CN116092519A - Vehicle fault detection method and system - Google Patents
Vehicle fault detection method and system Download PDFInfo
- Publication number
- CN116092519A CN116092519A CN202111295562.XA CN202111295562A CN116092519A CN 116092519 A CN116092519 A CN 116092519A CN 202111295562 A CN202111295562 A CN 202111295562A CN 116092519 A CN116092519 A CN 116092519A
- Authority
- CN
- China
- Prior art keywords
- sound data
- vehicle
- training
- detected
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 139
- 238000000034 method Methods 0.000 claims abstract description 108
- 239000013598 vector Substances 0.000 claims abstract description 81
- 238000001228 spectrum Methods 0.000 claims abstract description 67
- 230000008569 process Effects 0.000 claims abstract description 56
- 238000012545 processing Methods 0.000 claims abstract description 53
- 238000006243 chemical reaction Methods 0.000 claims abstract description 25
- 238000012549 training Methods 0.000 claims description 142
- 238000005070 sampling Methods 0.000 claims description 45
- 238000013528 artificial neural network Methods 0.000 claims description 25
- 238000012795 verification Methods 0.000 claims description 13
- 238000001914 filtration Methods 0.000 claims description 10
- 238000012937 correction Methods 0.000 claims description 5
- 238000010606 normalization Methods 0.000 abstract description 4
- 230000006870 function Effects 0.000 description 15
- 238000010586 diagram Methods 0.000 description 9
- 230000007547 defect Effects 0.000 description 8
- 238000012423 maintenance Methods 0.000 description 8
- 230000000694 effects Effects 0.000 description 6
- 238000010200 validation analysis Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000005236 sound signal Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000009432 framing Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 230000008439 repair process Effects 0.000 description 2
- 230000000717 retained effect Effects 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
- G10L25/51—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
- G10L25/18—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters the extracted parameters being spectral information of each sub-band
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
- G10L25/21—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters the extracted parameters being power information
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/27—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
- G10L25/30—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
A vehicle fault detection method and system, the method includes: acquiring the model of a vehicle to be detected of the vehicle to be detected and acquiring sound data to be detected in the running process of the vehicle to be detected; performing time-frequency conversion processing on the sound data to be detected to obtain a frequency power spectrum vector of the sound data to be detected; sharpening the frequency power spectrum vector of the sound data to be detected to obtain characteristic data of the sound data to be detected; and inputting the characteristic data of the sound data to be detected into a pre-trained vehicle fault detection model corresponding to the model of the vehicle to be detected, and obtaining a fault detection result of the vehicle to be detected, which is output by the model. The invention can be implemented without high-precision sound collection equipment, reduces the implementation cost of fault detection and is easy to popularize and apply. The embodiment of the invention also ensures that the detection processing is more targeted by means of sharpening processing, normalization processing and the like, reduces the calculated amount of a detection scheme and improves the robustness of vehicle fault detection.
Description
Technical Field
The invention relates to the technical field of vehicle fault detection, in particular to a vehicle fault detection method and system.
Background
Certain noise (referred to herein as sound data) may be generated during the travel of the vehicle, and may be related to the running state of the vehicle itself. For example, the noise of a vehicle varies when different types of faults (e.g., engine faults, belt faults, wheel faults, etc.) occur in the vehicle. By analyzing the sound in the running process of the vehicle, the method is helpful for helping to locate the fault type of the vehicle and providing references of related information for vehicle owners and maintenance personnel.
In the prior art, the fault type is determined based on noise data in the running process of the vehicle, and the fault type is usually determined by means of personal experience accumulated by professional maintenance staff in long-term maintenance work, so that the fault type is difficult to popularize and apply in a large range. Another related vehicle fault detection method provided by the prior art, such as a method for identifying a railway vehicle wheel set defect by utilizing a voiceprint AI technology disclosed in the chinese patent application publication No. CN111307939a, uses the AI technology to find common voiceprint characteristics in a large number of fault audio samples, so that the method is very sensitive to internal defects of some wheel sets, and can solve the problems that the existing railway industry has overlong period aiming at the wheel set defect detection method, cannot meet the train maintenance requirement in running condition at any moment and is easy to generate missed detection. Specifically, the method comprises the following steps:
1) Selecting a subway vehicle which is detected manually and has determined that the wheel set defect exists as an AI training input source of a voiceprint fault model, and installing a miniature pickup to collect voiceprint information when the wheel set passes at the position of the vertical face of the track.
2) The collected voiceprints are subjected to framing treatment, and the purpose of framing is to convert macroscopically unstable audio signals into microscopically stable voiceprint frame signals which can be subjected to mathematical treatment.
3) The obtained voiceprint frame signal is windowed in order to gradually change the amplitude of each frame signal to 0 at both ends, thereby improving the resolution.
4) The pre-emphasis processing is carried out on the voiceprint frame, so that the purpose is to reduce the influence of sharp noise and improve the high-frequency part.
5) And short-time energy of voiceprint data is obtained through formula calculation.
6) The zero-crossing rate of the voiceprint data is obtained through formula calculation, and the purpose of the method is to extract the times that the signal waveform passes through the horizontal axis (zero level) in one frame of voiceprint.
7) A training vector set of the GMM model and the target fault training audio is given; and calculating the similarity of the ith Gaussian distribution.
8) And carrying out sub-score statistics by using the similarity of Gaussian distribution and the mixture weighted value, the average value vector and the variance, wherein the new full statistics generated by the training data are used for updating the full statistics of the ith mixed member of the UBM to obtain the target vehicle fault voiceprint model.
9) The algorithm is used for preparing a wheel set voiceprint fault detection program, and a trained voiceprint fault model is generated into a database to serve as a standard sample for identifying wheel set defects.
10 When the vehicle wheel set defect detection is carried out, the miniature pickup is arranged at the vertical face of the track to collect voiceprint information when the wheel set passes through, the voiceprint information is converted into digital signals through the digital-analog collection module and is input into a computer, and the weights, the mean values and the variance vectors after being processed by the wheel set fault voiceprint detection program are compared with standard samples.
11 When the fitting degree of the three vector values of the measured value and the standard sample is more than 90%, the defect of the corresponding wheel set shown by the vehicle fault voiceprint model can be judged.
