A motor equipment-based abnormal data detection method and device and a terminal device

Abstract

The application is suitable for the technical field of fault diagnosis, and provides an abnormal data detection method and device based on motor equipment and terminal equipment, which comprises the following steps: obtaining initial vibration data of the motor equipment; performing time-frequency analysis on the initial vibration data of each dimension through short-time Fourier transform to construct time-frequency diagrams of three dimensions; fusing the time-frequency diagrams of three dimensions into a three-channel picture to input an improved U-Net model; the improved U-Net model encodes the three-channel picture into a first feature vector, and decodes and reconstructs the first feature vector to obtain a pseudo image; obtaining a second feature vector output by the improved U-Net model based on the pseudo image, and calculating the root mean square error of the first feature vector and the second feature vector; if the root mean square error is out of a preset range, the initial vibration data of the motor equipment is abnormal. The application can diagnose the abnormal vibration condition of the motor equipment under the condition of few negative samples or no negative sample, and fully considers the time-frequency characteristics of the vibration data.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and in particular to a method, apparatus and terminal equipment for detecting abnormal data of motor equipment. Background Technology

[0002] With the continuous development of technology, motors have been widely used in various fields. In actual operation, motors will vibrate, which may be normal or abnormal. Abnormal vibration can lead to damage to the motor itself and its auxiliary machinery, so it is necessary to detect and deal with the problem in a timely manner.

[0003] Therefore, motor fault diagnosis focuses on monitoring motor vibration, primarily by analyzing the collected vibration signals and further combining signal processing and deep learning methods to evaluate and diagnose the motor's operating status. However, deep learning requires a large amount of sample data for training and is a typical data-driven fault diagnosis method. In practical applications, it is difficult to collect a large amount of fault data for all fault types. Furthermore, current deep learning methods directly input vibration time-series data into the model for training, ignoring the time-frequency characteristics of the vibration data itself. Summary of the Invention

[0004] The main objective of this invention is to propose an abnormal data detection method, device, and terminal equipment for motor equipment, which solves the problems of traditional motor fault diagnosis methods that require the collection of a large amount of fault data for all fault types, and the direct use of vibration time series data during deep learning while ignoring the time-frequency characteristics of the vibration data itself.

[0005] To achieve the above objectives, a first aspect of the present invention provides a method for detecting abnormal data of motor equipment, comprising:

[0006] Acquire initial vibration data of the motor equipment based on spatial coordinate dimensions;

[0007] The initial vibration data for each dimension were analyzed using short-time Fourier transform to construct a three-dimensional time-frequency diagram.

[0008] The three-dimensional time-frequency graphs are fused into an improved U-Net model with a three-channel image input; the improved U-Net model is used to encode the three-channel image into a first feature vector, and to decode and reconstruct a pseudo image based on the first feature vector.

[0009] Obtain the second feature vector output by the improved U-Net model based on the pseudo image, and calculate the root mean square error between the first feature vector and the second feature vector;

[0010] If the root mean square error is outside the preset range, then the initial vibration data of the motor equipment is abnormal.

[0011] In conjunction with the first aspect of the present invention, in the first embodiment of the present invention, time-frequency analysis is performed on the initial vibration data in any dimension using short-time Fourier transform to construct a time-frequency diagram, including:

[0012] The initial vibration data is processed by short-time Fourier transform to obtain the short-time Fourier spectrum;

[0013] Obtain the energy density amplitude of the short-time Fourier spectrum;

[0014] After converting the energy density amplitude, the short-time Fourier spectrum is updated to obtain the time-frequency diagram.

[0015] In conjunction with the first embodiment of the first aspect of the present invention, in the second embodiment of the present invention, the initial vibration data is subjected to short-time Fourier transform processing to obtain the short-time Fourier spectrum, as shown in the formula:

[0016]

[0017] Where t represents time and f represents signal frequency;

[0018] The conversion formula for converting the energy density amplitude is as follows:

[0019]

[0020] Where s is the energy density amplitude and r is the data length of the initial vibration data.

[0021] In conjunction with the first aspect or the first embodiment of the first aspect of the present invention, in the third embodiment of the present invention, before fusing the three-dimensional time-frequency maps into a three-channel image input improved U-Net model, the following steps are included:

[0022] Based on the reconstruction loss and encoder loss, the generator model is trained using the training set to obtain the improved U-Net model;

[0023] The objective function of the generator model is:

[0024] loss=λ1l con +λ2l enc ;

[0025] Among them, l con For reconstruction loss, l enc λ is the encoder loss, and λ1 and λ2 are weighting parameters that adjust the influence of individual losses on the objective function.

