A rotor system fault sensitive feature extraction method and system

Abstract

The present invention relates to a method and system for extracting fault-sensitive features of a rotor system, which are characterized by comprising: acquiring a vibration acceleration signal of a rotor system to be tested in real time, and generating a fault sample after adding a fault label; inputting the generated fault sample into a pre-built feature extraction model based on a multi-channel one-dimensional residual network to determine the fault features of the rotor system to be tested. The present invention can accurately characterize the differences between different faults and the similarities between similar faults, and can be widely used in the field of rotor system detection.

Description

Technical Field

[0001] The present invention relates to the field of rotor system detection, and in particular to a method and system for extracting fault-sensitive features of a rotor system. Background Art

[0002] When diagnosing mechanical equipment faults, fault-sensitive feature extraction is a key step, and the quality of the final diagnosis results depends largely on the quality of the extracted features. Convolutional Neural Network (CNN) has become a "weapon" for data feature mining due to its unique local perception and parameter sharing structural characteristics.

[0003] However, deep CNNs also face the problems of gradient vanishing or exploding and network performance degradation. At the same time, in the field of fault diagnosis, existing studies mostly use single-channel 2D images as network input, but there is also the problem that the information covered by the single-channel signal is not comprehensive enough, and the conversion of 1D data into 2D images may lead to the loss of key information. Summary of the invention

[0004] In view of the above problems, an object of the present invention is to provide a method and system for extracting fault-sensitive features of a rotor system which covers comprehensive information and does not lose any information.

[0005] To achieve the above objectives, the present invention adopts the following technical solutions: In a first aspect, a method for extracting fault sensitive features of a rotor system is provided, comprising:

[0006] Acquire the vibration acceleration signal of the rotor system to be tested in real time, and generate fault samples after adding fault labels;

[0007] The generated fault samples are input into a pre-built feature extraction model based on a multi-channel one-dimensional residual network to determine the fault characteristics of the rotor system to be tested.

[0008] Furthermore, the fault labels of the rotor system include normal state, misalignment, imbalance and loose bearing seat.

[0009] Furthermore, the construction process of the feature extraction model based on the multi-channel one-dimensional residual network is:

[0010] Obtain the vibration acceleration signal of the rotor system, add fault labels and generate fault samples;

[0011] According to the preset ratio, the fault samples are randomly divided into training set and test set;

[0012] A feature extraction model based on a multi-channel one-dimensional residual network is constructed, and the constructed feature extraction model is trained and tested based on the obtained training set and test set to obtain a trained feature extraction model based on a multi-channel one-dimensional residual network.

[0013] Furthermore, the step of acquiring the vibration acceleration signal of the rotor system and generating a fault sample after adding a fault label includes:

[0014] Acceleration sensors are arranged on the bearing seats at both ends of the rotor system, and vibration acceleration signals of the rotor system in radial and vertical directions under different working conditions are obtained through the acceleration sensors;

[0015] Add fault tags to vibration acceleration signals;

[0016] The sliding window sampling strategy is adopted to divide the vibration acceleration signal after the fault label according to the preset sample length, step length and overlap length to obtain the fault sample corresponding to each fault label.

[0017] Furthermore, the feature extraction model based on the multi-channel one-dimensional residual network is constructed, and the constructed feature extraction model is trained and tested based on the obtained training set and test set to obtain a trained feature extraction model based on the multi-channel one-dimensional residual network, including:

[0018] Construct a feature extraction model based on a multi-channel one-dimensional residual network and determine the parameters of the feature extraction model;

[0019] Based on the obtained training set, the constructed feature extraction model based on the multi-channel one-dimensional residual network is trained;

[0020] Based on the obtained test set, the trained feature extraction model is tested to obtain a trained feature extraction model based on a multi-channel one-dimensional residual network.

[0021] Furthermore, the constructing of a feature extraction model based on a multi-channel one-dimensional residual network and determining parameters of the feature extraction model include:

[0022] Determine the network model structure of the feature extraction model based on a multi-channel one-dimensional residual network;

[0023] Determine the raw data dimension of the input to the feature extraction model based on multi-channel one-dimensional residual network;

[0024] Determining a multi-channel signal input of a feature extraction model based on a multi-channel one-dimensional residual network;

[0025] Determine the hyperparameters of a multi-channel 1D residual network based feature extraction model.

[0026] Furthermore, the feature extraction model based on the multi-channel one-dimensional residual network includes a Conv1 layer, a Conv2_x layer, a Conv3_x layer and a Conv4_x layer;

[0027] The Conv1 layer includes 64 1×7 convolution kernels with a stride of 2;

[0028] The Conv2_x layer includes a maximum pooling layer, a data dimension reduction layer, a feature extraction layer and three 3-layer residual learning modules, wherein the convolution kernel size of the maximum pooling layer is 1×3 and the step size is 2; the convolution kernel size of the data dimension reduction layer is 1×1 and the number of channels is 64; the convolution kernel size of the feature extraction layer is 1×3 and the number of channels is 64; the three 3-layer residual learning modules are composed of a data dimension increase layer with a convolution kernel size of 1×1 and a channel number of 256;

[0029] The Conv3_x layer includes a 3-layer residual learning module with the number of channels changed to 128, 128, and 512;

[0030] The Conv4_x layer includes a 3-layer residual learning module that changes the number of channels to 256, 256, and 1024.

[0031] In a second aspect, a rotor system fault sensitive feature extraction system is provided, comprising:

[0032] The data acquisition module is used to acquire the vibration acceleration signal of the rotor system to be tested in real time and generate fault samples after adding fault labels;

[0033] The fault feature extraction module is used to input the generated fault samples into a pre-built feature extraction model based on a multi-channel one-dimensional residual network to determine the fault characteristics of the rotor system to be tested.

[0034] In a third aspect, a processing device is provided, comprising computer program instructions, wherein the computer program instructions, when executed by the processing device, are used to implement the steps corresponding to the above-mentioned rotor system fault sensitive feature extraction method.

[0035] In a fourth aspect, a computer-readable storage medium is provided, on which computer program instructions are stored, wherein the computer program instructions, when executed by a processor, are used to implement the steps corresponding to the above-mentioned method for extracting fault-sensitive features of a rotor system.

[0036] The present invention adopts the above technical solution, which has the following advantages:

[0037] 1. The present invention constructs a multi-channel one-dimensional residual network and adopts 1D convolution operation to directly process the original vibration acceleration signal, automatically extract fault-sensitive features, and realize "end-to-end" fault diagnosis.

[0038] 2. The feature extraction model based on the multi-channel one-dimensional residual network constructed by the present invention adopts an improved ResNet-50 residual model structure, which reduces the number of residual blocks and removes the Conv5_x layer, reduces the network depth and the number of network parameters, thereby improving the efficiency of network calculation.

[0039] 3. The feature extraction model based on the multi-channel one-dimensional residual network in the present invention adopts multi-channel input, which can give full play to the advantages of multi-channel convolution and provide richer vibration acceleration signals for fault diagnosis.

[0040] 4. The present invention can accurately characterize the differences between different faults and the similarities between similar faults, and is suitable for fault classification tasks.

[0041] In summary, the present invention can be widely applied in the field of rotor system detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Throughout the accompanying drawings, the same reference numerals are used to represent the same components. In the accompanying drawings:

[0043] Figure 1 is a schematic diagram of a model building process provided by an embodiment of the present invention;

[0044] Figure 2 is a schematic diagram of a sample fault generation strategy provided by an embodiment of the present invention;

[0045] Figure 3 is a schematic diagram of a rotor fault diagnosis accuracy curve provided by an embodiment of the present invention;

[0046] Figure 4 is a schematic diagram of a rotor fault diagnosis loss function curve provided by an embodiment of the present invention;

[0047] Figure 5 It is a schematic diagram of feature visualization of raw data and outputs of each layer of a multi-channel 1D-ResNet provided by an embodiment of the present invention;