As can be seen, the prior art vehicle fault locating method generally has the following problems:
1) The fault location is carried out by experience or attempt of maintenance personnel, so that the fault location cost is high and the maintenance efficiency is low;
2) Solutions proposed for specific fault types of specific vehicle types (such as wheel set defects of rail vehicles) have no versatility;
3) The expensive high-precision recording equipment is required for sound collection of the sound source, and the detection cost is high, so that the wide application of the high-precision recording equipment is limited.
Disclosure of Invention
At least one embodiment of the invention provides a vehicle fault detection method and system, which can detect various types of vehicle faults and reduce fault detection cost.
According to one aspect of the present invention, at least one embodiment provides a vehicle fault detection method, including:
acquiring the model of a vehicle to be detected of the vehicle to be detected and acquiring sound data to be detected in the running process of the vehicle to be detected;
performing time-frequency conversion processing on the sound data to be detected to obtain a frequency power spectrum vector of the sound data to be detected; sharpening the frequency power spectrum vector of the sound data to be detected to obtain characteristic data of the sound data to be detected;
and inputting the characteristic data of the sound data to be detected into a pre-trained vehicle fault detection model corresponding to the model of the vehicle to be detected, and obtaining a fault detection result of the vehicle to be detected, which is output by the model.
Furthermore, according to at least one embodiment of the present invention, before acquiring the model of the vehicle to be inspected and the sound data to be inspected, the method further includes:
obtaining training data of at least one vehicle model, wherein the training data of each vehicle model comprises a plurality of pieces of training sound data of the vehicle model and a vehicle fault type corresponding to each piece of training sound data;
For each vehicle model, training a corresponding vehicle fault detection model according to the following steps:
performing time-frequency conversion processing on the training sound data to obtain frequency power spectrum vectors of the training sound data; sharpening the frequency power spectrum vector of the training sound data to obtain characteristic data of the training sound data; coding the vehicle fault type corresponding to the training sound data to obtain a fault code corresponding to the training sound data;
and training the artificial neural network by utilizing the characteristic data of the training sound data and the corresponding fault codes to obtain a vehicle fault detection model corresponding to the vehicle model.
Furthermore, according to at least one embodiment of the present invention, the time-frequency conversion process includes:
performing band-pass filtering processing on target sound data to obtain first sound data, normalizing the first sound data by using the absolute value of the maximum amplitude in the first sound data to obtain second sound data, wherein the target sound data is to-be-detected sound data or training sound data;
segmenting the second sound data according to a preset sampling frequency, and dividing the second sound data into a plurality of sampling point sequences with preset lengths;
And carrying out FFT processing on each sampling point sequence to obtain a first vector corresponding to each sampling point sequence, and calculating a first frequency power spectrum vector corresponding to the first vector to obtain a frequency power spectrum vector of the target sound data.
Furthermore, according to at least one embodiment of the present invention, before performing FFT processing on each sampling point sequence, the method further includes:
for each sample point sequence, a symmetric hanning window function is used for correction.
Further, according to at least one embodiment of the present invention, among the plurality of sampling point sequences, adjacent sampling point sequences have portions overlapping each other.
Furthermore, in accordance with at least one embodiment of the present invention, the sharpening process includes:
for each frequency power spectrum vector of target sound data, firstly setting the first element in the frequency power spectrum vector to 0, and then sequentially judging whether the element is larger than the left and right adjacent elements from the second element, wherein for the last element, judging whether the element is larger than the left adjacent element, if so, marking the element as invariable; and finally, keeping all the element values marked as invariable unchanged, and setting other element values to 0, thereby obtaining the characteristic data of the target sound data.
Furthermore, in accordance with at least one embodiment of the present invention, the sharpening process further comprises:
and setting elements with the median value of the characteristic data of the target sound data smaller than a preset threshold value to 0.
In addition, according to at least one embodiment of the present invention, in the process of training the artificial neural network by using the feature data of the training sound data and the corresponding fault codes, the detection performance of the vehicle fault detection model obtained by training is verified by using the verification data set, if the detection performance obtained after M consecutive cycles is not improved, the training process is ended, and a model with optimal detection performance is selected from the vehicle fault detection models obtained by current training, where M is a preset positive integer, as a final vehicle fault detection model.
In addition, according to at least one embodiment of the present invention, the artificial neural network includes an input layer, an output layer, and a plurality of hidden layers, wherein the number of nodes of the input layer is the same as the number of elements in the frequency power spectrum vector of the target sound data, and the number of nodes of the output layer is the same as the type of the vehicle fault type corresponding to the vehicle model.
Further, according to at least one embodiment of the present invention, the plurality of pieces of training sound data of the vehicle model include: the method comprises the steps of collecting original sound data in the running process of the vehicle model, and adding various real background noises to the original sound data to obtain expanded sound data.
Further, according to at least one embodiment of the present invention, the failure detection result is any one of the following:
only one type of fault;
a potential fault list comprising a plurality of fault types and probabilities for each fault type;
and selecting a preset number of fault types according to the probability order of the fault types.
According to another aspect of the present invention, at least one embodiment provides a vehicle fault detection system, comprising: the first acquisition module is used for acquiring the model of the vehicle to be detected and the sound data to be detected acquired in the running process of the vehicle to be detected;
the characteristic acquisition module is used for performing time-frequency conversion processing on the sound data to be detected to obtain a frequency power spectrum vector of the sound data to be detected; sharpening the frequency power spectrum vector of the sound data to be detected to obtain characteristic data of the sound data to be detected;
the fault detection module is used for inputting the characteristic data of the sound data to be detected into a pre-trained vehicle fault detection model corresponding to the model of the vehicle to be detected, and obtaining a fault detection result of the vehicle to be detected, which is output by the model.
Furthermore, according to at least one embodiment of the present invention, there is also provided:
Model training module for:
obtaining training data of at least one vehicle model, wherein the training data of each vehicle model comprises a plurality of pieces of training sound data of the vehicle model and a vehicle fault type corresponding to each piece of training sound data;
for each vehicle model, training a corresponding vehicle fault detection model according to the following steps:
performing time-frequency conversion processing on the training sound data to obtain frequency power spectrum vectors of the training sound data; sharpening the frequency power spectrum vector of the training sound data to obtain characteristic data of the training sound data; coding the vehicle fault type corresponding to the training sound data to obtain a fault code corresponding to the training sound data;
and training the artificial neural network by utilizing the characteristic data of the training sound data and the corresponding fault codes to obtain a vehicle fault detection model corresponding to the vehicle model.
Furthermore, according to at least one embodiment of the present invention, the time-frequency conversion process includes:
performing band-pass filtering processing on target sound data to obtain first sound data, normalizing the first sound data by using the absolute value of the maximum amplitude in the first sound data to obtain second sound data, wherein the target sound data is to-be-detected sound data or training sound data;
Segmenting the second sound data according to a preset sampling frequency, and dividing the second sound data into a plurality of sampling point sequences with preset lengths;
and carrying out FFT processing on each sampling point sequence to obtain a first vector corresponding to each sampling point sequence, and calculating a first frequency power spectrum vector corresponding to the first vector to obtain a frequency power spectrum vector of the target sound data.
Furthermore, according to at least one embodiment of the present invention, the time-frequency conversion process further includes:
before FFT processing is performed on each sample point sequence, correction is performed using a symmetrical hanning window function for each sample point sequence.
Further, according to at least one embodiment of the present invention, among the plurality of first sampling point sequences, adjacent first sampling point sequences have portions overlapping each other.
Furthermore, in accordance with at least one embodiment of the present invention, the sharpening process includes:
for each frequency power spectrum vector of target sound data, firstly setting the first element in the frequency power spectrum vector to 0, and then sequentially judging whether the element is larger than the left and right adjacent elements from the second element, wherein for the last element, judging whether the element is larger than the left adjacent element, if so, marking the element as invariable; and finally, keeping all the element values marked as invariable unchanged, and setting other element values to 0, thereby obtaining the characteristic data of the target sound data.
Furthermore, in accordance with at least one embodiment of the present invention, the sharpening process further comprises:
and setting elements with the median value of the characteristic data of the target sound data smaller than a preset threshold value to 0.
Furthermore, according to at least one embodiment of the present invention, there is also provided:
the model verification module is used for verifying the detection performance of the vehicle fault detection model obtained by training by utilizing the verification data set in the process of training the artificial neural network by utilizing the feature data of the training sound data and the corresponding fault codes, ending the training process if the detection performance obtained after M continuous cycles is not improved, and selecting the model with the optimal detection performance from the vehicle fault detection models obtained by current training as the final vehicle fault detection model, wherein M is a preset positive integer.
In addition, according to at least one embodiment of the present invention, the artificial neural network includes an input layer, an output layer, and a plurality of hidden layers, wherein the number of nodes of the input layer is the same as the number of elements in the frequency power spectrum vector of the target sound data, and the number of nodes of the output layer is the same as the type of the vehicle fault type corresponding to the vehicle model.
Further, according to at least one embodiment of the present invention, the plurality of pieces of training sound data of the vehicle model include: the method comprises the steps of collecting original sound data in the running process of the vehicle model, and adding various real background noises to the original sound data to obtain expanded sound data.
Further, according to at least one embodiment of the present invention, the failure detection result is any one of the following:
only one type of fault;
a potential fault list comprising a plurality of fault types and probabilities for each fault type;
and selecting a preset number of fault types according to the probability order of the fault types.
According to another aspect of the invention, at least one embodiment provides a computer-readable storage medium having stored thereon a program which, when executed by a processor, implements the steps of the method as described above.
Compared with the prior art, the vehicle fault detection method and system provided by the embodiment of the invention can be implemented without high-precision sound acquisition equipment, so that the implementation cost of fault detection is reduced, and the vehicle fault detection method and system are easy to popularize and apply. In addition, the embodiment of the invention ensures that the detection processing is more targeted by means of sharpening processing, normalization processing and the like, reduces the calculated amount of a detection scheme and improves the robustness of vehicle fault detection.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to designate like parts throughout the figures. In the drawings:
fig. 1 is a schematic flow chart of a vehicle fault detection method according to an embodiment of the present invention;
FIG. 2 is a schematic flow chart of model training in an embodiment of the invention;
FIG. 3 is a schematic flow chart of a time-frequency conversion process according to an embodiment of the present invention;
FIG. 4 is a diagram illustrating an exemplary architecture of an artificial neural network according to an embodiment of the present invention;
FIG. 5 is a diagram of a simulated example of training loss and validation loss in accordance with an embodiment of the present invention;
FIG. 6 is a diagram of a simulation example of training accuracy and verification accuracy in accordance with an embodiment of the present invention;
FIG. 7 is a schematic diagram of a vehicle fault detection system according to an embodiment of the present invention;
FIG. 8 is a schematic diagram of a vehicle fault detection system according to an embodiment of the present invention;
FIG. 9 is a schematic diagram of another configuration of a vehicle fault detection system according to an embodiment of the present invention;
Fig. 10 is a schematic structural diagram of a vehicle fault detection system according to an embodiment of the present invention.
Detailed Description
Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
The terms first, second and the like in the description and in the claims, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that embodiments of the present application described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus. "and/or" in the specification and claims means at least one of the connected objects.
The following description provides examples and does not limit the scope, applicability, or configuration as set forth in the claims. Changes may be made in the function and arrangement of elements discussed without departing from the spirit and scope of the disclosure. Various examples may omit, substitute, or add various procedures or components as appropriate. For example, the described methods may be performed in an order different than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.
As described in the background art, the prior art vehicle fault detection method, or the proposed solution for a specific fault type of a specific vehicle type, has no versatility or requires expensive and complex equipment to collect sound data, thereby limiting its wide application. According to the vehicle fault detection method provided by the embodiment of the invention, high-precision sound collection equipment is not needed, sound data can be collected by means of equipment widely used in daily life (such as simple equipment like a mobile phone microphone) and the like, previous experience is learned by utilizing algorithms such as deep learning and the like, and a maintenance engineer or a user is assisted to quickly know possible faults of the vehicle. The invention is easy to popularize and use as expensive professional acquisition equipment is not needed.
Referring to fig. 1, a vehicle fault detection method provided by an embodiment of the present invention includes:
and 11, acquiring the model of the vehicle to be detected and the sound data to be detected acquired in the running process of the vehicle to be detected.
Here, the embodiment of the invention can collect the sound generated in the running process of the vehicle to be detected by using the mobile phone microphone and other devices widely used by drivers and passengers to obtain the sound data to be detected. In addition, it is also easy to obtain the vehicle model of the vehicle to be inspected. It can be seen that the embodiments of the present invention can be implemented without expensive sound collection equipment. Specifically, for sound data acquisition, the sampling rate may be 48000Hz. The sampled sound data may be saved as a WAV formatted file. Of course, other sampling rates may be used in the embodiments of the present invention, and stored in other formats.
