When you click on links to various merchants on this site and make a purchase, this can result in this site earning a commission. Affiliate programs and affiliations include, but are not limited to, the eBay Partner Network.
Had a little dyno-session with my FC yesterday. Wasn't able to tune the car completely due to a slipping clutch, but when I was on the dyno I found that it was extremely difficult to detect engine knock.
So, what are other people using for knock detection, and what are your experiences with it?
Some details about my car:
- S4 TII rotor housings (So no mounts for a knock sensor)
- S4 center iron and late S5 front and rear iron, stock ports
- S4 9.4:1 rotors, custom LIM, RX-8 4-port UIM
- 98 octane fuel (in europe, so comparable to 93 fuel in the USA), no AI yet.
- SXE362 turbo, 76mm turbine, 0.91A/R, twin tial wastegates set to 0,5 bar
- V-mount IC, dual 3" exhausts, surge tank with 044 pump 4x ID1000 injectors, IGN-1A coils
- MS3X ecu
So I'm currently using 3 bosch knock sensors, part number 0261231046, these ones do not have an internal bandpass filter.
Since I'm using S4 rotor housings I have one sensor in the oem location (On the middle iron), and the other sensors are on the mounting bosses on top of the rotor housings (Where the rats nest usually mount)
2 of these sensors are for the MS3X, I have the knock board installed and have the following settings in the ECU: (The .msq is also attached)
The 3rd sensor is for a phormula knock analyser pro with headphones.(Phormula - Advanced Engine Knock Detection)
Through the headphones I can hear more engine noise at higher loads, which is normal, but anywhere over 80kpa there's like a thumping / clipping sound.
I don't think it's actual knock (Dialing back ignition doesn't affect it and if it was knocking the engine would probably be dead by now).
These are some of the datalogs:
Rolling on throttle, dyno set to a steady 2500rpm
Rolling on throttle, dyno set to a steady 3000rpm
Rolling on throttle, dyno set to 3500rpm but clutch started slipping at around 380Nm
Current ignition table, I dialed back the timing a lot because of the knock readings, if I dial it back much further EGT's start getting toasty under boost.
So, not sure what to try next. Maybe just dial down the gain levels in the megasquirt and keep an eye on them.
Any thoughts?
I would redirect this thread to the megasquirt specific subforum. You asked a calibration question with an answer specific to your ecu in a subforum that's for discussions centered around single turbos.
Although detecting engine knock is not megasquirt specific the engine managment forum is probably a better location, I wasn't able to redirect the thread myself so I've asked a moderator to do it.
Anyway, I did find this nice nugget of information that describes how the knock sensor amplifier actually works and how to set it up: :
Starting at the knock sensor, the signal is fed into the ECU and to the periphery of the knock processor, where it goes through a fixed analogue gain stage. The purpose of the analogue gain stage is to present a signal that is neither too small nor too large for the following stages; if it is too small then the signal to noise ratio will be poor whereas if it is too large then it will get clipped (which ads considerable spectral energy at higher frequencies and could "drown out" the signal we are interested in).
Next it goes into an anti aliasing filter to remove spectral energy above the nyquist frequency of following analogue to digital converter. There are no settings for this stage, it is controlled by the system clock speed and you don't need to worry about it, it will be right for the system sampling rate.
The quantised signal then goes through a programmable gain stage, this is what you're programming with the gain setting; the aim here is to keep the signal to noise ratio sensible for the next stage. It allows the system to be matched to the engine without needing to change components in the analogue gain stage.
Once through the digital gain stage the signal goes through a programmable bandpass filter, this is what you're programming with the frequency setting. Engine knock is normally an inverse chirp signal, it starts at a high frequency and drops in both frequency and energy with time, so the centre of the bandpass filter should be set so that its passband permits as much spectral energy from the chirp through as possible. Sampling knock and running it through a spectrum analyser will give you a good idea of what your engine does when it knocks.
Once filtered, the signal is rectified. There is nothing for you to adjust on this stage, it just does what it does.
The final stage is the digital integrator. "Physically" it is controlled by an integrate input and the integrator time constant. "Logically" we want to listen for knock when it is likely to happen, and to do this we define a window of crank angles within which we realistically expect knock to occur, there's no point in listening to knock during the exhaust stroke! The integrate input is asserted by the timing logic when the crank is at the start angle; this causes the integrator to reset and to start integrating. The integrate input is deasserted by the timing logic when the crank is at the end angle; this causes the integrator to cease integrating and to hold the result. The rate at which the output of the integrator rises for a given input is determined by the integrator time constant, for simplicity just think of it as the time that it would take the integrator to swing from zero to full scale if a full scale input signal was present; the lower the number the quicker it rises, it's a bit like a gain but it's used for fine tuning time. If it's too low we can end up saturating the integrator output before the window has finished, if it's too high then we could find that even with heavy knock we still don't get a particularly high output from the knock processor.
Once the timing logic deasserts the integrate input, the knock processor's output is read by the knock control strategy to update the knockBase for the cylinder that just fired. For simplicity you can consider the knockBase to be a running average value for a given cylinder. We do knock discrimination on the premise that the knockBase is an ambient noise level for that cylinder in the absence of knock. The current knock value is then compared to the acceptable values in the Event Detection Level and Severe Event Detection Level tables; if it exceeds those values then a knock event is generated / deemed to have occurred and something must be done about it. This is very important because the premise is that knockBase represents background noise, and must not become contaminated with knock signal energy; if that happens then the discrimination will not work.
In terms of getting the right numbers in there, start with something in the region of a gain of 0.5, bandpass frequency around 6kHz, integrator time constant around 200, start angle around 0 degrees, end angle around 30 degrees. Turn on the knock detection at idle and see what sort of figures you're getting... they should be less than 10 but not 1 or 2, between 4 and 6 is a good starting point; you may need to play about with the gain to get the figures into this range. Do some runs with conservative timing and det cans to make sure there is no knock, leave knock control off but keep detection on and datalog the runs. Ideally you don't want to see more than 35 on the run. You can fine tune the individual cylinder trims at this point too so they all read the same. You may need to adjust the gain and/or time constant to get the readings in the right ballpark. Log the knock for each cylinder at a high sample rate (100Hz should be fine), so you can use SView to work out where you are / need to be with the event detection level settings... have a look at each knock trace in turn and also use a math function to get an average [you can use runavg(channel, samples) or filter(channel, filter_constant)]... then you can use another math function to subtract the average from the actual value to see what sort of "spikes" are normal, SView will automagically highlight the maximum present in your present view of the log. Once you have your acceptable "spike" thresholds for a clean run you can populate the event detection levels, allowing a little more than the "spikes" of course The default values are normally around 1.5x knockBase but you may find that's either too sensitive or not sensitive enough for your engine. Once you have event detection levels configured you can turn on knock control and test it with the cans on; it shouldn't be pulling timing as it is, if it is then you need to re-check your levels. Once it is OK you can start adding timing, you should see it pull timing before you can actually hear it, that is normal because if configured correctly it should be more sensitive than your hearing.
Reading the above article maybe my settings are too sensitive, I'll try adjusting the integrator constant and the gain so knock levels at idle are down to 4-6 and do some more testing next time on the dyno.
my notes show that the block resonates at 3500 cps. you are set at 5500. it does, however, look like you are getting proper data.
the peak #s generated at 2500 (48) may be serious "knock." ... or not. it all depends on your settings which will become better defined as you use the car..
re your 2500 pull, it appears to me that your timing (8) might benefit by being advanced a bit... see what happens by adding 2 or 3 degrees of advance.
what is your split at 150 MAP? what plugs? how do they look?
you have injector capacity for 350 rw rotary hp and a 550 rw rotary hp turbo... dangerous. what size are your wastegates?.
"350 rotary rwhp requires 2763 CC/Min (gas) into the motor after 85% Injector Duty Cycle and 20% Lag. In order to make up for 85% Max Duty Cycle and 20% Lag you need 4061 GROSS/Nominal Injector Capacity."
ideally you want an ACTIVE knock system.. i e it cuts timing at a certain knock threshold. in order for this to work properly the ECU should cancel the timing cut if the TPS is below a certain threshold. that way you sidestep false cuts.
good for you for instrumenting your setup. don't forget AI.
Last edited by Howard Coleman; 03-01-20 at 09:28 AM.
I've adjusted the frequency to 3500cps, the signal looks much cleaner without all the spikes. I think the 5500cps setting was wrong and it was picking up a lot of noise.
I'm also going to try to add a 3500cps bandfilter to my phormula headphone sensor to hopefully hear knock better next time I'm on the dyno.
I agree that my ignition map is very retarded and can use some more advance, will do so when I'm more confident that I can detect knock.
Right now I'm running about 5 degrees of split at 100kpa, tapering to 15 degrees at 170kpa, so at 150 it's about 12.
I'm using BUR9EQ's all around, I'll replace them and have a look tomorrow.
The exhaust manifold are 2 tapered runners into a twin scroll turbine housing. Each exhaust runner has a tial F38 wastegate on it. Wastegate exhaust is not re-routed to the main exhaust but dumped through a straight-through open muffler, so controlling boost shouldn't be a problem.
I agree the the fueling is insufficiënt to max out the turbo. Will add another pair of 1000cc injectors and water/meth injection later. I am running the injectors at a base pressure of 60psi.
In the meantime I'll try to keep the boost sane, and set up the overboost fuel cut and AFR / fuel pressure safety systems accordingly.
9 heat range plugs have no place on a single turbo app. they exacerbate spark plug mountain which is the primary cause of chatter. chatter is ultimately death to rotor housings.
twin 38 WGs generally are insufficient on an SX-E unless the manifold is biased to the WG at the expense of driving the turbine wheel.
9 heat range plugs have no place on a single turbo app. they exacerbate spark plug mountain which is the primary cause of chatter. chatter is ultimately death to rotor housings.
twin 38 WGs generally are insufficient on an SX-E unless the manifold is biased to the WG at the expense of driving the turbine wheel.
This looks like your personal opinion, unsupported by fact. Can you support your opinion?
I like the BUR9EQ's to get the car up and running on low boost and get the majority of the tuning done as they don't foul as easily. Will swap them out R7420's when the boost gets upped.
Will see if the wastegates hold up, this is a street car, so the main turbo exhaust has a lot of mufflers, (3 resonators and one muffler actually, it is as quiet as stock), and the wastegates are pretty much dumped open to atmosphere, which should help controlling boost.
It's also a stock port engine with a longer-than-stock intake manifold, so low VE at higher rpm's -> Less energy driving the turbine.
Yes, of course you can set the megasquirt up to retard ignition when it detects engine knock. Retarding ignition is the easy part, reliably detecting knock in all situations is not as easy.
Last edited by John Huijben; 03-06-20 at 07:31 AM.
I like the BUR9EQ's to get the car up and running on low boost and get the majority of the tuning done as they don't foul as easily. Will swap them out R7420's when the boost gets upped.
Will see if the wastegates hold up, this is a street car, so the main turbo exhaust has a lot of mufflers, (3 resonators and one muffler actually, it is as quiet as stock), and the wastegates are pretty much dumped open to atmosphere, which should help controlling boost.
It's also a stock port engine with a longer-than-stock intake manifold, so low VE at higher rpm's -> Less energy driving the turbine.
Yes, of course you can set the megasquirt up to retard ignition when it detects engine knock. Retarding ignition is the easy part, reliably detecting knock in all situations is not as easy.
I have found that fouling R7420-11 is extremely hard, however I do run a CDI box. I can't say if two 38mm wastegates are insuficient, what I do know is that if the turbo manifold was design in favor of spool, and not wastegate placement, you WILL see boost creep, even if dumping the wastegate gasses to atmosphere.
Update, I've upgraded the clutch, tuned the engine at 0.5 bar of boost and put around 2000km's on it so far.
So sign of boost creep, it's perfectly stable all the way to redline. Got the knock settings dialed in also, seems to be working well.
The SXE362 is spooling way quicker than I expected which is nice, full boost around 2250rpm.
For what it's worth, I think knock detection on a modded Rx-7 is a guessing game at best and in many cases a lost cause. I have worked in knock system development for a couple OEM's on turbo and nonturbo piston engines. To detect knock, we would literally build an engine just for the purpose. On an engine dyno we'd advance the spark until it knocks so hard you can hear it clearly outside the dyno room with the naked ear. We'd record the knock sensor signals and the cylinder pressure knock calculations to determine it.
For a single channel system (single sensor) it needs to be placed where it can pick up the vibration. So we'd stick lab-grade accelerometers all over the block to see which locations actually detect the knock, so that we know we put the sensor in the right place. Then we'd have a test vehicle drive the car around in the desert with the engine heatsoaked to confirm it was still picking up knock.
There's really no way to get good knock detection without actually intentionally knocking the engine. You could blow a motor doing that. All this stuff about background noise algorithms etc... that stuff has to be set after the knocking thresholds. And you can't set the thresholds unless you without a doubt make it knock.
mostly agree with you but my FD knock experience dating back to 2003 has been very helpful.
of course the Power FC doesn't have an active system but i learned by watching it carefully i was able to get a good read on timing and other potential problems.
my ViPEC V 88 does have an active feature and i run two Bosch/Mazda RX8 newer style sensors. of course w the BREW Mazda has done all the homework as to location.
i had a situation a few years ago where i had less than optimal gas and had done a power run. looking at the log immediately after, i noticed a knock reading of 100, right where knock should be, at peak torque... normally my knock readings are 10 to 20 around peak torque. the scary thing about it was i had no indication from the driver's seat. i dropped the timing about 4 degrees and did a second run . that run was an exact copy of the first run, AFR and Boost both matched the first run to the decimal point. knock was 10--20. if i hadn't had the knock data the winter gas could have broken my motor at a higher boost level.
the second benefit of having knock data is that it is fundamental as to construction of the timing map. many factors that could be specific to particular car's system influence timing. pasting in someone's map is like wearing someone else's clothes. knock data shows the way. a proper system generates gradations. you can sneak up on knock.
the highest knock reading i have seen since 2012 with my V88 has been 350. it was evident. at that time i was running my AI through fuel injectors. the 20% water component in the water meth shut down the fuel injectors in less than 5 days and with it my AI delivery.. i was very lucky my motor escaped harm as i was running around 2000 CC AI. so 350 is big knock for me.
just like any tool, a knock system's value depends on how it is used.
i did have one other incident where the active feature stepped in... i was adjusting my boost duty cycle and stepped on the right pedal at an RPM and gear where i had 100% duty. the log shows i gained almost 10 psi in less than a second, i had a knock event and thought i had hit the kill switch on my motor. nope... my timing had gone to zero. the knock reacted and all was well.
with the advent of way better ECUs many of us have active knock systems. my bet is few are set to be active. for those who haven't gotten to know their knock system i recommend you introduce yourself. when you are setting it up to be active i recommend you qualify it to be active only when your TPS is somewhere around 40%. this may help eliminate false driveline related readings.
Last edited by Howard Coleman; 05-14-20 at 09:00 AM.
mostly agree with you but my FD knock experience dating back to 2003 has been very helpful.
of course the Power FC doesn't have an active system but i learned by watching it carefully i was able to get a good read on timing and other potential problems.
my ViPEC V 88 does have an active feature and i run two Bosch/Mazda RX8 newer style sensors. of course w the BREW Mazda has done all the homework as to location.
i had a situation a few years ago where i had less than optimal gas and had done a power run. looking at the log immediately after, i noticed a knock reading of 100, right where knock should be, at peak torque... normally my knock readings are 10 to 20 around peak torque. the scary thing about it was i had no indication from the driver's seat. i dropped the timing about 4 degrees and did a second run . that run was an exact copy of the first run, AFR and Boost both matched the first run to the decimal point. knock was 10--20. if i hadn't had the knock data the winter gas could have broken my motor at a higher boost level.
the second benefit of having knock data is that it is fundamental as to construction of the timing map. many factors that could be specific to particular car's system influence timing. pasting in someone's map is like wearing someone else's clothes. knock data shows the way. a proper system generates gradations. you can sneak up on knock.
the highest knock reading i have seen since 2012 with my V88 has been 350. it was evident. at that time i was running my AI through fuel injectors. the 20% water component in the water meth shut down the fuel injectors in less than 5 days and with it my AI delivery.. i was very lucky my motor escaped harm as i was running around 2000 CC AI. so 350 is big knock for me.
just like any tool, a knock system's value depends on how it is used.
i did have one other incident where the active feature stepped in... i was adjusting my boost duty cycle and stepped on the right pedal at an RPM and gear where i had 100% duty. the log shows i gained almost 10 psi in less than a second, i had a knock event and thought i had hit the kill switch on my motor. nope... my timing had gone to zero. the knock reacted and all was well.
with the advent of way better ECUs many of us have active knock systems. my bet is few are set to be active. for those who haven't gotten to know their knock system i recommend you introduce yourself. when you are setting it up to be active i recommend you qualify it to be active only when your TPS is somewhere around 40%. this may help eliminate false driveline related readings.
Hi Howard,
I am setting up a Link G4+ Thunder with rx8 knock sensors at the moment. Would you be willing to share your map??, in particular the knock setup you have done on your vipec. I have never setup knock detection before and any pointers would be greatly appreciated.
02-27-20, 02:26 PM
The default values are normally around 1.5x knockBase but you may find that's either too sensitive or not sensitive enough for your engine. Once you have event detection levels configured you can turn on knock control and test it with the cans on; it shouldn't be pulling timing as it is, if it is then you need to re-check your levels. Once it is OK you can start adding timing, you should see it pull timing before you can actually hear it, that is normal because if configured correctly it should be more sensitive than your hearing.
