stress/repeatabilty.

Originally Posted by vex (Post 123527)

Oh believe me I know about samples and windowing. It's annoying as crap ;) Hence why I asked the question.

You are over my head with windowing functions.

As for the 3 million/min... why do you make me do math? I hate math! so you'd be sampling at 48khz unless you wish to avoid windowing then you'd need to sample at 96khz. Is the timing sensor hall or optical? (I'm trying to estimate the sensor error to guesstimate the perturbed error)

The system, I think is 80Khz, and it uses an optical pickup.

so your calibration uncertainty is 0.04% which ain't bad. If you can nail down the timing uncertainty it will be easy enough to get an uncertainty plot for the pressure distribution.

Originally Posted by Barry Bordes (Post 123594)

Originally Posted by vex (Post 123598)

Windowing occurs when you sample something too slowly and the result is a different reading altogether. For instance say you take a sample of a simple sine wave with a frequency (i'll keep it simple) of 4khz. If you sample that same sine wave at some other frequency that is less than 4khz you could get some surprizing results. Windowing would occur and you may end up looking at a cosine wave with a frequency of 2.3khz and a different amplitude that fluctuates over time. Not exactly what you're looking at? For DAQ's it is a good rule of thumb to sample at twice the frequency than what your expecting to be as your function. IE: if you are ever going to see 48khz in a function you'd want to sample double that to ensure you don't have window the function.

There is also another windowing occurance which usually helps in DAQ systems and is fairly easily explained in the following wiki article:
http://en.wikipedia.org/wiki/Window_function

Originally Posted by vex (Post 123598)

For DAQ's it is a good rule of thumb to sample at twice the frequency than what your expecting to be as your function. IE: if you are ever going to see 48khz in a function you'd want to sample double that to ensure you don't have window the function.

Originally Posted by scotty305 (Post 123814)

I don't mean to be contrary, but I received the same 'sample at double the input signal frequency' advice from textbooks and some professors. Other professors with more real-world industry experience advised that data will be much more useful when sampled at no less than 5-10x the signal frequency. Try sampling a 100Hz sine wave at 200Hz, the plotting the results... it's not going to look much like a sine wave. If you logged a fuel injector its operating frequency, the indicated duty cycles would always be 0%, 50% or 100%... Personally I'd like to at least know +/-10%.

Originally Posted by NoDOHC (Post 125370)

The curve is ugly, but I like the timing! You should be about peak power at this point (12-18 degrees ATDC for peak pressure).

Keep up the good work! (Try not to hurt your engine, I don't know how safe it is to run a turbo at optimal ignition timing. Most turbo cars I have seen run REALLY retarded.

Originally Posted by NoDOHC (Post 125556)

Peak pressure still at 38 degrees, if that engine were NA, I recommend that you advance the timing more. Turbo, you probably have it pretty much maxed out.
Actually 851psi @ 40ATDC

Are you sure that the exhaust port opening time on the chart is accurate? I would have expected the pressure to drop significantly more than 12 psi in 15 degrees of eccentric shaft rotation if the exhaust port were open. This is very close to the curve that would be followed with the port closed. I know that the exhaust port opens 75 degrees before BBDC = 195 ATDC, but I suspect that the trace may be inaccurate in scaling (which is weird, because the no-fire trace line looks Ok).
The 5000 psi sensor's error of 1% would be 50 psi. It is probably having a difficult time with resolution down there. The exhaust port opening is written into the engine spec for the program. I estimated it at 196º ATDC. The pressure still has work to do driving the turbine wheel, it probably would show a more distinct drop off if NA. We probably need another sensor on that side of the housing (hopefully in the future one each in the exhaust and intake ports).


Can you continue the trace out to 250 ATDC? Your sensor should be able to read about that long.
No that is the standard span for the program's chart.

Originally Posted by Barry_Bordes

The 5000 psi sensor's error of 1% would be 50 psi. It is probably having a difficult time with resolution down there. The exhaust port opening is written into the engine spec for the program. I estimated it at 196º ATDC. The pressure still has work to do driving the turbine wheel, it probably would show a more distinct drop off if NA. We probably need another sensor on that side of the housing (hopefully in the future one each in the exhaust and intake ports).

Originally Posted by Libor (Post 151968)

Bump this up.

Barry, are you going to do some tests with tighter or zero split?

Also I think there is slight misconception when should peak pressure occur. IIRC at 45° ATDC, rotor has most mechanical leverage over e-shaft, but peak pressure for optimal efficiency should occur much earlier.

Otherwise best of luck with your testing

Originally Posted by Barry Bordes (Post 152029)

But I have exactly the same thoughts as you. Looking at the chart, if we could have the ignition pressure rise as a continuation of the initial compression curve it would seem to be logical.

Barry