WO2023206860A1 - 一种机械设备故障的确定方法及装置 - Google Patents

一种机械设备故障的确定方法及装置 Download PDF

Info

Publication number
WO2023206860A1
WO2023206860A1 PCT/CN2022/111885 CN2022111885W WO2023206860A1 WO 2023206860 A1 WO2023206860 A1 WO 2023206860A1 CN 2022111885 W CN2022111885 W CN 2022111885W WO 2023206860 A1 WO2023206860 A1 WO 2023206860A1
Authority
WO
WIPO (PCT)
Prior art keywords
fault
joint distribution
data
time
neural network
Prior art date
Application number
PCT/CN2022/111885
Other languages
English (en)
French (fr)
Inventor
王新梦
王宗文
王红星
Original Assignee
山东瑞美油气装备技术创新中心有限公司
烟台杰瑞石油装备技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 山东瑞美油气装备技术创新中心有限公司, 烟台杰瑞石油装备技术有限公司 filed Critical 山东瑞美油气装备技术创新中心有限公司
Publication of WO2023206860A1 publication Critical patent/WO2023206860A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation

Definitions

  • the present application relates to the field of mechanical equipment detection, specifically, to a method and device for determining mechanical equipment faults.
  • Vibration signals are commonly used fault status response data for reciprocating or rotating mechanical equipment and are widely used in the industry. Ideally, the vibration signal should be responded to in time when a fault occurs. However, due to the influence of various conditions such as on-site noise and multi-working conditions on-site, the vibration signal fault response is not obvious, or there are difficulties in extracting fault features. At present, there have been studies, most of which are based on vibration signals for time domain statistical index extraction, frequency domain statistical index extraction, time-frequency domain joint distribution matrix extraction, signal decomposition and other methods to extract fault features and combine various types of neural networks for fault prediction. However, the prediction accuracy is generally not high, and the fault characteristics are only reflected in a certain dimension, which cannot comprehensively and timely capture the short-term instantaneous characteristics of the fault and the long-term decay and degradation characteristics of the fault.
  • the main purpose of this application is to provide a method and device for determining mechanical equipment faults, so as to solve the technical problem of the lack of high-precision detection means for mechanical equipment faults in related technologies.
  • a method for determining a mechanical equipment fault includes: obtaining vibration data corresponding to the target test position of the equipment and a preset neural network model; inputting the vibration data into the preset neural network model, and obtaining the output results of the preset neural network model; determining the output results contained in the Target tag, where the target tag is either a fault tag or a non-fault tag; if the target tag is a fault tag, it is determined that a fault occurs at the target test location; otherwise, it is determined that a fault does not occur at the target test location.
  • the method before obtaining the preset neural network model, also includes: constructing an initial preset neural network model, wherein the initial preset neural network model includes a multi-layer convolution layer, a multi-layer pooling layer and a multi-layer fully connected layer. layer, each neural network layer is connected through the ReLU activation function, and the fully connected layer connected to the initial preset neural network model is connected with the LogSoftmax function; determine the training data set used to train the initial preset neural network model, and pass the training data Collect and train the initial preset neural network model to obtain the preset neural network model.
  • the initial preset neural network model includes a multi-layer convolution layer, a multi-layer pooling layer and a multi-layer fully connected layer. layer, each neural network layer is connected through the ReLU activation function, and the fully connected layer connected to the initial preset neural network model is connected with the LogSoftmax function; determine the training data set used to train the initial preset neural network model, and pass the training data Collect and train the initial preset neural network model to obtain the prese
  • determining the training data set used to train the initial preset neural network model includes: obtaining source data, where the source data is data collected through vibration sensors set on the mechanical equipment, and the source data is fault type data and non- Any of the fault type data; based on the source data, construct a 3D time-frequency joint distribution cube containing preset dimensions, where the preset dimensions at least include long time domain dimensions, short time domain dimensions, and frequency domain dimensions. Perform dimensionality reduction feature extraction on the 3D time-frequency joint distribution cube to obtain the LST-FD joint distribution matrix set; divide multiple LST-FD joint distribution matrices contained in the LST-FD joint distribution matrix set to obtain the training data set and Test data set for testing preset neural network models.
  • the method before dividing multiple LST-FD joint distribution matrices contained in the LST-FD joint distribution matrix set to obtain a training data set and a test data set for testing the preset neural network model, the method also includes : Determine the tag type corresponding to the LST-FD joint distribution matrix based on the source data type corresponding to the LST-FD joint distribution matrix, where the tag type is either a fault tag or a non-fault tag.
  • multiple 3D time-frequency joint distribution cubes containing preset dimensions are constructed, including: determining the first time granularity, and performing data segmentation on the source data through the first time granularity to obtain a fault sample data set. And a non-fault sample data set, where the fault sample data set contains multiple fault data subsets, the non-fault sample data set contains multiple non-fault data subsets, and the granularity at the first time includes at least multiple rotating integers corresponding to the mechanical equipment.
  • the total time corresponding to the cycle determine the second time granularity; divide each fault data subset according to the second time granularity to obtain multiple first short time period data; based on the second time granularity, divide each fault data subset into Split the non-fault data subset to obtain multiple second short-time period data; multiple first short-time period data corresponding to the fault sample data set, and multiple second short-time period data corresponding to the non-fault sample data set Processed to obtain a 3D time-frequency joint distribution cube.
  • multiple first short-time period data corresponding to the fault sample data set and multiple second short-time period data corresponding to the non-fault sample data set are processed to obtain a 3D time-frequency joint distribution cube, including: Perform STFT transformation processing on multiple first short time period data to obtain multiple first time domain joint distribution matrices; perform STFT transformation processing on multiple second short time period data to obtain multiple second time domain joint distribution matrices. ; According to the first time granularity and the preset sequence, stack multiple first time domain joint distribution matrices and multiple second time domain joint distribution matrices into a time domain joint distribution matrix set; According to the time domain joint distribution matrix set, Construct a 3D time-frequency joint distribution cube.
  • dimensionality reduction feature extraction is performed on the 3D time-frequency joint distribution cube to obtain the LST-FD joint distribution matrix set, including: based on the long time domain dimension, dimensionality reduction feature extraction is performed on a pair of D time-frequency joint distribution cubes through the formula to obtain LST-FD joint distribution matrix, where Formula 1 is:
  • z(k, t, f re ) is the time-frequency joint distribution matrix under the long-term dimension of the kth layer in the 3D time-frequency joint distribution cube
  • t is the short time domain dimension corresponding to the 3D time-frequency joint distribution cube
  • F re _d is the frequency-time domain average difference calculated by the 3D time-frequency joint distribution cube based on the long-time domain dimension of frequency replication.
  • the preset neural network model is a 2D-CNN neural network model.
  • a device for determining mechanical equipment failure includes: a first acquisition unit configured to acquire vibration data corresponding to the target test position of the device and a preset neural network model; a first input unit configured to input the vibration data to the preset neural network model, and Obtain the output result of the preset neural network model; the first determination unit is configured to determine the target label included in the output result, where the target label is either a fault label or a non-fault label; the second determination unit is It is configured to determine that a fault occurs at the target test location when the target label is a fault label; otherwise, it is determined that a fault does not occur at the target test location.
  • a computer-readable storage medium includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the claims.
  • a processor is provided, which is characterized in that the processor is used to run a program, wherein when the program is run, the machine of any one of claims 1 to 7 is executed. How to determine equipment failure.
  • the following steps are adopted: obtain the vibration data corresponding to the target test position of the equipment and the preset neural network model; input the vibration data into the preset neural network model, and obtain the output results of the preset neural network model; determine the output The target tag included in the result, where the target tag is either a fault tag or a non-fault tag; if the target tag is a fault tag, it is determined that the target test location is faulty; otherwise, it is determined that the target test location is not faulty.
  • Figure 1 is a flow chart of a method for determining a mechanical equipment failure according to an embodiment of the present application.
  • Figure 2 is a flow chart corresponding to another method for determining mechanical equipment failure provided by this application;
  • Figure 3 is a schematic diagram of a mechanical equipment fault determination device provided according to an embodiment of the present application.
  • a method for determining a mechanical equipment fault is provided.
  • Figure 1 is a flow chart of a method for determining a mechanical equipment failure according to an embodiment of the present application. As shown in Figure 1, the invention includes the following steps:
  • Step S101 obtain the vibration data corresponding to the target test position of the device and the preset neural network model
  • Step S102 input the vibration data into the preset neural network model, and obtain the output result of the preset neural network model
  • Step S103 determine the target tag included in the output result, where the target tag is any one of a fault tag and a non-fault tag;
  • Step S104 If the target tag is a fault tag, it is determined that a fault occurs at the target test location; otherwise, it is determined that a fault does not occur at the target test location.
  • this application provides a method for determining mechanical equipment faults by inputting the vibration data of the part to be tested of the equipment into a preset neural network model, and determining whether a fault occurs at the location to be tested through the labels output by the neural network.
  • the vibration signal of the part to be tested in this application is mainly obtained by installing a vibration sensor at the target detection position of the mechanical equipment, and collecting and transmitting back the vibration data through data acquisition software for storage.
  • a one-way acceleration vibration sensor is installed in the horizontal direction of the input side of the reduction box (close to the input side bearing position) to collect vibration data. It is marked AI1-32, the sensor sampling frequency is 51.2KHZ, and the speed of the motor at the power end of the reduction gearbox is known.
  • the method before obtaining the preset neural network model, further includes: constructing an initial preset neural network model, wherein, The initial preset neural network model includes multi-layer convolution layers, multi-layer pooling layers and multi-layer fully connected layers. Each neural network layer is connected through the ReLU activation function and is connected to the fully connected layer of the initial preset neural network model. There is a LogSoftmax function; determine the training data set used to train the initial preset neural network model, and train the initial preset neural network model through the training data set to obtain the preset neural network model.
  • the preset neural network model is a 2D-CNN neural network model.
  • a 2D-CNN neural network model needs to be built first.
  • the network structure of the 2D-CNN neural network model includes a total of 7 layers. , including 3 convolutional layers, 2 pooling layers, and 2 fully connected layers.
  • the ReLU activation function is used between neurons, and finally the fully connected layer is connected to the LogSoftmax function to output the model prediction results.
  • the loss function of the network model uses the CrossEntropyLoss function to calculate the error between the input data and the prediction result.
  • the optimizer corresponding to the model uses the Adam function to optimize the model neuron connection weights.
  • this application provides a method for determining the training data set of a preset neural network model, which specifically includes the following steps:
  • S201 Obtain source data, where the source data is collected through a vibration sensor installed on the mechanical equipment, and the source data is any one of fault type data and non-fault type data;
  • S202 Based on the source data, construct a 3D time-frequency joint distribution cube containing preset dimensions, where the preset dimensions at least include long time domain dimensions, short time domain dimensions, and frequency domain dimensions;
  • S203 Perform dimensionality reduction feature extraction on the 3D time-frequency joint distribution cube to obtain the LST-FD joint distribution matrix set;
  • S204 Divide multiple LST-FD joint distribution matrices contained in the LST-FD joint distribution matrix set to obtain a training data set and a test data set for testing the preset neural network model.
  • this application provides a mechanical equipment fault prediction method based on the LST-FD (Long short time-Frequency difference) matrix.
  • the LST-FD matrix starts from the two time particle dimensions of short time (instantaneous) and long time, constructs a 3D time-frequency joint distribution cube under the long and short time dimensions, and performs time domain average difference on the frequency signals in the long time dimension. value calculation, dimensionality reduction to extract feature values, and obtain the LST-FD matrix, which is used as the feature matrix for fault classification prediction of the preset neural network model to perform equipment fault classification prediction.
  • the LST-FD matrix adds another long time domain dimension based on the time-frequency joint distribution matrix, and performs feature dimensionality reduction and extraction under this time domain dimension.
  • the prediction accuracy is relatively more accurate.
  • multiple 3D time-frequency joint distribution cubes containing preset dimensions are constructed based on the source data, including: determining the first Time granularity, and perform data segmentation on the source data through the first time granularity to obtain a fault sample data set and a non-fault sample data set, where the fault sample data set contains multiple fault data subsets, and the non-fault sample data set contains For multiple non-fault data subsets, the first time granularity at least includes the total time corresponding to multiple complete rotation cycles of the mechanical equipment; determine the second time granularity; according to the second time granularity, divide each fault data subset into Divide to obtain multiple first short-time period data; divide each non-fault data subset according to the second time granularity to obtain multiple second short-time period data; perform multiple data sets corresponding to the fault sample data set The first short-time period data and multiple second short-time period data corresponding to the
