Fuel injection woes

[joostvdw]

Hi!

I need some help figuring out what's wrong with the fuel injection on my car. The problem is that recently it's been running very rich (about 20L/100km with an 1800cc turbo engine) and it has a very inconsistent power output. Varying from scary fast (the acceleration surprises me even after 6 years) to "where-did-the-turbo-go" slow.

The problem is that the ECU cannot be read at the dealer, it only has a self diagnosis system, which indicates no trouble. It's a quite basic system, with a MAF, TPS, no lambda sensor and batch fired injectors. Strangely enough it is self learning which means it adjusts the mapping according to my driving. Everytime I unplug the battery for a while it runs better until it "learns" it's incorrect settings again and it starts running worse and worse.

The coolant sensor is still within range from the manual (resistance measured) and the injectors were dipped in the ultrasonic cleaner, but not flow checked. I also swapped MAFs and checked fuel pressure (both fine). Plugs, HT leads and distributor cap as well as other maintenance items are good.

I've tried everything I could think of and I'm starting to believe that the only way to fix this is going to standalone management (megasquirt for instance)

Oh almost forgot! It's a 1987 Subaru XT turbo, with the EA82T engine. Parts are almost impossible to find and the spare parts I have are not interchangeable (strangely enough).

[vitesseefi]

So, if this thing is self learning, it must some kind of feedback device - I guess an O2sensor? You say no sensor - but I cannot see how it can possibly do any meaningful self learning without?

If the sensor is sick it will be telling the ECU lies and adjustments will be made on the basis of wrong information. An exhaust leak upstream of the sensor could have a similar effect.

The other thing that occurs to me is that it may be nothing directly to do the FI, but a problem with the boost control/wastegate?

Cheers

Nick

[joostvdw]

It definately does not have an O2/lambda sensor, I once fitted one, but came to the conclusion that the wiring loom in the car doesn't even have a wire for it! And even after running a new wire the ECU didn't have the pin in the connector! Apparently only models equipped with a catalytic converter had an O2 sensor to keep the cat from clogging or burning up and since mine had none from stock, no sensor.

The only other thing it can learn from is the TPS sensor, looking at how agressively you punch the throttle I guess..

You are thinking in the right direction with the boost though, since I do have boost problems, but it's more an overboosting issue (boost creep), it always hits (and overshoots) the stock 0.7bar pressure.

[scaryoldcortina]

Are you sure it's self learning? '87 seems a bit early for that (as well as no lambda=no closed loop). Do you have a spec sheet that states it is, or are you assuming it does because a reset "cures" it for a while?

I'm guessing it could be a faulty ECU, how long does it take to go bad again after a reset, and how long are you unplugging it for?

[scruff]

What is the ignition - electronic or dizzy?
I had a dizzy mech advance stick on my Porker, took ages to find as it would periodically unstick and behave for a bit.

That would run OK at idle but then chug like a tractor as the revs rose, then it would unstick and run fine for a bit then drop dead at the next set of lights as the dizzy was stuck at max advance.

[joostvdw]

I have no spec sheet stating that it is self learning, but a lot of people think it is, and after some highway driving it always seems "slower" than after a quick blast around town, but of course, we all know the accuracy of the butt dyno... When I unplug it it's for atleast a full night and the ECU learns within hours how to misbehave again.

I just realised, the car has an immobiliser which is somehow connected to the ECU since the check engine light does not light when the immobiliser is active. Could this be effecting the ECU? Or perhaps over time (immobiliser fitted in '96 I think) it messed with some settings?

Scruff, the ignition is fully electronic, there is a slotted wheel inside the distributor with 2 "tracks". One track has 360 slots and the other 4, so this sensor works as a crank and camshaft position sensor. On top there is a traditional distributor cap with rotor and plug wires.

I have determined that it's probably a bosch LH-jetronic derived unit, so perhaps this helps some way?

[scaryoldcortina]

I think most modern FI can be considered to be LH jetronic derived Doesn't help too much with the system capabilities though, ECUs vary widely in spec. I still think yours is fairly basic because of the lack of a lambda sensor. I did wonder if you have a knock sensor though (seeing as it it a turbo) these can wreak absolute havoc when faulty (or loose!) because the engine thinks it's detonating and reacts to preserve itself...

