Use of structure-borne sound sensing device for online emission modeling in pressure profile analysis

Abstract

The invention relates to the use of a structure-borne sound sensor for in-line emission modeling in pressure profile analysis, in particular to a control unit for an internal combustion engine, which is designed for real-time monitoring of exhaust gas components of a combustion process, the real-time monitoring includes collecting, by a structure-borne acoustic sensor, acoustic signals for a combustion process in a cylinder of an internal combustion engine, and generating an emission prediction based on the acoustic signals, where the emission prediction describes an exhaust gas composition of the combustion process.

Description

Technical Field

[0001] The invention relates to a control unit, a method, a computer program, an exhaust gas monitoring system and a motor vehicle. Background Art

[0002] In order to monitor pollutant emissions in road traffic, it is desirable to measure the pollutants released during operation of individual motor vehicles as accurately as possible.

[0003] Based on such measurements, the implementation of government regulations for avoiding emissions can be monitored.

[0004] Furthermore, on the basis of such measurements, a control of the combustion process during operation of the motor vehicle can be achieved, so that undesirable pollutants can be avoided.

[0005] However, installing a measuring sensor device to directly measure the pollutant concentration in the exhaust gas is difficult to achieve.

[0006] It is known that in-service pollutant monitoring can be achieved by installing chemical measurement technology to directly measure the exhaust gas components in the exhaust system of a motor vehicle.

[0007] Another known alternative is to measure the pressure during the combustion process in the cylinder by installing a pressure sensor in the cylinder head and subsequently model the combustion process based on the measured pressure.

[0008] The installation of exhaust gas measuring sensor systems in the exhaust system and of pressure sensors in the cylinder head are associated with considerable costs and effort.

[0009] A known method for avoiding the need for sensor systems that are difficult to provide for installation is to evaluate acoustic signals that can be provided in many motor vehicles, for example by knock or structure-borne noise sensors.

[0010] Thus, the publication US 2004 / 134450 A1 describes a method for regulating an internal combustion engine based on thermodynamic modeling of a combustion process. In particular, the publication describes that "for efficient operational regulation of the combustion process in HCCI operation, a modeled combustion process that can be described by internal state variables is formed using process influencing variables" and further describes that "the process influencing variables are then regulated using initial variables of the actual and modeled combustion process".

[0011] In the publication US 2004 / 134450 A1, a structure-borne noise sensor is used to determine the maximum pressure in the cylinder. However, the maximum pressure only provides a less reliable basis for calculating the combustion process occurring during operation.

[0012] Publication US2009 / 112449 A1 describes "a method for controlling an internal combustion engine, in particular a diesel internal combustion engine, in which at least one variable is formed in a cylinder-specific manner, each being characterized by the course of combustion in a sub-combustion chamber, and the control of cylinder-specific fuel injection parameters depends on at least one variable characterizing the course of combustion to have an impact."

[0013] Publication US2009 / 112449 A1 in particular describes that "characteristic variables" can be determined based on the signals of knock sensors.

[0014] However, for the precise determination of emissions, it is advantageous to make the most of the information content of the acoustic signals as comprehensively as possible. Summary of the Invention

[0015] The object of the present invention is to provide a method for emissions monitoring that at least partially overcomes the above-mentioned drawbacks.

[0016] This object is solved by a control unit for an internal combustion engine, a method for real-time monitoring of the exhaust gas components of a combustion process in an internal combustion engine, a computer program for real-time monitoring of the exhaust gas components of a combustion process in an internal combustion engine, an exhaust gas monitoring system for real-time monitoring of the exhaust gas components of a combustion process, and a motor vehicle.

[0017] Further advantageous embodiments of the present invention result from the following description of the preferred embodiments of the present invention. Brief Description of the Drawings

[0018] Embodiments of the present invention are described by way of example and with reference to the drawings, wherein:

[0019] Figure 1 An embodiment of a method for directly generating an emissions prediction from an acoustic signal is schematically shown; and

[0020] Figure 2 An embodiment of a method for generating an emissions prediction from a pressure curve, wherein the pressure curve is determined from an acoustic signal, is schematically shown; and

[0021] Figure 3 An embodiment of a method for controlling a combustion process based on an emissions prediction is schematically shown; and

[0022] Figure 4 An embodiment of a method for generating training data is schematically shown; and

[0023] Figure 5 The composition of the training data in the first embodiment is schematically shown; and

[0024] Figure 6 The composition of the training data in the second embodiment is schematically shown; and

[0025] Figure 7 Schematically shows the composition of the training data in the third embodiment; and

[0026] Figure 8 Shows an illustration of an internal combustion engine having a structure-borne sound sensor for performing the method according to the disclosure; and

[0027] Figure 9 Shows an illustration of a motor vehicle according to the present disclosure; and

[0028] Figure 10 Shows a symbolic view of the pressure curve. DETAILED DESCRIPTION

[0029] Before a more precise description of the embodiments of the present disclosure is a general comment.

[0030] A control unit for an internal combustion engine is configured for real-time monitoring of the exhaust gas components of a combustion process, wherein the real-time monitoring includes: collecting (Erheben) acoustic signals for the combustion process in the cylinder block of the internal combustion engine by a structure-borne sound sensor, and

[0031] generating an emission prediction based on the acoustic signals, wherein the emission prediction describes the exhaust gas components of the combustion process.

[0032] The control unit of the internal combustion engine can be a control device of a vehicle, such as an electronic control unit (ECU) or an electronic control module (ECM).

[0033] According to the present disclosure, the control unit can be a control device configured to control a vehicle or a component of a vehicle. The control unit is configured to receive and process sensor data, in particular data from a structure-borne sound sensor. The control unit is also configured to acquire and process characteristic parameters of the internal combustion engine, such as the internal pressure in the cylinder block, the rotational speed or the temperature.

[0034] An internal combustion engine in the sense of the present disclosure is a prime mover that converts chemical energy into force or mechanical energy by combustion and is generally used to drive a motor vehicle. In particular, the internal combustion engine is a piston engine, which is configured as a gasoline engine (spark ignition) or a diesel engine (self-ignition).

[0035] Real-time monitoring means monitoring during the operation of the internal combustion engine, wherein the method steps provided for monitoring do not require any external post-processing and are available to the control unit during operation such that the immediate continuous operation of the internal combustion engine can be controlled based on the monitoring.

[0036] Exhaust gas components are the gaseous and solid chemical components, in particular molecules, of the exhaust gas generated by the operation of the internal combustion engine.

[0037] The combustion process is referred to as the combustion process resulting from ignition during the working cycle of the cylinder block of an internal combustion engine. The combustion process can also be referred to as the working process or working cycle.

[0038] An acoustic signal is a digital representation of a physical sound in a digitally processable form. An acoustic signal is in particular an acoustic signal in the frequency domain, which is transformed from an acoustic signal in the time domain into the frequency domain by means of a Fourier transform in a scanning window (Abtastfenster).

[0039] The scanning window can be a pre-determined value, for example 30 ms. The scanning window can also be calculated based on the rhythm or clock frequency of the control unit, the structure-borne sound sensor or the clock frequency of the internal combustion engine or the frequency of the working cycle of the cylinder block.

[0040] According to the present disclosure, this conversion can be performed by a structure-borne sound sensor or a control unit. Accordingly, the acoustic signal can be acquired by the control unit as a signal in the time domain or a signal in the frequency domain. The acoustic signal can be processed or be processed by filtering, signal decomposition or component analysis.

[0041] The term "collection" means that the control unit records, acquires or receives the acoustic signal of the structure-borne sound sensor, wherein the acoustic signal is based on the measurement of the sound in the scanning window by the structure-borne sound sensor.

[0042] According to the present disclosure, the structure-borne sound sensor is a known sensor which is configured to acquire vibrations in solids, in particular high-frequency vibrations, and convert them into an electrical signal or a digital signal.

[0043] The structure-borne sound sensor is for example configured as a piezoelectric acceleration sensor and maps a frequency domain larger than the frequency domain of typical knock combustion as a broadband probe.

[0044] Emission prediction is a model that describes the quantity of exhaust gas products generated during the combustion process. The exhaust gas products include chemical compounds including at least one of nitrogen (N 2 ), carbon dioxide (CO 2 ), water (H 2 O), carbon monoxide (CO), nitrogen oxides (NOx), ammonia (NH 3 ), methane (CH 4 ), formaldehyde (HCHO), nitrous oxide (N 2 O) and hydrocarbons. The exhaust gas products can also include solids, such as soot particles.

[0045] In other words, emission prediction is the prediction of emission products at the end of the combustion process in an internal combustion engine.

[0046] Emission prediction is based on the real-time monitoring of the combustion process.

[0047] There are embodiments in which the acoustic signal is an average acoustic signal and / or the combustion process is an average combustion process.

[0048] The combustion process can be the sole combustion process. The combustion process can also be an average combustion process. An average combustion process is the average superposition of all combustion processes occurring during a scan window. For example, the average combustion process is the average superposition of all combustion processes during the ignition sequences of all cylinders of an internal combustion engine.

[0049] Alternatively, the average combustion process is the average superposition of all combustion processes over multiple revolutions of an internal combustion engine.

[0050] The average combustion process can be generated from an average acoustic signal. The average acoustic signal can be the physical superposition of all acoustic signals generated by all combustion processes in an internal combustion engine within a scan window.

[0051] Emission prediction can be described as the quantity of exhaust gas products during the combustion process, e.g., described by mass.

[0052] Emission prediction can also describe the quantity of exhaust gas products during the average combustion process, e.g., described by mass. However, according to the present disclosure, emission prediction can also describe the exhaust gas products as a total amount during a monitoring time period. Alternatively, emission prediction can present the exhaust gas products as the concentration of exhaust gas products in the total exhaust gas volume. According to the present disclosure, other suitable presentations are possible.

[0053] A control unit according to the present disclosure can directly control the combustion process based on such real-time monitoring, thereby achieving low pollutant emissions considering current engine and operating parameters. Alternatively, the control unit can control the combustion process based on such real-time monitoring, thereby achieving low component wear considering current engine and operating parameters.

[0054] Alternatively, the control unit can transmit the real-time calculated emission values to an external receiver. For example, according to the emission values calculated according to the present disclosure, national or private control agencies can monitor the emissions and / or pollutants discharged by motor vehicles.

[0055] In the context of aftertreatment, a tailpipe emission prediction can be created based on the emission prediction. Different from the emission prediction, the tailpipe emission prediction describes the emissions at the moment when the exhaust gas is released into the atmosphere, i.e., after exhaust gas aftertreatment.

[0056] Exhaust gas aftertreatment is performed in a conventional manner through filters, optionally soot particle filters, catalytic converters, etc.

[0057] By generating an emissions prediction from the acoustic signals collected by the structure-borne sound sensor, a large number of sensors that would otherwise be necessary can be saved. For example, the need to install chemical measurement sensing devices, such as carbon monoxide sensors, hydrocarbon sensors, or particulate sensors, may be eliminated.

[0058] There are embodiments in which the control unit is further configured to perform control of the combustion process in the internal combustion engine based on the emissions prediction.

[0059] The control can be carried out by manipulating existing engine regulating devices.

[0060] There are embodiments in which the control unit is further configured to control the combustion process such that low emissions are achieved.

[0061] This control can be achieved, for example, by a delay adjustment of the ignition timing carried out by the control unit. For example, a delay adjustment of the ignition timing can achieve a reduction in nitrogen oxide emissions, but at the cost of a slightly increased fuel consumption.

[0062] There are embodiments in which the control unit is further configured to control the combustion process such that low component wear is achieved.

[0063] The components can be, for example, exhaust gas aftertreatment components, such as known catalytic converters, etc.

[0064] The exhaust gas matrix, or the composition of the raw exhaust gas generated from the combustion process, directly affects the conversion behavior of the installed exhaust gas aftertreatment components.

[0065] The behavior of the exhaust gas aftertreatment components is known or described in a model. Therefore, the exhaust gas aftertreatment components can be controlled via an emissions model and protected from the formation of undesired exhaust gas products, such as N 2 O or NH 3 .

[0066] This control can be carried out within the scope of an operating strategy that is created based on the emissions prediction.

[0067] There are embodiments in which controlling the combustion process includes adjusting at least one of the following regulating devices: throttle valve, camshaft, exhaust gas recirculation valve, wastegate, pre-throttle valve, charge motion flap, intake manifold length, exhaust throttle valve, injection duration, rail pressure, ignition timing, injection start, variable turbine geometry position, injection pattern.

[0068] There are embodiments in which generating the emissions prediction includes: determining a pressure curve from the acoustic signal and analyzing the pressure curve to generate the emissions prediction.

[0069] In a conventional sense, a pressure curve maps the value of the internal pressure of the cylinder during the combustion process as a function of the crank angle.

[0070] It is possible, but not necessary, to convert the pressure curve to a time-dependent view.

[0071] The pressure curve in the combustion chamber may also be referred to as the pressure curve in the combustion chamber.

[0072] The pressure curve is determined with the aid of a pressure curve model. To this end, the acoustic signal is first mapped onto a representative parameter set. The representative parameter set is then mapped onto a set of sample curves, the most likely of which represents the actual pressure curve in the cylinder.

[0073] Representative parameter sets include scalar values ​​that are characteristic and linearly independent of each other.

[0074] The parameter group also presets how the sample curves are calculated so that they map onto the real pressure curve.

[0075] In one specific embodiment, in particular as a pressure curve model, a mathematical mapping of the acoustic signal directly onto the pressure curve in the combustion chamber can be provided.

[0076] In one embodiment variant, the representative parameter set is derived via a physical model.

[0077] One embodiment of a physical model is a simple single-mass oscillator. The single-mass oscillator model describes the system or transmission behavior between pressure and acoustic signals. Thus, the pressure curve can be presented as a function of a basic or sample curve. To this end, the pressure curve is first mapped onto a set of parameters, for example onto characteristic eigenvalues ​​of the physical model.

[0078] The basic functions can then be calculated from the parameter sets in order to map them onto the measured pressure curve. The parameter sets can be found, for example, via the method of least squares.

