A vehicle fault diagnosis and early warning method and system based on vibration and noise

Abstract

The present invention discloses a vehicle fault diagnosis and early warning method and system based on vibration and noise, the method comprising the following steps: collecting vibration signals and noise signals corresponding to different vehicle fault types; obtaining corresponding time-frequency feature data, converting all time-frequency feature data into feature image data; constructing an image-based deep learning model, and using the feature image data for training to obtain a trained vehicle fault diagnosis model; using the trained vehicle fault diagnosis model to perform vehicle fault diagnosis and early warning. In the scheme of the present invention, the time-frequency feature image provides information on the fault signal in time and frequency, making the analysis of the fault signal more comprehensive. Through time-frequency analysis, transient features and frequency components in the signal can be identified, helping the model to better understand the fault mode. At the same time, the time-frequency feature image can effectively capture the characteristics of non-order vibration and noise, thereby providing more accurate information for fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle fault diagnosis, and in particular to a vehicle fault diagnosis and early warning method and system based on vibration and noise. Background Art

[0002] At present, the safety and comfort of automobile operation have become issues that major OEMs and consumers attach great importance to. Automobile operation safety includes active fault monitoring system and passive fault emergency system. Passive fault emergency system can help drivers and passengers avoid serious personal injury and increase the survival probability of people in the car after a vehicle accident. Active fault monitoring system mainly refers to fault warning issued in the form of alarm in the early stage of fault.

[0003] In recent years, the explosive growth of data and the emergence of high-performance computers have made it possible to discover vehicle faults in the early stages, allowing the industry to use artificial intelligence algorithms to process image, sound and other types of data to replace manual troubleshooting. In many cases, the assembly relationship of the mechanical structure of a vehicle will change in the early stages of a fault, resulting in abnormal vibration or noise problems. Based on the correspondence between vibration and sound characteristics and vehicle faults, a vibration and audio recognition neural network is designed to learn the characteristics of fault vibration and sound encoding and classification, and to determine the causes of abnormal vibration and noise in the vehicle, which can greatly help industry personnel locate problems early. The implementation of this method will greatly improve the efficiency of fault diagnosis. Therefore, it is necessary to provide a vehicle fault diagnosis and early warning solution based on vibration and noise to improve the efficiency of fault diagnosis. Summary of the invention

[0004] The purpose of the present invention is to provide a vehicle fault diagnosis and early warning method and system based on vibration and noise, so as to improve the efficiency of fault diagnosis.

[0005] To achieve the purpose of the present invention, the technical solution provided by the present invention is as follows:

[0006] First aspect

[0007] The present invention provides a vehicle fault diagnosis and early warning method based on vibration and noise, comprising the following steps:

[0008] Step S1: collecting vibration signals and noise signals corresponding to different vehicle fault types;

[0009] Step S2: performing short-time Fourier feature extraction, continuous wavelet transform feature extraction, and Mel-frequency cepstrum feature extraction on the vibration signal and the noise signal, and obtaining corresponding time-frequency feature data respectively, and converting all the time-frequency feature data into feature image data;

[0010] Step S3: constructing an image-based deep learning model, and using the feature image data for training to obtain a trained vehicle fault diagnosis model; wherein the image-based deep learning model is a dual attention mechanism neural network model based on a multi-head self-attention mechanism and a multi-scale attention mechanism;

[0011] Step S4: Use the trained vehicle fault diagnosis model to perform vehicle fault diagnosis and early warning.

[0012] Second aspect

[0013] The present invention provides a vehicle fault diagnosis and early warning system based on vibration and noise, comprising the following units: a signal acquisition unit, an image data acquisition unit, a model building unit, and a diagnosis and early warning unit;

[0014] The signal acquisition unit is used to collect vibration signals and noise signals corresponding to different vehicle fault types;

[0015] The image data acquisition unit is used to perform short-time Fourier feature extraction, continuous wavelet transform feature extraction, and Mel cepstrum feature extraction on the vibration signal and the noise signal, and obtain corresponding time-frequency feature data respectively, and convert all time-frequency feature data into feature image data;

[0016] The model building unit is used to build an image-based deep learning model and use the feature image data for training to obtain a trained vehicle fault diagnosis model; wherein the image-based deep learning model is a dual attention mechanism neural network model based on a multi-head self-attention mechanism and a multi-scale attention mechanism;

[0017] The diagnosis and early warning unit is used to perform vehicle fault diagnosis and early warning using the trained vehicle fault diagnosis model.

