Fault diagnosis method and training method and device of fault diagnosis model

Abstract

The invention provides a fault diagnosis method and a training method and device of a fault diagnosis model, and is applied to the technical field of fault diagnosis and artificial intelligence. The fault diagnosis method comprises the following steps: in response to a received to-be-detected signal under a target working condition, reconstructing the to-be-detected signal to obtain N intermediate to-be-detected signals corresponding to the to-be-detected signal; converting the N intermediate to-be-detected signals to obtain respective to-be-detected images of the N intermediate to-be-detected signals, the N to-be-detected images being used for representing the to-be-detected signals under the target working condition; and inputting the N to-be-detected images into the target domain fault diagnosis model to obtain a fault diagnosis result, the fault diagnosis result comprising a fault type. The target domain fault diagnosis model is obtained by training the source domain fault diagnosis model according to a target domain sample signal under a target working condition; the source domain fault diagnosis model is obtained by training an initial fault diagnosis model according to a source domain sample signal under an initial working condition.

Description

Technical Field

[0001] The present disclosure relates to the technical fields of fault diagnosis and artificial intelligence, and more particularly to a fault diagnosis method, a training method for a fault diagnosis model, and an apparatus. Background Art

[0002] With the development of artificial intelligence technology, a fault diagnosis model can be used to diagnose faults of a device. For example, a signal to be detected in the working state of the device can be collected and input into the fault diagnosis model to obtain a fault diagnosis result. The fault diagnosis result may include a fault type.

[0003] In the process of implementing the concept of the present disclosure, the inventors found that there are at least the following problems in the related art: due to the influence of other devices or the surrounding environment when the device is working, the signal to be detected collected contains noise signals in addition to the signals characterizing the device fault features. This noise signal reduces the signal-to-noise ratio of the signal to be detected, thereby increasing the difficulty of fault diagnosis, and further affecting the efficiency and accuracy of fault diagnosis. In addition, for the signals to be detected under different working conditions of the device, the generalization ability of the fault diagnosis model is poor. Summary of the Invention

[0004] In view of the above problems, the present disclosure provides a fault diagnosis method and a training method for a fault diagnosis model.

[0005] According to a first aspect of the present disclosure, there is provided a fault diagnosis method, including: in response to receiving a signal to be detected under a target working condition, reconstructing the signal to be detected to obtain N intermediate signals to be detected corresponding to the signal to be detected, where N is a positive integer; converting the N intermediate signals to be detected to obtain respective detected images of the N intermediate signals to be detected, where the N detected images are used to characterize the signal to be detected under the target working condition; and inputting the N detected images into a target domain fault diagnosis model to obtain a fault diagnosis result, the fault diagnosis result including a fault type; where the target domain fault diagnosis model is trained based on a target domain sample signal under the target working condition for a source domain fault diagnosis model; and the source domain fault diagnosis model is trained based on a source domain sample signal under an initial working condition for an initial fault diagnosis model.

[0006] According to another aspect of the present disclosure, there is provided a method for training a fault diagnosis model, including: in response to receiving a target domain sample signal under a target working condition, reconstructing the target domain sample signal to obtain M target domain intermediate sample signals corresponding to the target domain sample signal, where each of the M target domain intermediate sample signals has first fault label information, the first fault label information includes a fault type, and M is a positive integer; converting the M target domain intermediate sample signals to obtain target domain sample images of the M target domain intermediate sample signals respectively, where the M target domain sample images are used to represent the target domain sample signal under the target working condition; and training a source domain fault diagnosis model using the M target domain sample images and the first fault label information to obtain a target domain fault diagnosis model. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Through the following description of the embodiments of the present disclosure with reference to the drawings, the above content and other objects, features, and advantages of the present disclosure will become clearer. In the drawings:

[0008] Figure 1 Schematically shows a system architecture to which a fault diagnosis method and a method for training a fault diagnosis model can be applied according to an embodiment of the present disclosure.

[0009] Figure 2 Schematically shows a flowchart of a fault diagnosis method according to an embodiment of the present disclosure.

[0010] Figure 3 Schematically shows a fault diagnosis process according to an embodiment of the present disclosure.

[0011] Figure 4 Schematically shows a flowchart of a method for training a fault diagnosis model according to an embodiment of the present disclosure.

[0012] Figure 5 Exemplarily shows a schematic structural diagram of a source domain fault diagnosis model according to an embodiment of the present disclosure.

[0013] Figure 6 Schematically shows a training process of a fault diagnosis model according to an embodiment of the present disclosure.

[0014] Figure 7 Schematically shows a structural block diagram of a fault diagnosis device according to an embodiment of the present disclosure.

[0015] Figure 8 Schematically shows a structural block diagram of a device for training a fault diagnosis model according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, for the sake of explanation, numerous specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, evidently, one or more embodiments can also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts of the present disclosure.

[0017] The terms used herein are merely for describing specific embodiments and are not intended to limit the present disclosure. The terms "including", "comprising", etc. used herein indicate the presence of the described features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0018] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0019] In the case of using expressions such as "at least one of A, B, and C, etc.", generally, it should be interpreted according to the meaning commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include, but is not limited to, a system having only A, only B, only C, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.).

[0020] With the development of artificial intelligence technology, a fault diagnosis model can be used to diagnose faults in a device. For example, a signal to be detected in the working state of the device can be collected, and the aforementioned signal to be detected can be input into the fault diagnosis model to obtain a fault diagnosis result. Among them, the fault diagnosis result can include the fault type.

[0021] According to an embodiment of the present disclosure, the device can include engineering equipment, medical equipment, household electrical appliances, automation equipment, electronic equipment, etc. The signal to be detected can include electrical signals, optical signals, electromagnetic signals, sound signals, image signals, network signals, signals obtained by using sensors, etc. (such as temperature signals, pressure signals, humidity signals, displacement signals, vibration signals, etc.).

[0022] According to an embodiment of the present disclosure, the device can include a bearing. In the case of using the fault diagnosis model to diagnose faults in the bearing, the signal to be detected can include a vibration signal.

[0023] For example, vibration signals (i.e., signals to be detected) during the operation of a bearing can be collected and input into a fault diagnosis model to obtain a fault diagnosis result. The fault diagnosis result may include the type of fault.

[0024] In the process of implementing the concept of the present disclosure, the inventors found that at least the following problems exist in the related art: Due to the influence of other devices or the surrounding environment during the operation of the bearing, the collected vibration signals contain noise signals in addition to harmonic signals and impact signals that characterize the bearing fault characteristics. This noise signal reduces the signal-to-noise ratio of the vibration signal, thereby increasing the difficulty of fault diagnosis and further affecting the efficiency and accuracy of fault diagnosis. In addition, the generalization ability of the fault diagnosis model for vibration signals under different working conditions of the bearing (such as the initial working condition and the target working condition) is poor.

[0025] To at least partially solve the technical problems existing in the related art, the present disclosure provides a fault diagnosis method and a training method for a fault diagnosis model, which can be applied to the fields of fault diagnosis and artificial intelligence technology.

[0026] Figure 1 Schematically shows the system architecture to which the fault diagnosis method and the training method for the fault diagnosis model according to the embodiments of the present disclosure can be applied. It should be noted that Figure 1 The shown is only an example of the system architecture to which the embodiments of the present disclosure can be applied, to help those skilled in the art understand the technical content of the present disclosure, but it does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0027] As Figure 1 shown, the system architecture 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 is used to provide a medium for communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0028] Users can use at least one of the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications may be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only as examples).

