Method for determining a target variable, sensor arrangement and vehicle

Abstract

Method for determining a target variable by means of a vehicle sensor (14) and an evaluation device (23), which method comprises the following steps: A) The vehicle sensor detects a sound and generates a first signal (x(t)) depending on the detected sound; B) a predetermined first number of time-limited first signal ranges (W1, W2, W3, W4; W1 - Wn) of the first signal (x(t)) is generated from the first signal (x(t)), C) first spectra (FFT-W1, FFT-W2, FFT-W3, FFT-W4; Log[FFT-W1], Log[FFT-W2], Log[FFT-W3], Log[FFT-W4]) are calculated from the first signal ranges (W1, W2, W3, W4; W1 - Wn) by a first Fourier transformation; D) using an aggregation method, a second spectrum (MEAN-FFT) is calculated from the first spectra (FFT-W1, FFT-W2, FFT-W3, FFT-W4; Log[FFT-W1], Log[FFT-W2], Log[FFT-W3], Log[FFT-W4]); E) a cepstrum is calculated from the second spectrum by a second Fourier transform, which is an inverse Fourier transform; F) depending on the cepstrum, the target variable is determined by a classification procedure, whereby in the classification procedure a confidence factor is assigned to the individual cepstra according to a predetermined procedure, and in which the determination of the target variable takes place depending on the confidence factors assigned to the cepstra.

Description

[0001] The invention relates to a method for determining a target variable, a sensor arrangement for carrying out such a method and a vehicle with such a sensor arrangement.

[0002] EP 3 084 418 B1 shows a method for determining the condition of a road and the condition of a tire, in which a measurement of an acoustic signal is carried out, a spectral power density of the acoustic signal is determined, a frequency interval is segmented, a date is assigned to each frequency band, and a condition of the road and the tire is determined by means of a discriminant analysis of the data.

[0003] US 2017 / 0 309 175 A1 discloses a method for determining a target variable, comprising the following steps: A vehicle sensor detects a sound, and a signal is generated depending on the detected sound. A number of time-limited first signal ranges of the first signal (x(t)) are generated from the first signal (x(t)). Property values are extracted from the signal ranges in the time domain and frequency domain, wherein the property values are values of a cepstrum. The target variable is determined from the cepstrum.

[0004] DE 10 2018 125 713 A1, DE 10 2017 214 409 A1, DE 10 2018 216 557 A1, and DE 10 2016 100 736 A1 disclose further methods for determining a target variable based on sound. For example, DE 10 2017 214 409 A1 discloses a repeated or multiple Fourier transformation to determine a cepstrum. DE 10 2018 125 713 A1 discloses methods for autonomous driving based on audio signals. DE 10 2016 100 736 A1 concerns the classification of a road surface based on sound captured by microphones. DE 10 2018 216 557 A1 discloses a method for acoustic monitoring of the faultlessness of a motor vehicle based on sound signals.

[0005] It is therefore an object of the invention to provide a new method for determining a target variable, a sensor arrangement for carrying out such a method and a vehicle with such a sensor arrangement.

[0006] This object is achieved by a method according to claims 1, 3 and 6, by a sensor arrangement according to claim 18 and a vehicle according to claim 19.

[0007] The methods for determining a target variable by means of a vehicle sensor and an evaluation device according to claims 1, 3 and 6 each comprise the following steps: A) The vehicle sensor detects a sound and generates a first signal depending on the detected sound; B) a predetermined first number of time-limited first signal ranges of the first signal are generated from the first signal, C) first spectra are calculated from the first signal ranges by a first Fourier transformation; D) a second spectrum is calculated from the first spectra using an aggregation procedure; E) a cepstrum is calculated from the second spectrum by a second Fourier transform, which is an inverse Fourier transform; F) Depending on the cepstrum, the target variable is determined using a classification procedure.

[0008] Analysis based on the cepstra allows for accurate classification. The evaluation of steps B) to F) is preferably performed in the evaluation device.

[0009] The inverse Fourier transform differs in its basic mathematical formula from the “normal” Fourier transform in the sign of the exponent.

[0010] According to claim 1, in the classification method, a confidence factor is assigned to the individual cepstra according to a predetermined method, and in which the determination of the target variable takes place as a function of the confidence factors assigned to the cepstra.

[0011] According to claim 3, the second spectrum is calculated by averaging the first spectra or by weighting the first spectra.