In order to facilitate processing of sound data, the embodiment of the invention transforms the sound data to be detected from a time domain to a frequency domain to obtain a frequency power spectrum vector thereof. In addition, the characteristic data of the sound data to be detected are obtained by sharpening the frequency power spectrum vector. The sharpening process can filter out the frequency characteristics with lower importance, so that the subsequent detection process is more targeted, and the subsequent calculation amount can be reduced.
And 13, inputting the characteristic data of the sound data to be detected into a pre-trained vehicle fault detection model corresponding to the model of the vehicle to be detected, and obtaining a fault detection result of the vehicle to be detected, which is output by the model.
Because vehicles of different models possibly have different structures and materials, the spectral characteristics of sound signals generated in respective running processes are also greatly different, and therefore, the embodiment of the invention trains corresponding vehicle fault detection models in advance for each vehicle model, inputs the characteristic data of the sound data to be detected obtained in the step 12 into the vehicle fault detection model corresponding to the vehicle model to be detected, and the models output fault detection results, thereby realizing detection of vehicle faults.
Here, the failure detection result output by the vehicle failure detection model may be a specific vehicle failure type, or may be a plurality of vehicle failure types and probability values of each vehicle failure type, which is not particularly limited in the embodiment of the present invention. That is, the failure detection result may be any one of the following:
1) And only one fault type is adopted, wherein the fault type is the vehicle fault type finally judged by the model, for example, the fault type with highest probability.
2) The potential fault list comprises a plurality of fault types and probabilities of each fault type, and the potential fault list provides a plurality of fault types, wherein the fault types can be fault types with probability values larger than a certain preset threshold by model judgment.
3) And selecting a preset number of fault types according to the probability order of the fault types. Here, the model selects L fault types according to the order of the probability of the fault types from high to low, and outputs the L fault types. L is a preset positive integer, such as 2 or 3.
Of course, the fault detection result may also take other forms, which are not particularly limited in the embodiment of the present invention.
Through the steps, the fault detection method can be implemented without depending on high-precision sound collection equipment, so that the implementation cost of fault detection is reduced, and the method is easy to popularize and apply. In addition, the embodiment of the invention ensures that the detection processing is more targeted by sharpening processing, thereby reducing the calculated amount of the detection scheme.
Before the step 11, the embodiment of the invention can be used for training in advance to obtain the corresponding vehicle fault detection model aiming at various vehicle models. Specifically, as shown in fig. 2, the model training process includes:
And 21, obtaining training data of at least one vehicle model, wherein the training data of each vehicle model comprises a plurality of pieces of training sound data of the vehicle model and a vehicle fault type corresponding to each piece of training sound data.
Here, the training sound data may be collected by a device such as a microphone of a mobile phone, and similarly, the sampling rate may be 48000Hz. The sampled sound data may be saved as a WAV formatted file. The training sound data that is typically collected may take on a wide variety of coding formats or different sampling frequencies. For example, sound data may be present in the video, which is extracted and stored in WAV format. For example, the sound data is stored in MP3 format, and the embodiment of the present invention may also convert the sound data into WAV format. For sound data samples with sampling rates different from 48000Hz, the corresponding frequency power spectrum may be converted into a power spectrum with sampling rate of 48000Hz by an approximately aligned method. Of course, other sampling rates may be used in the embodiments of the present invention, and stored in other formats.
In addition, the type of the vehicle fault corresponding to each training sound data can be obtained by a professional repair engineer through manual detection, and the data can be obtained from a database of the vehicle repair records.
Then, for each vehicle model, a corresponding vehicle fault detection model is trained according to the following steps:
Here, the vehicle fault type corresponding to the training sound data is encoded, and a one-hot code (one-hot) encoding may be used, that is, each element of the encoded vector corresponds to one fault type.
And step 23, training an artificial neural network by utilizing the feature data of the training sound data and the corresponding fault codes to obtain a vehicle fault detection model corresponding to the vehicle model.
Through the steps, the embodiment of the invention can train the corresponding vehicle fault detection models aiming at various vehicle models, so that when the fault detection is needed to be carried out on the sound data of a certain vehicle model, the relevant characteristic data can be input into the corresponding vehicle fault detection models for detection according to the flow shown in the figure 1.
The system of the embodiment of the invention does not require clean environment without background noise or high-precision recording equipment, but can record vehicle noise by using a microphone of a smart phone, so that various background noises exist in the vehicle noise. Background noise presents difficulties in feature extraction. For convenience of distinction, the vehicle operation noise data is referred to as sound data, the background noise is referred to as noise data, and noise is often mixed in the sound data.
In both the above steps 12 and 22, the time-frequency conversion processing is required for the sound data, and the sharpening processing is required for the frequency power spectrum vector, and the time-frequency conversion processing and the sharpening processing will be described in detail below. The following description is merely an example of the processing manner that may be adopted by the present invention, and is not intended to limit the present invention.
The extraction of the characteristic data mainly comprises two parts of representation transformation and noise reduction of the sound data. The embodiments of the present invention will use the frequency power spectrum of sound data as a feature related to the type of vehicle fault. A fast fourier transform (Fast Fourier Transform, FFT) will be used to convert the representation of the sound data from the time domain to the frequency domain, resulting in a frequency power spectrum.
As an example, as shown in fig. 3, the time-frequency conversion process in the above step 12 and step 22 includes:
Here, the high-frequency and low-frequency parts in the target sound data may be removed by band-pass filtering, for example, the cut-off frequency of the band-pass filtering may be set to 20Hz and 12000Hz, that is, the allowable passing signal frequency is 20 to 12000Hz. Of course, the band-pass range can be enlarged or reduced according to the processing performance of the device. After bandpass filtering, the target sound data is normalized with the maximum absolute value of the amplitude to improve the robustness of the subsequent detection process.
And step 32, segmenting the second sound data according to a preset sampling frequency, and dividing the second sound data into a plurality of sampling point sequences with preset lengths.
Here, the second sound data may be sampled at a preset sampling frequency (e.g., 48000 Hz) to obtain a plurality of consecutive sampling points, and then the sampling points may be divided into a plurality of sampling point sequences of a preset length. Each sample point sequence has the same length, and the specific length can be set according to the design of the FFT transformation process. For example, each sample point sequence has a length of 4096, i.e., 4096 sample points are included. In addition, in order to prevent loss of information, adjacent sampling point sequences among the plurality of sampling point sequences have portions overlapping each other, for example, two adjacent blocks overlap by 50%. When dividing the sampling point sequence, if the length of the last sampling point sequence is smaller than the preset length, discarding the last sampling point sequence, so that all the finally obtained sampling point sequences have the same length.
