Rotary machinery zero fault sample diagnosis method, device, equipment and medium

Abstract

The invention relates to a rotary machine zero fault sample diagnosis method and device, equipment and a medium. The method comprises the following steps: 1) constructing n sampling parameters including sampling frequency, host rotating speed and sampling point positions, setting each sampling parameter to have m adjustment values, collecting vibration monitoring data of the rotating machine in a normal state under different adjustment value combinations, and constructing a plurality of groups of normal state data sets; 2) constructing a plurality of supervised learning tasks based on the SEResNet34 network, performing network optimization and verification by using the normal state data set, and obtaining an optimal model under each supervised learning task; 3) extracting a model capable of judging vibration monitoring data parameters from the obtained models, inputting data, and outputting a state prediction label; and 4) fusing the output of each group of models, and judging whether the running state of the rotating machinery is abnormal or normal according to the output state predication label. The method has the advantages of small dependence on fault samples, high detection accuracy and strong practical applicability.

Description

Technical Field

[0001] The present invention belongs to the technical field of mechanical equipment fault detection, and in particular relates to a rotating machinery zero-fault sample diagnosis method, device, equipment and medium. Background Art

[0002] During the operation of large-scale mechanical equipment, the working state of rotating machinery directly affects the operation safety of the entire equipment. Rotating machinery is prone to various abnormal conditions after long-term operation. Therefore, accurate state monitoring is crucial for the safe and reliable operation of the overall equipment. Traditional rotating machinery fault diagnosis methods mostly rely on supervised learning and require a large amount of fault data for training. However, it is often difficult to obtain sufficient fault data in practical applications. In addition, traditional methods rely on manual feature extraction, and fault identification capabilities are highly dependent on high-quality feature extraction and classifiers. Summary of the invention

[0003] The object of the present invention is to provide a rotating machinery zero-fault sample diagnosis method, device, equipment and medium, which has low dependence on fault samples, high detection accuracy and strong adaptability.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is: a zero-fault sample diagnosis method for rotating machinery, characterized in that it includes the following steps: 1) Construct n sampling parameters including sampling frequency, main engine speed, and sampling point. Assume that each sampling parameter has m adjustment values. Use the control variable method to fix n-1 sampling parameters, change the nth sampling parameter, and collect vibration monitoring data of rotating machinery in normal state under different adjustment value combinations. A total of m vibration monitoring data of rotating machinery in normal state are obtained. n-1 group vibration monitoring data; normalize and two-dimensionally transform the obtained vibration monitoring data, with the goal of identifying different sampling parameters, construct multiple groups of normal state data sets, and assign sampling parameter labels to the normal state data sets; 2) Build multiple supervised learning tasks based on the SEResNet34 network, use the normal state data set to optimize and verify the network, and obtain the best model for each supervised learning task; 3) After processing the vibration monitoring data of the rotating machinery, record the sampling parameter labels of the vibration monitoring data, then extract a model that can judge the vibration monitoring data parameters from the model obtained in step 2) and input the data. If the model output is consistent with a sampling parameter in the vibration monitoring data, then output the corresponding state prediction label. The output label value "0" indicates that the monitored object is in a normal state, otherwise the output label value "1" indicates that the monitored object is in an abnormal state; 4) The outputs of each group of models are fused. If the label values ​​output by each model are all "0", the final judgment is that the operating state of the rotating machinery is normal. Otherwise, the operating state of the rotating machinery is abnormal.

[0005] Furthermore, the implementation method of step 1) is: At a fixed sampling frequency f i and other sampling parameters, changing the host speed r1, r2, ..., r m , collect vibration monitoring data of the rotating machinery under normal conditions by means of vibration sensors arranged at multiple sampling points on the rotating machinery; traverse all sampling frequencies to obtain vibration monitoring data of the rotating machinery under normal conditions at multiple sampling points with different sampling frequencies, main engine speeds and other sampling parameters; The obtained vibration monitoring data is normalized and two-dimensionally converted, the one-dimensional vibration monitoring data is converted into a two-dimensional grayscale image, and then the processed vibration monitoring data is reorganized to construct m n-1 Normal state data sets, each normal state data set contains the normal state data of all sampling points at the same sampling frequency and the same speed, where n represents a total of n sampling parameters, and m represents a total of m adjustment values ​​for each sampling parameter; Each normal state data set is divided into a training set and a validation set according to a set ratio, which is used for the training and validation of the SEResNet34 network model for the subsequent corresponding supervised learning tasks.