[0012] According to claim 6, vehicle data relating to a vehicle's speed and acceleration are evaluated, and depending on the evaluation, it is determined whether the temporally assigned, time-limited first signal ranges are suitable for the classification method. This makes the method more robust.

[0013] According to a preferred embodiment, the target variable comprises at least a first target variable from the target variable group consisting of: • Tread depth of a tire, • Aging condition of a tire, • Tire type, • Tire manufacturers, • Strength of brake squeal, • Dryness between tires and road surface, • Moisture between tires and road surface, • Wetness between tires and road, • Quantitative wetness between tires and road surface, • Presence of ice on the road, • Presence of snow on the road, • Presence of contamination between the tyre and the road surface, • Quantitative information on the contamination between tires and road surface, • Condition of the road surface, and • Porosity of the road surface.

[0014] These targets enable increased security and needs-based service. All subcombinations of this target group are possible.

[0015] According to a preferred embodiment, in step E), the cepstrum is calculated using a discrete cosine transform. The discrete cosine transform can be calculated faster and with less computational effort than a full Fourier transform.

[0016] According to a preferred embodiment, in step F), the cepstrum is compared with stored cepstra of known target variables. This allows for a precise determination of the target variable.

[0017] According to a preferred embodiment, the stored cepstra are generated by measuring known target variables. Such generation allows for a more accurate result than, for example, purely synthetic generation.

[0018] According to a preferred embodiment, in step A), the vehicle sensor detects structure-borne sound or airborne sound. These sound types are relevant for the desired target variables. Vehicle sensors for detecting both structure-borne sound and airborne sound can also be used.

[0019] According to a preferred embodiment, in step B), the first signal ranges overlap at least partially in time. Experiments have shown that this can lead to an improvement in the result.

[0020] According to a preferred embodiment, the first signal is generated as a digital first signal. Generating a digital first signal facilitates further processing.

[0021] According to a preferred embodiment, the first Fourier transform and the second Fourier transform are calculated as discrete Fourier transforms. The accuracy of discrete Fourier transforms is sufficient, and they can be performed quickly.

[0022] According to a preferred embodiment, the first spectra in step C) are calculated based on the logarithm of the first Fourier transform of the first signal ranges of the first signal. Using the logarithm advantageously takes into account different amplitudes over a large range. Alternatively or cumulatively, the frequency of the spectra can also be considered logarithmically.

[0023] According to a preferred embodiment, first data are transmitted from a plurality of vehicles via a network to at least one server in order to enable an evaluation of the first data by the at least one server, which first data comprises at least one piece of information from: • first signal, • first signal areas, • first spectra, • second spectra, • Cepstral.

[0024] The transmission of this data to the server enables the function to be checked and the classification process to be improved.

[0025] According to a preferred embodiment, second data is transmitted from the server to at least two vehicles via a network. This second data is intended to influence the evaluation device, wherein the predetermined target variable is determined in the evaluation device as a function of the second data. The accuracy of the classification method can be improved by the second data.

[0026] According to a preferred embodiment, the determined target value is transmitted to a driver assistance system.

[0027] According to a preferred embodiment, the determined target value is fed to an information display.

[0028] According to a preferred embodiment, the target variable comprises information about the wheels of a vehicle or about the condition of the road on which the vehicle is moving.

[0029] The sensor arrangement according to claim 18 comprises a vehicle sensor and an evaluation device and is designed to carry out such a method.

[0030] The vehicle according to claim 19 has such a sensor arrangement. Such a sensor arrangement is particularly well suited for vehicles.

[0031] Further details and advantageous developments of the invention will become apparent from the exemplary embodiments described below and illustrated in the drawings, which are in no way to be understood as limiting the invention, as well as from the dependent claims. They show: Fig. 1 a vehicle with a sensor arrangement, Fig. 2 shows a schematic representation of the signal processing process, Fig. 3 a signal from a sound sensor with time-limited first signal ranges, Fig. 4 to Fig. 7 a result of a Fourier transformation of the time-limited first signal ranges, Fig. 8 the result of an averaging of the results of Fig. 4 to Fig. 7, Fig. 9 the result of the calculation of the cepstrum from the result of Fig. 8, Fig. 10 an embodiment of a classification method, Fig. 11 an embodiment with a server communication, Fig. 12 a definition of a normal range for cepstra, and Fig. 13 an exemplary representation of the procedure for an example data set from Cepstra with classification relative to the normal range.

