CN115577258A - Vibration signal recognition model training method, motor fault detection method and device - Google Patents

Vibration signal recognition model training method, motor fault detection method and device Download PDF

Info

Publication number
CN115577258A
CN115577258A CN202211098693.3A CN202211098693A CN115577258A CN 115577258 A CN115577258 A CN 115577258A CN 202211098693 A CN202211098693 A CN 202211098693A CN 115577258 A CN115577258 A CN 115577258A
Authority
CN
China
Prior art keywords
vibration signal
sample
category
frequency domain
signal
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.)
Withdrawn
Application number
CN202211098693.3A
Other languages
Chinese (zh)
Inventor
沈浩
赵德欣
王磊
成莎莎
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China Telecom Corp Ltd
Original Assignee
China Telecom Corp Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by China Telecom Corp Ltd filed Critical China Telecom Corp Ltd
Priority to CN202211098693.3A priority Critical patent/CN115577258A/en
Publication of CN115577258A publication Critical patent/CN115577258A/en
Withdrawn legal-status Critical Current

Links

Images

Classifications

    • 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

Abstract

The embodiment of the application discloses a vibration signal recognition model training method, a motor fault detection method and a device, wherein the vibration signal recognition model training method comprises the following steps: performing constant Q transformation processing on each vibration signal sample to convert each vibration signal sample into a frequency domain signal sample; performing feature extraction processing on the frequency domain signal sample according to a vibration signal identification model to be trained to obtain a sample vector corresponding to the frequency domain signal sample, and calculating a category center vector corresponding to each vibration signal category according to the sample vector; iterative training is carried out on the vibration signal recognition model to be trained according to the distance between the sample vector and the category center vector to obtain the vibration signal recognition model after training, so that data which are more convenient for model processing can be input, the characteristic extraction effect of the model on the vibration signal is improved, the condition that the model is over-fit due to unbalanced sample number in each vibration signal category is avoided, and the model training effect is improved.

