TD_D mod journey. From stock to the 'bastardmobile'

[HolyCrapItsFast]

I usually just start with another injector's values that has the same value as yours at 14 volts and then make adjustments from there. It's not the most ideal but it is the only method I know.

The only other way would be to drive the injectors at the other voltages and make adjustments for each. Not at all practical

 

[Td_d]

^ That's exactly what I've done - got a reference point of another injector with 1ms at 14v, and checked the curve as well versus the stock numbers, tracking relatively well. I'll use that and take it from there.

 

[Td_d]

So, with the HG in the clear, and the injectors on, I'm about to get to the whole point of this exercise

Boost!

Just trying to set a workflow up for myself - once I've nailed injectors scaling, latencies and MAF scaling down, I want to crank the boost up - I should be able to go up to 2 bar (not that I'm going to yet!) with the built engine, no problem given the spec. Given that higher target boost is reached at around mid 4000's, I assume that in tuning in the higher boost, it would be prudent as a starting point to drop all timing from that load/RPM point onwards down? 1* to 2* I would think - and then check for no knock, and slowly edge up?

 

[Td_d]

Thanks Fuji - would make life easier if I had a reference point. I've come to realize that I'm likely to do my own tune far more thoroughly than any tuner, sadly...

As a matter of interest for anybody running FMIC's and having that 'rich dip' that is so renowed with that - NSFW's seems to have found a whole additional range of logic and variables that come into play in the ECU that cause it (whereas it was always assumed it was due to the physical lengthening of the charge pipes and associated delay / greater air volume). Interesting stuff, still hurting my head though...

 

[Td_d]

It looks like the rule of thumb is roughly to reduce 1* per additional lb of boost. Since my timing tables are currently setup for 1.5 - 1.55bar, if I aim to move up 1.8 bar that's a 4 PSI increase - so would indicate a 4* timing across those areas where boost would now be higher.

My intuition would tell me that I should probably reduce timing a little less, say 1* - 2* in the earlier areas (3500-4000) where boost is not yet peak target, tapering to 3-4* once it's hit peak and up mid 4000' onwards.

 

[HolyCrapItsFast]

Thanks Fuji - would make life easier if I had a reference point. I've come to realize that I'm likely to do my own tune far more thoroughly than any tuner, sadly...

As a matter of interest for anybody running FMIC's and having that 'rich dip' that is so renowed with that - NSFW's seems to have found a whole additional range of logic and variables that come into play in the ECU that cause it (whereas it was always assumed it was due to the physical lengthening of the charge pipes and associated delay / greater air volume). Interesting stuff, still hurting my head though...

I am very interested in this... What thread is it?

 

[HolyCrapItsFast]

It looks like the rule of thumb is roughly to reduce 1* per additional lb of boost. Since my timing tables are currently setup for 1.5 - 1.55bar, if I aim to move up 1.8 bar that's a 4 PSI increase - so would indicate a 4* timing across those areas where boost would now be higher.

My intuition would tell me that I should probably reduce timing a little less, say 1* - 2* in the earlier areas (3500-4000) where boost is not yet peak target, tapering to 3-4* once it's hit peak and up mid 4000' onwards.

I don't use rule of thumb :lol:

I have been using Fugi's method to build boost but once boost is full, I throw what ever timing at it that the thing can handle

 

[Td_d]

The thread is this one : http://www.romraider.com/forum/viewtopic.php?f=25&t=7442

I don't use rule of thumb :lol:

I have been using Fuji's method to build boost but once boost is full, I throw what ever timing at it that the thing can handle

Chuckle, agreed - I plan on test driving that method too - I'm just worried that the current timing which is already quite aggressive will cause knock with boost being upped quite a bit... So, I'd rather give myself some margin to play with, and start stepping it up thereafter. Or (other than the Fuji 'trick'), am I likely to be ok, and only have to worry about it at the very top end (I'm likely to need a new row at the fueling and load / timing end of the tables now)?

