Computer-implemented method and computer program product for diagnosing the acoustic detection capability of at least one acoustic sensor and computer-implemented method for diagnosing the acoustic detection capability of a vehicle

Abstract

Computer-implemented method for diagnosing the acoustic detection capability of at least one acoustic sensor (mic0-mic7), the method comprising the steps of: generating at least one acoustic sensor model (ASM), the model (ASM) comprising audio data of the at least one acoustic sensor (mic0-mic7); obtaining current mean values ​​(µ c ) and current standard deviations (σ c ) of the sound field quantity for the frequency components (f) of a current sequence of the sound field quantity, wherein the current sequence was recorded during a journey of the vehicle (1) at a current speed with the acoustic sensor (mic0-mic7) (V3); comparison of the current mean values ​​(µ c ) and current standard deviations (σ c ) with the mean values ​​(µ m ) and standard deviations (σ m) from the speed range of the acoustic sensor model (ASM) in which the current speed lies, where the current mean values ​​(µ) have statistical significance c ) and current standard deviations (σ c ) the acoustic detection capability of the acoustic sensor (mic0-mic7) is diagnosed as anomalous (V4).

Description

[0001] The invention relates to a computer-implemented method and a computer program product for diagnosing an acoustic detection capability of at least one acoustic sensor. Furthermore, the invention relates to a computer-implemented method for diagnosing an acoustic detection capability of a vehicle.

[0002] The following definitions, descriptions and statements retain their respective meaning for and apply to the entire disclosed subject matter of the invention.

[0003] Sensors, such as acoustic sensors and microphones, can fail. If a defect goes undetected, this can have negative consequences. For example, in an automated vehicle, such as an autonomous vehicle that uses external microphones to detect the environment, such as a special emergency signal, such as a siren, this signal may no longer be detected if a membrane in the microphone is torn.

[0004] Electrical failures such as short circuits, cable breaks, undervoltage or similar can usually be identified because the microcontrollers used detect and report these errors, or the digital audio signal is transmitted only as 0s.

[0005] However, if the fault occurs before the analog-to-digital conversion, for example, due to a torn membrane in the microphone or a sensor clogged with snow, the audio signal is simply severely attenuated. This attenuation can render the signal unusable. The electronics cannot detect such a fault. If the sensor is simply clogged, a trip through a car wash can solve the problem.

[0006] An analogous example is a person who, after diving, has water in their ear and can only perceive their surroundings very muffled and unclearly. As a countermeasure, the head is tilted to allow the water to drain away. The detection of water in the ear is achieved neither by special sensing elements in the ear nor by a sound actively generated by the person, but rather by the passive detection of an unusually low ambient noise level.

[0007] In the field of acoustics, it is a routine task to check whether a microphone is functioning properly. For this purpose, the frequency response is typically measured by generating a noise excitation or sweep with a calibrated loudspeaker, and the microphone records this signal. The signal is then displayed in the frequency domain and compared with a reference curve.

[0008] An acoustic sensor is a sensor that detects mechanical vibrations, for example caused by airborne sound waves, and converts them into a processable signal, for example, an electrical signal such as an electrical voltage. The acoustic sensor includes an analog and / or digital signal output. The conversion takes place in two stages. In a first acoustic-mechanical conversion stage, the airborne sound is converted into the movement of an object, for example, a membrane of the acoustic sensor, according to a specific reception principle. In the second mechanical-electrical conversion stage, the movement of the object is converted into the electrical signal according to a specific transducer principle. Examples of acoustic sensors include an arrangement of a magnet and an electrical coil, microphones, accelerometers, piezo sensors, or strain gauges.A micro-electro-mechanical system, abbreviated MEMS, comprising an array of semiconductor elements that absorb vibrations, can also be used as an acoustic sensor.

