HOW-TO: replace knock sensor on EJ25 turbo (& maybe others)

[Yowie]

INTRODUCTION

During my recent dyno tune, with a slow/hard dyno ramp rate the knock sensor threw the toys out of the cot and pulled 11.5 degrees (max) timing. Among other things, this set off my 100 degrees coolant temp alarm as a load of still-burning mixture exited the heads via the poor brand-new exhaust valves. Not great.

The tuner said that Subaru knock sensors 200,000km+ vehicles are known for their unreliability. The problem can be hidden in turbo cars, but is very obvious in naturally aspirated 4 cylinder cars with limited power to begin with.

This description is consistent with my experience driving both my car (modified 2008 Forester turbo) and Mrs Yowie's car (relatively stock 2009 Forester turbo). Any medium-grade long uphill section in either car required a shift from 4th to 3rd gear or the car would "spit the dummy".

Various diagnostic tools would also allow the detection of excessive amounts of knock-sensor activity, which could be assumed to be false positives if the sensor is old, the fuel is high quality and the driving conditions are not too tough.

This is a quick guide to knock sensor replacement at home.


KNOCK SENSOR

This little beasty (new at top, old at bottom of images):





Gregory's Service and Repair Manual (Impreza, WRX, STI 2002-2014) describes this as:

"a piezoelectric crystal that oscillates in proportion to engine vibration which produces a voltage output that is monitored by the PCM. This retards the ignition timing when the oscillation exceeds a certain threshold"

On my understanding, the manufacture has figured out that knock (the sharp explosion of fuel in the cylinder rather than the desired rapid burn) corresponds with certain sound frequencies. The knock sensor is calibrated to generate a voltage when the crystal/s vibrate to those frequencies. Sometimes the knock sensor is overwhelmed by noise/vibration at "full power" so the car will be tuned (from factory or otherwise) to ignore the knock sensor after (eg) certain high revs and just rely on ignition retard built into that performance map.

A tuner with a microphone on the block and a set of ear phones is obviously going to do a much better job of actually hearing knock (and its precursors), but for actually driving around the factory knock sensor is not too bad - while it works.

The pics below are from the relatively stock Forester.


TOOLS

Usual home garage tools, but specifically:

- 10mm and 12mm sockets
- long socket extension bar
- torque wrench suitable for 24Nm
- radiator hose remover (screwdriver-type bent spike for removing coolant hoses)
- as much illumination as you can muster (incl. compact light sources)
- magnetic dropped-parts getter
- fender guards
- the usual tools for removing the intercooler & small coolant hoses


KNOCK SENSOR LOCATION

On the EJ25 turbo (2008ish with the cross-flow intercooler and the non-centre throttle body) the knock sensor is bolted to the top of the block, at the back in the vicinity of cylinder 4 - under the inside edge of the throttle body.

It is highly likely that other Subaru engines have the knock sensor in the same location, but with different obstructions in the way.

The first picture shows it amongst the clutter (bolt marked with red paint, grey plug end at right of image):




The second picture shows the knock sensor after some obstructions have been removed and some light cleaning of the muck & leaves that accumulated on top of the block (bolt and plug marked red):




CLEARING THE WAY

First, remove the engine cover and intercooler.

Second, clear any minor/annoying obstructions or decide to work around them. I removed the blow-off valve ("BoV") and its small vacuum hose. I also unbolted from the manifold (but did not remove) the large diameter BoV air-return hose.

Third, remove the throttle body. The knock sensor might be removable with the throttle body in the way, but it will be impossible to apply the correct torque on re-installation.

(a) use a paint pen or similar to mark how the throttle body hose is "clocked" to the throttle body. This will save the pain-in-backside factor on re-installation;

(b) remove the throttle body hose;

(c) disconnect the electrical plug from the throttle body

(d) undo the two small coolant lines (at the non-throttle body ends) at the metal spigots on/near the top of the gearbox. Use plyers on the constant-tension clamps. a radiator hose remover is handy to break the seal. You will lose a little bit of coolant so have a rag handy.

(e) undo the four 10mm-headed long bolts that hold the throttle body to the manifold.

(f) remove the throttle body from the engine bay.


If you haven't already, consider doing the "throttle body coolant bypass mod" on reinstallation.

The picture below shows the coolant path that previously went through the throttle body now going straight from source to destination. The longer of the two OEM small coolant hoses shortens up nicely to match the shape you need.




REMOVE OLD KNOCK SENSOR

First, mark the angle of the existing knock sensor relative to the block. Gregory's says:

"Note the location of the sensor (the pigtail harness should be at approximately 60 degree angle to the back of the engine)"

[EDIT - I haven't figured out the 60 degree thing. Knock sensors on this model are installed at the pictured angle, but I can't work out where 60 degrees comes into it. /rant]

Second, unplug the 2-wire electrical plug.

Third, remove the 12mm headed bolt with a long extension and remove the old knock sensor.

Fourth, clean up the ring of crud:




INSTALL NEW KNOCK SENSOR

First, position the new knock sensor and start the bolt (this is fiddly). With hand tools, lightly tighten the bolt and make sure the desired "60 degree" angle doesn't shift as the bolt starts to clamp the sensor.



Second, use a torque wrench set to 24Nm to set final torque on the 12mm headed bolt. Check it hasn't clocked away from 60 degrees.



Third, connect the electrical plug.


You can then re-install the throttle body (incl electrical connection & your chosen coolant line solution), throttle-body hose (both ends!), BoV hose, intercooler, BoV and engine cover.


The job took me the best part of the afternoon, cleaning things as I went and in no particular rush.

It's early days post-installation, but the car seems to not be protesting on medium-grade uphill sections in 4th gear any more.


Thanks for reading.

[RX25SE]

Excellent write up!

Second, use a torque wrench set to 24Nm to set final torque on the 12mm headed bolt. Check it hasn't clocked away from 60 degrees.

This is a very important step.
As piezoelectric crystals are used in other types of pressure sensors as well (such as MAP sensors and electronic scales) they are sensitive to pressure exerted on them, hence correctly torquing the mounting bolt.

[Yowie]

Cheers RX25SE.

Also, presumably it goes without saying to not install it with a rattle-gun.

As a further update - the car is driving very nicely post-swap. "Feels like it has correct ignition advance" is the short version. More responsive, will lug along in 4th (at light load), handles hills without a sweat, etc. Presumably over the medium term it will save fuel too.

[rexhunta]

Great writeup.

Thank god the 2.5i was a lot easier. just the bolt on top, and you can reach it without moving everything. You can reach it with a knuckle adaptor and extension.

[bigBADbenny]

Did you find any cracks in the old knock sensor?
Apparently that’s one fault mode.

I replaced my knock sensor at around 300kkm, coincidentally the engine almost completely ceased to fine knock learn around the same time, but I’d also fixed the cracked primary pcv hoses, which had been in effect for years but gradually got worse to the point that the hose leaked blow-by oil.

I’d have to reinstall the old knock sensor to see if it was the issue or not.
No cracks in the old sensor, but that doesn’t necessarily exclude material breakdown of the sensor itself over time.
What’s interesting is its yet another example of a failure mode that calls attention to itself, generally self protective.

Bear in mind that some of the torque specs in the Factory Service Manual are… optimistic, and using extensions generally decreases torque at the bolt, hopefully one aspect takes care of the other, or chase the threads if in doubt.

Definitely an item to add to the ever growing pre tune health check lists.

[Yowie]

On visual inspection the old knock sensor was discoloured but not damaged. The bolt had a blue tinge (heat?) which was odd. I much try to cut it in half later for sh*t-hey information.

On your description below and my recent experience, I'd suggest that the old knock sensor was the culprit rather than cracked breather plumbing.

The socket extension and the light WD40 probably cancelled each other out If the torque is less-than-spec that would make a less sensitive knock sensor which is fine by me. It's the relatively stock family car (with 98 octane fuel and a tune to match) so it's rarely having its neck wrung.

Interestingly, the tuner mentioned "the old days" (before max retard could be adjusted via tune) when they put an o-ring under the knock sensor. Following that logic, a less-tight bolt would err on the side of less sensitive.

[bigBADbenny]

It sure is a fascinating subject, easily overcome by just following book procedure, as did I.

However there’s always room for experimentation next time around.

I say that because my car still has some Feedback Knock Count (FBKC) but practically no Fine Learning Knock Count (FLKC).

Both are load and RPM dependent, the ECU essentially filters the knock data, but FBKC can be said to favour mechanical knock (I have a noisy throwout bearing and a very tired engine), and FLKC can be said to favour fuel trim knock. Source: https://cobbtuning.atlassian.net/wiki/s ... nitor+List

Inlet tract leaks include pcv system vacuum leaks: unmetered air causing negative fuel trims in closed loop around 5-10%.
That’s the ecu reducing fuel delivery by that percentage (!) due to unmetered air.

Anyhow that’s all fixed meaning the only way to “see” the difference between my newer and ancient knock sensors is to datalog them, also at various torque settings and or try orings, washers and all the other home brew “fixes”. Edit: better still, just hook them up to an oscilloscope and log the outputs in far higher resolution than the oem ecu.

On that note, I can also datalog Knock Sum, a Subaru engine parameter that might be thought of as the unfiltered “root” knock signal. Usually this parameter is recorded as an incrementing, resetting parameter, but with the magic of Megalog Viewer HD and the help of Jeff Huckstead on Guild of Subaru Tuners FB group, I’ve made a formula to turn Knock Sum into a parameter that might be useful for this test, to see if old knock sensors are more sensitive, and to see what differences are encountered with soft mounts and less than factory torque spec.

[rexhunta]

Did you find any cracks in the old knock sensor?
Apparently that’s one fault mode.

my gen 1 threw knock codes, the sensor was split in half on both sides.

[CirioP]

It sure is a fascinating subject, easily overcome by just following book procedure, as did I.

However there’s always room for experimentation next time around.

I say that because my car still has some Feedback Knock Count (FBKC) but practically no Fine Learning Knock Count (FLKC).

Both are load and RPM dependent, the ECU essentially filters the knock data, but FBKC can be said to favour mechanical knock (I have a noisy throwout bearing and a very tired engine), and FLKC can be said to favour fuel trim knock. Source: https://cobbtuning.atlassian.net/wiki/spaces/PRS/pages/97687600/Subaru+Monitor+List

Inlet tract leaks include pcv system vacuum leaks: unmetered air causing negative fuel trims in closed loop around 5-10%.
That’s the ecu reducing fuel delivery by that percentage (!) due to unmetered air.

Anyhow that’s all fixed meaning the only way to “see” the difference between my newer and ancient knock sensors is to datalog them, also at various torque settings and or try orings, washers and all the other home brew “fixes”. Edit: better still, just hook them up to an oscilloscope and log the outputs in far higher resolution than the oem ecu.

On that note, I can also datalog Knock Sum, a Subaru engine parameter that might be thought of as the unfiltered “root” knock signal. Usually this parameter is recorded as an incrementing, resetting parameter, but with the magic of Megalog Viewer HD and the help of Jeff Huckstead on Guild of Subaru Tuners FB group, I’ve made a formula to turn Knock Sum into a parameter that might be useful for this test, to see if old knock sensors are more sensitive, and to see what differences are encountered with soft mounts and less than factory torque spec.

It sounds like you've delved deep into the intricacies of knock detection and tuning. It's impressive that you've been able to navigate this subject by following book procedures. Your experience with FBKC and FLKC sheds light on the nuances of load and RPM dependency, with the ECU acting as a filter for knock data.

Your insights about FBKC favoring mechanical knock and FLKC favoring fuel trim knock are valuable. The information about inlet tract leaks causing vacuum issues in the PCV system is particularly interesting, as it directly impacts fuel trims in closed-loop operation.

It's great to hear that you've resolved these issues. Your suggestion to datalog and experiment with torque settings, or even using an oscilloscope for higher-resolution data, shows a commendable commitment to fine-tuning your setup.

[bigBADbenny]

I absolutely used fsm procedure when I replaced my knock sensor.
No reason to to replace it other than age, and I had a new spare sensor.

I’d only experiment with angle, washers or sensor bolt torque, or aftermarket vs oem knock sensors, if I had already established the procedure for oscilloscope datalogging of that sensor’s output, along with other piezoelectric contact sensors placed eg on block and heads, with the output data merged with obdii logging of relevant parameters, including FLKC, FKC and Knock Sum, on a known good engine.

In essence initially reverse engineering the oem knock management strategy from data, then trying to influence that, as opposed to delving into the ROM file?