[0029] The first terminal device 101, the second terminal device 102, and the third terminal device 103 may be various electronic devices with a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0030] The server 105 may be a server that provides various services. For example, it may be a background management server (only an example) that supports the websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process data such as user requests received, and feedback the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0031] It should be noted that the fault diagnosis method and the training method of the fault diagnosis model provided in the embodiments of the present disclosure can generally be executed by the server 105. Correspondingly, the fault diagnosis device and the training device of the fault diagnosis model provided in the embodiments of the present disclosure can generally be set in the server 105. The fault diagnosis method and the training method of the fault diagnosis model provided in the embodiments of the present disclosure can also be executed by a server or a server cluster different from the server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or the server 105. Correspondingly, the fault diagnosis device and the training device of the fault diagnosis model provided in the embodiments of the present disclosure can also be set in a server or a server cluster different from the server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or the server 105.

[0032] Alternatively, the fault diagnosis method and the training method of the fault diagnosis model provided in the embodiments of the present disclosure can also be executed by the first terminal device 101, the second terminal device 102, or the third terminal device 103, or can also be executed by other terminal devices different from the first terminal device 101, the second terminal device 102, or the third terminal device 103. Correspondingly, the fault diagnosis device and the training device of the fault diagnosis model provided in the embodiments of the present disclosure can also be set in the first terminal device 101, the second terminal device 102, or the third terminal device 103, or set in other terminal devices different from the first terminal device 101, the second terminal device 102, or the third terminal device 103.

[0033] It should be understood that Figure 1 the numbers of terminal devices, networks, and servers in

[0034] It should be noted that the serial numbers of the respective operations in the following methods are only for representing the operations for description purposes and should not be regarded as indicating the execution order of the respective operations. Unless explicitly stated, the method does not need to be executed exactly in the order shown.

[0035] Figure 2 Schematically shown is a flowchart of a fault diagnosis method according to an embodiment of the present disclosure.

[0036] As Figure 2 shown, the fault diagnosis method of this embodiment includes operation S210 to operation S230.

[0037] In operation S210, in response to receiving a signal to be detected under a target working condition, the signal to be detected is reconstructed to obtain N intermediate signals to be detected corresponding to the signal to be detected, where N is a positive integer.

[0038] In operation S220, the N intermediate signals to be detected are converted to obtain respective images to be detected of the N intermediate signals to be detected, where the N images to be detected are used to characterize the signal to be detected under the target working condition.

[0039] In operation S230, the N images to be detected are input into a fault diagnosis model for the target domain to obtain a fault diagnosis result, where the fault diagnosis result includes a fault type; the fault diagnosis model for the target domain is obtained by training a source domain fault diagnosis model based on a target domain sample signal under the target working condition; the source domain fault diagnosis model is obtained by training an initial fault diagnosis model based on a source domain sample signal under an initial working condition.

[0040] According to an embodiment of the present disclosure, the working condition can be understood as operating conditions and environmental factors, etc. when the device is running. The working conditions of the device can include an initial working condition and a target working condition. Among them, the initial working condition can be understood as a simulated application scenario (such as a laboratory scenario), and the target working condition can be understood as an actual application scenario.

[0041] Taking only a bearing as an example, the rotational speed, load, temperature and other working states of the bearing under different working conditions may be different, resulting in possible differences in the vibration signals of the bearing under different working conditions.

[0042] According to an embodiment of the present disclosure, the signal to be detected can include an original signal obtained by using a sensor or the like when the device is operating under the target working condition. In the case of using the fault diagnosis model for the target domain to diagnose the fault of the bearing, the signal to be detected can include the vibration signal of the bearing.

[0043] Exemplarily, the vibration signal of the bearing can be collected by using a sensor, and the obtained vibration signal can be understood as the signal to be detected. Among them, in addition to the harmonic signal and impact signal, etc. that characterize the fault characteristics of the bearing, the vibration signal also includes a noise signal.

[0044] According to an embodiment of the present disclosure, for a bearing operating under a target working condition, the original signal (i.e., the signal to be detected) can be collected according to a preset sampling duration and a preset sampling frequency. Based on this, the signal to be detected can be understood as a time-domain signal, that is, the signal to be detected can include a plurality of sampling moments and the vibration signals respectively corresponding to the plurality of sampling moments. Among them, those skilled in the art can reasonably set the sampling duration and the sampling frequency according to actual needs or application scenarios, which will not be limited here.

[0045] For example, when the bearing speed is 1797 rpm, the original signal can be obtained with 10 seconds as a sampling unit at a specified sampling frequency (12 kHz). Approximately 400 signal points can be collected when the bearing rotates one circle, and the original signal can include 120,000 signal points.

[0046] According to an embodiment of the present disclosure, in response to receiving the signal to be detected under the target working condition, the signal to be detected can be reconstructed to obtain N intermediate signals to be detected corresponding to the signal to be detected. Among them, the intermediate signal to be detected can be understood as the signal to be detected with a selected signal length.

[0047] For example, the signal to be detected can be preprocessed such as smoothed and denoised to obtain the signal to be detected after removing the noise signal, so as to better characterize the fault characteristics of the equipment, which is beneficial to improving the accuracy of the fault diagnosis result. For example, the signal to be detected after removing the noise signal can be sampled to obtain N intermediate signals to be detected with a selected signal length, so as to reduce the amount of data to be processed subsequently, which is beneficial to improving the efficiency of the fault diagnosis.

[0048] According to an embodiment of the present disclosure, for each intermediate signal to be detected, the intermediate signal to be detected can include T sampling moment information and the amplitude information of each of the T sampling moment information, where T is a positive integer. The N intermediate signals to be detected can be converted to obtain the images to be detected of the N intermediate signals to be detected respectively, where the N images to be detected are used to characterize the signal to be detected under the target working condition.

[0049] According to an embodiment of the present disclosure, for each intermediate signal to be detected, by converting the T sampling moment information and the amplitude information of each of the T sampling moment information into an image to be detected, the time-series data is converted into image data, so that the time correlation of the signal to be detected can be effectively retained, and further, the limitation of directly using single time-series data for fault diagnosis can be at least partially overcome. Based on this, the image to be detected can more comprehensively characterize the signal to be detected under the target working condition, and is convenient for the target-domain fault diagnosis model to extract features, which can be beneficial to improving the accuracy of the fault diagnosis.

[0050] According to an embodiment of the present disclosure, the fault diagnosis result may include a fault type (e.g., may include normal or faulty). In the case where the fault type is characterized as faulty, the fault diagnosis result may further include a fault area (e.g., may include a fault location and a fault size).

[0051] According to an embodiment of the present disclosure, by reconstructing the signal to be detected, N intermediate signals to be detected corresponding to the signal to be detected can be obtained, so as to better characterize the fault characteristics of the device and reduce the amount of data to be processed subsequently, thereby facilitating improving the accuracy and efficiency of fault diagnosis of the device. By converting the intermediate signal to be detected into an image to be detected, the image to be detected can more comprehensively characterize the signal to be detected under the target working condition, and is convenient for the target domain fault diagnosis model to extract features, thereby facilitating improving the accuracy of fault diagnosis. In addition, by using the target domain sample detection signals under different target working conditions to train the source domain fault diagnosis model, target domain fault diagnosis models corresponding to different target working conditions can be obtained respectively, thereby improving the generalization ability and overall performance of the target fault diagnosis model, and further facilitating improving the accuracy of the fault diagnosis result obtained by the target fault diagnosis model for the signals to be detected under different working conditions of the device.