Optionally, in the above step 32, in order to reduce the interference signal, a symmetrical hanning window function may be used to perform correction processing for each sampling point sequence, and then step 33 may be performed.
And step 33, performing FFT processing on each sampling point sequence to obtain a first vector corresponding to each sampling point sequence, and calculating a first frequency power spectrum vector corresponding to the first vector to obtain a frequency power spectrum vector of the target sound data.
Here, assuming that each sample point sequence has a length of 4096, invoking FFT for each sample point sequence can result in a first vector f=including 2049 elements<f 0 ,f 1 ,…,f 2048 >Wherein f i The i-th element in the first vector F is represented, and the value range of i is 0-2049. For the first vector F, its corresponding frequency-power spectral vector s=can be calculated as follows<s 0 ,s 1 ,…,s 2048 >，s i Representing the i-th element in the vector:
In the above formula, 2E-5 represents 2×10 -5 ，abs(f i ) Representing the element f i Modulo, c is a window compensation value, L represents the length of the sample point sequence (e.g., 4096), and sum (hann) represents the sum of the hanning window functions.
After obtaining the frequency power spectrum vector of the target sound data, the embodiment of the present invention further needs to sharpen the frequency power spectrum vector of the target sound data in the above steps 12 and 22. Here, the sharpening process specifically includes:
For each frequency power spectrum vector of the target sound data, firstly, setting the first element in the frequency power spectrum vector to 0, and then, starting from the second element, judging whether the element is larger than the left adjacent element and the right adjacent element in sequence, wherein for the last element, judging whether the element is larger than the left adjacent element, that is, for the last element, only comparing with the one adjacent element because only one adjacent element exists. If the element is greater than its left and right neighbors (for the last element, only the neighbors to the left are needed), then the element is marked as immutable; and finally, keeping all the element values marked as invariable unchanged, and setting the other element values which are not marked as invariable to 0, thereby obtaining the characteristic data of the target sound data. Step 13 or step 33 may then be entered.
Considering that some background noise may exist in the sound signal, the amplitude of the background noise is usually smaller, so a threshold may be set here, and in the embodiment of the present invention, elements with median values of feature data of the target sound data smaller than a preset threshold may be set to 0, and then step 13 or step 33 is performed, so that the influence of the background noise may be filtered, and the subsequent calculation amount may be reduced.
Through the above processing, the embodiment of the invention converts the WAV data into the representation in the frequency domain through Fast Fourier Transform (FFT), the frequency power spectrum generated by the FFT is sharpened through the above processing to reduce noise, and the processed frequency power spectrum is used as the characteristic data of model training.
In the sharpening process, the embodiment of the invention reduces the influence of noise by sharpening and filtering noise by using the threshold value. For example, in the sharpening process, only those elements that are greater than the left and right adjacent element values are retained, and those elements that are ignored are reset by using a particular default value (e.g., zero). In this way, only local maxima may be retained after the sharpening process. In the filtering process based on the preset threshold, elements having a value smaller than the threshold may be ignored.
In addition, the robustness of vehicle fault detection is improved in a plurality of ways by taking the difference between the sound collection device and the background environment into consideration. For example, normalization processing is performed on sound data, the first element in the frequency-power spectrum vector is set to a zero value (the first element represents a zero frequency term), and so on. In addition, the embodiment of the invention can also improve the robustness by expanding the training samples. For example, the sample is expanded by intentionally adding noise, at which time the pieces of training sound data of the vehicle model include: the method comprises the steps of collecting original sound data in the running process of the vehicle model, and adding various real background noises to the original sound data to obtain expanded sound data.
By the above processing, taking the sample point sequence length of 4096 as an example, a feature matrix can be finally obtained. Assuming that given target sound data (training sound data or to-be-detected sound data) is divided into N blocks, the feature matrix has N rows and 2049 columns.
The embodiment of the invention adopts an artificial neural network (Artificial Neural Network, ANN) of a Multi-Layer Perception (MLP) to establish a vehicle fault detection model. In addition, early-stop techniques are used in the model training process to find the most appropriate model. During model training, the detection accuracy and error of model training and testing reflect the performance of the model, and the embodiment of the invention performs model verification by verifying the prediction result of the data set. The verification data set may include a plurality of verification sound data and a vehicle fault type corresponding to each verification sound data for performing performance verification on the detection accuracy of the model.
The artificial neural network used as the vehicle fault detection model comprises an input layer, an output layer and a plurality of hidden layers, wherein the number of nodes of the input layer is the same as the number of elements in the frequency power spectrum vector of the target sound data, and the number of nodes of the output layer is the same as the type of the vehicle fault type corresponding to the vehicle model. Fig. 4 shows an example of an artificial neural network according to an embodiment of the present invention, where the number of input layer nodes of the network is 2049 and the number of output layer nodes is equal to the number N of detectable vehicle fault types.
As shown in fig. 4, the MLP is a key part of the overall network model. The feature data may be processed with a BatchNorm function before entering the MLP; the output of the MLP may be activated with a Sigmoid function. The MLP in fig. 4 is designed with 3 hidden layers, 4096, 1024 and 256 nodes, respectively. Each layer contains one dense linear operation and one ReLU activation function.
The embodiment of the invention can use the BCEWITHLogitsLoss function as a loss function of training optimization, and the detection precision is used as a measure of the detection capability. Here, the detection accuracy may be defined as a ratio of the number of prediction types of faults exactly the same as the real types to the total number of predictions.
Early stop techniques were used during model training. Specifically, in the process of training the artificial neural network by using the feature data of the training sound data and the corresponding fault codes, verifying the detection performance of the vehicle fault detection model obtained by training by using a verification data set, ending the training process if the detection performance obtained after continuous M times of cycles is not improved, and selecting a model with optimal detection performance from the vehicle fault detection models obtained by current training as a final vehicle fault detection model, wherein M is a preset positive integer.
For example, it may be set that once the validation loss (epochs) is no longer reduced for 150 cycles (epochs), training will automatically stop and a training model that has minimal validation loss up to now is saved. As one example, during training, the batch size of the feature input is set to 32, the learning rate 1E-7, and the learning rate decay rate 0.1.
