Rotating machinery fault diagnosis method and device based on double-path parallel fusion and cross-validation

Abstract

This invention belongs to the field of fault diagnosis technology, specifically relating to a method and device for diagnosing rotating machinery faults using dual-path parallel fusion and cross-validation. The method collects thermal and vibration data of the rotating machinery from multiple sensors, constructs a dual-path neural network architecture combining collaborative multi-scale decomposition and a Chirplet-CNN interpretable network, and employs a decision-level cross-validation mechanism for the diagnostic results output by the dual-path network structure, ultimately outputting the rotating machinery fault diagnosis result. This invention achieves technological improvements in multiple dimensions, including fault detection capability, diagnostic accuracy, result interpretability, system reliability, response timeliness, and field applicability, effectively solving the problems faced by existing rotating machinery fault detection technologies in practical applications.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis technology, specifically relating to a method and device for diagnosing rotating machinery faults using dual-path parallel fusion and cross-verification. Background Technology

[0002] Rotating machinery, as core equipment in modern industrial production systems, is widely used in key fields such as power, petrochemicals, metallurgy, coal, and aerospace. Typical rotating machinery includes various electric motors, gas turbines, steam turbines, centrifugal compressors, blowers, pumps, and gearboxes. These devices typically operate continuously for extended periods under harsh conditions of high temperature, high pressure, and high speed, and their operating status directly affects the safety and economic efficiency of the entire production system.

[0003] Statistics show that unplanned downtime caused by rotating machinery failures accounts for 20% to 30% of total industrial production downtime. Equipment failures result in significant economic losses for enterprises annually, as well as serious threats to the personal safety of on-site workers. Therefore, real-time monitoring and fault diagnosis of rotating machinery's operating status, timely detection of potential faults, and implementation of preventative maintenance measures have become core technical means of modern industrial equipment management.

[0004] Existing detection technologies utilize sensor signals, including vibration signals, infrared thermal imaging signals, noise signals, and acoustic imaging signals, to detect fault signals in rotating machinery. Furthermore, a range of portable rotating machinery inspection devices have emerged on the market. These devices typically integrate a single type of sensor and are widely used in on-site inspections and rapid check-up scenarios due to their small size, light weight, and ease of operation.

[0005] However, these devices still have some shortcomings, such as limited information dimensions from a single sensor, a high false negative rate, a lack of edge AI reasoning capabilities in existing portable devices, and a lack of interpretable analysis mechanisms. Therefore, the industry urgently needs a rotating machinery fault detection technology solution that can integrate information from multiple sensors, achieve interpretable intelligent diagnosis, possess decision-level cross-validation capabilities, and be easy to use in the field. Summary of the Invention

[0006] In order to solve at least one of the above-mentioned technical problems in the prior art, the present invention provides a method and device for diagnosing rotating machinery faults by dual-path parallel fusion and cross-verification.

[0007] This invention employs the following technical solution: a rotating machinery fault diagnosis method based on dual-path parallel fusion and cross-validation, comprising: Collect and process the thermal and vibration data of rotating machinery during operation. Based on a collaborative multi-scale decomposition and bi-branch convolutional fusion network, a first fault diagnosis model is constructed. The thermal data of rotating machinery operation is input into the first fault diagnosis model, and the output results are used as the first fault diagnosis conclusion of rotating machinery. A second fault diagnosis model is constructed based on the Chirplet-CNN interpretable neural network. The vibration data of rotating machinery is input into the second fault diagnosis model, and the output result is used as the second fault diagnosis conclusion of rotating machinery. Spatial alignment and consistency scores are calculated for the first and second diagnostic conclusions of rotating machinery. Based on the calculated consistency score, the final diagnostic conclusion for rotating machinery is output.

[0008] Preferably, the process involves collecting and processing thermal and vibration data of the rotating machinery during operation, including: Infrared thermal images of the outer shell of rotating machinery are acquired by a thermal imager during operation, and a thermal imaging heat map is generated. The acoustic imager uses a built-in camera and microphone array to collect sound source localization data generated during the operation of rotating machinery and generate a sound source heat map. Vibration signals generated during the operation of rotating machinery are collected by vibration sensors and converted into a vibration time-frequency diagram through short-time Fourier transform. Acoustic signals generated during the operation of rotating machinery are collected by noise sensors and converted into a time-frequency waveform of acoustic patterns through short-time Fourier transform.

[0009] Preferably, a first fault diagnosis model is constructed based on a collaborative multi-scale decomposition and bi-branch convolutional fusion network. The thermal data of the rotating machinery is input into the first fault diagnosis model, and the output result serves as the first diagnostic conclusion for the rotating machinery, including: A first fault diagnosis model is constructed based on a collaborative multi-scale decomposition and dual-branch convolutional fusion network; among which... The first fault diagnosis model includes a multi-scale decomposition module and a dual-branch convolution module; The multi-scale decomposition module is used to perform multi-cooperative scale decomposition on thermal imaging heatmaps and sound source heatmaps to construct a shared multi-scale image pyramid. The low-scale levels in the shared multi-scale image pyramid use a sparse sampling strategy to suppress background noise, while the high-scale levels in the shared multi-scale image pyramid use a dense sampling strategy to retain minor fault information. The dual-branch convolution module includes a first-branch convolution unit for receiving thermal imaging heatmaps after multi-scale decomposition, and a second-branch convolution unit for receiving sound source heatmaps after multi-scale decomposition. Both the first branch convolutional unit and the second branch convolutional unit include multi-scale sub-networks corresponding to the shared multi-scale image pyramid levels, and each multi-scale sub-network is provided with a dual-channel feature fusion layer. The multi-scale fused feature maps are flattened into vectors using global average pooling to obtain the total feature vector; The total feature vector is input into the fully connected layer, and the output result is the first diagnostic conclusion of the rotating machinery. The first diagnostic conclusion of the rotating machinery includes the first fault type, the first fault confidence level, and the fault spatial location.

[0010] Preferably, a second fault diagnosis model is constructed based on the Chirplet-CNN interpretable neural network. The vibration data of the rotating machinery is input into the second fault diagnosis model, and the output result serves as the second diagnostic conclusion for the rotating machinery, including: In a three-layer convolutional neural network, the first layer kernel is replaced with Chirplet basis functions to construct a dual-channel Chirplet-CNN network as the second fault diagnosis model; the second fault diagnosis model includes the first and second channels, which are both Chirplet-CNN network models. The vibration time-frequency map in the vibration data of rotating machinery is input into the first channel of the dual-channel Chirplet-CNN interpretable neural network, which outputs a vibration feature map. The acoustic fingerprint time-frequency map in the vibration data of rotating machinery is input into the second channel of the dual-channel Chirplet-CNN interpretable neural network, and the output is an acoustic fingerprint feature map. The vibration feature map and the acoustic signature feature map are stitched together, input into a global average pooling and fully connected layer, and the output result is the second rotating machinery diagnostic conclusion; the second rotating machinery diagnostic conclusion includes the second fault type, the second fault confidence level, and the fault frequency.

[0011] Preferably, the spatial alignment and consistency score calculation for the first and second rotating machinery diagnostic conclusions includes: The fault spatial location in the first rotating machine diagnostic conclusion is converted into the physical location of the first rotating machine through a pre-calibrated perspective transformation matrix; The fault frequency in the diagnostic conclusion of the second rotating machine is compared with the preset theoretical fault frequency database, and the fault location corresponding to the fault frequency is mapped to the physical location of the second rotating machine. Determine the spatial position matching degree based on the fault spatial location and the physical location of the first rotating machinery; Determine the fault type matching degree based on the first fault type and the second fault type; Based on spatial location matching degree and fault type matching degree, a consistency score is calculated for the diagnostic conclusions of the first rotating machinery and the diagnostic conclusions of the second rotating machinery.

[0012] Preferably, spatial location matching degree The formula is expressed as: in, Indicates the physical position of the first rotating machine. Indicates the physical position of the second rotating machine. express ; The fault type matching degree The formula is expressed as: in, Indicates the first fault type. Indicates the second fault type; [·] represents Iverson brackets; The consistency score The formula is expressed as: in, and These are the weights of the spatial location matching degree and the fault type matching degree, respectively. + =1.