[0052] According to an embodiment of the present disclosure, the above-mentioned reconstructing the signal to be detected in response to receiving the signal to be detected under the target working condition to obtain N intermediate signals to be detected corresponding to the signal to be detected includes: in response to receiving the signal to be detected under the target working condition, performing a Fourier transform on the signal to be detected to obtain a Fourier spectrum to be detected corresponding to the signal to be detected; segmenting the Fourier spectrum to be detected according to preset parameters to obtain P continuous signal frequency bands to be detected, where P is an integer greater than or equal to 2; filtering the P continuous signal frequency bands to be detected respectively to obtain a first number of intermediate signal frequency bands to be detected; and reconstructing the first number of intermediate signal frequency bands to be detected to obtain N intermediate signals to be detected.

[0053] According to an embodiment of the present disclosure, since the signal to be detected can be understood as a time-domain signal, the time-domain signal can be converted into a frequency-domain signal through Fourier transform to respectively characterize each frequency component such as harmonic signals, impact signals, and noise signals in the signal to be detected, thereby facilitating processing such as denoising and subsequent feature extraction of the signal to be detected. Among them, the Fourier spectrum to be detected can be understood as a frequency-domain signal obtained by performing a Fourier transform on the signal to be detected, and the Fourier spectrum to be detected can be used to characterize the spectral characteristics of the signal to be detected, as well as the phase and amplitude information of each frequency component.

[0054] According to an embodiment of the present disclosure, the Fourier spectrum to be detected can be segmented according to preset parameters to obtain P consecutive signal frequency bands to be detected. Those skilled in the art can select reasonable preset parameters according to the signal characteristics of the Fourier spectrum to be detected, which are not limited herein. For example, the Fourier spectrum to be detected can be segmented based on several maximum points in the Fourier spectrum to be detected to obtain P consecutive signal frequency bands to be detected. Based on this, the frequency bands can be adaptively selected according to the signal characteristics of the Fourier spectrum to be detected, so as to at least partially overcome the mode mixing problem caused by the discontinuous time-frequency scale of the signal to be detected.

[0055] According to an embodiment of the present disclosure, a filter corresponding to the signal frequency band to be detected can be adaptively constructed based on the signal characteristics of each signal frequency band to be detected. Exemplarily, the aforementioned filter can be selected as a wavelet filter.

[0056] According to an embodiment of the present disclosure, each of the P consecutive signal frequency bands to be detected can be filtered according to the filter corresponding to each signal frequency band to be detected, so as to obtain the first number of intermediate signal frequency bands to be detected. Among them, the intermediate signal frequency band to be detected can be understood as a frequency domain signal after attenuating the frequency components corresponding to the noise signal. It can be understood that since the intermediate signal frequency band to be detected attenuates the frequency components corresponding to the noise signal and retains the frequency components corresponding to harmonic signals and impulse signals, etc., which characterize the equipment fault characteristics, the intermediate signal frequency band to be detected can better characterize the fault characteristics of the equipment, thus facilitating the improvement of the accuracy of the fault diagnosis result.

[0057] According to an embodiment of the present disclosure, by performing a Fourier transform on the signal to be detected to obtain the Fourier spectrum to be detected, so as to convert the time-domain signal into a frequency signal, it is convenient to perform noise reduction and subsequent feature extraction on the signal to be detected. By adaptively selecting frequency bands according to the signal characteristics of the Fourier spectrum to be detected and segmenting the Fourier spectrum to be detected, P consecutive signal frequency bands to be detected can be obtained, so as to at least partially overcome the mode mixing problem caused by the discontinuous time-frequency scale of the signal to be detected. By filtering each of the P consecutive signal frequency bands to be detected according to the filter corresponding to each signal frequency band to be detected, the first number of intermediate signal frequency bands to be detected can be obtained, so that the intermediate signal frequency band to be detected can better characterize the fault characteristics of the equipment. On this basis, the first number of intermediate signal frequency bands to be detected can be reconstructed to obtain N intermediate signals to be detected. Thus, the N intermediate signals to be detected can better characterize the fault characteristics of the equipment, which is conducive to improving the accuracy of fault diagnosis of the equipment. In addition, compared with the signal to be detected, the N intermediate signals to be detected can also reduce the amount of data to be processed subsequently, which is conducive to improving the efficiency of fault diagnosis.

[0058] According to an embodiment of the present disclosure, the reconstruction of the first number of intermediate signals to be detected in frequency bands to obtain N intermediate signals to be detected includes: extracting the first number of intermediate signals in frequency bands to obtain the second number of amplitude-modulated and frequency-modulated component signals to be detected; screening the second number of amplitude-modulated and frequency-modulated component signals to be detected according to a preset fault feature mapping to obtain the third number of feature component signals to be detected, where the preset fault feature mapping includes a fault type and component signal features corresponding to the fault type, and the third number of feature component signals to be detected is used to characterize the fault type; performing an addition process on the third number of feature component signals to be detected to obtain a composite signal to be detected, where the signal length of the composite signal to be detected is the same as the signal length of the signal to be detected; and performing a first sampling process on the composite signal to be detected according to a preset sampling length to obtain N intermediate signals to be detected.

[0059] According to an embodiment of the present disclosure, the second number of amplitude-modulated and frequency-modulated component signals to be detected can be used to respectively characterize component signals in different frequency ranges. The preset fault feature mapping can be used to determine component signals corresponding to device fault features. Only as an example, the preset fault feature mapping can include a fault type and component signal features corresponding to the fault type. Among them, the component signal features can include frequency.

[0060] For example, the second number of amplitude-modulated and frequency-modulated component signals to be detected can be screened based on a preset frequency threshold to obtain component signals in several specific frequency ranges. For example, the foregoing component signals in several specific frequency ranges can be matched with the preset fault feature mapping, and according to the fault type and component signal features corresponding to the fault type, the third number of feature component signals to be detected is obtained. Among them, the third number of feature component signals to be detected can be used to characterize the fault type.

[0061] According to an embodiment of the present disclosure, the third number of feature component signals to be detected that characterize the device fault features can be obtained according to the preset fault feature mapping. An addition process can be performed on the third number of feature component signals to be detected, so as to obtain a composite signal to be detected from which the signal components corresponding to the noise signals are removed. Thus, the composite signal to be detected can further better characterize the fault features of the device, which can further help improve the accuracy of the fault diagnosis result.

[0062] According to an embodiment of the present disclosure, a first sampling process can be performed on the composite signal to be detected according to a preset sampling length to obtain N intermediate signals to be detected. Those skilled in the art can set a reasonable sampling length according to actual requirements or application scenarios, etc., which is not limited herein.

[0063] For example, the preset sampling length can be selected as 4096 signal points. For example, the first sampling process can be performed on the synthetic signal to be detected based on a sliding window or a random start sampling method to obtain N intermediate signals to be detected. Among them, the signal length of each intermediate signal to be detected is 4096 signal points.

[0064] According to the embodiments of the present disclosure, through the above settings, a synthetic signal to be detected from which the signal components corresponding to the noise signal are removed can be obtained, and the first sampling process can be performed on the synthetic signal to be detected according to the preset sampling length to obtain N intermediate signals to be detected. Thus, the N intermediate signals to be detected can further better characterize the fault characteristics of the device, which can further contribute to improving the accuracy of the fault diagnosis result. In addition, compared with the signal to be detected, the N intermediate signals to be detected can also reduce the amount of data to be processed subsequently, which is beneficial to improving the efficiency of fault diagnosis.