In the model training examples shown in fig. 5 and 6, the early stop is closed. It can be seen that the validation loss value decreases from the beginning and then increases after training 220 cycles. That is, the model is overfitted after 200 cycles of training, presumably to achieve minimum validation loss at round 213 training.
From the foregoing, it can be seen that embodiments of the present invention provide a method for analyzing vehicle operating noise to detect vehicle faults using machine learning techniques. Compared with the prior art, the embodiment of the invention is different in technical realization and application scene. In terms of technical implementation, the embodiment of the invention provides different feature extraction methods and designs different artificial neural network structures, and an Artificial Neural Network (ANN) of multi-layer perception (MLP) is used. From the perspective of application, the solution of the embodiment of the invention is not aimed at a certain part of the vehicle (such as an engine) but is aimed at the whole vehicle, and in addition, expensive high-precision recording equipment is not needed. Based on the above method, the embodiment of the invention can enable a user (such as a vehicle maintenance worker or a vehicle owner) to conveniently access the service provided based on the method so as to quickly determine the vehicle state.
The foregoing describes various methods of embodiments of the present invention. A system for implementing the above method will be further provided below.
Fig. 7 is a schematic diagram of an architecture of a vehicle fault detection system according to an embodiment of the present invention, where the system may include two subsystems, i.e., an on-line subsystem and an off-line subsystem. In the on-line subsystem, a user records and preprocesses the running sound of the repairing (but the fault is not clear) vehicle through the smart phone APP, and then sends the result to the vehicle fault detection service. The service checks to receive content submitted by the user and returns information such as a type code (id) of the potential failure. In the off-line subsystem, firstly, faults of various vehicle models and corresponding sound data are collected and stored, then the sound data are processed, the characteristic data of each fault are extracted, the extracted characteristic data and fault type information are sent to a neural network to train an MLP model, and finally, the trained model is deployed as an on-line vehicle fault detection service.
Referring to fig. 8, another structure of a vehicle fault detection system according to an embodiment of the present invention includes:
the first obtaining module 81 is configured to obtain a model of a vehicle to be inspected and sound data collected during operation of the vehicle to be inspected;
The feature acquisition module 82 is configured to perform time-frequency conversion processing on the sound data to be detected, so as to obtain a frequency power spectrum vector of the sound data to be detected; sharpening the frequency power spectrum vector of the sound data to be detected to obtain characteristic data of the sound data to be detected;
the fault detection module 83 is configured to input the feature data of the sound data to be detected to a pre-trained vehicle fault detection model corresponding to the model of the vehicle to be detected, and obtain a fault detection result of the vehicle to be detected, which is output by the model.
As shown in fig. 9, the vehicle fault detection system of the embodiment of the invention further includes:
obtaining training data of at least one vehicle model, wherein the training data of each vehicle model comprises a plurality of pieces of training sound data of the vehicle model and a vehicle fault type corresponding to each piece of training sound data;
for each vehicle model, training a corresponding vehicle fault detection model according to the following steps:
performing time-frequency conversion processing on the training sound data to obtain frequency power spectrum vectors of the training sound data; sharpening the frequency power spectrum vector of the training sound data to obtain characteristic data of the training sound data; coding the vehicle fault type corresponding to the training sound data to obtain a fault code corresponding to the training sound data;
And training the artificial neural network by utilizing the characteristic data of the training sound data and the corresponding fault codes to obtain a vehicle fault detection model corresponding to the vehicle model.
Optionally, the time-frequency conversion process includes:
performing band-pass filtering processing on target sound data to obtain first sound data, normalizing the first sound data by using the absolute value of the maximum amplitude in the first sound data to obtain second sound data, wherein the target sound data is to-be-detected sound data or training sound data;
segmenting the second sound data according to a preset sampling frequency, and dividing the second sound data into a plurality of sampling point sequences with preset lengths;
and carrying out FFT processing on each sampling point sequence to obtain a first vector corresponding to each sampling point sequence, and calculating a first frequency power spectrum vector corresponding to the first vector to obtain a frequency power spectrum vector of the target sound data.
Optionally, the time-frequency conversion process further includes:
before FFT processing is performed on each sample point sequence, correction is performed using a symmetrical hanning window function for each sample point sequence.
Optionally, among the plurality of first sampling point sequences, adjacent first sampling point sequences have portions overlapping each other.
Optionally, the sharpening process includes:
for each frequency power spectrum vector of target sound data, firstly setting the first element in the frequency power spectrum vector to 0, and then sequentially judging whether the element is larger than the left and right adjacent elements from the second element, wherein for the last element, judging whether the element is larger than the left adjacent element, if so, marking the element as invariable; and finally, keeping all the element values marked as invariable unchanged, and setting other element values to 0, thereby obtaining the characteristic data of the target sound data.
Optionally, the sharpening process further includes:
and setting elements with the median value of the characteristic data of the target sound data smaller than a preset threshold value to 0.
As shown in fig. 9, the vehicle fault detection system of the embodiment of the invention further includes:
the model verification module 85 is configured to verify, by using the verification data set, the detection performance of the trained vehicle fault detection model in the process of training the artificial neural network by using the feature data of the training sound data and the corresponding fault codes, and if the detection performance obtained after M consecutive cycles is not improved, end the training process, and select a model with the optimal detection performance from the vehicle fault detection models obtained by current training, as a final vehicle fault detection model, where M is a preset positive integer.
Optionally, the artificial neural network includes an input layer, an output layer and a plurality of hidden layers, wherein the number of nodes of the input layer is the same as the number of elements in the frequency power spectrum vector of the target sound data, and the number of nodes of the output layer is the same as the type of the vehicle fault type corresponding to the vehicle model.
Optionally, the plurality of training sound data of the vehicle model includes: the method comprises the steps of collecting original sound data in the running process of the vehicle model, and adding various real background noises to the original sound data to obtain expanded sound data.
Optionally, the fault detection result is any one of the following:
only one type of fault;
a potential fault list comprising a plurality of fault types and probabilities for each fault type;
and selecting a preset number of fault types according to the probability order of the fault types.
The system in this embodiment corresponds to the vehicle fault detection method, and the implementation manner in each of the embodiments described above is applicable to the embodiment of the system, so that the same technical effects can be achieved. The system provided by the embodiment of the invention can realize all the method steps realized by the embodiment of the method and can achieve the same technical effects, and the parts and the beneficial effects which are the same as those of the embodiment of the method in the embodiment are not described in detail.
Referring to fig. 10, a schematic structural diagram of another vehicle fault detection system according to an embodiment of the present invention includes: a processor 1001, a transceiver 1002, a memory 1003, a user interface 1004 and a bus interface.
In an embodiment of the present invention, the system further includes: a program stored in the memory 1003 and executable on the processor 1001.
The transceiver 1002 is configured to receive and transmit data under the control of the processor;
the processor 1001 is configured to read the computer program in the memory and perform the following operations:
acquiring the model of a vehicle to be detected of the vehicle to be detected and acquiring sound data to be detected in the running process of the vehicle to be detected;
performing time-frequency conversion processing on the sound data to be detected to obtain a frequency power spectrum vector of the sound data to be detected; sharpening the frequency power spectrum vector of the sound data to be detected to obtain characteristic data of the sound data to be detected;
and inputting the characteristic data of the sound data to be detected into a pre-trained vehicle fault detection model corresponding to the model of the vehicle to be detected, and obtaining a fault detection result of the vehicle to be detected, which is output by the model.
It can be appreciated that in the embodiment of the present invention, the computer program, when executed by the processor 1001, may implement the processes of the above-described embodiments of the vehicle fault detection method, and achieve the same technical effects, so that repetition is avoided and no further description is given here.
In fig. 10, a bus architecture may be comprised of any number of interconnected buses and bridges, and in particular, one or more processors represented by the processor 1001 and various circuits of the memory represented by the memory 1003. The bus architecture may also link together various other circuits such as peripheral devices, voltage regulators, power management circuits, etc., which are well known in the art and, therefore, will not be described further herein. The bus interface provides an interface. The transceiver 1002 may be a number of elements, i.e. comprising a transmitter and a receiver, providing a means for communicating with various other apparatus over a transmission medium. The user interface 1004 may also be an interface capable of interfacing with an inscribed desired device for a different user device, including but not limited to a keypad, display, speaker, microphone, joystick, etc.
The processor 1001 is responsible for managing the bus architecture and general processing, and the memory 1003 may store data used by the processor 1001 in performing operations.
The apparatus in this embodiment corresponds to the method shown in fig. 4, and the implementation manner in each embodiment is applicable to the embodiment of the apparatus, so that the same technical effects can be achieved. In this device, the transceiver 1002 and the memory 1003, and the transceiver 1002 and the processor 1001 may be communicatively connected through a bus interface, and the functions of the processor 1001 may be implemented by the transceiver 1002, and the functions of the transceiver 1002 may be implemented by the processor 1001. It should be noted that, the above device provided in the embodiment of the present invention can implement all the method steps implemented in the method embodiment and achieve the same technical effects, and detailed descriptions of the same parts and beneficial effects as those of the method embodiment in the embodiment are omitted herein.
In some embodiments of the present invention, there is also provided a computer-readable storage medium having stored thereon a program which, when executed by a processor, performs the steps of:
acquiring the model of a vehicle to be detected of the vehicle to be detected and acquiring sound data to be detected in the running process of the vehicle to be detected;
performing time-frequency conversion processing on the sound data to be detected to obtain a frequency power spectrum vector of the sound data to be detected; sharpening the frequency power spectrum vector of the sound data to be detected to obtain characteristic data of the sound data to be detected;
and inputting the characteristic data of the sound data to be detected into a pre-trained vehicle fault detection model corresponding to the model of the vehicle to be detected, and obtaining a fault detection result of the vehicle to be detected, which is output by the model.
When the program is executed by the processor, all the implementation modes in the vehicle fault detection method can be realized, the same technical effects can be achieved, and the repeated description is omitted here for avoiding the repetition.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, and are not repeated herein.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiment of the present invention.
In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method described in the embodiments of the present invention. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a ROM, a RAM, a magnetic disk, or an optical disk, etc.
The foregoing is merely illustrative of the present invention, and the present invention is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.
Claims (13)
1. A vehicle failure detection method, characterized by comprising:
acquiring the model of a vehicle to be detected of the vehicle to be detected and acquiring sound data to be detected in the running process of the vehicle to be detected;
performing time-frequency conversion processing on the sound data to be detected to obtain a frequency power spectrum vector of the sound data to be detected; sharpening the frequency power spectrum vector of the sound data to be detected to obtain characteristic data of the sound data to be detected;
and inputting the characteristic data of the sound data to be detected into a pre-trained vehicle fault detection model corresponding to the model of the vehicle to be detected, and obtaining a fault detection result of the vehicle to be detected, which is output by the model.
2. The method according to claim 1, wherein before acquiring the model of the vehicle to be inspected and the sound data to be inspected, the method further comprises:
obtaining training data of at least one vehicle model, wherein the training data of each vehicle model comprises a plurality of pieces of training sound data of the vehicle model and a vehicle fault type corresponding to each piece of training sound data;
for each vehicle model, training a corresponding vehicle fault detection model according to the following steps:
Performing time-frequency conversion processing on the training sound data to obtain frequency power spectrum vectors of the training sound data; sharpening the frequency power spectrum vector of the training sound data to obtain characteristic data of the training sound data; coding the vehicle fault type corresponding to the training sound data to obtain a fault code corresponding to the training sound data;
and training the artificial neural network by utilizing the characteristic data of the training sound data and the corresponding fault codes to obtain a vehicle fault detection model corresponding to the vehicle model.
3. The method of claim 2, wherein the time-to-frequency conversion process comprises:
performing band-pass filtering processing on target sound data to obtain first sound data, normalizing the first sound data by using the absolute value of the maximum amplitude in the first sound data to obtain second sound data, wherein the target sound data is to-be-detected sound data or training sound data;
segmenting the second sound data according to a preset sampling frequency, and dividing the second sound data into a plurality of sampling point sequences with preset lengths;
and carrying out FFT processing on each sampling point sequence to obtain a first vector corresponding to each sampling point sequence, and calculating a first frequency power spectrum vector corresponding to the first vector to obtain a frequency power spectrum vector of the target sound data.
4. The method of claim 3, further comprising, prior to FFT processing each sample point sequence:
for each sample point sequence, a symmetric hanning window function is used for correction.
5. A method as claimed in claim 3, wherein adjacent ones of the plurality of sequences of sample points have portions that overlap one another.