  • embodiments of the present application provide two time granularities, including a first time granularity and a second time granularity.
  • the second time granularity t 1 is determined.
  • the second time granularity is determined according to the rotation period of the device, and generally covers the time length of one entire rotation cycle of the device.
  • the second time granularity perform data division on each of the above-mentioned fault data subsets and non-fault data subsets to obtain multiple first short-time period data and second short-time period data, for example: x_ (f2,) fault data
  • the subset according to the time granularity t 2 , is divided into The above processing.
  • a plurality of first short time period data corresponding to the fault sample data set, and a plurality of first short time period data corresponding to the non-fault sample data set are Processing multiple second short time period data to obtain a 3D time-frequency joint distribution cube includes: performing STFT transformation processing on multiple first short time period data to obtain multiple first time domain joint distribution matrices; STFT transformation is performed on the second short time period data to obtain multiple second time domain joint distribution matrices; according to the first time granularity and the preset sequence, multiple first time domain joint distribution matrices and multiple third time domain joint distribution matrices are obtained The two time-domain joint distribution matrices are stacked into a time-domain joint distribution matrix set; based on the time-domain joint distribution matrix set, a 3D time-frequency joint distribution cube is constructed.
  • x_ (f2) ⁇ x_ (t11,) x_ (t12,) ...,x_ t1k ⁇
  • k time-frequency joint distribution matrices contained in each data set with k (i.e. t 2 time length) as the third dimension, and k time-frequency joint distribution matrices are stacked in order of t 2 time to generate 3D Time-frequency joint distribution cube construction. That is, each sample data in the above two data sets corresponds to a 3D time-frequency joint distribution cube (k, t, fre).
  • the first dimension of the 3D time-frequency joint distribution cube is the long time domain dimension of the vibration signal (including at least 10 complete rotation cycles of the device), the second dimension is the short time domain dimension of the vibration signal (covering one rotation cycle of the device), and the third dimension is the short time domain dimension of the vibration signal (covering one rotation cycle of the device).
  • the dimension is the frequency domain dimension of the vibration signal.
  • dimensionality reduction feature extraction is performed on the 3D time-frequency joint distribution cube to obtain the LST-FD joint distribution matrix set, including: Based on the long time domain dimension, dimensionality reduction feature extraction is performed on a pair of D time-frequency joint distribution cubes through the formula to obtain the LST-FD joint distribution matrix, where Formula 1 is:
  • z (k, t, fre) is the time-frequency joint distribution matrix under the long time dimension (t 1 time dimension) of the kth layer in the 3D time-frequency joint distribution cube
  • t is the time-frequency joint distribution matrix corresponding to the 3D time-frequency joint distribution cube.
  • Fre_ d is the frequency-time domain average difference calculated by the 3D time-frequency joint distribution cube based on the frequency replication calculation in the long time domain dimension.
  • dimensionality reduction is performed to extract features.
  • the frequency values at different short time points in the second and third dimensions are corresponding to the amplitudes, and the frequency time domain average difference Fre_d is calculated sequentially along the first dimension.
  • each 3D time-frequency joint distribution cube perform dimensionality reduction feature extraction on each 3D time-frequency joint distribution cube to obtain each LST-FD joint distribution matrix.
  • the shape of the matrix is (t, Fre_d), and t is the second dimension of the original 3D time-frequency joint distribution cube.
  • the short time domain dimension of the vibration signal, Fre_d is the frequency time domain average difference calculated from the frequency amplitude based on the first dimension of the original 3D time-frequency joint distribution cube.
  • the method also includes: determining the label type corresponding to the LST-FD joint distribution matrix according to the source data type corresponding to the LST-FD joint distribution matrix, where , the label type is either a fault label or a non-fault label.
  • the data (each data subset) is processed into a one-to-one corresponding set of LST-FD joint distribution matrices.
  • the corresponding label data sets labeled 1 and 0 are generated for the LST-FD joint distribution matrix generated from the non-fault data and fault data respectively.
  • 1 indicates that the device status is faulty
  • 0 indicates that the device status is normal.
  • the LST-FD joint distribution matrix set corresponding to the normal data and the fault data and the respective label data are divided according to the ratio of 1:9, respectively, to generate a test set and a training set.
  • the training data set into the above-mentioned neural network model.
  • the data is calculated in each layer of the network, it is output to the next layer of neural network layer through the ReLU activation function, and the last layer's LogSoftmax function outputs the calculation result.
  • input the calculation results and real data into the CrossEntropyLoss loss function, and the loss function calculates the loss value.
  • the optimization function Adam backpropagates the value in the gradient direction in the direction of reducing the loss value according to the loss value. Update the network connection weights of each layer; when the loss function value is less than the set threshold ⁇ , the neural network training ends and the network structure and neuron information at all levels are saved;
  • the experimental data selects the horizontal direction AI1-32 sensor data on the input side of the vehicle-mounted plunger pump reduction box of the fracturing vehicle, and the sensor sampling frequency is 51.2kHZ.
  • a total of 30 hours of normal operation AI1-32 sensor data of the reduction gearbox and 18 hours of reduction gearbox bearing roller failure data were obtained.
  • Model training and model testing are performed based on the above training data and test data.
  • the relevant parameters are as follows:
  • Epoch (number of model training times) 100 learning rate 0.0001 Loss value threshold ⁇ 0.01 AUC value threshold ⁇ 0.90
  • the same model directly uses time-frequency diagrams or time-domain signal statistical indicators as feature inputs to perform fault classification prediction results as follows:
  • the 2D_CNN model based on LST-FD has the highest prediction accuracy and the largest AUC value.
  • the model prediction accuracy based on time domain signal statistical indicators is relatively low, and the AUC value has just exceeded the set threshold. .
  • 3D joint distribution cube containing three dimensions: long time domain dimension, short time domain dimension and frequency domain dimension is constructed. This cube can capture the short-time instantaneous spectrum fault characteristics of the fault, and can also Capture the long-term degradation trend characteristics of faults, enable early detection of incipient faults and monitor fault degradation trends for a long time;
  • This application also provides another method for determining mechanical equipment faults, as shown in Figure 2.
  • t2 in Figure 2 is the total time corresponding to multiple full rotation cycles of the mechanical equipment, and t1 is equipment 1.
  • the length of an entire rotation cycle, the method provided in Figure 1 also solves the technical problem of the lack of high-precision detection methods for mechanical equipment faults in related technologies, and thereby captures and responds to vibration signals in response to faults from multiple dimensions. It can not only capture the instantaneous characteristics of faults, but also reflect the technical effects of long-term degradation trend characteristics of faults.
  • An embodiment of the present application provides a method for determining mechanical equipment faults by obtaining vibration data corresponding to the target test position of the equipment and a preset neural network model; inputting the vibration data into the preset neural network model, and obtaining the preset neural network model The output result of the network model; determine the target label included in the output result, where the target label is either a fault label or a non-fault label; when the target label is a fault label, determine that the target test location has failed, and vice versa , it is determined that no fault has occurred at the target test position, and the technical problem of the lack of high-precision detection means of mechanical equipment faults in related technologies is solved. This achieves the technical effect of capturing and reacting the response of vibration signals to faults from multiple dimensions, which can capture the instantaneous characteristics of faults and reflect the long-term degradation trend characteristics of faults.
  • the embodiment of the present application also provides a device for determining a mechanical equipment failure. It should be noted that the device for determining a mechanical equipment failure in the embodiment of the present application can be used to perform the method for determining a mechanical equipment failure provided by the embodiment of the present application. How to determine equipment failure. The following is an introduction to a mechanical equipment fault determination device provided by an embodiment of the present application.
  • Figure 3 is a schematic diagram of a mechanical equipment fault determination device according to an embodiment of the present application.
  • the device includes: an acquisition unit 301, configured to acquire vibration data corresponding to the target test position of the device, and a preset neural network model; an input unit 302, configured to input the vibration data to the preset neural network model network model, and obtains the output result of the preset neural network model; the first determination unit 303 is configured to determine the target label contained in the output result, where the target label is any one of a fault label and a non-fault label; The second determination unit 304 is configured to determine that a fault occurs at the target test location when the target tag is a fault tag; otherwise, determine that a fault does not occur at the target test location.
  • the building unit is configured to build an initial preset neural network model before acquiring the preset neural network model, wherein the initial preset neural network model includes a multi-layer convolution layer and a multi-layer The pooling layer and the multi-layer fully connected layer are connected through the ReLU activation function.
  • the fully connected layer connected to the initial preset neural network model is connected with the LogSoftmax function;
  • the third determination unit is configured to determine the Train the training data set of the initial preset neural network model, and train the initial preset neural network model through the training data set to obtain the preset neural network model.
  • the third determination unit includes: an acquisition subunit, configured to acquire source data, wherein the source data is data collected through a vibration sensor provided on the mechanical equipment, and the source data is fault Any one of type data and non-fault type data; the construction subunit is configured to construct a 3D time-frequency joint distribution cube containing preset dimensions based on the source data, where the preset dimensions at least include long time domain dimensions and short time domain dimensions. time domain dimension and frequency domain dimension.
  • the extraction subunit is configured to perform dimensionality reduction feature extraction on the 3D time-frequency joint distribution cube to obtain the LST-FD joint distribution matrix set; the dividing subunit is configured to extract multiple LSTs contained in the LST-FD joint distribution matrix set.
  • -FD joint distribution matrix is divided to obtain a training data set and a test data set for testing the preset neural network model.
  • the divided subunits include: a first determination module configured to determine the label type corresponding to the LST-FD joint distribution matrix according to the source data type corresponding to the LST-FD joint distribution matrix, where , the label type is either a fault label or a non-fault label.
  • the construction subunit includes: a second determination module configured to determine a first time granularity, and perform data segmentation on the source data through the first time granularity to obtain a fault sample data set And a non-fault sample data set, where the fault sample data set contains multiple fault data subsets, the non-fault sample data set contains multiple non-fault data subsets, and the granularity at the first time includes at least multiple rotating integers corresponding to the mechanical equipment.
  • the total time corresponding to the cycle; the third determination module is configured to determine the second time granularity; the first segmentation module is configured to segment each fault data subset according to the second time granularity to obtain multiple third a short time period data; a second segmentation module configured to segment each non-fault data subset according to the second time granularity to obtain multiple second short time period data; a processing module configured to segment the fault data A plurality of first short-time period data corresponding to the sample data set and a plurality of second short-time period data corresponding to the non-fault sample data set are processed to obtain a 3D time-frequency joint distribution cube.
  • the processing module includes: a first processing submodule configured to perform STFT transformation processing on a plurality of first short time period data to obtain a plurality of first time domain joint distribution matrices;
  • the second processing sub-module is configured to perform STFT transformation processing on a plurality of second short-time period data to obtain a plurality of second time-domain joint distribution matrices;
  • the stacking sub-module is configured to perform STFT transformation processing based on the first time granularity and predetermined Assuming the sequence, multiple first time domain joint distribution matrices and multiple second time domain joint distribution matrices are stacked into a time domain joint distribution matrix set;
  • the construction submodule is configured to construct a 3D based on the time domain joint distribution matrix set. Time-frequency joint distribution cube.
  • the extraction subunit includes: an extraction module configured to perform dimensionality reduction feature extraction on a D time-frequency joint distribution cube based on the long time domain dimension through the formula to obtain the LST-FD joint distribution Matrix, where Formula 1 is:
  • z(k, t, fre) is the time-frequency joint distribution matrix under the long-term dimension of the kth layer in the 3D time-frequency joint distribution cube
  • t is the short time domain dimension corresponding to the 3D time-frequency joint distribution cube
  • Fre_d It is the frequency-time domain average difference obtained by calculating the frequency replication of the 3D time-frequency joint distribution cube based on the long time domain dimension.