[0065] According to the embodiments of the present disclosure, each intermediate signal to be detected includes T sampling time information and the amplitude information of each of the T sampling time information, where T is a positive integer; the conversion of the N intermediate signals to be detected to obtain the images to be detected of the N intermediate signals to be detected respectively includes: performing normalization processing on the N intermediate signals to be detected respectively to obtain N normalized signals to be detected; encoding the N normalized signals to be detected respectively to obtain T polar coordinate vectors of each of the N normalized signals to be detected, where each polar coordinate vector includes cosine information of the angle and radius information, the cosine information of the angle is used to characterize the amplitude information, and the radius information is used to characterize the sampling time information; performing dot product calculation on the cosine information of the angle and the radius information of the T polar coordinate vectors of each of the N normalized signals to be detected respectively to obtain Gram matrices of each of the N normalized signals to be detected; performing a second sampling process on the N Gram matrices according to a preset resampling method to obtain the images to be detected of the N intermediate signals to be detected respectively.

[0066] According to the embodiments of the present disclosure, normalization processing can be performed on the N intermediate signals to be detected respectively, and the N intermediate signals to be detected are respectively mapped to the amplitude interval of [-1, 1] to obtain N normalized signals to be detected. On this basis, encoding can be performed on the N normalized signals to be detected respectively to obtain T polar coordinate vectors of each of the N normalized signals to be detected. Among them, each polar coordinate vector includes cosine information of the angle and radius information, the cosine information of the angle is used to characterize the amplitude information, and the radius information is used to characterize the sampling time information.

[0067] According to an embodiment of the present disclosure, dot product calculations can be respectively performed on the cosine information and radius information of the T polar coordinate vectors of each of the N normalized signals to be detected, to obtain the Gram matrix of each of the N normalized signals to be detected. The N Gram matrices can be second sampled according to a preset resampling method to obtain the images to be detected of each of the N intermediate signals to be detected.

[0068] For example, the signal length of each intermediate signal to be detected can be selected as 4096 signal points, and thus the signal length of each normalized signal to be detected is also 4096 signal points. For each normalized signal to be detected, dot product calculations can be respectively performed on the cosine information and radius information of the T polar coordinate vectors of each normalized signal to be detected, to obtain the Gram matrix (4096*4096) of each normalized signal to be detected.

[0069] For example, the N Gram matrices can be second sampled according to resampling methods such as nearest neighbor interpolation, bilinear interpolation, cubic convolution interpolation, etc., to generate the images to be detected of each of the N intermediate signals to be detected (the size of the image to be detected can be selected as 224*224 pixels), so as to input the N images to be detected into the target domain fault diagnosis model, thereby obtaining the fault diagnosis result.

[0070] Table 1 exemplarily shows the fault diagnosis result according to an embodiment of the present disclosure.

[0071] As shown in Table 1, the fault diagnosis result can include normal and fault. In the case where the fault diagnosis result indicates a device fault, the fault diagnosis result further includes the fault area.

[0072] Number 0 1 2 3 4 Fault diagnosis result B007 fault B014 fault B021 fault IR007 fault IR014 fault Number 5 6 7 8 9 Fault diagnosis result IR021 fault OR007 fault OR014 fault OR021 fault Normal

[0073] Table 1

[0074] Exemplarily, in Table 1, the letter part of the fault diagnosis result represents the fault location, and the numerical part represents the fault size. Among them, B represents that the fault location is the bearing rolling element, IR represents that the fault location is the bearing inner ring, and OR represents that the fault location is the bearing outer ring. For example, the B007 fault represents a bearing rolling element fault with a fault diameter of 7 mils; the IR014 fault represents a bearing inner ring fault with a fault diameter of 14 mils; the OR021 fault represents a bearing outer ring fault with a fault diameter of 21 mils. Among them, 1 mil = 0.001 inch, and 1 inch = 25.4 mm.

[0075] According to an embodiment of the present disclosure, since the image to be detected is generated based on the Gram matrix of each of the N polar coordinate vectors, and since the cosine information of the angle of each polar coordinate vector is used to characterize the amplitude information and the radius information of each polar coordinate vector is used to characterize the sampling time information, the image to be detected can effectively retain the time correlation of the signal to be detected. Based on this, the image to be detected can more comprehensively characterize the signal to be detected under the target working condition, and is convenient for the target domain fault diagnosis model to extract features, thereby facilitating improving the accuracy of fault diagnosis.

[0076] Figure 3 Schematically shows a fault diagnosis process according to an embodiment of the present disclosure.

[0077] As Figure 3 shown, in response to receiving the signal 301 to be detected under the target working condition, the signal 301 to be detected can be subjected to Fourier transform to obtain the Fourier spectrum 302 to be detected corresponding to the signal 301 to be detected. The Fourier spectrum to be detected can be segmented according to preset parameters to obtain P consecutive signal frequency bands 303 to be detected, where P is an integer greater than or equal to 2. The P consecutive signal frequency bands 303 to be detected can be filtered respectively to obtain the first number of intermediate signal frequency bands 304 to be detected.

[0078] According to an embodiment of the present disclosure, the first number of intermediate signal frequency bands 304 to be detected can be extracted to obtain the second number of AM - FM component signals 305 to be detected. The second number of AM - FM component signals 305 to be detected can be screened according to a preset fault feature mapping 306 to obtain the third number of feature component signals 307 to be detected. Among them, the preset fault feature mapping 306 includes the fault type and the component signal features corresponding to the fault type, and the third number of feature component signals 307 to be detected is used to characterize the fault type.

[0079] According to an embodiment of the present disclosure, the third number of feature component signals 307 to be detected can be added to obtain the composite signal 308 to be detected. Among them, the signal length of the composite signal 308 to be detected is the same as the signal length of the signal 301 to be detected. On this basis, the composite signal 308 to be detected can be first sampled according to a preset sampling length 309 to obtain N intermediate signals 310 to be detected.

[0080] According to an embodiment of the present disclosure, each intermediate signal to be detected 310 includes T sampling time information and amplitude information of each of the T sampling time information, where T is a positive integer. The N intermediate signals to be detected 310 can be respectively normalized to obtain N normalized signals to be detected 311. The N normalized signals to be detected 311 can be respectively encoded to obtain T polar coordinate vectors 312 of each of the N normalized signals to be detected 311. Each polar coordinate vector 312 includes cosine information 312_1 and radius information 312_2, where the cosine information 312_1 is used to represent the amplitude information, and the radius information 312_2 is used to represent the sampling time information.

[0081] According to an embodiment of the present disclosure, the dot product calculation can be respectively performed on the cosine information 312_1 and the radius information 312_2 of the T polar coordinate vectors 312 of each of the N normalized signals to be detected 311 to obtain the Gram matrix 313 of each of the N normalized signals to be detected 311. The second sampling process can be performed on the N Gram matrices 313 according to a preset resampling method to obtain the images to be detected 314 of each of the N intermediate signals to be detected 310.

[0082] According to an embodiment of the present disclosure, the N images to be detected 314 can be input into the target domain fault diagnosis model 315 to obtain a fault diagnosis result 316, and the fault diagnosis result 316 includes a fault type. The target domain fault diagnosis model 315 is trained based on the target domain sample signals under the target working condition for the source domain fault diagnosis model; the source domain fault diagnosis model is trained based on the source domain sample signals under the initial working condition for the initial fault diagnosis model.

[0083] The above is only an exemplary embodiment, but is not limited thereto. Other fault diagnosis methods known in the art may also be included, as long as the accuracy and efficiency of fault diagnosis of the device can be improved.

[0084] Figure 4 Schematically shows a flowchart of a training method of a fault diagnosis model according to an embodiment of the present disclosure.