[0013] Preferably, based on the calculated consistency score, the final diagnostic conclusion for the rotating machinery is output, including: If the consistency score is greater than the first preset consistency score threshold, the confidence level of the first fault is greater than the preset first fault confidence level threshold, and the confidence level of the second fault is greater than the preset second fault confidence level threshold, then the fault diagnosis result of the rotating machinery is either the first fault type or the second fault type. If the consistency score is less than or equal to the second preset consistency score threshold, and simultaneously the first fault confidence level is greater than the preset first fault confidence threshold or the second fault confidence level is greater than the preset second fault confidence threshold, then the re-examination process is executed; where, The re-inspection process includes: Check the data quality of the thermal data and vibration data of the rotating machinery. If the data quality is abnormal, notify manual maintenance. After the maintenance is completed, thermal data and vibration data of the rotating machinery are collected again, and a consistency score is calculated. If the recalculated consistency score is less than or equal to the second preset consistency score threshold, the fault type of the rotating machinery fault diagnosis result will be downgraded to the second fault type, and manual review will be requested. If the confidence level of the first fault is less than or equal to the preset first fault confidence level threshold, and the confidence level of the second fault is less than or equal to the preset second fault confidence level threshold, then no rotating machinery fault diagnosis result will be output, and manual intervention will be requested.

[0014] This invention discloses a rotating machinery fault diagnosis device with dual-path parallel fusion and cross-verification, comprising: The data acquisition module is used to collect and process the thermal and vibration data of the rotating machinery during operation. The first rotating machinery diagnostic conclusion output module is used to construct a first fault diagnosis model based on a collaborative multi-scale decomposition and dual-branch convolutional fusion network. The rotating machinery operation thermodynamic data is input into the first fault diagnosis model, and the output result is used as the first rotating machinery diagnostic conclusion. The second rotating machinery diagnostic conclusion output module is used to construct a second fault diagnosis model based on the Chirplet-CNN interpretable neural network. The rotating machinery operation vibration data is input into the second fault diagnosis model, and the output result is used as the second rotating machinery diagnostic conclusion. The calculation module is used to calculate the spatial alignment and consistency score of the first and second rotating machinery diagnostic conclusions; The output module is used to output the final diagnostic conclusion for rotating machinery based on the calculated consistency score.

[0015] The present invention discloses a computer device, including an input / output unit, a memory, and a processor. The memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor performs the steps in the method described in the foregoing embodiments.

[0016] The present invention discloses a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps in the method described in the foregoing embodiments.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention provides a dual-path parallel fusion and cross-validation method for rotating machinery fault diagnosis. It collects thermal and vibration data of the rotating machinery from multiple sensors, constructs a dual-path neural network architecture combining collaborative multi-scale decomposition and a Chirplet-CNN interpretable network, and employs a decision-level cross-validation mechanism for the diagnostic results output by the dual-path network structure, ultimately outputting the rotating machinery fault diagnosis result. This invention achieves technological improvements in multiple dimensions, including fault detection capability, diagnostic accuracy, result interpretability, system reliability, response timeliness, and field applicability, effectively addressing the problems faced by existing rotating machinery fault detection technologies in practical applications. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating a rotating machinery fault diagnosis method based on dual-path parallel fusion and cross-verification provided by the present invention.

[0020] Figure 2 This is a schematic diagram of the structure of a rotating machinery fault diagnosis device with dual-path parallel fusion and cross-verification provided by the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] It should be noted that the structures, proportions, sizes, etc., shown in the accompanying drawings of this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed in the specification, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportional relationships, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should fall within the scope of the technical content disclosed in the present invention. It should be noted that in this specification, relational terms such as "first" and "second" are only used to distinguish one entity from several other entities, and do not necessarily require or imply any actual relationship or order between these entities.

[0023] This invention provides an embodiment: such as Figure 1 As shown, this invention provides a method for diagnosing rotating machinery faults using dual-path parallel fusion and cross-validation, comprising: S110: Collect and process the thermal and vibration data of rotating machinery during operation.

[0024] In this embodiment of the invention, a gearbox is used as an example to illustrate the rotating machinery. Specifically, the steps for collecting and processing the operating data of the rotating machinery include: Infrared thermal images of the outer shell of rotating machinery are collected by a thermal imager during operation, generating a thermal imaging heat map.

[0025] The thermal imager involved in this invention is an FLIRE40 infrared thermal imager with a temperature measurement range of -20°C to +650°C, an accuracy of ±2°C, a frame rate of 30Hz, and a resolution of 320×240. The thermal imager is mounted on the side of the gearbox, 0.5m to 1m away from the gearbox body, with the optical axis perpendicular to the surface being measured.

[0026] The process of generating a thermal imaging heat map is as follows: the temperature matrix is ​​obtained from the thermal imager, emissivity correction is performed (the emissivity of the gearbox cast iron shell is set to 0.85), normalized to the range of [0,255], and stored as a grayscale image.

[0027] The acoustic imager, equipped with a built-in camera and microphone array, collects sound source localization data generated during the operation of rotating machinery, and generates a sound source heat map. The acoustic imager involved in this invention features a microphone array of 64 MEMS microphones arranged in a ring, with a sampling rate of 48kHz.

[0028] The process of generating a sound source heatmap is as follows: beamforming (delay summation algorithm) is performed on the sound signals collected by the microphone array to generate a sound source intensity distribution map, which is then superimposed on the image of the visual camera to obtain a color heatmap with the same resolution as the heatmap.

[0029] Vibration signals generated during the operation of rotating machinery are collected by a vibration sensor and converted into a vibration time-frequency diagram through short-time Fourier transform. The vibration sensor involved in this invention is an IEPE accelerometer, model PCB352C33, with a sensitivity of 100mV / g, a range of ±50g, and a sampling rate of 10kHz. It is mounted on the bearing seats of the input and output shafts of the gearbox via a magnetic base.

[0030] The process of generating the vibration time-frequency diagram is as follows: The vibration signal generated during the operation of rotating machinery is collected by the vibration sensor and subjected to short-time Fourier transform (STFT) with a window length of 1024 points, an overlap rate of 50%, and a Hanning window to obtain the time-frequency diagram (time × frequency), which is then converted into a grayscale image (224 × 224).

[0031] Acoustic signals generated during the operation of rotating machinery are collected by noise sensors and converted into a time-frequency waveform of acoustic patterns through short-time Fourier transform.

[0032] The process of generating the voiceprint time-frequency map is as follows: Perform STFT on one reference microphone signal (or the enhanced signal after beamforming) in the microphone array, with the same parameters as the vibration signal, to obtain a time-frequency map of the same size.

[0033] S120: Based on the collaborative multi-scale decomposition and bi-branch convolutional fusion network, a first fault diagnosis model is constructed. The thermal data of rotating machinery operation is input into the first fault diagnosis model, and the output result is used as the first diagnostic conclusion of rotating machinery.

[0034] In an embodiment of the present invention, the first fault diagnosis model is constructed based on a collaborative multi-scale decomposition and bi-branch convolutional fusion network, including a multi-scale decomposition module and a bi-branch convolutional module.

[0035] The multi-scale decomposition module is used to perform multi-cooperative scale decomposition on thermal imaging heatmaps and sound source heatmaps to construct a shared multi-scale image pyramid. The low-scale levels in the shared multi-scale image pyramid use a sparse sampling strategy to suppress background noise, while the high-scale levels in the shared multi-scale image pyramid use a dense sampling strategy to retain minor fault information.

[0036] In this invention, the shared multi-scale image pyramid has a three-layer structure. The lowest layer is the original image; the middle layers are obtained by sparsely sampling the original image in the lowest layer, with a sampling factor of 3, meaning that a point is taken every three pixels in the original image; the highest layer is obtained by densely sampling the images in the middle layers, with a sampling factor of 1.5. Each layer contains both a thermal imaging heatmap and a sound source heatmap.

[0037] The dual-branch convolution module includes a first-branch convolution unit for receiving thermal imaging heatmaps after multi-scale decomposition, and a second-branch convolution unit for receiving sound source heatmaps after multi-scale decomposition. Both the first and second branch convolutional units include multi-scale sub-networks corresponding to the shared multi-scale image pyramid levels, and each multi-scale sub-network is equipped with a dual-channel feature fusion layer.

[0038] In this invention, the first branch convolutional unit and the second branch convolutional unit respectively process thermal imaging thermal maps and sound source thermal maps, and adopt the same network structure.

[0039] Specifically, the first branch convolutional unit includes three multi-scale sub-networks. The input data of the first multi-scale sub-network is the original image of the thermal imaging heatmap in the lowest layer of the shared multi-scale image pyramid. After convolution processing by the first multi-scale sub-network, a first feature map is obtained. The input data of the second multi-scale sub-network is the sparsely sampled image of the thermal imaging heatmap in the middle layer of the shared multi-scale image pyramid. After convolution processing by the second multi-scale sub-network, a second feature map is obtained. The input data of the third multi-scale sub-network is the image of the thermal imaging heatmap in the middle layer of the highest layer of the shared multi-scale image pyramid after dense sampling. After convolution processing by the third multi-scale sub-network, a third feature map is obtained. The second branch convolutional unit includes three multi-scale sub-networks. The input data of the fourth multi-scale sub-network is the original image of the sound source heatmap in the lowest layer of the shared multi-scale image pyramid. After convolution processing by the fourth multi-scale sub-network, the fourth feature map is obtained. The input data of the fifth multi-scale sub-network is the sparsely sampled image of the sound source heatmap in the middle layer of the shared multi-scale image pyramid. After convolution processing by the fifth multi-scale sub-network, the fifth feature map is obtained. The input data of the sixth multi-scale sub-network is the image of the sound source heatmap in the middle layer of the highest layer of the shared multi-scale image pyramid after dense sampling. After convolution processing by the sixth multi-scale sub-network, the sixth feature map is obtained.

[0040] At each scale, the output feature maps of the two branches are fused, and the resulting fused feature maps are flattened into vectors by global average pooling to obtain the total feature vector. The total feature vector is input into the fully connected layer, and the output result is the first diagnostic conclusion of the rotating machinery. The first diagnostic conclusion of the rotating machinery includes the first fault type, the first fault confidence level, and the fault spatial location.

[0041] The first type of failure includes categories such as normal, pitting, broken teeth, wear, and poor lubrication.

[0042] In embodiments of the present invention, a process of training a first fault diagnosis model is further included. The training data consists of historical thermal imaging heatmaps and historical sound source heatmaps labeled with fault types and locations, and the loss function is: in Cross-entropy classification loss, Let be the mean squared error of the location regression. The KL divergence is used to calibrate the confidence level.

[0043] The optimizer is Adam, and the learning rate is... Batch size 32, training 50 rounds.

[0044] S130: Based on the Chirplet-CNN interpretable neural network, a second fault diagnosis model is constructed. The vibration data of the rotating machinery is input into the second fault diagnosis model, and the output result is used as the second diagnostic conclusion of the rotating machinery.

[0045] In a three-layer convolutional neural network, the first layer kernel is replaced with Chirplet basis functions to construct a dual-channel Chirplet-CNN network as the second fault diagnosis model; the second fault diagnosis model includes the first and second channels, which are both Chirplet-CNN network models. The vibration time-frequency map in the vibration data of rotating machinery is input into the first channel of the dual-channel Chirplet-CNN interpretable neural network, which outputs a vibration feature map. The acoustic fingerprint time-frequency map in the vibration data of rotating machinery is input into the second channel of the dual-channel Chirplet-CNN interpretable neural network, and the output is an acoustic fingerprint feature map. The vibration feature map and the acoustic signature feature map are stitched together, input into a global average pooling and fully connected layer, and the output result is the second rotating machinery diagnostic conclusion; the second rotating machinery diagnostic conclusion includes the second fault type, the second fault confidence level, and the fault frequency.

[0046] From the center frequencies learned by the Chirplet convolutional kernels, the kernel with the strongest response to the fault category is selected as the output. Simultaneously, the attention weights of the kernels are calculated, and the final frequency value is obtained through a weighted average.

[0047] S140: Calculate the spatial alignment and consistency score for the first and second rotating machinery diagnostic conclusions.

[0048] Specifically, the spatial location of the fault in the diagnostic conclusion of the first rotating machine is converted into the physical location of the first rotating machine through a pre-calibrated perspective transformation matrix.

[0049] The fault frequency in the diagnostic conclusion of the second rotating machine is compared with the preset theoretical fault frequency database, and the fault location corresponding to the fault frequency is mapped to the physical location of the second rotating machine.

