This study describes the analysis of an internal combustion (ic) engine knocking phenomena that was developed using recorded vibration signals.
In the first part of this article, the analysis of the available signals is described. This analysis is the basis for the design of a special knocking detection algorithm. The main goal of this task was the determination of the characteristics and the parameters of the knocking noise or vibration from signal processing point of view.
The last part of the article shows the design of a knocking detection algorithm that fulfills the following requirements:
- Detection of the knocking phenomena that was represented by the recorded vibration signals.
- As input signals one or more noise or vibration signals and one synchronous impulse should be used.
- Revolution speed independence
- Calculation of one knocking indication signal for each cylinder
- Each knocking indication signal shows the occurrence and the strength of the knocking that comes from the corresponding cylinder.
- The knocking detection algorithm can be implemented easily on an industrial-proofed signal analysis system.
The presented detection algorithm fulfills all those requirements.1. Data Used in Study
All the work presented in the following is based on the following signal files (Table 1
Table 1: List of Knocking Signals used for analysis
The data had the following properties:
- On the 1st channel the vibration signal was recorded using a sample rate of 48 kHz. On the 2nd channel a synchronous signal was recorded at the same sample rate.
- The synchronous signal consists of one pulse per engine cycle.
- For the analysis those files were divided into files with a signal length of 5 seconds.
The recorded synchronous pulses were transformed into time instants using a linear interpolation of the rising flank’s threshold-crossing.
2. Data Analysis: Time Domain
At first the vibration signals were analyzed in time domain without any pre-processing to get a first impression of the knocking phenomena. The following picture shows a short section of the signals which were recorded at 2000 rpm with no, a few and hard knocking (Figure 1):
Figure 1: Knocking time domain signals at 2000 rpm: Top – Signal with no knocking, Middle – Signal with a few knocks, Bottom – Signal with hard knocking.
The diagrams show a section of 0.1 seconds. All signals shown in the diagrams have periodic components and characteristic periodic occurring bursts. Because of the burst period of approximately 0.015 seconds this component originates from the combustion process.
Some bursts of the signal with hard knocking show significant spikes. There are nearly no differences visible between the signal with no knocking and the signal with few knocking.
The spikes also are visible at the 5000 rpm signal with hard knocking. But the 5000 rpm signals do not show the periodic combustion bursts. They seem to be overlaid or hidden by other signal components that become stronger at higher revolution speeds (Figure 2).
Figure 2: Knocking time domain signals from 5000 rpm: Top – No knocking, Bottom – Hard knocking.
The spikes which can be seen in the hard knocking signals have to be insulated from the whole vibration signal to get a knocking criterion.
3. Data Analysis: Angle Domain
Because the knocking of the engine occurs during the combustion process the detection of the spikes can be improved by angle synchronous signal processing in two ways:
- Spikes which are coming from a combustion process can be separated from other spikes if only those spikes will be taken into concern which occur during the combustion process. This is possible if the crankshaft angle is known, where a certain spike occurs.
- If the angle of a certain spike is known it can be associated with one certain combustion process of one certain cylinder.
For doing angle domain analysis the vibration signal has been digital resampled using the synchronous information. The result is a vibration signal with a fixed number of angle equidistant samples per each engine cycle.
The figures below show representative resampled vibration data for each available rpm and rating (Figure 3). It can be seen that the periodic burst are related to certain angle ranges. Each of those angle ranges is related to the combustion process of one certain cylinder. The 2000 rpm signal with hard knocking shows significant spikes at the 4th cylinder.
Figure 3: Knocking angle domain signals at 2000 rpm: Top – Signal with no knocking, Middle – Signal with a few knocks, Bottom – Signal with hard knocking.
At the 2000 rpm signal with few knocking the 4th cylinder also shows higher amplitudes as the other ones. But it is difficult to derive a reliable knocking criterion from the shown signals, because the difference between “no knocking” and “few knocking” is too small.
The signals at 5000 rpm are disturbed by a higher overall noise. That makes it more difficult to detect knocking also at the signal with hard knocking (Figure 4).
Figure 4: Knocking angle domain signals from 5000 rpm: Top – No knocking, Bottom – Hard knocking.
Additionally a transient component at approximately 320 degree crankshaft occurs that does not correspond with any combustion cycle. But this component will not be taken into account because it is known that no combustion occurs at this angle range.
As a first conclusion it can be seen, that angle domain analysis improves knocking detection because spikes coming from knocking can be separated from other spikes. Additionally, the knocking cylinder can be determined.
The main problem that has to be solved now is to distinguish between a normal combustion and a knocking combustion in a more reliable way as it is possible with angle domain analysis
.4. Data Analysis: Angle-Order Domain Analysis
This differentiation requires a deeper analysis of the combustion signal not only in angle, but also in frequency or order domain
. If the frequency characteristic of a spike coming from knocking is known it can be better separated from the signal emitted by a normal combustion.
To evaluate the frequency components of the combustion signals the angle synchronous Wigner-Ville spectrum was used. This method is well suited for such analysis tasks. The angle synchronous Wigner-Ville spectrum allows the choice of a high angle resolution and a high order/frequency resolution at the same time.