[0032] In the following, identical or equivalent parts are provided with the same reference symbols and are usually described only once. The description builds on each figure to avoid unnecessary repetition.

[0033] Fig. 1 shows a vehicle 10 with four wheels 12 and four vehicle sensors 14, each associated with a wheel 12, which are each connected to an evaluation device 23 via an associated data line 16. The evaluation device 23 comprises a computing unit, whereby the calculation can also be performed partially in hardware, for example, in sub-computing units associated with the vehicle sensors 14. The evaluation device 23 can thus be centralized, decentralized, or partially centralized.

[0034] The vehicle sensors 14 and the evaluation device 23 together form a sensor arrangement 20.

[0035] The evaluation device 23 is connected, for example, to a driver assistance system 25 and to an information display 27. Furthermore, the evaluation device 23 is preferably connected to a data transmission unit 29.

[0036] The aim of the evaluation device 23 is to determine a target variable as a function of the signals from the vehicle sensors 14. The target variable preferably comprises information about the wheels 12 or about the condition of the road.

[0037] Fig. Figure 2 shows schematically the basic signal processing process.

[0038] A double arrow 18 indicates a vibration, such as that which occurs due to the dynamic contact of the wheels 12 with the road.

[0039] In step S100, a first signal is generated from the vibration 18 by a sensor. The sensor is preferably a structure-borne sound sensor or an airborne sound sensor. Additionally, signal preprocessing or filtering may take place, for example, to attenuate high-frequency components. Hardware filters and / or software filters may be used for this filtering.

[0040] Subsequently, in step S102, the first signal is transmitted. The transmission preferably occurs as a cyclic transmission.

[0041] In step S104, the first signal is evaluated and the target variable is determined.

[0042] In step S106, further steps can be performed. For example, the determined target value can be sent to the driver assistance system 25 by Fig. 1 so that it can react to sudden moisture or wetness, thereby increasing safety. A feature extraction can be performed, whereby the cepstrum can already be used as a feature. It is also possible to display information about the determined target variable with the information display 27 of Fig. 1 and warn the driver, for example. Another possibility is to use the data transmission unit 29 from Fig. 1 to transmit data to a server in order to enable further processing of the data in the server and, if necessary, also to transmit information from the server via the data transmission unit 29 to the evaluation device 23 of Fig. 1 to be transferred.

[0043] The target variable is preferably at least a first target variable from the target variable group consisting of: • Tread depth of a tire, • Aging condition of a tire, • Tire type, • Tire manufacturers, • Strength of brake squeal, • Dryness of the tires, • Moisture on the tires, • Wetness on the tires, • Quantitative wetness on the tires, • Presence of ice, • Presence of snow, • Presence of pollution, • Quantitative information on pollution, • Road surface, and • Surface porosity.

[0044] The tread depth of a vehicle tire influences the noise generated while driving, and therefore the tread depth can be inferred from the noise.

[0045] As a tire ages, the tire material changes, for example, it becomes more brittle, hardens, and cracks. This also changes the noise the tire produces.

[0046] Tire types such as summer tires, winter tires or ultra-high-performance tires (UHP tires) also influence the noise generated while driving.

[0047] Tire manufacturers use different tire material compounds and different tread patterns, which also result in different noises.

[0048] The condition of the brakes (temperature, wear, wetness) leads to different noises, such as squeaking.

[0049] Humidity and wetness also influence the noise generated by tires while driving, and these noises can be used to determine whether the vehicle is dry, humid, or wet. Approximately, qualitative and quantitative information about wetness is also possible.

[0050] The presence of ice, snow, and dirt also influences the noise generated while driving. Both qualitative and quantitative information about dirt can be provided.

[0051] The nature of the road surface affects the noise, and surfaces such as asphalt, concrete, gravel, paving stones and sand result in different noises.

[0052] The porosity of the road surface leads to different noises, and from the noises it can be concluded whether the road is smooth or porous.

[0053] The distribution of the individual steps to the existing hardware can be variably selected. For example, if the vehicle sensor 14 is Fig. 1 is designed to carry out a Fourier transformation, the first Fourier transformation can already take place in the vehicle sensor 14.