[0050] The spatial position matching degree is determined based on the fault spatial location and the physical location of the first rotating machinery.

[0051] Spatial location matching degree The formula is expressed as: in, Indicates the physical position of the first rotating machine. Indicates the physical position of the second rotating machine. express ; The fault type matching degree is determined based on the first fault type and the second fault type.

[0052] Fault type matching degree The formula is expressed as: in, Indicates the first fault type. Indicates the second fault type; [·] represents Iverson brackets; Based on spatial location matching degree and fault type matching degree, a consistency score is calculated for the diagnostic conclusions of the first rotating machinery and the diagnostic conclusions of the second rotating machinery.

[0053] Consistency score The formula is expressed as: in, and These are the weights for spatial location matching degree and fault type matching degree, respectively. + =1. In this invention, the values ​​of both are 0.6 and 0.4.

[0054] S150: Based on the calculated consistency score, output the final diagnostic conclusion for rotating machinery.

[0055] If the consistency score is greater than the first preset consistency score threshold, the first fault confidence is greater than the preset first fault confidence threshold, and the second fault confidence is greater than the preset second fault confidence threshold, then the fault diagnosis result of the rotating machinery is either the first fault type or the second fault type.

[0056] Specifically, if the consistency score S > 0.7, the confidence level of the first fault C1 > 0.6, and the confidence level of the second fault C2 > 0.6, then the fault diagnosis result of the rotating machinery is either the first fault type or the second fault type.

[0057] If the consistency score is less than or equal to the second preset consistency score threshold, and the first fault confidence level is greater than the preset first fault confidence level or the second fault confidence level is greater than the preset second fault confidence level, then the re-inspection process is executed.

[0058] Specifically, if the consistency score S≤0.3 and the confidence level of the first fault C1>0.6 or the confidence level of the second fault C2>0.6, then the re-examination process is executed.

[0059] The re-inspection process includes: Check the data quality of the thermal data and vibration data of the rotating machinery. If the data quality is abnormal, notify manual maintenance. After the maintenance is completed, thermal data and vibration data of the rotating machinery are collected again, and a consistency score is calculated. If the recalculated consistency score is less than or equal to the second preset consistency score threshold, the fault type of the rotating machinery fault diagnosis result will be downgraded to the second fault type, and manual review will be requested. If the confidence level of the first fault is less than or equal to the preset first fault confidence level threshold, and the confidence level of the second fault is less than or equal to the preset second fault confidence level threshold, then no rotating machinery fault diagnosis result will be output, and manual intervention will be requested.

[0060] Specifically, if the confidence level of the first fault C1 ≤ 0.6 and the confidence level of the second fault C2 ≤ 0.6, then no fault diagnosis result for the rotating machinery will be output, and manual intervention will be requested.

[0061] After determining the fault diagnosis conclusion for the rotating machinery, the built-in fault severity rule base is used to classify the fault diagnosis conclusion into three levels: severe, moderate, and moderate. A structured diagnostic report is generated based on the severity classification. This report must include the fault type, overall confidence level, key fault frequency, and fault occurrence event. When the severity classification is severe, an alarm email is sent to a preset email address via email service.

[0062] like Figure 2 As shown, this invention proposes a rotating machinery fault diagnosis device with dual-path parallel fusion and cross-verification, comprising: The data acquisition module 210 is used to collect and process the thermal data and vibration data of the rotating machinery during operation. The first rotating machinery diagnostic conclusion output module 220 is used to construct a first fault diagnosis model based on a collaborative multi-scale decomposition and dual-branch convolutional fusion network, input the rotating machinery operation thermodynamic data into the first fault diagnosis model, and output the result as the first rotating machinery diagnostic conclusion. The second rotating machinery diagnostic conclusion output module 230 is used to construct a second fault diagnosis model based on the Chirplet-CNN interpretable neural network, input the rotating machinery operation vibration data into the second fault diagnosis model, and output the result as the second rotating machinery diagnostic conclusion. Calculation module 240 is used to calculate the spatial alignment and consistency score of the first and second rotating machinery diagnostic conclusions; Output module 250 is used to output the final diagnostic conclusion for rotating machinery based on the calculated consistency score.

