A pump machine state monitoring method and device based on a machine learning SVM model

Abstract

The application provides a pump state monitoring method based on a machine learning SVM model, which comprises the following steps: collecting audio signals of a training pump; preprocessing the audio signals; setting a custom mel filter bank, calling the custom mel filter bank to process each frame of signal, extracting the mel frequency cepstral coefficient features of each frame, and combining the mel frequency cepstral coefficient features of all frames into a first feature matrix; standardizing the extracted first feature matrix; taking the first feature matrix as input, training independent one-class support vector machine models according to the running state of the training pump; and using the one-class support vector machine models to judge the running state of a pump to be tested. The application has the advantages of: low misjudgment rate due to independent training of multiple models; improved feature recognition and abnormal recognition accuracy due to the use of non-contact audio acquisition and a custom filter bank; support for end-to-end real-time monitoring, and suitability for deployment of various pumps.

Description

Technical Field

[0001] This application relates to the field of industrial equipment condition monitoring and fault diagnosis technology, and more specifically, to a pump condition monitoring method and device based on a machine learning SVM model. Background Technology

[0002] Pumps are key power equipment in industrial production, and their operating status directly affects production safety and efficiency. Currently, pump condition monitoring mainly employs the following methods:

[0003] Monitoring pump vibration signals requires installing contact vibration sensors on the pump surface to determine the equipment status by collecting vibration signals. The disadvantages of this method are: complex sensor installation, high cost, and unsuitability for enclosed or high-speed rotating pump structures.

[0004] The pump's status is determined by setting thresholds for one or more physical quantities. The drawback of this method is that it can only identify extreme faults, cannot distinguish between early anomalies and normal fluctuations, and has a high false alarm rate.

[0005] Monitoring is based on traditional multi-class support vector machines. The drawback of this method is that it requires collecting training samples for all fault types, but obtaining various fault samples in industrial settings is difficult and costly, resulting in poor model generalization and making practical deployment challenging.

[0006] In summary, there is a need in this field to provide a pump condition monitoring method and device based on a machine learning SVM model to overcome the shortcomings of existing technologies. Summary of the Invention

[0007] This application provides a pump condition monitoring method and apparatus based on a machine learning SVM model, which can solve the problems existing in the prior art. The objective of this application is achieved through the following technical solution.

[0008] In a first aspect, one embodiment of this application provides a pump condition monitoring method based on a machine learning SVM model, which includes several steps:

[0009] Step S1: Collect the audio signal of the training pump operation;

[0010] Step S2: Preprocess the audio signal; wherein, the preprocessing includes at least noise reduction, amplitude normalization and frame segmentation.

[0011] Step S3: Set up a custom Mel filter bank, call the custom Mel filter bank to process each frame of signal, extract the Mel frequency cepstral coefficient features of each frame, and combine the Mel frequency cepstral coefficient features of all frames into the first feature matrix; where, the custom Mel filter bank refers to a triangular filter bank with non-standard Mel frequency scale and linear frequency division method that is optimized for the audio characteristics of the pump.

[0012] Step S4: Standardize the extracted first feature matrix; the standardization process uses Z-score standardization, Min-Max normalization, or MaxAbs normalization.

[0013] Step S5: Using the first feature matrix as input, train independent single-class support vector machine models according to the operating states of the training pump; wherein, the operating state includes at least one state representing the normal operation of the pump, and each operating state corresponds to an independent single-class support vector machine model.

[0014] Step S6: Use a single-class support vector machine model to determine the operating state of the pump under test; wherein, when the single-class support vector machine model determines that the operating state of the pump under test is consistent with the operating state corresponding to its training, it outputs a first value; otherwise, it outputs a second value.

[0015] According to one embodiment of the present application, the pump status monitoring method includes the following steps in determining the status of the pump under test using a single-class support vector machine model:

[0016] Step S61: Collect the audio signal of the pump under test during operation;

[0017] Step S62: Preprocess the audio signal of the pump under test;

[0018] Step S63: Call the custom Mel filter bank to process the signal of each frame, extract the Mel frequency cepstral coefficient features of each frame, and combine the Mel frequency cepstral coefficient features of all frames into the second feature matrix;

[0019] Step S64: Based on the second feature matrix, use the single-class support vector machine model corresponding to each operating state to determine the operating state of the pump under test.

[0020] According to the pump condition monitoring method provided in one embodiment of this application, setting a custom Mel filter bank includes the following steps:

[0021] Step S31: Initialize the custom Mel filter bank, set the number of filters p to 48, the number of FFT points to 1024, and the frequency coverage range to 0 to 1 / 2 of the sampling rate;

[0022] Step S32: Linearly divide the frequency range from 0 to Fs / 2 into p+2 equally divided frequency points;

[0023] Step S33: Calculate the frequency resolution based on the number of FFT points, and convert the frequency points into FFT frequency domain index values;

[0024] Step S34: Generate the response curves of multiple filters in the form of a triangular window function to form a filter bank matrix.

[0025] According to the pump status monitoring method provided in one embodiment of the present application, the operating status further includes one or more states representing pump abnormalities, and each operating state corresponds to an independent single-class support vector machine model.

[0026] According to one embodiment of the present application, a pump status monitoring method is provided, wherein the enframe function is used, with a frame length of 1024 and a frame shift of 512.

[0027] According to one embodiment of the present application, the pump condition monitoring method includes a first feature matrix consisting of 6 to 48 Vimer frequency cepstral coefficient feature matrices.

[0028] The pump condition monitoring method according to one embodiment of this application further includes:

[0029] Step S7: Calculate the false positive rate and false positive type, and generate a monitoring report.

[0030] According to the pump status monitoring method provided in one embodiment of this application, the preprocessing further includes windowing each frame of signal using a Hanning window, and then performing a fast Fourier transform on each frame of signal.

[0031] Secondly, one embodiment of this application provides a pump condition monitoring device based on a machine learning SVM model, which includes:

[0032] Training acquisition module: used to acquire audio signals from the training pump operation;

[0033] The training audio preprocessing module is used to preprocess the audio signal; the preprocessing includes at least noise reduction, amplitude normalization and frame segmentation.

[0034] Training Feature Matrix Generation Module: Used to set up a custom Mel filter bank, call the custom Mel filter bank to process the signal of each frame, extract the Mel frequency cepstral coefficient features of each frame, and combine the Mel frequency cepstral coefficient features of all frames into the first feature matrix; where, the custom Mel filter bank refers to a triangular filter bank with non-standard Mel frequency scale and linear frequency division method that is optimized for the audio characteristics of the pump.

[0035] Training data standardization module: used to standardize the extracted first feature matrix; the standardization process uses Z-score standardization, Min-Max normalization or MaxAbs normalization.

[0036] Training module: Used to train independent single-class support vector machine models based on the first feature matrix as input and the operating state of the training pump; wherein, the operating state includes at least one state representing the normal operation of the pump, and each operating state corresponds to an independent single-class support vector machine model;

[0037] Monitoring module: used to determine the operating status of the pump under test using a single-class support vector machine model; wherein, when the single-class support vector machine model determines that the operating status of the pump under test is consistent with the operating status corresponding to its training, it outputs a first value; otherwise, it outputs a second value.

[0038] According to one embodiment of the present application, a pump condition monitoring device is provided, wherein the monitoring module includes:

[0039] Monitoring and acquisition submodule: used to acquire audio signals of the pump under test during operation;

[0040] Audio preprocessing submodule for monitoring: used to preprocess the audio signal of the pump under test;

[0041] Monitoring data standardization submodule: used to call a custom Mel filter bank to process each frame of signal, extract the Mel frequency cepstral coefficient features of each frame, and combine the Mel frequency cepstral coefficient features of all frames into a second feature matrix;

[0042] Monitoring and Judgment Submodule: Based on the second feature matrix, this module uses the single-class support vector machine model corresponding to each operating state to determine the operating state of the pump under test.

[0043] According to one embodiment of the present application, the pump status monitoring device, wherein processing each frame of signal by calling a custom Mel filter bank and extracting the Mel frequency cepstral coefficient features of each frame includes calling the following modules:

[0044] Initialization submodule: Initialize the custom Mel filter bank, set the number of filters p to 48, the number of FFT points to 1024, and the frequency coverage range to 0 to 1 / 2 of the sampling rate;

[0045] Frequency division point submodule: linearly divides the frequency range from 0 to Fs / 2 into p+2 equal frequency points;

[0046] Frequency conversion submodule: Calculates the frequency resolution based on the number of FFT points and converts the frequency points into FFT frequency domain index values;

[0047] The filter matrix submodule generates the response curves of multiple filters in the form of a triangular window function, forming a filter bank matrix.

[0048] According to the pump status monitoring device provided in one embodiment of the present application, the operating status further includes one or more states representing pump abnormalities, and each operating state corresponds to an independent single-class support vector machine model.

[0049] According to one embodiment of the present application, a pump condition monitoring device is provided, wherein the first feature matrix is ​​a 6 to 48 Vimer frequency cepstral coefficient feature matrix.

[0050] The pump condition monitoring device according to one embodiment of this application further includes a pump condition monitoring method that includes:

[0051] The reporting module is used to calculate the false positive rate and false positive type, and generate monitoring reports.

