A rotating machine fault discrimination method, computer equipment and storage medium

Abstract

The application belongs to the technical field of rotating machinery fault diagnosis, and discloses a rotating machinery fault discrimination method, a computer device and a storage medium, the rotating machinery fault discrimination method based on a time sequence deep fusion network comprises the following steps: adopting a time-frequency domain analysis method and a sliding translation window method to process vibration signals of rotating machinery, obtaining a processing result corresponding to the vibration signals, and adopting a trained convolutional neural network to analyze fault features in the processing result, and obtaining fault information of the rotating machinery. The trained convolutional neural network is adopted to analyze the fault features, and then the fault diagnosis of the rotating machinery is obtained, which is helpful to realize real-time intelligent fault diagnosis of the rotating machinery.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of rotating machinery fault diagnosis, and particularly relates to a rotating machinery fault diagnosis method, a computer device and a storage medium. BACKGROUND

[0002] At present, with the development of industrial technology, the technical level and complexity of rotating machinery equipment are improving. The failure of these devices will cause huge economic losses and even casualties. The purpose of fault diagnosis is to detect, isolate and identify faults when they occur, which is the key to ensuring the safe and reliable operation of the system. It is crucial for ensuring the safe and reliable operation of large rotating machinery and helps to develop a reasonable maintenance plan to reduce equipment maintenance costs.

[0003] Most of the existing rotating machinery diagnosis methods are to analyze the vibration signal. Most deep networks based on deep learning usually pay great attention to the quasi-static features of industrial fault signals, i.e. the amplitude and phase of each frequency in a relatively short time, but this is not enough. The waveform of rotating machinery in a short time window is a periodic signal, which is the quasi-static information corresponding to the state at a certain time, and is a manifestation of the state. In a longer period much larger than the collection period, the state of the rotating machinery may change significantly. If a fault occurs, the vibration waveform may not contain obvious fault information at a certain state, or the fault information may be submerged in other information. At this time, it is difficult to rely on randomly extracted state information for diagnosis. If the fault feature is not absolutely clear, the method based on the state waveform at a certain time cannot distinguish these faults well. Most fault signals change over time, and in the case of a large change in operating conditions, the fault will cause the rotor's working frequency vibration amplitude and phase to change regularly over time. These time series features of the signal are often ignored by deep networks, resulting in the inability of existing methods to achieve better results. Some existing CNN-based fault diagnosis methods use multi-scale convolution to solve the problem of fault time scale to some extent, but they cannot automatically learn fault time series features directly and well. Some deep learning methods use time series networks to process rotor time series features, but they do not fully utilize the depth of quasi-static features. Therefore, it is very important to organically combine the two types of features while deep networks are used for fault diagnosis.

[0004] Through the above analysis, the problems and defects of the prior art are:

[0005] (1) The existing technology has low accuracy in obtaining fault information of rotating machinery in rotating machinery fault diagnosis.

[0006] (2) In the prior art, in the case of large changes in operating conditions, the time sequence characteristics of the signal are often ignored by the deep network, resulting in that the existing method cannot obtain accurate signal data. SUMMARY

[0007] In view of the problems in the prior art, the present application provides a rotating machinery fault discrimination method, a computer device and a storage medium.

[0008] Specifically relates to a rotating machinery fault discrimination method based on a time sequence deep fusion network.

[0009] The present application is realized in this way, a rotating machinery fault discrimination method based on a time sequence deep fusion network comprises:

[0010] The time-frequency domain analysis method and the sliding translation window method are used to process the vibration signal of the rotating machinery, to obtain the processing result corresponding to the vibration signal, and the trained convolutional neural network is used to analyze the fault features in the processing result, to obtain the fault information of the rotating machinery.

[0011] Further, before the vibration signal of the rotating machinery is processed by the time-frequency domain analysis method and the sliding translation window method, the following steps need to be performed: collecting the vibration signal of the rotating machinery; setting a vibration displacement sensor on the rotating machinery, and in the movement process of the rotating machinery, the vibration displacement sensor collects the vibration signal of the rotating machinery in a predetermined sampling manner.

[0012] Further, the vibration displacement sensor is arranged along the radial direction of the shaft of the rotating machinery, to detect the vibration of the rotating machinery along the radial direction; the vibration displacement sensor is an eddy current displacement sensor, arranged along the radial direction of the shaft;

[0013] The sampling manner includes sampling length and sampling frequency, and synchronous whole cycle sampling is used, the sampling frequency is 2 n times the rotating speed of the rotating machinery, n is a positive integer greater than or equal to 6; the sampling time length is 2k rotation periods, k is a positive integer greater than or equal to 3.

[0014] The sampling frequency is 2 n times the rotating speed of the rotating machinery, n is a positive integer greater than or equal to 6; the sampling length is 2 k rotation periods, k is a positive integer greater than or equal to 3.

[0015] Further, the processing of the vibration signal of the rotating machinery by the time-frequency domain analysis method and the sliding translation window method specifically includes:

[0016] According to the input form of the pre-trained temporal deep fusion network model, the collected vibration signal is subjected to wavelet transform, and the information matrix is ​​combined to obtain the short time-frequency domain information matrix of the vibration signal.

[0017] According to the input form of the pre-trained temporal deep fusion network model, the collected vibration signal is subjected to short-time Fourier transform to extract the amplitude and phase of the characteristic frequency, and the temporal change information vector is obtained by sliding translation.

[0018] Furthermore, the specific steps for obtaining the short-time-frequency domain information matrix of the vibration signal include:

[0019] The vibration signal is truncated to a length of 2. n A one-dimensional array, where n is a positive integer greater than or equal to 6;

[0020] A wavelet transform is performed on the one-dimensional array to extract vibration signals with frequencies up to eight times the rotational speed; the one-dimensional array is then expanded to a 2D array. k ×2 n A matrix of size, serving as the short-time information matrix F of the vibration signal. x k is a positive integer greater than or equal to 3.

[0021] Furthermore, the specific steps for obtaining the temporal variation information vector of the vibration signal include:

[0022] The vibration signal was sequentially segmented into 2 parts according to the time sequence. k A length of 2 n The row vectors are denoted as follows: k is a positive integer greater than or equal to 3, and n is a positive integer greater than or equal to 6;

[0023] For these 2 k vectors Perform a Fourier transform to obtain the vibration signal of the frequency.

[0024] The amplitude and phase of the vibration signal at the characteristic frequency are extracted to obtain two signals of length 2. k The time-series vector;

[0025] Then, the sliding translation method is used to obtain 2m time series vectors. Its p m Let m be the m-th time series vector.

[0026] Furthermore, the fault information of the rotating machinery is obtained by analyzing the fault features in the processing results using a trained convolutional neural network, including:

[0027] The vibration comprehensive information matrix is input into the pre-trained time sequence deep fusion network model, the network model performs diagnostic analysis on the received vibration information matrix and vector, and a fault diagnosis result of the rotating machinery is obtained;

[0028] The pre-training process of the convolutional neural network model includes the following steps:

[0029] The rotating machinery is subjected to a fault experiment, and vibration displacement signals are collected for different fault experiments to obtain fault signals under different fault conditions;

[0030] The fault signals under different fault conditions are analyzed in the frequency domain and the time domain to obtain a short-time-time-frequency domain information matrix and a time sequence change information vector, which are used as inputs of the convolutional neural network model;

[0031] According to the occurrence of each fault diagnosis in the fault experiment, a corresponding output matrix is constructed;

[0032] The parameters of the convolutional neural network model are set, the input and output matrices are input into the time sequence deep fusion network model for training, and a pre-trained fusion neural network model for fault diagnosis and identification of the rotating machinery is obtained.

[0033] Further, the convolutional neural network includes a deep block, a time sequence block and a fusion block;

[0034] The deep block includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, an unfolding layer and a full connection layer; wherein the first convolutional layer performs matrix convolution operation on the original three-dimensional matrix to extract matrix numerical features under different convolution kernel dimensions to obtain a first feature matrix, and the features are transmitted to the first pooling layer; the first pooling layer performs pooling operation on the first feature matrix to reduce the scale of the feature matrix to obtain a second feature matrix, which is transmitted to the second convolutional layer; the second convolutional layer performs second matrix convolution operation on the second feature matrix to extract matrix feature numerical features under different convolution kernel dimensions again to obtain a third feature matrix; the second pooling layer performs pooling operation on the third feature matrix to reduce the scale of the feature matrix again to obtain a fourth feature matrix; the unfolding layer expands each value of the fourth feature matrix to a one-dimensional long vector; and the full connection layer connects the one-dimensional long vector to the fusion layer. In summary, the deep block extracts the quasi-static features in the original input three-dimensional matrix and transmits them to the fusion layer.

[0035] The time sequence block includes an input layer, a time sequence convolutional layer and a full connection layer; wherein the input layer receives time sequence information and transmits it to the time sequence convolutional layer; the time sequence convolutional layer uses a strong time sequence feature extraction network to perform convolution calculation thereon to extract time sequence change features therein; and the full connection layer integrates the features into a one-dimensional long vector to be transmitted to the fusion layer.

[0036] The fusion layer comprises a full connection layer and a classifier. The fusion layer fully connects the features of the deep block and the time sequence block, integrates the features through two-dimensional convolution operation of several full connection layers, and classifies the connected features by using the classifier.

[0037] Another object of the present application is to provide a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the computer program is executed by the processor to enable the processor to execute the rotating machinery fault discrimination method based on the time sequence deep fusion network.

[0038] Another object of the present application is to provide a computer readable storage medium storing a computer program, and the computer program is executed by a processor to enable the processor to execute the rotating machinery fault discrimination method based on the time sequence deep fusion network.