[0063] To implement the embodiments, the present invention also proposes a computer device, including an input / output unit, a memory, and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor performs the steps in the methods of the foregoing embodiments.

[0064] To implement the embodiments, the present invention also proposes a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the methods described in the foregoing embodiments.

[0065] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for rotating machinery fault diagnosis with double-path parallel fusion and cross-validation, characterized in that, include: Collect and process the thermal and vibration data of rotating machinery during operation. A first fault diagnosis model is constructed based on a collaborative multi-scale decomposition and dual-branch convolutional fusion network. The thermal data of the rotating machinery operation is input into the first fault diagnosis model, and the output result is used as the first diagnostic conclusion of the rotating machinery. A second fault diagnosis model is constructed based on the Chirplet-CNN interpretable neural network. The vibration data of the rotating machinery is input into the second fault diagnosis model, and the output result is used as the second fault diagnosis conclusion of the rotating machinery. Spatial alignment and consistency scores are calculated for the first and second diagnostic conclusions of rotating machinery. Based on the calculated consistency score, the final diagnostic conclusion for rotating machinery is output.

2. The method according to claim 1, wherein, Collect and process the thermal and vibration data of rotating machinery during operation, including: Infrared thermal images of the outer shell of rotating machinery are acquired by a thermal imager during operation, and a thermal imaging heat map is generated. The acoustic imager uses a built-in camera and microphone array to collect sound source localization data generated during the operation of rotating machinery and generate a sound source heat map. Vibration signals generated during the operation of rotating machinery are collected by vibration sensors and converted into a vibration time-frequency diagram through short-time Fourier transform. Acoustic signals generated during the operation of rotating machinery are collected by noise sensors and converted into a time-frequency waveform of acoustic patterns through short-time Fourier transform.

3. The method according to claim 2, wherein, A first fault diagnosis model is constructed based on a collaborative multi-scale decomposition and bi-branch convolutional fusion network. The thermal data of the rotating machinery is input into the first fault diagnosis model, and the output result is used as the first diagnostic conclusion for the rotating machinery, including: The first fault diagnosis model is constructed based on a collaborative multi-scale decomposition and dual-branch convolutional fusion network; wherein... The first fault diagnosis model includes a multi-scale decomposition module and a bi-branch convolution module; The multi-scale decomposition module is used to perform multi-cooperative scale decomposition on the thermal imaging heatmap and the sound source heatmap to construct a shared multi-scale image pyramid. The low-scale level of the shared multi-scale image pyramid adopts a sparse sampling strategy to suppress background noise, while the high-scale level of the shared multi-scale image pyramid adopts a dense sampling strategy to retain minor fault information. The dual-branch convolution module includes a first branch convolution unit for receiving thermal imaging heatmaps after multi-scale decomposition, and a second branch convolution unit for receiving sound source heatmaps after multi-scale decomposition. Both the first branch convolutional unit and the second branch convolutional unit include a multi-scale sub-network corresponding to the shared multi-scale image pyramid level, and each multi-scale sub-network is provided with a dual-channel feature fusion layer; The multi-scale fused feature maps are flattened into vectors using global average pooling to obtain the total feature vector; The total feature vector is input into the fully connected layer, and the output result is the first diagnostic conclusion of the rotating machinery; the first diagnostic conclusion of the rotating machinery includes the first fault type, the first fault confidence level, and the fault spatial location.

4. The method according to claim 2, wherein, A second fault diagnosis model is constructed based on the Chirplet-CNN interpretable neural network. The vibration data of the rotating machinery is input into the second fault diagnosis model, and the output result is used as the second diagnostic conclusion for the rotating machinery, including: In a three-layer convolutional neural network, the first layer convolutional kernel is replaced with Chirplet basis functions to construct a dual-channel Chirplet-CNN network as the second fault diagnosis model; wherein, the second fault diagnosis model includes a first channel and a second channel, which are both Chirplet-CNN network models; The vibration time-frequency map from the vibration data of the rotating machinery is input into the first channel of the dual-channel Chirplet-CNN interpretable neural network, which outputs a vibration feature map. The acoustic fingerprint time-frequency map in the vibration data of the rotating machinery is input into the second channel of the dual-channel Chirplet-CNN interpretable neural network, and the acoustic fingerprint feature map is output. The vibration feature map and the acoustic signature feature map are stitched together, input into a global average pooling and fully connected layer, and the output result is the second rotating machinery diagnostic conclusion; the second rotating machinery diagnostic conclusion includes a second fault type, a second fault confidence level, and a fault frequency.