Finally drove the car back today, with the 1200's on with rough scaling and latencies - after some learning, drove perfectly fine - other than my old friend the stumble reappearing. Sigh. Well, I expected that. So - my approach is once again to zero the load comps, and start logging to tuck in closed loop nice and tight again, after which I'll nail the load compensations, and move on to open loop. I think the scaling is relatively close, however - open loop fueling, which I was watching like a hawk, was actually tracking pretty close to what the fueling targets are.

I've also decided to redo tip-in from scratch - I got lazy last time, so this time, basement floor. Starting with the stock figures, multiplied by a correction factor of the new to stock injectors (hehehe, 45% less!). Other than the usual stumble, car feels perfectly fine, really looks like I got lucky here.

 

[Td_d]

Yeehahaha... So, I'm being naughty - dialing in both closed loop and open loop at the same time, but this time round I can see the trends in the data, so I feel comfortable doing so.

Slowly dialing up the boost - currently at 1.6bar (~23psi). Zero knock in high load high boost areas, and the car is flying, holy shit. And I'm still way too rich at the top - I found that using the 1200cc as the scaling was actually too lean (as the explanation in Romraider also attests, and a fair amount of reading online - the actual scalar is not technically the real flow - but an estimation - and a lot of people have found that a value 10% to 15% lower was actually a better peg). Started with a much lower figure - 960's - closed loop was much better, but very rich at the top, so I've settled on a scalar of 1040, which is now getting me close to target at the top end. Also found that the advertised latencies (1ms@14v) were not working too well for me - downward facing slope, so I'm slowly raising them to settle at a value that works for me.

I think I'm going to peg the boost at 1.8 bar for the moment (26psi) - I could take it up to 2bar (29psi) max, my setup can handle it, but I think that might be pushing it too close to the edge. Never mind that at 1.55bar, I was already getting early 400whp.

IDC's at full tilt, 1.6bar, highest torque areas? 65%

 

[HolyCrapItsFast]

Nice! I did some open loop tuning the other day on my car and I was noticing the IDC were very low for E85. I have 272 cams, ported and polished spec c heads, 60lbs turbo, 1000CC injectors, headers, turbo-back, big maf and my IDC's on e85 are 75% @ 20psi.

The only explanation I can come up with off the top of my fat head is that the Deatschwerks 300lph fuel pump is responsible. That's really the only fuel related component I replaced in the last year and it kind of makes sense. I recently instructed someone to install a fuel pump because his IDC's were kissing 105%. He did just that and his IDC dropped significantly to 85%. No other mods were performed, just fuel pump. I was amazed.

 

[Td_d]

Nice! I did some open loop tuning the other day on my car and I was noticing the IDC were very low for E85. I have 272 cams, ported and polished spec c heads, 60lbs turbo, 1000CC injectors, headers, turbo-back, big maf and my IDC's on e85 are 75% @ 20psi.

The only explanation I can come up with off the top of my fat head is that the Deatschwerks 300lph fuel pump is responsible. That's really the only fuel related component I replaced in the last year and it kind of makes sense. I recently instructed someone to install a fuel pump because his IDC's were kissing 105%. He did just that and his IDC dropped significantly to 85%. No other mods were performed, just fuel pump. I was amazed.

Jeez - 20% drop, that's a lot. Seems our setups are quite sensitive to the fuel pump and FPR components - just think what a major impact just the fuel duty % variable made...

Can't tell you how nice it is to be back in the STI after the Kia I was testing, irresponsibly of course, the open loop fueling, and got a chance to open her up in 4th nearly all the way to 7000 RPMs - I'm not kidding, you start feeling like tunnel vision!

Tune her out!

:twisted:

 

[Td_d]

Oh, by the way Fuji - I'm well tucked in to disassembling your Rom. I don't obsess much.

 

[Td_d]

Damn it, I'm going to need to make a plan and learn how to use Matlab and those scripts that Nil5 posted on the Romraider forums. These CAN log crunches are killing me - quad core machine busy for an hour on a collection of logs using Excel. It's killing me.