  • An embodiment of the present application provides a device for determining mechanical equipment faults.
  • the acquisition unit 301 is configured to acquire vibration data corresponding to the target test position of the equipment and a preset neural network model;
  • the input unit 302 is configured to obtain the vibration data.
  • the first determination unit 303 is configured to determine the target label included in the output result, where the target label is a fault label and a non-fault label. Any one;
  • the second determination unit 304 is configured to determine that the target test location is faulty when the target tag is a fault tag, and conversely, determine that the target test location is not faulty.
  • a device for determining mechanical equipment faults includes a processor and a memory.
  • the above-mentioned acquisition unit 201 and so on are stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to implement corresponding functions.
  • the processor contains a core, which retrieves the corresponding program unit from the memory.
  • One or more cores can be set, and the technical problem of the lack of high-precision detection means for mechanical equipment faults in related technologies is solved by adjusting the core parameters.
  • Memory may include non-permanent memory in computer-readable media, random access memory (RAM) and/or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM).
  • RAM random access memory
  • ROM read-only memory
  • flash RAM flash memory
  • Embodiments of the present application provide a storage medium on which a program is stored.
  • the program is executed by a processor, a method for determining a mechanical equipment failure is implemented.
  • Embodiments of the present application provide a processor, which is used to run a program, wherein when the program is run, a method for determining a mechanical equipment fault is executed.
  • the embodiment of the present application provides a device.
  • the device includes a processor, a memory, and a program stored in the memory and executable on the processor.
  • the processor executes the program, it implements the following steps: Obtain vibration data corresponding to the target test position of the device. , and the preset neural network model; input vibration data into the preset neural network model, and obtain the output results of the preset neural network model; determine the target labels included in the output results, where the target labels are fault labels and non-fault labels Any one of them; when the target label is a fault label, it is determined that the target test position has failed, otherwise, it is determined that the target test position has not failed.
  • the method before obtaining the preset neural network model, further includes: constructing an initial preset neural network model, wherein the initial preset neural network model includes a multi-layer convolution layer and a multi-layer pooling layer. layer and multi-layer fully connected layers. Each neural network layer is connected through the ReLU activation function. The fully connected layer connected to the initial preset neural network model is connected with the LogSoftmax function; determine the training used to train the initial preset neural network model. data set, and train the initial preset neural network model through the training data set to obtain the preset neural network model.
  • determining the training data set used to train the initial preset neural network model includes: obtaining source data, where the source data is data collected through a vibration sensor set on the mechanical equipment, the source The data is either fault type data or non-fault type data; based on the source data, a 3D time-frequency joint distribution cube containing preset dimensions is constructed, where the preset dimensions at least include long time domain dimensions and short time domain dimensions and Frequency domain dimension. Perform dimensionality reduction feature extraction on the 3D time-frequency joint distribution cube to obtain the LST-FD joint distribution matrix set; divide multiple LST-FD joint distribution matrices contained in the LST-FD joint distribution matrix set to obtain the training data set and Test data set for testing preset neural network models.
  • multiple LST-FD joint distribution matrices included in the LST-FD joint distribution matrix set are divided to obtain a training data set and test data for testing the preset neural network model.
  • the method also includes: determining the label type corresponding to the LST-FD joint distribution matrix according to the source data type corresponding to the LST-FD joint distribution matrix, where the label type is any one of a fault label and a non-fault label.
  • constructing multiple 3D time-frequency joint distribution cubes containing preset dimensions includes: determining the first time granularity, and performing data analysis on the source data through the first time granularity. Split to obtain a fault sample data set and a non-fault sample data set, where the fault sample data set contains multiple fault data subsets, and the non-fault sample data set contains multiple non-fault data subsets.
  • the first granularity includes at least mechanical The total time corresponding to multiple full rotation cycles of the equipment; determine the second time granularity; divide each fault data subset according to the second time granularity to obtain multiple first short time period data; based on the second time granularity Time granularity, divide each non-fault data subset to obtain multiple second short time period data; multiple first short time period data corresponding to the fault sample data set, and multiple first short time period data corresponding to the non-fault sample data set
  • the second short time period data is processed to obtain a 3D time-frequency joint distribution cube.
  • a plurality of first short-time period data corresponding to the fault sample data set and a plurality of second short-time period data corresponding to the non-fault sample data set are processed to obtain the 3D time-frequency
  • the joint distribution cube includes: performing STFT transformation processing on multiple first short time period data to obtain multiple first time domain joint distribution matrices; performing STFT transformation processing on multiple second short time period data to obtain multiple The second time domain joint distribution matrix; according to the first time granularity and the preset sequence, stack multiple first time domain joint distribution matrices and multiple second time domain joint distribution matrices into a time domain joint distribution matrix set; according to A set of time-domain joint distribution matrices to construct a 3D time-frequency joint distribution cube.
  • dimensionality reduction feature extraction is performed on the 3D time-frequency joint distribution cube to obtain the LST-FD joint distribution matrix set, including: based on the long time domain dimension, a pair of D time-frequency joint distribution cubes are obtained through the formula Dimensionality reduction feature extraction is performed to obtain the LST-FD joint distribution matrix, where Formula 1 is:
  • z(k, t, fre) is the time-frequency joint distribution matrix under the long-term dimension of the kth layer in the 3D time-frequency joint distribution cube
  • t is the short time domain dimension corresponding to the 3D time-frequency joint distribution cube
  • Fre_d It is the frequency-time domain average difference obtained by calculating the frequency replication of the 3D time-frequency joint distribution cube based on the long time domain dimension.
  • the devices in this article can be servers, PCs, PADs, mobile phones, etc.
  • This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program initialized with the following method steps: obtaining vibration data corresponding to the target test position of the device, and presetting the neural network model; The vibration data is input to the preset neural network model, and the output result of the preset neural network model is obtained; the target label included in the output result is determined, where the target label is either a fault label or a non-fault label; in the target label In the case of a fault label, it is determined that a fault occurs at the target test position; otherwise, it is determined that a fault does not occur at the target test position.
  • the method before obtaining the preset neural network model, further includes: constructing an initial preset neural network model, wherein the initial preset neural network model includes a multi-layer convolution layer and a multi-layer pooling layer. layer and multi-layer fully connected layers. Each neural network layer is connected through the ReLU activation function. The fully connected layer connected to the initial preset neural network model is connected with the LogSoftmax function; determine the training used to train the initial preset neural network model. data set, and train the initial preset neural network model through the training data set to obtain the preset neural network model.
  • determining the training data set used to train the initial preset neural network model includes: obtaining source data, where the source data is data collected through a vibration sensor set on the mechanical equipment, the source The data is either fault type data or non-fault type data; based on the source data, a 3D time-frequency joint distribution cube containing preset dimensions is constructed, where the preset dimensions at least include long time domain dimensions and short time domain dimensions and Frequency domain dimension. Perform dimensionality reduction feature extraction on the 3D time-frequency joint distribution cube to obtain the LST-FD joint distribution matrix set; divide multiple LST-FD joint distribution matrices contained in the LST-FD joint distribution matrix set to obtain the training data set and Test data set for testing preset neural network models.
  • multiple LST-FD joint distribution matrices included in the LST-FD joint distribution matrix set are divided to obtain a training data set and test data for testing the preset neural network model.
  • the method also includes: determining the label type corresponding to the LST-FD joint distribution matrix according to the source data type corresponding to the LST-FD joint distribution matrix, where the label type is any one of a fault label and a non-fault label.
  • constructing multiple 3D time-frequency joint distribution cubes containing preset dimensions includes: determining the first time granularity, and performing data analysis on the source data through the first time granularity. Split to obtain a fault sample data set and a non-fault sample data set, where the fault sample data set contains multiple fault data subsets, and the non-fault sample data set contains multiple non-fault data subsets.
  • the first granularity includes at least mechanical The total time corresponding to multiple full rotation cycles of the equipment; determine the second time granularity; divide each fault data subset according to the second time granularity to obtain multiple first short time period data; based on the second time granularity Time granularity, divide each non-fault data subset to obtain multiple second short time period data; multiple first short time period data corresponding to the fault sample data set, and multiple first short time period data corresponding to the non-fault sample data set
  • the second short time period data is processed to obtain a 3D time-frequency joint distribution cube.
  • a plurality of first short-time period data corresponding to the fault sample data set and a plurality of second short-time period data corresponding to the non-fault sample data set are processed to obtain the 3D time-frequency
  • the joint distribution cube includes: performing STFT transformation processing on multiple first short time period data to obtain multiple first time domain joint distribution matrices; performing STFT transformation processing on multiple second short time period data to obtain multiple The second time domain joint distribution matrix; according to the first time granularity and the preset sequence, stack multiple first time domain joint distribution matrices and multiple second time domain joint distribution matrices into a time domain joint distribution matrix set; according to A set of time-domain joint distribution matrices to construct a 3D time-frequency joint distribution cube.
  • dimensionality reduction feature extraction is performed on the 3D time-frequency joint distribution cube to obtain the LST-FD joint distribution matrix set, including: based on the long time domain dimension, a pair of D time-frequency joint distribution cubes are obtained through the formula Dimensionality reduction feature extraction is performed to obtain the LST-FD joint distribution matrix, where Formula 1 is:
  • z(k, t, fre) is the time-frequency joint distribution matrix under the long-term dimension of the kth layer in the 3D time-frequency joint distribution cube
  • t is the short time domain dimension corresponding to the 3D time-frequency joint distribution cube
  • Fre_d It is the frequency-time domain average difference obtained by calculating the frequency replication of the 3D time-frequency joint distribution cube based on the long time domain dimension.
  • embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment that combines software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
  • computer-usable storage media including, but not limited to, disk storage, CD-ROM, optical storage, etc.
  • These computer program instructions may also be stored in a computer-readable memory that causes a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction means, the instructions
  • the device implements the functions specified in a process or processes of the flowchart and/or a block or blocks of the block diagram.
  • These computer program instructions may also be loaded onto a computer or other programmable data processing device, causing a series of operating steps to be performed on the computer or other programmable device to produce computer-implemented processing, thereby executing on the computer or other programmable device.
  • Instructions provide steps for implementing the functions specified in a process or processes of a flowchart diagram and/or a block or blocks of a block diagram.
  • a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
  • processors CPUs
  • input/output interfaces network interfaces
  • memory volatile and non-volatile memory
  • Memory may include non-volatile memory in computer-readable media, random access memory (RAM) and/or non-volatile memory in the form of read-only memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.
  • RAM random access memory
  • ROM read-only memory
  • flash RAM flash memory
  • Computer-readable media includes both persistent and non-volatile, removable and non-removable media that can be implemented by any method or technology for storage of information.
  • Information may be computer-readable instructions, data structures, modules of programs, or other data.
  • Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), and read-only memory.
  • PRAM phase change memory
  • SRAM static random access memory
  • DRAM dynamic random access memory
  • RAM random access memory
  • read-only memory read-only memory
  • ROM read-only memory
  • EEPROM electrically erasable programmable read-only memory
  • flash memory or other memory technology
  • compact disc read-only memory CD-ROM
  • DVD digital versatile disc
  • Magnetic tape cassettes tape magnetic disk storage or other magnetic storage devices or any other non-transmission medium can be used to store information that can be accessed by a computing device.
  • computer-readable media does not include transitory media, such as modulated data signals and carrier waves.
  • embodiments of the present application may be provided as methods, systems or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
  • computer-usable storage media including, but not limited to, disk storage, CD-ROM, optical storage, etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

一种机械设备故障的确定方法及装置，包括：获取设备的目标测试位置对应的振动数据，以及预设神经网络模型(S101)；将振动数据输入至预设神经网络模型，并获取预设神经网络模型的输出结果(S102)；确定输出结果中包含的目标标签，其中，目标标签为故障标签和非故障标签中的任意一种(S103)；在目标标签为故障标签的情况下，确定目标测试位置发生故障，反之，确定目标测试位置未发生故障(S104)。由此，解决了相关技术中缺少高精度的机械设备故障的检测手段的技术问题。

Description

一种机械设备故障的确定方法及装置
本申请要求于2022年4月27日提交至中国国家知识产权局、申请号为202210452424.6且发明名称为“一种机械设备故障的确定方法及装置”的专利申请的优先权。
技术领域
本申请涉及机械设备检测领域，具体而言，涉及一种机械设备故障的确定方法及装置。
背景技术
相关技术中，机械设备状态检测和故障诊断目前是热门的研究方向，随着机械设备作业数据等各类数据的积累存储，基于大数据方式的设备故障预测方法也是百花齐放，各种神经网络模型应运而生。发明人知晓的一些方案中，大多数研究基于实验室环境，不能模拟和构建真实现场多因素影响的工况环境，导致模型很难工业化落地。且故障特征提取是一个难点，直接影响设备故障预测精度。
振动信号作为往复或者旋转类机械设备常用的故障状态反应数据，在行业内被广泛应用。理想情况下，振动信号在故障发生时，及时发生信号相应。但是现场噪声影响、现场多工况作业等各种条件影响导致振动信号故障相应不明显，或者故障特征提取存在困难。目前已有研究，大多基于振动信号进行时域统计指标提取、频域统计指标提取、时频域联合分布矩阵提取、信号分解等多种方式进行故障特征提取并结合各类神经网络进行故障预测。但是普遍预测精度不高，且故障特征仅仅体现在某一维度，不能全面及时的捕捉故障短时瞬时特征和故障长时间的衰变劣化特征。
针对相关技术中存在的上述，目前尚未提出有效的解决方案。
发明内容
本申请的主要目的在于提供一种机械设备故障的确定方法及装置，以解决了相关技术中缺少高精度的机械设备故障的检测手段的技术问题。
为了实现上述目的，根据本申请的一个方面，提供了一种机械设备故障的确定方法。该发明包括：获取设备的目标测试位置对应的振动数据，以及预设神经网络模型；将振动数据输入至预设神经网络模型，并获取预设神经网络模型的输出结果；确定输出结果中包含的目标标签，其中，目标标签为故障标签和非故障标签中的任意一种；在目标标签为故障标签的情况下，确定目标测试位置发生故障，反之，确定目标测试位置未发生故障。
进一步地，在获取预设神经网络模型之前，该方法还包括：构建初始预设神经网络模型，其中，初始预设神经网络模型包含多层卷积层和多层池化层以及多层全连接层，各个神经网络层之间通过ReLU激活函数连接，连接在初始预设神经网络模型的全连接层连接有LogSoftmax函数；确定用于训练初始预设神经网络模型的训练数据集，并通过训练数据集对初始预设神经网络模型进行训练以获得预设神经网络模型。
进一步地，确定用于训练初始预设神经网络模型的训练数据集，包括：获取源数据，其中，源数据通过设置在机械设备上的振动传感器采集到的数据，源数据为故障类型数据和非故障类型数据中的任意一种；依据源数据，构建包含预设维度的3D时频联合分布立方体，其中，预设维度至少包括长时域维度和短时域维度以及频域维度。对3D时频联合分布立方体进行降维特征提取以获得LST-FD联合分布矩阵集合；对LST-FD联合分布矩阵集合中包含的多个LST-FD联合分布矩阵进行划分，以获得训练数据集以及用于测试预设神经网络模型的测试数据集。
进一步地，在对LST-FD联合分布矩阵集合中包含的多个LST-FD联合分布矩阵进行划分，以获得训练数据集以及用于测试预设神经网络模型的 测试数据集之前，该方法还包括：依据LST-FD联合分布矩阵对应的源数据类型，确定LST-FD联合分布矩阵对应的标签类型，其中，标签类型为故障标签和非故障标签中的任意一种。
进一步地，依据源数据，构建多个包含预设维度的3D时频联合分布立方体，包括：确定第一时间颗粒度，并通过第一时间颗粒度对源数据进行数据分割以获得故障样本数据集以及非故障样本数据集，其中，故障样本数据集中包含多个故障数据子集，非故障样本数据集中包含多个非故障数据子集，第一时间颗粒度至少包括机械设备对应的多个旋转整周期对应的总时间；确定第二时间颗粒度；依据第二时间颗粒度，将每个故障数据子集进行分割以获得多个第一短时间周期数据；依据第二时间颗粒度，将每个非故障数据子集进行分割以获得多个第二短时间周期数据；对故障样本数据集对应的多个第一短时间周期数据，以及非故障样本数据集对应的多个第二短时间周期数据进行处理，以获得3D时频联合分布立方体。
进一步地，对故障样本数据集对应的多个第一短时间周期数据，以及非故障样本数据集对应的多个第二短时间周期数据进行处理，以获得3D时频联合分布立方体，包括：对多个第一短时间周期数据进行STFT变换处理，以获得多个第一时域联合分布矩阵；对多个第二短时间周期数据进行STFT变换处理，以获得多个第二时域联合分布矩阵；依据第一时间颗粒度以及预设先后顺序，将多个第一时域联合分布矩阵以及多个第二时域联合分布矩阵堆叠成时域联合分布矩阵集；依据时域联合分布矩阵集，构建3D时频联合分布立方体。
进一步地，对3D时频联合分布立方体进行降维特征提取以获得LST-FD联合分布矩阵集合，包括：基于长时域维度，通过公式一对D时频联合分布立方体进行降维特征提取以获得LST-FD联合分布矩阵，其中，公式一为：
Figure PCTCN2022111885-appb-000001
其中，z(k，t，f re)为3D时-频联合分布立方体中第k层长时间维度下的时-频联合分布矩阵，t为3D时频联合分布立方体对应的短时域维度，F re_d为3D 时频联合分布立方体基于长时域维度对频率复制计算求得的频率时域平均差值。
进一步地，所述预设神经网络模型为2D-CNN神经网络模型。
为了实现上述目的，根据本申请的另一方面，提供了一种机械设备故障的确定装置。该装置包括：第一获取单元，被配置为获取设备的目标测试位置对应的振动数据，以及预设神经网络模型；第一输入单元，被配置为将振动数据输入至预设神经网络模型，并获取预设神经网络模型的输出结果；第一确定单元，被配置为确定输出结果中包含的目标标签，其中，目标标签为故障标签和非故障标签中的任意一种；第二确定单元，被配置为在目标标签为故障标签的情况下，确定目标测试位置发生故障，反之，确定目标测试位置未发生故障。
为了实现上述目的，根据本申请的另一方面，提供了一种计算机可读存储介质，计算机可读存储介质包括存储的程序，其中，在程序运行时控制计算机可读存储介质所在设备执行权利要求1至7中任意一项一种机械设备故障的确定方法。
为了实现上述目的，根据本申请的另一方面，提供了一种一种处理器，其特征在于，处理器用于运行程序，其中，程序运行时执行权利要求1至7中任意一项一种机械设备故障的确定方法。
通过本申请，采用以下步骤：获取设备的目标测试位置对应的振动数据，以及预设神经网络模型；将振动数据输入至预设神经网络模型，并获取预设神经网络模型的输出结果；确定输出结果中包含的目标标签，其中，目标标签为故障标签和非故障标签中的任意一种；在目标标签为故障标签的情况下，确定目标测试位置发生故障，反之，确定目标测试位置未发生故障。解决了相关技术中缺少高精度的机械设备故障的检测手段的技术问题，进而达到了从多个维度捕捉和反应振动信号对于故障的响应，既能够捕捉故障瞬时特征，又能反应故障长期劣化趋势特征的技术效果。
附图说明
构成本申请的一部分的附图用来提供对本申请的进一步理解，本申请的示意性实施例及其说明用于解释本申请，并不构成对本申请的不当限定。在附图中：
图1是根据本申请实施例提供的一种机械设备故障的确定方法的流程图；以及
图2为本申请提供的另一种机械设备故障的确定方法对应的流程图；
图3是根据本申请实施例提供的一种机械设备故障的确定装置的示意图。
具体实施方式
需要说明的是，在不冲突的情况下，本申请中的实施例及实施例中的特征可以相互组合。下面将参考附图并结合实施例来详细说明本申请。
为了使本技术领域的人员更好地理解本申请方案，下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分的实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本申请保护的范围。
需要说明的是，本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本申请的实施例。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
根据本申请的实施例，提供了一种机械设备故障的确定方法。
图1是根据本申请实施例提供的一种机械设备故障的确定方法的流程 图。如图1所示，该发明包括以下步骤：
步骤S101，获取设备的目标测试位置对应的振动数据，以及预设神经网络模型；
步骤S102，将振动数据输入至预设神经网络模型，并获取预设神经网络模型的输出结果；
步骤S103，确定输出结果中包含的目标标签，其中，目标标签为故障标签和非故障标签中的任意一种；
步骤S104，在目标标签为故障标签的情况下，确定目标测试位置发生故障，反之，确定目标测试位置未发生故障。
上述地，本申请提供了一种机械设备故障的确定方法，通过将设备待测试部位的振动数据输入至预设神经网络模型中，通过神经网络输出的标签来确定待测试位置是否发生故障。
通过上述方法，解决了相关技术中缺少高精度的机械设备故障的检测手段的技术问题，进而达到了从多个维度捕捉和反应振动信号对于故障的响应，既能够捕捉故障瞬时特征，又能反应故障长期劣化趋势特征的技术效果。
具体地，本申请待测试部位的振动信号获取主要通过在机械设备目标检测位置安装振动传感器，通过数采软件进行振动数据采集和回传存储。
例如：在对压裂车车载柱塞泵减速箱输入端轴承故障预测的过程中，在减速箱输入侧水平方向(靠近输入侧轴承位置)安装一个单向加速度振动传感器进行振动数据采集，该传感器标记为AI1-32，传感器采样频率为51.2KHZ，且减速箱动力端电机转速已知。
在一个可选的实施例中，在本申请实施例提供的一种机械设备故障的确定方法中，在获取预设神经网络模型之前，该方法还包括：构建初始预设神经网络模型，其中，初始预设神经网络模型包含多层卷积层和多层池化层以及多层全连接层，各个神经网络层之间通过ReLU激活函数连接，连接在初始预设神经网络模型的全连接层连接有LogSoftmax函数；确定 用于训练初始预设神经网络模型的训练数据集，并通过训练数据集对初始预设神经网络模型进行训练以获得预设神经网络模型。
在一种可选的实施例中，预设神经网络模型为2D-CNN神经网络模型，上述方法中，首先需要搭建2D-CNN神经网络模型，2D-CNN神经网络模型的网络结构共包含7层，其中卷积层3层，池化层2层，全连接层2层。神经元之间通过ReLU激活函数，最后全连接层连接LogSoftmax函数，输出模型预测结果。
需要说明的是，在一个可选的实施例中，该网络模型的损失函数选用CrossEntropyLoss函数进行输入数据与预测结果误差计算。同时，该模型对应的优化器选用Adam函数进行模型神经元连接权重优化。
上述的，本申请提供了一种确定预设神经网络模型的训练数据集的方法，具体包括以下步骤：
S201：获取源数据，其中，源数据通过设置在机械设备上的振动传感器采集到的数据，源数据为故障类型数据和非故障类型数据中的任意一种；
S202：依据源数据，构建包含预设维度的3D时频联合分布立方体，其中，预设维度至少包括长时域维度和短时域维度以及频域维度；
S203：对3D时频联合分布立方体进行降维特征提取以获得LST-FD联合分布矩阵集合；
S204：对LST-FD联合分布矩阵集合中包含的多个LST-FD联合分布矩阵进行划分，以获得训练数据集以及用于测试预设神经网络模型的测试数据集。
上述地，本申请提供了一种基于LST-FD(Long short time-Frequency difference，即长短时-频差)矩阵的机械设备故障预测方法。LST-FD矩阵从短时间(瞬时)和长时间两个时间颗粒维度出发，构建长短双时间维度下的3D时-频联合分布立方体，并通过对长时间维度下的频率信号进行时域平均差值计算，降维提取特征值，获得LST-FD矩阵，以此作为预设神经网络模型故障分类预测的特征矩阵，进行设备故障分类预测。LST-FD矩阵， 基于时-频联合分布矩阵加入另一个长时域维度，并在该时域维度下进行特征降维提取，一方面放大了故障信号发生时时频波动特征，另一方面能够同时捕捉设备在故障发生初期短时瞬时特征响应及长期衰变特征波-动两种故障信号。相对传统时频图或者其他特征指标作为模型输入，预测精度相对更加准确。
在一个可选的实施例中，在本申请实施例提供的一种机械设备故障的确定方法中，依据源数据，构建多个包含预设维度的3D时频联合分布立方体，包括：确定第一时间颗粒度，并通过第一时间颗粒度对源数据进行数据分割以获得故障样本数据集以及非故障样本数据集，其中，故障样本数据集中包含多个故障数据子集，非故障样本数据集中包含多个非故障数据子集，第一时间颗粒度至少包括机械设备对应的多个旋转整周期对应的总时间；确定第二时间颗粒度；依据第二时间颗粒度，将每个故障数据子集进行分割以获得多个第一短时间周期数据；依据第二时间颗粒度，将每个非故障数据子集进行分割以获得多个第二短时间周期数据；对故障样本数据集对应的多个第一短时间周期数据，以及非故障样本数据集对应的多个第二短时间周期数据进行处理，以获得3D时频联合分布立方体。
具体地，本申请的实施例提供了两种时间粒度，其中，包括第一时间颗粒度以及第二时间颗粒度。
在一种具体地实施例中，根据设备动力端转速，按照时间颗粒度t 1(第一时间颗粒度)对源数据进行样本数据分割(时间颗粒度t 1长度至少包含10个设备旋转整周期时间长度)，从而获得分割以后的故障数据样本数据集X_ fault＝{x_ (f1,)x_ (f2,)…,x_ fi}和正常数据样本数据集X_ normal＝{x_ (n1,)x_ (n2,)…,x_ ni}，其中，x_ (f1,)等为故障数据子集，x_ (n1,)为非故障数据子集。
进一步地，确定第二时间颗粒度t 1，优选地，第二时间颗粒度根据设备旋转周期确定，一般覆盖设备1个旋转整周期时间长度。
依据第二时间颗粒度对上述的各个故障数据子集以及非故障数据子集分别进行数据划分得到多个第一短时间周期数据以及第二短时间周期数据，例如：x_ (f2,)故障数据子集，按照时间颗粒度t 2，划分为x_ (f2,)＝{x_ (t11,) x_ (t12,)…,x_ t1k},其中,k＝t 2/t 1，其他样本数据皆进行上述处理。
在一个可选的实施例中，在本申请实施例提供的一种机械设备故障的确定方法中，对故障样本数据集对应的多个第一短时间周期数据，以及非故障样本数据集对应的多个第二短时间周期数据进行处理，以获得3D时频联合分布立方体，包括：对多个第一短时间周期数据进行STFT变换处理，以获得多个第一时域联合分布矩阵；对多个第二短时间周期数据进行STFT变换处理，以获得多个第二时域联合分布矩阵；依据第一时间颗粒度以及预设先后顺序，将多个第一时域联合分布矩阵以及多个第二时域联合分布矩阵堆叠成时域联合分布矩阵集；依据时域联合分布矩阵集，构建3D时频联合分布立方体。
进一步地，对通过第二时间颗粒度进行划分得到的多个第一短时间周期数据和多个第二短时间周期数据进行STFT变换，其中，窗函数选用hann窗，窗函数长度为256，窗函数重叠数为50％。经过STFT变换之后获得时-频联合分布矩阵z (t,fre)，其中，t为时间长度，f re为频率范围，z中值为频率幅值。例：x_ (f2)＝{x_ (t11,)x_ (t12,)…,x_ t1k}，经过STFT变换后得到k个时-频联合分布矩阵{z_ (x_t11,)z_ (x_t12,)z_ (x_t13,)…,z_ x_t1k}。即X_ (fault)＝{x_ (f1,)x_ (f2,)…,x_ fi}和X_ (normal)＝{x_ (n1,)x_ (n2,)…,x_ ni}中每个短时间周期数据生成k个时-频联合分布矩阵。
进一步地，将X_ (fault)＝{x_ (f1,)x_ (f2,)…,x_ fi}和X_ (normal)＝{x_ (n1,)x_ (n2,)…,x_ ni}两个数据集中，每个数据集中包含的k个时-频联合分布矩阵，以k(即t 2时间长度)为第三维度，按照t 2时间先后顺序，堆叠k个时-频联合分布矩阵，生成3D时-频联合分布立方体构建。即上述两个数据集中的每一个样本数据对应一个3D时-频联合分布立方体(k,t,fre)。该3D时-频联合分布立方体中第一维度为振动信号长时域维度(至少包含10个设备旋转整周期)，第二维度为振动信号短时域维度(覆盖设备一个旋转周期)，第三维度为振动信号频域维度。
在一个可选的实施例中，在本申请实施例提供的一种机械设备故障的确定方法中，对3D时频联合分布立方体进行降维特征提取以获得LST-FD 联合分布矩阵集合，包括：基于长时域维度，通过公式一对D时频联合分布立方体进行降维特征提取以获得LST-FD联合分布矩阵，其中，公式一为：
Figure PCTCN2022111885-appb-000002
其中，z (k，t，fre)为3D时-频联合分布立方体中第k层长时间维度(t 1时间维度)下的时-频联合分布矩阵，t为3D时频联合分布立方体对应的短时域维度，Fre_ d为3D时频联合分布立方体基于长时域维度对频率复制计算求得的频率时域平均差值。
进一步的，将上述构建得到的3D时-频联合分布立方体
，基于第一维度，即振动信号长时域维度，进行降维提取特征。
即将第二和第三维度上的不同短时间点上的频率值对应得幅值，沿第一维度时间顺序计算频率时域平均差值Fre_d。
按照上述公式对各3D时-频联合分布立方体进行降维特征提取，获得各LST-FD联合分布矩阵，该矩阵形状为(t，Fre_d)，t为原3D时-频联合分布立方体第二维振动信号短时域维度，Fre_d为原3D时-频联合分布立方体基于第一维，对频率幅值求得的频率时域平均差值。
在一个可选的实施例中，在本申请实施例提供的一种机械设备故障的确定方法中，在对LST-FD联合分布矩阵集合中包含的多个LST-FD联合分布矩阵进行划分，以获得训练数据集以及用于测试预设神经网络模型的测试数据集之前，该方法还包括：依据LST-FD联合分布矩阵对应的源数据类型，确定LST-FD联合分布矩阵对应的标签类型，其中，标签类型为故障标签和非故障标签中的任意一种。
上述地，经过上述对源数据的处理，将源数据集合X fault＝{x f1，x f2，…，x fi}和 X normal＝{x n1，x n2，…，x ni}中的各样本数据(各个数据子集)被处理成一一对应的LST-FD联合分布矩阵集合。并对非故障数据数据和故障数据生成的LST-FD联合分布矩阵分别生成标注1和0的对应标签数据集。其中，1表示设备状态故障，0表示设备状态正常。按照1：9的比例分别对正常数据和故障数据对应的LST-FD联合分布矩阵集合以及各自标签数据进行分割，生成测试集和训练集。
进一步地，生成训练集和测试集后，将训练数据集输入上述神经网络模型，数据在各层网络进行计算后经过ReLU激活函数输出到下一层神经网络层，最后一层LogSoftmax函数输出计算结果，将计算结果与真实数据输入到CrossEntropyLoss损失函数，损失函数计算损失值，当损失值大于设定阈值β时，优化函数Adam根据损失值反向传播值在梯度方向上朝着降低损失值的方向更新各层网络连接权值；当损失函数值小于设定阈值β时，神经网络训练结束，保存网络结构和各级神经元信息；
将测试数据集输入训练好的神经网络模型中，输出测试结果和测试准确度量AUC值，如果AUC值小于设定阈值α则完成模型测试，如果AUC值大于阈值α则重新抽样选择样本数据重新进行模型训练和测试。
在本申请提供的一个具体实施例中，实验数据选取压裂车车载柱塞泵减速箱输入侧水平方向AI1-32传感器数据，传感器采样频率为51.2kHZ。共获取减速箱正常作业AI1-32传感器数据30小时，减速箱轴承滚子故障数据18小时。按照设备曲轴旋转整周期时间长度，分别设置时间颗粒度t2＝1s，t1＝10s。
训练过程：
并按照步骤1～5，生成正常数据LST-FD联合分布矩阵10800个以及相 同数量的正常数据标签，生成故障数据LST-FD联合分布矩阵6480个以及相同数量的故障数据标签。按照1：9的比例进行样本分割，共获得正常训练数据及标签各9720个，正常测试数据及标签各1080个，故障训练数据及标签各5832个，故障测试数据及标签各648个。
基于以上训练数据和测试数据进行模型训练以及模型测试，相关涉及参数如下所示：
Epoch(模型训练次数) 100
学习速率 0.0001
损失值阈值β 0.01
AUC值阈值α 0.90
模型预测结果及对比相同模型直接将时频图或者时域信号统计指标作为特征输入进行故障分类预测结果如下所示：
Figure PCTCN2022111885-appb-000003
因此，由上述具体实施例看出，基于LST-FD的2D_CNN模型预测准确率最高，AUC值也是最大，基于时域信号统计指标得模型预测准确率相对较低，且AUC值刚过设定阈值。
因此，本申请提供的一种机械设备故障的确定方法具备以下优势：
1：创新性的构建了LST-FD联合分布矩阵，并结合预设神经网络模型实现相对较高预测精度的设备故障预测；
2：从设备故障发生特征角度考虑，构建了包含长时域维度和短时域维度以及频域维度，三个维度的3D联合分布立方体，该立方体能捕捉故障短时间瞬时频谱故障特征，也能捕捉故障长时间劣化趋势特征，能够较早的发现初期故障且长时间监控故障劣化趋势；
3：对3D联合分布立方体进行长时域维度的降维主特征提取，生成LST-FD联合分布矩阵，既保留了原三维得数据特征，又降低了数据维度，提高了模型预测精度和运行效率。
本申请还提供了另一种机械设备故障的确定方法，如图2所示，需要说明的是，图2中的t2为机械设备对应的多个旋转整周期对应的总时长，t1为设备1个旋转整周期时间长度，通过图1提供的方法同样解决了相关技术中缺少高精度的机械设备故障的检测手段的技术问题，进而达到了从多个维度捕捉和反应振动信号对于故障的响应，既能够捕捉故障瞬时特征，又能反应故障长期劣化趋势特征的技术效果。
需要说明的是，在附图的流程图示出的步骤可以在诸如一组计算机可执行指令的计算机系统中执行，并且，虽然在流程图中示出了逻辑顺序，但是在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤。
本申请实施例提供的一种机械设备故障的确定方法，通过获取设备的目标测试位置对应的振动数据，以及预设神经网络模型；将振动数据输入至预设神经网络模型，并获取预设神经网络模型的输出结果；确定输出结果中包含的目标标签，其中，目标标签为故障标签和非故障标签中的任意一种；在目标标签为故障标签的情况下，确定目标测试位置发生故障，反之，确定目标测试位置未发生故障，解决了相关技术中缺少高精度的机械设备故障的检测手段的技术问题。进而达到了从多个维度捕捉和反应振动信号对于故障的响应，既能够捕捉故障瞬时特征，又能反应故障长期劣化趋势特征的技术效果。
需要说明的是，在附图的流程图示出的步骤可以在诸如一组计算机可执行指令的计算机系统中执行，并且，虽然在流程图中示出了逻辑顺序，但是在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤。
本申请实施例还提供了一种机械设备故障的确定装置，需要说明的是，本申请实施例的一种机械设备故障的确定装置可以用于执行本申请实施例所提供的用于一种机械设备故障的确定方法。以下对本申请实施例提供的一种机械设备故障的确定装置进行介绍。
图3是根据本申请实施例的一种机械设备故障的确定装置的示意图。如图3所示，该装置包括：获取单元301，被配置为获取设备的目标测试位置对应的振动数据，以及预设神经网络模型；输入单元302，被配置为将振动数据输入至预设神经网络模型，并获取预设神经网络模型的输出结果；第一确定单元303，被配置为确定输出结果中包含的目标标签，其中，目标标签为故障标签和非故障标签中的任意一种；第二确定单元304，被配置为在目标标签为故障标签的情况下，确定目标测试位置发生故障，反之，确定目标测试位置未发生故障。
在一种可选的实施例中，构建单元，被配置为在获取预设神经网络模型之前，构建初始预设神经网络模型，其中，初始预设神经网络模型包含多层卷积层和多层池化层以及多层全连接层，各个神经网络层之间通过ReLU激活函数连接，连接在初始预设神经网络模型的全连接层连接有LogSoftmax函数；第三确定单元，被配置为确定用于训练初始预设神经网络模型的训练数据集，并通过训练数据集对初始预设神经网络模型进行训练以获得预设神经网络模型。
在一种可选的实施例中，第三确定单元，包括：获取子单元，被配置为获取源数据，其中，源数据通过设置在机械设备上的振动传感器采集到的数据，源数据为故障类型数据和非故障类型数据中的任意一种；构建子单元，被配置为依据源数据，构建包含预设维度的3D时频联合分布立方体，其中，预设维度至少包括长时域维度和短时域维度以及频域维度。提取子单元，被配置为对3D时频联合分布立方体进行降维特征提取以获得LST-FD联合分布矩阵集合；划分子单元，被配置为对LST-FD联合分布矩阵集合中包含的多个LST-FD联合分布矩阵进行划分，以获得训练数据集以及用于测试预设神经网络模型的测试数据集。
在一种可选的实施例中，划分子单元，包括：第一确定模块，被配置为依据LST-FD联合分布矩阵对应的源数据类型，确定LST-FD联合分布矩阵对应的标签类型，其中，标签类型为故障标签和非故障标签中的任意一种。
在一种可选的实施例中，构建子单元，包括：第二确定模块，被配置为确定第一时间颗粒度，并通过第一时间颗粒度对源数据进行数据分割以获得故障样本数据集以及非故障样本数据集，其中，故障样本数据集中包含多个故障数据子集，非故障样本数据集中包含多个非故障数据子集，第一时间颗粒度至少包括机械设备对应的多个旋转整周期对应的总时间；第三确定模块，被配置为确定第二时间颗粒度；第一分割模块，被配置为依据第二时间颗粒度，将每个故障数据子集进行分割以获得多个第一短时间周期数据；第二分割模块，被配置为依据第二时间颗粒度，将每个非故障数据子集进行分割以获得多个第二短时间周期数据；处理模块，被配置为对故障样本数据集对应的多个第一短时间周期数据，以及非故障样本数据集对应的多个第二短时间周期数据进行处理，以获得3D时频联合分布立方体。
在一种可选的实施例中，处理模块，包括：第一处理子模块，被配置为对多个第一短时间周期数据进行STFT变换处理，以获得多个第一时域联合分布矩阵；第二处理子模块，被配置为对多个第二短时间周期数据进行STFT变换处理，以获得多个第二时域联合分布矩阵；堆叠子模块，被配置为依据第一时间颗粒度以及预设先后顺序，将多个第一时域联合分布矩阵以及多个第二时域联合分布矩阵堆叠成时域联合分布矩阵集；构建子模块，被配置为依据时域联合分布矩阵集，构建3D时频联合分布立方体。
在一种可选的实施例中，提取子单元，包括：提取模块，被配置为基于长时域维度，通过公式一对D时频联合分布立方体进行降维特征提取以获得LST-FD联合分布矩阵，其中，公式一为：
Figure PCTCN2022111885-appb-000004
其中，z(k，t，fre)为3D时-频联合分布立方体中第k层长时间维度下的时-频联合分布矩阵，t为3D时频联合分布立方体对应的短时域维度，Fre_d为3D时频联合分布立方体基于长时域维度对频率复制计算求得的频率时域平均差值。
本申请实施例提供的一种机械设备故障的确定装置，获取单元301， 被配置为获取设备的目标测试位置对应的振动数据，以及预设神经网络模型；输入单元302，被配置为将振动数据输入至预设神经网络模型，并获取预设神经网络模型的输出结果；第一确定单元303，被配置为确定输出结果中包含的目标标签，其中，目标标签为故障标签和非故障标签中的任意一种；第二确定单元304，被配置为在目标标签为故障标签的情况下，确定目标测试位置发生故障，反之，确定目标测试位置未发生故障。
一种机械设备故障的确定装置包括处理器和存储器，上述获取单元201等均作为程序单元存储在存储器中，由处理器执行存储在存储器中的上述程序单元来实现相应的功能。
处理器中包含内核，由内核去存储器中调取相应的程序单元。内核可以设置一个或以上，通过调整内核参数来解决了了相关技术中缺少高精度的机械设备故障的检测手段的技术问题。
存储器可能包括计算机可读介质中的非永久性存储器，随机存取存储器(RAM)和/或非易失性内存等形式，如只读存储器(ROM)或闪存(flash RAM)，存储器包括至少一个存储芯片。
本申请实施例提供了一种存储介质，其上存储有程序，该程序被处理器执行时实现一种机械设备故障的确定方法。
本申请实施例提供了一种处理器，处理器用于运行程序，其中，程序运行时执行一种机械设备故障的确定方法。
本申请实施例提供了一种设备，设备包括处理器、存储器及存储在存储器上并可在处理器上运行的程序，处理器执行程序时实现以下步骤：获取设备的目标测试位置对应的振动数据，以及预设神经网络模型；将振动数据输入至预设神经网络模型，并获取预设神经网络模型的输出结果；确定输出结果中包含的目标标签，其中，目标标签为故障标签和非故障标签中的任意一种；在目标标签为故障标签的情况下，确定目标测试位置发生故障，反之，确定目标测试位置未发生故障。