[0009] US 8 379 876 B2 discloses an audio device for fault detection in a microphone array. The microphone array comprises a first microphone and a second microphone, each generating a first audio signal and a second audio signal from the ambient audio signals. A fault detector detects functions of the first and second microphones based on the first and second audio signals and generates a status signal. A digital signal processor processes the first and second audio signals based on the status signal. If the status signal indicates that only the first microphone or the second microphone is faulty, the digital signal processor switches to a single-microphone mode in which only the remaining normal microphone is activated. If the status signal indicates that both the first and second microphones are faulty, the digital signal processor generates a fault indication signal and stops processing the first and second audio signals.The error detector detects speech activity in the first and second audio signals. The disadvantage of this audio device is that error detection is based on speech activity.

[0010] US 10 887 713 B1 discloses a method for testing a microphone, whereby a specific target signal is generated, which must then be recorded by a microphone. If the microphone is defective, it cannot record the reproduced target sound at the volume that would otherwise be possible. The disadvantage of this method is that it relies on a target signal.

[0011] Further prior art is shown in DE 10 2019 213 816 A1 and US 2021 / 0 375 305 A1.

[0012] The invention was based on the object of how an acoustic detection capability of an acoustic sensor which is arranged in use on a frame and / or an outer skin of a vehicle and is exposed to external influences, in particular while the vehicle is moving, can be diagnosed, in particular as normal, abnormally quiet or abnormally loud, in particular without actively playing a reference tone.

[0013] The subject matter of the independent and subordinate claims each solves this problem by means of a self-diagnosis of the acoustic sensor, in which the acoustic detection capability is determined by comparing it with data from the normal state of the acoustic sensor via recorded audio data. Specific ambient noise and other available information such as driving speed are recorded in a data-driven manner.

[0014] Advantageous embodiments of the invention emerge from the definitions, the subclaims, the drawings and the description of preferred embodiments.

[0015] According to one aspect, the invention provides a computer-implemented method for diagnosing the acoustic detection capability of at least one acoustic sensor. The acoustic sensor can comprise a microphone capsule and a transducer. The acoustic-mechanical conversion takes place in the microphone capsule. The microphone capsule comprises, for example, a membrane that is excited to vibrate by airborne sound. The mechanical-electrical conversion takes place in the transducer. The transducer is, for example, an electrodynamic transducer, such as in a moving-coil microphone, or an electrostatic transducer, such as in a condenser microphone. According to one aspect, the acoustic sensor is a MEMS microphone.

[0016] One step of the method comprises generating at least one acoustic sensor model. The model can be generated using real recorded audio data or simulated using synthetic audio data. Audio data is, for example, the raw output data of the acoustic sensor. The model comprises audio data from the at least one acoustic sensor, wherein the audio data comprises sequences of at least one sound field variable of a sound field outside a vehicle and the sequences are obtained while the vehicle is traveling at different speeds. The sequences can comprise, for example, a spectrogram being evaluated every 2 s. The time interval of 2 s can be sampled using a smoothing window or sliding window, for example with a duration of 0.2 s. Sound field variables describe the state of a sound field and include, for example, sound pressure, sound pressure level, and sound flow.The loudness of a sound event can be measured as sound pressure. A vehicle generally refers to a movable machine. A vehicle is, for example, a passenger vehicle or a commercial vehicle for road, land, air, or water. The vehicle can be electrically powered. The vehicle can be an automated vehicle. For example, the vehicle is an automated, e.g., autonomous, passenger transport system, such as a shuttle. Statistics on the audio data are determined by breaking down the audio data obtained in these speed ranges into frequency components and determining, for each frequency component, a mean value of the sound field size for this frequency component and a standard deviation of the sound field size for this frequency component. Speed ​​ranges can be, for example, the ranges (7-9) m / s, (9-11) m / s, (11-13) m / s, etc.The frequency components can be, for example, the 32 frequency components, also called bins, 750 Hz, 1500 Hz, ..., 24 kHz. This means that in each of the speed ranges mentioned above, for example, the aforementioned frequency bins are obtained, and for each of these frequency bins, a mean and a standard deviation of the sound field size are obtained. This creates submodels based on the speed, each with a mean and a standard deviation for each frequency bin. The audio data can be recorded over long periods of time, for example, days, weeks, or years, or a correspondingly large volume of audio data can be simulated. A long-term model is then generated using the statistics.