[0026] In conjunction with the third embodiment of the first aspect of the present invention, in the fourth embodiment of the present invention, the generator model includes a U-Net model with de-jumped connections and an encoder;

[0027] The U-Net model that removes skip connections is used to encode the three-channel image into a first feature vector, and to decode and reconstruct a pseudo image based on the first feature vector;

[0028] The encoder is used to re-encode the pseudo-image into a second feature vector and calculate the root mean square error between the first feature vector and the second feature vector.

[0029] In conjunction with the fourth embodiment of the first aspect of the present invention, in the fifth embodiment of the present invention, for the part of the U-Net model that de-jumps the jump connection, the reconstruction loss is used to recover the encoded features to generate a pseudo-image, wherein the reconstruction loss represents the Euclidean distance between the historical input image and the historical generated image; the historical input image is data in the training set; and the historical generated image is a pseudo-image obtained by decoding and reconstructing the first feature vector.

[0030] The formula for calculating the reconstruction loss is:

[0031]

[0032]

[0033] Where x is the historical input image, Generate images for history. Here, μ is a constraint function used to constrain the similarity between the historical input image and the historical generated image. x and Historical input image x and historical generated image The average value, It is the variance of x. yes The variance.

[0034] In conjunction with the fourth embodiment of the first aspect of the present invention, in the sixth embodiment of the present invention, for the encoder portion, the encoder loss is used for encoding learning, wherein the encoder loss represents minimizing the Euclidean distance between the normal image encoding and the pseudo-image encoding;

[0035] The formula for calculating the encoder loss is:

[0036]

[0037] Where z represents the encoded features of the historical input image. Encoded features for historical generated images.

[0038] A second aspect of the present invention provides an abnormal data detection device based on motor equipment, comprising:

[0039] The initial vibration data acquisition module is used to acquire initial vibration data of the motor equipment based on the spatial coordinate dimension.

[0040] The time-frequency analysis module is used to perform time-frequency analysis on the initial vibration data in each dimension through short-time Fourier transform, and construct a time-frequency diagram in three dimensions.

[0041] The encoding and reconstruction module is used to fuse the three-dimensional time-frequency graphs into an improved U-Net model with a three-channel image input; the improved U-Net model is used to encode the three-channel image into a first feature vector, and to decode and reconstruct a pseudo-image based on the first feature vector.

[0042] The second encoding module is used to obtain the second feature vector output by the improved U-Net model based on the pseudo image and to calculate the root mean square error between the first feature vector and the second feature vector.

[0043] An abnormal data detection module is used to indicate that the initial vibration data of the motor equipment is abnormal when the root mean square error is outside a preset range.

[0044] A third aspect of the present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method provided in the first aspect above.

[0045] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method provided in the first aspect above.

[0046] The present invention provides a method for detecting abnormal data of motor equipment. It performs time-frequency analysis on the initial vibration data of each spatial coordinate dimension using short-time Fourier transform. Specifically, the signal is segmented along the time axis in the time domain, and spectral analysis is performed on each segment. This results in a constructed time-frequency map with time resolution, reflecting the time-frequency characteristics of the signal in the initial vibration data. The three-dimensional time-frequency maps are then fused into a three-channel image and input into an improved U-Net model. The improved U-Net model encodes, reconstructs, and re-encodes the three-channel image to obtain a first feature vector based on the three-channel image and a second feature vector based on the pseudo-image. Since both the three-channel image and the pseudo-image are essentially time-frequency maps, the root mean square error between the first and second feature vectors can be used to determine whether the initial vibration data of the motor equipment is abnormal and whether the motor equipment exhibits abnormal vibration. Therefore, the improved U-Net model in this invention visualizes the time-frequency characteristics of motor vibration data, enabling the diagnosis of abnormal vibration of motor equipment even with few or no negative samples, thus solving the problem of difficult acquisition of abnormal motor data. Attached Figure Description

[0047] Figure 1 This is a schematic diagram illustrating the implementation process of the abnormal data detection method for motor equipment provided in an embodiment of the present invention;

[0048] Figure 2 A model framework diagram of the generator provided in this embodiment of the invention;

[0049] Figure 3 This is a schematic diagram of the composition structure of an abnormal data detection device based on motor equipment provided in an embodiment of the present invention.

[0050] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0051] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0052] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0053] In this document, suffixes such as "module," "part," or "unit" used to denote elements are used only for the purpose of illustrative purposes and do not have any specific meaning in themselves. Therefore, "module" and "part" can be used interchangeably.

[0054] like Figure 1 As shown, this embodiment of the invention provides a method for detecting abnormal data of motor equipment, characterized in that it includes:

[0055] S101. Obtain initial vibration data of the motor equipment based on spatial coordinate dimensions.

