Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
COMBUSTION NOISE DETECTION IN AN INTERNAL COMBUSTION ENGINE
Document Type and Number:
WIPO Patent Application WO/2019/016010
Kind Code:
A1
Abstract:
A method of monitoring combustion noise in a multi-cylinder internal combustion comprises: a1) monitoring combustion noise at a first cylinder of said engine by means of a knock sensor over a sensing window to determine a measured combustion noise intensity; b1) from an injector noise mapping dependent on fuel injection pressure, looking-up a set of noise parameters including an opening noise duration, an opening delay, a closing noise duration and a closing delay corresponding an injector event occurring on a second cylinder; c1) computing, based on the injection timing at the second cylinder and on said set of noise parameters, an estimate of opening noise and an estimate of closing noise falling within said sensing window; d1) computing a compensated noise intensity by subtracting said estimates of opening and closing noise from said measured combustion noise intensity; e1) using said compensated noise intensity for determining the presence of knock or pre-ignition in said first cylinder. Also disclosed is a method for calibrating the duration and delay of injector noise.

Inventors:
RANDAZZO, Stephane (6 rue Guy de Maupassant, Fameck, Fameck, 57290, FR)
DA GRACA, Mathieu (4 rue Moliere, Villerupt, Villerupt, 54190, FR)
BOUAITA, Abdelhamid (3 Boucle des Semailles, Elange, Elange, 57100, FR)
Application Number:
EP2018/068524
Publication Date:
January 24, 2019
Filing Date:
July 09, 2018
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
DELPHI AUTOMOTIVE SYSTEMS LUXEMBOURG SA (Avenue de Luxembourg, 4940 Bascharage, 4940, LU)
International Classes:
F02D35/02; F02D41/24; F02P5/152; G01L23/22
Domestic Patent References:
WO2003040677A12003-05-15
Foreign References:
EP1219805A22002-07-03
EP2339313A12011-06-29
US5408863A1995-04-25
DE102008041175A12010-02-18
Attorney, Agent or Firm:
DELPHI FRANCE SAS (Delphi Technologies - Campus Saint Christophe Bâtiment Galilée 2 -, 10 avenue de l'Entreprise, Cergy Pontoise Cedex, 95863, FR)
Download PDF:
Claims:
CLAIMS:

1 . A method of monitoring combustion noise in a multi-cylinder internal combustion, said method comprising:

a1 ) monitoring combustion noise at a first cylinder of said engine by means of a knock sensor over a sensing window to determine a measured combustion noise intensity;

b1 ) from an injector noise mapping dependent on fuel injection pressure, looking-up a set of noise parameters including an opening noise duration, an opening delay, a closing noise duration and a closing delay corresponding an injector event occurring on a second cylinder;

c1 ) computing, based on the injection timing at the second cylinder and on said set of noise parameters, an estimate of opening noise and an estimate of closing noise falling within said sensing window;

d1 ) computing a compensated noise intensity by subtracting said estimates of opening and closing noise from said measured combustion noise intensity;

e1 ) using said compensated noise intensity for determining the presence of knock or pre-ignition in said first cylinder.

2. The method according to claim 1 , wherein said injector noise mapping is further dependent on a frequency filter and said compensated noise intensity is a compounded intensity computed as a weighted sum of filter-specific compensated intensities.

3. The method according to claim 1 or 2, wherein said injector noise mapping comprises:

4. a calibrated duration map of said closing noise duration and said opening noise duration for a plurality of fuel injection pressures and preferably for one or more filter frequencies; and a calibrated delay map of said opening delay and said closing delay depending for a plurality of fuel injection pressure and preferably for one or more filter frequencies.

5. The method according to claim 3, wherein step c1 ) comprises estimating an amount of closing noise and of opening noise falling within said sensing window and looking-up from an opening noise mapping and closing noise mapping dependent on fuel injection pressure, said estimates of opening and closing noise.

6. The method according to claim 3 or 4, wherein said calibrated duration map and calibrated delay map are obtained by a method as claimed in any one of claims 6 to 10.

