A method and system for fault diagnosis of rotating machinery based on image fusion and ensemble networks

Abstract

This invention discloses a method and system for fault diagnosis of rotating machinery based on image fusion and integrated networks, belonging to the field of fault diagnosis technology. The fault diagnosis system includes: a system startup module, a data acquisition module, a data preprocessing module, a convolutional neural network training module, and an integrated neural network output module. The fault diagnosis method includes: Step 1, installing sensors, installing several vibration sensors at the rotation axis of the rotating machinery, and installing current sensors at the power supply cable of the drive equipment of the rotating machinery; Step 2, acquiring training data; Step 3, data preprocessing; Step 4, constructing and training neural network models to obtain several neural network models; Step 5, constructing an integrated network model; Step 6, running the integrated network model, inputting different types of fused time-frequency images obtained from real-time data preprocessing into the corresponding neural network model, and outputting fault diagnosis results.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and in particular to a method and system for diagnosing rotating machinery faults based on image fusion and integrated networks. Background Technology

[0002] Rotating machinery is widely used in various fields of the national economy and defense industry, such as aviation, construction, chemical industry, transportation, and processing and manufacturing. Fault detection and diagnosis of rotating machinery is a basic task to ensure safe and efficient industrial production.

[0003] Traditional intelligent fault diagnosis methods rely excessively on feature extraction of fault signals, requiring signal processing techniques, expert knowledge, and human resources. In recent years, with the development of hardware computing power and algorithms, deep learning algorithms have been increasingly widely applied in machine health monitoring and fault diagnosis. Deep learning architectures have the ability to automatically learn the complex features of input data, and have great potential to overcome the shortcomings of traditional intelligent fault diagnosis. However, simply using deep learning methods lacks interpretability and reliability. Furthermore, in real-world industrial scenarios, workloads are constantly changing, and environmental noise is unavoidable, leading to poor model robustness and difficulty in extracting and judging subtle fault signal features. Traditional filtering operations under strong noise also affect feature extraction by deep learning models, resulting in severe performance degradation of intelligent fault diagnosis methods.

[0004] The diagnostic method provided by Wang Xiuli et al. in their Chinese invention patent application "A Deep Learning-Based Method for Diagnosing Bearing Faults" (application number: CN202010962681.5) includes the following steps: collecting vibration signals of bearings with different fault types, filtering the vibration signals using wavelet denoising; performing an S-transform on the filtered signals to obtain the time-spectrum diagram of the j-th fault type; using the time-spectrum diagrams of different fault types to form a sample dataset T; using the sample dataset T and different fault types j as input to a classification model to train a bearing fault identification model; and inputting the time-spectrum diagram of the vibration signal into the bearing fault identification model for fault identification. This method only uses a CNN deep learning model to classify bearing fault features and cannot overcome the problem of severe performance degradation of deep learning models under strong noise.

[0005] The UAV bearing fault diagnosis method disclosed by Chen Kai et al. in their Chinese invention patent application "A Bearing Fault Diagnosis Method Based on CNN-Stacking" (application number: CN201910530342.7) first collects bearing signals, then preprocesses the bearing signals to extract time-domain and time-frequency domain signals; then, based on the time-domain and time-frequency domain signals, respectively, a time-domain weak classification model and a time-frequency domain weak classification model are constructed using an ensemble learning algorithm; finally, the membership probability value of the UAV bearing signal to be detected is predicted by cascading and fusing the time-domain weak classification model and time-frequency weak classification model, thereby realizing UAV bearing fault diagnosis. However, this method only predicts the membership probability value of the UAV bearing signal to be detected by cascading and fusing the time-domain weak classification model and cannot overcome the problem of severe performance degradation of deep learning models under strong noise.

[0006] Therefore, those skilled in the art are dedicated to developing a new method and system for diagnosing faults in rotating machinery, overcoming the problem of severe performance degradation of deep learning models under strong noise in existing technologies. Summary of the Invention

[0007] In view of the above-mentioned deficiencies of the prior art, the technical problem to be solved by the present invention is how to overcome the deficiencies of the prior art, such as the lack of correlation in the extraction of fault features, the low robustness of deep learning models for fault diagnosis under variable load and strong noise conditions, and the performance degradation.

[0008] To achieve the above objectives, this invention provides a method and system for diagnosing rotating machinery faults based on multi-sensor time-frequency image fusion and integrated networks, which improves the accuracy and stability of rotating machinery fault diagnosis and enhances the reliability of the diagnostic model.

