Methods for training a machine learning model and methods for diagnosing particle size in an exhaust gas stream

Abstract

The invention relates to a method for training a machine learning model to diagnose particle size in an exhaust gas stream. The method comprises the steps of: providing (S1) a labeled training dataset containing a plurality of sensor signals from a laser-induced incandescence sensor; training (S2) the machine learning model based on the training dataset; and providing (S3) the trained machine learning model.

Description

[0001] The invention relates to a method and a system for training a machine learning model for diagnosing a particle size in an exhaust stream. The invention relates to a method and a system for diagnosing a particle size in an exhaust stream. The invention relates to an evaluation and / or computing device for a laser-induced incandescence-based soot particle sensor. The invention relates to a computer program with program code and a computer-readable data carrier with program code of a computer program. State of the art

[0002] On-board diagnostics for the condition monitoring and / or diagnosis of a vehicle's particulate filter are increasingly being used in vehicles, partly due to EU-wide emissions legislation. To check the particulate filter for proper functioning, it is necessary to count the particles downstream of the filter in the exhaust system. Furthermore, estimating a particle size is also beneficial to improve the accuracy of the diagnostic result.

[0003] The principle of laser-induced incandescence (LII) for the detection of nanometer-sized particles in air is known in the literature and is applied, for example, on a laboratory scale for the characterization of combustion processes.

[0004] In addition to counting particles in the (exhaust) air, the estimation of the size of small particles (range 30 nm to 150 nm) based on LII sensor signals is known. Particle size can be determined, for example, by evaluating the duration of the signal, the width of the signal waveform, the amplitude of the signal, and / or the maximum rise or gradient of the signal, etc.

[0005] Although approaches for diagnosing particle number and / or particle size in an exhaust gas stream downstream of a particulate filter of a motor vehicle are already known, there is still a need for improvement.

[0006] DE 10 2018 221 701 A1 relates to a method for processing a signal of a particle sensor that operates according to the principle of laser-induced incandescence, an arrangement for processing such a signal, and a particle sensor.

[0007] US 2017 / 0 016 820 A1 relates to a particle detection device for measuring properties such as number, size or mass concentration of particles contained in the atmosphere or in the air in a clean room.

[0008] The invention is based on the object of providing an improved method for diagnosing a particle number and / or a particle size in an exhaust gas stream, in particular downstream of a particle filter, of a motor vehicle.

[0009] The problem is solved by a method for training a machine learning model according to the features of patent claim 1. The problem is solved by a method for diagnosing a particle number and / or a particle size in an exhaust gas stream using a trained machine learning model according to the features of patent claim 8. The problem is solved by a system for training a machine learning model according to the features of patent claim 11. The problem is solved by a system for diagnosing a particle number and / or a particle size in an exhaust gas stream using a trained machine learning model according to the features of patent claim 12. Disclosure of the invention

[0010] According to a first aspect, a method for training a machine learning model for diagnosing a particle number and / or a particle size in an exhaust gas stream, in particular downstream of a particulate filter, preferably of a motor vehicle, is provided. The method comprises the steps: - Providing a labeled training data set comprising a plurality of sensor signals of a laser-induced incandescence sensor; - Training the machine learning model based on the training dataset; - Deploy the trained machine learning model.

[0011] It is understood that the steps according to the invention, as well as other optional steps, do not necessarily have to be performed in the order shown, but can also be performed in a different order. Furthermore, additional intermediate steps can be provided. The individual steps can also comprise one or more substeps without thereby departing from the scope of the method according to the invention.

[0012] The quality of particle size estimation based on supervised ML methods is determined by labeled training data obtained from soot generator measurements. The soot generator preferably generates a specific, narrow, and monodisperse particle size distribution with a mean and a small standard deviation with respect to the particle size diameter. Even if particles of a specific size cannot be generated precisely, a large number of measurements is sufficient to provide the supervised ML method with a sufficient amount of high-quality training data.

[0013] This training data is preferably generated using particle measurements taken from specific, narrow, single-peak soot generator measurements. The particles are generated by a soot generator in the form of a specific, narrow, single-peak (also called monodisperse) particle size distribution. The particle size distribution has a mean µ and a preferably small standard deviation σ with respect to a particle size diameter. Therefore, the soot generator cannot generate particles of exactly one specific size, but only particles with sizes within a certain size range.