[0006] Furthermore, in step 2), construct m n-1 supervised learning task, denoted as , to distinguish the normal state data sets from different sources; for each supervised learning task, a SEResNet34 network model is constructed, and the corresponding normal state data sets are used for training and verification, and the model with the best verification accuracy under each supervised learning task is saved, and finally m is obtained. n-1 The best model.

[0007] Furthermore, the SEResNet34 network model includes a SE module, a convolutional layer, residual blocks of multiple stages, a pooling layer and a fully connected layer; The SE module includes a compression module, an excitation module and a scaling module; the compression module is represented as follows: Among them, z c is the compressed value of the cth channel, x c (i, j) is the value of the cth channel of the input feature map at position (i, j), H and W are the height and width of the feature map respectively; The excitation module is represented as follows: Among them, z is the channel compression value vector obtained after global average pooling, W1 and W2 are the weight matrices of the two fully connected layers, δ is the ReLU activation function, σ is the Sigmoid activation function, and s is the output channel weight vector; The scaling module is represented as follows: Among them, s c is the weight factor of the cth channel, x c is the cth channel of the input feature map, is the cth channel of the recalibrated feature map; The convolution operation of the convolutional layer is expressed as follows: Among them, Y(i, j) represents the value of the output feature map at position (i, j), X(i+p, j+q) is the value of the input feature map at position (i+p, j+q), W(p,q) represents the weight of the convolution kernel at position (p, q), b is the bias, and K is the number of convolution kernels; There are multiple residual blocks in each stage, and the residual operation of the residual block is expressed as follows: Where X is the input of the residual block, Y is the output, and F(X) represents the transformation operation performed by the residual block; The pooling operation of the pooling layer is maximum pooling, which is expressed as follows: Among them, Y(i, j) is the value of the output feature map at position (i, j), For the input feature map at position The value at , P represents the pooling window length, and S is the pooling stride; During the model training and verification process, the input image is passed to each layer of the network through forward propagation, and then passes through the residual blocks of each stage in turn. The cross entropy loss function is used to measure the classification error of the model. The model parameters are updated and tuned in combination with backpropagation and Adam optimizer. After the training is completed, the optimal parameters of the model are saved to obtain the optimal model.

[0008] Furthermore, the residual blocks of the multiple stages include: 3 residual blocks of Layer1 in the first stage, 4 residual blocks of Layer2 in the second stage, 6 residual blocks of Layer3 in the third stage, and 3 residual blocks of Layer4 in the fourth stage. Each residual block has two layers of convolution, and there are 32 convolution layers in total.

[0009] Furthermore, the implementation method of step 3) is: Acquire vibration monitoring data of rotating machinery under different sampling parameters, and perform normalization and two-dimensional conversion on the acquired vibration monitoring data; The processed vibration monitoring data is reorganized to obtain m n-1 Set the data to be tested; M n-1 The group of test data is input into the multiple groups of models obtained in step 2), and the state prediction labels corresponding to the vibration monitoring data are output to complete the preliminary identification of the state of the rotating machinery.

[0010] The present invention also provides a rotating machinery zero-fault sample diagnosis device for implementing the above method, characterized in that it comprises: The data acquisition and processing module is used to collect n sampling parameters including sampling frequency, host speed, sampling point position, and m different adjustment value combinations of the rotating machinery in the normal state. n-1 The obtained vibration monitoring data are normalized and converted into two dimensions, with the goal of identifying different sampling parameters, to construct multiple groups of normal state data sets, and to assign sampling parameter labels to the normal state data sets; The task construction module is used to build multiple supervised learning tasks based on the SEResNet34 network, use the normal state data set to optimize and verify the network, and obtain the best model for each supervised learning task; The fault diagnosis module is used to extract a model that can determine the parameters of the vibration monitoring data from the model obtained by the task construction module. If the model output is consistent with a sampling parameter in the vibration monitoring data, the corresponding state prediction label is output. The output label value "0" indicates that the monitored object is in a normal state, otherwise the output label value "1" indicates that the monitored object is in an abnormal state; then the outputs of each group of models are fused. If the label values ​​output by each model are all "0", the final judgment is that the operating state of the rotating machinery is normal, otherwise it is judged that the operating state of the rotating machinery is abnormal.

[0011] The present invention also provides a computer device, comprising: at least one processor, at least one memory, and computer program instructions stored in the memory, and the above method is implemented when the computer program instructions are executed by the processor.

[0012] The present invention also provides a computer-readable storage medium on which computer program instructions are stored. When the computer program instructions are executed by a processor, the above method is implemented.

[0013] Compared with the prior art, the present invention has the following beneficial effects: Traditional supervised learning requires a large amount of fault sample data for model training, while the data of the fault state of rotating machinery is difficult to obtain in practice. Therefore, the present invention proposes a rotating machinery fault diagnosis method based on zero-fault sample learning under the SEResNet34 network. The method only needs the data of the normal state of the rotating machinery for training, gets rid of the dependence on fault samples, effectively alleviates the problem of insufficient fault samples in engineering practice, and significantly improves the robustness and accuracy of zero-fault sample diagnosis of rotating machinery. The traditional supervised learning model needs to be trained with a specific type of fault data, and the obtained fault detection model is not suitable for other types of faults. The present invention establishes a model through normal state data under different sampling parameters, which has stronger practical applicability and can better cope with the intelligent anomaly detection requirements under zero-fault samples. The present invention uses multiple groups of normal sample data for training, and improves the accuracy of zero-fault sample diagnosis of rotating machinery through fusion judgment. In addition, the present invention combines multiple normal samples, conducts multi-recognition task training, introduces fusion fault judgment criteria, improves the generalization of the model, and has higher application potential in actual engineering scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a flow chart of a method implementation of an embodiment of the present invention; Figure 2 It is a schematic diagram of the basic structure of the SEResNet34 network in an embodiment of the present invention; Figure 3 It is a two-dimensional grayscale image under different states of a sampling frequency of 12.8 khz in an embodiment of the present invention; Figure 4 is a diagram of fault identification accuracy of the fusion model in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0016] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0017] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0018] like Figure 1 As shown, this embodiment provides a zero-fault sample diagnosis method for rotating machinery, comprising the following steps: S1. Construct n sampling parameters such as sampling frequency, main engine speed, sampling point, etc. Assume that each sampling parameter has m adjustment values. Use the control variable method to fix n-1 sampling parameters, change the nth sampling parameter, collect vibration monitoring data of rotating machinery in normal state under different adjustment value combinations, and obtain a total of m vibration monitoring data of rotating machinery in normal state. n-1 The obtained vibration monitoring data are normalized and two-dimensionally transformed to identify different sampling parameters, and multiple groups of normal state data sets are constructed, and sampling parameter labels are assigned to the normal state data sets.

[0019] In order to collect vibration data, multiple vibration sensors are arranged at different points on the rotating machinery.

[0020] At a fixed sampling frequency f i and other sampling parameters, changing the host speed r1, r2, ..., r m , the vibration monitoring data of the rotating machinery under normal conditions are collected by vibration sensors arranged at multiple sampling points on the rotating machinery. All sampling frequencies are traversed to obtain the vibration monitoring data of the rotating machinery under normal conditions at different sampling frequencies, main engine speeds and other sampling parameters and multiple sampling points.

[0021] Then, the obtained vibration monitoring data is normalized and two-dimensionally transformed, the one-dimensional vibration monitoring data is converted into a two-dimensional grayscale image, and then the processed vibration monitoring data is reorganized to construct m n-1 Normal state data sets, each normal state data set contains the normal state data of all sampling points at the same sampling frequency and the same speed, where n represents a total of n sampling parameters, and m represents a total of m adjustment values ​​for each sampling parameter.

[0022] Each normal state data set is divided into a training set and a validation set according to a set ratio, which is used for the training and validation of the SEResNet34 network model for the subsequent corresponding supervised learning tasks.