The picture below shows three Wigner-Ville spectra calculated from 2000-rpm signals of 5 seconds length (Figure 5
). Especially the 4th cylinder of the hard knocking signal shows significant amplitudes at the range above the 600th order. But also the 2nd cylinder shows such an effect but with a lower amplitude. This example shows, that the Wigner-Ville spectrum allows visualization of knocking processes that cannot be seen at the time domain and the angle domain analysis in such a clear way.
Figure 5: Wignerville angle-order domain signals at 2000 rpm. Y-axis is degrees of rotation, X-axis is orders, Z-axis (color) is amplitude: Left – Signal with no knocking, Middle – Signal with a few knocks, Right – Signal with hard knocking.
But also the signal with few knocking shows higher amplitudes at the 3rd cylinder. The characteristic of that process is not as distinct as it can be seen at the hard knocking signal.
Similar effects can be seen at the 5000-rpm examples shown below (Figure 6). The output of the Wigner-Ville spectra is as clear as at 2000 rpm. The disturbing noise is identified very effectively.
Figure 6: Knocking angle-order domain signals from 5000 rpm. Y-axis is degrees of rotation, X-axis is orders, Z-axis (color) is amplitude: Left – No knocking, Right – Hard knocking.
The pictures above show revolution synchronous Wigner-Ville spectra averaged over 5 seconds. That means, the spectral components at the pictures are averaged over several knocking processes. This method allows a good analysis of the common frequency characteristic of the knocking phenomena.
But, the revolution synchronous Wigner-Ville spectrum also is well suited to investigate one certain knocking combustion.
The following figure shows two Wigner-Ville spectra (Figure 7).
Figure 7: 5000 rpm data: Left – Signal and Wigner-Ville angle-order without knocking, Right – Signal and Wigner-Ville angle-order with hard knock
The right one shows the angle-order-analysis of the knocking process shown above. The left side of the picture shows a motor cycle without knocking as a reference.
It can be seen that it is possible to analyze single knocking combustion using revolution synchronous Wigner-Ville analysis.
The Wigner-Ville analysis is a good tool to do high sophisticated analysis of the knocking phenomena for example to improve the engine design or engine control.
A Simcenter Anovis system
allows calculation and off-line presentation of angle synchronous Wigner-Ville spectra as it can be seen in this article.4. The Knocking Detection Algorithm
The Wigner-Ville analysis shown above is a well suited tool for deeper analysis of the knocking phenomena. But this method cannot be implemented on a real-time signal processing system in an economical way. That means:
- For real-time calculation of Wigner-Ville spectra a high computation performance is necessary. Up-to-date signal processing systems which provide the needed computation power and system resources are very expensive.
- The output of the Wigner-Ville spectra is very reliable. But at the same time it is oversized if only the detection of knocking is required.
Starting at those points an algorithm has been designed that has the following advantages:
- Detection of combustion knocking in real-time.
- Calculation of one easy to handle knocking level indicator. This indicator not only says that a cylinder is knocking, it also says how strong the knocking is.
- This is done by a pass band time filtering at the frequency range identified by the Wigner-Ville analysis and transforming the filtered data into the angle domain.
- Using the angle domain, association of each certain knocking process with the corresponding cylinder is performed. That means that for example for a four cylinder engine four knocking level indicators are provided by the system.
- The algorithm runs stand-alone with an easy to handle parameterization to adapt the system to different input signals and/or engines (for example number of cylinders).
The following figures show the knocking level indicator outputs. It can be seen that the output corresponds with the rating supplied with the data (no, few or hard knocking) for 2000 rpm (Figure 8).
Figure 8: Knocking algorithm output (Y-axis) versus number of cycles (X-axis) for signals at 2000 rpm: Top– Signal with no knocking, Middle – Signal with a few knocks, Bottom – Signal with hard knocking.
The knocking indicator level shows an equal reliability for 5000 rpm as well (Figure 9).
Figure 9: Knocking algorithm output (Y-axis) versus number of cycles (X-axis) for signals at 5000 rpm: Top – No knocking, Bottom – Hard knocking.
The different colors of the lines seen in the pictures are related to the certain cylinders. The figures show, that the knocking detection algorithm allows to determine exactly which cylinder is knocking when.
This article shows that it is possible to detect and to analyze the knocking phenomena at the supplied vibration signals. The following conclusions can be made:
- The Wigner-Ville analysis is a good tool to do high sophisticated analysis of the knocking phenomena for example to improve the engine design or engine control.
- The knocking detection algorithm allows real-time knocking detection and evaluation.
- The knocking detection algorithm allows to determine exactly which cylinder is knocking when.
- The knocking indicator level is a reliable criterion to evaluate the knocking of an internal combustion (ic) engine independent to revolution speed. The knocking indicator level provides the information how strong a certain cylinder is knocking.
- The knocking detection algorithm can be implemented in the Simcenter Anovis system for real-time and stand-alone operation.
Questions? Email email@example.com
(Americas) or firstname.lastname@example.org
(Europe and Asia).Related Links: