Rotating machine fault diagnosis method and device

Abstract

The invention relates to a rotating machinery fault diagnosis method and device, and the method comprises the steps: obtaining original sample data which comprises acceleration vibration signal simulation data in a plurality of different working states; calculating an axis track according to the original sample data; inputting a one-dimensional array based on a time sequence in the axis track as an input feature vector into a long short-term memory network for model training, so as to obtain a rotating machinery fault diagnosis model; and performing fault diagnosis on a to-be-detected rotating machine based on the rotating machine fault diagnosis model, so as to improve the fault recognition capability of the rotating machine fault diagnosis model.

Description

Technical Field

[0001] The present invention relates to the field of neural network technology, and in particular to a rotating machinery fault diagnosis method and device. Background Art

[0002] When rotating machinery is rotating, mechanical faults can cause abnormal vibrations. Since vibration signals are easy to collect and understand, they are usually used as important parameter information for fault warning and diagnosis of rotating machinery. In fault analysis based on vibration signals, the axis trajectory of the rotating machinery rotor is a very important graphic feature, which usually contains two signals in perpendicular directions and has very obvious observability. With the development of information technology, the method of extracting graphic features from the axis trajectory to construct a classification model has gradually been adopted to realize the fault diagnosis of rotating machinery.

[0003] At present, in the process of building a classification model for diagnosing rotating machinery faults, the classification model is usually trained using feature vectors such as the center of gravity coordinates, central moments, HU invariant moments, and wavelet invariant moments extracted from the axis trajectory. There are usually two existing feature extraction methods: the first method is to save the axis trajectory data as an image first, and then use the saved image as the input feature vector of the classification model, but this method has low extraction efficiency and requires additional storage space for image storage; the second method is to first convert the axis trajectory data into a two-dimensional matrix, and then extract the invariant moment features based on the two-dimensional matrix, and use the extracted invariant moment features as the input feature vector. Although this method does not require additional storage space for image storage, the two-dimensional matrix conversion will be affected by the row and column spacing. If the row and column spacing is too large, the graphic features will be lost. If the row and column spacing is too small, too much noise will be introduced to cause classification interference. Therefore, in the existing feature extraction methods, both the first method and the second method will have different degrees of influence on the input feature vector of the classification model, thereby affecting the normal diagnosis of rotating machinery faults by the classification model. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a rotating machinery fault diagnosis method and device to improve the fault identification capability of a rotating machinery fault diagnosis model.

[0005] The technical solution adopted by the present invention to solve the technical problem is: to provide a rotating machinery fault diagnosis method, comprising:

[0006] S101, obtaining original sample data, wherein the original sample data includes acceleration vibration signal simulation data under multiple different working states;

[0007] S102, calculating the axis trajectory according to the original sample data;

[0008] S103, inputting the one-dimensional array based on the time series in the axis trajectory as an input feature vector into a long short-term memory network for model training to obtain a rotating machinery fault diagnosis model;

[0009] S104: Perform fault diagnosis on the rotating machinery to be tested based on the rotating machinery fault diagnosis model.

[0010] Optionally, the step S102 includes:

[0011] Performing a fast Fourier transform on the original sample data to convert the original sample data into a frequency domain distribution;

[0012] Determining a current frequency conversion according to a preset frequency interval of the frequency domain distribution;

[0013] Calculating the frequency conversion frequencies corresponding to different multiples within a preset multiple interval according to the current frequency conversion;

[0014] Calculate the vibration displacement integral result according to the rotation frequencies corresponding to different frequency multiples and the phase angles and amplitudes corresponding to the rotation frequencies;

[0015] Based on the vibration displacement integration result, an axis center trajectory is drawn on a two-dimensional coordinate axis.

[0016] Optionally, determining the current frequency conversion according to a preset frequency interval of the frequency domain distribution includes:

[0017] Calculate the frequency resolution of the frequency domain distribution according to the sampling frequency and the number of sampling points corresponding to the original sample data;

[0018] Determine a preset frequency interval according to the rated rotation frequency, the synchronous rotation frequency and the frequency resolution corresponding to the original sample data;

[0019] The maximum amplitude of the frequency domain distribution within the preset frequency interval is selected, and the frequency corresponding to the maximum amplitude is determined as the current rotation frequency.

[0020] Optionally, the calculation formula of the vibration displacement integral result is:

[0021]

[0022] Where n is the frequency multiplication number, f n is the frequency conversion frequency corresponding to the n-fold frequency, α is the difference in phase angles in different directions, and is the displacement amplitude in different directions.

[0023] Optionally, the calculation formula of the displacement amplitude is:

[0024]

[0025] Among them A n x and A n y The frequency f n The corresponding acceleration amplitude, c is the unit transformation constant.

[0026] Optionally, the step S103 includes:

[0027] Decomposing the axis trajectory within the same time length into two one-dimensional arrays in different directions;

[0028] The two one-dimensional arrays in different directions are concatenated to obtain an input feature vector;

[0029] The input feature vector is input into a long short-term memory network for model training, and the trained long short-term memory network is determined to be a rotating machinery fault diagnosis model.

[0030] Optionally, after step S101, the rotating machinery fault diagnosis method further includes:

[0031] Preprocessing the original sample data;

[0032] The step S102 includes:

[0033] The axis trajectory is calculated based on the preprocessed original sample data.

[0034] Optionally, the preprocessing of the original sample data includes:

[0035] Setting a movable window in the original sample data;

[0036] The sub-data framed by the movable window on the preset track are intercepted, and a set of all the sub-data is determined as the pre-processed original sample data.

[0037] Optionally, preprocessing the original sample data includes:

[0038] Gaussian noise is added to the original sample data to expand the original sample data.

[0039] The present invention also provides a rotating machinery fault diagnosis device, comprising:

[0040] An acquisition unit, used to acquire original sample data, wherein the original sample data includes acceleration vibration signal simulation data under multiple different working states;

[0041] A calculation unit, used for calculating the axis trajectory according to the original sample data;

[0042] A model training unit, used for inputting the one-dimensional array based on the time series in the axis trajectory as an input feature vector into the long short-term memory network for model training to obtain a rotating machinery fault diagnosis model;

[0043] The fault diagnosis unit is used to perform fault diagnosis on the rotating machinery to be tested based on the rotating machinery fault diagnosis model.

[0044] The implementation of the present invention has the following beneficial effects:

[0045] First, the original sample data is obtained, and the original sample data includes acceleration vibration signal simulation data under multiple different working conditions; then the axis trajectory is calculated according to the original sample data; then the one-dimensional array based on the time series in the axis trajectory is input as the input feature vector into the long short-term memory network for model training to obtain the rotating machinery fault diagnosis model; finally, the rotating machinery to be tested is diagnosed based on the rotating machinery fault diagnosis model. In this way, the axis trajectory can be converted into a one-dimensional array with a time series relationship, and the one-dimensional array is input into the long short-term memory network in the form of a time series for model training, and the axis trajectory does not need to be converted into an image or a two-dimensional matrix, thereby reducing the negative effects brought about by the image conversion or the two-dimensional matrix conversion process. And because the long short-term memory network has the characteristics of processing long-term dependence and anti-interference for the analysis and prediction of time series, the one-dimensional array with a time series relationship is used to train the rotating machinery fault diagnosis model, so as to improve the fault recognition ability of the rotating machinery fault diagnosis model. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0047] Figure 1 It is a schematic diagram of an embodiment of a rotating machinery fault diagnosis method of the present invention;

[0048] Figure 2 is a schematic diagram of another embodiment of the rotating machinery fault diagnosis method of the present invention;

[0049] Figure 3 It is a schematic diagram of an embodiment of a rotating machinery fault diagnosis device of the present invention;

[0050] Figure 4 It is a schematic diagram of the process of preprocessing the original sample data using the sliding window method in the present invention;

[0051] Figure 5 This is a schematic diagram of the axis trajectory of the present invention in a normal working state;

[0052] Figure 6It is a schematic diagram of a one-dimensional array change waveform in a normal working state of the present invention;

[0053] Figure 7 It is a schematic diagram of the axis trajectory of the horizontal misalignment working state of the present invention;

[0054] Figure 8 It is a schematic diagram of a one-dimensional array change waveform of the horizontal misalignment working state of the present invention;

[0055] Fig. 9 It is a schematic diagram of the long short-term memory network structure framework of the present invention;

[0056] Fig.10 It is a schematic diagram of the serial connection of the input feature vectors of the present invention. DETAILED DESCRIPTION

[0057] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, the specific embodiments of the present invention are now described in detail with reference to the accompanying drawings. Features defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the features. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0058] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0059] The present invention provides a rotating machinery fault diagnosis method and device, which are used to improve the fault identification capability of a rotating machinery fault diagnosis model.

[0060] The method provided by the present invention is applied to a system, a server or a terminal with logic analysis and processing capabilities for execution and implementation. The embodiment of the present invention is described by taking the system as an example. Figure 1 As shown, an embodiment of the rotating machinery fault diagnosis method of the present invention includes:

[0061] S101, obtaining original sample data, where the original sample data includes acceleration vibration signal simulation data under multiple different working states;

[0062] In this embodiment, the original sample data is acceleration vibration signal simulation data of a rotating machine collected by using an experimental simulation device, and the experimental simulation device can be a rotating machine modeled using simulation technology. The acceleration vibration signal simulation data is a pair of acceleration vibration signals collected by an acceleration vibration sensor when the rotating machine rotates in multiple different working states. The pair of acceleration vibration signals are acceleration vibration signals in two mutually perpendicular directions, such as: horizontal x direction and vertical y direction. The multiple different working states include fault states and load states, wherein the fault state can be: normal state, horizontal or vertical misalignment state, base loose state or dynamic eccentricity state, and the load state can be: no-load state, light load state, half-load state or heavy load state.

[0063] S102, calculating the axis trajectory according to the original sample data;

[0064] The axis trajectory refers to the motion trajectory of a point on the axis relative to the rotating machinery in a plane perpendicular to the axis system, which is formed by displacement signals in two mutually perpendicular directions. As an important graphical feature in the rotor vibration signal, the axis trajectory can conveniently and intuitively depict the operating status of the rotor. A specific axis trajectory graph corresponds to a specific rotating machinery vibration fault, such as Figure 5 and Figure 7 Schematic diagram of the axis trajectory shown. In this embodiment, the original sample data acquired by the system are acceleration vibration signals in two mutually perpendicular directions. Therefore, when drawing the axis trajectory, it is necessary to use fast Fourier transform to convert the original sample data into time domain and frequency domain. Specifically, the amplitude and phase angle of the rotation frequency and its multiples can be calculated first, and then brought into the cosine function through iterative addition to obtain the displacement signals in two mutually perpendicular directions, and then the axis trajectory can be drawn according to the displacement signal.

[0065] S103, inputting the one-dimensional array based on the time series in the axis trajectory as an input feature vector into the long short-term memory network for model training to obtain a rotating machinery fault diagnosis model;

[0066] The long short-term memory network is an improved version of the recursive neural network that can solve the long-term dependency problem of the original recursive neural network. It is currently widely used in the fields of time series, speech recognition, and text translation. Fig. 9 As shown in the figure, the LSTM network structure framework is constructed. Among them, for the input layer in the long short-term memory network structure framework, as Fig.10As shown in , since the axis trajectory is a two-dimensional array based on the time series, and the two-dimensional array based on the time series is composed of two one-dimensional arrays in perpendicular directions, the two-dimensional arrays within the same time length can be spliced ​​to obtain two one-dimensional arrays in different directions, and this one-dimensional array is used as the input feature vector of the input layer. Specifically, the axis trajectory in the normal working state and the one-dimensional array change waveform corresponding to the axis trajectory can be referred to Figure 5 and Figure 6 As shown; the axis trajectory of the horizontal misalignment working state and the one-dimensional array change waveform corresponding to the axis trajectory can be referred to Figure 7 and Figure 8 As shown. For example: the time length of one-dimensional array A is 0.6s, the time interval is 0.1s, and time series B is [0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6]. One-dimensional array A contains one-dimensional array a in the horizontal x direction and one-dimensional array b in the vertical y direction. The one-dimensional array a corresponding to time series B is [0.002, 0.000, -0.002], and the one-dimensional array b corresponding to time series B is [0.001, 0.003, -0.001]. For the hidden layer in the long short-term memory network structure framework, the hidden layer is constructed by two layers of LSTM units. The number of LSTM units in each layer can be set to 512, and only the last output of the hidden layer is retained. For the fully connected layer in the long short-term memory network structure framework, since the fully connected layer can realize the functions of dimension reduction, auxiliary classification decision and feature fusion, several layers of fully connected layers can be set after the hidden layer. For example, three layers of fully connected layers can be set according to the input data length, the number of hidden layer units and the number of categories, and their sizes are (512, 256), (256, 64) and (64, the number of categories) respectively. For the activation function layer in the long short-term memory network structure framework, the activation function layer can adopt the hyperbolic tangent activation function, which can enhance the network's acquisition of nonlinear relationships, so that more complex data patterns can be processed. For the classification layer in the long short-term memory network structure framework, the classification layer can adopt the Softmax function, which is one of the activation functions and is widely used in multi-classification models. After the long short-term memory network is trained using a one-dimensional array based on a time series, it can be determined that the trained LSTM model is a rotating machinery fault diagnosis model, which is used to perform fault diagnosis and prediction on rotating machinery.

[0067] S104, performing fault diagnosis on the rotating machinery to be tested based on the rotating machinery fault diagnosis model.

[0068] First, the acceleration vibration signals of a pair of mutually perpendicular directions of the rotating machinery to be tested are collected during the rotating motion, and then the axis trajectory is calculated according to the collected acceleration vibration signals, and then a one-dimensional array based on a time series is constructed from the axis trajectory, and the one-dimensional array based on the time series is input into the rotating machinery fault diagnosis model for fault diagnosis, and finally the output result of the rotating machinery fault diagnosis model is obtained, and the output result is determined to be the fault diagnosis result of the rotating machinery to be tested. The fault diagnosis result can be: the rotating machinery to be tested has a misalignment fault, or the tested machinery is operating normally without a fault, which is not limited here.

[0069] In this embodiment, the original sample data is first obtained, and the original sample data includes acceleration vibration signal simulation data under multiple different working states; then the axis trajectory is calculated according to the original sample data; then the one-dimensional array based on the time series in the axis trajectory is input as an input feature vector to the long short-term memory network for model training to obtain a rotating machinery fault diagnosis model; finally, the rotating machinery to be tested is diagnosed based on the rotating machinery fault diagnosis model. In this way, the axis trajectory can be converted into a one-dimensional array with a time series relationship, and the one-dimensional array is input into the long short-term memory network in the form of a time series for model training, and the axis trajectory does not need to be converted into an image or a two-dimensional matrix, thereby reducing the negative effects brought about by the image conversion or the two-dimensional matrix conversion process. And because the long short-term memory network has the characteristics of processing long-term dependence and anti-interference for the analysis and prediction of time series, the one-dimensional array with a time series relationship is used to train the rotating machinery fault diagnosis model, so that the fault recognition ability of the rotating machinery fault diagnosis model can be improved.

[0070] See also Figure 2 As shown, another embodiment of the rotating machinery fault diagnosis method of the present invention includes:

[0071] S201, obtaining original sample data, where the original sample data includes acceleration vibration signal simulation data under multiple different working states;

[0072] Step S201 in this embodiment is similar to the above Figure 1 Step S101 in the illustrated embodiment is similar and will not be described in detail here.

[0073] S202, preprocessing the original sample data;

[0074] Optionally, since the training of the neural network model requires a large amount of samples, in order to ensure the training accuracy of the neural network model, it is usually necessary to preprocess the training samples to increase the sample size. In this embodiment, the original sample data can be enhanced by sliding window method or noise enhancement.

[0075] Specifically, Figure 4 As shown, a movable window can be set in the original sample data; the sub-data selected by the movable window on the preset track is intercepted, and the set of all sub-data is determined to be the preprocessed original sample data. Among them, a movable window with a fixed window size can be set, and the movable window slides on the preset track according to a fixed moving step length, and the original sample data in the movable window is intercepted at each stop position of the movable window, and the original sample data in this part of the movable window is determined as sub-data. Obtaining the sub-data of all the stop positions of the movable window can achieve data enhancement of the original sample data. For example: set the window size of the movable window to 0.2s, the moving step to 0.1s, the length of the original sample data to 1s, and then the movable window is translated 0.1s each time until it reaches the end of the original sample data, and the original sample data within 0.2s is intercepted during each translation process, such as: the first intercepted sub-data is the original sample data within 0s-0.2s, the second intercepted sub-data is the original sample data within 0.1s-0.3s... The last intercepted sub-data is the original sample data within 0.8s-1.0s.

[0076] In another possible implementation, Gaussian noise can be added to the original sample data to expand the original sample data. The specific calculation formula is: i '=x i ×(1+θ),θ~N(0,σ 2 )

[0077] The x in the formula i is the original sample data at time i; the mean of θ is 0; the standard deviation is the parameter of σ normal distribution; the σ value is arbitrarily generated between (0, h), and the h value can be set manually based on historical information statistics. The original sample data can be expanded by adding Gaussian noise to improve the diversity of the original sample data.

[0078] S203, performing fast Fourier transform on the original sample data to convert the original sample data into frequency domain distribution;

[0079] S204, determining the current frequency conversion according to the preset frequency interval of the frequency domain distribution;

[0080] S205, calculating the frequency conversion frequencies corresponding to different multiples within a preset multiple interval according to the current frequency conversion;

[0081] S206, calculating the vibration displacement integral result according to the rotation frequencies corresponding to different frequency multiples and the phase angles and amplitudes corresponding to the rotation frequencies;

[0082] S207, drawing an axis trajectory on a two-dimensional coordinate axis based on the vibration displacement integration result;

[0083] Optionally, in this embodiment, fast Fourier transform calculation is first performed on two mutually perpendicular x directions and y directions in the original sample data to obtain frequency parameters in the frequency domain distribution, and phase parameters and amplitude parameters in the x and y directions.

[0084] Then, the current rotation frequency is determined according to the preset frequency interval. Specifically, the frequency resolution of the frequency domain distribution is calculated according to the sampling frequency and the number of sampling points corresponding to the original sample data; the preset frequency interval is determined according to the rated rotation frequency, synchronous rotation frequency and frequency resolution corresponding to the original sample data; the maximum amplitude of the frequency domain distribution within the preset frequency interval is selected, and the frequency corresponding to the maximum amplitude is determined as the current rotation frequency. Among them, the frequency resolution is the frequency interval on the frequency domain distribution, and the calculation formula of the frequency resolution df is: df = f s / N, where f s is the sampling frequency, and N is the number of sampling points. The preset frequency range can be set to where f rpmn is the rated frequency, f rpm The current frequency f r f r =f j , where j is the frequency of the independent variable in the frequency domain distribution and From formula A r =max(A j ) determines the maximum amplitude within the preset frequency range, and the frequency f corresponding to the maximum amplitude is j The current frequency f r .

[0085] Then, the maximum frequency multiplication number maxn is determined based on the type of rotating machinery corresponding to the original sample data. The preset frequency multiplication range can be set as The frequency corresponding to the n-fold frequency is f n =f j , where j is the frequency of the independent variable in the frequency domain distribution and From formula A n =max(A j ) determines the maximum amplitude within the preset frequency doubling interval, and the frequency f corresponding to the maximum amplitude is j The frequency f n The phase difference corresponding to the switching frequency is in is the phase angle in the x direction at the n-fold corresponding rotation frequency, is the phase angle in the y direction at the n-fold corresponding rotation frequency. The displacement amplitudes corresponding to the rotation frequency in the x and y directions are Among them A n x and An y The frequency f n The corresponding acceleration amplitude, c is the unit transformation constant.

[0086] Finally, according to the above-obtained rotation frequency, phase difference and displacement amplitude, the vibration displacement integral results in the x-direction and y-direction corresponding to the n-fold frequency can be calculated, which are:

[0087]

[0088] Since the frequency integral is affected by the frequency resolution when the acceleration vibration signal is integrated in the frequency domain, the unit conversion can be performed by using the displacement amplitude to reduce the error caused by too little data when the acceleration vibration signal is converted in the unit. In addition, this embodiment calculates the axis trajectory in the form of intervals instead of locating the specific frequency, thereby improving the performance of the vibration sensor and the adaptability of the rotating machinery working environment.

[0089] S208, decomposing the axis trajectory within the same time length into two one-dimensional arrays in different directions;

[0090] S209, concatenating two one-dimensional arrays in different directions to obtain an input feature vector;

[0091] S210, inputting the input feature vector into the long short-term memory network for model training, and determining that the trained long short-term memory network is a rotating machinery fault diagnosis model;

[0092] Optionally, in this embodiment, if Fig.10 As shown, the one-dimensional arrays of the same time length corresponding to the x-direction and y-direction in the two-dimensional array based on the time series can be concatenated to form a one-dimensional vector, and then the one-dimensional vector is input as an input feature vector into the long short-term memory network for model training. For example: within the time length of 1s, the one-dimensional array in the x-direction within the time length is [0.01, 0.04, 0.07], and the one-dimensional array in the y-direction within the time length is [0.06, 0.04, 0.02]. The one-dimensional vector obtained by concatenating in series in the form of x-direction first and y-direction last is [0.01, 0.04, 0.07, 0.06, 0.04, 0.02].

[0093] S212. Perform fault diagnosis on the rotating machinery to be tested based on the rotating machinery fault diagnosis model.

[0094] Step S212 in this embodiment is similar to the above Figure 1 Step S104 in the illustrated embodiment is similar and will not be described in detail here.

[0095] See also Figure 3 As shown, an embodiment of the rotating machinery fault diagnosis device of the present invention includes:

[0096] An acquisition unit 301 is used to acquire original sample data, where the original sample data includes acceleration vibration signal simulation data under multiple different working states;

[0097] A calculation unit 302, used for calculating the axis trajectory according to the original sample data;

[0098] A model training unit 303 is used to input the one-dimensional array based on the time series in the axis trajectory as an input feature vector into the long short-term memory network to perform model training to obtain a rotating machinery fault diagnosis model;

[0099] The fault diagnosis unit 304 is used to perform fault diagnosis on the rotating machine to be tested based on the rotating machine fault diagnosis model.

[0100] In this embodiment, the acquisition unit 301 acquires original sample data, which includes acceleration vibration signal simulation data under multiple different working states; the calculation unit 302 calculates the axis trajectory according to the original sample data; the model training unit 303 inputs the one-dimensional array based on the time series in the axis trajectory as an input feature vector into the long short-term memory network for model training to obtain a rotating machinery fault diagnosis model; the fault diagnosis unit 304 performs fault diagnosis on the rotating machinery to be tested based on the rotating machinery fault diagnosis model. In this way, the axis trajectory can be converted into a one-dimensional array with a time series relationship, and the one-dimensional array is input into the long short-term memory network in the form of a time series for model training, and the axis trajectory does not need to be converted into an image or a two-dimensional matrix, thereby reducing the negative effects brought about by the image conversion or the two-dimensional matrix conversion process. And because the long short-term memory network has the characteristics of processing long-term dependence and anti-interference for the analysis and prediction of time series, the training of the rotating machinery fault diagnosis model using the two-dimensional array with a time series relationship can improve the fault recognition ability of the rotating machinery fault diagnosis model.

[0101] It can be understood that the above embodiments only express the preferred implementation modes of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the patent scope of the present invention. It should be pointed out that, for ordinary technicians in this field, the above technical features can be freely combined without departing from the concept of the present invention, and several deformations and improvements can be made, which all belong to the protection scope of the present invention. Therefore, all equivalent changes and modifications made to the scope of the claims of the present invention should belong to the coverage of the claims of the present invention.

