Strain Gage Knock Sensor

[casuprock]

The OEM knock sensor is an accelerometer that detects motion in the block after a knock event. This causes many people problems with phantom knock after the stock block is modified or the internals are upgraded. The OEM sensor is designed as an indicator of extreme condition, not as a tuning aid. Additionally, many of the OEM knock sensors are old or have never been replaced. Unreliable data from a knock sensor can range from a tuning headache to engine damage in severe cases.

The acceleration detected by the OEM knock sensor is the result of a transfer function between force and motion in the block. This is sometimes referred to as transmissibility. The compliance (inverse of stiffness, e.g. 1/k) and dynamics (damping coefficient, inertia) of the block are different for forces applied each cylinder. They're also different for each degree of freedom in the system. The block can be modeled as a set of unique stiffness, mass, and damping matrices containing the coefficients of the differential equations of motion (see eigenvalue problem) that describe the block's response from force inputs from each cylinder.

So, in order to use the stock knock sensor to tell if you're really getting knock (and how bad it really is), you have to understand the frequency response of the OEM knock sensor, the frequency response function of the block with respect to each cylinder in each axis, and solve a multi-degree of freedom system of differential equations. Did I mention the damping coefficients change with temperature?

Instead of that nonsense...

You can measure strain (force) in the block directly. Strain is a better metric for determining knock. First, you're not going to pick up motion of the block as noise. Only force inputs to the block causing strain in the axis of a gage is going to excite a response in the sensor. For instance, a location near the crank bearings would be a good place to locate a strain-based knock sensor. It is easier to filter strain data to look for specific force events and their frequency content.

I am using semiconductor strain gages I produce for machining sensors to build a strain gage based knock sensor.

It's not going to be easy.
It does involve designing a sensor from the ground up.
It will take significant testing and determining proper location.
It will be awesome.
It is the Knock Master 5000.

You must be logged in to view this image or video.

 

[bastarddsm]

One issue I've read about in all of this is that normal combustion events look a lot like a detonation even to most all knock sensors. I read a several page paper on "knock detecting" It was by Bosch or GM.

I do know that the dsm ecu has a pretty specific window for it to see knock. It has a a notch filter to filter out noise, but it also only monitors the ks for about 20* before it fires the plug, to help filter noise out.

So your proposing measuring strain in the block to determine knock.. So theoretically a "normal combustion event will produce strain x, while a det event will produce a larger strain y? I would suspect you will have to map strain x to engine rpm and the power output?
I would also think that this would be sensitive to gear changes, or clutch engagement?

Why not look at cylinder pressure instead?

 

[LiquidX]

Take a read through this first.

BigMoose's Ramblings: Knock

 

[jtmcinder]

I don't follow the idea that it's easier to filter the output from a strain-gauge than from a piezo-ceramic accelerometer. Why do you make this claim? I also don't see the issue of single-axis measurement favoring strain-gauges, since a piezo-C accelerometer is also single-axis.

I'm not questioning the basic idea that there may be better ways to detected knock than the OE method. I'm just trying to follow the reasoning behind your alternative approach.

 

[casuprock]

Take a read through this first.

BigMoose's Ramblings: Knock

Great article!

I don't follow the idea that it's easier to filter the output from a strain-gauge than from a piezo-ceramic accelerometer. Why do you make this claim? I also don't see the issue of single-axis measurement favoring strain-gauges, since a piezo-C accelerometer is also single-axis.

I'm not questioning the basic idea that there may be better ways to detected knock than the OE method. I'm just trying to follow the reasoning behind your alternative approach.

Here's the reasoning: Accelerometer detects motion while strain gage detects stretching of the metal in the block. Motion is deterministic, strain is causal. Motion can be motivated by other moving things in the engine (in addition to knock) Since strain antecedent to motion, it is more direct to measure strain and there isn't as much "noise" from other non-knock sources.

Here's the prototype assembled and discussed. We'll get some preliminary testing done soon which will hopefully give measurement results from various locations on the engine. At the very least it will be interesting to see where this sensor picks up the most strain while comparing it to the OEM knock sensor readout.