[0014] A laser-induced incandescence sensor (often abbreviated to LII sensor) is a measuring instrument for determining particle properties, particularly soot particles in combustion processes. The method is based on the excitation of particles with a strong laser pulse. Due to the rapid and intense energy input, the particles evaporate almost instantaneously and emit light known as incandescence. This incandescence can be measured with suitable detectors. The basic procedure of the LII method is as follows: A short, intense laser pulse irradiates a portion of the medium under investigation (e.g., an exhaust jet). The soot particles in the irradiated volume absorb the laser energy and heat up very quickly. The heated particles emit incandescence light, whereby the intensity and the temporal decay of the light can provide information about the size, concentration, and other properties of the particles.This incandescent light is captured and analyzed by one or more detectors.

[0015] The application and functional range of the present invention is wide-ranging. For example, it can be used for monitoring in the automotive sector to diagnose the condition of particulate filters or gasoline particulate filters. Not only the counting of particles, but also the size of the counted particles can provide information about the harmfulness of emitted exhaust gases. Furthermore, the invention can detect the quantity and size of particles at busy intersections in the city center, thus complying with legal regulations for air pollution control.

[0016] In one embodiment, the plurality of sensor signals is determined by measuring and / or counting particles originating from monodisperse soot generator measurements using the laser-induced incandescence (LII) sensor. In this case, the LII sensor signal is analyzed based on labeled signal curves, for example, based on measurements generated with a dedicated soot generator, using, in particular, supervised machine learning (ML) methods. The soot generator generates, for example, specific, narrow and single-peaked or monodisperse particle size distributions, which form the basis for the subsequent analysis.

[0017] In one embodiment, the soot generator generates monodisperse particle size distributions. The particles generated during the provision of the training data are preferably generated by a soot generator that generates a specific, narrow and / or single-peaked, i.e., monodisperse, particle size distribution of soot. Particles with specific particle sizes from a soot generator cannot always have an exactly predetermined size, but preferably follow a specific, narrow and single-peaked particle size distribution. The sensor signals measured by the LII sensor can be labeled depending on the particle size, since each particle size can be assigned a monodisperse distribution. The large number of measurements thus generates, on average, good training data for a supervised ML training process.

[0018] In one embodiment, the machine learning model is trained on the basis of raw data from the plurality of sensor signals and / or on the basis of preprocessed and / or filtered data from the plurality of sensor signals and / or on the basis of at least one signal feature from the plurality of sensor signals.

[0019] In one embodiment, the at least one signal feature comprises a duration of the respective sensor signal and / or a width of a signal profile of the respective sensor signal and / or an amplitude of the respective sensor signal and / or a gradient of the respective sensor signal. From the signal profile of the respective sensor signal, after the particle event has been detected, in particular by comparison with a threshold, a signal shape is characterized by different signal features, a duration of the signal, a width of the signal profile, an amplitude or a height of the signal and / or a maximum increase of the signal, etc., and the particle size is estimated based thereon. This list of input features or signal features is not exhaustive and is intended merely as an indication of which signal features are possible. Furthermore, it is possible to use the LII raw signal itself as a signal feature.

[0020] In one embodiment, the training comprises supervised training. In principle, semi-supervised training or online training can also be used.

[0021] In one embodiment, the machine learning model comprises a regression model and / or a classification model, wherein the regression model preferably comprises a neural network regression and / or a support vector regression and / or a Gaussian process regression, and / or wherein the classification model preferably comprises a neural network classification and / or a support vector classification and / or a Gaussian process classification. A regression method, in particular a neural network regression and / or a support vector regression and / or a Gaussian process regression, etc., can be used as the ML method, among others. Such a regression method can be effectively used to directly estimate and / or predict the particle size. Alternatively, a classification method can also be used as the ML method.Such a classification method can, for example, be based on a neural network classification and / or a support vector classification and / or a Gaussian process classification, etc., and can be carried out in particular for different, predefined particle size ranges.

[0022] According to a second aspect, a method for diagnosing a particle number and / or a particle size in an exhaust gas stream, in particular downstream of a particulate filter, preferably of a motor vehicle, using a machine learning model is provided. The machine learning model is trained in the present manner. The method comprises: - Providing at least one sensor signal of a laser-induced incandescence sensor for measuring particles of an exhaust gas stream; - Determining and / or estimating a particle size of particles in the exhaust gas flow based on the sensor signal by the trained machine learning model.

[0023] The presently proposed supervised trained ML method for particle size estimation is preferably based on input features of a raw signal from a LII sensor and / or on the LII raw signal itself. This sensor signal is preferably acquired at a very high frequency (millions of Hz) and preferably has a characteristic shape for different particle sizes.

