Rolls, until I did some filtering on it yes I did get a graph similar. But if you collect data for each pulse width and do a statistical analysis of it you get the underlying line. Remember that your data will be out until you get the correct injector data and you will have all sorts of outlying data. Will send you a picture of my injector graph
Yeah even with lots of filtering I still get a negative x intercept and the slope is ~105lb/hr opposed to my 118lb/hr, now we know the ID1000 low slope should be closer to 120 than 105 indicating the trendline is wrong.
Seems too inaccurate with too many outliers to use it to calculate an offset correction yet.
edit: Perhaps another way to calculate offset correction is to backcalculate injectorms from the actual measured fuel mass (eg ms = fuelmass/lowslope), then multiply the injector ms by the current fuel trim. Make sure you filter the fuel mass so it is below your breakpoint so you are only plotting the low slope.
You can plot this curve which will be a perfect straight line and use this x intercept as your breakpoint adjustment. Curious if you do this method if you get the same answer from your data.
edit2: This all seems like a lot of guess work and not something I think that can be automated. I will keep going with my previous method and test it out by putting wrong injector data into my car, logging it, putting it into the calculator and seeing if the answer it spits out are correct.
Ultimately if you have the offset or breakpoint completely wrong your data will be so bad you won't be able to automatically calculate an improvement, it will only work if they are already fairly correct.
Last edited by rolls; 06-19-2016 at 07:29 PM.
How much negative was the intercept? You know at low pulse widths this has a major impact on the fuel trims. Secondly when I tried ID1000s in some ausy Fords the idle fuel trims were around +10% and I also tried using Ford data for GT500 injectors and had fuel trims +28% at idle but even Ford had many different values for this injector.Originally Posted by rolls
Now what type of filtering did you do? With all the data in you will end up the actual injector data in the tune (obviously) but if you back off a bit and collect the sample based on a rounded injector pulse width, adding all the calculated fuel masses for this pulse width and then using the data if you have a reasonable number of points (> 20 worked as a minimum) then I think you may get closer.
Happy to discuss and send you the sample code now that I have found it again
So you actually averaged each injector value for a given fuel mass and plotted that? I was simply doing a scatter plot however a scatter plot trendline should effectively give the same result.
I was filtering all samples below minimum pulse width (and the equivalent fuel mass), I was also filtering all values above the breakpoint (both fuel mass and pulse width).
I don't remember the x intercept but it was reasonably negative which makes no sense as my actual trims are positive in that fuel mass region.
Ah I think I misunderstood what you were saying. You weren't plotting an X,Y scatterplot you were summing all injector ms values for each fuel mass at various steps and then plotting that is that right?
That wasn't what I was doing but what you've described makes more sense. I will try that and see what result I get.
rolls, we are getting closer and I am really keen to see if you end up with a similar result.
I did all my calcs on injector pulse width. I rounded each pulse width to say 0.005ms and added all the fuels masses and kept a count on the number of sample points. Then I would use the data for each point if the number of samples was reasonable (>20) and as you say would divide the fuel mass by the number of sample points.
What I found really promising was that I kept changing injector data to test and this method kept bringing me back to the same injector data.
Ok turns out what you said and a scatterplot are actually a same thing. An XY scatterplot in excel averages sums all the x values for a specific y value so the answer is the same.
I did some more filtering graphing idle and it gives me a low slope of 113.4lb/h instead of the 118lb/h in the VCM, I still get a negative offset that is very close to 0 indicating my offset is probably close to spot on. The slope changing from 118 to 113 seems correct as that will only increase fueling at the breakpoing by maybe 2% which is about right.
Filtering the data is difficult because you have to remove transient data and outliers. I did it manually by trending the data and looking for a spot where RPM was close to static that was at the lowest possible injector ms and then finding some data near the breakpoint.
End of the day these techniques only work if we use the SCT to log airmass, if we stick with the VCM scanner data then we can only emulate this data and shift accordingly.
Interesting stuff!
Last edited by rolls; 06-20-2016 at 03:16 AM.
Filtering the data is not hard, just select the data when in closed loop. The VCM scanner can do this as well. You can make a good guess by just be looking at the LTFT/STFT values.
VCM scanner logs Airmass but at this stage HPT has not fixed it to log the correct value. All I need is to know if what it logs corresponds to anything we can convert back to Airmass.
I don't use excel for this, have programmed it directly. I took it a bit further and did the injector first and then calculated the new SD tables based on the fuel trims, and of course, the follow on effect of the new injector data.
And your "negative offset that is very close to 0" is the correction to the injector Offset.
Excellent work rolls.
I just found you can log "transient fuel" in SCT, that will make it much easier to use correct values and remove outliers.
Someone just asked if they can see the source code for the project. Here is a link the the source for the download in the OP if anyone is interested.
https://github.com/rolandh/DataAnaly...CMHelperBranch
is this able to be used with a wideband at all or only ltft?
You can pick anything you want at all. It loads a csv which lets you choose any variable you have logged on any axis, exactly like VCM scanner/live link does.
i probly didnt word that right, are the end results in the graphs the same as if you had used ltft?
nevermind
thanks for making this available to everyone!
Only if you disable long term fuel trims as the closed loop will mess with your afr.
yeah i understand that, that's why i was asking if it would work with a wideband. i had a better idea of it all after i had some time to actually load some logs.
I made some updates. More filtering options, optimized so it runs much faster with massive datalogs (loads and filters a 100mb csv file in a few seconds).
After lots of playing with auto calculating slopes and offset I gave up using the standard O2 sensors and ran everything in open loop and used a wideband. Going to see what kind of results I get now.
oh my god my head hurts after trying to comprehend what ive read here
Is there an updated download link for this? I need scatterplots in my life!
Reply With Quote