[0079] An alternative is to map the parameter sets via machine learning and thereby extend the validity range over the entire performance map.

[0080] By using machine learning, the mostly only limited validity of the parameter sets found by the least squares method, which originates from the system behavior that varies with the engine speed, can be avoided.

[0081] In another embodiment variant, the pressure curve model is a machine learning (ML) model. In particular, the ML model is a pre-trained ML model and in particular includes a back-propagation algorithm.

[0082] The calibration of the model for predicting the pressure curve is based on the pressure curve measured via the graduation measurement technique. The calibration of the model finds a parameter set that maps to the measured pressure curve in the combustion chamber based on the solid-borne sound signal in the frequency domain and based on the sample curve of the sound signal.

[0083] By determining the pressure curve during the combustion process based on the sound signal collected by the solid-borne sound sensor, the pressure curve does not have to be determined by means of a pressure sensor, so its installation in the cylinder block can be saved. In addition, the only solid-borne sound sensor for the internal combustion engine is sufficient to determine the pressure curve, while a pressure sensor for each cylinder block of the internal combustion engine is required to determine the pressure curve by means of a pressure sensor. According to the present disclosure, more than one solid-borne sound sensor can be provided.

[0084] The pressure curve analysis (DVA) takes into account in particular the engine adjustment variables that describe the current state of the internal combustion engine. These adjustment variables can be preset or calculated by the engine control device within the scope of its normal operation.

[0085] Regardless of the aging state of the engine, the pressure curve is used as the most important variable for deriving emissions formation.

[0086] DVA is a thermodynamic evaluation of the combustion process, for example in the case of using a reaction kinetics network or applying the laws of thermodynamics, and calculates the energy and temperature curves and the mass conversion in the combustion chamber based on the pressure curve.

[0087] By using a machine learning model for determining the pressure curve, the use of numerical models is avoided. The numerical calculation of thermodynamic systems is usually associated with high requirements for the computer infrastructure used or high time consumption. The high requirements for the computer infrastructure used and the high time consumption impede the use in motor vehicles and / or the real-time monitoring of thermodynamic processes.

[0088] In addition, in the case of a more complex case / test space, a cluster can also be applied if necessary. This can also be presented via an unmonitored ML method (k-means, self-organizing map, etc.) so that a dedicated, cluster-specific network can be trained immediately afterwards.

[0089] The pressure curve, the mass conversion and the temperature curve form the decisive basis for calculating emissions formation.

[0090] The generation of emissions prediction is based on emissions modeling.

[0091] Emissions modeling obtains some thermodynamic boundary conditions that are not measurable or not directly measured from DVA, in particular the pressure curve, the mass conversion and the temperature curve.

[0092] Based on these boundary conditions, the change in the amount of relevant combustion products in the combustion chamber is calculated using a known reaction mechanism (e.g., GRI 3.0). For example, this can be based on a numerical solution of the relevant reaction kinetics.

[0093] In this way, the accuracy and reliability of emission predictions can be improved.

[0094] There is an embodiment in which the determination of the pressure curve from the acoustic signal is performed based on a machine learning model that maps the acoustic signal to the pressure curve.

[0095] In another implementation variant of the ML model, as described above, a mathematical mapping of the acoustic signal directly to the pressure curve in the combustion chamber can be set here as the pressure curve model.

[0096] There is an embodiment in which the machine learning model is pre-trained using training data, where the training data includes a plurality of experimentally determined pressure curves with associated acoustic signals.

[0097] The training data is determined during various experimental combustion processes, during which input and output values are collected sensorially. For example, an internal combustion engine in a test bench can be subjected to a test run of a preset length, e.g., 1 hour or 10 hours, and the corresponding relevant values can be collected.

[0098] Preferably, the training data of multiple different test runs are combined with each other to generate the ML model, where the conditions are differently selected in each test run, such as the parameters of the internal combustion engine or the engine adjustment variables.

[0099] There is an embodiment in which the determination of the pressure curve from the acoustic signal includes mapping the acoustic signal to the pressure curve by means of a physical model.

[0100] In particular, the physical model can map the acoustic signal to a sample pressure curve.

[0101] Alternatively, the model can be a machine learning model that maps the acoustic signal to a sample pressure curve.

[0102] Alternatively, the model can again be a pure machine learning model that determines the pressure curve from the acoustic signal based on the training data without mapping the acoustic signal to a sample pressure curve.

[0103] The set of physical parameters can include expected chemical compositions, expected temperature developments, expected entropy, or other parameters.

[0104] There is an embodiment in which the analysis of the pressure curve is performed based on a second machine learning model that maps the pressure curve to an emission prediction.