[0085] As Figure 4 shown, the training method of the fault diagnosis model includes operations S410 to S430.

[0086] In operation S410, in response to receiving the target domain sample signals under the target working condition, the target domain sample signals are reconstructed to obtain M target domain intermediate sample signals corresponding to the target domain sample signals, where each of the M target domain intermediate sample signals has first fault label information, and the first fault label information includes a fault type, and M is a positive integer.

[0087] In operation S420, the M target domain intermediate sample signals are converted to obtain the target domain sample images of the respective M target domain intermediate sample signals, where the M target domain sample images are used to characterize the target domain sample signals under the target working conditions.

[0088] In operation S430, the source domain fault diagnosis model is trained using the M target domain sample images and the first fault label information to obtain the target domain fault diagnosis model.

[0089] According to an embodiment of the present disclosure, the target domain sample signal may include the original signal of the device operating under the target working conditions obtained by using sensors and the like. In the case of fault diagnosis of a bearing using the target domain fault diagnosis model, the target domain sample signal may include the vibration signal of the bearing.

[0090] According to an embodiment of the present disclosure, in response to receiving the target domain sample signal under the target working conditions, the target domain sample signal is reconstructed to obtain M target domain intermediate sample signals corresponding to the target domain sample signal, where M is a positive integer. On this basis, the M target domain intermediate sample signals can be converted to obtain the target domain sample images of the respective M target domain intermediate sample signals.

[0091] It should be noted that in the training method of the fault diagnosis model according to the embodiment of the present disclosure, the part of reconstructing the target domain sample signal and converting the M target domain intermediate sample signals corresponds to the part of reconstructing the signal to be detected and converting the N intermediate signals to be detected in the fault diagnosis method according to the embodiment of the present disclosure. The specific description of the part of reconstructing the target domain sample signal and converting the M target domain intermediate sample signals can refer to the description of the part of reconstructing the signal to be detected and converting the N intermediate signals to be detected, which will not be elaborated here.

[0092] According to an embodiment of the present disclosure, each of the M target domain intermediate sample signals has the first fault label information, and the first fault label information includes the fault type. The M target domain sample images can be used to characterize the target domain sample signals under the target working conditions. The source domain fault diagnosis model can be trained using the M target domain sample images and the first fault label information to obtain the target domain fault diagnosis model.

[0093] According to an embodiment of the present disclosure, since the target domain fault diagnosis model is obtained by training the source domain fault diagnosis model by using M target domain sample images characterizing the target domain sample signals under the target working conditions and the first fault label information including the fault types, it is possible to train the source domain fault diagnosis model by using the target domain sample detection signals under different target working conditions to obtain the target domain fault diagnosis models corresponding to different target working conditions respectively. Thereby, the generalization ability and overall performance of the target fault diagnosis model can be effectively improved, which is conducive to improving the accuracy of the fault diagnosis results obtained by the target fault diagnosis model for the signals to be detected under different working conditions of the equipment. Moreover, by training the source domain fault diagnosis model according to the target domain sample signals under the target working conditions to obtain the target domain fault diagnosis model, the calculation time and calculation resources can be effectively saved, which is conducive to improving the training efficiency of the target domain fault diagnosis model.

[0094] According to an embodiment of the present disclosure, the above-mentioned training of the source domain fault diagnosis model by using M target domain sample images and the first fault label information to obtain the target domain fault diagnosis model includes: dividing the M target domain sample images into a target domain sample training set and a target domain sample test set according to a first preset ratio; inputting the target domain sample training set into the source domain fault diagnosis model to obtain a fourth number of first fault diagnosis results, where the first fault results include the fault types; based on a first preset loss function, determining a first loss function value according to the fourth number of first fault diagnosis results and the first fault label information; adjusting the model parameters of the source domain fault diagnosis model according to the first loss function value until a first predetermined end condition is satisfied to obtain a candidate target domain fault diagnosis model; inputting the target domain sample test set into the candidate target domain fault diagnosis model to obtain a fifth number of second fault diagnosis results, where the second fault results include the fault types; and in the case where the fifth number of second fault diagnosis results satisfy a first preset condition, determining the candidate target domain fault diagnosis model as the target domain fault diagnosis model.

[0095] According to an embodiment of the present disclosure, the target domain sample training set can be used to train the source domain fault diagnosis model, and the target domain sample test set can be used to test the trained source domain fault diagnosis model. On this basis, the target domain sample training set can be input into the source domain fault diagnosis model to obtain a fourth number of first fault diagnosis results, where the first fault results include the fault types.

[0096] Exemplarily, those skilled in the art can reasonably set the first preset ratio according to actual needs or application scenarios, etc., which is not limited herein. For example, the first preset ratio can be set such that the ratio of the target domain sample training set to the target domain sample test set is 4:1.

[0097] According to an embodiment of the present disclosure, the first preset loss function may include at least one of the following: Cross Entropy Loss, Hinge Loss, Exponential Loss, etc. For example, the model parameters of the source domain fault diagnosis model may be adjusted according to the backpropagation algorithm or the stochastic gradient descent algorithm until the first predetermined condition is satisfied, and a candidate target domain fault diagnosis model is obtained. The first predetermined condition may include at least one of the convergence of the first loss function value and the training round reaching the maximum training round.

[0098] According to an embodiment of the present disclosure, after obtaining the candidate target domain fault diagnosis model, the target domain sample test set may be input into the candidate target domain fault diagnosis model to obtain a fifth number of second fault diagnosis results, where the second fault results include fault types. The candidate target domain fault diagnosis model may be evaluated based on the fault types characterized by the second fault diagnosis results and the fault types characterized by the first fault labels, and a first evaluation result corresponding to the fifth number of second fault diagnosis results is obtained. As an example, the first evaluation result may include the accuracy, precision, recall rate, etc. of the candidate target domain fault diagnosis model, which is not limited herein.

[0099] According to an embodiment of the present disclosure, the first preset condition may include that the first evaluation result is greater than or equal to the first preset threshold. As an example, when the first evaluation result is greater than or equal to the first preset threshold, the candidate target domain fault diagnosis model may be determined as the target domain fault diagnosis model. As another example, when the first evaluation result is less than the first preset threshold, the model parameters of the source domain fault diagnosis model may be continuously adjusted until the first predetermined condition is satisfied, and an updated candidate target domain fault diagnosis model is obtained.

[0100] According to an embodiment of the present disclosure, the target domain sample test set may be used to avoid overfitting during the training process of the target domain fault diagnosis model, which is beneficial to improving the overall performance of the target domain fault diagnosis model, and further helps to improve the accuracy of using the target domain fault diagnosis model to diagnose faults in equipment.

[0101] According to an embodiment of the present disclosure, the source domain fault diagnosis model is obtained by training an initial fault diagnosis model based on source domain sample signals under an initial working condition. Training the initial fault diagnosis model based on the source domain sample signals under the initial working condition to obtain the source domain fault diagnosis model includes: constructing an initial fault diagnosis model, where the initial fault diagnosis model includes at least one first residual module, at least one first downsampling module, and a first fully connected layer; in response to receiving the source domain sample signals under the initial working condition, reconstructing the source domain sample signals to obtain S source domain intermediate sample signals corresponding to the source domain sample signals, where the S source domain intermediate sample signals each have second fault label information, and the second fault label information includes the fault type, and S is a positive integer; converting the S source domain intermediate sample signals to obtain source domain sample images of the S source domain intermediate sample signals respectively, where the S source domain sample images are used to represent the source domain sample signals under the initial working condition; and training the initial fault diagnosis model using the S source domain sample images and the second fault label information to obtain the source domain fault diagnosis model.