<iframe width="640" height="360" src="http://www.youtube.com/embed/Y0dkuLsDr7A?feature=player_embedded" frameborder="0" allowfullscreen></iframe>

One issue I've read about in all of this is that normal combustion events look a lot like a detonation even to most all knock sensors. I read a several page paper on "knock detecting" It was by Bosch or GM.

I do know that the dsm ecu has a pretty specific window for it to see knock. It has a a notch filter to filter out noise, but it also only monitors the ks for about 20* before it fires the plug, to help filter noise out.

So your proposing measuring strain in the block to determine knock.. So theoretically a "normal combustion event will produce strain x, while a det event will produce a larger strain y? I would suspect you will have to map strain x to engine rpm and the power output?
I would also think that this would be sensitive to gear changes, or clutch engagement?

Why not look at cylinder pressure instead?

You're right, I would like to look at cylinder pressure, we'll get there. First thing is to make something that might be placed on the block maybe near a crank bearing to read out load. Obviously cylinder pressure and block strain is closely correlated. Most pressure sensors are actually strain sensors on a metal membrane, in our case the block (or head) will play the role of the "membrane". I talked with Thomas Dorris at ECM Tuning and he described (very similar to your description) a programmable gain amplifier and some filtering in analog inside the ECU. I don't know exactly what to expect until we start to take some measurements.

I don't think this will be as sensitive to clutch engagement or gear changes- if it's intelligently located. It will be interesting to see what effect location has on results. We'll see what happens.

 

[jtmcinder]

Motion is deterministic, strain is causal.

What?

 

[casuprock]

What?

Sorry, I'm not trying to be confusing. Acceleration is a result/consequence of a change in motion. Acceleration is also a consequence of force acting on (compressing or stretching) an elastic material. The latter is what we care about. the first is "noise" and interferes with measuring knock.

Basically, we're stripping out a source of mechanical noise by bypassing motion-based sensing and using direct strain measurement.

 

[pos4g63]

Looks cool now it just needs some kind of in car display for testing. You should try it next to the factory knock sensor. Induce some knock according to the factory sensor via timming changes and compare it to the 5000

 

[casuprock]

Looks cool now it just needs some kind of in car display for testing. You should try it next to the factory knock sensor. Induce some knock according to the factory sensor via timming changes and compare it to the 5000

We will play with advancing timing. The test vehicle belongs to kthackst. It has ECMlink and we will log the stock knock sensor during our tests. Depending on the preliminary results (i.e. if it works the first time), we might proceed with replacing the oem knock sensor with this prototype and logging strain from ECMlink.

 

[turboglenn]

J&S Electronics SafeGuard Indivdual Cylinder Knock Control

this is what i use, nothing better IMO, but i can't wait to see what you make!

 

[silvreclips]

Looks interesting. Cant wait to see some of the results. This could be very beneficial for people running stand alone units that dont allow control of the stock knock sensor.

 

[casuprock]

Between work and travel schedules, it's been busy! Last week, we installed this on kthackst's car. Rapping on the block produced a healthy looking signal on the oscilloscope. We installed the sensor near a head stud on the passenger side of the block. Pictures will be forthcoming.

The next step is to log the raw output of this sensor versus the OEM sensor. There was some discussion about using the ECU to log the strain signal, but a more direct comparison is to look at the raw data. We could do this with the oscilloscope, but in the spirit of making this accessible (and posting some results in a format everyone can look at) I'll whip together a special cable to scale/buffer the signals for logging on a PC line-in. This way, we will be able to record the signals time-aligned at 48kHz.

If you're following the testing, you'll want to download a program like Goldwave so you can see the results when they get posted.

 

[turboglenn]

Not sure if you know this or put any value to it but with my talks with John Pizuto of J&S safegaurd, supposedly the worst place to put a KS is where mitsu locates there's and that's midway in the cylinders. John had said that through his testing he found that "real knock" sound waves are best picked up in the area along the plain of the main journal "deck area" hard to explain but the area where the oil pan bolts screw into, think of that area and just put the KS 90* from the pan bolts along that plane. He said this is because the noise of true knock is of a frequency that will travel through the rods, to the crank and from there into the main journals and with the solid flat plane of the area right there it's the best place to get the vibrations transmitted to a sensor.