在一种可选的实施例中，在获取预设神经网络模型之前，该方法还包 括：构建初始预设神经网络模型，其中，初始预设神经网络模型包含多层卷积层和多层池化层以及多层全连接层，各个神经网络层之间通过ReLU激活函数连接，连接在初始预设神经网络模型的全连接层连接有LogSoftmax函数；确定用于训练初始预设神经网络模型的训练数据集，并通过训练数据集对初始预设神经网络模型进行训练以获得预设神经网络模型。
在一种可选的实施例中，确定用于训练初始预设神经网络模型的训练数据集，包括：获取源数据，其中，源数据通过设置在机械设备上的振动传感器采集到的数据，源数据为故障类型数据和非故障类型数据中的任意一种；依据源数据，构建包含预设维度的3D时频联合分布立方体，其中，预设维度至少包括长时域维度和短时域维度以及频域维度。对3D时频联合分布立方体进行降维特征提取以获得LST-FD联合分布矩阵集合；对LST-FD联合分布矩阵集合中包含的多个LST-FD联合分布矩阵进行划分，以获得训练数据集以及用于测试预设神经网络模型的测试数据集。
在一种可选的实施例中，在对LST-FD联合分布矩阵集合中包含的多个LST-FD联合分布矩阵进行划分，以获得训练数据集以及用于测试预设神经网络模型的测试数据集之前，该方法还包括：依据LST-FD联合分布矩阵对应的源数据类型，确定LST-FD联合分布矩阵对应的标签类型，其中，标签类型为故障标签和非故障标签中的任意一种。
在一种可选的实施例中，依据源数据，构建多个包含预设维度的3D时频联合分布立方体，包括：确定第一时间颗粒度，并通过第一时间颗粒度对源数据进行数据分割以获得故障样本数据集以及非故障样本数据集，其中，故障样本数据集中包含多个故障数据子集，非故障样本数据集中包含多个非故障数据子集，第一时间颗粒度至少包括机械设备对应的多个旋转整周期对应的总时间；确定第二时间颗粒度；依据第二时间颗粒度，将每个故障数据子集进行分割以获得多个第一短时间周期数据；依据第二时间颗粒度，将每个非故障数据子集进行分割以获得多个第二短时间周期数据；对故障样本数据集对应的多个第一短时间周期数据，以及非故障样本数据集对应的多个第二短时间周期数据进行处理，以获得3D时频联合分 布立方体。
在一种可选的实施例中，对故障样本数据集对应的多个第一短时间周期数据，以及非故障样本数据集对应的多个第二短时间周期数据进行处理，以获得3D时频联合分布立方体，包括：对多个第一短时间周期数据进行STFT变换处理，以获得多个第一时域联合分布矩阵；对多个第二短时间周期数据进行STFT变换处理，以获得多个第二时域联合分布矩阵；依据第一时间颗粒度以及预设先后顺序，将多个第一时域联合分布矩阵以及多个第二时域联合分布矩阵堆叠成时域联合分布矩阵集；依据时域联合分布矩阵集，构建3D时频联合分布立方体。
在一种可选的实施例中，对3D时频联合分布立方体进行降维特征提取以获得LST-FD联合分布矩阵集合，包括：基于长时域维度，通过公式一对D时频联合分布立方体进行降维特征提取以获得LST-FD联合分布矩阵，其中，公式一为：
Figure PCTCN2022111885-appb-000005
其中，z(k，t，fre)为3D时-频联合分布立方体中第k层长时间维度下的时-频联合分布矩阵，t为3D时频联合分布立方体对应的短时域维度，Fre_d为3D时频联合分布立方体基于长时域维度对频率复制计算求得的频率时域平均差值。
本文中的设备可以是服务器、PC、PAD、手机等。
本申请还提供了一种计算机程序产品，当在数据处理设备上执行时，适于执行初始化有如下方法步骤的程序：获取设备的目标测试位置对应的振动数据，以及预设神经网络模型；将振动数据输入至预设神经网络模型，并获取预设神经网络模型的输出结果；确定输出结果中包含的目标标签，其中，目标标签为故障标签和非故障标签中的任意一种；在目标标签为故障标签的情况下，确定目标测试位置发生故障，反之，确定目标测试位置未发生故障。
在一种可选的实施例中，在获取预设神经网络模型之前，该方法还包 括：构建初始预设神经网络模型，其中，初始预设神经网络模型包含多层卷积层和多层池化层以及多层全连接层，各个神经网络层之间通过ReLU激活函数连接，连接在初始预设神经网络模型的全连接层连接有LogSoftmax函数；确定用于训练初始预设神经网络模型的训练数据集，并通过训练数据集对初始预设神经网络模型进行训练以获得预设神经网络模型。
在一种可选的实施例中，确定用于训练初始预设神经网络模型的训练数据集，包括：获取源数据，其中，源数据通过设置在机械设备上的振动传感器采集到的数据，源数据为故障类型数据和非故障类型数据中的任意一种；依据源数据，构建包含预设维度的3D时频联合分布立方体，其中，预设维度至少包括长时域维度和短时域维度以及频域维度。对3D时频联合分布立方体进行降维特征提取以获得LST-FD联合分布矩阵集合；对LST-FD联合分布矩阵集合中包含的多个LST-FD联合分布矩阵进行划分，以获得训练数据集以及用于测试预设神经网络模型的测试数据集。
在一种可选的实施例中，在对LST-FD联合分布矩阵集合中包含的多个LST-FD联合分布矩阵进行划分，以获得训练数据集以及用于测试预设神经网络模型的测试数据集之前，该方法还包括：依据LST-FD联合分布矩阵对应的源数据类型，确定LST-FD联合分布矩阵对应的标签类型，其中，标签类型为故障标签和非故障标签中的任意一种。
在一种可选的实施例中，依据源数据，构建多个包含预设维度的3D时频联合分布立方体，包括：确定第一时间颗粒度，并通过第一时间颗粒度对源数据进行数据分割以获得故障样本数据集以及非故障样本数据集，其中，故障样本数据集中包含多个故障数据子集，非故障样本数据集中包含多个非故障数据子集，第一时间颗粒度至少包括机械设备对应的多个旋转整周期对应的总时间；确定第二时间颗粒度；依据第二时间颗粒度，将每个故障数据子集进行分割以获得多个第一短时间周期数据；依据第二时间颗粒度，将每个非故障数据子集进行分割以获得多个第二短时间周期数据；对故障样本数据集对应的多个第一短时间周期数据，以及非故障样本数据集对应的多个第二短时间周期数据进行处理，以获得3D时频联合分 布立方体。
在一种可选的实施例中，对故障样本数据集对应的多个第一短时间周期数据，以及非故障样本数据集对应的多个第二短时间周期数据进行处理，以获得3D时频联合分布立方体，包括：对多个第一短时间周期数据进行STFT变换处理，以获得多个第一时域联合分布矩阵；对多个第二短时间周期数据进行STFT变换处理，以获得多个第二时域联合分布矩阵；依据第一时间颗粒度以及预设先后顺序，将多个第一时域联合分布矩阵以及多个第二时域联合分布矩阵堆叠成时域联合分布矩阵集；依据时域联合分布矩阵集，构建3D时频联合分布立方体。
在一种可选的实施例中，对3D时频联合分布立方体进行降维特征提取以获得LST-FD联合分布矩阵集合，包括：基于长时域维度，通过公式一对D时频联合分布立方体进行降维特征提取以获得LST-FD联合分布矩阵，其中，公式一为：
Figure PCTCN2022111885-appb-000006
其中，z(k，t，fre)为3D时-频联合分布立方体中第k层长时间维度下的时-频联合分布矩阵，t为3D时频联合分布立方体对应的短时域维度，Fre_d为3D时频联合分布立方体基于长时域维度对频率复制计算求得的频率时域平均差值。
本领域内的技术人员应明白，本申请的实施例可提供为方法、系统、或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本申请是参照根据本申请实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、 专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
在一个典型的配置中，计算设备包括一个或多个处理器(CPU)、输入/输出接口、网络接口和内存。
存储器可能包括计算机可读介质中的非永久性存储器，随机存取存储器(RAM)和/或非易失性内存等形式，如只读存储器(ROM)或闪存(flash RAM)。存储器是计算机可读介质的示例。
计算机可读介质包括永久性和非永久性、可移动和非可移动媒体可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。计算机的存储介质的例子包括，但不限于相变内存(PRAM)、静态随机存取存储器(SRAM)、动态随机存取存储器(DRAM)、其他类型的随机存取存储器(RAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、快闪记忆体或其他内存技术、只读光盘只读存储器(CD-ROM)、数字多功能光盘(DVD)或其他光学存储、磁盒式磁带，磁带磁磁盘存储或其他磁性存储设备或任何其他非传输介质，可用于存储可以被计算设备访问的信息。按照本文中的界定，计算机可读介质不包括暂存电脑可读媒体(transitory media)，如调制的数据信号和载 波。
还需要说明的是，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、商品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、商品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括要素的过程、方法、商品或者设备中还存在另外的相同要素。
本领域技术人员应明白，本申请的实施例可提供为方法、系统或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
以上仅为本申请的实施例而已，并不用于限制本申请。对于本领域技术人员来说，本申请可以有各种更改和变化。凡在本申请的精神和原理之内所作的任何修改、等同替换、改进等，均应包含在本申请的权利要求范围之内。