[0039] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by the present application are analyzed from the following aspects:

[0040] Firstly, in view of the technical problems existing in the prior art and the difficulty in solving the problems, the technical solution to be protected by the present application and the results and data in the research and development process are combined to analyze in detail and profoundly how the technical solution solves the technical problems and brings some creative technical effects after solving the problems. The specific description is as follows:

[0041] The rotating machinery fault discrimination method based on the time sequence deep fusion network adopts a time-frequency domain analysis method and a sliding translation window method to process the vibration signal of the rotating machinery, so as to obtain a processing result corresponding to the vibration signal, and a trained convolutional neural network is used to analyze the fault features, so as to obtain the fault diagnosis of the rotating machinery, which is helpful to realize real-time intelligent fault diagnosis of the rotating machinery.

[0042] Secondly, from the perspective of the product or as a whole, the technical solution to be protected by the present application has the technical effects and advantages, which are described in detail as follows:

[0043] The present technology is suitable for rotating machinery fault diagnosis supported by sliding bearings, and automatically extracts the features of the fault by using a deep time sequence fusion algorithm, and automatically diagnoses the fault category. The technology can automatically extract the quasi-static and time sequence features in the running signal, and the diagnosis effect is better than that of the traditional deep learning algorithm.

[0044] Thirdly, as the creative auxiliary evidence of the claims of the present application, it is also embodied in the following important aspects:

[0045] The expected income and commercial value of the technical scheme of the application are: suitable for rotating machinery supported by sliding bearings. The fault diagnosis of rotating machinery such as steam turbines provides help and guidance for equipment management and operation personnel.

[0046] The technical scheme of the application fills the technical gap in the industry at home and abroad: for the first time, the algorithm is used in the deep learning algorithm to extract the quasi-static and time sequence features of the vibration signal, and the two features are fused, thereby improving the accuracy of fault diagnosis.

[0047] The technical scheme of the application solves the technical problems that people have been eager to solve but have always failed to succeed: for the first time, the algorithm is used in the deep learning algorithm to extract the quasi-static and time sequence features of the vibration signal, and the two features are fused, thereby improving the accuracy of fault diagnosis.

[0048] The technical scheme of the application overcomes the technical bias: for the first time, the algorithm is used in the deep learning algorithm to extract the quasi-static and time sequence features of the vibration signal, and the two features are fused, thereby improving the accuracy of fault diagnosis. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 It is a flow chart of the rotating machinery fault discrimination method based on the time sequence deep fusion network provided by the embodiment of the application;

[0050] Figure 2 It is a flow chart of obtaining the short-time-time frequency domain information matrix of the vibration signal provided by the embodiment of the application;

[0051] Figure 3 It is a flow chart of obtaining the time sequence change information vector of the vibration signal provided by the embodiment of the application;

[0052] Figure 4 It is a pre-training flow chart of the convolutional neural network model provided by the embodiment of the application;

[0053] Figure 5 It is a convolutional neural network structure diagram provided by the embodiment of the application;

[0054] Figure 6 It is a schematic diagram of the measuring point position and sensor arrangement provided by the embodiment of the application;

[0055] Figure 7 It is a trend diagram of the measuring point vibration frequency and power frequency provided by the embodiment of the application;

[0056] Figure 8 It is a vibration schematic diagram of the measuring point provided by the embodiment of the application;

[0057] Figure 9 It is a vibration diagnosis result diagram of the measuring point provided by the embodiment of the application;

[0058] Figure 10 is the shaft support vibration measuring point arrangement provided by the embodiment of the present application;

[0059] Figure 11 is the vibration trend monitoring result and graph provided by the embodiment of the present application;

[0060] Figure 12 is the vibration waveform graph provided by the embodiment of the present application;

[0061] Figure 13 is the vibration intelligent diagnosis result graph provided by the embodiment of the present application;

[0062] In the figure: 1, first convolution layer; 2, first pooling layer; 3, second convolution layer; 4, second pooling layer, 5, unfolding layer; 6, full connection layer; 7, input layer; 8, time series convolution layer; 9, first full connection layer; 10, second full connection layer; 11, classifier. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below with examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0064] I. Explanation of the embodiment. In order for those skilled in the art to fully understand how the present application is implemented, this part is an explanation of the embodiment of the technical scheme of the claims.

[0065] The rotating machinery fault diagnosis and identification method based on time series deep fusion network provided by the embodiment of the present application adopts time-frequency domain analysis method and sliding translation method to process the vibration signal of the rotating machinery, and adopts the trained time series deep fusion network model to perform fault diagnosis analysis on the processing result, and further obtains the fault diagnosis of the rotating machinery, to assist in realizing real-time intelligent diagnosis of rotating machinery fault.

[0066] Embodiment 1

[0067] As Figure 1 described, the rotating machinery fault diagnosis and identification method based on time series deep fusion network provided by the embodiment of the present application includes the following steps:

[0068] S101, collecting the vibration signal of the rotating machinery (sliding bearing without gear box and belt pulley).

[0069] In the embodiment of the present application, a vibration displacement sensor is arranged on the rotating machinery, and the vibration displacement sensor collects the vibration signal of the rotating machinery in a predetermined sampling manner during the movement of the rotating machinery.

[0070] The vibration displacement sensor is arranged along the radial direction of the shaft of the rotating machine to detect the vibration of the rotating machine along the radial direction. In this embodiment, the vibration displacement sensor is an eddy current displacement sensor arranged along the radial direction of the shaft.

[0071] The sampling mode includes a sampling length and a sampling frequency. In this embodiment, synchronous full-cycle sampling is used to reduce the influence of speed variation on the sampling set, n is 8, and k is 4. The sampling frequency of synchronous full-cycle sampling is 256 times the rotating speed of the rotating machine, and 16 cycles of vibration signals are collected each time.

[0072] S102, wavelet transform is performed on the collected vibration signal according to the input form of the pre-trained time series deep fusion network model, and information matrices are combined to obtain a short-time-time-frequency domain information matrix of the vibration signal.

[0073] S103, the collected vibration signal is subjected to short-time Fourier transform according to the input form of the pre-trained time series deep fusion network model, the amplitude and phase of the characteristic frequency are extracted, and a time series change information vector is obtained by means of sliding translation.

[0074] S104, the vibration comprehensive information matrix is input into the pre-trained time series deep fusion network model, and the network model performs diagnostic analysis on the received vibration information matrix and vector to obtain a rotating machine fault diagnosis result.

[0075] Embodiment 2

[0076] Based on the rotating machine fault diagnosis and recognition method based on the time series deep fusion network provided in Embodiment 1 of the present application, preferably, in step S101, the vibration signal is collected by the vibration displacement sensor arranged on the rotating machine, which is suitable for a sliding bearing without a gear box and a belt pulley.

[0077] The vibration displacement sensor is arranged horizontally or vertically along the rotating machine.

[0078] The vibration displacement sensor collects the vibration signal of the rotating machine in a predetermined sampling mode;

[0079] The predetermined sampling mode is synchronous full-cycle sampling, and the sampling frequency is 2 n times the rotating speed of the rotating machine, n is a positive integer greater than or equal to 6; the sampling time length is 2k rotation periods, and k is a positive integer greater than or equal to 3.

[0080] The sampling frequency is 2 nn is a positive integer greater than or equal to 6; the sampling length is 2 k rotational periods, k is a positive integer greater than or equal to 3.

[0081] Embodiment 3

[0082] Based on the rotating machinery fault diagnosis and recognition method based on the timing deep fusion network provided in Embodiment 1 of the present application, preferably, as shown in Figure 2 the specific steps of obtaining the short-time-time-frequency domain information matrix of the vibration signal in step S102 include:

[0083] S201, the vibration signal is cut into a one-dimensional array with a signal length of 2 n n is a positive integer greater than or equal to 6.

[0084] S202, wavelet transform is performed on the one-dimensional array to extract the vibration signal with a frequency within 8 times the rotational speed; the one-dimensional array is expanded into a 2 k ×2 n matrix with a size of 2 x k is a positive integer greater than or equal to 3.

[0085] Embodiment 4

[0086] Based on the rotating machinery fault diagnosis and recognition method based on the timing deep fusion network provided in Embodiment 1 of the present application, preferably, as shown in Figure 3 the specific steps of obtaining the time sequence change information vector of the vibration signal in step S102 include:

[0087] S301, the vibration signal is sequentially cut into 2 k row vectors with a length of 2 n , respectively denoted as k is a positive integer greater than or equal to 3, and n is a positive integer greater than or equal to 6.

[0088] S302, Fourier transform is performed on the 2 k vectors to obtain the vibration signal with a frequency within 8 times the rotational speed.

[0089] S303, the amplitude and phase of the vibration signal of the characteristic frequency are extracted to obtain two time sequence vectors with a length of 2 k .

[0090] S304, the sliding translation method is used to obtain 2m time sequence vectors , where p m is the mth time sequence vector.

[0091] Embodiment 5

[0092] Based on the rotating machinery fault diagnosis recognition method based on the time sequence deep fusion network provided in Embodiment 1 of the present application, preferably, as shown in Figure 4 As shown in the figure, in step S104, the pre-training process of the convolutional neural network model includes the following steps:

[0093] S401, performing a fault experiment on the rotating machinery, collecting vibration displacement signals for different fault experiments respectively, and obtaining a plurality of groups of fault signals under different fault conditions.

[0094] S402, performing frequency domain and time domain information analysis on the fault signals under a plurality of fault conditions to obtain a short-time-time-frequency domain information matrix and a time sequence change information vector as inputs of the convolutional neural network model.

[0095] S403, according to the occurrence of each fault diagnosis in the fault experiment, an output matrix corresponding to the occurrence is constructed.

[0096] S404, setting the parameters of the convolutional neural network model, inputting the input and output matrices into the time sequence deep fusion network model for training, and obtaining a pre-trained fusion neural network model for rotating machinery fault diagnosis recognition.

[0097] In Embodiment 5 of the present application, the step of performing frequency domain information analysis on the fault signal is the same as the above steps S102 and S103.

[0098] In Embodiment 5 of the present application, the faults include rotor imbalance, rotor misalignment, bearing seat loosening, dynamic and static rubbing, airflow excitation and the like. Any fault type with historical data can be used as a diagnosis result of fault diagnosis.

[0099] Embodiment 6

[0100] Based on the rotating machinery fault diagnosis recognition method based on the time sequence deep fusion network provided in Embodiment 1 of the present application, as shown in Figure 5 The fusion network model provided in the present application adopts a convolutional neural network model, which includes a deep block, a time sequence block and a fusion block.

[0101] The deep block includes a first convolutional layer 1, a first pooling layer 2, a second convolutional layer 3, a second pooling layer 4, an unfolding layer 5, and a full connection layer 6; the time sequence block includes an input layer 7, a time sequence convolutional layer 8, and a first full connection layer 9; the fusion layer includes a second full connection layer 10 and a classifier 11,

[0102] Each classifier judges the occurrence of a fault. Among them, the first convolution layer 1 performs matrix convolution operation on the original three-dimensional matrix, extracts matrix numerical features under different convolution kernel dimensions, obtains a first feature matrix, and transmits the features to the first pooling layer; the first pooling layer 2 performs pooling operation processing on the first feature matrix, reduces the scale of the feature matrix, obtains a second feature matrix, and transmits it to the second convolution layer; the second convolution layer 3 performs second matrix convolution operation on the second feature matrix, extracts matrix feature numerical features under different convolution kernel dimensions again, obtains a third feature matrix; the second pooling layer 4 performs pooling operation on the third feature matrix again, reduces the scale of the feature matrix again, obtains a fourth feature matrix; the unfolding layer 5 performs unfolding operation on each value of the fourth feature matrix, and changes it into a one-dimensional long vector; the fully connected layer 6 connects the one-dimensional long vector to the fusion layer. In summary, the role of the depth block is to extract the quasi-static features in the original input three-dimensional matrix and transmit them to the fusion layer. The input layer 7 receives time series information and transmits it to the time series convolution layer; the time series convolution layer 8 uses a strong time series feature extraction network to perform convolution calculation thereon to extract time series change features therein; the first fully connected layer 9 integrates the features into a one-dimensional long vector to be transmitted to the fusion layer. The fusion layer fully connects the features of the depth block and the time series block, and then integrates the features through several two-dimensional convolution operations of the second fully connected layer 10, and then uses the classifier 11 to classify the connected features.

[0103] Embodiment 7

[0104] Based on the fusion network provided in Embodiment 7 of the present application, in the present application, a rotor fault simulation test bench and industrial rotor operation data are used as training data of a convolutional neural network model, a plurality of groups of speed-up tests are performed on five faults (rotor imbalance, rotor misalignment, bearing seat loosening, dynamic and static rubbing, and airflow excitation) and one normal condition to obtain a large amount of fault data; the network is used for clustering training to obtain a model network for diagnosing each type of fault.

[0105] The rotating machinery fault discrimination method based on the time series depth fusion network provided by the present application adopts a time-frequency domain analysis method and a sliding window to process the vibration signal of the rotating machinery, and adopts a trained network model to perform fault diagnosis analysis on the processing result, and further obtains the fault category of the rotating machinery, to assist in realizing real-time intelligent diagnosis of rotating machinery faults.

[0106] II. Application Examples. In order to prove the creativity and technical value of the technical solutions of the present application, this part is an application example of the technical solutions of the claims on a specific product or related technology.

[0107] Application Example One:

[0108] (1) Measurement point arrangement

[0109] The unit is a supercritical, single-shaft, four-cylinder, four-steam, double-backpressure, eight-stage regenerative extraction, reaction steam turbine N1050-26.25 / 600 / 600 designed and manufactured by Shanghai Turbine Co., Ltd. and Germany Siemens AG for Jiangsu Changshu Power Generation Co., Ltd. 2x1050MW Unit Phase II No. 5 steam turbine. The designed rated output power of the unit is 1050MW.

[0110] The positions of the measuring points and the arrangement of the sensors are shown in Figure 6 Fig. 1. The relative shaft vibration and bearing vibration sensors were installed at 1-8 bearing positions of No. 5 unit. The vibration measuring points were synchronously and integrally sampled. The working frequency of the steam turbine was 50Hz, and the sampling length was 1024 points. The data source: the DSPi 408 data sampling and analysis device produced by Bentley was used for measurement. The measuring device was calibrated by the measurement unit approved by the State Bureau of Metrology, and met the relevant national regulations. The rotor shaft vibration and bearing seat vibration evaluation standards met GB / T11348.2-2007 and GB / T6075.2-2002.

[0111] (2) Monitoring results

[0112] 1) Vibration trend chart

[0113] From the trend of the vibration frequency and the working frequency of the measuring points Figure 7 , it can be seen that the vibration of each measuring point of the steam turbine is mostly within the reasonable range. When the speed is constant, the vibration value of the 6# bearing data point has a tendency to increase with the running time. And the vibration value of 5#7#8# bearing is less than 60um in the vibration trend chart at constant speed, and the vibration is stable without climbing trend, while the vibration value of 6# bearing is greater than 90um, with a climbing trend. Among them, Figure 7 (a) Data source of collection software: 5X bearing displacement vibration trend chart of Jiangsu Changshu Power Generation Co., Ltd. 2x1050MW Unit Phase II No. 5 steam turbine by Bentley 408 software; Figure 7 (b) Data source of collection software: 6X bearing displacement vibration trend chart of Jiangsu Changshu Power Generation Co., Ltd. 2x1050MW Unit Phase II No. 5 steam turbine by Bentley 408 software; Figure 7 (c) Data source of collection software: 7X bearing displacement vibration trend chart of Jiangsu Changshu Power Generation Co., Ltd. 2x1050MW Unit Phase II No. 5 steam turbine by Bentley 408 software Figure 7 (d) Data source of collection software: 8X bearing displacement vibration trend chart of Jiangsu Changshu Power Generation Co., Ltd. 2x1050MW Unit Phase II No. 5 steam turbine by Bentley 408 software.

[0114] 2) Vibration waveform bode chart of each point

[0115] To further determine the vibration of the unit, select the slightly larger 5#6#7#8# measuring point to show. From the bode diagram of the measuring point vibration, 5#7#8# signal is relatively stable, from the figure, the amplitude is not large during the speed process, 7# has the phenomenon of vibration increasing at the critical speed, but it is still within the reasonable range. 6# vibration increases continuously during the process of increasing the speed to the rated speed, and still has the increasing trend at the rated speed. Figure 8 From the bode diagram of the measuring point vibration, 5#7#8# signal is relatively stable, from the figure, the amplitude is not large during the speed process, 7# has the phenomenon of vibration increasing at the critical speed, but it is still within the reasonable range. 6# vibration increases continuously during the process of increasing the speed to the rated speed, and still has the increasing trend at the rated speed. Figure 8 (a) The data source of the software is collected by Bentley 408 software: Jiangsu Changshu Power Co., Ltd. 2x1050MW unit 5# steam turbine 5X Bode diagram; Figure 8 (b) The data source of the software is collected by Bentley 408 software: Jiangsu Changshu Power Co., Ltd. 2x1050MW unit 5# steam turbine 6X Bode diagram;

[0116] Figure 8 (c) The data source of the software is collected by Bentley 408 software: Jiangsu Changshu Power Co., Ltd. 2x1050MW unit 5# steam turbine 7X Bode diagram; Figure 8 (d) The data source of the software is collected by Bentley 408 software: Jiangsu Changshu Power Co., Ltd. 2x1050MW unit 5# steam turbine 8X Bode diagram.

[0117] (3) Intelligent diagnosis result

[0118] The slightly larger 6# wave original waveform data is input into the algorithm to diagnose the waveform, and the corresponding result is automatically diagnosed. The diagnosis result of 6# wave is not stable, and there is a rubbing phenomenon. In the order diagram, it can be found that the vibration is mainly at 1X frequency, and the value is large. In the frequency spectrum diagram, it can be obviously observed that the vibration climbing has a certain slope. For example Figure 9 The diagnosis results of 5#6# measuring point vibration are shown.

[0119] Application example two:

[0120] (1) Measuring point arrangement

[0121] The unit is a supercritical, intermediate reheating, four-cylinder, four-steam, single-shaft condensing steam turbine unit produced by Shanghai Steam Turbine and Generator Group Co., Ltd. The steam turbine model is N600-24.2 / 566 / 566, and the generator is a water-hydrogen-hydrogen generator with a model of QFS-600-2. The steam turbine is six supports, the generator and exciter are three supports, and the whole shafting is supported by nine radial bearings. The shafting support diagram is shown in Figure 10 The vibration measuring point arrangement diagram is shown.

[0122] The relative shaft vibration monitoring devices of 1#-7# turbine are installed randomly in the unit, and the shaft vibration monitoring device adopts the 3500 system of BENTLY company. When testing, the DAIU-408 data acquisition analyzer is connected to the output signal of 1#-7# turbine relative shaft vibration from the vibration monitoring device to monitor and analyze the vibration.