Claims

1. A method for diagnosing a rotating machinery fault, characterized in that: include: S101, obtaining original sample data, wherein the original sample data includes acceleration vibration signal simulation data under multiple different working states; S102, calculating the axis trajectory according to the original sample data; S103, inputting the one-dimensional array based on the time series in the axis trajectory as an input feature vector into a long short-term memory network for model training to obtain a rotating machinery fault diagnosis model; S104: Perform fault diagnosis on the rotating machinery to be tested based on the rotating machinery fault diagnosis model.

2. The rotating machinery fault diagnosis method according to claim 1, characterized in that: The step S102 includes: Performing a fast Fourier transform on the original sample data to convert the original sample data into a frequency domain distribution; Determining a current frequency conversion according to a preset frequency interval of the frequency domain distribution; Calculating the frequency conversion frequencies corresponding to different multiples within a preset multiple interval according to the current frequency conversion; Calculate the vibration displacement integral result according to the rotation frequencies corresponding to different frequency multiples and the phase angles and amplitudes corresponding to the rotation frequencies; Based on the vibration displacement integration result, an axis center trajectory is drawn on a two-dimensional coordinate axis.

3. The rotating machinery fault diagnosis method according to claim 2, characterized in that: The determining the current frequency conversion according to the preset frequency interval of the frequency domain distribution includes: Calculate the frequency resolution of the frequency domain distribution according to the sampling frequency and the number of sampling points corresponding to the original sample data; Determine a preset frequency interval according to the rated rotation frequency, the synchronous rotation frequency and the frequency resolution corresponding to the original sample data; The maximum amplitude of the frequency domain distribution within the preset frequency interval is selected, and the frequency corresponding to the maximum amplitude is determined as the current rotation frequency.

4. The rotating machinery fault diagnosis method according to claim 2, characterized in that: The calculation formula of the vibration displacement integral result is: Where n is the frequency multiplication number, f n is the frequency conversion frequency corresponding to the n-fold frequency, α is the difference in phase angles in different directions, and is the displacement amplitude in different directions.

5. The rotating machinery fault diagnosis method according to claim 4, characterized in that: The calculation formula of the displacement amplitude is: Among them A n x and A n y The frequency f n The corresponding acceleration amplitude, c is the unit transformation constant.

6. The rotating machinery fault diagnosis method according to claim 1, characterized in that: The step S103 includes: Decomposing the axis trajectory within the same time length into two one-dimensional arrays in different directions; The two one-dimensional arrays in different directions are concatenated to obtain an input feature vector; The input feature vector is input into a long short-term memory network for model training, and the trained long short-term memory network is determined to be a rotating machinery fault diagnosis model.

7. The rotating machinery fault diagnosis method according to claim 1, characterized in that: After step S101, the rotating machinery fault diagnosis method further includes: Preprocessing the original sample data; The step S102 includes: The axis trajectory is calculated based on the preprocessed original sample data.

8. The rotating machinery fault diagnosis method according to claim 7, characterized in that: The preprocessing of the original sample data comprises: Setting a movable window in the original sample data; The sub-data framed by the movable window on the preset track are intercepted, and a set of all the sub-data is determined as the pre-processed original sample data.

9. The rotating machinery fault diagnosis method according to claim 7, characterized in that: Preprocessing the original sample data includes: Gaussian noise is added to the original sample data to expand the original sample data.

10. A rotating machinery fault diagnosis device, characterized in that: include: An acquisition unit, used to acquire original sample data, wherein the original sample data includes acceleration vibration signal simulation data under multiple different working states; A calculation unit, used for calculating the axis trajectory according to the original sample data; A model training unit, used for inputting the one-dimensional array based on the time series in the axis trajectory as an input feature vector into the long short-term memory network for model training to obtain a rotating machinery fault diagnosis model; The fault diagnosis unit is used to perform fault diagnosis on the rotating machinery to be tested based on the rotating machinery fault diagnosis model.

Images