[0048] Figure 6 It is a schematic diagram comparing the diagnostic results of different methods provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0050] It should be understood that the terms used herein are only for the purpose of describing specific example embodiments and are not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms "one", "an" and "said" as used herein may also be meant to include plural forms. The terms "include", "comprise", "contain", and "have" are inclusive, and therefore specify the existence of stated features, steps, operations, elements and / or parts, but do not exclude the existence or addition of one or more other features, steps, operations, elements, parts, and / or combinations thereof. The method steps, processes, and operations described herein are not interpreted as necessarily requiring them to be performed in the specific order described or illustrated, unless the execution order is clearly indicated. It should also be understood that additional or alternative steps may be used.

[0051] Although the terms first, second, third, etc. can be used in the text to describe multiple elements, components, regions, layers and / or sections, these elements, components, regions, layers and / or sections should not be limited by these terms. These terms can only be used to distinguish an element, component, region, layer or section from another region, layer or section. Unless the context clearly indicates, terms such as "first", "second" and other numerical terms do not imply order or sequence when used in the text. Therefore, the first element, component, region, layer or section discussed below can be referred to as the second element, component, region, layer or section without departing from the teaching of the example embodiments.

[0052] The rotor system fault sensitive feature extraction method and system provided by the embodiment of the present invention introduce "residual learning" on the basis of CNN through the residual network (Residual Network, ResNet), and solve the above-mentioned problems faced by deep CNN by constructing a "natural identity mapping". At the same time, in order to address the information loss problem that may be encountered during the data dimension conversion process, a 1D-ResNet (One-dimensional Residual Network, 1D-ResNet) is constructed to directly process the original one-dimensional vibration acceleration signal, and multi-channel input is used to obtain more comprehensive fault information, thereby improving the effect of fault diagnosis and realizing the extraction of rotor system fault sensitive features based on a multi-channel one-dimensional residual network.

[0053] Example 1

[0054] This embodiment provides a method for extracting fault sensitive features of a rotor system, comprising the following steps:

[0055] 1) If Figure 1 As shown, a feature extraction model based on a multi-channel one-dimensional residual network is pre-built, specifically:

[0056] 1.1) Obtain the vibration acceleration signal of the rotor system and generate fault samples after adding fault labels.

[0057] 1.1.1) Acceleration sensors are installed on the bearing seats at both ends of the rotor system. The vibration acceleration signals of the rotor system in radial and vertical directions under different working conditions are obtained through the acceleration sensors. The vibration acceleration signals obtained by the acceleration sensors are sent to the host computer for analysis through the data acquisition device.

[0058] 1.1.2) Add fault tags to vibration acceleration signals.

[0059] Specifically, the fault labels of the rotor system include four types: normal state (NS), misalignment (MA), imbalance (IB) and bearing seat looseness (BSL).

[0060] 1.1.3) If Figure 2 As shown, a sliding window sampling strategy is adopted to divide the vibration acceleration signal after the fault label according to the preset sample length, step length and overlap length to obtain the fault sample corresponding to each fault label.

[0061] 1.2) Randomly divide the fault samples according to the preset ratio to obtain the training set and test set for model training and testing.

[0062] 1.3) Construct a feature extraction model based on a multi-channel one-dimensional residual network, and train and test the constructed feature extraction model based on the obtained training set and test set to obtain a trained and tested feature extraction model based on a multi-channel one-dimensional residual network, specifically:

[0063] 1.3.1) Construct a feature extraction model based on a multi-channel one-dimensional residual network, and determine the network depth, convolution kernel size and number, optimizer type, learning rate and other parameters of the feature extraction model:

[0064] 1.3.1.1) Determine the network model structure of the feature extraction model based on the multi-channel one-dimensional residual network.

[0065] Specifically, the residual network is composed of multiple residual units stacked together, mainly including direct mapping and residual parts. F(x) is the residual part, and the network output H(x) is the sum of the residual part and the direct mapping. In ResNet, the residual function F(x) is set as the goal of network learning. Through multiple iterative training, F(x) is minimized and made close to 0, and the identity mapping relationship between input and output H(x) = can be achieved:

[0066] F(x)=H(x)-x(1)

[0067] At present, commonly used residual network structures include ResNet-34, ResNet-50, and ResNet-101. Among them, the residual learning module in ResNet-34 uses two convolutional layers with a convolution kernel size of 3×3 and the same number of output channels to learn data features; while ResNet-50 uses a three-layer residual learning module, that is, first use a 1×1 convolution kernel to reduce the dimension of the data, then use a 3×3 convolution kernel to extract features, and finally use a 1×1 convolution kernel to increase the dimension of the data. This method can not only ensure the accuracy of the network, but also reduce the number of parameters, and is more suitable for deep networks. However, the traditional ResNet-50 model has a complex structure, many stacked residual blocks, and low network calculation efficiency. To address this problem, the ResNet-50 model is improved.

[0068] Specifically, the network model structure of the feature extraction model based on the multi-channel one-dimensional residual network constructed in this embodiment adopts the improved ResNet-50 model, which can effectively solve the problems of the complex structure of the existing ResNet model, the large number of stacked residual blocks, and the low network calculation efficiency. The network model structure of the feature extraction model based on the multi-channel one-dimensional residual network constructed in this embodiment is shown in Table 1 below, which records the network level of the improved ResNet-50 model, as well as the size and number of convolution kernels corresponding to each layer. In the improvement, the 3-layer residual learning module form of the ResNet-50 model is still adopted, but the original five-layer network structure is changed to four layers. The Conv1 and Conv2_x layers in the original network structure are kept unchanged, and one residual block is retained in each Conv3_x and Conv4_x layer, and the Conv5_x layer is removed to obtain the feature extraction model of the multi-channel one-dimensional residual network. The above modifications can effectively reduce the network depth and the amount of network parameters, thereby improving the efficiency of network calculation.

[0069] Specifically, the feature extraction model based on the multi-channel one-dimensional residual network includes a Conv1 layer, a Conv2_x layer, a Conv3_x layer and a Conv4_x layer, wherein the Conv1 layer includes 64 1×7 convolution kernels with a step size of 2; the Conv2_x layer includes a maximum pooling layer, a data dimension reduction layer, a feature extraction layer and three 3-layer residual learning modules, wherein the convolution kernel size of the maximum pooling layer is 1×3 and the step size is 2; the convolution kernel size of the data dimension reduction layer is 1×1 and the number of channels is 64; the convolution kernel size of the feature extraction layer is 1×3 and the number of channels is 64; the three 3-layer residual learning modules are composed of a data dimension increase layer with a convolution kernel size of 1×1 and a channel number of 256; the Conv3_x layer includes a 3-layer residual learning module that changes the number of channels to 128, 128, and 512; the Conv4_x layer includes a 3-layer residual learning module that changes the number of channels to 256, 256, and 1024.

[0070] Table 1: Model structure parameters

[0071]

[0072] 1.3.1.2) Determine the original data dimension of the input of the feature extraction model based on the multi-channel one-dimensional residual network.

[0073] Specifically, generally speaking, ResNet mainly uses 2D input for the recognition of image data, that is, the convolution kernel is in the form of a two-dimensional matrix. However, when the monitoring data is a one-dimensional time series signal, the above-mentioned convolution kernel is obviously no longer applicable. To solve this problem, this embodiment directly processes one-dimensional raw data by constructing a one-dimensional residual network (1D-ResNet). Compared with the two-dimensional residual network, the main feature of 1D-ResNet is that the size of the convolution kernel is 1×W, as shown below:

[0074] ω 1×W =[ω 11 … 1W ](2)

[0075] Among them, ω 1×W is the convolution kernel weight matrix; W is the convolution kernel size length.

[0076] 1.3.1.3) Determine a multi-channel signal input for a feature extraction model based on a multi-channel one-dimensional residual network.