[0105] The mapping of the pressure curve onto the emission prediction can be accomplished by applying a trained model that acts as a black box and has the pressure curve as the input value and the associated emission prediction (which is determined based on the training data) as the output value. This trained model can be referred to as the second machine learning model.

[0106] In one embodiment, a feedforward network can be used for this purpose, which is configured to map the corresponding reaction kinetics (e.g., the known reaction kinetics GRI3.0). The calibration of this feedforward network can be achieved based on measurement data or previous numerical calculations, e.g., a numerical solution of the relevant reaction kinetics by means of the backpropagation algorithm.

[0107] By using a machine learning model for generating emission predictions, the use of numerical models is avoided. Numerical calculations of thermodynamic systems are usually associated with high requirements for the computer infrastructure used or high time consumption. The high requirements for the computer infrastructure used and the high time consumption hinder the use in motor vehicles and / or the real-time monitoring of thermodynamic processes.

[0108] There is an embodiment in which the second machine learning model is pre-trained using training data, where the training data includes a plurality of experimentally determined emission data with associated pressure curves.

[0109] A method for real-time monitoring of the exhaust gas components of a combustion process in an internal combustion engine, comprising:

[0110] Collecting an acoustic signal by a solid-borne sound sensor for a combustion process in a cylinder block of the internal combustion engine, and

[0111] Generating an emission prediction based on the acoustic signal,

[0112] where the emission prediction describes the exhaust gas components of the combustion process.

[0113] The method according to the present disclosure can be executed by the control unit of a conventional motor vehicle, provided that the control unit has the required computing power for this purpose.

[0114] A computer program for real-time monitoring of the exhaust gas components of a combustion process in an internal combustion engine, which, when implemented by a computer, causes the computer to implement the following method:

[0115] Collecting an acoustic signal by a solid-borne sound sensor for a combustion process in a cylinder block of the internal combustion engine, and

[0116] Generating an emission prediction based on the acoustic signal, where the emission prediction describes the exhaust gas components of the combustion process.

[0117] The computer program can be digitally introduced into the control system of a motor vehicle such that the control system is configured to implement the method according to the present disclosure.

[0118] An exhaust gas monitoring system for real-time monitoring of exhaust gas components in a combustion process includes a control unit and a solid acoustic sensor according to the present disclosure.

[0119] A motor vehicle according to the present disclosure includes an internal combustion engine and an exhaust gas monitoring system according to the present disclosure.

[0120] Returning to the drawings, Figure 1 Schematically shown is a method for generating an emission prediction according to the present disclosure, which can be executed by a control unit according to the present disclosure. The method includes a collection step S11 and an analysis step S12. In the collection step S11, an acoustic signal is collected by the solid acoustic sensor.

[0121] The acoustic signal is an electronically machine-readable signal that can be processed in digital form by an electronic data processing device (e.g., by a computer).

[0122] To collect the acoustic signal, first the solid acoustic sensor measures the raw signal. Here, the raw signal is recorded during the operation of the internal combustion engine such that the raw signal maps at least one combustion process of the internal combustion engine.

[0123] The raw signal can be provided for use by the solid acoustic sensor in the time domain or the frequency domain. For example, the raw signal in the time domain is a signal that correlates the volume value of the sound in decibels measured by the solid acoustic sensor with a time value. Instead of the volume value, the energy value of the sound can be correlated with the time value.

[0124] If the raw signal is an acoustic signal in the time domain, the raw signal is transformed into an acoustic signal in the frequency domain. For this purpose, a scanning window is defined. The scanning window is a time window of the measured acoustic signal that should be transformed into the frequency domain. The scanning window can be, for example, 30 ms, however longer or shorter scanning windows are also possible according to the present invention. For example, the scanning window can include the time length of a working cycle.

[0125] Then, by means of a known method, such as the fast Fourier transform (FFT), the raw signal in the time domain is transformed into the frequency domain. The raw signal transformed into the frequency domain can be called the frequency raw signal.

[0126] The collection step S11 further includes signal conditioning (Signalaufbereitung). The signal conditioning processes the frequency raw signal (e.g., by filtering, signal decomposition, component analysis, etc.) to isolate the relevant frequency domain.

[0127] The signal conditioning identifies and isolates the signal portion relevant to the reconstruction of the cylinder pressure curve and thereby improves the model accuracy.

[0128] Filtering removes the interfering signal component from the original signal. For example, a low-pass filter is used to remove, more precisely, the frequency range that is relevant for knock recognition and irrelevant for the pressure curve.

[0129] According to known principal component analysis (PCA), the signal decomposition is performed via principal component analysis. Thereby, a multi-dimensional signal can be decomposed into multiple linear signal components. The relevance or lack of interest of the signal can be checked via the variance.