7. A method of determining noise in a multi-cylinder internal combustion engine, wherein a knock sensor is used to determine combustion noise such as knock or pre-ignition, the method comprising:

a2) operating a first engine cylinder in a normal combustion mode and monitoring with said knock sensor a combustion noise for said first cylinder over a predetermined sensing window per combustion event;

b2) in a second engine cylinder, performing a plurality of injector events by sweeping a timing of said injector events over a test period overlapping with said sensing window, in order to obtain a noise data set indicative of noise intensity over time as recorded by said knock sensor over said sensing window;

c2) processing said noise data set in order to determine therefrom noise parameters related to at least one of injector opening and/or injector closing in said second cylinder.

8. Method according to claim 6, wherein at step c2) said noise data set is processed to identify a duration of closing noise, a delay of closing noise, a duration of opening noise and a delay of opening noise. 9. Method according to claim 6 or 7, wherein at step c2) said noise data set is processed to identify intensities of opening noise and closing noise.

10. Method according to claim 6, 7 or 8, wherein said sweeping in step b2) is repeated for a plurality of fuel injector pressures.

1 1 . Method according to any one of claims 6 to 9, wherein said processing step includes :

identifying a first increase ramp in the data set from a stable level to a first plateau, the duration of said first increase corresponding to the duration of said closing noise, and the distance between the end of said first increase ramp and the start of said sensing window representing the closing noise delay;

identifying a later, second increase ramp in the data set from a stable level to a second plateau, the duration of said second increase corresponding to the duration of said opening noise, and the distance between the end of said second increase ramp and the start of said sensing window representing the opening noise delay.

Description:
COMBUSTION NOISE DETECTION IN AN INTERNAL COMBUSTION

ENGINE

TECHNICAL FIELD

[0001 ] The invention relates to combustion noise detection in a multi- cylinder internal combustion engine. The invention more particularly relates to a method for monitoring combustion noise in an internal combustion by means of a knock sensor for the purpose of detecting knock or pre-ignition.

BACKGROUND OF THE INVENTION

[0002] Knock and pre-ignition are two well-known undesired combustion modes, which create explosions waves that are over dimensioned for a regular engine piston, ultimately causing an irreversible destruction of the engine parts.

[0003] These undesired combustions modes are usually detected by means of a knock sensor. A knock sensor is typically a vibration sensor located in the lower engine block.

[0004] As the knock sensor is directly connected to the engine, it records all kinds of vibrations that are produced in this environment. It is thus important to discriminate irrelevant vibrations and/or background noise from vibrations related to knock detonations. The separation between background noise and knock is commonly realized by setting a knock sensing window and a noise detection threshold.

[0005] The sensing window is the period (crank angle segment) during which the knock sensor records the vibrations of the engine. Knock detonations only occur around the ignition-combustion phases during an engine cycle. It is thus unnecessary to consider the recordings of the sensor in remote phases of the engine cycle. There are commonly two different sensing windows set for a knock sensor, a first window for pre- ignition sensing and a second window for knock sensing. [0006] The detection threshold is predetermined for each sensing window depending on the mean background noise that is recorded for the engine cycle without knock or pre-ignition.

[0007] The background noise includes all kinds of noises generated by the engine components. Particularly critical for combustion noise monitoring is the operational noise of a fuel injector in a neighbor cylinder, when operated during the knock sensing window. It has indeed been observed that the noise intensity due to injector actuation may be high enough to cross the predetermined knock detection threshold, hence provoking false knock detection. False knock detection usually results in a wrong adjustment of the ignition timing and a poor performance of the engine.

[0008] Solutions have been proposed to address this problem. For example, US 7,007,663 discloses an internal combustion engine knock control method in which, when the fuel injection period and the knock sensing window overlap, the knock control is temporarily disabled.

[0009] The major drawback of this solution is obviously that it creates blind time periods during which no knock control can be detected, implying a risk of missing an undesired combustion. Another drawback is that when the knock determination window and the injection pulse overlap, the injection timing is not set to an optimum value leading to a loss in engine performance.

[0010] Another solution is proposed in document US 6,557,527. In the knock detection method presented therein, the knock determination threshold is modified when entering in a situation that could create false positive detection of knock.

[001 1 ] Here, the major drawback is that by setting a higher knock determination threshold, the system has a very poor knock determination ability when the knock determination window and the injection pulse overlap, resulting in risks of missing undesired combustions. OBJECT OF THE INVENTION

[0012] An object of the present invention is to provide an improved method of monitoring combustion noise that does not suffer from the above-mentioned drawbacks

GENERAL DESCRIPTION OF THE INVENTION

[0013] The present invention relies on the idea of compensating measured noise intensities for injector operational noise, rather than disabling or modifying detection thresholds when an injector event overlaps with the knock detection window.