[0052] According to one embodiment of the present application, the pump status monitoring device further includes preprocessing by applying a Hanning window to each frame of signal and then performing a fast Fourier transform on each frame of signal.

[0053] Thirdly, one embodiment of this application provides an electronic device comprising:

[0054] One or more processors;

[0055] Storage device, on which one or more programs are stored,

[0056] When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any of the implementations of the first aspect.

[0057] Fourthly, one embodiment of this application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by one or more processors, implements the method as described in any implementation of the first aspect.

[0058] The advantages of the pump condition monitoring method and apparatus based on the machine learning SVM model according to the embodiments of this application are as follows: it is applicable to pump audio collected in a non-contact manner, can be adapted to various pumps, and reduces deployment costs; the custom filter bank improves the recognizability of pump audio features and increases the accuracy of anomaly identification; it allows training the model using only normal samples, solving the problem of insufficient fault samples in industrial sites; the end-to-end process supports embedded real-time deployment, realizing online determination of pump condition; multiple models are trained independently, which can independently judge different types of pump anomalies, reducing the false judgment rate. Attached Figure Description

[0059] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0060] Figure 1 An exemplary system architecture diagram is shown, in which one embodiment of this application can be applied.

[0061] Figure 2 A flowchart of a pump condition monitoring method based on a machine learning SVM model according to one embodiment of this application is shown.

[0062] Figure 3 A flowchart illustrating the determination of the state of a pump under test using a single-class support vector machine model according to one embodiment of this application is shown.

[0063] Figure 4 The flowchart illustrates a process according to one embodiment of this application, invoking a custom Mel filter bank to process each frame of signal, extracting the Mel frequency cepstral coefficient features of each frame, and combining the Mel frequency cepstral coefficient features of all frames into a first feature matrix.

[0064] Figure 5 A schematic diagram of one embodiment of a pump condition monitoring device based on a machine learning SVM model according to one embodiment of this application is shown.

[0065] Figure 6 A schematic diagram of the structure of a computer system of an electronic device according to one embodiment of this application is shown. Detailed Implementation

[0066] The specific embodiments of this application are described below with reference to the accompanying drawings and examples. Through the content described in this specification, those skilled in the art can clearly and completely understand the technical solution, the technical problem solved, and the resulting technical effects of this application. It is understood that the specific embodiments described herein are only for explaining this application and not for limiting it. Furthermore, for ease of description, only the parts related to this application are shown in the accompanying drawings.

[0067] It should be noted that the structures, proportions, sizes, etc., shown in the accompanying drawings are only for the purpose of assisting those skilled in the art in understanding and reading the contents described in the specification, and are not intended to limit the conditions under which this application can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportional relationships, or adjustments to the size should fall within the scope of the technical content disclosed in this application, provided that they do not affect the effects and purposes that this application can produce.

[0068] The use of terms such as "first," "second," and "the" does not imply quantity limitation and may indicate singular or plural. The terms "comprising," "including," "having," and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules is not limited to the listed steps or units, but may also include steps or units not listed, or other steps or units inherent to these processes, methods, products, or devices. The terms "connected," "linked," and "coupled" used in this application are not limited to physical or mechanical connections, but may also include direct or indirect electrical connections.

[0069] Figure 1 An exemplary system architecture is shown, illustrating an embodiment of the pump condition monitoring method of this application. For example... Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0070] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104, thereby loading pages to be tested, etc. Terminal devices 101, 102, and 103 can be hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various electronic devices with displays, including but not limited to smartphones, tablets, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 players (Moving Picture Experts Group Audio Layer IV), laptops, and desktop computers, etc. When terminal devices 101, 102, and 103 are software, they can be installed on the terminal devices listed above. They can be implemented as multiple software programs or software modules (e.g., used to run pump status monitoring software) or as a single software program or software module. No specific limitations are made here.

[0071] In some cases, the pump status monitoring method provided in this application can be executed by terminal devices 101, 102, and 103, and correspondingly, the pump status monitoring device can be installed in terminal devices 101, 102, and 103. In this case, the system architecture 100 may not include server 105.

[0072] In some cases, the pump status monitoring method provided in this application can be jointly executed by terminal devices 101, 102, 103 and server 105. This application does not limit this. Correspondingly, the pump status monitoring device can also be separately installed in terminal devices 101, 102, 103 and server 105.

[0073] In some cases, the pump status monitoring method provided in this application can be executed by server 105. Accordingly, the pump status monitoring device can also be set in server 105. In this case, the system architecture 100 may not include terminal devices 101, 102, and 103.