[0130] In analysis step S12, the analysis can also be referred to as processing the frequency raw signal of the acoustic signal to generate an emissions prediction. In one embodiment, the emissions prediction can be based on a data-driven (black box) calculation of data formed from the frequency raw signal directly from the structure-borne sound sensor. The data-driven calculation can be implemented by a statistical model that maps the acoustic signal to a set of previously experimentally determined emissions data.

[0131] The method is applied in real time during the operation of the internal combustion engine. The control unit of the motor vehicle can be used to execute the method. The control unit can be a digital data processing device maintained in the motor vehicle, or multiple digital data processing devices. The control unit can be a monitoring unit of the vehicle components, which can be configured to execute the method. The control unit can be, for example, an engine control unit or an equivalent unit. In particular, the control unit can be an engine control device.

[0132] By generating an emissions prediction based on the acoustic signal by means of the method according to the present disclosure, the necessity of maintaining a chemical sensing device to measure the emissions during the operation of the internal combustion engine is eliminated.

[0133] The emissions prediction generated in analysis step S12 is further used at the end of the method.

[0134] Figure 2 Schematically shown is a method according to the present disclosure for generating an emissions prediction based on an acoustic signal in a second embodiment. The method includes a collection step S21, a pressure curve calculation S22, and an emissions calculation step S23. The collection step S21 can be the same as Figure 1 the collection step S11 in

[0135] The emissions prediction generated in the emissions calculation step S23 is further used at the end of the method.

[0136] Immediately following the collection step S21, in this embodiment, different from the method titled in Figure 1 first, the pressure curve is determined from the acoustic signal in the pressure curve calculation S22. The determination of the pressure curve is performed based on a model.

[0137] The model may be a physical model, as described above. In particular, as described above, the physical model may map the acoustic signal onto a sample pressure curve.

[0138] Alternatively, the model may be a machine learning model that maps acoustic signals onto sample pressure curves.

[0139] Alternatively, the model can again be a pure machine learning model which determines the pressure curve from the acoustic signal based on training data without mapping the acoustic signal onto a sample pressure curve.

[0140] The model derives the cylinder pressure from the acoustic signal which is collected in a collection step S21 and processed if necessary.

[0141] The functional relationship between cylinder pressure and time is the pressure curve. The pressure curve is determined based on the acoustic signal.

[0142] The pressure curve is determined with the aid of a pressure curve model. To this end, the acoustic signal is first mapped onto a representative parameter set. Next, the representative parameter set is mapped onto a set of sample curves which most likely represent the actual pressure curve in the cylinder.

[0143] The parameter set includes eigenvalues ​​and scalar values ​​that are linearly independent of each other.

[0144] In the emission calculation step S23 , an analysis of the pressure curve determined in the pressure curve calculation S22 is performed (pressure curve analysis, DVA).

[0145] In a first embodiment, the temperature curve and mass conversion of the fuel in the combustion chamber can be first derived from the determined pressure curve. Based on the temperature curve and mass conversion, the reaction kinetics of the relevant chemical reactions in the combustion chamber are numerically calculated, and the residual emissions at the end of the combustion process are thus calculated.

[0146] In a second embodiment, the ML model can also be used to replace the numerical calculation of the reaction kinetics. Then, the numerical calculation of the reaction kinetics provides a training database related to training the ML model.

[0147] In a third embodiment, characteristic parameters are calculated from a determined pressure curve (e.g. maximum cylinder pressure, maximum cylinder pressure gradient, etc.) and correlated with the measured emission values. With the aid of an ML model and a training database, the emission formation is thus calculated directly from the modeled cylinder pressure curve. A thermodynamic evaluation of the pressure curve can be omitted.

[0148] Figure 3 FIG. 4 shows a control of a combustion process according to the present disclosure based on emission prediction. In this embodiment, first, in steps S31, S32 and S33, based on Figure 2 The method described in ( Figure 2S21, S22, S23) in generate an emission prediction. In control step S34, the combustion process is then controlled based on the determined emission prediction.

[0149] The control is performed by adjusting engine adjustment variables via a control unit or an engine control unit. In particular, the combustion process is adjusted by means of at least one of the following adjustment devices:

[0150] Throttle valve, camshaft, exhaust gas recirculation valve, wastegate, pre-throttle valve, charge motion flap, intake manifold length, exhaust throttle valve, injection duration, rail pressure, ignition timing, injection start.

[0151] In an alternative embodiment, the control is performed such that as low a pollutant emission as possible is generated corresponding to the emission prediction. Since the emission prediction is performed in real time during the operation of the internal combustion engine, the control can also be performed in real time, so that the adjustment devices are adjusted at each operating point to achieve low pollutant emissions.

[0152] The control can also include shutting down and / or preventing the operation of the internal combustion engine when a determined pollutant limit value is exceeded. The control unit according to the present disclosure can be configured to receive and further process a data signal that transmits the pollutant limit value from an external source (e.g., from an external transmitter or transducer) to the control unit. Alternatively, the control unit can include a data storage unit in which the pollutant limit value is stored or can be stored, and the control is performed based on the pollutant limit value.