[0023] S2. Build multiple supervised learning tasks based on the SEResNet34 network, use the normal state data set to optimize and verify the network, and obtain the best model for each supervised learning task.

[0024] In this embodiment, construct m n-1 supervised learning task, denoted as , to distinguish the normal state data sets from different sources; for each supervised learning task, a SEResNet34 network model is constructed, and the corresponding normal state data sets are used for training and verification, and the model with the best verification accuracy under each supervised learning task is saved, and finally m is obtained. n-1 The best model.

[0025] like Figure 2 As shown, the SEResNet34 network model includes a SE (Squeeze-and-Excitation) module, a convolutional layer, multiple stages of residual blocks, a pooling layer, and a fully connected layer.

[0026] 1. SE (Squeeze-and-Excitation) module The SE module is a mechanism for enhancing the feature representation capability of convolutional neural networks, including three modules: Squeeze, Excitation, and Scale.

[0027] The compression module is represented as follows: Among them, z c is the compressed value of the cth channel, x c (i, j) is the value of the cth channel of the input feature map at position (i, j), and H and W are the height and width of the feature map, respectively.

[0028] The excitation module is represented as follows: Among them, z is the channel compression value vector obtained after global average pooling, W1 and W2 are the weight matrices of the two fully connected layers, δ is the ReLU activation function, σ is the Sigmoid activation function, and s is the output channel weight vector.

[0029] The scaling module is represented as follows: Among them, s c is the weight factor of the cth channel, x c is the cth channel of the input feature map, is the cth channel of the recalibrated feature map.

[0030] 2. Convolution operation The convolution operation of the convolutional layer is expressed as follows: Among them, Y(i, j) represents the value of the output feature map at position (i, j), X(i+p, j+q) is the value of the input feature map at position (i+p, j+q), W(p,q) represents the weight of the convolution kernel at position (p, q), b is the bias, and K is the number of convolution kernels.

[0031] 3. Residual calculation In this embodiment, the residual blocks of the multiple stages include: 3 residual blocks of Layer 1 in the first stage, 4 residual blocks of Layer 2 in the second stage, 6 residual blocks of Layer 3 in the third stage, and 3 residual blocks of Layer 4 in the fourth stage. Each residual block has two layers of convolution, and there are 32 convolution layers in total.

[0032] The residual operation of the residual block is expressed as follows: Among them, X is the input of the residual block, Y is the output, and F(X) represents the transformation operation performed by the residual block.

[0033] 4. Pooling The pooling operation of the pooling layer is maximum pooling, which is expressed as follows: Among them, Y(i, j) is the value of the output feature map at position (i, j), For the input feature map at position The value at , P represents the pooling window length, and S is the pooling stride.

[0034] During the model training and verification process, the input image is passed to each layer of the network through forward propagation, and then passes through the residual blocks of each stage in turn. The cross entropy loss function is used to measure the classification error of the model. The model parameters are updated and tuned in combination with backpropagation and Adam optimizer. After the training is completed, the optimal parameters of the model are saved to obtain the optimal model.

[0035] S3, after processing the vibration monitoring data of the rotating machinery, record the sampling parameter labels of the vibration monitoring data, then extract the model that can judge the vibration monitoring data parameters from the model obtained in step S2 and input the data, if the model output is consistent with a sampling parameter in the vibration monitoring data, then output the corresponding state prediction label, the output label value "0" indicates that the monitored object is in a normal state, otherwise the output label value "1" indicates that the monitored object is in an abnormal state. The specific implementation method is: 3.1. Obtain vibration monitoring data of rotating machinery under different sampling parameters, and perform normalization and two-dimensional conversion on the obtained vibration monitoring data.

[0036] 3.2. Reorganize the processed vibration monitoring data to obtain mn-1 Set the data to be tested; 3.3、m n-1 The group of test data is input into the multiple groups of models obtained in step 2), and the state prediction labels corresponding to the vibration monitoring data are output to complete the preliminary identification of the state of the rotating machinery.