[0077] Specifically, in the field of image processing, RGB three-channel images contain richer information and are more conducive to image recognition. Similarly, in mechanical fault diagnosis, the use of multi-channel input can collect more comprehensive fault information, thereby improving the accuracy of fault diagnosis. When a fault occurs, the vibration conditions at different locations of the equipment are different, so for the same type of fault, the time domain and frequency domain information of each channel are not the same. When data from multiple channels are input into the network at the same time, the fault information of the vibration acceleration signal at each location can be covered, which is more conducive to the implementation of fault classification tasks.

[0078] 1.3.1.4) Determine the hyperparameters of the feature extraction model based on the multi-channel one-dimensional residual network.

[0079] Specifically, the method of controlling variables is used to test the learning rate, batch size, and number of iterations of the feature extraction model. The accuracy and running time of the model results are comprehensively considered. When the model accuracy can achieve the highest, the parameter combination with the shortest running time is selected.

[0080] 1.3.2) Based on the obtained training set, the constructed feature extraction model based on the multi-channel one-dimensional residual network is trained.

[0081] Specifically, the network weights and biases are initialized, the training set is input into the feature extraction model in batches for forward and backward propagation, the fault features are extracted and the errors are calculated, and the training is repeated to update the network weights and biases.

[0082] 1.3.3) Based on the obtained test set, the trained feature extraction model is tested to obtain a trained and tested feature extraction model based on a multi-channel one-dimensional residual network.

[0083] Specifically, the trained feature extraction model is used to classify the test set, and the model performance is quantitatively evaluated based on the classification results, thereby obtaining a trained and tested feature extraction model based on a multi-channel one-dimensional residual network.

[0084] 2) Obtain the vibration acceleration signal of the rotor system to be tested in real time, and generate fault samples after adding fault labels.

[0085] 3) The generated fault samples are input into a pre-built feature extraction model based on a multi-channel one-dimensional residual network to determine the fault characteristics of the rotor system to be tested.

[0086] The following describes the rotor system fault sensitive feature extraction method of the present invention in detail using the Bently RK4 rotor test bench to carry out a fault diagnosis test as a specific embodiment:

[0087] In this test, the rotor system is driven by a DC motor and a coupling, and the motor is regulated by a speed regulator. When conducting the test, the equipment is first started to gradually increase the speed to 3000 rpm. After running for a period of time to allow the system to reach a stable state, the vibration acceleration signal is collected, and the sampling frequency is 20kHz. Acceleration sensors are installed on the bearing seats at both ends of the rotor to collect vibration acceleration signals of 4 channels in the radial and vertical directions of the rotor.

[0088] Experimental setup:

[0089] The test simulated four states of the rotor system: normal state, misalignment, imbalance and loose bearing seat. An additional weight of 1.2g was added to the rotor to simulate the rotor imbalance state; a 0.5mm feeler gauge was placed between the bearing seat and the base surface to simulate the rotor misalignment state; and the bolts of one bearing seat were loosened to simulate the loose state of the bearing seat.

[0090] Fault sample generation:

[0091] In order to obtain a large number of fault samples for model training, the sliding window sampling strategy is used to divide the acquired vibration acceleration signal. The sample length is set to 1000, 500 fault samples are generated for each state, and the training set and test set are randomly divided according to the ratio of 4:1. There are 1600 training samples and 400 test samples in the four fault states.

[0092] Model construction and hyperparameter selection:

[0093] This embodiment constructs a multi-channel one-dimensional residual network to carry out fault diagnosis experiments. The structural parameters are shown in Table 1 above. In this experiment, the Adam optimizer is selected, and the other hyperparameters are specifically set as follows:

[0094] The learning rates are 0.01, 0.001, and 0.0001, respectively, the batch sizes are 40, 50, and 60, respectively, and the number of iterations is 50 and 100. The test results of each parameter setting are shown in Table 2. The results show that different parameter settings have a great impact on the diagnosis results. For example, when the learning rate is 0.01 and the batch size is 40, the accuracy of 50 iterations is only 72.5%, and the diagnosis effect is not ideal. When the iteration is 50 times, the learning rate is 0.001, and the batch size is 40 or 50, the diagnosis accuracy can reach 100%. Considering the running time, the time spent when the batch size is set to 50 is shorter. Therefore, the network hyperparameter settings are: learning rate is 0.001, batch size is 50, and the number of iterations is 50.