[0153] In another alternative embodiment, the control of the emission prediction is such that as low a component wear as possible is achieved. For example, if the generation of combustion products indicates an operating mode of a specific damaged component of the internal combustion engine, the aforementioned adjustment devices can be adjusted to avoid or reduce the operating mode of the specific damaged component, thereby avoiding further wear.

[0154] Instead of method steps S31, S32, and S33, the emission prediction generated in step S33 can be generated by means of Figure 1 method steps S11 and S12 in.

[0155] Figure 4 shows a method for generating an ML model, which can be used for Figure 1 method step S12 in, Figure 2 step S22 and S23 in or Figure 2 step S32 and S33 in.

[0156] Note that this view relates to the generation of the ML model and does not relate to the execution of the method according to the present method or the working principle of the device according to the present disclosure. Instead, the method and the device operate the ML model, which is generated in the manner explained below.

[0157] The method described herein is an illustrative example of the training process for an ML model.

[0158] In a first step S41, a set of training data is generated that associates input values with output values respectively. The identity of the input values and output values is described below with reference to Figure 5 and 6 7.

[0159] Then, in step S42, the set of training data is evaluated by corresponding ML software that generates a corresponding ML model according to known methods. Then, the ML model thus generated maps one or more input values (corresponding to the input values of a previously determined set of training data) to one or more output values that can be expected during the experimental training phase.

[0160] For example, in the application of step S12 according to Figure 1 , an acoustic signal can be mapped to an emission prediction.

[0161] Alternatively, in the application of step S22 according to Figure 2 , an acoustic signal can be mapped to a pressure curve.

[0162] Alternatively, in the application of step S23 according to Figure 2 , reaction kinetics can be learned in the ML model.

[0163] Additionally, in the application of step S23 according to Figure 2 , a pressure curve can be mapped to an emission prediction.

[0164] Figure 5 shows a schematic diagram of a first set of training data 100 used in the ML model for step S12 in Figure 1 .

[0165] In this implementation variant, the set of training data includes acoustic signals 110 that are respectively individually associated with emission measurements 120. The acoustic signals 110 are acquired and processed by a structure-borne sound sensor during a test run, where, as described above, the conditioning is processed to isolate the relevant frequency domain.

[0166] The emission data can be determined from the engine exhaust gas at the exhaust pipe by means of chemical methods. In this example, the acoustic signals 110 correspond to the input values. The emission measurements 120 correspond to the output values.

[0167] Figure 6 shows a schematic diagram of a second set of training data 200 used in the ML model for step S22 in Figure 2 .

[0168] In this implementation variant, the training data set includes acoustic signals 210, which are individually associated with pressure measurement results 220, respectively. The acoustic signals 210 are acquired and processed by a structure-borne sound sensor during a test run, where, as described above, the conditioning is processed to isolate the relevant frequency domain.

[0169] The pressure measurement results presenting the pressure curve can be determined by means of a pressure measuring device introduced into the cylinder block. The pressure curve is associated with the internal pressure of the cylinder block and the measurement time point (see below Figure 10 ). In this example, the acoustic signals 210 correspond to the input values. The pressure measurement results 220 correspond to the output values.

[0170] Figure 7 Shows a schematic diagram of the third training data set 300 used in the ML model for step S23 in Figure 2 .

[0171] In this implementation variant, the training data set includes pressure measurement results 310, which are individually associated with emission measurement results 320, respectively. The pressure measurement results 310 are determined by means of a pressure measuring device introduced into the cylinder block during a test run. The collection of the current third training data set can be carried out simultaneously with the second training data set. The pressure measurement results presenting the pressure curve can be determined by means of a pressure measuring device introduced into the cylinder block. The pressure curve is associated with the internal pressure of the cylinder block and the measurement time point (see below Figure 10 ). In this example, the pressure measurement results 310 correspond to the input values. The emission measurement results 320 correspond to the output values.

[0172] Figure 8 Shows a structure-borne sound sensor 10, which is used to collect acoustic signals in the sense of the present disclosure.

[0173] The structure-borne sound sensor is installed at the crankcase of the internal combustion engine 20 so that the sound generated during the combustion process can be measured by the structure-borne sound sensor. This installation corresponds to the traditional installation position of the structure-borne sound sensor in known motor vehicles.

[0174] Therefore, the present disclosure can be used in traditional motor vehicles.

[0175] Figure 9 Shows a motor vehicle equipped with a control unit according to the present disclosure. By using the method according to the present disclosure, such a motor vehicle can perform real-time monitoring of emissions generated during operation. According to this real-time monitoring, the control unit can control the combustion process considering the current engine parameters and operating parameters, thereby achieving lower pollutant emissions. Alternatively, the control unit can control the combustion process according to this real-time monitoring, thereby achieving low component wear considering the current engine parameters and operating parameters.

