john banks

Whilst trying to put the final touches to the code for my electronic boost controller I have been doing some fiddling about to work out how exactly to write the code. I won't bore you with the details, lets just say I tried the following. It is on a MY00, and I am not responsible if you blow your engine up.

WHAT TO DO

1. Remove the hose that goes from the T-piece to the wastegate solenoid and cap it off with say a suitable screw.

2. Take a one inch long piece of hose and put a restrictor inside like the one in the hose off the turbo nipple.

3. Attach this hose where the wastegate hose in step one came off (to the T-piece).

Then gradually increase the hole in the restrictor until you get the boost you want. I would advise going up to 1 PSI below fuel cut.

Should cost about 10p. You can make the restrictors out of a suitable screw shank.

ALTERNATIVELY

Alternatively, pay £7+VAT for a restrictor valve but make sure it has an orifice of about 4mm or more or you won't get your boost very high.

RESULTS

The boost is smooth and spike free, fairly consistent between gears with slower spool up than a Dawes, but much earlier and held longer than standard. Part throttle boost is fine and the fuelling is OK on my car.

I have previously tried reducing the original restrictor size, but with spiky results. It seems a restrictor AND a bleed are required.

I think this method is not new, but I have not tried it like this before, and is the best you will get this side of a Dawes which benefits from quicker spool up.

Please post if you try it or have done anything similar with good results.

DIFFERENCES TO OTHER METHODS

This is not the regular type of bleed valve where you leave the solenoid connected.

TECHIE BIT

The main aim of this experiment was to work out if a static duty cycle would be a good starting point for all revs, which is then modified by target based PI loops.

T-uk

John,

do you think your EBC will be ready for Sunday?if it is,could I get your Dawes?,so that I can get some figures for my car+Dawes to post,as I have already got some consistent Star figures to compare.

john banks

I'm trying, but it will be tight and it won't be a permanent fixture on the car until it is on a proper board.

If you want to borrow my valve to try the above you can. It is what I was trying to achieve on the day we last went out with your car, but I was using restriction rather than bleed to turn up the boost. That didn't work this does.

[Edited by john banks - 1/18/2002 6:43:24 PM]

T-uk

if I get time I will come through,when is the best time to phone tomorrow,do not want to wake you from a night shift.

john banks

Not working until Monday Phone anytime after 9am. You can come for some runs with the EBC I hope - just needs wiring in now and ready to try.

T-uk

"not working until monday",well stop posting here and get the EBC finished

Floyd

iTrader: (9)

John

Surely this will be affected by ambient temperature and density, so fine tuning one day will not mean that it will be OK the next. This happens to all bleed valves doesn't it?

F (the questioning one)

john banks

Yes it will - like the original ECU to some degree - but it doesn't matter hugely unless you are right on fuel cut hence my recommendation to stay 1PSI away.

It will not be any worse than boost control systems which use a static duty cycle map - such as the current Unichip system.

Of course my electronic boost controller will only use the set duty cycle if there is not error. During spool up it uses maximum duty cycle until it gets near to target.

[Edited by john banks - 1/18/2002 10:40:16 PM]

EMS

John,

Look out to apply max duty cycle when the actual boost is lower than target! This works fine when WOT, but when part throttle your turbo is working like mad trying to make the boost. If you are familiar with compressor maps of a turbo: you pass the surge line, as the boost pressure will be quite high at low flow rate.

Mark Verhoeven.

john banks

Thanks Mark. I am running up to 95% duty cycle during spool up (not higher mainly to spare the solenoid) if the gain is high enough. I am planning to add TPS adjustment almost as soon as I get it on the car after tomorrow.

Apparently if you hit the surge line you know about it. I have had no such issues with the Dawes which should be at least as bad as my electronic design at low revs. I find unless I have pretty full throttle openings I don't make enough boost low down, so I'm not too bothered about the surge line. I checked 18 PSI at 2400rpm which I get at WOT on the nearest TD04L compressor map I could find and there was loads to spare. On part throttle I am making little very little boost at this point.

It seems most of the EBCs about except the AVC-R do not have any means of knowing throttle position or engine speed, so that includes MRT and HKS, and folk do not seem to have issues.

However, there are three main reasons for embarking on this EBC - 1. more linear throttle response than Dawes by using TPS 2. no environmental factors 3. ability to make RPM dependent boost map.

1 + 3 will be implemented after I get the WOT design done which is the trickiest.

The above experiment was encouraging because it made me realise I could use one duty cycle for full throttle over the whole rev range and this alone would hold pretty stable boost.

The proportion loop pushes up the duty cycle during spool up and then corrects errors - at least in simulation!

If I need to add integration I have the code waiting. I removed it because I wanted the minimum number of variables to set up. If I get close with just a duty cycle map a la Unichip, I can then correct using only one other factor if the corrections are small I think. We'll see.