[0014] To implement such compensation, a preliminary step is predict the operational noise that will be generated by the actuation of the fuel injector due to an injector event occurring at a given injection timing.

[0015] Accordingly, in one aspect the present invention provides a method for determining noise in a multi-cylinder internal combustion engine, which is designed to identify noise parameters related to the opening phase of the injector and to the closing phase thereof. It may be noted here that intensity, duration and propagation of noise mainly depend on the fuel injection pressure in the fuel injector.

[0016] These noise parameters can then be used to estimate the amount of injector operational noise that falls into the noise sensing window of the knock sensor, and hence perturb the sensor reading.

[0017] In another aspect, the present invention provides a method for identifying the noise parameters related to opening and closing the injector.

[0018] So, according to the first aspect, a method of monitoring combustion noise in a multi-cylinder internal combustion comprises the steps of:

a1 ) monitoring combustion noise at a first cylinder of the engine by means of a knock sensor over a sensing window to obtain a measured combustion noise intensity;

b1 ) from an injector noise mapping dependent on fuel injection pressure, looking-up a set of noise parameters including an opening noise duration, an opening delay, a closing noise duration and a closing delay corresponding an injector event occurring on a second cylinder;

c1 ) computing, based on the injection timing at the second cylinder and on said set of noise parameters, an estimate of opening noise and an estimate of closing noise falling within said sensing window;

d1 ) computing a compensated noise intensity by subtracting said estimates of opening and closing noise from said measured combustion noise intensity; and

e1 ) preferably using said compensated noise intensity for detecting the presence of knock or pre-ignition in said first cylinder.

[0019] The major benefit of the present method is the possibility of conducting combustion noise analysis with combustion noise values (i.e. the compensated noise intensity) that are unaffected by injector operational noise falling in the sensing window. The present method is applicable to detection of both knock and pre-ignition. The sensing window may thus either be a knock detection window or a pre-ignition sensing window, where the duration of the sensing period (in degrees of Crank Angle -CA) and the positioning of the sensing period are adapted accordingly.

[0020] In practice, the noise recorded by the knock sensor is preferably filtered to take into account combustion specific noises. Accordingly, steps b1 ) to d1 ) of the method are preferably carried out for one or more frequency ranges. Therefore, the injector noise mapping is advantageously further dependent on a frequency filter. In such case, the compensated noise intensity is a compounded intensity computed as a weighted sum of the individual filter-specific compensated intensities.

[0021 ] Preferably, the injector noise mapping comprises:

a calibrated duration map of the closing noise duration and the opening noise duration for a plurality of fuel injection pressures and preferably for one or more filter frequencies; and

a calibrated delay map of the opening delay and the closing delay depending for a plurality of fuel injection pressure and preferably for one or more filter frequencies. [0022] Such calibrated maps are typically learned by conducting calibration tests on a test bench for a given engine design. They are advantageously determined by the method disclosed below in relation to the second aspect. The calibrated duration and noise maps can thus be loaded in the engine at installation. Although bench calibration is preferred for expedience, it is also possible to devise a learning procedure when the engine is running in a user vehicle, in order to learn the values and populate the maps. In embodiments, the set of parameters further includes an intensity of noise.

[0023] In an embodiment, the method preferably also uses an opening noise mapping and closing noise mapping dependent on fuel injection pressure, and preferably on frequency. At step c1 ), an amount of closing noise and an amount of opening noise falling within said sensing window are first determined based on the injection timing at the second cylinder and on the set of noise parameters, i.e. opening noise duration, an opening delay, a closing noise duration and a closing delay. Then the estimate of closing noise and estimate of opening noise are looked up from the opening noise mapping and closing noise mapping, for the relevant fuel pressure and filter.

[0024] Here also these mappings are preferably calibrated on the test bench. In this embodiment, a concept of "amount of opening noise" is used, which represents the proportion of opening noise falling in the sensing window (also later referred to as opening/closing part in window). The same applies to the amount of closing noise. These amounts thus indicate to which extent a sensing window is impacted by the injector opening and/or closing noise, but is a value that does not immediately reflect the intensity or amplitude of the noise. The opening noise mapping and closing noise mapping however contain this injection pressure dependent information of noise amplitude/quantity in function of the "amount" of opening/closing noise.

[0025] According to the second aspect of the invention, a method of determining noise in a multi-cylinder internal combustion engine is proposed, wherein a knock sensor is used to determine combustion noise such as knock or pre-ignition. The method comprises the steps of:

a2) operating a first engine cylinder in a normal combustion mode and monitoring with said knock sensor a combustion noise for said first cylinder over a predetermined sensing window per combustion cycle;

b2) in a second engine cylinder, performing a plurality of injector events by sweeping a timing of said injector events over a test period overlapping with said sensing window, in order to obtain a noise data set indicative of noise intensity vs. time as recorded by said knock sensor over said sensing window;

c2) processing said noise data set in order to determine therefrom noise parameters related to at least one of injector opening and injector closing in said second cylinder.

[0026] A merit of this method is that it allows separating the respective contributions of injector closing and opening from the recorded knock sensor signal. In particular, the noise data set is processed to identify a duration of closing noise, a delay of closing noise, a duration of opening noise and a delay of opening noise. The intensities of closing noise and opening noise can also readily be determined from such data set.

[0027] As injector operational noise is typically dependent on fuel injection pressure, the sweeping step b2) is preferably repeated for a plurality of fuel injector pressures. Accordingly, a noise delay map and a noise duration map depending on fuel pressure can be built.

[0028] Raw sensor measurements are preferably filtered according to predetermined frequency ranges to retain noise occurring at relevant frequencies, i.e. known to correspond to combustion noise.

[0029] The processing step c2) may include:

identifying a first increase ramp in the data set from a stable level to a first plateau, the duration of the first increase corresponding to the duration of the closing noise, and the distance between the end of said first increase ramp and the start of said sensing window representing the closing noise delay; and identifying a later, second increase ramp in the data set from a stable level to a second plateau, the duration of the second increase corresponding to the duration of the opening noise, and the distance between the end of the second increase ramp and the start of the sensing window representing the opening noise delay.

[0030] As used herein, the term "normal combustion" designates combustion conditions leading to combustion of the air/fuel mixture and providing torque, without occurrence of knock or pre-ignition. In practice, the whole engine is operated in normal combustion mode for the purpose of determining the noise parameters in accordance with the present method.

BRIEF DESCRIPTION OF THE DRAWINGS

[0031 ] Further details and advantages of the present invention will be apparent from the following detailed description of not limiting embodiments with reference to the attached drawings, wherein:

Fig.1 is a combined graph representing (a) injector current as well as

(b) knock sensor feedback and (c) knock sensor acquisition status vs. crank angle position;

Fig.2: is a graph representing knock intensity vs. CA, as obtained by an injector timing sweeping procedure, as used in the present method;

Fig.3: is a combined graph representing (a) the knock sensor feedback,

(b) injector opening and closing noise and (c) the knock sensor acquisition status vs. CA, plotted for the injector closing time;

Fig.4: is a combined graph representing the (a) the knock sensor feedback, (b) injector opening and closing noise and (cl the knock sensor acquisition status vs. CA, plotted for the injector opening time;

Fig.5: is a flow chart representing one embodiment of an algorithm adapted for determining amounts of closing noise in the sensing window; and Fig.6: is a flow chart representing an embodiment of algorithm for compensating the knock sensor feedback.

DESCRIPTION OF PREFERRED EMBODIMENTS

[0032] As it is known in the art, in a gasoline operated multi-cylinder engine, a knock sensor is mounted on the engine block in order to detect vibrations on the engine. For a 3 or 4-cylinder engine, one knock sensor is generally considered sufficient. The knock sensor generates an output signal that is representative of engine block vibrations and that is used in the ECU or other module to detect the presence of knock, but also of pre- ignition. Conventional combustion strategies seek to avoid these two phenomena since they may lead to irreversible physical damages to the engine.

[0033] Since conventional knock sensors are sensible to a variety of noises, the output signal of the noise sensor is typically filtered out (by pass band filters) to retain only one or several frequency ranges that are known to correspond to combustion related noises, and specifically to the sources of noise of interest, i.e. knock or pre-ignition.

[0034] Knock detection during a combustion cycle can be perturbed by the occurrence of an injector event in another cylinder of the engine block, i.e. a neighboring fuel injector has been actuated and generates noise due to the mechanical impact of the needle against abutment surfaces at opening and closing. This injector operational noise is often generated at frequencies corresponding to the combustion frequencies of the knock sensor, and may cause a false detection of knock if the injector event overlaps with the knock sensor sensing period in the observed combustion cycle.

[0035] The perturbation of knock detection due to injector operational noise is illustrated on Fig.1 . The bottom graph c) corresponds to the acquisition status of a knock sensor, defining a sensing window extending over a predetermined crank angle segment. The position (start/end) of the sensing window is defined in order to monitor knock in a first cylinder of the engine, typically depending on combustion timing. Here the knock sensor will record noise between a sensing period ranging from 10 to 60° of CA (crank angle).

[0036] For the purpose of this graph, combustion in the first cylinder is a normal combustion, i.e. there is no knock nor pre-ignition. The top graph a) represents an injector command signal by way of the current flowing through a fuel injector associated with another fuel cylinder in the engine, whereas graph b) represents the output signal (voltage) of the knock sensor over the shown period, i.e. between -60 to 100° CA. As can be observed, the knock sensor detection voltage is affected by the actuation of the fuel injector. As a matter of fact, one can distinguish that the sensor signal is perturbed due to two events: injector opening and injector closing (respectively corresponding to rise and fall of the injector current). There is however a slight delay between the injector current edges and the detection of injector operational noise, due to some delay of injector actuation.

[0037] The combined graph of Fig.1 clearly illustrates how, in such case where the injector of another cylinder is actuated at a time that overlaps with a sensing window of a monitored cylinder, knock detection will be affected by injector operational noise.

[0038] It may be noted in passing that the injector noise caused by the injector of the cylinder under current noise monitoring will not impact the knock sensing window, because fuel injection in gasoline engines occurs early in the combustion cycle, so that the operational noise of this injector cannot impact the sensing period.

[0039] Traditionally, knock detection consists in comparing the measured noise intensity to a predetermined threshold value that is typically set as a multiple of the background noise recorded by the knock sensor. When the fuel injector of another cylinder is actuated at a timing that is too close from, or overlaps with the sensing period of the knock sensor, its operational noise will increase the background noise and lead to false knock detection unless counter-measures are taken.

[0040] Terminology. For ease of explanation, the following terminology will be used in the following description: first cylinder: is the cylinder under monitoring by means of the knock sensor, to record combustion noise occurring during a given combustion cycle.

second cylinder: is another cylinder of the engine, where an injector event is performed during the same combustion cycle.

injector event: is the fact of actuating an injector at a predetermined timing by applying a drive current in accordance with an injector drive pulseof known length, in order to cause lifting of the needle during a certain time period, causing opening and closing of the injector.

- sensing period: is the period during which the knock sensor is active (acquisition status=1 ) for recording combustion noise at the first cylinder over a defined crank angle segment.

[0041 ] Still to be noticed at this point, the intensity or amplitude of noise that is conventionally used in the ECU or knock module is computed by integrating the measured sensor signal over the sensing window. Such values of noise intensity are conventionally computed for each frequency filter.

[0042] Turning now to the present invention, as explained above, it encompasses two aspects:

[0043] The estimation of noise attributable to the closing of the injector and to the opening of the injector in the second cylinder, which requires calibration of noise parameters; and

[0044] The compensation: measured combustion noise is compensated for injector noise overlapping the sensing window by subtracting estimates of injector opening and closing noise from the measured combustion noise.

[0045] A. determination of injector noise - calibration

[0046] The aim of the calibration procedure is to determine the impact, on the measured knock sensor signal, of operational noise of the fuel injector installed on the second cylinder, whereas combustion is being monitored on the first cylinder. The calibration procedure is typically done on a test bench. [0047] As discussed in relation to Fig .1 , operational noise from the injector on the second cylinder will be generated at injector opening and at closing, due to mechanical impact. The present method is based on the following relationship, which acknowledges the fact that the noise recorded by the knock sensor comprises a noise due to injector actuation that can be split in two parts:

Noise_intensity = Background_part + I n j_cl os i n g_pa rt + lnj_opening_part

[eq. 1]

[0048] Where:

- Noisejntensity is the intensity measured by the knock sensor;

- Background_part is the intensity when an alternate injection event is far from the sensing period;

- inj_closing_part is the intensity increase due to injector closing noise; and - inj opening part is the intensity increase due to injector opening noise.