Description

Vibration signal recognition model training method, motor fault detection method and device
Technical Field
The application relates to the technical field of computers, in particular to a vibration signal recognition model training method, a motor fault detection method and a device.
Background
With the rapid development of modern industrial technologies, mechanical equipment is rapidly developed towards high speed, precision, automation and integration, a rolling bearing of a motor is one of the most important supporting and rotating components in the mechanical equipment, various faults are prone to occur due to the influence of working load and the like, and the running precision and safety reliability of the rolling bearing of the motor directly influence the overall performance of the mechanical equipment, so that fault diagnosis of the motor is of great importance.
At present, whether a motor has a fault is generally distinguished by adopting an artificial listening method on a motor production line, the cost is high, and the repeated and monotonous listening work is easy to cause personnel fatigue, so that the condition of misjudgment is easy to occur, and the judgment effect is poor.
Disclosure of Invention
In order to solve the technical problem, embodiments of the present application provide a vibration signal recognition model training method, a motor fault detection method and a device, so as to improve efficiency and accuracy of motor fault detection.
According to an aspect of an embodiment of the present application, there is provided a vibration signal recognition model training method, including: acquiring a vibration signal sample set, wherein vibration signal samples in the vibration signal sample set are marked with vibration signal types; performing constant Q transformation processing on each vibration signal sample to convert each vibration signal sample into a frequency domain signal sample; performing feature extraction processing on the frequency domain signal sample according to a vibration signal identification model to be trained to obtain a sample vector corresponding to the frequency domain signal sample, and calculating a category center vector corresponding to each vibration signal category according to the sample vector; and performing iterative training on the vibration signal identification model to be trained according to the distance between the sample vector and the category center vector to obtain the trained vibration signal identification model and the category center vector corresponding to each vibration signal category.
In some embodiments, performing a constant Q transform process on each vibration signal sample to convert each vibration signal sample to a frequency domain signal sample comprises: preprocessing the vibration signal samples to obtain a vibration signal frame corresponding to each vibration signal sample; filtering the vibration signal frame according to a preset central frequency and bandwidth ratio to obtain an intermediate frequency domain signal sample; and carrying out logarithmic calculation on the intermediate frequency domain signal samples to obtain frequency domain signal samples corresponding to each vibration signal sample.
In some embodiments, preprocessing the vibration signal samples to obtain a vibration signal frame corresponding to each vibration signal sample includes: carrying out pre-emphasis processing on the vibration signal sample to obtain an emphasized vibration signal sample; performing frame processing on the weighted vibration signal samples to obtain a plurality of vibration signal frames; and windowing the plurality of vibration signal frames to obtain a plurality of continuous vibration signal frames.
In some embodiments, performing feature extraction processing on a frequency domain signal sample according to a vibration signal recognition model to be trained to obtain a sample vector corresponding to the frequency domain signal sample, includes: selecting samples of frequency domain signal samples belonging to the same vibration signal category to obtain a sample support set and a sample query set; performing feature extraction processing on frequency domain signal samples in the sample support set according to a vibration signal identification model to be trained to obtain a category center vector of a vibration signal category corresponding to the sample support set; performing feature extraction processing on the frequency domain signal samples in the sample query set according to the vibration signal identification model to be trained to obtain a sample vector corresponding to each frequency domain signal sample in the sample query set; the iterative training is carried out on the vibration signal identification model to be trained, and the trained vibration signal identification model and the category center vector corresponding to each vibration signal category are obtained, and the iterative training comprises the following steps: calculating the distance between a sample vector corresponding to each frequency domain signal sample in the sample query set and the category center vector of each vibration signal category; calculating a loss value corresponding to each frequency domain signal sample in the sample query set according to a preset loss function and the distance; and performing iterative training on the vibration signal recognition model to be trained according to the loss value so as to obtain the vibration signal recognition model after training and the class center vector corresponding to each vibration signal class when a preset training finishing condition is reached.
In some embodiments, the obtaining a sample support set and a sample query set by performing sample selection on frequency domain signal samples belonging to the same vibration signal category includes: confirming the type of a vibration signal to be input in current iterative training to obtain the type of a target vibration signal; acquiring frequency domain signal samples belonging to the category of target vibration signals to obtain a candidate sample set; and randomly selecting the frequency domain signal samples in the candidate sample set to respectively obtain a sample support set and a sample query set, wherein the frequency domain signal samples between the sample support set and the sample query set are different.
In some embodiments, performing feature extraction processing on a frequency domain signal sample in a sample support set according to a vibration signal recognition model to be trained to obtain a class center vector of a vibration signal class corresponding to the sample support set, including: performing feature extraction processing on the frequency domain signal samples in the sample support set according to the vibration signal identification model to be trained to obtain a sample vector corresponding to each frequency domain signal sample in the sample support set; and calculating the average value of the sample vectors corresponding to each frequency domain signal sample in the sample support set to obtain the class central vector of the vibration signal class corresponding to the sample support set.
In some embodiments, the preset loss function comprises an AM-Softmax loss function.
According to one aspect of the embodiments of the present application, a motor fault detection method includes: obtaining a motor vibration signal to obtain a vibration signal to be identified; performing constant Q transformation processing on the vibration signal to be identified so as to convert the vibration signal to be identified into a frequency domain signal to be identified; inputting the frequency domain signal to be identified into a vibration signal identification model to obtain a vibration signal vector output by the vibration signal identification model, and acquiring a category center vector corresponding to each vibration signal category; the vibration signal recognition model and the category center vector corresponding to each vibration signal category are obtained according to the vibration signal recognition model training method; calculating the distance between the vibration signal vector and the class center vector corresponding to each vibration signal class to obtain the vibration signal class corresponding to the vibration signal to be identified according to the distance; and generating a motor fault detection result according to the vibration signal category corresponding to the vibration signal to be identified.
According to an aspect of an embodiment of the present application, there is provided a vibration signal recognition model training apparatus, including: the sample acquisition module is configured to acquire a vibration signal sample set, and vibration signal samples in the vibration signal sample set are labeled with vibration signal types; a sample conversion module configured to perform constant Q transform processing on each vibration signal sample to convert each vibration signal sample into a frequency domain signal sample; the sample feature extraction module is configured to perform feature extraction processing on the frequency domain signal sample according to the vibration signal identification model to be trained to obtain a sample vector corresponding to the frequency domain signal sample, and calculate a category center vector corresponding to each vibration signal category according to the sample vector; and the training module is configured to perform iterative training on the vibration signal identification model to be trained according to the distance between the sample vector and the category center vector to obtain the trained vibration signal identification model and the category center vector corresponding to each vibration signal category.
According to an aspect of an embodiment of the present application, there is provided a motor failure detection apparatus including: the signal acquisition module is configured to acquire a motor vibration signal to obtain a vibration signal to be identified; the signal conversion module is configured to perform constant Q conversion processing on the vibration signal to be identified so as to convert the vibration signal to be identified into a frequency domain signal to be identified; the vector acquisition module is configured to input the frequency domain signal to be identified into the vibration signal identification model to obtain vibration signal vectors output by the vibration signal identification model, and acquire a category center vector corresponding to each vibration signal category; the vibration signal recognition model and the category center vector corresponding to each vibration signal category are obtained according to the vibration signal recognition model training method; the signal identification module is configured to calculate the distance between the vibration signal vector and the category center vector corresponding to each vibration signal category so as to obtain the vibration signal category corresponding to the vibration signal to be identified according to the distance; and the detection result generation module is configured to generate a motor fault detection result according to the vibration signal category corresponding to the vibration signal to be identified.
According to an aspect of embodiments of the present application, there is provided a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements a vibration signal recognition model training method or a motor fault detection method as above.
According to an aspect of an embodiment of the present application, there is provided an electronic device including one or more processors; a storage device for storing one or more programs that, when executed by the electronic device, cause the electronic device to implement the vibration signal recognition model training method or the motor failure detection method as described above.
In the technical scheme provided by the embodiment of the application, the vibration signal sample is subjected to constant Q transformation processing, the characteristics of the vibration signal can be effectively extracted, then the frequency domain signal sample subjected to constant Q transformation processing is input into the vibration signal identification model to be trained, the category central vector corresponding to each vibration signal category is calculated according to the sample vector output by the vibration signal identification model to be trained, then the vibration signal identification model to be trained is subjected to iterative training according to the distance between the sample vector and the category central vector, the vibration signal identification model after training is obtained, data convenient for model processing is input, the characteristic extraction effect of the model on the vibration signal is improved, the condition that the model is over-fit due to the fact that the number of samples in each vibration signal category is unbalanced is avoided, and the model training effect is improved.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and together with the description, serve to explain the principles of the application. It is obvious that the drawings in the following description are only some embodiments of the application, and that for a person skilled in the art, other drawings can be derived from them without inventive effort. In the drawings:
FIG. 1 is a schematic diagram of an application environment of a vibration signal recognition model training method in an exemplary embodiment of the present application;
FIG. 2 is a flow chart of a vibration signal recognition model training method shown in an exemplary embodiment of the present application;
FIG. 3 is a flow chart of a vibration signal recognition model training method shown in another exemplary embodiment of the present application;
FIG. 4 is a schematic diagram of signal conversion shown in an exemplary embodiment of the present application;
FIG. 5 is a schematic diagram illustrating vibration signal recognition model training and deployment in accordance with an exemplary embodiment of the present application;
FIG. 6 is a schematic diagram of an application environment of a motor fault detection method according to an exemplary embodiment of the present application;
FIG. 7 is a flow chart illustrating a motor fault detection method in accordance with an exemplary embodiment of the present application;
FIG. 8 is a block diagram of a vibration signal recognition model training apparatus shown in an exemplary embodiment of the present application;
FIG. 9 is a block diagram of a motor fault detection arrangement shown in an exemplary embodiment of the present application;
FIG. 10 is a block diagram of a computer system suitable for use in implementing the electronic device of an embodiment of the present application.
Detailed Description
Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. The following description refers to the accompanying drawings in which the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments identical to the present application. Rather, they are merely examples of the same devices and methods of some aspects of the present application, as detailed in the appended claims.
The block diagrams shown in the figures are functional entities only and do not necessarily correspond to physically separate entities. I.e. the functional entities may be implemented in the form of application programs or in one or more hardware modules or integrated circuits or in different networks and/or processor means and/or microcontroller means.
The flow charts shown in the drawings are merely illustrative and do not necessarily include all of the contents and operations/steps, nor do they necessarily have to be performed in the order described. For example, some operations/steps may be decomposed, and some operations/steps may be combined or partially combined, so that the actual execution sequence may be changed according to the actual situation.
In the present application, the term "plurality" means two or more. "and/or" describes the association relationship of the associated object, indicating that there may be three relationships, for example, a and/or B may indicate: a exists alone, A and B exist simultaneously, and B exists alone. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship.
Optionally, in this embodiment, the vibration signal recognition model training method may be applied to the environment shown in fig. 1. As shown in fig. 1, the implementation environment includes a terminal 110 and a server 120, and the terminal 110 and the server 120 may be directly or indirectly connected through wired or wireless communication, and the application is not limited herein.
The terminal 110 communicates with the server 120 through a network. The training of the vibration signal identification model may be executed by the terminal 110 or the server 120, for example, the terminal 110 or the server 120 obtains a vibration signal sample set, and trains the vibration signal identification model according to the vibration signal sample set, so as to finally obtain the trained vibration signal identification model. The trained vibration signal recognition model may be deployed on the server 120, for example, in the cloud, and called by the terminal 110 to perform vibration signal recognition. Alternatively, the trained vibration signal recognition model may be downloaded locally by the terminal 110 for use.
The terminal 110 may be, but is not limited to, a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart watch, a vehicle-mounted terminal, and the like. The terminal may refer to one of a plurality of terminals, and this embodiment is only illustrated by the terminal 110, and those skilled in the art may know that the number of the terminals may be only one, or several tens or several hundreds of the terminals, or more, and at this time, the implementation environment of the vibration signal recognition model training method further includes other terminals. The number of terminals and the type of the device are not limited in the embodiments of the present application.
The server 120 may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a Network service, cloud communication, a middleware service, a domain name service, a security service, a Content Delivery Network (CDN), a big data and artificial intelligence platform, and the like. The server 120 is used to provide background services for the applications running on the terminal 110.
Optionally, the wireless or wired networks described above use standard communication techniques and/or protocols. The Network is typically the Internet, but can be any Network including, but not limited to, a Local Area Network (LAN), a Metropolitan Area Network (MAN), a Wide Area Network (WAN), a mobile, wired or wireless Network, a private Network, or any combination of virtual private networks. In some embodiments, data exchanged over a network is represented using techniques and/or formats including HyperText Mark-up Language (HTML), extensible Mark-up Language (XML), and the like. All or some of the links may also be encrypted using conventional encryption techniques such as Secure Socket Layer (SSL), transport Layer Security (TLS), virtual Private Network (VPN), internet Protocol Security (IPsec). In other embodiments, custom and/or dedicated data communication techniques may also be used in place of, or in addition to, the data communication techniques described above.
In some embodiments, the terminal 110 may be loaded with APP (Application) applications with vibration signal identification function, including applications that conventionally need to be installed separately, applet applications that can be used without downloading and installation, and the terminal may identify the vibration signal in the environment where the terminal 110 is located through the APP.
The trained vibration signal recognition model can be deployed locally at the terminal, the terminal calls the vibration signal recognition model to recognize the vibration signal so as to output a recognition result, or the trained vibration signal recognition model can also be deployed at the server, the server recognizes the vibration signal by using the trained vibration signal recognition model so as to output the recognition result, the recognition result is transmitted to the terminal, and the terminal displays the recognition result, which is not limited in the application.
Optionally, the server 120 undertakes training of the primary vibration signal recognition model, and the terminal 110 undertakes training of the secondary vibration signal recognition model; or, the server 120 undertakes the training work of the secondary vibration signal identification model, and the terminal 110 undertakes the training work of the primary vibration signal identification model; alternatively, the server 120 or the terminal 110 may respectively undertake the vibration signal recognition model training separately, which is not limited in this application.
Referring to fig. 2, fig. 2 is a flowchart illustrating a vibration signal recognition model training method according to an exemplary embodiment of the present application. The vibration signal recognition model training method can be applied to the implementation environment shown in fig. 1, and is specifically executed by the server 120 in the implementation environment. It should be understood that the method may be applied to other exemplary implementation environments and is specifically executed by devices in other implementation environments, and the embodiment does not limit the implementation environment to which the method is applied.
The following describes in detail a vibration signal recognition model training method proposed in an embodiment of the present application with a server as a specific implementation subject.
As shown in fig. 2, in an exemplary embodiment, the vibration signal recognition model training method at least includes steps S210 to S240, which are described in detail as follows:
step S210, a vibration signal sample set is obtained, and vibration signal samples in the vibration signal sample set are marked with vibration signal types.
The vibration signal sample set includes a plurality of vibration signal samples with sample labels, and the sample labels are used to indicate vibration signal classes of corresponding vibration signal samples, for example, the vibration signal classes include normal vibration signals and abnormal vibration signals.
For example, the vibration signal receiver acquires the vibration signal according to a preset sampling rate to obtain initial vibration signal data. And then reading the initial vibration signal data to obtain a plurality of vibration signals, wherein the number of sampling points contained in each vibration signal is the same, so that the dimension consistency of data input into a vibration signal identification model to be trained during training is ensured. Further, labeling each vibration signal to obtain a vibration signal sample with a sample label.
Step S220, performing constant Q transform processing on each vibration signal sample to convert each vibration signal sample into a frequency domain signal sample.
Constant Q Transform (CQT) means that the center frequency is exponentially distributed, and the obtained spectrum is obtained by filtering through a filter bank with different filter bandwidths but the ratio of the center frequency to the bandwidth is Constant Q, and the obtained spectrum is based on log2 as a base, and the length of the filter window can be changed according to the difference of the spectral line frequencies.
Because the vibration signal is usually preserved with the sampling point form, often contain tens of thousands of sampling points in a vibration signal, it is relatively poor to carry out the effect of neural network training, and this application embodiment carries out constant Q transform processing earlier the vibration signal sample, can effectively extract the characteristic of vibration signal, and the frequency domain signal sample that obtains can directly regard as the input of the vibration signal recognition model of treating the training to improve the effect of model training.
Step S230, performing feature extraction processing on the frequency domain signal sample according to the vibration signal identification model to be trained to obtain a sample vector corresponding to the frequency domain signal sample, and calculating a category center vector corresponding to each vibration signal category according to the sample vector.
It should be noted that the vibration signal recognition model to be trained refers to a model that is not completely trained, and after sufficient training and parameter adjustment, the model can be used as a vibration signal recognition model. And the category central vector corresponding to the vibration signal category is used for representing the integral characteristic of the vibration signal sample of which the classification result is the vibration signal category.
The vibration signal identification model is used for encoding an input frequency domain signal sample to obtain a sample vector with dimension M, wherein M includes but is not limited to 128, 256, 512 dimensions and the like. The specific form and category of the vibration signal identification model are not limited in the embodiment of the present application, the feature extraction layer in the vibration signal identification model may adopt a Convolutional Neural Network (CNN), a Residual Neural Network (ResNet), a Recurrent Neural Network (RNN), and the like, and the neuron number of the output layer of the Network is consistent with the dimension M of the sample vector.
In the training process of the vibration signal identification model to be trained, the category central vectors of the vibration signal categories are obtained according to the sample vectors corresponding to the vibration signal categories, different category central vectors correspond to different vibration signal categories, for example, the vibration signal samples comprise a positive sample representing a normal vibration signal and a negative sample representing an abnormal vibration signal, the positive sample corresponds to one category central vector, and the negative sample corresponds to one category central vector.
For example, when the sample vector of a vibration signal sample is one, the server determines the sample vector of the vibration signal sample as the class center vector of the vibration signal class corresponding to the vibration signal sample, and when the vibration signal class contains a plurality of vibration signal samples, the server determines the mean value of the sample vectors of the respective vibration signal samples as the class center vector corresponding to the vibration signal class.
Step S240, iteratively training the vibration signal recognition model to be trained according to the distance between the sample vector and the category center vector, to obtain the vibration signal recognition model after training and the category center vector corresponding to each vibration signal category.
It should be noted that, when not having been trained sufficiently, the vibration signal recognition result obtained by the vibration signal recognition model to be trained has a certain difference from the true correct result (i.e. the sample label), and by calculating the loss value and adjusting the model parameters of the vibration signal recognition model to be trained based on the loss value, the model can learn how to correctly perform vibration signal recognition and give the correct vibration signal recognition result.
It can be understood that the smaller the distance between the sample vector and the category center vector is, the more similar the vibration signal sample corresponding to the sample vector and the vibration signal category corresponding to the category center vector are; the greater the distance between the sample vector and the category center vector, the less similar the vibration signal sample corresponding to the sample vector and the vibration signal category corresponding to the category center vector are.
According to the embodiment of the application, the distance between the sample vectors and the category center vectors is calculated, then iterative training is carried out on the vibration signal recognition model to be trained by taking the purposes of reducing the distance between the sample vectors belonging to the same vibration signal category and expanding the distance between the sample vectors not belonging to the same vibration signal category as a target, so that the vibration signal recognition model after training is obtained, and the category center vectors of all the vibration signal categories corresponding to the vibration signal recognition model after training are obtained.
The vibration signal sample is subjected to constant Q transformation processing, the characteristics of the vibration signal can be effectively extracted, then the frequency domain signal sample subjected to constant Q transformation processing is input into the vibration signal recognition model to be trained, the category central vector corresponding to each vibration signal category is calculated according to the sample vector output by the vibration signal recognition model to be trained, iterative training is carried out on the vibration signal recognition model to be trained according to the distance between the sample vector and the category central vector, the vibration signal recognition model after training is obtained, on the premise that the characteristic extraction effect of the model on the vibration signal is improved, the condition that the model is over-trained and is not fit due to the unbalanced number of samples in each vibration signal category is avoided, and the model training effect is improved.
In some embodiments, step S220 may be further implemented by steps S221 to S223 shown in fig. 3, and the details are as follows:
step S221, preprocessing the vibration signal samples to obtain a vibration signal frame corresponding to each vibration signal sample.
Differences may exist between different vibration signal samples, for example, time domain lengths are different, and therefore, in order to facilitate model training on the vibration signal samples, data in the vibration signal samples need to be preprocessed before training.
In some embodiments, preprocessing the vibration signal samples to obtain a vibration signal frame corresponding to each vibration signal sample includes: pre-emphasis processing is carried out on the vibration signal sample to obtain an emphasized vibration signal sample; performing frame processing on the emphasized vibration signal sample to obtain a plurality of vibration signal frames; and windowing the plurality of vibration signal frames to obtain a plurality of continuous vibration signal frames.
Before processing the vibration signal samples, the audio signal may be pre-emphasized using a high-pass filter whose functional expression may be:
H(z)＝1-μz -1
where z represents the audio signal and μ represents the set hyper-parameter.
Since the purpose of pre-emphasis is to balance the spectrum to emphasize the high frequency signals, the corresponding time domain expression may be:
y(n)＝x(n)-αx(n-1)
where x (n) represents the time domain at time n, x (n-1) represents the time domain at time n-1, and y (n) represents the difference between time n and time n-1, and α is a constant, typically 0.97.
Therefore, pre-emphasis processing is performed before framing processing is performed on the vibration signal samples, so that high-frequency signals are highlighted, and reduction of attenuation loss of the signals is facilitated.
Then, the weighted vibration signal samples are subjected to framing processing to obtain a plurality of vibration signal frames.
In the embodiment of the present application, the framing processing refers to dividing the vibration signal to be evaluated into N pieces of vibration signals with fixed size, and each piece of vibration signal is referred to as a frame, and the frame length is generally 10ms to 30ms.
In the frame division, an overlapped segmentation method may be adopted, in which frames are moved to the overlapped part of the previous frame and the next frame, and the continuity is maintained by using the short-time stationarity of the signal to make the transition between the frames smooth. The weighted vibration signal samples are subjected to framing processing to obtain a plurality of vibration signal frames, and the processing of the vibration signal samples is facilitated.
Furthermore, after the weighted vibration signal samples are subjected to framing, a plurality of vibration signal frames obtained by framing are subjected to windowing processing, and a plurality of continuous vibration signal frames are obtained. The window function used in the windowing process includes, but is not limited to, a hann window (hann), a hamming window (hamming), a blackman-harris window (blackman-harris), and the like, and the preset window length is a width of the analysis window, and is not limited specifically, for example, when the sampling frequency is 48kHz, the preset window length may be 21.3ms. After sampling of the analysis window is completed each time, the analysis window is slid backward by a distance for sampling again, the slid distance is a frame shift, and the specific size of the frame shift is not limited, and may be, for example, half the window length.
For example, each frame is substituted into a window function, the window speech signal sw (n) = s (n) × w (n), s (n) represents a plurality of audio frames, w (n) represents an additive window function, and sw (n) represents a result of windowing the s (n) segment of the signal.
By windowing the multiple vibration signal frames, the continuous multiple vibration signal frames can be obtained, and the influence of signal discontinuity possibly caused by two ends of each frame can be eliminated.
Step S222, filtering the vibration signal frame according to a preset ratio of the center frequency to the bandwidth to obtain an intermediate frequency domain signal sample.
And then performing constant Q transformation processing on each vibration signal sample, wherein Q refers to the ratio of the center frequency of the filter to the bandwidth, performing constant Q transformation on the time domain signal to obtain a middle frequency domain signal sample corresponding to each vibration signal sample, wherein the frequency domain signal sample after the transformation has higher frequency spectrum resolution at lower frequency and higher time resolution at higher frequency.
Step S223, performing logarithm calculation on the intermediate frequency domain signal sample to obtain a frequency domain signal sample corresponding to each vibration signal sample.
For example, an average value of a spectrogram corresponding to the intermediate frequency-domain signal sample may be calculated first, and then a logarithm calculation may be performed on the intermediate frequency-domain signal sample, so as to obtain a frequency-domain signal sample corresponding to each vibration signal sample.
The functional expression for the logarithmic calculation of the intermediate frequency domain signal samples may be:
log|CQT(x)| 2
where x is the vibration signal sample.
To facilitate a general understanding of the signal conversion process of the present application, referring to fig. 4, fig. 4 schematically shows a flow of steps of signal conversion provided by an embodiment of the present application. As shown in fig. 4, when signal conversion is performed on a vibration signal sample, pre-emphasis processing, framing processing, windowing processing, constant Q transformation, and logarithm processing are performed in sequence, so as to obtain a frequency domain signal sample for easier model training, and the processing procedure of each step may refer to the technical solution of the foregoing embodiment.
In some embodiments, performing feature extraction processing on a frequency domain signal sample according to a vibration signal recognition model to be trained to obtain a sample vector corresponding to the frequency domain signal sample, so as to calculate a category center vector corresponding to each vibration signal category according to the sample vector, including: selecting samples of frequency domain signal samples belonging to the same vibration signal category to obtain a sample support set and a sample query set; performing feature extraction processing on frequency domain signal samples in the sample support set according to a vibration signal identification model to be trained to obtain a category center vector of a vibration signal category corresponding to the sample support set; performing feature extraction processing on the frequency domain signal samples in the sample query set according to the vibration signal identification model to be trained to obtain a sample vector corresponding to each frequency domain signal sample in the sample query set; the iterative training of the vibration signal recognition model to be trained is carried out according to the distance between the sample vector and the category center vector, so as to obtain the vibration signal recognition model after training and the category center vector corresponding to each vibration signal category, and the iterative training comprises the following steps: calculating the distance between a sample vector corresponding to each frequency domain signal sample in the sample query set and the category center vector of each vibration signal category; calculating a loss value corresponding to each frequency domain signal sample in the sample query set according to a preset loss function and the distance; and performing iterative training on the vibration signal recognition model to be trained according to the loss value so as to obtain the vibration signal recognition model after training and a category central vector corresponding to each vibration signal category when a preset training finishing condition is reached.
In the embodiment of the present application, the vibration signal samples in the vibration signal sample set are input into the vibration signal recognition model to be trained in batches according to the batch size (batch size), so as to perform model training on the vibration signal recognition model to be trained.
For example, the vibration signal category includes an abnormal vibration signal and a normal vibration signal, and a frequency domain signal sample set S = { (x) is obtained by performing constant Q transform processing on each vibration signal sample 1 ,y 1 ),...,(x N ,y 2 )}，x i I.e. a frequency domain signal sample generated by constant Q transformation on a representative vibration signal. Wherein y is i ∈{1,2}，y 1 Representing the normal vibration signal class, y 2 Representing an abnormal vibration signal category.
In some embodiments, the performing sample selection on frequency domain signal samples belonging to the same vibration signal category to obtain a sample support set and a sample query set includes: confirming the type of a vibration signal to be input in current iterative training to obtain the type of a target vibration signal; acquiring frequency domain signal samples belonging to the category of target vibration signals to obtain a candidate sample set; and randomly selecting frequency domain signal samples in the candidate sample set to respectively obtain a sample support set and a sample query set, wherein the frequency domain signal samples between the sample support set and the sample query set are different.
The target vibration signal category can be confirmed according to the training condition of the current vibration signal recognition model, for example, the target vibration signal category is confirmed according to the accuracy of the vibration signal recognition model for each vibration signal category, and the vibration signal category with the lowest accuracy is used as the target vibration signal category, so that the model training effect is improved.
In each iteration training of the vibration signal identification model to be trained, the frequency domain signal samples belonging to the target vibration signal category are subjected to sample selection according to the batch number to obtain a sample support set S k And sample query set Q k . Sample support set S k Is N randomly extracted from all samples with the target vibration signal class k s Strip samples, and samples query set Q k Removing a sample support set S which is trained by the iteration of the current round from a sample with the target vibration signal class k k Selected samples, and then randomly selected N from the remaining samples Q A bar sample wherein k is y 1 Or y 2 。
And then, performing feature extraction processing on the frequency domain signal samples in the sample support set according to the vibration signal identification model to be trained to obtain a category center vector of the vibration signal category corresponding to the sample support set. The calculation formula of the category center vector can be expressed as:
Figure BDA0003838586400000131
wherein, C k Is a class center vector of vibration signal class k, f w And identifying a feature extraction layer of the model for the vibration signal to be trained.
Further, calculating the sample vector corresponding to each frequency domain signal sample in the sample query set and the vibration signal categoryDistance between class center vectors. If the sample vector corresponding to the frequency domain signal sample in the query set is f w (x) Then the distance between the sample vector and the class center vector with the vibration signal class k is d (f) w (x),c k )). The vector distance calculation formula includes, but is not limited to, euclidean distance, manhattan distance, chebyshev distance, cosine distance, and the like, which is not limited in the present application.
It can be understood that, in the embodiment of the present application, for the training target of the vibration signal identification model to be trained, sample vectors corresponding to vibration signal samples output by the vibration signal identification model and belonging to the same vibration signal category are closer to each other, and sample vectors corresponding to vibration signal samples not belonging to the same vibration signal category are further away from each other, so that according to a preset loss function and a distance between a sample vector corresponding to each frequency domain signal sample in the query set and a category center vector of each vibration signal category, a loss value corresponding to each frequency domain signal sample in the sample query set is calculated, and then, according to the loss value, iterative training is performed on the vibration signal identification model to be trained, so that when a preset training end condition is reached, the trained vibration signal identification model and the category center vector of each vibration signal category corresponding to the current vibration signal identification model are obtained.
The preset training end condition may be loss function convergence, or may be that the number of model iterations reaches a preset number, which is not limited in the present application.
It can be understood that, each iterative training is performed with a selection of a sample support set and a sample query set, a category center vector of each vibration signal category of each iterative training is updated once, and when a preset training end condition is reached, a category center vector of each vibration signal category corresponding to the current vibration signal identification model may be a category center vector of each vibration signal category updated by the current iterative training, or may be a category center vector of each vibration signal category updated by the current vibration signal identification model after feature extraction processing is performed on all vibration signal samples by the current vibration signal identification model, and then a vector average value is calculated according to sample vectors corresponding to all frequency domain signal samples included in each vibration signal category to obtain a category center vector corresponding to the vibration signal category.
In some embodiments, performing feature extraction processing on a frequency domain signal sample in a sample support set according to a vibration signal recognition model to be trained to obtain a class center vector of a vibration signal class corresponding to the sample support set, including: performing feature extraction processing on the frequency domain signal samples in the sample support set according to the vibration signal identification model to be trained to obtain a sample vector corresponding to each frequency domain signal sample in the sample support set; and calculating the average value of the sample vectors corresponding to each frequency domain signal sample in the sample support set to obtain the class central vector of the vibration signal class corresponding to the sample support set.
And extracting a sample vector of each frequency domain signal sample in the sample support set according to a preset feature extraction algorithm. The preset feature extraction algorithm is an algorithm capable of processing the frequency domain signal samples to extract features, for example, the preset feature extraction algorithm is a convolutional neural network including at least one convolutional layer, the frequency domain signal samples are input into the convolutional neural network to perform convolution operation, the convolutional neural network outputs a feature map (feature map) with a size of W × H × C, and then dimension reduction processing is performed on the feature map with a size of W × H × C to obtain a sample vector with a size of 1 × 1 × (n × C). Wherein, C is the depth of convolution layer output, i.e. the number of convolution kernels, C is a hyper-parameter of the convolution neural network, which can be set by a user as required, W is the width of feature map, and H is the height of feature map.
The value of n may be determined by the adopted dimension reduction method, for example, after the convolutional neural network, the output three-dimensional feature map is subjected to dimension reduction through a global pooling layer, so as to obtain a sample vector with the size of 1 × 1 × C, and the global pooling layer may adopt global average pooling or global maximum pooling, etc. For another example, after the convolution neural network, a three-dimensional feature map is subjected to a flattening (flattening) operation to obtain a one-dimensional vector, i.e., a first feature vector having a size of 1 × 1 × (n × C), and n = W × H.
Then, calculating the average value of the sample vectors corresponding to each frequency domain signal sample in the sample support set to obtain the class center vector of the vibration signal class corresponding to the sample support set.
In some embodiments, the predetermined loss function includes an AM-Softmax loss function, which may decrease the class interval and increase the class interval.
Referring to fig. 5, fig. 5 is a schematic diagram of training and deployment of a vibration signal recognition model, as shown in fig. 5, a vibration signal sample set in a data storage is obtained, a constant Q transform process is performed on vibration signal samples in the vibration signal sample set to obtain a frequency domain signal sample set, and model training is performed on the vibration signal recognition model to be trained according to the frequency domain signal sample set to obtain a trained vibration signal recognition model. And then deploying the vibration signal identification model to execute vibration signal identification operation according to the deployed vibration signal identification model.
Optionally, in this embodiment, the motor fault detection method may be applied to an environment shown in fig. 6. As shown in fig. 6, the implementation environment includes a motor 610 and a vibration signal recognition server 620, a vibration signal recognition model is deployed in the motor 610 or the vibration signal recognition server 620, and the motor 610 and the vibration signal recognition server 620 may be directly or indirectly connected through a wired or wireless communication manner, which is not limited herein. The vibration signal recognition server 620 may be the same server as the server 120 for model training, or may be a different server, which is not limited in this application.
Referring to fig. 7, fig. 7 is a flowchart illustrating a motor fault detection method according to an exemplary embodiment of the present application. The motor failure detection method may be applied to the implementation environment shown in fig. 6, and is specifically performed by the vibration signal recognition server 620 in the implementation environment. It should be understood that the method may also be applied to other exemplary implementation environments and specifically executed by devices in other implementation environments, for example, if the vibration signal recognition model is disposed in the motor 610, the vibration signal recognition model may be specifically executed by the motor 610, and the embodiment is not limited to the implementation environment to which the method is applied.
The following describes in detail a vibration signal recognition model training method proposed in an embodiment of the present application with a vibration signal recognition server as a specific implementation subject.
As shown in fig. 7, in an exemplary embodiment, the vibration signal recognition model training method at least includes steps S710 to S750, which are described in detail as follows:
and step S710, obtaining a motor vibration signal to obtain a vibration signal to be identified.
The motor is correspondingly provided with a vibration signal collector to collect vibration signals generated by the motor through the vibration signal collector, and the collected motor vibration signals are sent to a vibration signal identification server as vibration signals to be identified.
Step S720, constant Q transformation processing is carried out on the vibration signal to be identified so as to convert the vibration signal to be identified into a frequency domain signal to be identified.
And the server performs constant Q transformation processing on the vibration signal to be identified to obtain a frequency domain signal to be identified so as to facilitate the vibration signal identification model to identify the vibration signal.
Step S730, inputting the frequency domain signal to be identified into a vibration signal identification model to obtain a vibration signal vector output by the vibration signal identification model, and acquiring a category center vector corresponding to each vibration signal category; the vibration signal recognition model and the category center vector corresponding to each vibration signal category are obtained according to a vibration signal recognition model training method.
And the vibration signal identification model performs characteristic extraction on the input frequency domain signal to be identified to obtain a vibration signal vector output by the vibration signal identification model.
Step S740, calculating a distance between the vibration signal vector and the category center vector corresponding to each vibration signal category, so as to obtain the vibration signal category corresponding to the vibration signal to be identified according to the distance.
And comparing the distances between the vibration signal vectors and the class center vectors corresponding to the vibration signal classes according to the obtained vibration signal vectors, and taking the vibration signal class with the minimum distance with the vibration signal vector as the vibration signal class corresponding to the vibration signal to be identified.
For example, the category center vector includes an abnormal vibration signal center vector and a normal vibration signal center vector, and the vector distance between the vibration signal vector and the abnormal vibration signal center vector is calculated as d 1 Calculating the vector distance d between the vibration signal vector and the normal vibration signal center vector 2 By judging d 1 Greater than d 2 And if so, the vibration signal category corresponding to the vibration signal to be identified is a normal vibration signal.
And step S750, generating a motor fault detection result according to the vibration signal category corresponding to the vibration signal to be identified.
For example, the vibration signal of the motor is judged in real time, and the obtained judgment result shows that the vibration signal is at t 1 To t 2 If the vibration signal category corresponding to the vibration signal in the time period is an abnormal vibration signal, it indicates that the motor has a fault in the time period, and t can be obtained 1 To t 2 To generate a motor fault detection result according to the motor operation record.
This application is through carrying out constant Q transform processing with the vibration signal sample, can effectively extract vibration signal's characteristic, then with the frequency domain signal sample input vibration signal recognition model that treats training after the constant Q transform processing, with the categorised central vector that each vibration signal category corresponds is calculated to the sample vector of vibration signal recognition model output according to treating training, and then treat the vibration signal recognition model of training according to the distance between sample vector and the categorised central vector and carry out the iterative training, obtain the vibration signal recognition model that the training was accomplished, with the data of the model processing of being convenient for more of input, improve the characteristic extraction effect of model to vibration signal, and avoid leading to the condition of model training overfitting because of the sample quantity imbalance in each vibration signal category, improve the effect of model training.
Fig. 8 is a block diagram illustrating a vibration signal recognition model training apparatus according to an embodiment of the present application, as shown in fig. 8, the apparatus including:
the sample obtaining module 810 is configured to obtain a vibration signal sample set, where vibration signal samples in the vibration signal sample set are labeled with vibration signal categories;
a sample conversion module 820 configured to perform a constant Q transform process on each vibration signal sample to convert each vibration signal sample into a frequency domain signal sample;
the sample feature extraction module 830 is configured to perform feature extraction processing on the frequency domain signal sample according to the vibration signal identification model to be trained to obtain a sample vector corresponding to the frequency domain signal sample, so as to calculate a category center vector corresponding to each vibration signal category according to the sample vector;
the training module 840 is configured to perform iterative training on the vibration signal recognition model to be trained according to the distance between the sample vector and the category center vector, so as to obtain the vibration signal recognition model after training and the category center vector corresponding to each vibration signal category.
In one embodiment of the present application, the sample conversion module 820 may include:
the preprocessing unit is configured to preprocess the vibration signal samples to obtain a vibration signal frame corresponding to each vibration signal sample;
the filtering processing unit is configured to filter the vibration signal frame according to a preset central frequency and bandwidth ratio to obtain an intermediate frequency domain signal sample;
and the logarithm calculation unit is configured to perform logarithm calculation on the intermediate frequency domain signal samples to obtain frequency domain signal samples corresponding to each vibration signal sample.
In one embodiment of the present application, the preprocessing unit may include:
the pre-emphasis processing unit is configured to perform pre-emphasis processing on the vibration signal sample to obtain an emphasized vibration signal sample;
a framing processing unit configured to perform framing processing on the emphasized vibration signal samples to obtain a plurality of vibration signal frames;
and the windowing processing unit is configured to carry out windowing processing on the plurality of vibration signal frames to obtain a plurality of continuous vibration signal frames.
In one embodiment of the present application, the sample feature extraction module 830 may include:
the sample selection unit is configured to perform sample selection on frequency domain signal samples belonging to the same vibration signal category to obtain a sample support set and a sample query set;
the characteristic extraction unit is configured to perform characteristic extraction processing on the frequency domain signal samples in the sample support set according to the vibration signal identification model to be trained to obtain a category central vector of a vibration signal category corresponding to the sample support set; performing feature extraction processing on the frequency domain signal samples in the sample query set according to the vibration signal identification model to be trained to obtain a sample vector corresponding to each frequency domain signal sample in the sample query set;
training module 840 may include:
the distance calculation unit is configured to calculate the distance between a sample vector corresponding to each frequency domain signal sample in the sample query set and a category center vector of each vibration signal category;
the loss value calculation unit is configured to calculate a loss value corresponding to each frequency domain signal sample in the sample query set according to a preset loss function and the distance;
and the iterative training unit is configured to perform iterative training on the vibration signal identification model to be trained according to the loss value so as to obtain the vibration signal identification model after training and the category center vector corresponding to each vibration signal category when a preset training end condition is reached.
In one embodiment of the present application, the sample selecting unit may include:
the target vibration signal type confirmation unit is configured to confirm the type of a vibration signal to be input in current iterative training to obtain the type of the target vibration signal;
the candidate sample set confirming unit is configured to obtain frequency domain signal samples belonging to the category of the target vibration signal to obtain a candidate sample set;
and the random selection unit is configured to randomly select the frequency domain signal samples in the candidate sample set to respectively obtain a sample support set and a sample query set, wherein the frequency domain signal samples between the sample support set and the sample query set are different.
In one embodiment of the present application, the feature extraction unit may include:
the single sample vector acquisition unit is configured to perform feature extraction processing on the frequency domain signal samples in the sample support set according to the vibration signal identification model to be trained to obtain a sample vector corresponding to each frequency domain signal sample in the sample support set;
and the average value calculating unit is configured to calculate an average value of the sample vector corresponding to each frequency domain signal sample in the sample support set to obtain a category center vector of the vibration signal category corresponding to the sample support set.
In one embodiment of the present application, the preset loss function comprises an AM-Softmax loss function.
Fig. 9 is a block diagram illustrating a motor failure detection apparatus according to an embodiment of the present application, and as shown in fig. 9, the apparatus includes:
the signal acquisition module 910 is configured to acquire a motor vibration signal to obtain a vibration signal to be identified;
the signal conversion module 920 is configured to perform constant Q transformation processing on the vibration signal to be identified so as to convert the vibration signal to be identified into a frequency domain signal to be identified;
a vector obtaining module 930, configured to input the frequency domain signal to be identified into the vibration signal identification model, obtain a vibration signal vector output by the vibration signal identification model, and obtain a category center vector corresponding to each vibration signal category; the vibration signal identification model and the category center vector corresponding to each vibration signal category are obtained according to any vibration signal identification model training method;
the signal identification module 940 is configured to calculate a distance between the vibration signal vector and a category center vector corresponding to each vibration signal category, so as to obtain the vibration signal category corresponding to the vibration signal to be identified according to the distance;
the detection result generating module 950 is configured to generate a motor fault detection result according to the vibration signal category corresponding to the vibration signal to be identified.
It should be noted that the vibration signal recognition model training device and the motor fault detection device provided in the foregoing embodiment belong to the same concept as the vibration signal recognition model training method and the motor fault detection method provided in the foregoing embodiment, and specific ways in which the respective modules and units perform operations have been described in detail in the method embodiments, and are not described herein again. In practical applications, the vibration signal recognition model training device and the motor fault detection device provided in the above embodiments may distribute the above functions through different functional modules as needed, that is, divide the internal structure of the device into different functional modules to complete all or part of the above described functions, which is not limited herein.
FIG. 10 illustrates a schematic structural diagram of a computer system suitable for use in implementing the electronic device of an embodiment of the present application.
It should be noted that the computer system 1000 of the electronic device shown in fig. 10 is only an example, and should not bring any limitation to the functions and the application scope of the embodiments of the present application.
As shown in fig. 10, the electronic device 1000 is in the form of a general purpose computing device. The components of the electronic device 1000 may include, but are not limited to: the at least one processing unit 1010, the at least one memory unit 1020, a bus 1030 connecting different system components (including the memory unit 1020 and the processing unit 1010), and a display unit 1040.
Where the storage unit stores program code that may be executed by the processing unit 1010 to cause the processing unit 1010 to perform the steps according to various exemplary embodiments of the present disclosure described in the "exemplary methods" section above in this specification.
The memory unit 1020 may include readable media in the form of volatile memory units, such as a random access memory unit (RAM) 1021 and/or a cache memory unit 1022, and may further include a read-only memory unit (ROM) 1023.
Storage unit 1020 may also include a program/utility 1024 having a set (at least one) of program modules 1025, such program modules 1025 including, but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may comprise an implementation of a network environment.
Bus 1030 may be any bus representing one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.
The electronic device 1000 may also communicate with one or more external devices 1070 (e.g., keyboard, pointing device, bluetooth device, etc.), with one or more devices that enable a user to interact with the electronic device 1000, and/or with any devices (e.g., router, modem, etc.) that enable the electronic device 1000 to communicate with one or more other computing devices. Such communication may occur through input/output (I/O) interfaces 1050. Also, the electronic device 1000 may communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network such as the internet) via the network adapter 1060. As shown, the network adapter 1060 communicates with the other modules of the electronic device 1000 over the bus 1030. It should be appreciated that although not shown, other hardware and/or application modules may be used in conjunction with the electronic device 1000, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, data backup storage systems, and the like.
In particular, according to embodiments of the present application, the processes described above with reference to the flow diagrams may be implemented as computer applications. For example, embodiments of the present application include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising a computer program for performing the method illustrated by the flow chart. Which when executed by the processing unit 1010, performs various functions defined in the system of the present application.
It should be noted that the computer readable medium shown in the embodiments of the present application may be a computer readable signal medium or a computer readable storage medium or any combination of the two. The computer readable storage medium may be, for example, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples of the computer readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a Read-Only Memory (ROM), an Erasable Programmable Read-Only Memory (EPROM), a flash Memory, an optical fiber, a portable Compact Disc Read-Only Memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present application, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In this application, however, a computer readable signal medium may include a propagated data signal with a computer program embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the foregoing.
The units described in the embodiments of the present application may be implemented by an application program or by hardware, and the described units may also be disposed in a processor. Wherein the names of the elements do not in some way constitute a limitation on the elements themselves.
Another aspect of the present application also provides a computer-readable storage medium on which a computer program is stored, which, when executed by a processor, implements the vibration signal recognition model training method or the motor fault detection method as before. The computer-readable storage medium may be included in the electronic device described in the above embodiment, or may exist separately without being incorporated in the electronic device.
Another aspect of the application also provides a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the vibration signal recognition model training method or the motor fault detection method provided in the above-described embodiments.
The above description is only a preferred exemplary embodiment of the present application, and is not intended to limit the embodiments of the present application, and one of ordinary skill in the art can easily make various changes and modifications according to the main concept and spirit of the present application, so that the protection scope of the present application shall be subject to the protection scope of the claims.