[0018] Compared with the prior art, the technical advantages of the present invention are as follows:

[0019] 1. In the solution of the present invention, the time-frequency feature image provides information about the fault signal in time and frequency, making the analysis of the fault signal more comprehensive. Through time-frequency analysis, the transient characteristics and frequency components in the signal can be identified, helping the model to better understand the fault mode. At the same time, the time-frequency feature image can effectively capture the characteristics of non-order vibration and noise, thereby providing more accurate information for fault diagnosis;

[0020] 2. In the solution of the present invention, the dual attention mechanism neural network model based on the multi-head self-attention mechanism and the multi-scale attention mechanism can simultaneously consider the local and global features of the data at different levels, so as to better capture the fault information contained in the signal and learn more complex and abstract feature representations, which helps to improve the performance of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A schematic diagram of a method flow chart provided by an embodiment of the present invention;

[0022] Figure 2 This is a structural schematic diagram of a dual attention mechanism neural network model based on a multi-head self-attention mechanism and a multi-scale attention mechanism in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] like Figure 1 As shown, this embodiment provides a vehicle fault diagnosis and early warning method based on vibration and noise, comprising the following steps:

[0025] Step S1: collecting vibration signals and noise signals corresponding to different vehicle fault types;

[0026] It should be noted that in step S1, the vibration signal collected is a vibration signal close to the faulty structure of the vehicle, and the noise signal collected includes noise signals close to the faulty structure of the vehicle, the driver's ear, the front passenger's ear, and the rear passenger's ear.

[0027] It should be noted that in step S1, vibration signals and noise signals corresponding to different vehicle fault types in the vehicle idling state, uniform speed state, and rapid acceleration state are collected.

[0028] Step S2: performing short-time Fourier feature extraction, continuous wavelet transform feature extraction, and Mel-frequency cepstrum feature extraction on the vibration signal and the noise signal, and obtaining corresponding time-frequency feature data respectively, and converting all the time-frequency feature data into feature image data;

[0029] Step S3: constructing an image-based deep learning model, and using the feature image data for training to obtain a trained vehicle fault diagnosis model; wherein the image-based deep learning model is a dual attention mechanism neural network model based on a multi-head self-attention mechanism and a multi-scale attention mechanism;

[0030] like Figure 2As shown, the dual attention mechanism neural network model based on the multi-head self-attention mechanism and the multi-scale attention mechanism includes a first fully connected layer and an average division and recombination layer, a splicing and recombination layer, a second fully connected layer, a multi-scale attention module, a third fully connected layer, softmax layer;

[0031] In order to expand the receptive field and guide the algorithm to explore more features at different scales, the first fully connected layer is used to transform the input signal in the length dimension. The expansion is 12 times, and the average division and recombination layer is used to divide it into h The query matrix of the attention heads , key matrix , and value matrix , , , The dimensions are the same, where Indicates Attention heads, 1≤ ≤ h， Perform multi-head self-attention mechanism calculation to obtain the attention output of each attention head , the formula is as follows:

[0032] ;

[0033] in, represents the dimension of the key matrix, T Represents a matrix transpose operation;

[0034] The splicing and reorganization layer is used to h The outputs of the attention heads are concatenated together to obtain the concatenated attention encoding , the formula is as follows:

[0035] ;

[0036] The second fully connected layer is used to Back-project back to the original space to get the attention encoding output , the formula is as follows:

[0037] ;

[0038] It should be noted that the multi-head self-attention mechanism allows the model to focus on different positions during the encoding process, while giving the attention layer output containing the encoded representation information in different self-spaces, thereby enhancing the model's expressiveness. In addition, in order to improve the performance of the model and accelerate the convergence of the model, according to and , through residual connection and layer normalization operation, the attention encoding reorganization output is obtained , the formula is as follows:

[0039] ;