[0049] Depending on the distance between the injector event and the sensing window, the contribution of the injector operational noise may be more or less important. This will be more apparent in the light of Fig. 2, which shows the evolution of noise intensity at the first cylinder depending on the timing of injection at the second cylinder. The horizontal axis is graduated in degrees of crank angle. The vertical axis indicates the noise intensity recorded by means of a knock sensor for a fixed sensing period, as indicated below the axis. All of the measurement points in this graph represent a noise intensity that has been measured by the knock sensor within the indicated sensing window. But the peculiarity of this graph is that each intensity value is positioned along the horizontal axis at the moment that corresponds to the end of injection (i.e. downward edge of injector drive pulse) for the second cylinder injector.

[0050] That is, point A in Fig.2 represents the noise intensity recorded during the sensing period for an injector event that ended at -20° before fire top dead center (TDC-f). Point B represents the noise intensity recorded during the sensing window for an injection ending at -10°. Point C is the recorded intensity for an injector event ending at about -5°. For the intensities represented by points D and E the injector event ended at about 30 and 55°, respectively.

[0051 ] As it will be understood, such graph is obtained by operating the first cylinder in a normal combustion mode, i.e. without knock. Injector events on the second cylinder are performed with a constant pulse width (duration of injection), but the injection timing is swept according to a test period that encompasses the sensing period. Here for example the test period extends from -20° to 80° CA. The sweeping was carried out with by incrementally increasing the start/end of injection by +2°CA, starting from End of injection = -20° CA.

[0052] In this graph, the intensities before -10° and after 72° correspond to the background part in eq.1 . Beyond these limits the sensed noise is not impacted by the actuation of the fuel injector in the second cylinder.

[0053] In between these limits, the noise intensity progressively increases and decreases. It may be noted that the intensity of the noise is expressed by choice in an arbitrary unit, here as a percentage. The data points have been fitted with lines, although this is not required.

[0054] As the injection timing is swept to the right, the noise contribution of the injector increases from the background value B by following a ramp to a first plateau (containing point C): there the entire noise due to injector closing is detected by the knock sensor.

[0055] As the injector timing is further swept to higher values of CA, the noise increases again due to the fact that the injector opening noise starts impacting the sensing period. At about 23° CA the maximum intensity is reached: both the opening and closing noise fall entirely into the sensing period and lead to a maximum perturbation of the sensor signal (noise intensity of point D). As sweeping is continued by incrementally moving the injection timing to higher values, the intensity decreases because the noise due to injector closing and then to injector opening, will progressively move out of the sensing window. At about 52°, the injector closing noise no longer affects the sensing window and the subsequent plateau (level of point E) corresponds to the maximum noise due to injector opening. [0056] From Fig.2 it can thus be seen that sweeping the timing of injection over a test window overlapping, and preferably encompassing, the sensing window provides a separation effect, by which it is possible to determine the respective noise contributions of both the opening and the closing of the injector.

[0057] From such a curve one can extract a duration of opening/closing noise, an intensity of opening/closing noise and a delay of opening/closing noise, as will be more apparent below.

[0058] Let us now turn to figs 3 and 4, the horizontal axis is graduated in degrees of CA and the same sensing period for the knock sensor is represented (graph c). At the top (graph a), one will recognize the curve representing the total injector noise having the same shape as in Fig.2, the data points have been fitted with lines.

[0059] For both Figures, a single calibration procedure has been performed, which comprises:

- Operating the first cylinder in normal combustion mode and monitoring the combustion noise over the sensing period,

- at the second cylinder, performing injection events by sweeping the injection timing over a test window overlapping the sensing window.

[0060] In Fig.3, the measured noise intensity is plotted over the CA timing corresponding to the end of injection (EOI). In Fig.4, the measured noise intensity is plotted over the CA timing corresponding to the start of injection (SOI).

[0061 ] At graph b) the first curve, designated 2, represents the noise contribution of the close of injection, whereas the second curve, designated 4, represents the noise contribution of the opening of injection. These curves are easily deduced from graph a). The graph is normalized for the ramp of interest and graduated in %.

[0062] In Fig. 3, the injector closing noise starts impacting the window at point EOM (corresponding to an injector drive pulse ending at -15° CA), where intensity is thus 0%, and is fully recorded by the sensing window at point EOI_2 (corresponding to an injector drive pulse ending at -0°), where intensity is 100%. [0063] We can also deduce two other noise parameters from the graph:

- the duration of the injector closing noise: Duration = EOI_2 -

EOM

- the delay between the start of detection of the closing noise and the end of the drive pulse: Delay = Win_start - EOI_2, where win_start is the start of the sensing period.