[0056] In this embodiment of the invention, initial vibration data of the motor equipment is acquired through an explosion-proof wireless sensor, and then the following steps S102 to S105 are implemented through a cloud algorithm to complete the abnormal data detection of the motor equipment. The initial vibration data, based on spatial coordinate dimensions, includes data on the x-axis, y-axis, and z-axis.

[0057] S102. Perform time-frequency analysis on the initial vibration data of each dimension using short-time Fourier transform to construct a three-dimensional time-frequency diagram.

[0058] In practical applications, the spectral information obtained by traditional Fourier transform lacks time resolution and cannot reflect the time-frequency characteristics of the signal. Therefore, in this embodiment of the invention, short-time Fourier transform is used, that is, the signal is segmented in the time domain along the time axis, and spectral analysis is performed on each segment of the time domain signal, so that the constructed time-frequency diagram has time resolution and reflects the time-frequency characteristics of the signal in the initial vibration data.

[0059] In one embodiment, for initial vibration data in any dimension, the detailed implementation of step S102 is as follows:

[0060] The initial vibration data is processed by short-time Fourier transform to obtain the short-time Fourier spectrum;

[0061] Obtain the energy density amplitude of the short-time Fourier spectrum;

[0062] After converting the energy density amplitude, the short-time Fourier spectrum is updated to obtain the time-frequency diagram.

[0063] In the above steps, the embodiments of the present invention convert the initial motor vibration data into a time-frequency diagram that integrates time-frequency features and power spectral density, thereby obtaining rich feature information and improving the calculation accuracy of the improved U-Net model in the following steps.

[0064] The initial vibration data is subjected to short-time Fourier transform processing to obtain the short-time Fourier spectrum, as shown in the formula:

[0065]

[0066] Where t represents time and f represents signal frequency;

[0067] The conversion formula for converting the energy density amplitude is as follows:

[0068]

[0069] Where s is the energy density amplitude and r is the data length of the initial vibration data.

[0070] S103. An improved U-Net model that integrates three-dimensional time-frequency graphs into a three-channel image input.

[0071] S104. Obtain the second feature vector output by the improved U-Net model based on the pseudo-image encoding.

[0072] In steps S103 and S104 above, the improved U-Net model is used to encode the three-channel image into a first feature vector, and to decode and reconstruct a pseudo-image based on the first feature vector. The improved U-Net model also encodes the output second feature vector again based on the pseudo-image, and calculates the root mean square error between the first feature vector and the second feature vector.

[0073] It should be noted that the pseudo-images, as generated images from the improved U-Net model, are essentially still time-frequency images.

[0074] S105. If the root mean square error is outside the preset range, the initial vibration data of the motor equipment is abnormal.

[0075] It is conceivable that if the root mean square error is outside the preset range, then the initial vibration data of the motor equipment is normal, and the vibration of the motor equipment is normal.

[0076] This embodiment of the invention also illustrates a training method for the improved U-Net model, namely, prior to step S103 above, the method further includes:

[0077] Based on the reconstruction loss and encoder loss, the generator model is trained using the training set to obtain the improved U-Net model;

[0078] The objective function of the improved U-Net model is:

[0079] loss=λ1l con +λ2l enc ;

[0080] Among them, l con For reconstruction loss, l encλ is the encoder loss, and λ1 and λ2 are weighting parameters that adjust the influence of individual losses on the objective function.

[0081] For example, during training, λ1 = 50 and λ2 = 1.

[0082] like Figure 2 As shown in the embodiment of the present invention, the model framework of the generator model used is also illustrated. It is mainly divided into two parts. The first part is the U-Net model for de-jumping connections, which is used to encode the three-channel image into a first feature vector and to decode and reconstruct the pseudo image based on the first feature vector. The second part is the encoder, which is used to encode the pseudo image into a second feature vector again and to calculate the root mean square error between the first feature vector and the second feature vector.

[0083] The first and second parts together constitute the improved U-Net model.

[0084] In this embodiment of the invention, for the part of the U-Net model that removes skip connections, the reconstruction loss is used to recover the encoded features and generate a pseudo-image. The reconstruction loss represents the Euclidean distance between the historical input image and the historical generated image. The historical input image is data in the training set. The historical generated image is a pseudo-image obtained by decoding and reconstructing the first feature vector.

[0085] The formula for calculating the reconstruction loss is:

[0086]

[0087]

[0088] Where x is the historical input image, Generate images for history. Here, μ is a constraint function used to constrain the similarity between the historical input image and the historical generated image. x

[0089] and Historical input image x and historical generated image The average value, It is the variance of x. yes The variance.