[0009] This invention proposes a method for rotating machinery fault diagnosis based on multi-sensor time-frequency image fusion and an integrated network. The method measures and extracts vibration signals from the rotating machinery and current signals from the power equipment driving the machinery. One-dimensional convolution with dropout is performed on the current and vibration signals, and the convolved signals undergo various time-frequency analyses. The time-frequency images of multiple signals from the same period are fused into a three-channel time-frequency spectrum, which is then input into a convolutional neural network for joint feature extraction, enabling fault classification and prediction. Furthermore, multiple neural networks are trained using various time-frequency analysis methods, and these deep network models are combined in parallel to form an integrated network, which can then perform voting-based judgment of fault types or weighted prediction of fault size.

[0010] Specifically, the rotating machinery fault diagnosis system based on image fusion and integrated networks provided by the present invention includes:

[0011] The system startup module is responsible for starting the rotating machinery and the sensors used to collect the vibration signals of the rotating machinery and the current signals of the drive equipment, adjusting the operating parameters of the fault diagnosis system to the preset parameter values, and waiting for the fault diagnosis system to run normally.

[0012] The data acquisition module records the numerical information collected by the sensors and the current normal operation or different fault types of the rotating machinery; during the model training phase of the fault diagnosis system, the data acquisition module constructs the dataset required for neural network modeling training; during the model operation phase of the fault diagnosis system, the data acquisition module collects real-time equipment operation data for fault diagnosis.

[0013] The data preprocessing module preprocesses the collected numerical information by using a large-scale one-dimensional convolution kernel with dropout operation to convolve the original numerical information, and then performs time-frequency analysis and subsequent image fusion on the convolved numerical information using different methods to obtain several different types of fused time-frequency images.

[0014] A convolutional neural network training module is used to construct and train corresponding neural network models using several different types of fused time-frequency images. The weights and biases of the neural network models are repeatedly adjusted and trained. Training is completed when the sum of squared errors of the output layer of the neural network model is less than a specified error, and the weights and biases of the neural network model are saved. Each type of fused time-frequency image corresponds to a trained neural network model, thus obtaining several neural network models.

[0015] An integrated neural network output module is provided, which combines several neural network models in parallel. Real-time data is processed by the data preprocessing module, and the corresponding type of fused video image is input into the corresponding neural network model. An ensemble learning method is applied to fuse the output results of all neural network models to output a fault diagnosis result.

[0016] Furthermore, after performing time-frequency analysis on the numerical information using different methods in the data preprocessing module, time-frequency images of the three-channel current signal and the three-channel vibration signal are obtained. Then, image fusion is performed on the time-frequency images to obtain several different types of fused time-frequency images.

[0017] Furthermore, in the integrated neural network output module, the fault diagnosis results include, but are not limited to, rotating machinery fault location, fault type, and fault size.

[0018] Furthermore, in the system startup module, the sensor includes a vibration sensor and a current sensor;

[0019] in,

[0020] The vibration sensor is installed at the radial and axial positions of the rotating shaft of the rotating machinery;

[0021] The current sensor is installed at the power supply cable of the drive device of the rotating machinery.

[0022] Furthermore, the fault diagnosis system includes a host computer, and the data acquisition module includes a data acquisition unit. The data acquisition unit is connected to the sensor via a data transmission line, and the data acquisition unit transmits the acquired numerical information to the host computer.

[0023] This invention also provides a method for diagnosing rotating machinery faults based on image fusion and integrated networks, the method comprising the following steps:

[0024] Step 1: Install sensors. Install several vibration sensors at the rotation shaft position of the rotating machinery, and install current sensors at the power supply cable of the drive equipment of the rotating machinery.

[0025] Step 2: Collect training data. The data acquisition device transmits the numerical information collected by the sensors and the current normal operation or different fault types of the rotating machinery to the host computer for recording, and obtains standard data samples. The standard data samples include 3-channel vibration signal data and 3-channel current signal data.

[0026] Step 3: Data preprocessing. Based on the sampling frequency of the sensor, a time window is set for time-frequency analysis, and a grayscale time spectrum with time-frequency contour lines is output. The grayscale time spectrum is output for the 3-channel current signal data acquired by the current sensor, and the images are fused and synthesized into an RGB three-channel color image. The grayscale time spectrum is output for the 3-channel vibration signal data acquired by the vibration sensor, and the images are fused and synthesized into an RGB three-channel color time spectrum. As the numerical information is continuously acquired in Step 2, a fused time-frequency image is generated.