[0054] Fig. 3 shows an example of a signal x(t) of one of the sensors 14 of Fig. 1. The signal x(t) is plotted over a period of one second and divided into four observation periods T1, T2, T3 and T4. In the exemplary embodiment, each observation period T1 to T4 is divided into four windows W1, W2, W3 and W4. In the exemplary embodiment, the windows W1 to W4 are arranged next to one another, but they can also be spaced apart from one another, partially overlapping, or, for example, partially spaced apart, partially overlapping and partially directly adjacent to one another. Which of the variants is advantageous depends on the specific application. The number of windows to be considered in each case can be selected, and generally the windows W1 - Wn are taken into account, where n corresponds to a predetermined number of observation periods to be taken into account in a predetermined time period T.The amplitude A is standardized to values between 1.0 (standardized maximum positive amplitude) and -1.0 (standardized maximum negative amplitude). Each of the windows W1 to W4 corresponds to a temporally limited first signal range of the first signal x(t). Such first signal ranges are also referred to as windowed data.

[0055] Fig. 4 to Fig. 7 show the result of a first Fourier transformation of the time-limited first signal areas W1 ( Fig. 4), W2 ( Fig. 5), W3 ( Fig. 6) and W4 ( Fig. 7). The result of the first Fourier transform is a spectrum with an amplitude A, plotted against the respective frequency. The amplitude A is specified in dBFS, a logarithmic unit with an absolute scale, as used in audio systems. For this purpose, a logarithmic function is applied to the result of the first Fourier transform. The frequency f is specified in Hertz (Hz). The frequency range shown is advantageous. However, it may prove advantageous to consider a larger or smaller frequency range in individual applications.

[0056] The Fourier transformation is preferably performed as a Fast Fourier Transform, referred to as FFT. The notation Log[FFT-W1] thus refers, for example, to performing a Fourier transformation on the first signal x(t) in the first window W1 and then taking the logarithm of the result.

[0057] In the example, only the real part Re of the Fourier transformation is considered.

[0058] The spectrum Log(Re{FFT(x(t))} is mathematically represented. Alternatively, the magnitude spectrum Log(|FFT(x(t))|) can be used.

[0059] Fig. 8 shows the MEAN-FFT result of averaging the spectra of Fig. 4 to Fig. 7, and the result is an averaged second spectrum. The averaging can be performed using the logarithmic spectra as shown, but the spectra can also be averaged in a first step and then the logarithm function applied. During averaging, the amplitudes of the spectra are averaged at a given frequency and used as the averaged amplitude of the averaged second spectrum until the averaged amplitude is obtained over the entire frequency range.

[0060] The averaging results in noise signals in individual windows W1 to W4 being less significant, thus making the method more robust.

[0061] In the averaged second spectrum, the amplitude A is also plotted against the frequency f.

[0062] With a complete first Fourier transform, the spectrum contains both real and imaginary components. Both the real and imaginary components can be processed, or only the real component. Another option is to consider the power spectrum. For this, the squared magnitude of the result of the first Fourier transform is used as the basis for further calculations.

[0063] The logarithm to base 10 or the natural logarithm with base e can be used as a logarithm function.

[0064] Averaging is one possible aggregation method. Another possible aggregation method is, for example, to weight the first spectra by Fig. 4 to Fig. 7, whereby the weighting is determined, for example, by a situation filter or outlier detection. This allows the first spectra to be considered to varying degrees when generating the second spectrum.

[0065] Fig. Figure 9 shows the result of a second Fourier transform of the averaged second spectrum of Fig. 8. The amplitude A is plotted against time t or the quotient 1 / f. Since the first signal also represents an amplitude over time, the signal from Fig. 9 The axis is called the cepstrum, and the abscissa is called the quefrency. The terms cepstrum and quefrency are anagrams of the English terms spectrum and frequency, and were introduced by John Wilder Tukey in a 1963 publication.

[0066] The second Fourier transformation transforms in particular the results of the first Fourier transformation ( Fig. 4 to Fig. 7 or Fig. 8) occurring frequencies are determined.

[0067] The second Fourier transform is preferably an inverse Fourier transform (IFFT). If the result of the first Fourier transform (cf. Fig. 8) only the real part or the magnitude spectrum is taken into account, the second Fourier transformation corresponds to a cosine transformation if the spectrum is symmetrical with respect to the y-axis (f = 0), since the sine components have no effect without the imaginary part.

[0068] Mathematically, for example, Re{IFFT(Log(Re{FFT(x(t))}))} or Re{IFFT(log(|FFT(x(t))|))} calculated.

[0069] Experiments have shown that the cepstrum of Fig. 9 is well suited to determine the desired target variables.