EMS

John,

Look out, as you see things a little wrong!

You don´t pass the surge line at WOT but at high revs and part throttle! You have to measure the boost in front of the throttle body, to see what pressure the turbo is actually producing. If you do that at 5.000 rpm and 95% duty cycle you will be shocked. The flow in that case is very low because of the closed throttle and the boost is high because of the high D.C. High boost and low flow makes you pass the surge line (the wrong side).

I whould advise you only to use max duty cycle from 50% throttle and below that use the target duty cycle.

Mark Verhoeven.

Cosie Convert

Mark

This is the point I was making about the Dawes valve as it holds the wastegate shut until set point, regardless of load.
From the sounds I hear from my engine, the dump (BOV) valve helps when the throttle is at small openings by relieving the pressure back to the turbo suction. This maintains some flow and would appear to keep the compressor out of surge.

The only drawback is the increased exhaust backpressure before the turbo, at light loads, tends to hold exhaust temperatures higher than they would be if the wastegate was opening (as in standard set up)
Temperatures I have recorded are still safe >650 Deg C.

I think it is a small trade off considering the improved response
and would not seem to be a problem at 17 - 18 psi,

May well be an issue at higher boost ??

cc

john banks

Thanks Mark again. That system sounds very easy to implement - in fact I'll do it now before it goes on the car. I will check TPS and if it is less than 3V (approx 75%) I will use static DC and only ramp it up if there is an error AND the throttle if pretty wide open. I think the Link does similar at 60% TPS according to the manual. This seems easier than proportionally altering the duty cycle gain according to TPS as you would have to move the target boost as well, and this would require much more setting up.

Should we not be having issues with the Dawes then? At 5000rpm the boost is starting to drop from 18PSI even at WOT. On part throttle it makes much less boost using a larger bleed hole.

If at 5000rpm at 100% throttle taking into account estimated volumetric efficiencies at 16PSI from the nearest compressor map I could find, it looks like the turbo would be flowing about 270cfm. 100cfm is the surge line. So I reckon I would need to see 16 PSI at 37% throttle opening to hit the surge line if indeed the throttle is a linear device.

At 4000rpm things look closer, and with the Dawes this seems to be the point where you get big boost on part throttle if you have no bleed (on my controller this is turning the gain down). Here at 18PSI the flow is about 230cfm with the surge line about half that. So if you can make full boost at half throttle at 4000rpm you could surge.

At 3000rpm it is 180cfm at full throttle with surge at 110, but thankfully on 2/3 throttle at this engine speed it makes nowhere near as much boost.

Is my reasoning sound here?

Cosie Convert

John

Just to clarify. When you say it's not making much boost at part throttle are you measuring pressure on the Engine or Turbo side of the Throttle ?
The Turbo knows or cares not about the pressure in the manifold, as far as surge is concerned it is the pre-throttle pressure that may be the issue.
WOT = no problem, even at low RPM.
It's part throttle ie low flow CFM and high turbo discharge pressure that can put it into surge.

cc

EMS

The flow through the throttle is not linear to the opening position. If you are able to get a Select Monitor, you can read the amount of air which is going into the engine. A pressure gauge you can fit at the pipe directly at the output of the turbo. (look out to connect it in front of the brass restrictor)

I didn´t measure that, but I can assure you you get into trouble with high DC and low throttle.

Mark Verhoeven.

john banks

I am measuring on the turbo side of the throttle. So we are saying a partial throttle "stalls" the turbo because there is too much restriction?

Is a Dawes unsafe then? Effectively it is running 100% duty cycle until target boost no matter what the throttle opening.

I have a spare variable resistor on the board which will now be named throttle position reference and will control the switch to proportional error correction.

Incidentally for those reading on topic - these issues do not affect the system I described at the top of the thread.

[Edited by john banks - 1/19/2002 10:06:55 AM]

Cosie Convert

John

I've said all along I don't think that at 17 - 18 PSI it will be a problem. I was only making the point in case you were taking the pressure downstream of the throttle for your flow map calculations.

It's just that the first time I measured intercooler pressure on my cosworth, I got a real eye opener ! Dashboard gauge (downstream) was measuring 10" vacuum at steady 80 MPH but the intercooler pressure was 22 psi ! Opening the throttle equalised them both at 21 psi (and soon exceeded 80 MPH !)

Not having a go, just hoping to help you make it work well. I think there is the potential to make your EBC control better that any MBC given enough info and development time.

cc

john banks

The help is REALLY appreciated. Now forgive me, I can do the electronic bit, but the greasy bits elude me.

My boost gauge is T'd off the line to the MAP sensor, so it is seeing what the MAP sensor sees. This is before the throttle, correct?

Cosie Convert

No, the MAP is concerned primarily with 'Manifold' pressure therefore you are monitoring downstream of the throttle.

