CartTroubleshooting a Knock Sensor
By Nick Hibberd
Hibtech Auto-Electrical Diagnostics
The captures below show the result of a little experimentation. The same diagnostic signal is measured with a known good sensor connected then disconnected to deliberately introduce an open circuit.
Capture 5 demonstrates the ECM’s response when the known good knock sensor was manually disconnected to introduce an open circuit. The recorded signal structure (better illustrated in 6) doesn’t look like the one that was previously recorded, but this isn’t too much to worry about. It’s a result of the very long 20 second time interval over which the capture was taken, which makes it impossible to see the detail in the signal. Even with the scope running at its maximum sampling rate, with this huge time interval it’s just not capable of recording the entire signal structure as clearly as before. For example, using the recorded signal signature in capture 4:
| Full diagnostic signal cycle duration | = 7.203 ms |
| DC Oscillation within this cycle | = 4.853 ms |
| % of time the signal spends oscillating | = (4.853/7.203) x 100 |
| = 67.3% | |
| No. of diagnostic signal cycles within 20 seconds | = (1/0.007203) x 20 |
| = 2776 |
So, within the 20 second recording interval there are 2776 diagnostic signal cycles, and 67.3% of that time is taken up with a 5.6 kHz oscillation. There’s a tremendous amount of information here that the DSO is trying to display. When you set the sampling rate of any DSO capture, a longer total display interval requires a longer time interval between sample points. 32 000 samples in 500 ms gives great detail over a short period of time, whereas 32 000 samples over 20 seconds gives a much bigger general picture but with less detail. However, in this case, the captures give the information we need and with reasonable detail, still enabling us to pick out the basic signal structures as shown in capture 4.
The most important observation to come out of this exercise is the clear difference in the nominal carrier voltage level when a sensor is open-circuit. The major influence on this ECM signal change is the absence of the shunt resistor between sensor pins 1 and 2.
Our problem vehicle
Although at the beginning of this investigation we saw that the trouble code was logged almost immediately after key-on, it took several further attempts to get the fault to reappear. Captures 7 through to 10 show the story of one particular key-on stage during the fault.
By now it should be clear that all these captures were taken during the last diagnostic stage of the whole signal pattern. Starting with capture 7 you can see a real difference between the two signals as the fault is now present. As before, channel A (Blue) is monitoring the cylinder 1 2-7-8 knock sensor and channel B (Red) is monitoring the other. We see the same signal outline but a total absence of oscillations within. Remember that the ECM may still be trying to emit the oscillations, but the sensor circuit could be preventing it from doing so. During the fault period, our problem sensor was disconnected, and our guess was confirmed when oscillations then appeared from the ECM.
Captures 8 and 9 show a gradual breakthrough of the oscillations, becoming better the longer the signal was allowed to carry on. Capture 10 was about the best that this diagnostic signal ever got during the fault period, so it is not surprising that the ECM wasn’t happy with the circuit at key-on. Here, on channel A, we can see the carrier voltage level a little higher than normal, and this was with the sensor connected. Remember the earlier experiment when a manually induced open circuit affected the carrier voltage by raising it slightly. An answer to the poor signal structure came from disconnecting the sensor and checking continuity between all three sensor pins. As expected only pins 1 and 2 showed continuity, but the meter picked up a resistance that hovered around 2.5 MΩ; a level that almost indicates an open circuit and fits the “sensor disconnected” exercise results.
There was more than enough evidence to condemn the sensor and subsequently a new unit was installed.
The possible conclusions formed within this study were reached based purely on the real-time evidence shown. The purpose of studies like this is to offer a basis for future diagnostic work: what happens next time when a new sensor doesn’t solve the problem?
Right: Knock sensor with outer casing removed. Note the sandwich of separators. It’s capped by a nut which is torqued independently to the main fixing bolt used. The centre of the unit is part of the thick base which fixes face-to-face with the engine block.
Left: Assembly showing shunt resistor location buried deep within the unit’s insulation.
Sensor dismantled into all available pieces showing piezo element and conductive plates. One sensor cable was attached to each plate. The damage shown to the piezo element was suspected to have been caused by dismantling. If the piezo element was damaged before, this could have contributed to the sensor's poor response to the diagnostic signal. There’s some concern about these particular knock sensors failing due to water ingress at the base, because of their mounting location at the bottom of the engine V. Whilst its operational efficiency could easily be questioned if it had been subjected to water, it’s doubtful if water could have penetrated a substance that took my grinding wheel over 10 minutes to penetrate. Anyway, water ingress was not an issue in this case.
Email
Help!