[0017] A further step of the method comprises obtaining current mean values ​​and current standard deviations of the sound field quantity for the frequency components of a current sequence of the sound field quantity, wherein the current sequence was recorded with the acoustic sensor while the vehicle was traveling at a current speed. The current sequence can, for example, comprise 10 sample values. A sample value is a value of the sequence, i.e. a signal, or a series of values ​​at a specific point in time. If there is more than one acoustic sensor, according to one aspect a multi-channel signal is processed in the recorded sequence of audio data. The audio data from each acoustic sensor is processed in a respective channel. In a multi-channel signal, the sample per channel comprises a value or a series of values ​​at a specific point in time.Using the current sequence, the current mean value, and the current standard deviation, a short-term model is generated relative to the long-term model. Using the current mean values ​​and the current standard deviations, the current short-term models can be quantitatively differentiated. This makes it possible to assign and derive a model that corresponds to the system state. According to one aspect, several trips are carried out, each at a current speed, and the process step described here is repeated several times. A trip at a current speed can also be repeated on different days, for example under different weather conditions, and / or at different times of day. The different conditions generate a large standard deviation of the long-term model relative to the short-term model.

[0018] A further step of the method comprises comparing the current mean values ​​and current standard deviations with the mean values ​​and standard deviations from the speed range of the acoustic sensor model in which the current speed lies. If the current mean values ​​and current standard deviations are statistically significant, the acoustic detection capability of the acoustic sensor is diagnosed as anomalous. For example, if the current speed is 52 km / h, the current mean value and the current standard deviation are compared with the mean value and standard deviation from the speed range (13-15) m / s of the acoustic sensor model. The short-term model is compared with the long-term model. In one aspect, the long-term model comprises the mean values ​​and standard deviations in the form of lookup tables, with one table being provided for each of the aforementioned speed ranges.Each table contains the respective mean values ​​and standard deviations for each of the aforementioned frequency components. If multiple acoustic sensors are used, each table contains the respective mean values ​​and standard deviations for each frequency component for each channel, i.e., for each of the acoustic sensors. Statistical significance exists if the current mean values ​​and current standard deviations deviate so significantly from the corresponding values ​​of the long-term model that the assumption of normal acoustic detection capability of the acoustic sensor at the current speed must be rejected. If the acoustic sensor is clogged, for example, due to the influence of foreign particles, its acoustic detection capability will be abnormally quieter. This can be diagnosed using this method by comparing it with the long-term model, which can serve as a reference.

[0019] The method thus comprises three processes: pre-filtering the acoustic input data by breaking it down into frequency components, model building, and decision logic. Pre-filtering reduces the influence of transient noise sources on model building and on the comparison in the decision logic. The advantage of this method is that no self-testing of the acoustic sensor with test signals is required, as specific ambient noise and other available information such as driving speed are used. A further advantage of the method is that the audio recordings are linked to driving speeds. Driving speeds can be reliably acquired in a vehicle. This means that by linking the audio data to the driving speed, it can be determined at ASIL Level A instead of the previously usual ASIL Level QM.

[0020] According to a further aspect, the acoustic sensor model is generated from long-term recordings of the acoustic sensor and incremented with the current mean values ​​and current standard deviations. For example, during a trip at a speed of 42 km / h N a =10 samples are recorded. For each of the above-mentioned frequency components, an average value µ a and a standard deviation σ a In the long-term model model k, the appropriate speed range (11-13) m / s is selected, for example, by a memory logic in which the speed ranges are stored, for example, in the form of the aforementioned lookup tables. In the current recording, the long-term model in this speed range includes N b Samples, for example N b =1000 and for each of the above-mentioned frequency components an average value µ b and a standard deviation σ bThe incremented long-term model model k+1 is then defined by the corresponding values ​​N c , µc , σ c characterized, where the incrementation is realized, for example, by the following operations: Nc=Na∪b=Na+Nb; μc=μa∪b=Naμa+NbμbNa∪b; σa∪b2=Na(σa2+μa2)+Nb(σb2+μb2)Na∪b−μa∪b2⋅

[0021] According to another aspect, the sequences include a sound pressure level from a background noise. This allows the volume of sound events to be determined. The amplitude of the frequency components then indicates the volume at the respective frequency.

[0022] According to another aspect, the audio data is processed in blocks. A block size is given by the number of samples of the audio data at a specific point in time. The sequences comprise a number of blocks. A block can be a vector or a matrix of samples from consecutive times arranged one after the other. For example, in the case of audio data from multiple acoustic sensors being processed in individual channels, a block is a matrix whose columns correspond to the channels and contain the samples of the respective channels. For blocks, the first dimension is parsed as the time domain and the second dimension as channels. A block is also called a frame. For example, for an eight-channel signal, where each channel contains ten samples, a block is a 10 x 8 matrix. The number of samples in a block is also called the frame size. The number of blocks is, for example, 20.For example, the aforementioned sliding window can comprise 20 blocks. For example, the block period is 0.01 s. If a block comprises 10 samples, the sampling time of the individual samples is 0.001 s. Block-wise processing advantageously allows multiple samples to be processed at once, thus reducing overhead.

[0023] According to another aspect, the audio data in each block is broken down into equal frequency components. The mean values ​​of the sound field parameters for each frequency component are determined by summing the values ​​of the sound field parameters for this frequency component from the blocks and dividing the sum by the number of blocks. This corresponds to smoothing across multiple samples and has the advantage of reducing transients.

[0024] According to another aspect, the frequency components are calculated using the fast Fourier transform. As an alternative to pre-filtering by decomposing the signal into frequency components using the fast Fourier transform, pre-filtering can also be performed using cepstrum, mel-spectrum, or artificial neural networks.

[0025] According to a further aspect, for each frequency component, a probability density function, which is characterized by the respective current mean value and the respective current standard deviation, is calculated in a first range (µ m - 2σ m ; µ m + 2σ m ) and a second area (µ m + 2σ m ; +∞) integrated. µ m is the corresponding mean and σ mthe corresponding standard deviation from the acoustic sensor model, the long-term model. If the integral over the second range is greater than the integral over the first range, the acoustic detection capability of the acoustic sensor is diagnosed as abnormally louder. The probability density function, for example, is the normal distribution. f(x)=12πσ2exp(−(x−μ)22σ2)⋅ The first range contains 98% of all audio data recorded so far. Integrating the current probability density function over the first range of the long-term model's distribution represents a consistency probability. Integrating the current probability density function over the second range represents an anomaly probability.

[0026] According to a further aspect, in the case where the integral is over a third region (-∞; µ m - 2σ m) is greater than the integral over the first range, the acoustic detection capability of the acoustic sensor is diagnosed as abnormally quiet.

[0027] In principle, it is possible to compare whether the expected value of the short-term model is larger or smaller than that of the long-term model and thus to distinguish whether the short-term behavior tends to be quieter or louder, even with normal microphone behavior.

[0028] Due to the relative position of the short-term and long-term models, the second and third ranges add a sign, extending the output to the range -100% to +100%. This allows for detailed diagnostics, for example, louder for a transient external noise or quieter for a dirty sensor, which causes attenuation of the input signals.

[0029] According to a further aspect, if there are multiple acoustic sensors, an acoustic sensor model is generated for each of the acoustic sensors. In block-wise processing of the audio data, the blocks for each of the acoustic sensors comprise a channel as described above. Each channel comprises a number of samples of the audio data. In each channel, the mean values ​​are calculated as described above, and standard deviations are obtained. In each channel, the probability density function, which is characterized by the respective current mean value and the respective current standard deviation, is integrated over the first, second, and third ranges given by the respective acoustic sensor model, as described above. Based on a comparison of the respective integrals obtained in the individual channels, the acoustic detection capability of each acoustic sensor is diagnosed as normal, abnormally loud, or abnormally quiet.

[0030] According to a further aspect, the model is created before the acoustic sensor(s) are deployed in the vehicle. The model can also be created while the vehicle is in operation, i.e. online, using memory logic to increment the acoustic sensor model(s) with the current mean values ​​and current standard deviations. Another aspect concerns a two-stage approach combining offline trained functions and online functions, which are provided, for example, via over-the-air updates. Advantageously, this means that if an acoustic sensor diagnosed as abnormal is replaced, no calibration is required. Basic functionality can be ensured with the offline model. Online modeling increases functionality during operation by recording the sensor- and position-specific properties.

[0031] According to a further aspect, the invention provides a computer program for diagnosing an acoustic detection capability of at least one acoustic sensor.

[0032] The computer program comprises program instructions that cause the computer to carry out the method described above when the computer program is loaded onto the computer or executed. The computer program can be provided as interpretable code that is loaded onto the computer, for example directly into an internal memory of the computer, or as compiled code that is executed by the computer. The provision can be made, for example, on a machine-readable medium on which the computer program is stored or via a data carrier signal. The program instructions can be machine instructions, source code, or object code, written in assembly language, an object-oriented programming language, for example C++, or in a procedural programming language, for example C.

[0033] According to a further aspect, the invention provides a computer-implemented method for diagnosing the acoustic detection capability of a vehicle. A plurality of acoustic sensors are arranged on a frame and / or an outer skin of the vehicle. The acoustic detection capability of each of the acoustic sensors is diagnosed according to the method described above. The vehicle is, for example, an automated vehicle that perceives its driving environment using acoustic detection capability, possibly in addition to visual detection capabilities.

[0034] According to a further aspect, acoustic sensors that record a similar sound field due to their arrangement on the vehicle or that are exposed to similar airflow during vehicle travel, for example, at the front, on the roof, or at the rear, form a group. Anomalies in acoustic detection capability are diagnosed by comparing the acoustic detection capability within a group of acoustic sensors. The acoustic sensors in a group are correlated. This correlation improves diagnostic functionality. Model uncertainties and long transient sound events affect similarly installed acoustic sensors equally, so consistency testing within groups provides another symptom for detecting anomalies. The deviation from the reference model and the consistency of at least two similarly installed sensors can be used to differentiate between cases.

[0035] The invention is explained by way of example in the following figures. They show: Fig. 1 an example of a pre-filtering system, Fig. 2 an embodiment for incrementing a long-term model with a short-term model, Fig. 3 an embodiment of an anomaly detection disclosed here, Fig. 4a an embodiment of a normally diagnosed acoustic detection capability disclosed here, Fig. 4b shows an embodiment of an abnormally loud diagnosed acoustic detection capability disclosed here, Fig. 4c shows an embodiment of an acoustic detection capability diagnosed as abnormally too loud and abnormally too quiet disclosed here, Fig. 5 an embodiment of an arrangement of acoustic sensors on a vehicle, Fig. 6 an embodiment of an analysis of a self-diagnosis of acoustic sensors disclosed here, Fig. 7 shows a further embodiment of an analysis of a self-diagnosis of acoustic sensors disclosed here, Fig. 8 an embodiment of an arrangement of groups of acoustic sensors on a vehicle, Fig. 9 an embodiment of an analysis of a self-diagnosis of groups of acoustic sensors disclosed here and Fig. 10 a schematic representation of a method disclosed here.

[0036] In the figures, identical reference symbols designate identical or functionally similar parts. For clarity, the relevant reference parts are indicated in the respective figures.