[0123] The measurement adopts the DAIU-408 data acquisition analyzer produced by BENTLY company, and the measurement device is calibrated by the measurement unit approved by the State Bureau of Metrology, which meets the relevant national regulations.

[0124] The rotor shaft vibration and bearing seat vibration evaluation standard conforms to GB / T11348.2-2012 and GB / T6075.2-2002.

[0125] (2) Monitoring results

[0126] 1) The vibration trend chart is shown in Figure 11 . Figure 11 (a) The data source of the collection software is BENTLY 408 software: displacement vibration trend chart of 1X turbine of #8 unit of Guohua Taicang Power Co., Ltd.; Figure 11 (b) The data source of the collection software is BENTLY 408 software: displacement vibration trend chart of 1Y turbine of #8 unit of Guohua Taicang Power Co., Ltd.

[0127] 2) The vibration waveform bode chart is shown in Figure 12 . Figure 12 (a) The data source of the collection software is BENTLY 408 software: Bode chart of 1X turbine of #8 unit of Guohua Taicang Power Co., Ltd.; Figure 12 (b) The data source of the collection software is BENTLY 408 software: Bode chart of 1Y turbine of #8 unit of Guohua Taicang Power Co., Ltd.

[0128] The shaft vibration of 1# turbine is large during the high load period of the unit, and the vibration fluctuation is large. The shaft vibration of 1# turbine is large during the operation of the unit, mainly because the 1# turbine has large frequency vibration, mainly because the high and medium pressure rotor has large two-order mass imbalance. The vibration fluctuation of 1# turbine is large, mainly because the sequential valve operation of the unit produces steam excitation.

[0129] (3) Intelligent diagnosis results

[0130] The original waveform data of each measuring point of vibration is input into the intelligent diagnosis system to diagnose the waveform, and the corresponding results are obtained automatically: the vibration of 1# turbine is large, the 0.5 frequency amplitude jumps constantly and has strong fluctuation, and the repeatability is good under the current load, which verifies the diagnosis results, and the results are shown in Figure 13 .

[0131] It should be noted that embodiments of the present application can be realized by hardware, software, or a combination of software and hardware. The hardware portion can be realized by a special logic; the software portion can be stored in a memory and executed by a proper instruction execution system, such as a microprocessor or a specially designed hardware. A person of ordinary skill in the art can understand that the above-mentioned apparatus and method can be realized by computer executable instructions and / or included in processor control codes, such as a carrier medium, such as a magnetic disk, CD or DVD-ROM, a programmable memory, such as a read-only memory (firmware), or a data carrier, such as an optical or electronic signal carrier. The apparatus of the present application and its modules can be realized by a hardware circuit, such as a very large scale integrated circuit or a gate array, a semiconductor, such as a logic chip, a transistor, or a programmable hardware device, such as a field programmable gate array, a programmable logic device, or the like, by software executed by various types of processors, or by a combination of the above-mentioned hardware circuit and software, such as firmware.

[0132] The above description is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any modification, equivalent replacement, and improvement within the technical range disclosed by the present application, and within the spirit and principle of the present application, should be included in the protection scope of the present application.