Fuji - you still remember how the basics? I need some pointers.

 

[Td_d]

So, made an interesting discovery trying to nail down load compensations again. I had a very severe stumble at 3000-3200 RPMs, but this time I could see exactly what was going on since I have a wideband gauge now. Tracks fine and then leans out terribly - with the higher boost and bigger injectors, interestingly, it's pushed the resonance out by about 200 rpms. And here's where the discovery happened...

I could not get rid of the leaning out using the load compensations spreadsheet (Airboy's) no matter how many compensations I was making, eventually, the spreadsheet was indicating that my AF correction curve was tracking very nicely - but the severe stumble was still there. And then it hit me - if the resonance had moved upwards in the RPM range, and the spreadsheet was basing compensations purely on AF corrections and learning, once you went into open loop, there could still be a resonance issue - but the ECU would not correct it. On this hunch, I made a fresh batch of data, but using the wideband to commanded fuel delta as 'AF correction', and a zero learning value (since it captures the full error) - and hey presto - this time the spreadsheet showed 'corrections' of over 17% in the stumble area. Adjusted the maps, and stumble gone.

 

[Td_d]

Wow... I got the Matlab scripts to work finally. George - I know you said you'll stick with what you know, but look at this - this was on an 82000 line, 16Mb log. Fuck me... like Fuji said - Matlab is a beast when it comes to crunching numbers.

>> el_comp
Loading data...
(1) log0125.csv
Elapsed time is 0.709293 seconds.
>> write_excel_table ('test.csv',mrp,rpm,new_comp_matrix)
>> write_excel_table ('test.xls',mrp,rpm,new_comp_matrix)

 

[Td_d]

Once I got my head around how the Matlab scripts worked, I put in an attempt to reversion the load compensation scripts to do closed loop scaling, which by far is the most time consuming using excel. After some figuring out, I managed to get a working script. What took Excel 20 - 25 minutes to do on a 4Mb file took Matlab 26 seconds... absolutely brilliant, this was always the most frustrating part of crunching the numbers. The great thing is also that the script scours the directory for all *.csv files and loads them up simultaneously into the same 'pot' - as long as the headings are what the script expects (Matlab is very picky about syntax). I used holy's filtering criteria, plus left in the throttle delta filter from the load compensation script - I figure that removing tip-in can only make the scaling cleaner. If I have the patience, I'll see if I can create a script to handle tip-in - would be great if I could use any log to calculate tip-in errors, as opposed to having to go out and specifically take a drive and generate various % angles.

Awesome! And if anyone is interested, one of the admins at the Romraider forums has altered the scripts to work with Octave - an opensource, free version of Matlab.

 

[HolyCrapItsFast]

So, made an interesting discovery trying to nail down load compensations again. I had a very severe stumble at 3000-3200 RPMs, but this time I could see exactly what was going on since I have a wideband gauge now. Tracks fine and then leans out terribly - with the higher boost and bigger injectors, interestingly, it's pushed the resonance out by about 200 rpms. And here's where the discovery happened...

I could not get rid of the leaning out using the load compensations spreadsheet (Airboy's) no matter how many compensations I was making, eventually, the spreadsheet was indicating that my AF correction curve was tracking very nicely - but the severe stumble was still there. And then it hit me - if the resonance had moved upwards in the RPM range, and the spreadsheet was basing compensations purely on AF corrections and learning, once you went into open loop, there could still be a resonance issue - but the ECU would not correct it. On this hunch, I made a fresh batch of data, but using the wideband to commanded fuel delta as 'AF correction', and a zero learning value (since it captures the full error) - and hey presto - this time the spreadsheet showed 'corrections' of over 17% in the stumble area. Adjusted the maps, and stumble gone.

That's awesome. I will have to try this method. So did you do the closed loop corrections separately from the open loop and just compile them into one all encompassing result?