[0024] In one embodiment, particle sizes in the exhaust stream can be diagnosed in a size range of 30 nm to 150 nm. The present method can be used to estimate and / or predict particularly small particles in an exhaust stream.

[0025] According to a third aspect, an evaluation and / or computing device of a laser-induced incandescence-based soot particle sensor is provided, which is configured to execute a present method for diagnosing a particle size in an exhaust gas stream and / or on which the presently trained machine learning model can be executed. The invention could preferably be used in a LII-based soot particle sensor having an integrated microcontroller on which the trained machine learning model can be executed.

[0026] According to a fourth aspect, a system for training a machine learning model for diagnosing a particle number and / or a particle size in an exhaust gas flow, in particular downstream of a particle filter, preferably of a motor vehicle, is specified, the system comprising an evaluation and / or control device which is designed to carry out the following steps: - Providing a labeled training data set comprising a plurality of sensor signals of a laser-induced incandescence sensor; - Training the machine learning model based on the training dataset; - Deploying the trained machine learning model

[0027] According to a fifth aspect, a system for diagnosing a particle number and / or a particle size in an exhaust gas flow, in particular downstream of a particle filter, preferably of a motor vehicle, by means of a trained machine learning model is specified, the system comprising an evaluation and / or control device which is designed to carry out the following steps: - Providing at least one sensor signal of a laser-induced incandescence sensor for measuring particles of an exhaust gas stream; - Determining and / or estimating a particle size of particles in the exhaust gas flow based on the sensor signal by the trained machine learning model.

[0028] The statements made for the procedure apply accordingly to the system. It is understood that linguistic modifications of procedurally formulated features can be reformulated for the system according to common linguistic practice, without such formulations having to be explicitly listed here.

[0029] The invention also claims a computer program with program code for executing at least parts of the method according to the invention in one of its embodiments when the computer program is executed on a computer. In other words, the invention provides a computer program (product) comprising instructions that, when executed by a computer, cause the computer to execute the method / steps of the method according to the invention in one of its embodiments.

[0030] According to the invention, a computer-readable data carrier with program code of a computer program is also proposed for executing at least parts of the method according to the invention in one of its embodiments when the computer program is executed on a computer. In other words, the invention relates to a computer-readable (storage) medium comprising instructions which, when executed by a computer, cause the computer to execute the method / steps of the method according to the invention in one of its embodiments.

[0031] The described designs and further training courses can be combined as desired.

[0032] Further possible embodiments, developments and implementations of the invention also include combinations of features of the invention described previously or below with regard to the embodiments that are not explicitly mentioned. Short description of the drawings

[0033] The accompanying drawings are intended to provide a further understanding of embodiments of the invention. They illustrate embodiments and, in conjunction with the description, serve to explain principles and concepts of the invention.

[0034] Other embodiments and many of the aforementioned advantages will become apparent upon review of the drawings. The elements illustrated in the drawings are not necessarily drawn to scale.

[0035] They show: Fig. 1 a schematic flow diagram of a method for training a machine learning model; Fig. 2 a sensor signal of a LII sensor with a particle event; Fig. 3 an example threshold definition and a particle event detection; Fig. 4 exemplary particle size distributions for several monodisperse measurements; and Fig. 5 Training and test results for training a machine learning model.

[0036] In the figures of the drawings, the same reference symbols designate the same or functionally equivalent elements, parts or components, unless otherwise stated.

[0037] Fig. 1 shows a schematic flow diagram of a computer-implemented method for training a machine learning model to diagnose a particle size in an exhaust stream.

[0038] In any embodiment, the method, particularly a computer-implemented method, can be executed at least partially by a system 100, which for this purpose can comprise several components not shown in detail, for example, an evaluation and computing device. Furthermore, the system 100 can comprise a storage device and / or an output device and / or a display device and / or an input device.

[0039] According to the invention, the computer-implemented method comprises at least the following steps: In a step S1, a labeled training data set is provided, which comprises a plurality of sensor signals of a laser-induced incandescence sensor. In step S2, the machine learning model is trained based on the training data set. In step S3, the trained machine learning model is provided.

[0040] Fig. Figure 2 shows a sensor signal 200 from an LII sensor with a particle event. The ordinate represents time in µs. The abscissa represents an electrical voltage in mV. This imaged and measured particle signal was generated by a soot generator, which generates a specific, narrow, and single-peaked particle size distribution of soot. Particles with specific particle sizes from a soot generator cannot always have a precisely defined size, but rather follow a specific, narrow, and single-peaked particle size distribution. Nevertheless, this is used to label a specific signal shape. Due to the large number of measurements from the LII sensor, good training data is generated on average.

[0041] Fig. Figure 3 shows an example threshold definition and particle event detection. The ordinate represents time in µs. The abscissa represents an electrical voltage in mV. A particle event 300 is preferably detected based on a threshold value 302. From this signal curve, after the particle event 300 has been detected by the at least one threshold value 302, the signal shape is characterized by various signal features.

[0042] Fig. Figure 4 shows exemplary particle size distributions for several monodisperse measurements. The ordinate represents the particle size in nm. The abscissa represents the particle number in number (#) / cm 2 The particle size distributions of different measurements of a specific setting for the mean particle size can be viewed.

[0043] Fig. Figure 5 shows training results (left) and test results (right) for training a machine learning model. Fig. Figure 5 shows a schematic classification diagram that maps the results to the training and test data. A regression model was trained that directly estimates / predicts the size of a particle based on labeled training data. The labels (or targets) of the training (left) and test data (right) are plotted on the x-axis, and the prediction of the regression model is plotted on the y-axis. Furthermore, the coefficient of determination R2 is given, which indicates a value between 0 and 1. The closer the coefficient of determination is to 1, the better the classification and thus the estimate of the particle size. It can be seen that a large value for the coefficient of determination R2 is achievable. QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited patent literature

[0000] DE 10 2018 221 701 A1

[0006] US 2017 / 0 016 820 A1

[0007]

Claims

[1] A method for training a machine learning model for diagnosing a particle size in an exhaust gas stream, the method comprising the steps of: - Providing a labeled training data set comprising a plurality of sensor signals of a laser-induced incandescence sensor; - Training the machine learning model based on the training dataset; - Deploy the trained machine learning model. [2] The method of claim 1, wherein the plurality of sensor signals are determined by measuring and / or counting particles resulting from monodisperse soot generator measurements using the laser-induced incandescence sensor. [3] The method of claim 2, wherein the soot generator produces monodisperse particle size distributions. [4] Method according to one of the preceding claims, wherein the training of the machine learning model is carried out on the basis of raw data of the plurality of sensor signals and / or on the basis of preprocessed and / or filtered data of the plurality of sensor signals and / or on the basis of at least one signal feature of the plurality of sensor signals. [5] Method according to claim 4, wherein the at least one signal feature comprises a duration of the respective sensor signal and / or a width of a signal curve of the respective sensor signal and / or an amplitude of the respective sensor signal and / or a gradient of the respective sensor signal. [6] A method according to any preceding claim, wherein the training comprises supervised training. [7] Method according to one of the preceding claims, wherein the machine learning model comprises a regression model and / or a classification model, wherein the regression model preferably comprises a neural network regression and / or a support vector regression and / or a Gaussian process regression, and / or wherein the classification model preferably comprises a neural network classification and / or a support vector classification and / or a Gaussian process classification. [8] A method for diagnosing a particle size in an exhaust gas stream using a machine learning model trained according to a method according to any one of claims 1 to 7, the method comprising: - Providing at least one sensor signal of a laser-induced incandescence sensor for measuring particles of an exhaust gas stream; - Determining and / or estimating a particle size of particles in the exhaust gas flow based on the sensor signal by the trained machine learning model. [9] The method according to claim 8, wherein particle sizes in a size range of 30 nm to 150 nm are diagnosable. [10] Evaluation and / or computing device of a laser-induced incandescence-based soot particle sensor, which is designed to carry out a method according to claim 8 and / or on which the machine learning model trained according to one of claims 1 to 7 can be executed. [11] System (100) for training a machine learning model for diagnosing a particle size in an exhaust gas stream, the system comprising an evaluation and / or control device which is designed to carry out the following steps: - Providing a labeled training data set comprising a plurality of sensor signals of a laser-induced incandescence sensor; - Training the machine learning model based on the training dataset; - Deploy the trained machine learning model. [12] System (100) for diagnosing a particle size in an exhaust gas stream by means of a trained machine learning model, the system comprising an evaluation and / or control device which is designed to carry out the following steps: - Providing at least one sensor signal of a laser-induced incandescence sensor for measuring particles of an exhaust gas stream; - Determining and / or estimating a particle size of particles in the exhaust gas flow based on the sensor signal by the trained machine learning model. [13] Computer program with program code to carry out at least parts of a method according to one of claims 1 to 9 when the computer program is executed on a computer. [14] Computer-readable data carrier with program code of a computer program for carrying out at least parts of a method according to one of claims 1 to 9 when the computer program is executed on a computer.

Images