[0176] Alternatively, the control unit can also transmit the emission values calculated in real time to an external receiver. For example, according to the emission values calculated according to the present disclosure, national or private control agencies can monitor the emissions and / or pollutants emitted by motor vehicles.

[0177] Figure 10 The symbolic visualization of the pressure curve 50 is shown, which can correspond to the pressure measurement result 220 or the pressure measurement result 310. The pressure curve 50 maps the values of the internal pressure (axis A1) of the cylinder block with respect to the crank angle (axis A2).

[0178] The pressure curve 50 is used as Figure 2 the output value of the model (or ML model) in step S22 in Figure 2 where it is determined by the acoustic signal and used as

[0179] List of reference symbols

[0180] 10 Solid acoustic sensor

[0181] 20 Internal combustion engine

[0182] 30 Motor vehicle

[0183] 50 Pressure curve

[0184] 100 Training data

[0185] 110 Acoustic signal

[0186] 120 Emission measurement

[0187] 200 Training data

[0188] 210 Acoustic signal

[0189] 220 Pressure measurement result

[0190] 300 Training data

[0191] 310 Pressure measurement result

[0192] 320 Emission measurement result.

Claims

1. A control unit for an internal combustion engine (20), configured for real-time monitoring of exhaust gas components of a combustion process, wherein the real-time monitoring comprises: Acoustic signals (50) are collected by a structure-borne sound sensor (10) for a combustion process in a cylinder of the internal combustion engine (20), and generating an emissions prediction based on the acoustic signal, Therein, the emission prediction describes the exhaust gas composition of the combustion process.

2. The control unit according to claim 1, wherein: The control unit is further configured to perform control of a combustion process in the internal combustion engine (20) based on the emission prediction.

3. The control unit according to claim 2, wherein: The control unit is further configured to control the combustion process in such a way that low emission emissions are achieved.

4. The control unit according to claim 2, wherein: The control unit is also designed to control the combustion process in such a way that low component wear is achieved.

5. A control unit according to any one of claims 3 or 4, wherein: The control of the combustion process comprises adjusting at least one of the following regulating devices: Throttle, camshaft, exhaust gas recirculation valve, wastegate, pre-throttle, charge movement flap, intake manifold length, exhaust throttle, injection duration, rail pressure, ignition timing, injection start, variable turbine geometry supercharger position, injection pattern.

6. A control unit according to any one of claims 1 to 5, wherein: The generation of the emission forecast includes: determining a pressure curve (50) from the acoustic signal, and The pressure curve (50) is analyzed to generate the emission prediction, wherein the pressure curve (50) is a pressure curve (50) in the combustion chamber.

7. The control unit according to claim 6, wherein: Determining the pressure curve (50) from the acoustic signal is performed based on a machine learning model that maps the acoustic signal onto the pressure curve.

8. A control unit according to any one of claims 1 to 7, wherein: The acoustic signal is an average acoustic signal and / or the combustion process is an average combustion process.

9. A control unit according to any one of claims 6 to 8, wherein: Determining the pressure curve (50) from the acoustic signal comprises mapping the acoustic signal onto the pressure curve (50) by means of a physical model.

10. A control unit according to any one of claims 6 to 9, wherein: Analysis of the pressure curve (50) is performed based on a second machine learning model that maps the pressure curve onto the emissions prediction.

11. The control unit according to claim 10, wherein: The second machine learning model is pre-trained using training data, wherein the training data includes a plurality of experimentally determined emission data with associated pressure curves.

12. A method for real-time monitoring of exhaust gas components of a combustion process in an internal combustion engine (20), wherein the method comprises: Acoustic signals are collected by a structure-borne sound sensor (10) for a combustion process in a cylinder of the internal combustion engine (20); and generating an emissions prediction based on the acoustic signal, Therein, the emission prediction describes the exhaust gas composition of the combustion process.

13. A computer program for real-time monitoring of exhaust gas components of a combustion process in an internal combustion engine (20), which, when implemented by a computer, causes the computer to perform the following method: Acoustic signals are collected by a structure-borne sound sensor (10) for a combustion process in a cylinder of the internal combustion engine (20); and generating an emissions prediction based on the acoustic signal, in, The emissions prediction describes the exhaust gas composition of the combustion process.

14. An exhaust gas monitoring system for real-time monitoring of exhaust gas components in a combustion process, comprising: A control unit and a structure-borne sound sensor (10) according to any one of claims 1 to 11.

15. A motor vehicle (30), comprising An internal combustion engine (20) and an exhaust gas monitoring system according to claim 14.

Images