[0037] Specifically, m n-1 The corresponding input of the group of samples to be tested is The input principle of the multiple models obtained below is that the sampling parameter labels contained in the sample to be tested have at least one corresponding relationship with the sampling parameter labels contained in the above model training set, that is, the sampling parameter labels of the model training set contain the sampling parameter labels of the sample to be tested. In addition, multiple sets of model result fusion recognition rules are set. If the parameter labels output by the model are consistent with the parameter labels of the sample to be tested, the recognition result is judged to be consistent, represented by "0". Otherwise, the recognition result is inconsistent, represented by "1". The sample to be tested is passed The recognition results obtained by the models are summarized in Table 1.

[0038] Table 1 S4. The outputs of each group of models are integrated. If the label values ​​output by each model are all "0", the operating state of the rotating machinery is finally determined to be normal. Otherwise, the operating state of the rotating machinery is determined to be abnormal.

[0039] Specifically, the predicted labels obtained after the test sample set consisting of the two-dimensional grayscale image of the fault signal is predicted in step S4 are combined according to the logic of step 3.2, and the resulting situation is shown in Table 2. When the "and" operation result of each recognition result is "0", it is considered that the current state of the rotating machinery is normal, otherwise, the rotating machinery state is judged to be abnormal.

[0040] Table 2 The present invention also provides a rotating machinery zero-fault sample diagnosis device for implementing the above method, characterized in that it comprises: The data acquisition and processing module is used to collect n sampling parameters including sampling frequency, host speed, sampling point position, and m different adjustment value combinations of the rotating machinery in the normal state. n-1 The obtained vibration monitoring data are normalized and converted into two dimensions, with the goal of identifying different sampling parameters, to construct multiple groups of normal state data sets, and to assign sampling parameter labels to the normal state data sets; The task construction module is used to build multiple supervised learning tasks based on the SEResNet34 network, use the normal state data set to optimize and verify the network, and obtain the best model for each supervised learning task; The fault diagnosis module is used to extract a model that can determine the parameters of the vibration monitoring data from the model obtained by the task construction module. If the model output is consistent with a sampling parameter in the vibration monitoring data, the corresponding state prediction label is output. The output label value "0" indicates that the monitored object is in a normal state, otherwise the output label value "1" indicates that the monitored object is in an abnormal state; then the outputs of each group of models are fused. If the label values ​​output by each model are all "0", the final judgment is that the operating state of the rotating machinery is normal, otherwise it is judged that the operating state of the rotating machinery is abnormal.

[0041] The present invention also provides a computer device, comprising: at least one processor, at least one memory, and computer program instructions stored in the memory, and the above method is implemented when the computer program instructions are executed by the processor.

[0042] The present invention also provides a computer-readable storage medium on which computer program instructions are stored. When the computer program instructions are executed by a processor, the above method is implemented.

[0043] The implementation process of the present invention is further described below with a specific embodiment and specific experimental data, and the effectiveness of the present invention is verified.

[0044] like Figure 1 As shown, in this embodiment, the method provided by the present invention is applied to the fault diagnosis of the ship stern bearing, and the implementation process is as follows: Step 1: Vibration sensors are set at different distribution points to obtain vibration monitoring data of the stern bearing under three sampling frequencies, three main engine speeds and three wear degrees.

[0045] Step 2: Perform two-dimensional conversion and normalization on the collected vibration data, and convert the one-dimensional vibration data into a two-dimensional matrix (grayscale image) for model training and testing. Each data set contains vibration data samples at different speeds under the same sampling frequency and fixed collection points. A total of 9 data sets are constructed, each containing 3,600 grayscale images.

[0046] Step 3: Build the SEResNet34 network model for each of the 9 training sample sets. The model structure is as follows: Figure 2 As shown in the figure. During the training process, each data set is divided into a training set and a test set according to the predetermined ratio. The highest accuracy is evaluated on the test set in each round of training, and the best model is saved as the best model for the training data set. A total of 9 best models (referred to as task1 to task9) are obtained.

[0047] Step 4: Then use the collected fault vibration data through their corresponding models to obtain the recognition results. The rule is that when the output is "0", it means that the model judges the fault signal as normal. When the output label is "1", it means that the model judges the vibration data to be tested as abnormal. Finally, the labels predicted by multiple models are saved.

[0048] Step 5: The predicted labels obtained from the vibration data to be tested at different points collected at the same time are fused and judged. The fusion judgment rule is to combine the predicted labels obtained from the fault vibration data at three different points. When the predicted label is "1", the vibration data to be tested is in an abnormal state.