[0070] Fig. 10 shows an example of a classification procedure by which, depending on the cepstrum of Fig. 9 the target variable is determined.

[0071] The classification method uses, for example, an input vector 70 of dimension 1x32, and a group of classes is defined in a field 72, each of which is assigned to the cepstrum of a given tire type TT at a given vehicle speed. The classification method then checks whether the cepstrum of Fig. 9 corresponds to one of the classes, or what degree of agreement the cepstrum has with the respective class. The result could, for example, be a good match with a winter tire at a speed of 80-90 km / h and a good match with a summer tire at 120-130 km / h, and a poorer match with the different tire types at other speeds. Additional consideration of the speed of the vehicle 10 during the noise measurement then enables the selection of the appropriate class. If the measurement was carried out at a speed of 82 km / h, for example, a winter tire is determined as the result of the target variable. Taking into account other parameters such as speed leads to a greater robustness of the classification procedure.

[0072] Class group 72 is further subdivided into classes that correspond to a new tire, a slightly worn tire, and a heavily worn tire. This subdivision also makes it possible to assign the first signal x(t) to a tire condition 74 (new), 75 (slightly worn), and 76 (heavily worn) during classification. These tire conditions are also referred to as the main classes of the classification process. This information can be used, for example, in the driver assistance system 25 by activating the safety systems more strongly in the case of a heavily worn tire or by preventing extreme driving situations. In addition, a display of the tire condition is provided in the information display 27 of Fig. 1 possible.

[0073] Possible classification methods include: • Cuboid classifier, • Distance classifier, • Nearest neighbor classification, • Clustering methods, • artificial neural network, and • latent class analysis.

[0074] Artificial neural networks are networks of artificial neurons. They therefore represent a field of artificial intelligence.

[0075] The classification process is preferably influenced by machine learning. For this purpose, sound measurements are conducted under known, predefined conditions, and the measurement data is evaluated to develop the classification process. Methods such as cross-validation are used.

[0076] A further possibility for increasing the robustness of the method is to evaluate vehicle data on a speed (e.g. vehicle speed, tire speed, engine speed) and an acceleration (e.g. vehicle acceleration, tire acceleration, engine acceleration) of the vehicle 10, and depending on the evaluation, it is determined whether the temporally assigned, time-limited first signal ranges W1-W4 are suitable for the classification method. In this case, for example, if predetermined limit values are exceeded or undershot, the corresponding signal range can be disregarded as an invalid sample, or it can be taken into account less than signal ranges where the speed or acceleration lies within a predetermined range. Such a filter can be referred to as a situation or stone chip filter.

[0077] It is also possible to filter out unusual early signal regions by performing outlier detection. This means that if an early signal region or the associated spectrum exhibits, for example, an unusual shape or unusually high amplitudes, they can be considered less or not at all.

[0078] Fig. 11 shows a preferred embodiment in which the evaluation device 23A of a first vehicle and the evaluation device 23B of a second vehicle 10 are in data connection with a server 80. Preferably, the data connection is wireless and is carried out, for example, via the data transmission unit 29 of Fig. 1. The evaluation devices 23 preferably transmit first data to the server 80. The first data preferably comprise at least one piece of information from: • first signal, • first signal areas, • first spectra, • second spectra, • Cepstral.

[0079] In this way, data from different vehicles can be evaluated in server 80, and depending on the initial data, an improvement of the classification method in server 80 is possible. The improved classification method can then be returned to the evaluation devices 23A and 23B via a data connection. This allows the evaluation devices 23 of different vehicles to contribute to an improvement of the classification method, and data on new classes (e.g., new tire types) can also be included in the classification method.

[0080] The server 80 can, for example, be provided by the manufacturer of the vehicle or as part of a cloud infrastructure.

[0081] Fig. Figure 12 shows another embodiment of a classification method. A closed curve 40, for example, indicates a normal range within which known measured cepstra 42 lie. Outside the normal range 40 are measured cepstra 41 that are unknown to the classification method. In other words, measured cepstra are expected to lie within the normal range 40. Remaining cepstra 41 are marked as outliers and not used for classification.

