BigMoose's Ramblings on Knock Sensors



This is a summary of a discussion we had on The Old One's Discussion Board

Now for a couple of SAE papers to get you guys started.

SAE 871911 "Individual Cylinder Knock Control by Detecting Cylinder Pressure" by Kunifumi Sawamoto, Yoshiaki Kawamura, Toru Kita and Kenjiro Matsush*ta of Nissan Motor Co. Ltd. This paper will show you the relationship of knock frequency with acoustic velocity in the combustion gasses, bore diameter and resonance mode. It will also characterize the frequency multiples of the modes, and it 'taint x2, x4 ,x8 etc. Paper will also show where in crank angle knock typically occurs.

SAE 890156 "Characterization of Knock in a Spark Ignited Engine" by Kwang Min Chun and John B. Heywood of MIT Fame. This paper has a couple of nice graphs and drawings that will show you why knock collapses the ring lands.

SAE 982477 "Knock in a S.I. Engines: A Comparison between Different Techniques for Detection and Control" by F. Millio and C. V Ferraro of the Politecnico di Torino, Departmento di Energetica... like in Italy. This work will show you that the resonances are Bessel wave functions ... also that predominant resonances are are engine speed dependent. Also has some really nice graphs that show you how valve chatter and piston slap overlay the knock signatures. That's sure to get you depressed!

Here is the beginning of the analyses I wanted to share with all of you. This is for a GM Part No. 10456288 Knock sensor which is the one I think comes with the MSD Knock Alert. It is a nice and sensitive sensor with a singular resonance at 10.547 KHz. It has an approximate bandwidth of 400 Hz centered around the resonance at 10.547 KHz. More to follow:

gm10456288.jpg

The results are from an impulse to the sensor. I am working on a calibrated impulse 'device' so that our voltage curves will have some relevance to the strength of the excitation. That characteristic is relevant to signal to noise ratio concerns.

This type of driving function (impulse) will reveal the frequency response of the sensors thru FFT analyses.

Neal the factories go to great efforts to find a good responsive place on the block to place the sensor, so first we want to put it where the factory put it. If my memory is correct the MSD comes with a GM sensor with a 1/8 in NPT on the end? GM uses it to plug the lower block drain hole. My first choice would be to retap the hole, but a close second would be if you can machine an adapter, a robust adapter out of say 1" dia steel with a male thread on the bottom and a female 1/8 NPT on the top. Make sure to surface the part of the adapter that will tighten against the block. I am assuming your stock sensor is the type with a bolt thru the middle that tightens against a machined surface on the block. You want this adapter to "look" like an extension of the block. Rigid. You don't want it to have a resonance of its own in the frequency of interest. 4KHz thru 14 KHz. It wouldn't hurt to turn the outside diameter to the diameter of the stock sensor that kisses the block.

OK, here is the first round analysis of the sensors above, from greasemonkee I think.

B18C1hondab18c1pixsm.jpgGM10456288gm10456288pixsm.jpg

FWIW the B18C1 Honda knock sensor is really manufactured by Delphi, part number AS10074. The GM10456288 is the sensor that comes with the MSD knock alert.

Here are the spectral analyses of these sensors: Honda B18C1:

hondab18c1fft.jpg

and the GM10456288 that greasmonke sent me... (note it is different than the Echlin equivalent that I posted before as an example...)

gm10456288fft.jpg

Here is the ECHLIN DKS224 which is sold by NAPA as an equivalent to the REAL GM 10456288:

echlin_dks224fft.jpg

Here is my favorite non resonant, or flat response sensor made by Delphi. I prefer to do my signal processing in the electronics and not with a resonant sensor.

delphipixsm.jpg

Notice how much different the response characteristics are:

delphifft.jpg

There is a slight resonance way up in the 30 KHz range, but it is not significant as it is out of our frequency spectrum of interest. The multiple strikes are just that. The impulse device makes multiple strikes around a millisecond or so apart. This artifact is visible in all the sensor graphs.