[0074] It should be noted that server 105 can be either hardware or software. When server 105 is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When server 105 is software, it can be implemented as multiple software programs or software modules, or as a single software program or software module. No specific limitations are made here.

[0075] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0076] Figure 2 A flowchart illustrating a pump condition monitoring method based on a machine learning SVM model according to one embodiment of this application is shown. Figure 2 As shown, the pump condition monitoring method includes several steps:

[0077] Step S1: Collect the audio signal of the training pump operation;

[0078] Step S2: Preprocess the audio signal; wherein, the preprocessing includes at least noise reduction, amplitude normalization and frame segmentation.

[0079] Step S3: Call the custom Mel filter bank to process the signal of each frame, extract the Mel-Frequency Cepstral Coefficients (MFCC features) of each frame, and combine the Mel-Frequency Cepstral Coefficients of all frames into the first feature matrix; where the custom Mel filter bank refers to a triangular filter bank with non-standard Mel frequency scale and linear frequency division method that is optimized for the audio characteristics of the pump.

[0080] Step S4: Standardize the extracted first feature matrix; the standardization process uses Z-score standardization, Min-Max normalization, or MaxAbs normalization.

[0081] Step S5: Using the first feature matrix as input, train independent one-class support vector machine models (i.e., OneClassSVM) according to the operating state of the training pump; wherein, the operating state includes at least one state representing the normal operation of the pump, and each operating state corresponds to an independent one-class support vector machine model.

[0082] Step S6: Use a single-class support vector machine model to determine the operating status of the pump under test; wherein, when the single-class support vector machine model determines that the operating status of the pump under test is consistent with the operating status corresponding to its training, the single-class support vector machine model outputs a first value (e.g., "1"); otherwise, it outputs a second value (e.g., "0").

[0083] According to the pump status monitoring method provided in one embodiment of this application, the single-class support vector machine model (i.e., OneClassSVM) refers to an unsupervised / semi-supervised learning model trained using only samples of a single class (i.e., normal state) to determine whether a new sample belongs to that class (i.e., whether it is abnormal). Its goal is to learn a compact boundary in the feature space for the training samples, such that normal samples are located inside the boundary and abnormal samples are located outside the boundary. This invention trains a OneClassSVM model for the normal state of the pump. The type and number of OneClassSVM models corresponding to the abnormal state can be selected based on the sample situation of the abnormal state. Each model only needs samples corresponding to the state to be trained. During inference, the current state of the pump is determined by combining the output results of all models.

[0084] According to the pump status monitoring method provided in one embodiment of this application, those skilled in the art can set the training parameters of OneClassSVM based on experience. Preferably, nu is 0.01 to 0.1, the kernel function is RBF, and gamma is set to 'scale' or determined according to the reciprocal of the feature dimension.

[0085] According to the pump status monitoring method provided in one embodiment of this application, the Mel frequency cepstral coefficient feature (i.e., MFCC feature) is an audio feature representation method based on human auditory perception characteristics or a custom frequency domain division method. In this invention, the MFCC feature extraction employs a custom linearly divided Mel filter bank, typically with 48 triangular filters covering the entire frequency band from 0 to half the sampling rate. Six to 48-dimensional MFCC coefficients are extracted from each frame to form a feature matrix. This feature can effectively characterize the differences in acoustic signature between normal pump operation and various fault states.

[0086] According to the pump condition monitoring method provided in one embodiment of this application, the first feature matrix is ​​a 6- to 48-dimensional Mehmed frequency cepstral coefficient feature matrix. The lower dimension is used for rapid monitoring, and the higher dimension is used for detailed diagnosis. Preferably, the first feature matrix is ​​a 24-dimensional Mehmed frequency cepstral coefficient feature matrix.

[0087] According to the pump status monitoring method provided in one embodiment of this application, the noise reduction processing can be selected by those skilled in the art based on the pump model and / or usage scenario. For example, noise reduction can be performed using a Butterworth bandpass filter with a cutoff frequency of 50Hz to 22kHz, or steady-state background noise can be removed using spectral subtraction.

[0088] Figure 3 A flowchart illustrating the determination of the state of a pump under test using a single-class support vector machine model according to one embodiment of this application is shown. Figure 3 As shown, determining the state of the pump under test using a single-class support vector machine model includes the following steps:

[0089] Step S61: Collect the audio signal of the pump under test during operation;

[0090] Step S62: Preprocess the audio signal of the pump under test;

[0091] Step S63: Call the custom Mel filter bank to process the signal of each frame, extract the Mel frequency cepstral coefficient features of each frame, and combine the Mel frequency cepstral coefficient features of all frames into the second feature matrix;

[0092] Step S64: Based on the second feature matrix, use the single-class support vector machine model corresponding to each operating state to determine the operating state of the pump under test.