[0082] Fig.Figure 13 shows an exemplary representation of the method for an example data set in which the measured cepstra form a snake-like two-dimensional curve. The cepstra 52 within the normal range 40 have a high confidence factor, and the cepstra 51 outside the normal range 40 have a comparatively low confidence factor. Preferably, the confidence factor is not only considered with 0 (no confidence or not certain) or 1 (full confidence or certain), but can also be set, for example, depending on the distance from the normal range 40, e.g., 0.0; 0.1; 0.7; 1.0. Considering the confidence factor for the respective cepstrum increases the robustness of the classification method.

[0083] Naturally, various variations and modifications are possible within the scope of this application.

[0084] For example, in addition to the cepstrum, the spectrum can also be used for the classification process. List of reference symbols 10 vehicles 12 wheels 14 Vehicle sensor 16 data lines 18 Vibration 20 Sensor arrangement 23 Evaluation device 25 Driver assistance system [e.g. active wheel suspension, active steering systems] 27 Information display 29 Data transmission unit 40 normal range 41 unknown cepstrum 42 known cepstrums 51 unknown cepstrum of a current measurement 52 known cepstrum of a current measurement T1-T4 observation periods W1-W4 windows

Claims

[1] Method for determining a target variable by means of a vehicle sensor (14) and an evaluation device (23), which method comprises the following steps: A) The vehicle sensor detects a sound and generates a first signal (x(t)) depending on the detected sound; B) from the first signal (x(t)) a predetermined first number of time-limited first signal ranges (W1, W2, W3, W4; W1 - Wn) of the first signal (x(t)) is generated, C) from the first signal ranges (W1, W2, W3, W4; W1 - Wn) first spectra (FFT-W1, FFT-W2, FFT-W3, FFT-W4; Log[FFT-W1], Log[FFT-W2], Log[FFT-W3], Log[FFT-W4]) are calculated by a first Fourier transformation; D) using an aggregation method, a second spectrum (MEAN-FFT) is calculated from the first spectra (FFT-W1, FFT-W2, FFT-W3, FFT-W4; Log[FFT-W1], Log[FFT-W2], Log[FFT-W3], Log[FFT-W4]); E) a cepstrum is calculated from the second spectrum by a second Fourier transform, which is an inverse Fourier transform; F) depending on the cepstrum, the target variable is determined by a classification procedure, whereby in the classification procedure a confidence factor is assigned to the individual cepstra according to a predetermined procedure, and in which the determination of the target variable takes place depending on the confidence factors assigned to the cepstra. [2] Method according to claim 1, wherein in the aggregation method the second spectrum (MEAN-FFT) is calculated by averaging the first spectra (FFT-W1, FFT-W2, FFT-W3, FFT-W4; Log[FFT-W1], Log[FFT-W2], Log[FFT-W3], Log[FFT-W4]). [3] Method for determining a target variable by means of a vehicle sensor (14) and an evaluation device (23), which method comprises the following steps: A) The vehicle sensor detects a sound and generates a first signal (x(t)) depending on the detected sound; B) from the first signal (x(t)) a predetermined first number of time-limited first signal ranges (W1, W2, W3, W4; W1 - Wn) of the first signal (x(t)) is generated, C) from the first signal ranges (W1, W2, W3, W4; W1 - Wn) first spectra (FFT-W1, FFT-W2, FFT-W3, FFT-W4; Log[FFT-W1], Log[FFT-W2], Log[FFT-W3], Log[FFT-W4]) are calculated by a first Fourier transformation; D) by means of an aggregation process, a second spectrum (MEAN-FFT) is calculated from the first spectra (FFT-W1, FFT-W2, FFT-W3, FFT-W4; Log[FFT-W1], Log[FFT-W2], Log[FFT-W3], Log[FFT-W4]) in such a way that the second spectrum (MEAN-FFT) is calculated by averaging the first spectra (FFT-W1, FFT-W2, FFT-W3, FFT-W4; Log[FFT-W1], Log[FFT-W2], Log[FFT-W3], Log[FFT-W4]) or by weighting the first spectra (FFT-W1, FFT-W2, FFT-W3, FFT-W4; Log[FFT-W1], Log[FFT-W2], Log[FFT-W3], Log[FFT-W4]); E) a cepstrum is calculated from the second spectrum by a second Fourier transform, which is an inverse Fourier transform; F) Depending on the cepstrum, the target variable is determined using a classification procedure. [4] Method according to claim 3, in which in the classification method a confidence factor is assigned to the individual cepstra according to a predetermined method, and in which the determination of the target variable takes place as a function of the confidence factors assigned to the cepstra. [5] Method according to one of claims 1-4, in which vehicle data on a speed and an acceleration of the vehicle (10) are evaluated, and in which it is determined as a function of the evaluation whether the temporally assigned, time-limited