[0049] In this embodiment, the experimental data is collected by vibration sensors distributed at three positions (2 o'clock, 12 o'clock, and 10 o'clock) on the stern bearing. Three sampling frequencies (12.8khz, 25.6khz, and 51.2khz) are selected, three main engine speeds (250rpm, 500rpm, and 750rpm) are set, and three different stern bearing wear states (0.01mm, 0.03mm, and 0.05mm) are simulated, and experiments are carried out respectively. The effectiveness of the method of the present invention and the specific operation process are verified by using the above experimental data as follows: Step 1: Vibration sensors are distributed at 2, 10, and 12 o'clock above the stern bearing. The vibration data of the stern bearing in normal state is continuously monitored at different sampling frequencies (12.8kHz, 25.6kHz, 51.2kHz) and different main engine speeds (250rpm, 500rpm, 750rpm) for 4 minutes.

[0050] Step 2: Arrange the collected vibration data according to the fixed sampling frequency and collection points to form 9 groups of normal state data sample sets, and use the sample sets to construct 9 groups of supervised learning tasks (task1~task9). For example, set the sampling frequency to 12.8kHz, and at a speed of 750rpm, use the normal vibration data at different sampling points to construct a group of normal state data sample sets, which contains three point sampling parameter labels, namely 2 points, 10 points, and 12 points. Then, the data samples are normalized and two-dimensionally converted. Specifically, the one-dimensional vibration data is converted into a two-dimensional grayscale image for easy model training and testing. The vibration data of each state generates 1200 grayscale images by sliding the signal window. Therefore, each training data set contains 3600 grayscale images, and a total of 9 training data sets are constructed. The obtained grayscale image Figure 3 The two-dimensional grayscale images of data samples at different sampling points are shown at a sampling frequency of 12.8kHz and a rotation speed of 750rpm. Figure 3In the figure, (a) is a two-dimensional grayscale image of a normal signal with 10 points at 750rpm; (b) is a two-dimensional grayscale image of a normal signal with 12 points at 750rpm; (c) is a two-dimensional grayscale image of a normal signal with 2 points at 750rpm.

[0051] Step 3: Establish 9 SEResNet34 models and use 9 training sets to optimize the model network parameters. Each group of optimal models corresponds to 3 groups of sampling frequencies and 3 groups of sampling points. For example, the model for identifying different speed sampling parameters at a sampling frequency of 12.8kHz and 2 points is task1; the model for identifying different speed sampling parameters at a sampling frequency of 12.8kHz and 10 points is task2; the model for identifying different speed sampling parameters at a sampling frequency of 12.8kHz and 12 points is task3. The model for identifying different speed sampling parameters at a sampling frequency of 25.6kHz and 2 points is task4; the model for identifying different speed sampling parameters at a sampling frequency of 25.6kHz and 10 points is task5; the model for identifying different speed sampling parameters at a sampling frequency of 25.6kHz and 12 points is task6. The model for identifying different speed sampling parameters at a sampling frequency of 51.2kHz and 2 points is task4; the model for identifying different speed sampling parameters at a sampling frequency of 51.2kHz and 10 points is task5; the model for identifying different speed sampling parameters at a sampling frequency of 51.2kHz and 12 points is task6. After each round of training, the validation samples are used for evaluation, and the model with the highest accuracy is saved as the best model for each group of supervised learning tasks.

[0052] Step 4: Use the corresponding optimal model to test 1,000 test samples of stern bearings with different wear conditions to obtain preliminary test labels. If the output label is "0", the test sample is normal; if the output label is "1", the test sample is abnormal. Record the test results of all models on the test samples.