I'm guessing your immobilizer interupts the ecu live (often shared with the fuel pump) which is why the check engine light doen't light with the immobilizer on.

[joostvdw]

Oh yes it has a knock sensor! I've unplugged it before but it didn't have any effect. It also should give a check engine light when it detects a knock which it never does.

Can I check this sensor in some way?

[joostvdw]

I had a quick peak inside the ECU but couldn't find anything obvious, besides from the empty socket (?) and reprogrammable memory bank...

[scaryoldcortina]

Knock sensor testing... Good description here

The IC with the sticker on is an EPROM - although it is programmable memory, it isn't programmable in circuit. It most likely contains the load mapping parameters. I'd guess the extra socket is for another one for applications where 256k isn't enough memory.

[joostvdw]

Thanks for the link. I had a look and should be able to do something with it. I can see the ignition retard/advance with a timing light and hitting something with a hammer is never a problem

[ratmangta]

Check for air leaks to the MAP sensor and the sensor itself . I very much doubt it is self learning .It would need AFR input at least .

[joostvdw]

I had a little tinkering time today and here are the results: there are no air leaks around the MAF and the ECU does not respond to knocking(!). I used a piece of pipe and a hammer to knock on the engine while it was running (10cm from the knock sensor) and the timing didn't budge.

Could it be that the ECU retards timing at random because the knock sensor gives faulty readings? How do I check or repair this?

[scaryoldcortina]

How expensive/difficult is it to get a new knock sensor? Test by substitution is the usual way unless you have a lot of diagnostic gear..

[joostvdw]

New from subaru is most probably very expensive, second hand is do-able by asking someone in the states to go junkyarding for me, but, I don't like calling in favors from people I don't know... Does it matter what type of knock sensor I put in? A quick ebay trawl showed me that most cars have sensors with a bolt through it and some (hondas for instance) have a screw in sensor like my subaru. Can I just grab a sensor from an old honda in the junkyard here? I guess it could never be worse than what's in there now...

And how about the high fuel consumption? Replacing the knock sensor won't fix this right?

[vitesseefi]

If the knock sensor is causing the ECU to go to minimum timing settings or limp-home mode it might make the car quite thirsty. Would have thought that limp-home mode (if it has such a thing) would show an engine check light/error code though?

Suspect you may have more than one thing going on here - still wonder about the wastegate/boost control.......

Nick

[joostvdw]

Wastegate/boost control is very simple, there is an actuator with a small tube going to the pressure side and that's it. The ECU does not control the boost pressure. I have a boostgauge (analog) fitted so that I can see that it boosts like it should. The only boost related "problem" I have is that it suffers from boost-creep/overboosting and consequently fuel-cut, but that's only if I keep my foot in it too long;)

[joostvdw]

A rummage through my parts stash yielded me a spare ECU, MAF, CTS, 2x TPS, distributor and most importantly, a knock sensor! Sadly though, I know for a fact that this ECU is from another year and perhaps model car, so it doesn't even turn over when fitted I do have 3 different factory service manuals which should help me getting around this, but none of them is exactly my model and year, so that won't be easy..

Can I test the knock sensor before fitting it? What sort of signal should it give when hit with a hammer?

[scaryoldcortina]

This is just a guess, but the knock sensor is basically just a piezo crystal (a microphone). It is going to be putting out a tiny AC signal (couple of mV tops) so you won't see anything meaningful with a multimeter. I would try attaching the sensor to an amplifier (like a guitar amp, or the "mic" input of a tape deck) and listening to see if your hammer taps get amplified.

[vitesseefi]

Simpler way might (I have read of this but not tried it) be to connect a crystal earpiece such as this
www.maplin.co.uk/crystal-earpiece-3256?c=froogle&u=3256&t=module
which is effectively the same thing in reverse. you connect the knock sensor to the earpiece, stick the earpiece in your ear and clout the block near the sensor with a small hammer, theory is you get a noise in the earpiece...... maybe.

Little bit here too.... www.explorerforum.com/forums/showthread.php?t=162549

Nick