[0037] In Fig. 1, the audio data is processed in blocks B. The time interval T for a measurement is, for example, T=2 s. This time interval T is sampled with a sliding window of duration t, for example, t=0.2 s. In each sliding window, a number NF of blocks B are recorded. A block B comprises a number N of samples, for example, N=10. The samples can be read out in one channel from one acoustic sensor mic0-mic7 or multi-channel, i.e., one channel for each of the multiple acoustic sensors mic0-mic7, for example, from 8 channels. In the single-channel case, block B is a vector with 10 entries. In the multi-channel case, the block is an 8 x 10 matrix. In each block B, the sequence of samples is represented in a frequency spectrum, for example, using a fast Fourier transform (FFT): Each block B can, for example, be decomposed into 32 frequency components. By averaging across blocks B, the influence of transient noise is reduced.The measurements can be repeated and performed at different driving speeds, allowing a long-term model to be generated.

[0038] Fig. Figure 2 shows the incrementation of a long-term model of a multi-channel system of acoustic sensors mic0-mic7 over N samples with a short-term model of 10 samples. The long-term model is accessible via lookup tables. The long-term model, for example, comprises 1000 samples. The recordings of the acoustic sensors in the short-term model are taken at a driving speed of 42 km / h. Each microphone channel is divided into NF blocks B. In each block, the sequences are decomposed using FFT, so that averaging in each microphone channel produces a current mean value µ for each frequency component. c and a current standard deviation σ ccan be determined. In the long-term model, the corresponding speed range is then selected via the lookup tables, in which the long-term model mean values ​​µ m and the long-term model standard deviations σ c By the union operations M, namely Nc=Na∪b=Na+Nb, μc=μa∪b=Naμa+NbμbNa∪b, σa∪b2=Na(σa2+μa2)+Nb(σb2+μb2)Na∪b−μa∪b2, the number of samples, mean and standard deviation for each frequency component f in each microphone channel of the incremented long-term model are obtained.

[0039] Fig. Figure 3 shows a schematic of the acoustic sensor model ASM. The current driving speed and the current mean values ​​µ c and current standard deviations σ cfor each frequency component f are read in and compared with the long-term model mean values ​​µ stored in the corresponding speed range of the long-term model m and long-term model standard deviations σ m each frequency component f is compared.

[0040] A consistency probability and an anomaly probability are calculated using the probability density function PDF of the short-term model, which is determined by the current mean values ​​µ c and current standard deviations σ c given, determined and compared to diagnose an anomaly, see Fig. 4a. The consistency probability is the value of the integral of the probability density function PDF of the short-term model over the first range (-∞;µ m + 2σ m ) of the long-term model. The anomaly probability is the value of the integral of the probability density function PDF of the short-term model over the second range (µm + 2σ m ; +∞) of the long-term model. In Fig. 4a, the short-term model is consistent with the long-term model, so the current acoustic detection capability of the acoustic sensor mic0-mic7 is normal.

[0041] Fig. Figure 4b shows another example of a consistency probability and anomaly probability. In Fig. 4b the short-term model is abnormally louder.

[0042] Fig. Figure 4c, for example, shows a normally loud and a normally quiet short-term model.

[0043] Fig. 5 shows an arrangement of the acoustic sensors mic0-mic7 on a vehicle 1. The acoustic sensors mic0-mic7 are arranged symmetrically to a vehicle longitudinal axis, two acoustic sensors mic5 and mic6 at the front on the shock absorbers, two acoustic sensors mic0 and mic1 at the front on the roof, two acoustic sensors mic2 and mic3 at the rear on the roof and two acoustic sensors mic4 and mic7 at the rear on the shock absorbers of vehicle 1.

[0044] Fig. Figure 6 shows an analysis of a self-diagnosis of the acoustic sensors mic0-mic7. For each of the acoustic sensors mic0-mic7, a short-term model is created, which is carried out at a speed of 48 km / h with, for example, the Fig. 5, is compared with a long-term model comprising 1212 samples. The current recordings, i.e., the current acoustic detection capabilities, for each of the acoustic sensors mic0-mic7 are consistent with the respective long-term model of the corresponding acoustic sensor mic0-mic7. The respective anomaly probabilities are negligible.