[0102] According to an embodiment of the present disclosure, the initial fault diagnosis model may include at least one model structure. The model structure may include at least one model sub-structure and the connection relationships between the respective model sub-structures. The model structure may be a structure obtained by connecting at least one model sub-structure based on the connection relationships between the model sub-structures. The at least one model sub-structure included in the model structure may be a structure from at least one operation layer. For example, the model structure may be a structure obtained by connecting at least one model sub-structure from at least one operation layer based on the connection relationships between the model sub-structures.

[0103] Exemplarily, the initial fault diagnosis model may include at least one first residual module, at least one first downsampling module, and a first fully connected layer.

[0104] According to an embodiment of the present disclosure, the source domain sample signals may include the original signals of the device operating under the initial working condition obtained by using sensors, etc. In the case of using the target domain fault diagnosis model to diagnose the faults of a bearing, the source domain sample signals may include the vibration signals of the bearing.

[0105] According to an embodiment of the present disclosure, in response to receiving the source domain sample signals under the initial working condition, the source domain sample signals are reconstructed to obtain S source domain intermediate sample signals corresponding to the source domain sample signals, where S is a positive integer. On this basis, the S source domain intermediate sample signals may be converted to obtain source domain sample images of the S source domain intermediate sample signals respectively.

[0106] It should be noted that in the training method of the fault diagnosis model according to the embodiments of the present disclosure, the part of reconstructing the source domain sample signal and transforming the S source domain intermediate sample signals corresponds to the part of reconstructing the signal to be detected and transforming the N intermediate signals to be detected in the fault diagnosis method according to the embodiments of the present disclosure. The specific description of the part of reconstructing the source domain sample signal and transforming the S source domain intermediate sample signals can refer to the description of the part of reconstructing the signal to be detected and transforming the N intermediate signals to be detected, and will not be elaborated here.

[0107] According to an embodiment of the present disclosure, each of the S source domain intermediate sample signals has second fault label information, and the second fault label information includes a fault type. The S source domain sample images can be used to characterize the source domain sample signals under the initial working condition. The initial fault diagnosis model can be trained by using the S source domain sample images and the second fault label information to obtain the source domain fault diagnosis model.

[0108] According to an embodiment of the present disclosure, by training the initial fault diagnosis model according to the source domain sample signals under the initial working condition to obtain the source domain fault diagnosis model, the calculation time and calculation resources can be effectively saved, which is beneficial to improving the training efficiency of the source domain fault diagnosis model.

[0109] According to an embodiment of the present disclosure, the above-mentioned training of the initial fault diagnosis model by using the S source domain sample images and the second fault label information to obtain the source domain fault diagnosis model includes: dividing the S source domain sample images into a source domain sample training set and a source domain sample test set according to a second preset ratio; inputting the source domain sample training set into the initial fault diagnosis model to obtain a sixth number of third fault diagnosis results, where the third fault diagnosis results include a fault type; based on a second preset loss function, determining a second loss function value according to the sixth number of third fault diagnosis results and the second fault label information; adjusting the model parameters of the initial fault diagnosis model according to the second loss function value until a second predetermined end condition is satisfied to obtain a candidate source domain fault diagnosis model; inputting the source domain sample test set into the candidate source domain fault diagnosis model to obtain a seventh number of fourth fault diagnosis results, where the fourth fault results include a fault type; and in the case where the seventh number of fourth fault diagnosis results satisfy a second preset condition, determining the candidate source domain fault diagnosis model as the source domain fault diagnosis model.

[0110] It should be noted that in the training method of the fault diagnosis model according to the embodiments of the present disclosure, the part of training the initial fault diagnosis model by using S source domain sample images and second fault label information corresponds to the part of training the source domain fault diagnosis model by using M target domain sample images and first fault label information in the fault diagnosis method according to the embodiments of the present disclosure. The specific description of the part of training the initial fault diagnosis model can refer to the description of the part of training the source domain fault diagnosis model in part. For the sake of brevity, the similarities between the part of training the initial fault diagnosis model and the part of training the source domain fault diagnosis model will not be elaborated. Next, the differences between the two will be described in detail.

[0111] According to the embodiments of the present disclosure, the source domain sample training set can be used to train the initial fault diagnosis model, and the source domain sample test set can be used to test the trained initial fault diagnosis model.

[0112] Exemplarily, the second preset ratio can be set such that the ratio of the source domain sample training set to the source domain sample test set is 4:1. The second predetermined condition can include at least one of the convergence of the second loss function value and the training round reaching the maximum training round.

[0113] According to the embodiments of the present disclosure, the candidate source domain fault diagnosis model can be evaluated based on the fault type characterized by the fourth fault diagnosis result and the fault type characterized by the second fault label to obtain a second evaluation result corresponding to the fifth number of second fault diagnosis results. As an example, the second evaluation result can include the accuracy rate, precision rate, recall rate, etc. of the candidate source domain fault diagnosis model, which are not limited herein.

[0114] According to the embodiments of the present disclosure, the second preset condition can include that the second evaluation result is greater than or equal to the second preset threshold. As an example, when the second evaluation result is greater than or equal to the second preset threshold, the candidate source domain fault diagnosis model can be determined as the source domain fault diagnosis model. As another example, when the second evaluation result is less than the second preset threshold, the model parameters of the initial fault diagnosis model can be continuously adjusted until the second predetermined condition is met to obtain an updated candidate source domain fault diagnosis model.

[0115] According to the embodiments of the present disclosure, the source domain sample test set can be used to avoid overfitting in the training process of the source domain fault diagnosis model, thereby facilitating the improvement of the overall performance of the source domain fault diagnosis model and the target domain fault diagnosis model, and further contributing to the improvement of the accuracy of fault diagnosis of the device by using the target domain fault diagnosis model.

[0116] According to an embodiment of the present disclosure, the source domain fault diagnosis model includes at least one second residual module, at least one second downsampling module, and a second fully connected layer. Adjust the model parameters of the source domain fault diagnosis model according to the first loss function value until the first predetermined end condition is satisfied, and obtain a candidate target domain fault diagnosis model, including: while keeping the model parameters of at least one second residual module and / or at least one second downsampling module unchanged, adjust the parameters of the second fully connected layer according to the first loss function value until the first predetermined end condition is satisfied, and obtain a candidate target domain fault diagnosis model.

[0117] Figure 5 Exemplarily shows a schematic structural diagram of a source domain fault diagnosis model according to an embodiment of the present disclosure.

[0118] As Figure 5 shown, the source domain fault diagnosis model 500 may sequentially include a convolutional layer 501 (Conv2d K2, s2), a second residual module 502 (ConvNeXt Block dim = 32), a second downsampling module 503 (Downsample), three second residual modules 502, a second downsampling module 503, a global average pooling layer 504 (Global Average Pooling), and a second fully connected layer 505 (Fully Connected). Among them, those skilled in the art can reasonably adjust the quantity and proportion of the second residual module 502 in the source domain fault diagnosis model, which is not limited herein.

[0119] According to an embodiment of the present disclosure, while keeping the model parameters of at least one second residual module 502 and / or at least one second downsampling module 503 unchanged, adjust the parameters of the second fully connected layer 505 according to the first loss function value until the first predetermined end condition is satisfied, obtain a candidate target domain fault diagnosis model, and based on this, a target domain fault diagnosis model can be obtained.

[0120] According to an embodiment of the present disclosure, through the above settings, the calculation time and calculation resources in the training process of the target domain fault diagnosis model can be effectively saved, thereby effectively improving the training efficiency of the target domain fault diagnosis model.