[0040] In order to further improve the learning ability and extract deeper fault features, this embodiment designs a multi-scale attention module. The multi-scale attention module includes a first convolution layer, a first multi-scale convolution layer, a fourth fully connected layer, a second multi-scale convolution layer, a fifth fully connected layer, and a second convolution layer;

[0041] The first convolutional layer is used to Perform convolution operation to get the output of the first convolution layer , the formula is as follows:

[0042] ;

[0043] Said The multi-scale convolution output is obtained by sequentially passing through the first multi-scale convolution layer, the fourth fully connected layer, and the second multi-scale convolution layer. , the formula is as follows:

[0044] ;

[0045] The fifth fully connected layer and the second convolutional layer are used to Back-projecting back to the original space, we get the output of the multi-scale attention mechanism , the formula is as follows:

[0046] ;

[0047] according to , through residual connection and layer normalization operations, the fault feature output is obtained O b :

[0048] ;

[0049] The third fully connected layer, softmax Layer, used to output fault characteristics O b Processing to obtain the final classification result , the formula is as follows:

[0050] .

[0051] Step S4: Use the trained vehicle fault diagnosis model to perform vehicle fault diagnosis and early warning.

[0052] In addition, corresponding to the above method, this embodiment also provides a vehicle fault diagnosis and warning system based on vibration and noise, including the following units: a signal acquisition unit, an image data acquisition unit, a model building unit, and a diagnosis and warning unit;

[0053] The signal acquisition unit is used to collect vibration signals and noise signals corresponding to different vehicle fault types;

[0054] The image data acquisition unit is used to perform short-time Fourier feature extraction, continuous wavelet transform feature extraction, and Mel cepstrum feature extraction on the vibration signal and the noise signal, and obtain corresponding time-frequency feature data respectively, and convert all time-frequency feature data into feature image data;

[0055] The model building unit is used to build an image-based deep learning model and use the feature image data for training to obtain a trained vehicle fault diagnosis model; wherein the image-based deep learning model is a dual attention mechanism neural network model based on a multi-head self-attention mechanism and a multi-scale attention mechanism;

[0056] The diagnosis and early warning unit is used to perform vehicle fault diagnosis and early warning using the trained vehicle fault diagnosis model.

[0057] The dual attention mechanism neural network model based on multi-head self-attention mechanism and multi-scale attention mechanism includes a first fully connected layer and an average division and recombination layer, a splicing and recombination layer, a second fully connected layer, a multi-scale attention module, a third fully connected layer, softmax layer;

[0058] The first fully connected layer is used to transform the input signal in the length dimension The expansion is 12 times, and the average division and recombination layer is used to divide it into h The query matrix of the attention heads , key matrix , and value matrix , , , The dimensions are the same, where Indicates Attention heads, 1≤ ≤ h， Perform multi-head self-attention mechanism calculation to obtain the attention output of each attention head , the formula is as follows:

[0059] ;

[0060] in, represents the dimension of the key matrix, TRepresents a matrix transpose operation;

[0061] The splicing and reorganization layer is used to h The outputs of the attention heads are concatenated together to obtain the concatenated attention encoding , the formula is as follows:

[0062] ;

[0063] The second fully connected layer is used to Back-project back to the original space to get the attention encoding output , the formula is as follows:

[0064] ;

[0065] according to and , through residual connection and layer normalization operation, the attention encoding reorganization output is obtained , the formula is as follows:

[0066] ;

[0067] The multi-scale attention module includes a first convolutional layer, a first multi-scale convolutional layer, a fourth fully connected layer, a second multi-scale convolutional layer, a fifth fully connected layer, and a second convolutional layer;

[0068] The first convolutional layer is used to Perform convolution operation to get the output of the first convolution layer , the formula is as follows:

[0069] ;

[0070] Said The multi-scale convolution output is obtained by sequentially passing through the first multi-scale convolution layer, the fourth fully connected layer, and the second multi-scale convolution layer. , the formula is as follows:

[0071] ;

[0072] The fifth fully connected layer and the second convolutional layer are used to Back-projecting back to the original space, we get the output of the multi-scale attention mechanism , the formula is as follows:

[0073] ;

[0074] according to , through residual connection and layer normalization operations, the fault feature output is obtained O b :

[0075] ;

[0076] The third fully connected layer, softmax Layer, used to output fault characteristics O b Processing to obtain the final classification result , the formula is as follows:

[0077] .