I will let you guys start noticing the differences.

alphajesse, yes I would also stick with the stock B18C1 sensor. The resonant frequency shift is more than I think is optimum at 2 KHz. 12.5 KHz to 10.5 KHz. Also as you said too the stock GM10456288 resonance at 10.547 versus the Echlin DKS224 at 11.328 at 800 Hz higher is not optimum but should be mostly functional. Note that the GM sensors are more sensitive, as in they have a larger output per input "g", but the signal processing electronics should even that out. Also the Honda B18C1 sensor has about twice the bandwidth of the GM sensor, so it will sample a little broader resonance input.

Notice that the resonant sensors ... well resonate. They ring like a bell, so they do not have the best ability to time differentiate the signal. For example the GM style sensors are ringing for around 2 mS. Now they sure were not designed for this application, but a V8 at 7000 RPM will have ignition events around 2.14 mS apart, the sensor ring is starting to carry over into the next ignition event, a bit. The flat response sensor does not resonate, therefore its a good transducer of what the block is really doing. It gives you a little edge on discerning knock at the higher RPM's as it does not smear the signal.

ngt, with the flat sensor it is done with Digital Signal Processing routines in a Texas Instruments DSP Chip. Variants on the FFT algorithms that are more suited to impulse type signals. I am moving into Wavelet Algorithms as I believe they have more potential "up higher" in RPM.

I knew sooner or later we would get to real circuit diagrams. What you showed will work when it is hammering like a son of a gun... but it won't differentiate VTEC changeover noise, piston slap, and valve clatter on the line. So I don't like it. Any real analog knock detection device will window the knock signal, somewhat filter it (even with a resonant sensor), precision rectify it, and then peak compare... That takes more than one chip, and access to the ignition signal. It will also require a micro in the box. I am not saying you won't get an "indication" with the simple detectors, just that I would not trust a 10K engine build on one.

I updated the graphs above and added a open literature calculation from an SAE paper for knock frequency for the first 5 modes. As you will note this simple calculation does not exactly line up with the sensors. A couple of reasons, it is a simple correlation, the actual engine knock frequencies have a bandwidth to them, ie they are not pure tones. The 101.6mm 9.4 CR calc is for the GM engine the sensor was originally intended for. The dominant knock mode will be a factor of compression ratio, engine speed, piston/head geometry, combustion temperature and a few other factors... but that is a whole other story.

I also bought my flat sensors directly from Delphi...

The joy is in the journey! So feel free to ask, poke, dig, and add your data and ovservations, and I'll try to answer to the best of my ability...

A nice contribution from B20C5Turbo:

KnockTable.JPG

Here is todays knock sensor. Thanks to Scott (oneslowalltrac) for it! It is off a 1988 Toyota 3SGTE

toyota3sgte.jpg

And here is our analysis:

toyota3sgtefft.jpg

Notice that this sensor is very different than all the ones we tested so far! Its resonant frequency is an almost direct hit to the calculated U11 resonant mode. It is also tuned to a much lower frequency than our previous sensors. It would be interesting to compare the head/piston countours and the placement of the spark plug on this engine versus those we have studied to date. ... so if anyone has a pix of it, send it or link it to me and I will add it to this presentation.

oneslowalltrac graciously provided these pix of the 3sgte piston and head:

ST165_Piston_1.JPG3SGTE-gen1.jpg

Now we have a Porsche 928 sensors Bosch Part Number 0 261 231 008 with thanks to Mike (YYY)

porsche928.jpg porsche928det.jpg

And here is our analysis:

porsche928fft.jpg

Well, this looks like another "flat" sensor, and a very good one at that. Similar to the nonresonant or flat Delphi sensor profiled above. The conclusion is that the Porsche folks are doing the signal processing in software in a DSP'ish device in the ECM.

Mike sent me 3 of these sensors, two that were pretty deteriorated with respect to the potting epoxy. They all performed with similar characteristics. Of course when exposed to engine operating conditions and contamination, their operation may well have been comprimised. IMHO all three sensors were probably operational when removed.

Thats all for now...

BigMoose



[Return To Main Menu]