[0121] According to an embodiment of the present disclosure, at least one first residual module and / or at least one second residual module includes: at least one convolutional layer, at least one normalization layer, at least one activation function layer, at least one global average pooling layer, and at least one fully connected layer.

[0122] According to an embodiment of the present disclosure, the normalization layer can be selected as the Layer Normalization (LN), and the activation function can be selected as the Gaussian Error Linear Unit (GELU).

[0123] According to the embodiments of the present disclosure, compared with the Batch Normalization (BN) which normalizes the same feature of different samples, since the Layer Normalization (LN) normalizes different features of a single sample, the Layer Normalization can be not limited by the number of samples. In addition, the Layer Normalization helps to stabilize the gradient of backpropagation. Compared with the Rectified Linear Unit (ReLU), since the Gaussian Error Linear Unit (GELU) has a non-zero gradient in the negative value region, it can avoid the "ReLU Dying" problem, that is, when an abnormal input occurs, a large gradient will be generated in backpropagation, resulting in the death of neurons and the disappearance of gradients. In addition, since the GELU is smoother than the ReLU in the region close to zero, it is easier to converge during the training process, which helps to improve the training efficiency of the target domain fault diagnosis model. Thus, through the above settings, it is beneficial to improve the overall performance of the source domain fault diagnosis model and the target domain fault diagnosis model, which helps to improve the accuracy and efficiency of using the target domain fault diagnosis model to diagnose faults in equipment.

[0124] Figure 6 Schematically shows the training process of the fault diagnosis model according to the embodiments of the present disclosure.

[0125] As Figure 6 shown, in response to receiving the source domain sample signal 601 under the initial working condition, the source domain sample signal 601 can be reconstructed to obtain S source domain intermediate sample signals 602 corresponding to the source domain sample signal 601. Among them, each of the S source domain intermediate sample signals 602 has the second fault label information 603.

[0126] According to the embodiments of the present disclosure, the S source domain intermediate sample signals 602 can be transformed to obtain the source domain sample images 604 of the S source domain intermediate sample signals 602 respectively. Among them, the S source domain sample images 604 are used to represent the source domain sample signal 601 under the initial working condition.

[0127] According to an embodiment of the present disclosure, S source domain sample images 604 may be divided into a source domain sample training set 605 and a source domain sample test set 606 according to a second preset ratio. The initial fault diagnosis model 607 may be trained based on the source domain sample training set 605, the source domain sample test set 606, and the second fault label information 603 to obtain a source domain fault diagnosis model 608.

[0128] According to an embodiment of the present disclosure, in response to receiving a target domain sample signal 609 under a target working condition, the target domain sample signal 609 may be reconstructed to obtain M target domain intermediate sample signals 610 corresponding to the target domain sample signal 609. Among them, each of the M target domain intermediate sample signals 610 has first fault label information 611.

[0129] According to an embodiment of the present disclosure, the M target domain intermediate sample signals 610 may be converted to obtain target domain sample images 612 corresponding to the M target domain intermediate sample signals 610. Among them, the M target domain sample images 612 are used to represent the target domain sample signal 609 under the target working condition.

[0130] According to an embodiment of the present disclosure, the M target domain sample images 612 may be divided into a target domain sample training set 613 and a target domain sample test set 614 according to a first preset ratio. The source domain fault diagnosis model 608 may be trained based on the target domain sample training set 613, the target domain sample test set 614, and the first fault label information 611 to obtain a target domain fault diagnosis model 615.

[0131] The above are only exemplary embodiments, but not limited thereto. Other training methods of fault diagnosis models known in the art may also be included, as long as they can improve the overall performance and generalization ability of the target domain fault diagnosis model.

[0132] Based on the above fault diagnosis method, the present disclosure also provides a fault diagnosis device.

[0133] Figure 7 Schematically shows a structural block diagram of a fault diagnosis device according to an embodiment of the present disclosure. As Figure 7 shown, the fault diagnosis device may include a first reconstruction module 710, a first conversion module 720, and a fault diagnosis module 730.

[0134] The first reconstruction module 710 is configured to, in response to receiving a signal to be detected under a target working condition, reconstruct the signal to be detected to obtain N intermediate signals to be detected corresponding to the signal to be detected, where N is a positive integer.

[0135] The first conversion module 720 is configured to convert N intermediate signals to be detected, so as to obtain the images to be detected corresponding to the N intermediate signals to be detected respectively, where the N images to be detected are used to characterize the signals to be detected under the target working condition.

[0136] The fault diagnosis module 730 is configured to input the N images to be detected into the target domain fault diagnosis model to obtain a fault diagnosis result, where the fault diagnosis result includes the fault type. The target domain fault diagnosis model is obtained by training the source domain fault diagnosis model according to the target domain sample signals under the target working condition; the source domain fault diagnosis model is obtained by training the initial fault diagnosis model according to the source domain sample signals under the initial working condition.

[0137] Based on the above training method of the fault diagnosis model, the present disclosure further provides a training device for the fault diagnosis model.

[0138] Figure 8 The structural block diagram of the training device for the fault diagnosis model according to an embodiment of the present disclosure is schematically shown. As Figure 8 shown, the training device 800 for the fault diagnosis model may include a second reconstruction module 810, a second conversion module 820, and a model training module 830.

[0139] The second reconstruction module 810 is configured to, in response to receiving the target domain sample signals under the target working condition, reconstruct the target domain sample signals to obtain M target domain intermediate sample signals corresponding to the target domain sample signals, where each of the M target domain intermediate sample signals has first fault label information, and the first fault label information includes the fault type, and M is a positive integer.

[0140] The second conversion module 820 is configured to convert the M target domain intermediate sample signals to obtain the target domain sample images corresponding to the M target domain intermediate sample signals respectively, where the M target domain sample images are used to characterize the target domain sample signals under the target working condition.

[0141] The model training module 830 is configured to use the M target domain sample images and the first fault label information to train the source domain fault diagnosis model to obtain the target domain fault diagnosis model.

[0142] It should be noted that the fault diagnosis device part in the embodiments of the present disclosure corresponds to the fault diagnosis method part in the embodiments of the present disclosure. For the description of the fault diagnosis device part, please refer to the fault diagnosis method part specifically, and details are not described herein again. The training device part of the fault diagnosis model in the embodiments of the present disclosure corresponds to the training method part of the fault diagnosis model in the embodiments of the present disclosure. For the description of the training device part of the fault diagnosis model, please refer to the training method part of the fault diagnosis model specifically, and details are not described herein again.

[0143] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions noted in the blocks may occur in a different order than that noted in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, and combinations of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or by a combination of dedicated hardware and computer instructions.

[0144] Those skilled in the art will appreciate that the features recited in the various embodiments and / or claims of the present disclosure may be combined or combined in various ways, even if such combinations or combinations are not explicitly recited in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features recited in the various embodiments and / or claims of the present disclosure may be combined and combined in various ways. All such combinations and / or combinations fall within the scope of the present disclosure.

[0145] The embodiments of the present disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although the embodiments have been described separately above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of the present disclosure is defined by the appended claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art can make various substitutions and modifications, and all such substitutions and modifications should fall within the scope of the present disclosure.