[0064] Turning to Fig. 4, the injector opening noise starts impacting the window at point SOM (injector drive pulse starting at -5° CA), where intensity is thus 0%, and is fully recorded by the knock sensor at point SOI_2 (corresponding to drive pulse starting at +5°CA, where intensity is 100%.

[0065] We can also deduce two other noise parameters from the graph:

- the duration of the injector opening noise: Duration = SOI 2-

SOM

- the delay between injection an noise: Delay = Win_start -

SOI_2, where win_start is the start of the sensing period.

[0066] For the purpose of calibration, the sweeping of injector events over the test window is carried out for a plurality of fuel supply pressures (rail pressure), preferable at least five. On can thus build a mapping of noise duration and noise delay for injector opening and closing.

At the second cylinder, the sweeping is carried out for constant combustion conditions, without knock. In particular the sweeping of the second injector is carried out at fixed engine speed, fixed load (constant drive pulse), fixed CA50 (fixed phasing) and fixed sensing window position.

[0067] In summary, the above-described procedure allows, by means of a sweeping of injector event timing, to identify a number of parameters such as:

- intensity of injector opening noise

- intensity of injector closing noise

- closing noise duration

- closing delay - opening noise duration

- opening delay

[0068] As it will be understood, depending on the timing of injection at the second cylinder, the noise due to injector event may be fully inside the sensing period or only partly.

[0069] From the above parameters, one can compute the amount (proportion) of noise that is present in the sensing period, as will be explained with reference to Fig.5.

[0070] Figure 5 is drawn-up for a given filter (box10) of a selected frequency range used for combustion noise monitoring, as explained above. When operating at a given fuel pressure (noted rail pressure in the Fig.), map 12 outputs values of injector opening delay and closing delay, whereas map 14 outputs values of injector opening duration and closing duration. The values of delay and duration are typically stored in time unit in the maps (ms) and are converted in units of °CA based on engine speed (RPM). Maps 12 and 14 contain calibrated values of duration and delay that have been obtained based on the sweeping procedure disclosed above, although other calibrations methods could be used.

[0071 ] For a given injector event at the second cylinder, the start and end of injection are known from the corresponding drive pulse that is available in the ECU. At box 16, the start, resp. end, of opening noise is computed by adding the corresponding delay to the timing of start of injection, resp. timing of end of injection. At box 18 the relevant end of noise timings are computed by adding the opening/closing noise duration to the previously computed start/end of noise.

[0072] Based on these four timing values (namely start of opening noise, end of opening noise, duration of opening noise and duration of closing noise) and on the start/end points of the sensing window, one can determine the amount of closing and opening noise that will be present in the sensing window. This is what is represented at box 20, which outputs values in terms of percentage of noise in the sensing window, namely:

- opening noise part in window

- closing noise part in window [0073] As indicated before, the same knock sensor can be used for detecting pre-ignition and knock, however the noise monitoring is carried out with a different sensing window. But the present approach can be used to determine the impact of injector operational noise in both windows.

[0074] In Fig.5, "window 1 " is indicated for parameters relating to one of the sensing periods, e.g. pre-ignition, and "window 2" is thus used for parameters relating to the knock sensing window.

[0075] The algorithm represented at figure 5 thus allows determining parameters referred to as "opening noise part in window" and "closing noise part in window" that represent the proportion of injector opening noise and injector closing noise that affects the sensing window and will correspondingly increase the measured noise intensity. They can be expressed in any unit; here the arbitrary unit is %.

[0076] It may be noted that when the used frequency filter is not impacted by the injector noise, the computation can be disabled for the corresponding filter and the returned noise value is zero.

[0077] In the other case, the amount/percentage of closing and opening noise is computed for each sensing window and each filter (as shown in

Fig.5) based on:

- injection timing (start/end of pulse)

- engine speed

- fuel injection pressure (rail pressure)

- sensing window position (start/end)

- filter in use.

[0078] As it will appear from chapter B below, the values of "opening noise part in window" and "closing noise part in window" are in this embodiment used as input for the compensating measured noise intensities.