[0078] Among them, the vibration signal collected by the signal collection unit is a vibration signal close to the faulty structure of the vehicle, and the collected noise signal includes noise signals close to the faulty structure of the vehicle, the driver's ear, the front passenger's ear, and the rear passenger's ear.

[0079] The signal acquisition unit acquires vibration signals and noise signals corresponding to different vehicle fault types when the vehicle is in an idle state, a uniform speed state, and a sudden acceleration state.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate and explain the present invention, and are not intended to limit the present invention to the scope of the described embodiments. In addition, those skilled in the art can understand that the present invention is not limited to the above embodiments, and more variations and modifications can be made according to the teachings of the present invention, and these variations and modifications all fall within the scope of the protection claimed by the present invention.

Claims

1. A vehicle fault diagnosis and early warning method based on vibration and noise, characterized in that: The steps include: Step S1: collecting vibration signals and noise signals corresponding to different vehicle fault types; Step S2: performing short-time Fourier feature extraction, continuous wavelet transform feature extraction, and Mel-frequency cepstrum feature extraction on the vibration signal and the noise signal, and obtaining corresponding time-frequency feature data respectively, and converting all the time-frequency feature data into feature image data; Step S3: constructing an image-based deep learning model, and using the feature image data for training to obtain a trained vehicle fault diagnosis model; wherein the image-based deep learning model is a dual attention mechanism neural network model based on a multi-head self-attention mechanism and a multi-scale attention mechanism; Step S4: using the trained vehicle fault diagnosis model to perform vehicle fault diagnosis and early warning; The dual attention mechanism neural network model based on the multi-head self-attention mechanism and the multi-scale attention mechanism includes a first fully connected layer and an average division and recombination layer, a splicing and recombination layer, a second fully connected layer, a multi-scale attention module, a third fully connected layer, and a softmax layer; The first fully connected layer is used to expand the input signal I by 12 times in the length dimension and divide it into query matrices q of h attention heads through the average division and recombination layer. i , key matrix k i , and value matrix v i ,q i , k i 、v i The dimensions are the same as , where i represents the i-th attention head, 1≤i≤h, and the multi-head self-attention mechanism is calculated to obtain the attention output a of each attention head i , the formula is as follows: Among them, d k represents the dimension of the key matrix, and T represents the matrix transpose operation; The concatenation and reorganization layer is used to concatenate the h attention head outputs together to obtain the concatenated attention code x a , the formula is as follows: x a =concat{a1,a2,...,a h } The second fully connected layer is used to transform x a Back-project back to the original space to get the attention encoding output O a , the formula is as follows: About a =FC(x a ) According to O a and I, through residual connection and layer normalization operations, we get the attention encoding reorganized output I a , the formula is as follows: I a =LayerNorm(O a )+I The multi-scale attention module includes a first convolutional layer, a first multi-scale convolutional layer, a fourth fully connected layer, a second multi-scale convolutional layer, a fifth fully connected layer, and a second convolutional layer; The first convolutional layer is used to a Perform a convolution operation to obtain the first convolution layer output C1, the formula is as follows: C1=Conv1(I a ) The C1 passes through the first multi-scale convolution layer, the fourth fully connected layer, and the second multi-scale convolution layer in sequence to obtain a multi-scale convolution output C M , the formula is as follows: C M =CBAM2{FC[CBAM1(C1)]} The fifth fully connected layer and the second convolutional layer are used to transform C M Back-projecting back to the original space, we get the multi-scale attention mechanism output O' b , the formula is as follows: O' b =Conv2[FC(C M )] According to O' b , through residual connection and layer normalization operations, the fault feature output O is obtained b : THE b =LayerNorm(O' b )+O a The third fully connected layer and the softmax layer are used to output the fault features. b Processing to obtain the final classification result The formula is as follows:

2. A vehicle fault diagnosis and early warning method based on vibration and noise according to claim 1, characterized in that: In step S1, the vibration signal collected is a vibration signal close to the faulty structure of the vehicle, and the noise signal collected includes noise signals close to the faulty structure of the vehicle, the driver's ear, the front passenger's ear, and the rear passenger's ear.