Claims

1. A fault diagnosis method, comprising: In response to receiving a signal to be detected under a target working condition, reconstructing the signal to be detected to obtain N intermediate signals to be detected corresponding to the signal to be detected, where N is a positive integer; Converting the N intermediate signals to be detected to obtain images to be detected of the N intermediate signals to be detected, wherein the N images to be detected are used to characterize the signals to be detected under target working conditions; and Inputting the N to-be-detected images into a target domain fault diagnosis model to obtain a fault diagnosis result, wherein the fault diagnosis result includes a fault type; Among them, the target domain fault diagnosis model is obtained by training the source domain fault diagnosis model according to the target domain sample signal under the target working condition; the source domain fault diagnosis model is obtained by training the initial fault diagnosis model according to the source domain sample signal under the initial working condition.

2. The method according to claim 1, wherein: In response to receiving the signal to be detected under the target working condition, reconstructing the signal to be detected to obtain N intermediate signals to be detected corresponding to the signal to be detected includes: In response to receiving the signal to be detected under the target working condition, performing Fourier transform on the signal to be detected to obtain a Fourier spectrum to be detected corresponding to the signal to be detected; The Fourier spectrum to be detected is divided according to preset parameters to obtain P continuous frequency bands of the signal to be detected, where P is an integer ≥ 2; Filtering the P consecutive signal frequency bands to be detected respectively to obtain a first number of intermediate signal frequency bands to be detected; and The first number of intermediate to-be-detected signal frequency bands are reconstructed to obtain the N intermediate to-be-detected signals.

3. The method according to claim 2, wherein: The reconstructing the first number of intermediate to-be-detected signal frequency bands to obtain the N intermediate to-be-detected signals comprises: Extracting the first number of intermediate signal frequency bands to be detected to obtain a second number of amplitude modulation and frequency modulation component signals to be detected; According to a preset fault feature map, the second number of amplitude modulation and frequency modulation component signals to be detected are screened to obtain a third number of characteristic component signals to be detected, wherein the preset fault feature map includes the fault type and component signal characteristics corresponding to the fault type, and the third number of characteristic component signals to be detected are used to characterize the fault type; Performing an addition process on the third number of characteristic component signals to be detected to obtain a composite signal to be detected, wherein the signal length of the composite signal to be detected is the same as the signal length of the signal to be detected; and The first sampling process is performed on the synthetic signal to be detected according to a preset sampling length to obtain the N intermediate signals to be detected.

4. According to the method of claim 1, each of the intermediate signals to be detected includes T sampling time information and amplitude information of each of the T sampling time information, where T is a positive integer; and converting the N intermediate signals to be detected to obtain images to be detected of each of the N intermediate signals to be detected includes: Normalizing the N intermediate signals to be detected respectively to obtain N normalized signals to be detected; Encoding the N normalized signals to be detected respectively to obtain T polar coordinate vectors of the N normalized signals to be detected, wherein each polar coordinate vector includes angle cosine information and radius information, the angle cosine information is used to characterize the amplitude information, and the radius information is used to characterize the sampling time information; Performing dot product calculations on the angle cosine information and radius information of the T polar coordinate vectors of each of the N normalized signals to be detected, respectively, to obtain a Gram matrix of each of the N normalized signals to be detected; A second sampling process is performed on the N Gram matrices according to a preset resampling method to obtain the images to be detected of each of the N intermediate signals to be detected.

5. A method for training a fault diagnosis model, comprising: In response to receiving a target domain sample signal under a target operating condition, reconstructing the target domain sample signal to obtain M target domain intermediate sample signals corresponding to the target domain sample signal, wherein each of the M target domain intermediate sample signals has first fault label information, the first fault label information includes a fault type, and M is a positive integer; Converting the M target domain intermediate sample signals to obtain target domain sample images of the M target domain intermediate sample signals, wherein the M target domain sample images are used to characterize the target domain sample signals under target working conditions; as well as The M target domain sample images and the first fault label information are used to train a source domain fault diagnosis model to obtain a target domain fault diagnosis model.

6. The method according to claim 5, wherein: The using the M target domain sample images and the first fault label information to train the source domain fault diagnosis model to obtain the target domain fault diagnosis model includes: Dividing the M target domain sample images into a target domain sample training set and a target domain sample test set according to a first preset ratio; Inputting the target domain sample training set into the source domain fault diagnosis model to obtain a fourth number of first fault diagnosis results, wherein the first fault results include the fault type; Based on a first preset loss function, determining a first loss function value according to the fourth number of first fault diagnosis results and the first fault label information; Adjusting the model parameters of the source domain fault diagnosis model according to the first loss function value until a first predetermined end condition is met, thereby obtaining a candidate target domain fault diagnosis model; Inputting the target domain sample test set into the candidate target domain fault diagnosis model to obtain a fifth number of second fault diagnosis results, wherein the second fault results include the fault type; and In a case where the fifth number of second fault diagnosis results satisfy a first preset condition, the candidate target domain fault diagnosis model is determined as the target domain fault diagnosis model.

7. The method according to claim 6, wherein: The source domain fault diagnosis model is obtained by training the initial fault diagnosis model according to the source domain sample signal under the initial working condition. The initial fault diagnosis model is trained according to the source domain sample signal under the initial working condition to obtain the source domain fault diagnosis model, including: Constructing the initial fault diagnosis model, wherein the initial fault diagnosis model includes at least one first residual module, at least one first downsampling module, at least one global average pooling layer and a first fully connected layer; In response to receiving the source domain sample signal under the initial working condition, reconstructing the source domain sample signal to obtain S source domain intermediate sample signals corresponding to the source domain sample signal, wherein each of the S source domain intermediate sample signals has second fault label information, the second fault label information includes the fault type, and S is a positive integer; Converting the S source domain intermediate sample signals to obtain source domain sample images of the S source domain intermediate sample signals, wherein the S source domain sample images are used to characterize the source domain sample signals under the initial working condition; and The initial fault diagnosis model is trained using the S source domain sample images and the second fault label information to obtain the source domain fault diagnosis model.

8. The method according to claim 7, wherein: The using the S source domain sample images and the second fault label information to train the initial fault diagnosis model to obtain the source domain fault diagnosis model comprises: Dividing the S source domain sample images into a source domain sample training set and a source domain sample test set according to a second preset ratio; Inputting the source domain sample training set into the initial fault diagnosis model to obtain a sixth number of third fault diagnosis results, wherein the third fault diagnosis results include the fault type; Based on a second preset loss function, determining a second loss function value according to the sixth number of third fault diagnosis results and the second fault label information; adjusting the model parameters of the initial fault diagnosis model according to the second loss function value until the second predetermined end condition is met, thereby obtaining a candidate source domain fault diagnosis model; Inputting the source domain sample test set into the candidate source domain fault diagnosis model to obtain a seventh number of fourth fault diagnosis results, wherein the fourth fault results include the fault type; and When the seventh number of fourth fault diagnosis results meets the second preset condition, the candidate source domain fault diagnosis model is determined as the source domain fault diagnosis model.

9. The method according to claim 8, wherein: The source domain fault diagnosis model includes at least one second residual module, at least one second downsampling module and a second fully connected layer, and the model parameters of the source domain fault diagnosis model are adjusted according to the first loss function value until a first predetermined end condition is met, and the candidate target domain fault diagnosis model is obtained, including: While keeping the model parameters of the at least one second residual module and / or the at least one second downsampling module unchanged, the parameters of the second fully connected layer are adjusted according to the first loss function value until the first predetermined end condition is met, so as to obtain the candidate target domain fault diagnosis model.

10. The method according to claim 9, wherein: The at least one first residual module and / or the at least one second residual module includes: at least one convolutional layer, at least one normalization layer, at least one Gaussian error activation function layer, at least one global average pooling layer and at least one fully connected layer.

Images