first signal ranges (W1, W2, W3, W4; W1 - Wn) are suitable for the classification method. [6] Method for determining a target variable by means of a vehicle sensor (14) and an evaluation device (23), which method comprises the following steps: A) The vehicle sensor detects a sound and generates a first signal (x(t)) depending on the detected sound; B) from the first signal (x(t)) a predetermined first number of time-limited first signal ranges (W1, W2, W3, W4; W1 - Wn) of the first signal (x(t)) is generated, C) from the first signal ranges (W1, W2, W3, W4; W1 - Wn) first spectra (FFT-W1, FFT-W2, FFT-W3, FFT-W4; Log[FFT-W1], Log[FFT-W2], Log[FFT-W3], Log[FFT-W4]) are calculated by a first Fourier transformation; D) using an aggregation method, a second spectrum (MEAN-FFT) is calculated from the first spectra (FFT-W1, FFT-W2, FFT-W3, FFT-W4; Log[FFT-W1], Log[FFT-W2], Log[FFT-W3], Log[FFT-W4]); E) a cepstrum is calculated from the second spectrum by a second Fourier transform, which is an inverse Fourier transform; F) depending on the cepstrum, the target variable is determined by a classification method, wherein vehicle data on a speed and an acceleration of the vehicle (10) are evaluated, and wherein depending on the evaluation it is determined whether the temporally assigned, time-limited first signal ranges (W1, W2, W3, W4; W1 - Wn) are suitable for the classification method. [7] Method according to one of the preceding claims, in which the target variable comprises at least a first target variable from the target variable group consisting of: • Tread depth of a tire, • Aging condition of a tire, • Tire type, • Tire manufacturers, • Intensity of brake squeal, • Dryness between tires and road, • Moisture between tires and road surface, • Wetness between tires and road, • Quantitative wetness between tires and road surface, • Presence of ice on the road, • Presence of snow on the road, • Presence of contamination between the tyres and the road surface, • Quantitative information on the contamination between tires and road surface, • Condition of the road surface, and • Porosity of the road surface. [8] Method according to one of the preceding claims, wherein in step E) the cepstrum is calculated by a discrete cosine transformation. [9] Method according to one of the preceding claims, in which in step F) a comparison of the cepstrum with stored cepstra of known target variables is carried out, wherein the stored cepstra are generated by measuring known target variables. [10] Method according to one of the preceding claims, in which in step A) a structure-borne sound or an airborne sound is detected with the vehicle sensor. [11] Method according to one of the preceding claims, in which in step B) the first signal ranges (W1, W2, W3, W4; W1 - Wn) at least partially overlap in time. [12] Method according to one of the preceding claims, in which the first signal (x(t)) is generated as a digital first signal (x(t)). [13] A method according to any one of the preceding claims, wherein the first Fourier transform and the second Fourier transform are calculated as discrete Fourier transforms. [14] Method according to one of the preceding claims, in which the first spectra (Log[FFT-W1], Log[FFT-W2], Log[FFT-W3], Log[FFT-W4]) in step C) are calculated on the basis of the logarithm of the first Fourier transform of the first signal ranges (W1, W2, W3, W4; W1 - Wn) of the first signal (x(t)). [15] Method according to one of the preceding claims, in which the target variable in the classification method is also determined as a function of vehicle data via - a speed of the vehicle (10), - an acceleration of the vehicle (10), or - a speed and an acceleration of the vehicle (10) are determined. [16] Method according to one of the preceding claims, in which first data are transmitted from a plurality of vehicles (10; 10A, 10B) via a network to at least one server (80) in order to enable an evaluation of the first data by the at least one server (80), which first data comprise at least one piece of information from: • first signal, • first signal areas, • first spectra, • second spectra, • Cepstral. [17] Method according to claim 16, wherein the server (80) transmits second data via a network to at least two vehicles (10; 10A, 10B), which second data are intended to influence the evaluation device (23), wherein the predetermined target variable is determined in the evaluation device (23) as a function of the second data. [18] Sensor arrangement (20) which comprises a vehicle sensor (14) and an evaluation device (23), and which is designed to carry out a method according to one of the preceding claims. [19] Vehicle (10) comprising a sensor arrangement (20) according to claim 18.

Images