Also that when mounted mid cylinder, its like putting a microphone along theside of an empty metal can so all the noise in it echos and is amplified byt the structure of the engine. If it weren't for the intermediate shaft for the left CV my KS would still be mounted right there in the hold behind the AC bracket where the intermediate shaft for the CV bolts up.

now that i've moved it back up to the OEM location i pick up so much more false/phantom kock that it drives me nuts...

i guess to re-idderate as to why the main journal plane is supposedly the best is that detonation is a masive pressure wave/pressure increase and is like hammeriung the top of the piston with, well, a hammer, so the energy/vibration naturally travels through the solid metal rod to the crank, to main jounal etc.. a lot better and more accurately than through empty space and water filled areas of the block and other such things between the OEM ks and where the detonation is making it's most impact

From the results i've seen this seems to hold true a far as what shows up on the J&S and what doesn't, with it back in the OEM location it picks up piston slap really easily again as well as picking up other random noises... if you think about it the journal plane is a lot more solid and sturdy than the cylinder wall area which in comparison could be called flexible/flimsy which would resonate other noises easier, or at least it makes sense to me and my findings are only based on testing with my J&S and backed up with my knowledge of looking for detonation signs on plugs and such to verify if what i'm picking up is knock or not

well, i'm probably rambling and repeating the same thing over and over but hopefully it's usefull and sound knowledge and not just opinions of some one i trust that i backed up by chance with personal experience



edit just read post #5 by you and it seems you're thinking along the same lines as far as where to get some of your readings, this puts a little more faith in me behind the knowledge that i just shared, since it was only told to me by one person before and ihave no scientific basis to say it's fact, it's not something i go around spreading much in case it's not that accurate, but now i feel a tad better

also, i'm not implying that you copy the J&S or imlying to give you access to it for that reason (for several reasons but mainly i highly respect and like john of J&S, he's a great guy and has a great product.. I hope you come up with something just as good and usefull because there's just not much out there that's worth it's weight in scrap metal when it comes to knock sensing equipment, hell det cans are better than a lot of the crap out there you can build/buy on the net ), but if there's anything i can help you out with through what i have to test with i'm more than willing to jump on board and lend a hand

 

[casuprock]

turboglenn: Since we're looking for strain, we did put it near the deck... only because it was near a stud location. Like you said, another place to put it would be near the crank bearings because they bear the load from cylinder pressure. Maybe that's why J&S was suggesting the oil pan location (since the response of the bottom side is correlated to the crank bearing loads).

One thing to note is areas with high strain don't necessarily have high acceleration... so an area that works well for the strain knock sensor could be bad for an accelerometer type sensor.

 

[turboglenn]

capsurock, I'mnot sure what "type" of sensor i'm using but it's one designated for a saturn 4 banger with a cylinder bore of near the same as our engines (iirc it's like 88mm at most) but it's a single wire unit if that tells you anything. And in case I mis-stated where i was talking about mounting it was on the plane of the lower deck along the main journal area (same as i think you're talking about)

Like i said before too, I got a lot cleaner signal from there than i do from the OEM location where i seem to pick anything from piston slap to any other odd noise in the engine or things that vibrate/transfer into the block area. Luckily with the J&S i can filter out up to a certain amoungt but i like to leave it some what sensitive to noises while setting up my baseline, then "deafen" it a little after all is said and done.

can't wait to see more on your project, i've never liked knock sensors as i've often posted about but i do think having something reliable to let you know of noie in the engine is nice to have, but as far as the mitsu sensor i'd rather tune without one in general as opposed to trying to tune around that thing and it's jumpy readings

 

[casuprock]

Update: we might get some testing time in next week... Maybe friday/saturday Hoping work schedules work out so that we can spend some time with this. All work, no play lately

 

[NHerron]

I'm pretty interested in the KM5K, and thanks for sharing it here

Any updates casuprock ?

 

[1983gsp]

Any updates on the km5k?

 

[1983gsp]

Any updates?

 

[gprix1]

Bring forth the Knock Master 5000!

 

[lowell]

Did this ever work?