Claims

1. A rotating machinery fault discrimination method based on a time-series deep fusion network, characterized in that, The rotating machinery fault discrimination method based on the time-series deep fusion network comprises: The time-frequency domain analysis method and the sliding translation window method are used to process the vibration signal of the rotating machinery, to obtain a processing result corresponding to the vibration signal, and a trained convolutional neural network is used to analyze fault features in the processing result, to obtain fault information of the rotating machinery. The time-frequency domain analysis method and the sliding translation window method are used to process the vibration signal of the rotating machinery, to obtain a processing result corresponding to the vibration signal, and a trained convolutional neural network is used to analyze fault features in the processing result, to obtain fault information of the rotating machinery. The time-frequency domain analysis method and the sliding translation window method are used to process the vibration signal of the rotating machinery, to obtain a processing result corresponding to the vibration signal, and a trained convolutional neural network is used to analyze fault features in the processing result, to obtain fault information of the rotating machinery. The vibration signal collected is subjected to wavelet transform according to an input form of a time-series deep fusion network model trained in advance, and information matrices are combined to obtain a short-time-time-frequency domain information matrix of the vibration signal. The vibration signal collected is subjected to short-time Fourier transform according to an input form of a time-series deep fusion network model trained in advance, and amplitude and phase of a characteristic frequency are extracted, and a time-series change information vector is obtained by using a sliding translation method. The vibration signal is intercepted to a one-dimensional array with a signal length of 2 n n is a positive integer greater than or equal to 6; The one-dimensional array is wavelet-transformed to extract a vibration signal with a frequency within 8 times of a frequency of a rotation speed; the one-dimensional array is extended to a 2 k ×2 n dimensional matrix as a short-time information matrix F of the vibration signal x , k is a positive integer greater than or equal to 3. The time-frequency domain analysis method and the sliding translation window method are used to process the vibration signal of the rotating machinery, to obtain a processing result corresponding to the vibration signal, and a trained convolutional neural network is used to analyze fault features in the processing result, to obtain fault information of the rotating machinery. The vibration signal is sequentially intercepted as two k row vectors with a length of 2 n in time sequence, respectively denoted as ; k is a positive integer greater than or equal to 3, and n is a positive integer greater than or equal to 6; For these 2 k vectors Fourier transform is done to get the frequency vibration signal; The amplitude and phase of the vibration signal of the characteristic frequency are extracted to obtain two time sequence vectors with a length of 2 k . Again, the sliding translation method is used to obtain 2n time series vectors ; The time-frequency domain analysis method and the sliding translation window method are used to process the vibration signal of the rotating machinery, to obtain a processing result corresponding to the vibration signal, and a trained convolutional neural network is used to analyze fault features in the processing result, to obtain fault information of the rotating machinery. The convolutional neural network comprises a deep block, a time-series block and a fusion block. The deep block comprises a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, an unfolding layer and a full connection layer. The time-series block comprises an input layer, a time-series convolutional layer and a full connection layer. The fusion layer comprises a full connection layer and a classifier. 2.The rotating machinery fault diagnosis method based on the time-series deep fusion network according to claim 1, characterized in that, The first convolutional layer performs matrix convolution operation on the original three-dimensional matrix, extracts matrix numerical features in different convolution kernel dimensions, obtains a first feature matrix, and transmits the features to the first pooling layer; the first pooling layer performs pooling operation on the first feature matrix, reduces the size of the feature matrix, obtains a second feature matrix, and transmits the second feature matrix to the second convolutional layer; the second convolutional layer performs second matrix convolution operation on the second feature matrix, extracts matrix feature numerical features in different convolution kernel dimensions again, obtains a third feature matrix; the second pooling layer performs pooling operation on the third feature matrix again, reduces the size of the feature matrix again, obtains a fourth feature matrix; the unfolding layer unfolds each value of the fourth feature matrix, and changes the value into a one-dimensional long vector; and the full connection layer connects the one-dimensional long vector to the fusion layer.

3. The rotating machinery fault diagnosis method based on the time-series deep fusion network according to claim 2, characterized in that, Before the time-frequency domain analysis method and the sliding translation window method are used to process the vibration signal of the rotating machinery, the vibration signal of the rotating machinery is collected; a vibration displacement sensor is arranged on the rotating machinery, and the vibration displacement sensor collects the vibration signal of the rotating machinery in a predetermined sampling mode during movement of the rotating machinery. The vibration displacement sensor is arranged in a horizontal or vertical direction along a radial direction of a shaft of the rotating machinery, to detect vibration of the rotating machinery along the radial direction; and the vibration displacement sensor is an eddy current displacement sensor, and is arranged in the horizontal and vertical directions along the radial direction of the shaft. The sampling mode comprises a sampling length and a sampling frequency, and synchronous whole-cycle sampling is used, The sampling frequency is 2 n times of the rotating speed of the rotating machine, n is a positive integer greater than or equal to 6; and the sampling time length is 2k rotating periods, k is a positive integer greater than or equal to 3. The sampling frequency is 2 n times the rotating speed of the rotating machine, n is a positive integer greater than or equal to 6, the sampling length is 2 k rotating periods, and k is a positive integer greater than or equal to 3.

4. The rotating machinery fault diagnosis method based on the timing depth fusion network according to claim 1, characterized in that, The trained convolutional neural network is used to analyze the fault features in the processing result, and fault information of the rotating machinery is obtained. The vibration comprehensive information matrix is input into a pre-trained time series deep fusion network model, the network model performs diagnostic analysis on the received vibration information matrix and vector, and a fault diagnosis result of the rotating machinery is obtained. The pre-training process of the convolutional neural network model includes the following steps: Fault experiments are performed on the rotating machinery, and vibration displacement signals are collected for different fault experiments, so as to obtain fault signals under different fault conditions; The fault signals under different fault conditions are analyzed in the frequency domain and the time domain to obtain a short-time-time-frequency domain information matrix and a time series change information vector, which are used as inputs of the convolutional neural network model; According to the occurrence of each fault diagnosis in the fault experiment, a corresponding output matrix is constructed; The parameters of the convolutional neural network model are set, the input and output matrices are input into the time series deep fusion network model for training, and a pre-trained fusion neural network model for fault diagnosis and identification of the rotating machinery is obtained.

5. A computer device, comprising: The computer device includes a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to make the processor execute the rotating machinery fault discrimination method based on the time series deep fusion network in any one of claims 1-4.

6. A computer readable storage medium storing a computer program, the computer program being executed by a processor to make the processor execute the rotating machinery fault discrimination method based on the time series deep fusion network in any one of claims 1-4.

Images