Claims (10)

1. A vibration signal recognition model training method is characterized by comprising the following steps:
obtaining a vibration signal sample set, wherein vibration signal samples in the vibration signal sample set are marked with vibration signal types;
performing constant Q transformation processing on each vibration signal sample to convert each vibration signal sample into a frequency domain signal sample;
performing feature extraction processing on the frequency domain signal sample according to a vibration signal identification model to be trained to obtain a sample vector corresponding to the frequency domain signal sample, and calculating a category central vector corresponding to each vibration signal category according to the sample vector;
and performing iterative training on the vibration signal identification model to be trained according to the distance between the sample vector and the category center vector to obtain the vibration signal identification model after training and the category center vector corresponding to each vibration signal category.
2. The method of claim 1, wherein said performing a constant Q transform on each of said vibration signal samples to convert each of said vibration signal samples into frequency domain signal samples comprises:
preprocessing the vibration signal samples to obtain a vibration signal frame corresponding to each vibration signal sample;
filtering the vibration signal frame according to a preset central frequency and bandwidth ratio to obtain an intermediate frequency domain signal sample;
and carrying out logarithmic calculation on the intermediate frequency domain signal samples to obtain frequency domain signal samples corresponding to each vibration signal sample.
3. The method of claim 2, wherein the pre-processing the vibration signal samples to obtain a vibration signal frame corresponding to each vibration signal sample comprises:
pre-emphasis processing is carried out on the vibration signal sample to obtain an emphasized vibration signal sample;
performing frame processing on the weighted vibration signal samples to obtain a plurality of vibration signal frames;
and windowing the plurality of vibration signal frames to obtain a plurality of continuous vibration signal frames.
4. The method according to claim 1, wherein the performing feature extraction processing on the frequency domain signal samples according to the vibration signal recognition model to be trained to obtain sample vectors corresponding to the frequency domain signal samples, so as to calculate class center vectors corresponding to respective vibration signal classes according to the sample vectors, includes:
selecting samples of frequency domain signal samples belonging to the same vibration signal category to obtain a sample support set and a sample query set;
performing feature extraction processing on the frequency domain signal samples in the sample support set according to the vibration signal identification model to be trained to obtain a category center vector of a vibration signal category corresponding to the sample support set; performing feature extraction processing on the frequency domain signal samples in the sample query set according to the vibration signal identification model to be trained to obtain a sample vector corresponding to each frequency domain signal sample in the sample query set;
the iterative training of the vibration signal recognition model to be trained is performed according to the distance between the sample vector and the category center vector to obtain the vibration signal recognition model after training and the category center vector corresponding to each vibration signal category, and the iterative training comprises the following steps:
calculating the distance between a sample vector corresponding to each frequency domain signal sample in the sample query set and the category center vector of each vibration signal category;
calculating a loss value corresponding to each frequency domain signal sample in the sample query set according to a preset loss function and the distance;
and performing iterative training on the vibration signal identification model to be trained according to the loss value so as to obtain the vibration signal identification model after training and a category center vector corresponding to each vibration signal category when a preset training finishing condition is reached.
5. The method of claim 4, wherein the sampling the frequency domain signal samples belonging to the same vibration signal category to obtain a sample support set and a sample query set comprises:
confirming the type of a vibration signal to be input in current iterative training to obtain the type of a target vibration signal;
acquiring frequency domain signal samples belonging to the category of the target vibration signal to obtain a candidate sample set;
and randomly selecting frequency domain signal samples in the candidate sample set to respectively obtain a sample support set and a sample query set, wherein the frequency domain signal samples between the sample support set and the sample query set are different.
6. The method according to claim 4, wherein the performing feature extraction processing on the frequency domain signal samples in the sample support set according to the vibration signal recognition model to be trained to obtain a category center vector of a vibration signal category corresponding to the sample support set includes:
performing feature extraction processing on the frequency domain signal samples in the sample support set according to the vibration signal identification model to be trained to obtain a sample vector corresponding to each frequency domain signal sample in the sample support set;
and calculating the average value of the sample vectors corresponding to each frequency domain signal sample in the sample support set to obtain the class center vector of the vibration signal class corresponding to the sample support set.
7. The method of claim 4, wherein the preset loss function comprises an AM-Softmax loss function.
8. A motor fault detection method, comprising:
obtaining a motor vibration signal to obtain a vibration signal to be identified;
performing constant Q transformation processing on the vibration signal to be identified so as to convert the vibration signal to be identified into a frequency domain signal to be identified;
inputting the frequency domain signal to be identified into a vibration signal identification model to obtain a vibration signal vector output by the vibration signal identification model, and acquiring a category center vector corresponding to each vibration signal category; the vibration signal recognition model and the class center vector corresponding to each vibration signal class are obtained according to the vibration signal recognition model training method of any one of claims 1 to 7;
calculating the distance between the vibration signal vector and the class center vector corresponding to each vibration signal class to obtain the vibration signal class corresponding to the vibration signal to be identified according to the distance;
and generating a motor fault detection result according to the vibration signal category corresponding to the vibration signal to be identified.
9. A vibration signal recognition model training apparatus, characterized in that the apparatus comprises:
the system comprises a sample acquisition module, a data processing module and a data processing module, wherein the sample acquisition module is configured to acquire a vibration signal sample set, and vibration signal samples in the vibration signal sample set are marked with vibration signal categories;
a sample conversion module configured to perform constant Q transform processing on each of the vibration signal samples to convert each of the vibration signal samples into frequency domain signal samples;
the sample feature extraction module is configured to perform feature extraction processing on the frequency domain signal sample according to a vibration signal identification model to be trained to obtain a sample vector corresponding to the frequency domain signal sample, and calculate a category center vector corresponding to each vibration signal category according to the sample vector;
and the training module is configured to perform iterative training on the vibration signal recognition model to be trained according to the distance between the sample vector and the category center vector to obtain the trained vibration signal recognition model and the category center vector corresponding to each vibration signal category.
10. A motor fault detection apparatus, the apparatus comprising:
the signal acquisition module is configured to acquire a motor vibration signal to obtain a vibration signal to be identified;
the signal conversion module is configured to perform constant Q conversion processing on the vibration signal to be identified so as to convert the vibration signal to be identified into a frequency domain signal to be identified;
the vector acquisition module is configured to input the frequency domain signal to be identified into a vibration signal identification model, obtain a vibration signal vector output by the vibration signal identification model, and acquire a category center vector corresponding to each vibration signal category; the vibration signal recognition model and the class center vector corresponding to each vibration signal class are obtained according to the vibration signal recognition model training method of any one of claims 1 to 7;
the signal identification module is configured to calculate a distance between the vibration signal vector and a category center vector corresponding to each vibration signal category so as to obtain the vibration signal category corresponding to the vibration signal to be identified according to the distance;
and the detection result generation module is configured to generate the motor fault detection result according to the vibration signal category corresponding to the vibration signal to be identified.
CN202211098693.3A 2022-09-08 2022-09-08 Vibration signal recognition model training method, motor fault detection method and device Withdrawn CN115577258A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211098693.3A CN115577258A (en) 2022-09-08 2022-09-08 Vibration signal recognition model training method, motor fault detection method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211098693.3A CN115577258A (en) 2022-09-08 2022-09-08 Vibration signal recognition model training method, motor fault detection method and device