[0027] Step 4: Construct and train a neural network model. Aiming for high accuracy and structural simplicity, construct the layers of the neural network model and select parameters. Use fully connected layers as output layers and employ the softmax function to predict fault types and sizes. Set the training batch size to the same value as the number of output types. During testing, use overall statistical data instead of mini-batch data. Use the backpropagation algorithm to repeatedly adjust and train the weights and biases of the neural network model. Training is complete when the sum of squared errors of the output layers of the neural network model is less than a specified error. Save the weights and biases of the neural network model. Each type of fused time-frequency image corresponds to one trained neural network model, thus obtaining several neural network models.

[0028] Step 5: Construct an integrated network model by combining several neural network models in parallel to form the integrated network model. Use voting rules to output the classification problem and use weighted average to output the numerical prediction problem of fault size type. The weighting weight is the accuracy of the individual network.

[0029] Step 6: Run the integrated network model. The real-time data is processed by the data preprocessing in Step 3 to obtain different types of fused time-frequency images, which are then input into the corresponding neural network model. The ensemble learning method is applied to fuse the outputs of all neural network models, output the fault diagnosis result, and send the fault diagnosis result to the human-machine interface for device response.

[0030] Furthermore, in step 1, the vibration sensor includes three sensors, of which two are installed radially on the rotating shaft of the rotating machinery, one is installed axially on the rotating shaft of the rotating machinery, and the installation directions of the three vibration sensors are perpendicular to each other.

[0031] Furthermore, in step 2, the standard data sample also includes the rotating machinery condition, rotating machinery fault type, and rotating machinery fault size.

[0032] Furthermore, the time window is selected as a time window of 1 to 2 seconds.

[0033] Furthermore, the time-frequency analysis methods include, but are not limited to, continuous wavelet transform, short-time Fourier transform, Hilbert-Huang transform, Fourier compressed synchronous transform, and Wigner-Ville distribution.

[0034] The rotating machinery fault diagnosis system and method based on image fusion and integrated networks provided by this invention have at least the following technical advantages:

[0035] 1. The rotating machinery fault diagnosis system and method provided by this invention features streamlined and automated processing. The current signal of the drive equipment and the vibration signal of the rotating machinery are extracted using time-frequency features and used as input data for a deep network. Multi-sensor time-frequency image fusion is achieved through convolutional neural network to automatically extract features, discover the correlation of fault features, and have higher understanding and accuracy in judging the type of subtle fault signals. This gives the fault diagnosis model better predictive performance, detects hidden faults in advance, and improves the generalization of the model through large-scale one-dimensional convolutional preprocessing training with dropout operation. This makes the model decision robust in the actual noise environment of the factory. The integrated learning method combines neural networks in parallel to improve the accuracy and reliability of the model.

[0036] 2. The rotating machinery fault diagnosis system provided by this invention can effectively replace the shortcomings of single feature extraction methods or neural network detection methods, such as low automation, low reliability, and poor robustness. It improves the accuracy of fault diagnosis, speeds up the fault diagnosis process, and enhances the reliability of fault diagnosis. It can detect and determine abnormal working conditions and potential faults early, and is more suitable for actual industrial sites with high noise and complex working conditions.

[0037] The following will further explain the concept, specific structure, and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the installation positions of the data acquisition module and sensor provided in a preferred embodiment of the present invention;

[0039] Figure 2 This is a system module information transmission structure diagram provided in a preferred embodiment of the present invention;

[0040] Figure 3 This is a flowchart of the fault diagnosis system structure provided in a preferred embodiment of the present invention;

[0041] Figure 4 A schematic diagram of a time-frequency image construction method provided in a preferred embodiment of the present invention;

[0042] Figure 5 This is a schematic diagram of an integrated neural network structure provided for a preferred embodiment of the present invention. Detailed Implementation