Wow... I got the Matlab scripts to work finally. George - I know you said you'll stick with what you know, but look at this - this was on an 82000 line, 16Mb log. Fuck me... like Fuji said - Matlab is a beast when it comes to crunching numbers.

>> el_comp
Loading data...
(1) log0125.csv
Elapsed time is 0.709293 seconds.
>> write_excel_table ('test.csv',mrp,rpm,new_comp_matrix)
>> write_excel_table ('test.xls',mrp,rpm,new_comp_matrix)

Once I got my head around how the Matlab scripts worked, I put in an attempt to reversion the load compensation scripts to do closed loop scaling, which by far is the most time consuming using excel. After some figuring out, I managed to get a working script. What took Excel 20 - 25 minutes to do on a 4Mb file took Matlab 26 seconds... absolutely brilliant, this was always the most frustrating part of crunching the numbers. The great thing is also that the script scours the directory for all *.csv files and loads them up simultaneously into the same 'pot' - as long as the headings are what the script expects (Matlab is very picky about syntax). I used holy's filtering criteria, plus left in the throttle delta filter from the load compensation script - I figure that removing tip-in can only make the scaling cleaner. If I have the patience, I'll see if I can create a script to handle tip-in - would be great if I could use any log to calculate tip-in errors, as opposed to having to go out and specifically take a drive and generate various % angles.

Awesome! And if anyone is interested, one of the admins at the Romraider forums has altered the scripts to work with Octave - an opensource, free version of Matlab.

I have to know more about what you are talking about here. To be honest this is way over my head but I want so much to learn. Is Matlab integrated with excel or do you use them separately. WTF is Matlab anyway?

Sorry for my ignorance!

 

[Td_d]

That's awesome. I will have to try this method. So did you do the closed loop corrections separately from the open loop and just compile them into one all encompassing result?

Yup - it's somewhat manual in that I trust the A/F correction and learning figures that the ECU throws at AFR more than a straighforward calculation of wideband AFR minus targeted fueling (since there might be other compensations that we don't know about being thrown in) - so I did the normal closed loop approach, and then ran it again using the correction as my wideband figure minus the final fueling converted to a percentage, and threw that back into the spreadsheet. It's a bit of a pain in the ass, since you have to look through the data to see if you have massive spikes (from tip-in and deceleration), but I'll figure out a neater way of doing it.

Upshot is that the compensations, especially the positive ones around 2800 to 3200 are actually required all the way into positive manifold pressure territory - just like the stock maps.

I have to know more about what you are talking about here. To be honest this is way over my head but I want so much to learn. Is Matlab integrated with excel or do you use them separately. WTF is Matlab anyway?

Sorry for my ignorance!

Holy, I have no doubt you can learn this, trust me - you're the engineer I was getting so friggin frustrated using Excel, so I decided to battle my way through Matlab. It's a mathematical scripting language (in a GUI format) that's specifically targeted at technical computing, and it processes numbers at an absolutely beastly speed. The scripts that were posted basically do what the spreadsheet does - it 'bins' the data - so with MAF according to the headings of the table; then it applies a whole range of filters, as one would do manually in Excel (I plugged yours into the script), and then goes through the data and puts the relevant number in each bin (and then averages it).

I spent a good couple of nights shouting at my computer screen out of frustration But eventually, and with some tips that were posted on the thread, I got the original script working, and then braved writing it to do the closed loop. Saves so much time... I can PM you about the software and scripts.

 

[Td_d]

Hey Jordan! It was such a long delay between getting them and installing that I forgot to give you some feedback - treating me very well! Latency tuning was dead easy actually, even though I didn't have the adjacent voltage latencies. Ironically, I finally reduced my idle back to 850 (since the 'tuner' couldn't get the 750's dialed in, he ramped idle up to 1100, and I simply forgot to drop it), and the car idles better with 1200's than with 750!

I've only just started scratching the surface of what I can do with these injectors - at 1.6bar and only hit upper 60's with IDCs, plenty headroom (which I intend to fully abuse).

 

[HolyCrapItsFast]

Five0 FTW!