[0045] Fig. Figure 7 shows a further analysis of a self-diagnosis of the acoustic sensors mic0-mic7, with the short-term model being retracted at a speed of 53 km / h. The short-term models for the acoustic sensors mic5 and mic6 and the short-term models for the acoustic sensors mic4 and mic7 are consistent with the respective long-term models. The short-term models for the acoustic sensors mic2 and mic3 and the short-term models for the acoustic sensors mic0 and mic1 deviate from the respective long-term models. The short-term models for the acoustic sensors mic2 and mic3 are similarly louder. The short-term models for the acoustic sensors mic0 and mic1 are also similarly louder. This is based on the similar arrangement position and similar airflows of the pairs of acoustic sensors mic5 and mic6, mic4 and mic7, mic2 and mic3, and mic0 and mic1. These pairs are considered as groups BG of acoustic sensors. Fig. Figure 6 shows that the acoustic sensors of groups BG are similarly affected by transient noise and model inconsistencies. The short-term models of the acoustic sensors mic2 and mic3 are both similarly louder, analogous to mic0 and mic1.

[0046] In the Fig. The respective groups BG are marked in the arrangement of the acoustic sensors mic0-mic7 shown in Figure 8.

[0047] As a further verification option, Fig. 9 the analysis of a self-diagnosis of groups of acoustic sensors using pair consistency.

[0048] Fig. Figure 10 schematically shows the diagnostic procedure. In step V1, a long-term model is generated for at least one of the acoustic sensors. In step V2, the long-term model mean values ​​µ are calculated for each frequency component f. m and the long-term model standard deviations σ mdetermined. In step V3, a short-term model is run at a current speed. In step V4, the short-term model is checked for consistency with the long-term model to diagnose anomalies. Reference symbol T Duration of a measurement recording t Time size sliding window N number of samples FFT fast Fourier transform B Block NF block number µ mean σ standard deviation f frequency component M union operation ASM acoustic sensor model PDF probability density function mic0-mic7 acoustic sensors BG Group 1 vehicle V1-V4 process steps