5. The rotating machinery fault diagnosis method based on dual-path parallel fusion and cross-validation according to claim 1, characterized in that, Spatial alignment and consistency scoring is calculated for the first and second diagnostic conclusions of rotating machinery, including: The fault spatial location in the first rotating machine diagnostic conclusion is converted into the physical location of the first rotating machine through a pre-calibrated perspective transformation matrix; The fault frequency in the diagnostic conclusion of the second rotating machine is compared with the preset theoretical fault frequency database, and the fault location corresponding to the fault frequency is mapped to the physical location of the second rotating machine. Based on the fault location and the physical location of the first rotating machinery, the spatial location matching degree is determined; Determine the fault type matching degree based on the first fault type and the second fault type; Based on the spatial location matching degree and the fault type matching degree, a consistency score is calculated for the first rotating machinery diagnostic conclusion and the second rotating machinery diagnostic conclusion.

6. The method according to claim 5, wherein, The spatial position matching degree The formula is expressed as: wherein, represents the first rotating machine physical position, represents the second rotating machine physical position, represents ; The fault type matching degree The formula is expressed as: wherein represents a first fault type, represents a second fault type; [·] is an Iverson bracket; the consistency score is expressed by the formula: wherein, and are weights of the spatial position matching degree and the fault type matching degree, respectively, and + = 1.

7. The method according to claim 6, wherein, Based on the calculated consistency score, the final diagnostic conclusion for rotating machinery is output, including: If the consistency score is greater than the first preset consistency score threshold, the first fault confidence is greater than the preset first fault confidence threshold, and the second fault confidence is greater than the preset second fault confidence threshold, then the rotating machinery fault diagnosis result is either the first fault type or the second fault type. If the consistency score is less than or equal to a second preset consistency score threshold, and simultaneously the first fault confidence level is greater than a preset first fault confidence threshold or the second fault confidence level is greater than a preset second fault confidence threshold, then a re-examination process is executed; wherein... The re-inspection process includes: Check the data quality of the thermal data and vibration data of the rotating machinery. If the data quality is abnormal, notify manual maintenance. After the maintenance is completed, thermal data and vibration data of the rotating machinery are collected again, and a consistency score is calculated. If the recalculated consistency score is less than or equal to the second preset consistency score threshold, the fault type of the rotating machinery fault diagnosis result will be downgraded to the second fault type, and manual review will be requested. If the confidence level of the first fault is less than or equal to the preset first fault confidence level threshold, and the confidence level of the second fault is less than or equal to the preset second fault confidence level threshold, then no rotating machinery fault diagnosis result will be output, and manual intervention will be requested.

8. A rotating machinery fault diagnosis device with dual-path parallel fusion and cross-verification, characterized in that, include: The data acquisition module is used to collect and process the thermal and vibration data of the rotating machinery during operation. The first rotating machinery diagnostic conclusion output module is used to construct a first fault diagnosis model based on a collaborative multi-scale decomposition and dual-branch convolutional fusion network, input the operating thermal data of the rotating machinery into the first fault diagnosis model, and output the result as the first rotating machinery diagnostic conclusion. The second rotating machinery diagnostic conclusion output module is used to construct a second fault diagnosis model based on the Chirplet-CNN interpretable neural network, input the rotating machinery operation vibration data into the second fault diagnosis model, and output the result as the second rotating machinery diagnostic conclusion. The calculation module is used to calculate the spatial alignment and consistency score of the first and second rotating machinery diagnostic conclusions; The output module is used to output the final diagnostic conclusion for rotating machinery based on the calculated consistency score.

9. A computer device, comprising: The method includes an input / output unit, a memory, and a processor, wherein the memory stores computer-readable instructions that, when executed by the processor, cause the processor to perform the steps of the method as described in any one of claims 1 to 7.

10. A storage medium storing computer readable instructions, wherein, When the computer-readable instructions are executed by one or more processors, they cause the one or more processors to perform the steps in the method as described in any one of claims 1 to 7.

Images