If you put a gauge as Mark suggests, on the turbo outlet nipple, you will see a completely different pressure profile.

Not 100% sure on your MY but this should be the same place that the Dawes inlet takes its supply.

john banks

Will need to pick your brains on this tomorrow. I don't have a clue with the greasy bits, which seems to be your forte!

I have configured some LEDs on the board to tell me when I am near target boost. I will initially set the duty cycle with these. Then I will up the proportional gain to get the response I want (ie max out the duty cycle until near target for a Dawes like response), but gain is zero if the throttle is not above the setpoint.

Will only have to cut one wire at the ECU and tap into TPS and MAP - just trying to see if my NPN transistor will source rather than sink current to allow a common earth and only one wire to cut rather than two.

paul w

Hi John,

I love reading your threads on boost control,reminds me of myself when i used to play about with my uno turbo's,i need a bit more boost and have tried the conventional bleed valve method but tends to overboost when car is full of wife and little pair.
Going back to top of thread did you use a bleed valve at the t piece or just a straight through pipe.

Keep up the good work.

Paul

James_PowerMad

John,

You mention that you are considering using an Integral component in your controller...

Remembering my control theory courses at uni, I think than an integral component could make your system less stable.

If we examine the behaviour of the system in open-loop:

The input to the 'plant' is the duty-cycle of the solenoid.

The output from the 'plant' is the air pressure in the intercooler (=MAP when WOT).


If we want to control the plant in order to maintain a target level at its output, then an invaluable exercise is to model the open-loop behaviour.

Now, forming an accurate model of this complex system would be a bit of a nightmare, but...

We can imagine that if the duty-cycle is low, and the output pressure is low (turbo is spinning slowly), then we immediately change the duty-cycle to high, then the subsequent rise in output pressure will lag behind the change in the duty-cycle. This is called the step-response of the plant, and indicates that the plant is predominantly laggy (due to the inertia of the turbo).


If the plant is controlled using a PI controller, then (imagining the system starting out with low output pressure)...

The error signal (desired boost - actual boost) will be high, and will increase the duty-cycle a bit (proportional gain).

As time proceeds, the integral component of the controller will ramp-up (the area under the error curve), and increase the duty-cycle more and more (probably saturate the duty-cycle quite quickly).

.
.
.

Now, the output boost is rapidly approaching the target level, and the Proportional component is tailing off. But the Integral is still well 'charged-up', and drives the output pressure right through the target level.

Now, the error signal changes sign, and the Proportional component tries to reduce the duty-cycle, but is insignificant compared to the large 'wound-up' Integral component.

.
.
.

The output boost has swung well above the target for some time now, and the Integral component has ramped back down, so the duty-cycle has dropped back. The turbo starts to spin down.

.
.
.

Again, because of both the plant and the controller being laggy, the output swings back straight through the target, and overshoots. We basically have an oscillator.


Having said that... An Integral component can be very beneficial in any control system, because it forces the average error to be zero. You just need to make sure it doesn't cause the above instability.


Using a PD controller alone would be much more stable, because as the laggy turbo is increasing the output pressure towards the target, then the Derivative component would help to stop overshoot by dropping the duty-cycle based on the rate at which the output pressure is rising (it will dampen a sharply rising output).

The downside to PD, is that at roughly steady state, the controller would be fairly limp about correcting any small error in the output pressure.


My thoughts lead towards a predominantly PD controller, with a small I component (to force the average error to zero). The instability problems associated with the Integral component can usually be helped greatly by limiting the amount that it can 'wind-up' (limit the range of the variable used to represent I in your code), and keep the I gain reasonably low.


Aside from all this stuff... I myself did a small control system project once using a microcontroller. My biggest tip would probably be to always think carefully about the resolution of the internal calculations... Ensuring that you cannot get overflows, or times when rounding errors will make a low-gain component of the control system do nothing.

Work some example error (pressure - target) scenarios through the code on paper, to check that the control paths are actually within their dynamic-range.

It is probably quite easy to put lots of effort into designing the perfect controller, but to find that in practice it is behaving like a simple 'bang-bang' controller (if a PI controller is used on a laggy plant, with much I gain, and no wind-up protection, then it may be worse than 'bang-bang' control).


I hope this can be of some use towards your goal m8.

PS. Have you considered attaching another pressure sensor (used as your feedback signal rather than the MAP) to the turbo side of the throttle? This could help you steer clear of surge. I guess they are quite expensive though.

john banks

Paul I was using a valve, but it could be done with restrictors.

James - thanks very much. I was going to use PD, but then everyone put me off so I did PI. In my PI code I had antiwindup protection.

I have just got back from my first run with the EBC on the car and I am pleased to say it works, and may even work quite well when properly set up.

With just a duty cycle and no error correction it does overshoot a bit (this is effectively the mod I described at the beginning of the thread).