Publications (1)

Publication Number Publication Date
CN115577258A true CN115577258A (en) 2023-01-06

Family

ID=84580661

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211098693.3A Withdrawn CN115577258A (en) 2022-09-08 2022-09-08 Vibration signal recognition model training method, motor fault detection method and device

Country Status (1)

Country Link
CN (1) CN115577258A (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109726740A (en) * 2018-12-05 2019-05-07 国网浙江省电力公司湖州供电公司 A kind of trade power consumption behavior analysis method based on clustering
CN109819282A (en) * 2017-11-22 2019-05-28 腾讯科技（深圳）有限公司 A kind of video user classification recognition methods, device and medium
CN111260032A (en) * 2020-01-14 2020-06-09 北京迈格威科技有限公司 Neural network training method, image processing method and device
CN112353402A (en) * 2020-10-22 2021-02-12 平安科技（深圳）有限公司 Training method of electrocardiosignal classification model, electrocardiosignal classification method and device
CN114118370A (en) * 2021-11-19 2022-03-01 北京的卢深视科技有限公司 Model training method, electronic device, and computer-readable storage medium
CN114722812A (en) * 2022-04-02 2022-07-08 尚蝉(浙江)科技有限公司 Method and system for analyzing vulnerability of multi-mode deep learning model

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109819282A (en) * 2017-11-22 2019-05-28 腾讯科技（深圳）有限公司 A kind of video user classification recognition methods, device and medium
CN109726740A (en) * 2018-12-05 2019-05-07 国网浙江省电力公司湖州供电公司 A kind of trade power consumption behavior analysis method based on clustering
CN111260032A (en) * 2020-01-14 2020-06-09 北京迈格威科技有限公司 Neural network training method, image processing method and device
CN112353402A (en) * 2020-10-22 2021-02-12 平安科技（深圳）有限公司 Training method of electrocardiosignal classification model, electrocardiosignal classification method and device
CN114118370A (en) * 2021-11-19 2022-03-01 北京的卢深视科技有限公司 Model training method, electronic device, and computer-readable storage medium
CN114722812A (en) * 2022-04-02 2022-07-08 尚蝉(浙江)科技有限公司 Method and system for analyzing vulnerability of multi-mode deep learning model

Similar Documents

Publication Publication Date Title
US20230081645A1 (en) Detecting forged facial images using frequency domain information and local correlation
CN112785016A (en) New energy automobile maintenance and fault monitoring and diagnosis method based on machine learning
CN110995475A (en) Power communication network fault detection method based on transfer learning
CN108831505B (en) Method and device for identifying use scenes of application
CN105488539B (en) The predictor method and device of the generation method and device of disaggregated model, power system capacity
CN113886181B (en) Dynamic threshold prediction method, device and medium applied to AIOps fault early warning
CN111863014A (en) Audio processing method and device, electronic equipment and readable storage medium
CN113125135A (en) Fault diagnosis method for rotary machine, storage medium, and electronic device
CN114048787B (en) Method and system for intelligently diagnosing bearing fault in real time based on Attention CNN model
CN113450828A (en) Music genre identification method, device, equipment and storage medium
CN115238740A (en) Fault diagnosis method and device based on time-frequency distribution and deep learning
CN112489623A (en) Language identification model training method, language identification method and related equipment
CN109088793B (en) Method and apparatus for detecting network failure
CN114004996A (en) Abnormal sound detection method, abnormal sound detection device, electronic equipment and medium
CN113870903A (en) Pathological voice recognition method, device, equipment and storage medium
CN113758709A (en) Rolling bearing fault diagnosis method and system combining edge calculation and deep learning
CN110988673B (en) Motor rotor fault detection method and device and terminal equipment
CN113129122A (en) Financial risk early warning auditing method and device, electronic equipment and storage medium
WO2023185801A1 (en) Vibration signal identification method and apparatus
CN115577258A (en) Vibration signal recognition model training method, motor fault detection method and device
CN117079005A (en) Optical cable fault monitoring method, system, device and readable storage medium
CN116741159A (en) Audio classification and model training method and device, electronic equipment and storage medium
CN116386669A (en) Machine running acoustic state monitoring method and system based on block automatic encoder
CN113555031B (en) Training method and device of voice enhancement model, and voice enhancement method and device
CN113035176B (en) Voice data processing method and device, computer equipment and storage medium

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
WW01 Invention patent application withdrawn after publication

Application publication date: 20230106

WW01 Invention patent application withdrawn after publication