6. The method of claim 3, wherein the sharpening process comprises:
for each frequency power spectrum vector of target sound data, firstly setting the first element in the frequency power spectrum vector to 0, and then sequentially judging whether the element is larger than the left and right adjacent elements from the second element, wherein for the last element, judging whether the element is larger than the left adjacent element, if so, marking the element as invariable; and finally, keeping all the element values marked as invariable unchanged, and setting other element values to 0, thereby obtaining the characteristic data of the target sound data.
7. The method of claim 6, wherein the sharpening process further comprises:
and setting elements with the median value of the characteristic data of the target sound data smaller than a preset threshold value to 0.
8. The method according to claim 2, wherein in the training of the artificial neural network using the feature data of the training sound data and the corresponding fault codes, the verification data set is used to verify the detection performance of the vehicle fault detection model obtained by training, and if none of the detection performances obtained after M consecutive cycles is improved, the training process is ended, and the model with the optimal detection performance is selected from the vehicle fault detection models obtained by current training, as the final vehicle fault detection model, wherein M is a preset positive integer.
9. The method of claim 3, wherein the artificial neural network includes an input layer, an output layer, and a plurality of hidden layers, wherein the number of nodes of the input layer is the same as the number of elements in the frequency-power spectrum vector of the target sound data, and the number of nodes of the output layer is the same as the kind of the vehicle fault type corresponding to the vehicle model.
10. The method of claim 2, wherein the plurality of pieces of training sound data of the vehicle model includes: the method comprises the steps of collecting original sound data in the running process of the vehicle model, and adding various real background noises to the original sound data to obtain expanded sound data.
11. The method of claim 1, wherein the fault detection result is any one of:
only one type of fault;
a potential fault list comprising a plurality of fault types and probabilities for each fault type;
and selecting a preset number of fault types according to the probability order of the fault types.
12. A vehicle failure detection system, characterized by comprising:
the first acquisition module is used for acquiring the model of the vehicle to be detected and the sound data to be detected acquired in the running process of the vehicle to be detected;
the characteristic acquisition module is used for performing time-frequency conversion processing on the sound data to be detected to obtain a frequency power spectrum vector of the sound data to be detected; sharpening the frequency power spectrum vector of the sound data to be detected to obtain characteristic data of the sound data to be detected;
the fault detection module is used for inputting the characteristic data of the sound data to be detected into a pre-trained vehicle fault detection model corresponding to the model of the vehicle to be detected, and obtaining a fault detection result of the vehicle to be detected, which is output by the model.
13. The system as recited in claim 12, further comprising:
Model training module for:
obtaining training data of at least one vehicle model, wherein the training data of each vehicle model comprises a plurality of pieces of training sound data of the vehicle model and a vehicle fault type corresponding to each piece of training sound data;
for each vehicle model, training a corresponding vehicle fault detection model according to the following steps:
performing time-frequency conversion processing on the training sound data to obtain frequency power spectrum vectors of the training sound data; sharpening the frequency power spectrum vector of the training sound data to obtain characteristic data of the training sound data; coding the vehicle fault type corresponding to the training sound data to obtain a fault code corresponding to the training sound data;
and training the artificial neural network by utilizing the characteristic data of the training sound data and the corresponding fault codes to obtain a vehicle fault detection model corresponding to the vehicle model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111295562.XA CN116092519A (en) | 2021-11-03 | 2021-11-03 | Vehicle fault detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111295562.XA CN116092519A (en) | 2021-11-03 | 2021-11-03 | Vehicle fault detection method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116092519A true CN116092519A (en) | 2023-05-09 |
Family
ID=86206883
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111295562.XA Pending CN116092519A (en) | 2021-11-03 | 2021-11-03 | Vehicle fault detection method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116092519A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117409816A (en) * | 2023-12-14 | 2024-01-16 | 湖南华夏特变股份有限公司 | Equipment fault detection method and system based on sound signals |
-
2021
- 2021-11-03 CN CN202111295562.XA patent/CN116092519A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117409816A (en) * | 2023-12-14 | 2024-01-16 | 湖南华夏特变股份有限公司 | Equipment fault detection method and system based on sound signals |
| CN117409816B (en) * | 2023-12-14 | 2024-03-26 | 湖南华夏特变股份有限公司 | Equipment fault detection method and system based on sound signals |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108648748B (en) | Acoustic event detection method under hospital noise environment | |
| KR101493552B1 (en) | Signal judgment method, signal judgment apparatus, program, and signal judgment system | |
| CN109949823B (en) | DWPT-MFCC and GMM-based in-vehicle abnormal sound identification method | |
| CN106653032B (en) | Based on the animal sounds detection method of multiband Energy distribution under low signal-to-noise ratio environment | |
| CN102623009B (en) | Abnormal emotion automatic detection and extraction method and system on basis of short-time analysis | |
| CN104916289A (en) | Quick acoustic event detection method under vehicle-driving noise environment | |
| CN105841797A (en) | Window motor abnormal noise detection method and apparatus based on MFCC and SVM | |
| CN110398647B (en) | Transformer state monitoring method | |
| CN113077005B (en) | Anomaly detection system and method based on LSTM self-encoder and normal signal data | |
| CN104978507A (en) | Intelligent well logging evaluation expert system identity authentication method based on voiceprint recognition | |
| CN1949364B (en) | System and method for testing identification degree of input speech signal | |
| CN108198561A (en) | A kind of pirate recordings speech detection method based on convolutional neural networks | |
| CN113823293B (en) | Speaker recognition method and system based on voice enhancement | |
| CN112529177A (en) | Vehicle collision detection method and device | |
| CN112599134A (en) | Transformer sound event detection method based on voiceprint recognition | |
| CN113740635A (en) | Electrical equipment fault diagnosis method, terminal and multi-probe sensing device | |
| CN115618205A (en) | Portable voiceprint fault detection system and method | |
| CN116092519A (en) | Vehicle fault detection method and system | |
| CN116778964A (en) | Power transformation equipment fault monitoring system and method based on voiceprint recognition | |
| CN113505524B (en) | Rail transit noise contribution determining method, system, device and storage medium | |
| CN112149498A (en) | Online intelligent recognition system and method for abnormal sound of automobile complex part | |
| Zhang et al. | Fault diagnosis method based on MFCC fusion and SVM | |
| CN113963719A (en) | Deep learning-based sound classification method and apparatus, storage medium, and computer | |
| CN107993666B (en) | Speech recognition method, speech recognition device, computer equipment and readable storage medium | |
| KR101382356B1 (en) | Apparatus for forgery detection of audio file |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |