Q for the enginmeering wizards in the group re EGT's

[heinz wittenbecher]

I'm collecting egt data in the hope of being able to graph it at some point and take note of changes with time.

However, now I'm not sure it'll work.

I'm collecting, every 10 seconds at the moment, all 8 egt's and also a spread between coldest and hottest.

My thought/hope was that if I maintain the same spread for same rpm, manifold press, etc it would be an indication of engine health as it would be unlikely that all plugs fouled same, etc. so if a ring was failing or some other horrible thing, the egt should show it and it should show up in the spread. Or if a plug lead falls off if should be noticable as well. ( don't laugh, it's happened

What I'm getting though is variations when I didn't expect any. I.e. it'll be a 200 degree spread and all of a sudden jump to 400 and then back to 200 (not accurate, figures are only for illustration purposes but they do vary that much). The high one is only momentary, perhaps only 1 or 2 hits but the do exist.

Is it possible that firing instances are such, occasionally, that one is just firing and another sucking cold?
I havn't checked the actual temps at that time yet... still working on my database/graphing scheme.

perhaps 10 seconds is too coarse? I'm going to go for constant on the next leg of my trip, but was just wondering if I'm totally off-base in my hope and quest of decreasing the chance of unexpected failure in this manner.

Thanks in advance for thought and comments.

Heinz
...with too much information

 

[jim white]

> > >spread between coldest and hottest.
> >
>
> >i think your egt probes are to far from the head....should be in the
area of 2-3 inches...
jw

 

[heinz wittenbecher]

>

>
>spread between coldest and hottest.
>

>
>Heinz -- I not clear as to what triggers the taking of the exhaust gas
>temperature of each cylinder. Is it the timer that takes it exactly every
10
>seconds, or is it being taken with the piston at exactly the same point
(such
>as when the plug fires) approximately every 10 seconds?

The timer is stricly time, so as you suggest could be at a different point
of the stroke each and every time. I would think it's accurate but is
timeclock based, or clock based.

I can specify the interval that data is written to disk and I believe if set
to 2 seconds it'll be continous as the it scans all sensors every 2 seconds.

I record outside air and while barometric, fuel etc would have an effect I
would expect it to be minute compared to impending failure or failure of
something... but then I'm just guessing

I also record speed, rpm, manifold pressure so hope to be able to twll if
relatively flat ground or storming up the grapevine.

>
>Your timer might have just enough variance so that the piston might be in a
>different position when the sensor takes the reading. There can be quite a
>bit of difference in the instaneous temperature readings during the various
>parts of the four stroke cycle.

That's a definate I would think.

>
>Perhaps you can smooth the data by taking an average over a time period or
>perhaps using one of the smoothing techniques such as least means squared
or
>a variety of other analytical techniques.
>
>Also - when you want to compare data from one day to the other you might
>encounter changes due to changes in the barometric pressure of the outside
>air and with the temperature of the outside air. Also the fuel can change
>from tank to tank even if you use the same pumps at the same dealer.
>
>My son-in-law is an automotive engineer and a manager at E.F. Roush in
>Dearborn, MI. I'll ask him if he or any of his staff is familar with what
>you are trying to do and let you know.

Thanks very much.
I figure if some folks can do it with a simple vacuum gauge (analize health
of engine) I ought to be able to figure out a way to do it too

The whole object of the came is to improve chances of breaking down at home
vs clear accross the country [g]
It probably won't work out that way, but it's a goal.

Heinz

 

[hdavis]

>I'm collecting egt data in the hope of being able to graph it at some
point and take note of changes with time.
>
>However, now I'm not sure it'll work.
>
>I'm collecting, every 10 seconds at the moment, all 8 egt's and also a
spread between coldest and hottest.
>
- -snip-

>perhaps 10 seconds is too coarse? I'm going to go for constant on the next
leg of my trip, but was just wondering if I'm totally off-base in my hope
and quest of decreasing the chance of unexpected failure in this manner.
>
>Thanks in advance for thought and comments.
>
>Heinz
>...with too much information
>

Heinz,

this is one time that you have too little information ;>

Here's what I'd suggest that you do and why:

1. increase your sampling rate to 2x the max permissible engine RPM

The underlying events are syncronized to the RPM. Assuming that each
revolution has a frequency component that is of interest you need to sample
at twice the frequency (you'll see this referenced as the Nyquist rate in
books on signal processing).

2. collect samples at the higher frequency for 2-3 times the amount of time
required for you to currently see the EGT excursions occur

3. try filtering the sampled data with a series of low pass filters (LPF)
to see which one(s) give you representative indications of the EGTs

4. now do some real time signal processing to produce the LPF results from
#3 and save THOSE results at a frequency that seems resonable to you.

5. capture excursions that exceed the normal operating conditions of your
engine. Do so by capturing the time and EGTs. You will probably want to
keep an "anomaly" log separate from your periodic samples. You can always
resample the sampled log if further processing needs to be performed.

You may also want to capture other engine parameters such as vacuum,
coolant temp etc to further correlate engine performance vs excursions.

Henry

 

[hdavis]

>Ouch, my head hurts Henry... but thank you for your excellent comments.

Sorry to make your head hurt!

>
>I do collect more stuff, like vacuum, rpm, speed, water temp, 3 temp
probes on the block itself, and a few more.
>
>the monitor runs like a hobbs meter and sends the data at the interval you
set. 2 seconds is probably continous. I'm going to try that.
>
>We then filter the data into a csv file with hobbs time as row and all
parameters as columns in Excel.
>
>And that's where it gets interesting... i.e. condensing it into a graph
are normal/abnormal indication.

I'll see if I still have any papers on realtime engine monitoring. There
have been a number of research projects presented at automotive conferences.

>
>We're in the process of reading the data realtime on a laptop, displaying
the parameters we want and writing out a filtered csv file as well.

One of the reasons to implement a low pass filter is that you can band
limit the data that you collect. So, the data that you keep has some
relationship to the ra input. Of course, if the momentary excursions are
really meaningful, you have no choice but to sample at the higher frequency
and retain the higher frequency data. Based on what I know of the auto
industry's monitoring program during development, you'll be fine collecting
data at a slower rate - but sampling at a higher rate. Chances are that the
excursions that you see are actually there in every cylinder for every
ignition sequence. That's where the LPF will help you. By filtering you'll
smooth out the excursions and only see the longer duration excursions.

Henry Davis Consulting, Inc / new product consulting
PO Box 1270 / product readiness reviews
Soquel, Ca 95073 / IP reviews
ph: (831) 462-5199 / full service marketing
fax: (831) 462-5198
http://www.henry-davis.com/ http://www.henry-davis.com