[0093] According to the pump condition monitoring method provided in one embodiment of this application, setting a custom Mel filter bank includes the following steps:

[0094] Step S31: Initialize the custom Mel filter bank; wherein, the initialization includes setting the number of filters p to 48, the number of FFT points to 1024, and the frequency coverage range to 0 to 1 / 2 of the sampling rate;

[0095] Step S32: Linearly divide the frequency range from 0 to Fs / 2 into p+2 equally divided frequency points;

[0096] Step S33: Calculate the frequency resolution based on the number of FFT points, and convert the frequency points into FFT frequency domain index values;

[0097] Step S34: Generate the response curves of multiple filters in the form of a triangular window function to form a filter bank matrix.

[0098] According to the pump status monitoring method provided in one embodiment of this application, the Mel frequency cepstral coefficient features of each frame can be obtained by taking the logarithm of the energy output of each filter and then performing a discrete cosine transform (DCT) to finally obtain the Mel frequency cepstral coefficient features of each frame (this general method will lose a lot of information for industrial equipment audio acquisition and judgment).

[0099] According to the pump status monitoring method provided in one embodiment of this application, the acquisition device does not need to be in direct contact with the pump when acquiring audio signals of the pump under test and the pump under training. The type and location of the acquisition device can be selected by those skilled in the art based on their professional knowledge. For example, an omnidirectional condenser microphone with a sampling rate of 48kHz and a bit depth of 16 bits can be used, with the microphone placed 0.5 meters in front of the pump housing and axially aligned with the pump bearing.

[0100] According to the pump condition monitoring method provided in one embodiment of this application, the number of filters can be adjusted according to actual monitoring needs, for example, 48 or 32.

[0101] According to the pump condition monitoring method provided in one embodiment of this application, the operating state further includes one or more states representing pump abnormalities, each operating state corresponding to an independent single-class support vector machine model. Preferably, the number of abnormal states is greater than one. Abnormal states include, but are not limited to, bearing failure, impeller wear, cavitation, rotor imbalance, gear failure, poor lubrication, and loose foundation.

[0102] The pump condition monitoring method according to one embodiment of this application further includes:

[0103] Step S7: Calculate the misjudgment rate and misjudgment type, and generate a monitoring report; where the misjudgment rate refers to the proportion of the number of incorrectly judged samples on the validation set to the total number of test samples.

[0104] According to the pump status monitoring method provided in one embodiment of this application, the preprocessing further includes windowing each frame of signal using a Hanning window, and then performing a fast Fourier transform on each frame of signal.

[0105] Secondly, Figure 5 A schematic diagram of one embodiment of a pump condition monitoring device based on a machine learning SVM model, according to one embodiment of this application, is shown. Figure 5 As shown, the pump condition monitoring device includes:

[0106] Training acquisition module 201: Used to acquire audio signals from the training pump operation;

[0107] Training audio preprocessing module 202: used to preprocess audio signals; wherein, the preprocessing includes at least noise reduction, amplitude normalization and frame segmentation.

[0108] Training Feature Matrix Generation Module 203: Used to set up a custom Mel filter bank, call the custom Mel filter bank to process each frame of signal, extract the Mel frequency cepstral coefficient features of each frame, and combine the Mel frequency cepstral coefficient features of all frames into the first feature matrix; where, the custom Mel filter bank refers to a triangular filter bank with non-standard Mel frequency scale and linear frequency division method that is optimized for the audio characteristics of the pump.

[0109] Training data standardization module 204: used to standardize the extracted first feature matrix; wherein, the standardization process adopts Z-score standardization, Min-Max normalization or MaxAbs normalization;

[0110] Training module 205: Used to train independent single-class support vector machine models based on the first feature matrix and the operating state of the training pump, respectively; wherein, the operating state includes at least one state representing the normal operation of the pump, and each operating state corresponds to an independent single-class support vector machine model.

[0111] Monitoring module 206: Used to determine the operating status of the pump under test using a single-class support vector machine model.

[0112] It should be noted that, for clarity, the various sub-modules of the pump condition monitoring device are not listed in the table below. Figure 5 As shown in the image.

[0113] According to one embodiment of the present application, a pump condition monitoring device is provided, wherein the monitoring module includes:

[0114] Monitoring and acquisition submodule: Acquires audio signals from the pump under test during operation;

[0115] Audio preprocessing submodule for monitoring: preprocesses the audio signal of the pump under test;

[0116] Monitoring data standardization submodule: calls a custom Mel filter bank to process each frame of signal, extracts the Mel frequency cepstral coefficient features of each frame, and combines the Mel frequency cepstral coefficient features of all frames into a second feature matrix;

[0117] Monitoring and Judgment Submodule: Based on the second feature matrix, the single-class support vector machine model corresponding to each operating state is used to judge the operating state of the pump under test.

[0118] According to one embodiment of the present application, the pump status monitoring device, wherein processing each frame of signal by calling a custom Mel filter bank and extracting the Mel frequency cepstral coefficient features of each frame includes calling the following modules:

[0119] Initialization submodule: Initialize the custom Mel filter bank, set the number of filters p to 48, the number of FFT points to 1024, and the frequency coverage range to 0 to 1 / 2 of the sampling rate;

[0120] Frequency division point submodule: linearly divides the frequency range from 0 to Fs / 2 into p+2 equal frequency points;

[0121] Frequency conversion submodule: Calculates the frequency resolution based on the number of FFT points and converts the frequency points into FFT frequency domain index values;

[0122] The filter matrix submodule generates the response curves of multiple filters in the form of a triangular window function, forming a filter bank matrix.

[0123] According to the pump status monitoring device provided in one embodiment of the present application, the operating status further includes one or more states representing pump abnormalities, and each operating state corresponds to an independent single-class support vector machine model.

[0124] According to the pump status monitoring method provided in one embodiment of this application, the enframe function is used, and the frame length and frame shift can be set according to actual monitoring requirements, for example: frame length 1024, frame shift 512; or frame length 512, frame shift 256.

[0125] According to one embodiment of the present application, a pump condition monitoring device is provided, wherein the first feature matrix is ​​a 6 to 48 Vimer frequency cepstral coefficient feature matrix.

[0126] The pump condition monitoring device according to one embodiment of this application further includes a pump condition monitoring method that includes:

[0127] Report module 207 is used to calculate the false positive rate and false positive type, and generate monitoring reports.

[0128] According to one embodiment of the present application, the pump status monitoring device further includes preprocessing by applying a Hanning window to each frame of signal and then performing a fast Fourier transform on each frame of signal.

[0129] Thirdly, one embodiment of this application provides an electronic device comprising:

[0130] One or more processors;

[0131] Storage device, on which one or more programs are stored,

[0132] When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any of the implementations of the first aspect.

[0133] Fourthly, Figure 6 A schematic diagram of the structure of a computer system for implementing embodiments of the present application is shown. Figure 6 The computer system 500 shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this application. Figure 6 As shown, the computer system 500 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 501, which is capable of performing various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the computer system 500. The processing device 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0134] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 508 including, for example, magnetic tapes, hard disks, etc.; and communication devices 509. Communication device 509 allows computer system 500 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 A computer system 500 with various electronic devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0135] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including functions for executing... Figure 1 The flowchart illustrates the program code for the method. In such an embodiment, the computer program can be downloaded and installed from a network via communication device 509, or installed from storage device 508, or installed from ROM 502. When the computer program is executed by processing device 501, it performs the functions defined in the method of the embodiments of this application.

[0136] It should be noted that the two embodiments given below are merely illustrative, and the values ​​assigned to the various parameters are not intended to be limiting. Those skilled in the art can reasonably set the relevant parameters when implementing the technical solutions in this application according to the actual circumstances.

[0137] Example 1: Training an independent OneClassSVM model and using it

[0138] 1. Pump audio signal acquisition: Read the pump operation audio file wav and convert the signal into double-precision floating-point type xx=double(x).

[0139] 2. Signal preprocessing and framing: The audio signal is preprocessed for denoising by using enframe (xx,1024,512) framing, with a frame length of 1024 and a frame shift of 512, to adapt to the temporal characteristics of the pump's audio.

[0140] 3. Custom Mel filter bank feature extraction: Call customFilterBank(fs) to generate a filter bank with 48 filters covering the frequency domain from 0 to fs / 2; add Hanning window and FFT transform to each frame of signal, and extract 6-48 dimension features after filtering by the filter bank to obtain a high-recognition MFCC feature matrix m.

[0141] 4. Feature standardization and model training: Merge the MFCC features of multiple pump states and standardize them using StandardScaler; train independent OneClassSVM models for pump states 1 to 7, with parameters nu=0.05, kernel function rbf, and gamma=scale, and save the models and standardizers.

[0142] 5. State Reasoning and Anomaly Detection: Input the audio of the pump to be tested, extract the same features and standardize them, and use the pump state corresponding model of 1 to 7 in sequence for prediction; the output is 1 and judged as normal, and the output is -1 and judged as abnormal. At the same time, the misjudgment rate and misjudgment type are statistically analyzed.