[0079] However, the noise percentages obtained according to the above algorithm can be compounded in the same way as conventionally filtered intensities in order to determine a global injector noise percentage per window: Inj_Noise(W x ) = i=i ki (Closing noise(W x , F ) + Opening noise W x , F ))

[eq.2]

[0080] Where:

- lnj_Noise (Wx) is the global/compound injection noise in the sensing Window x (either pre-ignition or knock);

- Closing noise (Wx, Fi) is the injector closing noise in the Window x for frequency filter I;

- Opening noise (Wx, Fi) is the injector opening noise in the Window x for frequency filter I;

- ki is a weighting factor used to compound intensities from each filter; and

- n: is the number of frequency filters used for monitoring knock or pre-ignition. [0081] B. Noise detection with compensation for injector noise

[0082] The calibration procedure described in chapter A above allows determining noise due to injector opening and/or closing in a second cylinder (other than the cylinder under current monitoring). The impact of injector opening and closing will vary depending on the distance between the injector event and the sensing window.

[0083] Based on the noise part in the sensing window (reflected by previously determined "opening noise part in window" and "closing noise part in window"), fuel pressure and filter in use, it is now possible to predict the noise contribution of injector opening and closing noise that is included in the raw noise intensity as measured by the knock sensor.

[0084] In order to provide noise intensities that are not influenced by injector noise, injector raw intensities may thus be compensated based on the following formula:

Ic( x > Fi = Iraw x > Fi ~ Closing noise(W x , F{)— Opening noise

[eq. 3]

[0085] where - lc(Wx, Fi) is the compensated intensity for window x and frequency filter i;

- lraw(Wx, Fi) is the measured/raw noise intensity (e.g. as determined by the ECU based on sensor signal, by integration over window x) for window x and frequency filter i;

- Closing noise (Wx, Fi) is estimated the injector closing noise in the Window x for frequency filter i;

- Opening noise (Wx, Fi) is the estimated injector opening noise in the Window x for frequency filter i.

[0086] Then, compensated individual knock intensities may be compounded in order to provide a single value of noise intensity per sensing window to the knock control module or ECU.

Icp d (Wx) =∑?=i ki Xl c (W x , F [eq. 4]

[0087] Where

- lcpd(Wx) is the compounded intensity in window x;

- lc(Wx, Fi) is the compensated intensity for window x and frequency filter i;

- ki is a weighting factor used to compound intensities from each filter;

- n: is the number of frequency filters in use.

[0088] It will thus be appreciated that the present method provides a value of noise intensity l cp d that is compensated for injector operational noise occurring when an injector event overlaps with the knock sensing window or pre-ignition sensing window.

[0089] The fact that this value of noise intensity l cp d is not influenced by injector noise allows keeping enabled the knock control during injection timing transiently entering in the knock sensing window, without risking any false detection.

[0090] An embodiment of an algorithm for computing the compound intensity l cp d will now be explained with reference to Fig.6. At the top of Fig. 6 one will recognize 4 input values, namely the opening noise part and closing noise part for each window. As explained above, these values may be expressed as percentage of the total opening or closing noise that is inside the sensing window.

[0091 ] Reference sign 22 designates an opening noise map containing values of opening noise intensity vs. percentage in window, this for several fuel pressures and each frequency filter. Closing noise map 24, in turn, contains values of closing noise intensity vs. percentage in window, this for several fuel pressures and each frequency filter.

[0092] Maps 22 and 24 are obtained during the above described calibration process. Referring back to Fig.3, the fist ramp of the injector closing noise has been graduated in terms of % in window. In Fig.4, the ramp of the injector opening noise has also been graduated in terms of % in window. Hence, the parameters "opening noise part in window" and "closing noise part in window" corresponds to an amount of noise affecting the sensing window, but are only expressed in a relative unit, because the intensity of injector noise is dependent on the fuel pressure. Based on the amounts of noise, estimates of injector noise intensities are looked-up in maps 22 and 24, which depend on fuel pressure.

[0093] The compensated intensity is then calculated in box 26 based on the measured intensities (Raw), for each sensing window, using equation 3.

[0094] Once the compensated intensity has been computed for each filter in use, the compounded intensity l cp d for each sensing window is computed (box 28) by means of equation 4.

The obtained intensity l cp d can then be advantageously used instead of the raw noise intensity in conventional knock or pre-ignition detection algorithms. For example, the obtained intensity l cp d is compared to a threshold which, when exceeded, indicates the presence of knock.