[0095] Table 2: Test results of parameter settings

[0096]

[0097] Troubleshooting:

[0098] The fault samples of the rotor system are input into the constructed feature extraction model based on multi-channel one-dimensional residual network at a sample ratio of 4:1. The results are shown in Figure 3 , Figure 4 When the number of iterations reaches 20, the accuracy and loss function tend to be flat and the loss function value drops below 0.02, indicating that the constructed feature extraction model based on multi-channel one-dimensional residual network can converge in a relatively short time, and the final classification accuracy reaches 100%, which can fully reflect the differences between various types of faults.

[0099] The t-distributed stochastic neighbor embedding (t-SNE) algorithm is used to reduce the dimension and cluster the features of the original vibration acceleration signal and the output of each layer in the feature extraction model based on the multi-channel one-dimensional residual network. The visualization results are shown in Figure 2. Figure 5 As shown in the figure, it can be seen that the original vibration acceleration signal is chaotic and disordered, and it is difficult to distinguish the fault types. After the first convolutional layer operation, the various samples tend to cluster. With further layer-by-layer feature mining, the boundaries between different fault categories gradually become clear, and samples of the same fault type gradually gather together. It can be seen from the fully connected layer feature visualization that the four faults have been clearly distinguished, and the fault samples of the same type are also compactly clustered together.

[0100] Method comparison:

[0101] Single-channel and multi-channel samples were constructed, and fault diagnosis tests were carried out using the model of the present invention (1D-ResNet) and ResNet-18, VGG16 and "artificial features + SVM" in the prior art. The results are as follows: Figure 6As shown in the figure, in the "artificial feature + SVM" method, 7 time domain features (variance, effective value, skewness, kurtosis, peak factor, waveform factor, margin factor) and 5 frequency domain features (center of gravity frequency, root mean square frequency, frequency variance, spectrum skewness, spectrum kurtosis) of the vibration acceleration signal are extracted. The results show that compared with single-channel input, multi-channel input improves the diagnostic accuracy of each method by about 11 to 18 percentage points due to the integration of fault information at multiple locations collected by multiple sensors. In addition, the diagnostic methods based on deep networks (1D-ResNet, ResNet-18, VGG16) have higher accuracy than traditional artificial feature extraction methods. Compared with the traditional diagnostic method based on "feature engineering", the accuracy of the intelligent diagnostic method based on multi-channel 1D-ResNet has increased by 24.31 percentage points. Therefore, the rotor system fault sensitive feature extraction method of the present invention can directly process one-dimensional vibration acceleration signals, solve the problem of time-consuming and labor-intensive conversion of 1D data into 2D images and possible information loss, and at the same time, the use of multi-channel input can cover richer fault information, which is more conducive to the implementation of fault diagnosis tasks.

[0102] Example 2

[0103] This embodiment provides a rotor system fault sensitive feature extraction system, including:

[0104] The data acquisition module is used to acquire the vibration acceleration signal of the rotor system to be tested in real time and generate fault samples after adding fault labels.

[0105] The fault feature extraction module is used to input the generated fault samples into a pre-built feature extraction model based on a multi-channel one-dimensional residual network to determine the fault characteristics of the rotor system to be tested.

[0106] The system provided in this embodiment is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for specific processes and detailed contents, which will not be repeated here.

[0107] Example 3

[0108] This embodiment provides a processing device corresponding to the rotor system fault sensitive feature extraction method provided in this embodiment 1. The processing device can be applicable to a client processing device, such as a mobile phone, a laptop computer, a tablet computer, a desktop computer, etc., to execute the method of embodiment 1.

[0109] The processing device includes a processor, a memory, a communication interface and a bus, and the processor, the memory and the communication interface are connected through the bus to complete mutual communication. The memory stores a computer program that can be run on the processing device, and the processing device executes the rotor system fault sensitive feature extraction method provided in this embodiment 1 when running the computer program.

[0110] In some implementations, the memory may be a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory.

[0111] In some other implementations, the processor may be a central processing unit (CPU), a digital signal processor (DSP), or other general-purpose processors of various types, which are not limited herein.

[0112] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0113] Those skilled in the art will understand that the structure of the above-mentioned computing device is only a partial structure related to the scheme of the present application, and does not constitute a limitation on the computing device to which the scheme of the present application is applied. The specific computing device may include more or fewer components, or combine certain components, or have a different arrangement of components.

[0114] Example 4

[0115] This embodiment provides a computer program product corresponding to the rotor system fault sensitive feature extraction method provided in this embodiment 1. The computer program product may include a computer-readable storage medium on which computer-readable program instructions for executing the rotor system fault sensitive feature extraction method described in this embodiment 1 are loaded.

[0116] Computer readable storage media can be tangible devices that hold and store instructions used by instruction execution devices. Computer readable storage media can be, for example, but not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any combination thereof.

[0117] The above embodiment provides a computer-readable storage medium, whose implementation principle and technical effect are similar to those of the above method embodiment, and will not be repeated here.

[0118] 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.

[0119] 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 1 A function specified in one or more boxes.

[0120] 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.

[0121] The above embodiments are only used to illustrate the present invention, wherein the structure, connection mode and manufacturing process of each component may be changed. Any equivalent transformations and improvements based on the technical solution of the present invention should not be excluded from the protection scope of the present invention.

Claims

1. A method for extracting sensitive features of rotor system faults, It is characterized in that include: Acquire the vibration acceleration signal of the rotor system to be tested in real time, and generate fault samples after adding fault labels; The generated fault samples are input into a pre-built feature extraction model based on a multi-channel one-dimensional residual network to determine the fault characteristics of the rotor system to be tested; The construction process of the feature extraction model based on the multi-channel one-dimensional residual network is: Obtain the vibration acceleration signal of the rotor system, add fault labels and generate fault samples; According to the preset ratio, the fault samples are randomly divided into training set and test set; Constructing a feature extraction model based on a multi-channel one-dimensional residual network, and training and testing the constructed feature extraction model based on the obtained training set and test set to obtain a trained feature extraction model based on a multi-channel one-dimensional residual network; The method comprises constructing a feature extraction model based on a multi-channel one-dimensional residual network, and training and testing the constructed feature extraction model based on the obtained training set and test set to obtain a trained feature extraction model based on a multi-channel one-dimensional residual network, including: Construct a feature extraction model based on a multi-channel one-dimensional residual network and determine the parameters of the feature extraction model; Based on the obtained training set, the constructed feature extraction model based on the multi-channel one-dimensional residual network is trained; Based on the obtained test set, the trained feature extraction model is tested to obtain a trained feature extraction model based on a multi-channel one-dimensional residual network; The feature extraction model based on the multi-channel one-dimensional residual network includes a Conv1 layer, a Conv2_x layer, a Conv3_x layer and a Conv4_x layer; The Conv1 layer consists of 64 The convolution kernel has a step size of 2; The Conv2_x layer includes a maximum pooling layer, a data dimension reduction layer, a feature extraction layer and three 3-layer residual learning modules, wherein the convolution kernel size of the maximum pooling layer is , the step size is 2; the convolution kernel size of the data dimensionality reduction layer is , the number of channels is 64; the convolution kernel size of the feature extraction layer is , the number of channels is 64; the three 3-layer residual learning modules are composed of convolution kernels of size , a data dimension-enhancing layer with 256 channels; The Conv3_x layer includes a 3-layer residual learning module with the number of channels changed to 128, 128, and 512; The Conv4_x layer includes a 3-layer residual learning module that changes the number of channels to 256, 256, and 1024.

2. A rotor system fault sensitive feature extraction method as claimed in claim 1, It is characterized in that Fault tags for the rotor system include normal condition, misalignment, imbalance, and loose bearing seats.

3. A rotor system fault sensitive feature extraction method as claimed in claim 1, It is characterized in that The step of obtaining the vibration acceleration signal of the rotor system and generating a fault sample after adding a fault label comprises: Acceleration sensors are arranged on the bearing seats at both ends of the rotor system, and vibration acceleration signals of the rotor system in radial and vertical directions under different working conditions are obtained through the acceleration sensors; Add fault tags to vibration acceleration signals; The sliding window sampling strategy is adopted to divide the vibration acceleration signal after the fault label according to the preset sample length, step length and overlap length to obtain the fault sample corresponding to each fault label.

4. A rotor system fault sensitive feature extraction method as claimed in claim 1, It is characterized in that The method comprises constructing a feature extraction model based on a multi-channel one-dimensional residual network and determining parameters of the feature extraction model, including: Determine the network model structure of the feature extraction model based on a multi-channel one-dimensional residual network; Determine the raw data dimension of the input to the feature extraction model based on multi-channel one-dimensional residual network; Determining a multi-channel signal input of a feature extraction model based on a multi-channel one-dimensional residual network; Determine the hyperparameters of a multi-channel 1D residual network based feature extraction model.

5. A rotor system fault sensitive feature extraction system, It is characterized in that include: The data acquisition module is used to acquire the vibration acceleration signal of the rotor system to be tested in real time and generate fault samples after adding fault labels; A fault feature extraction module is used to input the generated fault samples into a pre-built feature extraction model based on a multi-channel one-dimensional residual network to determine the fault features of the rotor system to be tested; The construction process of the feature extraction model based on the multi-channel one-dimensional residual network is: Obtain the vibration acceleration signal of the rotor system, add fault labels and generate fault samples; According to the preset ratio, the fault samples are randomly divided into training set and test set; Constructing a feature extraction model based on a multi-channel one-dimensional residual network, and training and testing the constructed feature extraction model based on the obtained training set and test set to obtain a trained feature extraction model based on a multi-channel one-dimensional residual network; The method comprises constructing a feature extraction model based on a multi-channel one-dimensional residual network, and training and testing the constructed feature extraction model based on the obtained training set and test set to obtain a trained feature extraction model based on a multi-channel one-dimensional residual network, including: Construct a feature extraction model based on a multi-channel one-dimensional residual network and determine the parameters of the feature extraction model; Based on the obtained training set, the constructed feature extraction model based on the multi-channel one-dimensional residual network is trained; Based on the obtained test set, the trained feature extraction model is tested to obtain a trained feature extraction model based on a multi-channel one-dimensional residual network; The feature extraction model based on the multi-channel one-dimensional residual network includes a Conv1 layer, a Conv2_x layer, a Conv3_x layer and a Conv4_x layer; The Conv1 layer consists of 64 The convolution kernel has a step size of 2; The Conv2_x layer includes a maximum pooling layer, a data dimension reduction layer, a feature extraction layer and three 3-layer residual learning modules, wherein the convolution kernel size of the maximum pooling layer is , the step size is 2; the convolution kernel size of the data dimensionality reduction layer is , the number of channels is 64; the convolution kernel size of the feature extraction layer is , the number of channels is 64; the three 3-layer residual learning modules are composed of convolution kernels of size , a data dimension-enhancing layer with 256 channels; The Conv3_x layer includes a 3-layer residual learning module with the number of channels changed to 128, 128, and 512; The Conv4_x layer includes a 3-layer residual learning module that changes the number of channels to 256, 256, and 1024.

6. A processing device, It is characterized in that The method comprises computer program instructions, wherein the computer program instructions are used to implement the steps corresponding to the method for extracting fault-sensitive features of a rotor system according to any one of claims 1 to 4 when the computer program instructions are executed by a processing device.

7. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores computer program instructions, wherein the computer program instructions, when executed by a processor, are used to implement steps corresponding to the method for extracting fault-sensitive features of a rotor system according to any one of claims 1 to 4.

Images