[0143] Example 2: customFilterBank Feature Extraction Method

[0144] Step 1: Parameter Initialization and Sampling Rate Configuration

[0145] The custom filter bank construction function takes the audio sampling rate Fs as the input parameter; if no sampling rate is input, the default industrial standard sampling rate of 48000Hz is used; set the number of filters p=48, the number of FFT transform points nfft=1024, and the frequency coverage range from 0 to 1 / 2 of the sampling rate.

[0146] Step 2: Linear division of frequency range

[0147] The frequency range is divided into equal segments using a linear method, generating p+2 equal frequency points within the range of 0 to Fs / 2. This ensures that the filter bank uniformly covers the motor's full-frequency operating acoustic signal and adapts to different rotor frequencies and cylinder resonant frequency bands.

[0148] Step 3: Frequency Domain Index Calculation

[0149] The frequency resolution df = Fs / nfft is calculated based on the number of FFT points. The divided frequency points are converted into FFT frequency domain index values, and the left boundary index n1, center index n0, and right boundary index n2 of each filter are obtained respectively. This is adapted to the MATLAB array indexing rule that starts from 1, where nfft is the number of FFT transform points.

[0150] Step 4: Generating the triangular filter

[0151] Initialize a filter bank matrix with 24 rows, for example, 48K data points and nfft / 2 columns; iterate through the 48 filters and construct the response curve for each filter using a triangular window function:

[0152] • The response value increases linearly from the left boundary to the center of the filter;

[0153] • The response value decreases linearly from the center of the filter to the right boundary; finally, a triangular filter bank with bandpass filtering characteristics is generated to realize frequency domain weighting and feature enhancement of the motor acoustic signal.

[0154] Step 5: Filter bank output

[0155] The constructed filter bank matrix is ​​used as the output result for subsequent frequency domain filtering and feature extraction of motor acoustic signals.

[0156] The advantages of the pump condition monitoring method and apparatus based on the machine learning SVM model according to the embodiments of this application are as follows: it is applicable to pump audio collected in a non-contact manner, can be adapted to various pumps, and reduces deployment costs; the custom filter bank improves the recognizability of pump audio features and increases the accuracy of anomaly identification; it allows training the model using only normal samples, solving the problem of insufficient fault samples in industrial sites; the end-to-end process supports embedded real-time deployment, realizing online determination of pump condition; multiple models are trained independently, which can independently judge different types of pump anomalies, reducing the false judgment rate.

[0157] Although this application has been described and illustrated with reference to specific embodiments thereof, such descriptions and illustrations are not intended to limit the application. It will be readily understood by those skilled in the art that various changes can be made and equivalent elements can be substituted within embodiments without departing from the scope of protection of this application as defined by the claims. Differences may exist between the technical representation in this application and actual equipment due to variables in the manufacturing process, etc. Other embodiments of this application may exist that are not specifically described. The specification and illustrations should be considered illustrative rather than restrictive, and modifications can be made to suit the purpose and spirit of this application, all of which are within the scope of the claims. While the methods disclosed herein have been described with reference to specific operations performed in a particular order, it should be understood that these operations can be rearranged, subdivided, or arranged to form equivalent methods without departing from the teachings of this application. Therefore, unless specifically indicated herein, the order and grouping of operations do not limit the application.

Claims

1. A pump condition monitoring method based on a machine learning SVM model, characterized in that, It includes several steps: Step S1: Collect the audio signal of the training pump operation; Step S2: Preprocess the audio signal; wherein the preprocessing includes at least noise reduction, amplitude normalization and frame segmentation. Step S3: Set up a custom Mel filter bank, call the custom Mel filter bank to process each frame of signal, extract the Mel frequency cepstral coefficient features of each frame, and combine the Mel frequency cepstral coefficient features of all frames into a first feature matrix; wherein, the custom Mel filter bank refers to a triangular filter bank with non-standard Mel frequency scale and linear frequency division method that is optimized for the audio characteristics of the pump. Step S4: Standardize the extracted first feature matrix; wherein the standardization process uses Z-score standardization, Min-Max normalization or MaxAbs normalization. Step S5: Using the first feature matrix as input, train independent single-class support vector machine models according to the operating state of the training pump; wherein, the operating state includes at least one state representing the normal operation of the pump; Step S6: Use the single-class support vector machine model to determine the operating state of the pump under test, wherein when the single-class support vector machine model determines that the operating state of the pump under test is consistent with the operating state corresponding to its training, it outputs a first value; otherwise, it outputs a second value.