When I up the proportional gain too much it oscillates, I think there is a sweet spot which I will need to find.

I may yet put D in and see what it does for me.

However, in three days (I took two days off work to do this!) I have got a working system that seems to work quite well. Just the refinements to add now

My first microcontroller project - could have picked something simpler like flashing lights, but hey it works!

john banks

James I think you are right. I need some differential control here. Whilst I can dial out a lot of the spike by using a suitable size of restrictor and a reasonable baseline duty cycle, adding proportional gain only makes it worse if you try and hit target.

If I have D , I suspect there will be less need for I as I can use appropriate proportional gain to dial out offsets.

Whilst each variable added of PID seems to improve the curve I don't want to have three of them for simplifying setup.

Any more ideas welcome all.

Oh also hit fuel cut once when it was very spiky. It was over 19PSI (PPP ECU). I am no longer a fuel cut virgin although that is not a good thing

James_PowerMad

John m8,

Is there any convenient way for you to trace the variables in your system when it is actually going, or get it to store some values to be downloaded later?

I ask this because it would be interesting to know how noisy the pressure signal from the MAP is after sampling.

I guess you may have been put off of Derivative control because of its suseptability (can't spell long words - sorry) to noise. If you know that the feedback signal (MAP) after sampling changes nice and smoothly, then D will be a winner for improving stability.

BTW. What approx sample rate are you using? Or are you just running the control loop flat-out, without fixing the sample rate?

Oh, yes... I seem to remember another thread from when you were initially thinking up this EBC stuff... I think you said something about how you were setting the PWM duty-cycle, and that the calculated control value was used to increment/decrement the duty-cycle??? If so, then that is another lag in the system.

As you can probably tell, this project of yours combines rather a lot of my interests (Electronics, Programming, control systems, and modifying sexy Scoobies)!!! :-)

I will gladly attempt to help if I can.

James.

James_PowerMad

Another thought... Although I said that you want a smoothly changing pressure feedback signal to make good D control...

You still need to be able to get a good workable D value when the pressure is shootin up.

What I mean is, if you were sampling very quickly compared to how the nice smooth pressure curve is changing, then the classical D estimation of X(n) - X(n-1) will barely give any output...

Say pressure (X(n)) was doing...

10, 10, 10, 11, 11, 12, 12...

Then the D would be pretty useless.

In which case you may need to estimate D as something like:

X(n) - X(n-10)

Or some other scheme.

May the torque be with you. :-)

James.

john banks

The device does have 256 bytes of EEPROM, which I have not yet used at all. I am using about 1/3 of the 4K available Flash program space so far, but about 2/3 of what I can fill with the free (but very good otherwise) demo compiler I have.

The chip runs at 8MIPS (RISC @ 8MHz) but I am only sampling at 15.3 Hz which I have locked to the PWM cycle. The chip is sitting in a do-loop probably >>99% of the time waiting for the PWM timer interrupt.

I will try D and see how it goes.

What you could help with in particular right now is this: I am using a static duty cycle (about 60-70% with my present restrictor) and adding the integer value which is the output from my control loop. I did this so that when there is no error it would sit at the static duty cycle to avoid oscillations. Otherwise when there was no error the duty cycle would come to zero which would not maintain anything like target boost. Is there another (simple - remember I am GP!) way to do it?

john banks

From attaching my multimeter which measures at 3Hz the output is smooth with no more than 10mV wandering. I hope 15Hz is also.

The output from TPS is silky smooth - you can adjust 10mV (out of 4V range) on your multimeter with a tiny twitch of my size 9.

john banks

SUDDEN FLASH OF INSPIRATION ALERT: (I don't have these often honest) just looking at some example PID code -

I could add the current PD integer output to the PREVIOUS duty cycle rather than a FIXED duty cycle and let the process find its own duty cycle. Part throttle could be modified by altering boost targets rather than worrying about duty cycles.

I already have overflow protection on my duty cycles so that is not a concern.

Then I would not need to worry about calibrating duty cycles at all and could remove one variable, leaving me with only P and D. I am sure this is the way Mr PID intended I just fluffed it up.

James_PowerMad

John,

It sounds like you are doing the right thing by adding your 60-70% offset... This is effectively the zero line of your controller output.

You mentioned that you tried proportional control, and it was a bit unstable... It would be really good to be able to measure what duty-cycle is being set by your controller while this is going on.

Is there any way you can hook some LED's to a spare port on the processor to use as a bargraph (which you can use to indicate the approx duty-cycle signal level).

Or attach an RC lowpass filter, and needle voltmeter to display the average voltage of the PWM.

You may find that the system is 'bang-banging' from max duty-cycle to min as the error goes positive and negative (hence the proportional gain needs reducing, or add some D).

BTW. Are you using floating point calculations on your controller, or is it totally integer?