[0043] The following description, with reference to the accompanying drawings, illustrates several preferred embodiments of the present invention to make its technical content clearer and easier to understand. The present invention can be embodied in many different forms, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0044] This invention overcomes the shortcomings of existing technologies, such as lack of correlation in fault feature extraction, low robustness of deep learning models for fault diagnosis under variable load and high noise conditions, and performance degradation. It provides a method and system for rotating machinery fault diagnosis based on multi-sensor time-frequency image fusion and an integrated network, improving the accuracy and stability of rotating machinery fault diagnosis and enhancing the reliability of the diagnostic model. Specifically, the technical solution proposed in this invention introduces the concept of multi-sensor feature signal time-frequency image fusion. A convolutional neural network automatically extracts features from the fused time-frequency images, uncovering the correlation between fault features and solving the problem of difficulty in extracting and judging subtle fault signal feature types generated by single signals. Furthermore, an integrated network method is introduced to fuse and judge the diagnostic results of multiple fused time-frequency images, achieving more accurate and reliable fault diagnosis results. This invention effectively addresses the problem of difficulty in accurately extracting and judging fault signal features in actual industrial scenarios, proposing an accurate, reliable, and more suitable method for rotating machinery fault diagnosis in industrial settings.

[0045] The present invention provides a rotating machinery fault diagnosis system based on image fusion and integrated networks, including, as follows: Figure 2 As shown:

[0046] The system startup module is responsible for starting the rotating machinery and the sensors used to collect vibration signals from the rotating machinery and current signals from the drive equipment. It adjusts the operating parameters of the fault diagnosis system to the preset parameter values ​​and waits for the fault diagnosis system to run normally.

[0047] The data acquisition module records the numerical information collected by the sensors and the current normal operation or different fault types of the rotating machinery. During the model training phase of the fault diagnosis system, the data acquisition module constructs the dataset required for neural network modeling training. During the model operation phase of the fault diagnosis system, the data acquisition module collects real-time equipment operation data for fault diagnosis.

[0048] The data preprocessing module preprocesses the collected numerical information by using a large-scale one-dimensional convolution kernel with dropout to convolve the original numerical information, and then performs time-frequency analysis and subsequent image fusion on the convolved numerical information using different methods to obtain several different types of fused time-frequency images.

[0049] The convolutional neural network training module constructs and trains corresponding neural network models using several different types of fused time-frequency images. It repeatedly adjusts and trains the weights and biases of the neural network models. Training is complete when the sum of squared errors of the output layer of the neural network model is less than the specified error, and the weights and biases of the neural network model are saved. Each type of fused time-frequency image corresponds to a trained neural network model, thus obtaining several neural network models.

[0050] The integrated neural network output module combines several neural network models in parallel. Real-time data is processed by the data preprocessing module, and the corresponding type of fused video images are input into the corresponding neural network model. The ensemble learning method is applied to fuse the output results of all neural network models and output the fault diagnosis result.

[0051] In the data preprocessing module, time-frequency images of three-channel current signals and three-channel vibration signals are obtained after performing time-frequency analysis on the numerical information using different methods. Then, image fusion is performed on the time-frequency images to obtain several different types of fused time-frequency images.

[0052] Among them, the fault diagnosis results in the integrated neural network output module include, but are not limited to, the location of rotating machinery faults, fault type, and fault size.

[0053] The system startup module includes a vibration sensor and a current sensor.

[0054] in,

[0055] Vibration sensors are installed at the radial and axial positions of the rotating shaft of the rotating machinery;

[0056] The current sensor is installed at the power supply cable of the drive equipment of the rotating machinery.

[0057] The fault diagnosis system includes a host computer, and the data acquisition module includes a data acquisition unit. The data acquisition unit is connected to the sensor through a data transmission line, and the data acquisition unit transmits the collected numerical information to the host computer.

[0058] This invention also provides a method for diagnosing rotating machinery faults based on image fusion and integrated networks, the method comprising the following steps:

[0059] Step 1: Install the sensor, such as Figure 1 As shown, several vibration sensors are installed at the rotation axis of the rotating machinery, and a current sensor is installed at the power supply cable of the drive equipment of the rotating machinery. The vibration sensors include three sensors, of which two vibration sensors are installed at the radial position of the rotation axis of the rotating machinery, and one vibration sensor is installed at the axial position of the rotation axis of the rotating machinery. The installation directions of the three vibration sensors are perpendicular to each other.

[0060] Step 2: Collect training data. The data acquisition device transmits the numerical information collected by the sensors and the current normal operation or different fault types of the rotating machinery to the host computer for recording, and obtains standard data samples. The standard data samples include 3-channel vibration signal data, 3-channel current signal data, rotating machinery status, rotating machinery fault type, and rotating machinery fault size.

[0061] Step 3, data preprocessing, such as... Figure 4 As shown, based on the sensor's sampling frequency, a time window is set for time-frequency analysis, outputting a grayscale time spectrum filled with time-frequency contour lines. One-dimensional convolution with dropout is performed on the current signal and vibration signal. Specifically, the grayscale time spectrum of the 3-channel current signal data acquired by the current sensor is output and then fused into an RGB three-channel color image. Similarly, the grayscale time spectrum of the 3-channel vibration signal data acquired by the vibration sensor is output and then fused into an RGB three-channel color time spectrum. As numerical information is continuously acquired in step 2, a fused time-frequency image is generated.

[0062] The time window is selected as a time window of 1 to 2 seconds.

[0063] The time-frequency analysis methods include, but are not limited to, continuous wavelet transform, short-time Fourier transform, Hilbert-Huang transform, Fourier compressed synchronous transform, and Wigner-Ville distribution.

[0064] Step 4: Build and train the neural network model, such as Figure 3 As shown, with the goals of high accuracy and simple structure, the hierarchical structure of the neural network model is constructed and parameters are selected. The fully connected layer is used as the output layer, and the softmax function is used to predict the fault type and size. The training batch size is set to the same value as the number of output types. During testing, the overall statistical data is used instead of mini-batch. The backpropagation algorithm is used to repeatedly adjust the weights and biases of the neural network model to make the output vector as close as possible to the expected vector. The training is completed when the sum of squared errors of the output layer of the neural network model is less than the specified error. The weights and biases of the neural network model are saved. Each type of fused time-frequency image corresponds to a trained neural network model, that is, several neural network models are obtained, specifically: a neural network trained based on continuous wavelet transform fused time-frequency image, a neural network trained based on short-time Fourier transform fused time-frequency image, a neural network trained based on Fourier compressed synchronous transform fused time-frequency image, and neural networks trained based on various time-frequency transform methods for image fusion, etc.

[0065] Step 5: Construct an ensemble network model, such as Figure 5 As shown, several neural network models are combined in parallel to construct an ensemble network model. Voting rules are used to output the classification problem, and weighted average is used to output the numerical prediction problem of fault size type. The weighting weight is the accuracy of the individual network.

[0066] Step 6: Run the integrated network model. The real-time data, after being preprocessed in Step 3, is used to obtain different types of fused time-frequency images. These images are then input into the corresponding neural network models. An ensemble learning method is applied to fuse the outputs of all neural network models, outputting the fault diagnosis results. The fault diagnosis results are then sent to the human-machine interface and the device responds.

[0067] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A rotating machinery fault diagnosis system based on image fusion and integrated networks, characterized in that, include: The system startup module is responsible for starting the rotating machinery and the sensors used to collect the vibration signals of the rotating machinery and the current signals of the drive equipment, adjusting the operating parameters of the fault diagnosis system to the preset parameter values, and waiting for the fault diagnosis system to run normally. The data acquisition module records the numerical information collected by the sensor and the current normal operation or different fault types of the rotating machinery. During the model training phase of the fault diagnosis system, the data acquisition module constructs the dataset required for neural network modeling and training. During the model operation phase of the fault diagnosis system, the data acquisition module collects real-time equipment operation data for fault diagnosis. The data preprocessing module preprocesses the collected numerical information by using a large-scale one-dimensional convolution kernel with dropout operation to convolve the original numerical information, and then performs time-frequency analysis and subsequent image fusion on the convolved numerical information using different methods to obtain several different types of fused time-frequency images. A convolutional neural network training module is used to construct and train corresponding neural network models using several different types of fused time-frequency images. The weights and biases of the neural network models are repeatedly adjusted and trained. Training is completed when the sum of squared errors of the output layer of the neural network model is less than a specified error, and the weights and biases of the neural network model are saved. Each type of fused time-frequency image corresponds to a trained neural network model, thus obtaining several neural network models. An integrated neural network output module is provided, which combines several neural network models in parallel. Real-time data is processed by the data preprocessing module, and the corresponding type of fused time-frequency image is input into the corresponding neural network model. An ensemble learning method is applied to fuse the output results of all neural network models to output a fault diagnosis result.

2. The rotating machinery fault diagnosis system based on image fusion and integrated networks as described in claim 1, characterized in that, After performing time-frequency analysis on the numerical information using different methods in the data preprocessing module, time-frequency images of three-channel current signals and three-channel vibration signals are obtained. Then, image fusion is performed on the time-frequency images to obtain several different types of fused time-frequency images.

3. The rotating machinery fault diagnosis system based on image fusion and integrated networks as described in claim 1, characterized in that, In the integrated neural network output module, the fault diagnosis results include, but are not limited to, rotating machinery fault location, fault type, and fault size.

4. The rotating machinery fault diagnosis system based on image fusion and integrated network as described in claim 1, characterized in that, In the system startup module, the sensors include a vibration sensor and a current sensor; in, The vibration sensor is installed at the radial and axial positions of the rotating shaft of the rotating machinery; The current sensor is installed at the power supply cable of the drive device of the rotating machinery.

5. The rotating machinery fault diagnosis system based on image fusion and integrated networks as described in claim 1, characterized in that, The fault diagnosis system includes a host computer, and the data acquisition module includes a data acquisition device. The data acquisition device is connected to the sensor through a data transmission line, and the data acquisition device transmits the acquired numerical information to the host computer.

6. A method for fault diagnosis of rotating machinery based on image fusion and integrated networks, characterized in that, The method includes the following steps: Step 1: Install sensors. Install several vibration sensors at the rotation shaft position of the rotating machinery, and install current sensors at the power supply cable of the drive equipment of the rotating machinery. Step 2: Collect training data. The data acquisition device transmits the numerical information collected by the sensors and the current normal operation or different fault types of the rotating machinery to the host computer for recording, and obtains standard data samples. The standard data samples include 3-channel vibration signal data and 3-channel current signal data. Step 3: Data preprocessing. Based on the sampling frequency of the sensor, a time window is set for time-frequency analysis, and a grayscale time spectrum with time-frequency contour lines is output. The grayscale time spectrum is output for the 3-channel current signal data acquired by the current sensor, and the images are fused and synthesized into an RGB three-channel color image. The grayscale time spectrum is output for the 3-channel vibration signal data acquired by the vibration sensor, and the images are fused and synthesized into an RGB three-channel color time spectrum. As the numerical information is continuously acquired in Step 2, a fused time-frequency image is generated. Step 4: Construct and train a neural network model. Aiming for high accuracy and structural simplicity, construct the layers of the neural network model and select parameters. Use fully connected layers as output layers and employ the softmax function to predict fault types and sizes. Set the training batch size to the same value as the number of output types. During testing, use overall statistical data instead of mini-batch data. Use the backpropagation algorithm to repeatedly adjust and train the weights and biases of the neural network model. Training is complete when the sum of squared errors of the output layers of the neural network model is less than a specified error. Save the weights and biases of the neural network model. Each type of fused time-frequency image corresponds to one trained neural network model, thus obtaining several neural network models. Step 5: Construct an integrated network model by combining several neural network models in parallel to form the integrated network model. Use voting rules to output the classification problem and use weighted average to output the numerical prediction problem of fault size type. The weighting weight is the accuracy of the individual network. Step 6: Run the integrated network model. The real-time data is processed by the data preprocessing in Step 3 to obtain different types of fused time-frequency images, which are then input into the corresponding neural network model. The ensemble learning method is applied to fuse the outputs of all neural network models, output the fault diagnosis result, and send the fault diagnosis result to the human-machine interface for device response.

7. The rotating machinery fault diagnosis method based on image fusion and integrated networks as described in claim 6, characterized in that, In step 1, the vibration sensor comprises three components, wherein... Two of the vibration sensors are installed radially on the rotating shaft of the rotating machinery, one vibration sensor is installed axially on the rotating shaft of the rotating machinery, and the installation directions of the three vibration sensors are perpendicular to each other.

8. The rotating machinery fault diagnosis method based on image fusion and integrated networks as described in claim 6, characterized in that, In step 2, the standard data sample also includes the rotating machinery condition, rotating machinery fault type, and rotating machinery fault size.

9. The rotating machinery fault diagnosis method based on image fusion and integrated networks as described in claim 6, characterized in that, The time window is selected from a time window of 1 to 2 seconds.

10. The rotating machinery fault diagnosis method based on image fusion and integrated networks as described in claim 6, characterized in that, The time-frequency analysis methods include, but are not limited to, continuous wavelet transform, short-time Fourier transform, Hilbert-Huang transform, Fourier compressed synchronous transform, and Wigner-Ville distribution.

Images