2. The pump condition monitoring method according to claim 1, characterized in that, The process of determining the state of the pump under test using the single-class support vector machine model includes the following steps: Step S61: Collect the audio signal of the pump under test during operation; Step S62: Perform the aforementioned preprocessing on the audio signal of the pump under test. Step S63: Set a custom Mel filter bank, call the custom Mel filter bank to process each frame of signal, extract the Mel frequency cepstral coefficient features of each frame, and combine the Mel frequency cepstral coefficient features of all frames into a second feature matrix; Step S64: Based on the second feature matrix, the operating state of the pump under test is determined by using the single-class support vector machine model corresponding to each operating state.

3. The pump condition monitoring method according to claim 2, characterized in that, The process of setting a custom Mel filter bank, calling the custom Mel filter bank to process each frame of signal, and extracting the Mel frequency cepstral coefficient features of each frame includes the following steps: Step S31: Initialize the custom Mel filter bank, set the number of filters p to 48, the number of FFT points to 1024, and the frequency coverage range to 0 to 1 / 2 of the sampling rate; Step S32: Linearly divide the frequency range from 0 to Fs / 2 into p+2 equally divided frequency points; Step S33: Calculate the frequency resolution based on the number of FFT points, and convert the frequency points into FFT frequency domain index values; Step S34: Generate the response curves of multiple filters in the form of a triangular window function to form a filter bank matrix.

4. The pump condition monitoring method according to claim 1, characterized in that, The operating state also includes one or more states representing pump malfunctions, and each operating state corresponds to an independent single-class support vector machine model.

5. The pump condition monitoring method according to claim 3, characterized in that, The first characteristic matrix is ​​a cepstral coefficient characteristic matrix of 6 to 48 Vimer frequencies.

6. The pump condition monitoring method according to claim 1, characterized in that, The pump status monitoring method also includes: Step S7: Calculate the false positive rate and false positive type, and generate a monitoring report.

7. The pump condition monitoring method according to claim 1, characterized in that, The preprocessing also includes windowing each frame of signal using a Hanning window, and then performing a fast Fourier transform on each frame of signal.

8. A pump condition monitoring device based on a machine learning SVM model, characterized in that, It includes: Training acquisition module: used to acquire audio signals from the training pump operation; Training audio preprocessing module: used to preprocess the audio signal; wherein, the preprocessing includes at least noise reduction, amplitude normalization and frame segmentation; Training Feature Matrix Generation Module: Used to set up a custom Mel filter bank, call the custom Mel filter bank to process each frame of signal, extract the Mel frequency cepstral coefficient features of each frame, and combine the Mel frequency cepstral coefficient features of all frames into a first feature matrix; wherein, the custom Mel filter bank refers to a triangular filter bank with non-standard Mel frequency scale and linear frequency division method that is optimized for the audio characteristics of pumps. Training data standardization module: used to standardize the extracted first feature matrix; wherein, the standardization process adopts Z-score standardization, Min-Max normalization or MaxAbs normalization; Training module: used to train independent single-class support vector machine models based on the first feature matrix as input and the operating state of the training pump; wherein, the operating state includes at least one state representing the normal operation of the pump, and each operating state corresponds to an independent single-class support vector machine model; Monitoring module: used to determine the operating state of the pump under test using the single-class support vector machine model; wherein, when the single-class support vector machine model determines that the operating state of the pump under test is consistent with the operating state corresponding to its training, it outputs a first value; otherwise, it outputs a second value.

9. The pump condition monitoring device according to claim 8, characterized in that, The monitoring module includes: Monitoring and acquisition submodule: used to acquire the audio signal of the pump under test during operation; Audio preprocessing submodule for monitoring: used to preprocess the audio signal of the pump under test; Monitoring data standardization submodule: used to call a custom Mel filter bank to process each frame of signal, extract the Mel frequency cepstral coefficient features of each frame, and combine the Mel frequency cepstral coefficient features of all frames into a second feature matrix; Monitoring and Judgment Submodule: Used to judge the operating status of the pump under test based on the second feature matrix and using the single-class support vector machine model corresponding to each operating state.

10. The pump condition monitoring device according to claim 9, characterized in that, The process of calling a custom Mel filter bank to process each frame of signal and extract the Mel frequency cepstral coefficient features of each frame includes calling the following modules: Initialization submodule: Initialize the custom Mel filter bank, set the number of filters p to 48, the number of FFT points to 1024, and the frequency coverage range to 0 to 1 / 2 of the sampling rate; Frequency division point submodule: linearly divides the frequency range from 0 to Fs / 2 into p+2 equal frequency points; Frequency conversion submodule: Calculates the frequency resolution based on the number of FFT points and converts the frequency points into FFT frequency domain index values; The filter matrix submodule generates the response curves of multiple filters in the form of a triangular window function, forming a filter bank matrix.

Images