[0053] Step 5: Combine the models under task1~task3 and name it as multi-source fusion Model1; combine task4~task6 and name it as multi-source fusion Model2; combine task7~task9 and name it as multi-source fusion Model3. The three recognition results of Model1 are fused. If "1" appears in the recognition result, the test sample is judged to be in an abnormal state. Similarly, the detection labels of Model2 and Model3 are fused, and finally the fault detection results of different wear states of the stern bearing under Model1, Model2, and Model3 are obtained, such as Figure 4 shown. Figure 4In the figure, (a) is the fault recognition accuracy of multi-source fusion Model 1; (b) is the fault recognition accuracy of multi-source fusion Model 2; (c) is the fault recognition accuracy of multi-source fusion Model 3. Figure 4 It can be seen that at the sampling frequency of 12.8kHz, except for the significant decrease in accuracy at 750rpm host speed and higher wear (0.03mm and 0.05mm), the accuracy in other cases remains at a high level. It can be seen that at low and medium host speeds, the model performs well in identifying mild and moderate wear, but the model performance decreases significantly under high speed and high wear conditions. At the sampling frequencies of 25.6kHz and 51.2kHz, except for the slightly lower accuracy at 250rpm speed and mild wear (0.01mm), the accuracy of the former remains at a high level in other cases; the latter has excellent diagnostic effect under low and medium speed conditions. However, under high speed and high wear (0.03mm and 0.05mm) conditions, the accuracy decreases slightly, but still remains at a high level.

[0054] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0055] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0056] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1A function specified in one or more boxes.

[0057] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0058] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A zero-fault sample diagnosis method for rotating machinery, characterized in that: The following steps are involved: 1) Construct n sampling parameters including sampling frequency, main engine speed, and sampling point. Assume that each sampling parameter has m adjustment values. Use the control variable method to fix n-1 sampling parameters, change the nth sampling parameter, and collect vibration monitoring data of rotating machinery in normal state under different adjustment value combinations. A total of m vibration monitoring data of rotating machinery in normal state are obtained. n-1 Group vibration monitoring data; The obtained vibration monitoring data is normalized and converted into two dimensions, with the goal of identifying different sampling parameters, to construct multiple sets of normal state data sets, and to assign sampling parameter labels to the normal state data sets; 2) Build multiple supervised learning tasks based on the SEResNet34 network, use the normal state data set to optimize and verify the network, and obtain the best model for each supervised learning task; 3) After processing the vibration monitoring data of the rotating machinery, record the sampling parameter labels of the vibration monitoring data, then extract the model that can judge the vibration monitoring data parameters from the model obtained in step 2) and input the data. If the model output is consistent with a sampling parameter in the vibration monitoring data, then output the corresponding state prediction label. The output label value "0" indicates that the monitored object is in a normal state, otherwise the output label value "1" indicates that the monitored object is in an abnormal state; 4) The outputs of each group of models are fused. If the label values ​​output by each model are all "0", the final judgment is that the operating state of the rotating machinery is normal. Otherwise, the operating state of the rotating machinery is abnormal.

2. A rotating machinery zero-fault sample diagnosis method according to claim 1, characterized in that: The implementation method of step 1) is: At a fixed sampling frequency f i and other sampling parameters, changing the host speed r1, r2, ..., r m , collect vibration monitoring data of the rotating machinery under normal conditions by means of vibration sensors arranged at multiple sampling points on the rotating machinery; traverse all sampling frequencies to obtain vibration monitoring data of the rotating machinery under normal conditions at multiple sampling points with different sampling frequencies, main engine speeds and other sampling parameters; The obtained vibration monitoring data is normalized and two-dimensionally converted, the one-dimensional vibration monitoring data is converted into a two-dimensional grayscale image, and then the processed vibration monitoring data is reorganized to construct m n-1 Normal state data sets, each normal state data set contains the normal state data of all sampling points at the same sampling frequency and the same speed, where n means there are n kinds of sampling parameters, and m means there are m kinds of adjustment values ​​for each sampling parameter; Each normal state data set is divided into a training set and a validation set according to a set ratio, which is used for the training and validation of the SEResNet34 network model for the subsequent corresponding supervised learning tasks.

3. A zero-fault sample diagnosis method for rotating machinery according to claim 1, characterized in that: In step 2), construct m n-1 supervised learning task, denoted as , to distinguish the normal state data sets from different sources; for each supervised learning task, a SEResNet34 network model is constructed, and the corresponding normal state data sets are used for training and verification, and the model with the best verification accuracy under each supervised learning task is saved, and finally m is obtained. n-1 The best model.