3. A vehicle fault diagnosis and early warning method based on vibration and noise according to claim 2, characterized in that: In step S1, vibration signals and noise signals corresponding to different vehicle fault types in the vehicle idling state, uniform speed state, and rapid acceleration state are collected.

4. A vehicle fault diagnosis and warning system based on vibration and noise, characterized in that: It includes the following units: signal acquisition unit, image data acquisition unit, model building unit and diagnosis and early warning unit; The signal acquisition unit is used to collect vibration signals and noise signals corresponding to different vehicle fault types; The image data acquisition unit is used to perform short-time Fourier feature extraction, continuous wavelet transform feature extraction, and Mel cepstrum feature extraction on the vibration signal and the noise signal, and obtain corresponding time-frequency feature data respectively, and convert all time-frequency feature data into feature image data; The model building unit is used to build an image-based deep learning model and use the feature image data for training to obtain a trained vehicle fault diagnosis model; wherein the image-based deep learning model is a dual attention mechanism neural network model based on a multi-head self-attention mechanism and a multi-scale attention mechanism; The diagnosis and early warning unit is used to perform vehicle fault diagnosis and early warning using the trained vehicle fault diagnosis model; The dual attention mechanism neural network model based on the multi-head self-attention mechanism and the multi-scale attention mechanism includes a first fully connected layer and an average division and recombination layer, a splicing and recombination layer, a second fully connected layer, a multi-scale attention module, a third fully connected layer, and a softmax layer; The first fully connected layer is used to expand the input signal I by 12 times in the length dimension and divide it into query matrices q of h attention heads through the average division and recombination layer. i , key matrix k i , and value matrix v i ,q i , k i 、v i The dimensions are the same as , where i represents the i-th attention head, 1≤i≤h, and the multi-head self-attention mechanism is calculated to obtain the attention output a of each attention head i , the formula is as follows: Among them, d k represents the dimension of the key matrix, and T represents the matrix transpose operation; The concatenation and reorganization layer is used to concatenate the h attention head outputs together to obtain the concatenated attention code x a , the formula is as follows: x a =concat{a1,a2,...,a h } The second fully connected layer is used to transform x a Back-project back to the original space to get the attention encoding output O a , the formula is as follows: About a =FC(x a ) According to O a and I, through residual connection and layer normalization operations, we get the attention encoding reorganized output I a , the formula is as follows: I a =LayerNorm(O a )+I The multi-scale attention module includes a first convolutional layer, a first multi-scale convolutional layer, a fourth fully connected layer, a second multi-scale convolutional layer, a fifth fully connected layer, and a second convolutional layer; The first convolutional layer is used to a Perform a convolution operation to obtain the first convolution layer output C1, the formula is as follows: C1=Conv1(I a ) The C1 passes through the first multi-scale convolution layer, the fourth fully connected layer, and the second multi-scale convolution layer in sequence to obtain a multi-scale convolution output C M , the formula is as follows: C M =CBAM2{FC[CBAM1(C1)]} The fifth fully connected layer and the second convolutional layer are used to transform C M Back-projecting back to the original space, we get the multi-scale attention mechanism output O' b , the formula is as follows: O' b =Conv2[FC(C M )] According to O' b , through residual connection and layer normalization operations, the fault feature output O is obtained b : THE b =LayerNorm(O' b )+O a The third fully connected layer and the softmax layer are used to output the fault features. b Processing to obtain the final classification result The formula is as follows:

5. A vehicle fault diagnosis and warning system based on vibration and noise according to claim 4, characterized in that: The vibration signal collected by the signal collection unit is a vibration signal close to the faulty structure of the vehicle, and the noise signal collected includes noise signals close to the faulty structure of the vehicle, the driver's ear, the front passenger's ear, and the rear passenger's ear.

6. A vehicle fault diagnosis and warning system based on vibration and noise according to claim 5, characterized in that: The signal acquisition unit acquires vibration signals and noise signals corresponding to different vehicle fault types when the vehicle is in an idle state, a uniform speed state, and a sudden acceleration state.

Images