Claims

[1] Computer-implemented method for diagnosing an acoustic detection capability of at least one acoustic sensor (mic0-mic7), the method comprising the steps: • Generating at least one acoustic sensor model (ASM), the model (ASM) comprising audio data of the at least one acoustic sensor (mic0-mic7), wherein ○ the audio data comprise sequences of at least one sound field variable of a sound field outside a vehicle (1) and the sequences are obtained while the vehicle (1) is traveling at different speeds (V1); ○ Statistics on the audio data are determined by dividing the audio data obtained in speed ranges into frequency components (f) and calculating an average value (µ m ) of the sound field size for this frequency component (f) and a standard deviation (σ m) the sound field size for this frequency component (f) can be determined (V2); • Obtaining current mean values ​​(µ c ) and current standard deviations (σ c ) the sound field size for the frequency components (f) of a current sequence of the sound field size, wherein the current sequence was recorded (V3) with the acoustic sensor (mic0-mic7) while the vehicle (1) is traveling at a current speed; • Compare the current mean values ​​(µ c ) and current standard deviations (σ c ) with the mean values ​​(µ m ) and standard deviations (σ m ) from the speed range of the acoustic sensor model (ASM) in which the current speed lies, whereby in case of statistical significance of the current mean values ​​(µ c ) and current standard deviations (σ c ) the acoustic detection capability of the acoustic sensor (mic0-mic7) is diagnosed as abnormal (V4). [2] Method according to claim 1, wherein the acoustic sensor model (ASM) is generated from long-term recordings of the acoustic sensor (mic0-mic7) and is compared with the current mean values ​​(µ c ) and current standard deviations (σ c ) is incremented. [3] Method according to one of the preceding claims, wherein the sequences comprise a sound pressure of a background noise. [4] Method according to one of the preceding claims, wherein the audio data are processed block by block, wherein a block size is given by a number (N) of samples of the audio data at a certain time and the sequences comprise a block number (NF) of blocks (B). [5] Method according to claim 4, wherein the audio data in each block (B) are divided into equal frequency components (f) and the mean values ​​(µ m) of the sound field quantities per frequency component (f) are determined by adding the values ​​of the sound field quantities for this frequency component (f) from the blocks (B) and dividing the sum by the number of blocks (NF). [6] Method according to one of the preceding claims, wherein the frequency components (f) are calculated using the fast Fourier transform (FFT). [7] Method according to one of the preceding claims, wherein for each frequency component (f) a probability density function (PDF) which is determined by the respective current mean value (µ c ) and the respective current standard deviation (σ c ), in a first area (µ m - 2σ m ; µ m + 2σ m ) and a second area (µ m + 2σ m ; +∞) is integrated, where µ m the corresponding mean and σ mthe corresponding standard deviation from the acoustic sensor model (ASM), and in the case that the integral over the second range (µ m + 2σ m ; +∞) is greater than the integral over the first region (-∞; µ m + 2σ m ), the acoustic detection capability of the acoustic sensor (mic0-mic7) is diagnosed as abnormally louder. [8] Method according to claim 7, wherein in the case that the integral is over a third range (-∞; µ m - 2σ m ) is greater than the integral over the first region (µ m - 2σ m ; µ m + 2σ m ), the acoustic detection capability of the acoustic sensor (mic0-mic7) is diagnosed as abnormally quieter. [9] Method according to one of the preceding claims, wherein • if there are several acoustic sensors (mic0-mic7), an acoustic sensor model (ASM) is generated for each of the acoustic sensors (mic0-mic7); • when processing the audio data in blocks, the blocks (B) comprise one channel for each of the acoustic sensors (mic0-mic7); • each channel comprises a number of samples of the audio data; • in each channel the mean values ​​(µ m ) are formed according to claim 5 and standard deviations (σ m ) are obtained; • in each channel the probability density function (PDF) which is determined by the respective current mean value (µ c ) and the respective current standard deviation (σ c ), over the first (µ m - 2σ m ; µ m + 2σ m ), second (µ m + 2σ m ; +∞) and third area (-∞; µ m - 2σ m ) is integrated according to claim 7 and claim 8; • based on a comparison of the integrals obtained in each channel, the acoustic detection capability of each acoustic sensor (mic0-mic7) is diagnosed as normal, abnormally louder, or abnormally quieter. [10] Method according to one of the preceding claims, wherein the model formation takes place before use of the acoustic sensor(s) (mic0-mic7) in the vehicle (1) or with a memory logic by means of which the acoustic sensor model(s) (ASM) are / are linked with the current mean values ​​(µ c ) and current standard deviations (σ c ) are incremented during operation of the vehicle (1). [11] Computer program for diagnosing an acoustic detection capability of at least one acoustic sensor (mic0-mic7), the computer program comprising program instructions which cause the computer to carry out the method according to one of the preceding claims when the computer program is loaded onto the computer or executed. [12] Computer-implemented method for diagnosing an acoustic detection capability of a vehicle (1), wherein a plurality of acoustic sensors (mic0-mic7) are arranged on a frame and / or an outer skin of the vehicle (1) and for each of the acoustic sensors (mic0-mic7) its acoustic detection capability is diagnosed according to the method according to one of claims 1 to 10. [13] Method according to claim 12, wherein the acoustic sensors (mic0-mic7) which record a similar sound field due to their arrangement on the vehicle (1) or which are exposed to a similar air flow during a journey of the vehicle (1) form a group and anomalies in the acoustic detection capability are diagnosed by comparing the acoustic detection capability within a group of acoustic sensors (mic0-mic7).

Images