4. A rotating machinery zero-fault sample diagnosis method according to claim 1, characterized in that: The SEResNet34 network model includes a SE module, a convolutional layer, residual blocks of multiple stages, a pooling layer and a fully connected layer; The SE module includes a compression module, an excitation module and a scaling module; the compression module is represented as follows: Among them, z c is the compressed value of the cth channel, x c (i, j) is the value of the cth channel of the input feature map at position (i, j), and H and W are the height and width of the feature map respectively; The excitation module is represented as follows: Among them, z is the channel compression value vector obtained after global average pooling, W1 and W2 are the weight matrices of the two fully connected layers, δ is the ReLU activation function, σ is the Sigmoid activation function, and s is the output channel weight vector; The scaling module is represented as follows: Among them, s c is the weight factor of the cth channel, x c is the cth channel of the input feature map, is the cth channel of the recalibrated feature map; The convolution operation of the convolutional layer is expressed as follows: Among them, Y(i, j) represents the value of the output feature map at position (i, j), X(i+p, j+q) is the value of the input feature map at position (i+p, j+q), W(p, q) represents the weight of the convolution kernel at position (p, q), b is the bias, and K is the number of convolution kernels; There are multiple residual blocks in each stage, and the residual operation of the residual block is expressed as follows: Where X is the input of the residual block, Y is the output, and F(X) represents the transformation operation performed by the residual block; The pooling operation of the pooling layer is maximum pooling, which is expressed as follows: Among them, Y(i, j) is the value of the output feature map at position (i, j), For the input feature map at position The value at , P represents the pooling window length, and S is the pooling stride; During the model training and verification process, the input image is passed to each layer of the network through forward propagation, and then passes through the residual blocks of each stage in turn. The cross entropy loss function is used to measure the classification error of the model. The model parameters are updated and tuned in combination with backpropagation and Adam optimizer. After the training is completed, the optimal parameters of the model are saved to obtain the optimal model.

5. A zero-fault sample diagnosis method for rotating machinery according to claim 4, characterized in that: The residual blocks of the multiple stages include: 3 residual blocks of Layer1 in the first stage, 4 residual blocks of Layer2 in the second stage, 6 residual blocks of Layer3 in the third stage, and 3 residual blocks of Layer4 in the fourth stage. Each residual block has two layers of convolution, and there are 32 convolution layers in total.

6. A zero-fault sample diagnosis method for rotating machinery according to claim 1, characterized in that: The implementation method of step 3) is: Acquire vibration monitoring data of rotating machinery under different sampling parameters, and perform normalization and two-dimensional conversion on the acquired vibration monitoring data; The processed vibration monitoring data is reorganized to obtain m n-1 Set the data to be tested; M n-1 The group of test data is input into the multiple groups of models obtained in step 2), and the state prediction labels corresponding to the vibration monitoring data are output to complete the preliminary identification of the state of the rotating machinery.

7. A rotating machinery zero-fault sample diagnosis device for implementing the method according to any one of claims 1 to 6, characterized in that: include: The data acquisition and processing module is used to collect n sampling parameters including sampling frequency, host speed, sampling point position, and m different adjustment value combinations of the rotating machinery in the normal state. n-1 The obtained vibration monitoring data are normalized and converted into two dimensions, with the goal of identifying different sampling parameters, to construct multiple groups of normal state data sets, and to assign sampling parameter labels to the normal state data sets; The task construction module is used to build multiple supervised learning tasks based on the SEResNet34 network, use the normal state data set to optimize and verify the network, and obtain the best model for each supervised learning task; The fault diagnosis module is used to extract a model that can judge the parameters of the vibration monitoring data from the model obtained by the task construction module. If the model output is consistent with a sampling parameter in the vibration monitoring data, the corresponding state prediction label is output. The output label value "0" indicates that the monitored object is in a normal state, otherwise the output label value "1" indicates that the monitored object is in an abnormal state; then the outputs of each group of models are fused. If the label values ​​output by each model are all "0", the final judgment is that the operating state of the rotating machinery is normal, otherwise it is judged that the operating state of the rotating machinery is abnormal.

8. A computer device, characterized in that: include: At least one processor, at least one memory and computer program instructions stored in the memory, when the computer program instructions are executed by the processor, implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images