[0090] In practical applications, relying solely on Constraints, historical generated images, i.e., pseudo-images There are still differences in blur and brightness. Therefore, in the above-mentioned calculation formula for reconstruction loss, the embodiments of the present invention introduce SSIM as a constraint to make the pseudo-images more similar.

[0091] It should be noted that the U-Net model without skip connections is essentially a U-Net model that removes skip connections. In practical applications, skip connections are used to concatenate and fuse encoded and decoded feature information. However, the presence of skip connections directly exposes information of the same dimension from the encoder to the decoder, which degrades the decoder's ability to decode features. Moreover, the U-Net model with skip connections has a strong learning ability, which can lead to overfitting. That is, the three-channel image used as input may be exactly the same as the pseudo-image used as generated. In practical applications, if there is even a slight anomaly in the input image, the output of the generative adversarial network will be judged as anomaly, thus failing to handle complex and varied real-world situations. This embodiment of the invention improves the decoding ability of the first feature vector and the performance of the final improved U-Net model by removing skip connections, avoiding the situation where the accuracy is high on the training set but low in practical applications.

[0092] For the encoder portion, encoding learning is performed using the encoder loss, which represents minimizing the normal image encoding z and the pseudo-image encoding. The Euclidean distance between them;

[0093] The formula for calculating the encoder loss is:

[0094]

[0095] Where z represents the encoding features of the historical input image and z represents the encoding features of the historical generated image.

[0096] It should be noted that the U-Net model that removes the jump connection itself has an encoder and decoder, while the encoder and decoder in the above steps is a newly introduced encoder in this embodiment of the invention. After introducing the new encoder, if the initial vibration data of the motor equipment is normal, the three-channel image before inputting the improved U-Net model and the pseudo-image obtained by decoding and reconstructing based on the first feature vector are very similar. Therefore, the second feature vector output by this encoder is very similar to the first feature vector encoded from the three-channel image. Similarly, if the initial vibration data of the motor equipment is abnormal, the three-channel image before inputting the improved U-Net model and the pseudo-image obtained by decoding and reconstructing based on the first feature vector are significantly different. The difference between the encoded first feature vector and the second feature vector will be large. Therefore, the difference between the encoded first feature vector and the second feature vector can be used as a standard to judge whether the data is normal.

[0097] like Figure 3 As shown, this embodiment of the invention also provides an abnormal data detection device 30 based on motor equipment, comprising:

[0098] The initial vibration data acquisition module 31 is used to acquire the initial vibration data of the motor equipment based on the spatial coordinate dimension;

[0099] The time-frequency analysis module 32 is used to perform time-frequency analysis on the initial vibration data of each dimension through short-time Fourier transform, and construct a time-frequency diagram of three dimensions.

[0100] The encoding and reconstruction module 33 is used to fuse the three-dimensional time-frequency graphs into a three-channel image input improved U-Net model; the improved U-Net model is used to encode the three-channel image into a first feature vector, and to decode and reconstruct a pseudo image based on the first feature vector;

[0101] The second encoding module 34 is used to obtain the second feature vector output by the improved U-Net model based on the pseudo image and to calculate the root mean square error between the first feature vector and the second feature vector.

[0102] The abnormal data detection module 35 is used to indicate that the initial vibration data of the motor equipment is abnormal when the root mean square error is outside the preset range.

[0103] This invention also provides a terminal device including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of the machine translation method described in the above embodiments.

[0104] This invention also provides a storage medium, which is a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the various steps of the machine translation method described in the above embodiments.

[0105] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the foregoing embodiments have described the present invention in detail, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for detecting abnormal data in motor equipment, characterized in that, include: Acquire initial vibration data of the motor equipment based on spatial coordinate dimensions; The initial vibration data for each dimension were analyzed using short-time Fourier transform to construct a three-dimensional time-frequency diagram. The three-dimensional time-frequency graphs are fused into an improved U-Net model with a three-channel image input; the improved U-Net model is used to encode the three-channel image into a first feature vector, and to decode and reconstruct a pseudo image based on the first feature vector. Obtain the second feature vector output by the improved U-Net model based on the pseudo image, and calculate the root mean square error between the first feature vector and the second feature vector; If the root mean square error is outside the preset range, the initial vibration data of the motor equipment is abnormal. Before fusing the three-dimensional time-frequency plots into the improved U-Net model with a three-channel image input, the following steps are included: Based on the reconstruction loss and encoder loss, the generator model is trained using the training set to obtain the improved U-Net model; The objective function of the generator model is: ； in, To reconstruct the loss, λ1 and λ2 are weighted parameters that adjust the influence of individual losses on the objective function, where λ is the encoder loss. The generator model includes a U-Net model with de-jumped connections and an encoder; The U-Net model that removes skip connections is used to encode the three-channel image into a first feature vector, and to decode and reconstruct a pseudo image based on the first feature vector; The encoder is used to encode the pseudo image again into a second feature vector, and to calculate the root mean square error between the first feature vector and the second feature vector; For the U-Net model part that removes skip connections, the reconstruction loss is used to recover the encoded features and generate a pseudo-image, where the reconstruction loss represents the Euclidean distance between the historical input image and the historical generated image; the historical input image is the data in the training set; and the historical generated image is the pseudo-image obtained by decoding and reconstructing the first feature vector. The formula for calculating the reconstruction loss is: Where x is the historical input image, Generate images for history. Here, a constraint function is used to constrain the similarity between the historical input image and the historical generated image. and Historical input images and historical generated images The average value, yes variance yes The variance.

2. The abnormal data detection method based on motor equipment as described in claim 1, characterized in that, Time-frequency analysis is performed on the initial vibration data in any dimension using short-time Fourier transform to construct a time-frequency graph, including: The initial vibration data is processed by short-time Fourier transform to obtain the short-time Fourier spectrum; Obtain the energy density amplitude of the short-time Fourier spectrum; After converting the energy density amplitude, the short-time Fourier spectrum is updated to obtain the time-frequency diagram.

3. The abnormal data detection method based on motor equipment as described in claim 2, characterized in that, The initial vibration data are processed by short-time Fourier transform to obtain the short-time Fourier spectrum, as shown in the formula: ； Where t represents time and f represents signal frequency; The conversion formula for converting the energy density amplitude is as follows: ； Where s is the energy density amplitude and r is the data length of the initial vibration data.

4. The abnormal data detection method based on motor equipment as described in claim 3, characterized in that, For the encoder portion, encoding learning is performed using the encoder loss, which represents minimizing the Euclidean distance between normal image encoding and pseudo-image encoding. The formula for calculating the encoder loss is: ； Where z represents the encoded features of the historical input image. Encoded features for historical generated images.

5. An abnormal data detection device based on motor equipment, characterized in that, include: The initial vibration data acquisition module is used to acquire initial vibration data of the motor equipment based on the spatial coordinate dimension. The time-frequency analysis module is used to perform time-frequency analysis on the initial vibration data in each dimension through short-time Fourier transform, and construct a time-frequency diagram in three dimensions. The encoding and reconstruction module is used to fuse the three-dimensional time-frequency graphs into an improved U-Net model for three-channel image input; the improved U-Net model is used to encode the three-channel image into a first feature vector, and to decode and reconstruct a pseudo-image based on the first feature vector. The second encoding module is used to obtain the second feature vector output by the improved U-Net model based on the pseudo image and to calculate the root mean square error between the first feature vector and the second feature vector. An abnormal data detection module is used to indicate that the initial vibration data of the motor equipment is abnormal when the root mean square error is outside a preset range. Before fusing the three-dimensional time-frequency plots into the improved U-Net model with a three-channel image input, the following steps are included: Based on the reconstruction loss and encoder loss, the generator model is trained using the training set to obtain the improved U-Net model; The objective function of the generator model is: ； in, To reconstruct the loss, λ1 and λ2 are weighted parameters that adjust the influence of individual losses on the objective function, where λ is the encoder loss. The generator model includes a U-Net model with de-jumped connections and an encoder; The U-Net model that removes skip connections is used to encode the three-channel image into a first feature vector, and to decode and reconstruct a pseudo image based on the first feature vector; The encoder is used to encode the pseudo image again into a second feature vector, and to calculate the root mean square error between the first feature vector and the second feature vector; For the U-Net model part that removes skip connections, the reconstruction loss is used to recover the encoded features and generate a pseudo-image, where the reconstruction loss represents the Euclidean distance between the historical input image and the historical generated image; the historical input image is the data in the training set; and the historical generated image is the pseudo-image obtained by decoding and reconstructing the first feature vector. The formula for calculating the reconstruction loss is: Where x is the historical input image, Generate images for history. Here, a constraint function is used to constrain the similarity between the historical input image and the historical generated image. and Historical input images and historical generated images The average value, yes variance yes The variance.

6. A terminal device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements each step of the abnormal data detection method based on motor equipment as described in any one of claims 1 to 4.

7. A storage medium, said storage medium being a computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements each step of the abnormal data detection method based on motor equipment as described in any one of claims 1 to 4.

Images