Claims (11)

  1. 一种机械设备故障的确定方法，其中，包括：
    获取设备的目标测试位置对应的振动数据，以及预设神经网络模型；
    将所述振动数据输入至所述预设神经网络模型，并获取所述预设神经网络模型的输出结果；
    确定所述输出结果中包含的目标标签，其中，所述目标标签为故障标签和非故障标签中的任意一种；
    在所述目标标签为所述故障标签的情况下，确定所述目标测试位置发生故障，反之，确定所述目标测试位置未发生所述故障。
  2. 根据权利要求1所述的方法，其中，在获取预设神经网络模型之前，所述方法还包括：
    构建初始预设神经网络模型，其中，所述初始预设神经网络模型包含多层卷积层和多层池化层以及多层全连接层，各个神经网络层之间通过ReLU激活函数连接，连接在所述初始预设神经网络模型的所述全连接层连接有LogSoftmax函数；
    确定用于训练所述初始预设神经网络模型的训练数据集，并通过所述训练数据集对所述初始预设神经网络模型进行训练以获得所述预设神经网络模型。
  3. 根据权利要求2所述的方法，其中，确定用于训练所述初始预设神经网络模型的训练数据集，包括：
    获取源数据，其中，所述源数据通过设置在机械设备上的振动传感器采集到的数据，所述源数据为故障类型数据和非故障类型数据中的任意一种；
    依据所述源数据，构建包含预设维度的3D时频联合分布立方体，其中，所述预设维度至少包括长时域维度和短时域维度以及频域维度；
    对所述3D时频联合分布立方体进行降维特征提取以获得LST-FD 联合分布矩阵集合；
    对所述LST-FD联合分布矩阵集合中包含的多个LST-FD联合分布矩阵进行划分，以获得所述训练数据集以及用于测试所述预设神经网络模型的测试数据集。
  4. 根据权利要求3所述的方法，其中，在对所述LST-FD联合分布矩阵集合中包含的多个LST-FD联合分布矩阵进行划分，以获得所述训练数据集以及用于测试所述预设神经网络模型的测试数据集之前，所述方法还包括：
    依据所述LST-FD联合分布矩阵对应的源数据类型，确定所述LST-FD联合分布矩阵对应的标签类型，其中，所述标签类型为故障标签和非故障标签中的任意一种。
  5. 根据权利要求3所述的方法，其中，依据所述源数据，构建多个包含预设维度的3D时频联合分布立方体，包括：
    确定第一时间颗粒度，并通过所述第一时间颗粒度对所述源数据进行数据分割以获得故障样本数据集以及非故障样本数据集，其中，所述故障样本数据集中包含多个故障数据子集，所述非故障样本数据集中包含多个非故障数据子集，所述第一时间颗粒度至少包括所述机械设备对应的多个旋转整周期对应的总时间；
    确定第二时间颗粒度；
    依据所述第二时间颗粒度，将每个所述故障数据子集进行分割以获得多个第一短时间周期数据；
    依据所述第二时间颗粒度，将每个所述非故障数据子集进行分割以获得多个第二短时间周期数据；
    对所述故障样本数据集对应的多个所述第一短时间周期数据，以及所述非故障样本数据集对应的多个所述第二短时间周期数据进行处理，以获得所述3D时频联合分布立方体。
  6. 根据权利要求5所述的方法，其中，对所述故障样本数据集对应的多个所述第一短时间周期数据，以及所述非故障样本数据集对应的多个所述第二短时间周期数据进行处理，以获得所述3D时频联合分布立方体，包括：
    对多个所述第一短时间周期数据进行STFT变换处理，以获得多个第一时域联合分布矩阵；
    对多个所述第二短时间周期数据进行STFT变换处理，以获得多个第二时域联合分布矩阵；
    依据所述第一时间颗粒度以及预设先后顺序，将多个所述第一时域联合分布矩阵以及多个所述第二时域联合分布矩阵堆叠成时域联合分布矩阵集；
    依据所述时域联合分布矩阵集，构建所述3D时频联合分布立方体。
  7. 根据权利要求3所述的方法，其中，对所述3D时频联合分布立方体进行降维特征提取以获得LST-FD联合分布矩阵集合，包括：
    基于所述长时域维度，通过公式一对所述D时频联合分布立方体进行降维特征提取以获得LST-FD联合分布矩阵，其中，所述公式一为：
    Figure PCTCN2022111885-appb-100001
    其中，z(k，t，f re)为3D时-频联合分布立方体中第k层长时间维度下的时-频联合分布矩阵，t为所述3D时频联合分布立方体对应的短时域维度，F re_d为所述3D时频联合分布立方体基于所述长时域维度对频率复制计算求得的频率时域平均差值。
  8. 根据权利要求1至7任意一项所述的方法，其中，所述预设神经网络模型为2D-CNN神经网络模型。
  9. 一种机械设备故障的确定装置，其中，包括：
    获取单元，被配置为获取设备的目标测试位置对应的振动数据， 以及预设神经网络模型；
    输入单元，被配置为将所述振动数据输入至所述预设神经网络模型，并获取所述预设神经网络模型的输出结果；
    第一确定单元，被配置为确定所述输出结果中包含的目标标签，其中，所述目标标签为故障标签和非故障标签中的任意一种；
    第二确定单元，被配置为在所述目标标签为所述故障标签的情况下，确定所述目标测试位置发生故障，反之，确定所述目标测试位置未发生所述故障。
  10. 一种计算机可读存储介质，其中，所述计算机可读存储介质包括存储的程序，其中，在所述程序运行时控制所述计算机可读存储介质所在设备执行权利要求1至8中任意一项所述一种机械设备故障的确定方法。
  11. 一种处理器，其中，所述处理器用于运行程序，其中，所述程序运行时执行权利要求1至8中任意一项所述一种机械设备故障的确定方法。
PCT/CN2022/111885 2022-04-27 2022-08-11 一种机械设备故障的确定方法及装置 WO2023206860A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210452424.6 2022-04-27
CN202210452424.6A CN114722965A (zh) 2022-04-27 2022-04-27 一种机械设备故障的确定方法及装置

Publications (1)

Publication Number Publication Date
WO2023206860A1 true WO2023206860A1 (zh) 2023-11-02

Family

ID=82245854

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/111885 WO2023206860A1 (zh) 2022-04-27 2022-08-11 一种机械设备故障的确定方法及装置

Country Status (2)

Country Link
CN (1) CN114722965A (zh)
WO (1) WO2023206860A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114722965A (zh) * 2022-04-27 2022-07-08 烟台杰瑞石油装备技术有限公司 一种机械设备故障的确定方法及装置
CN115828087B (zh) * 2023-02-21 2023-05-09 山东大学 一种用于机械设备故障诊断的信号特征提取方法及系统

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103761569A (zh) * 2013-12-20 2014-04-30 上海电机学院 一种风力发电机故障诊断方法及装置
CN109612728A (zh) * 2018-12-20 2019-04-12 上海理工大学 基于振谱图和深度卷积神经网络的轴承故障诊断方法
CN109635677A (zh) * 2018-11-23 2019-04-16 华南理工大学 基于多标签分类卷积神经网络的复合故障诊断方法及装置
US20200225655A1 (en) * 2016-05-09 2020-07-16 Strong Force Iot Portfolio 2016, Llc Methods, systems, kits and apparatuses for monitoring and managing industrial settings in an industrial internet of things data collection environment
CN112084974A (zh) * 2020-09-14 2020-12-15 北京工商大学 一种基于元学习的多标签滚动轴承故障诊断方法
CN112395968A (zh) * 2020-11-11 2021-02-23 北京京航计算通讯研究所 基于神经网络的机械旋转部件故障诊断方法及装置
CN112903294A (zh) * 2021-01-07 2021-06-04 泰华宏业(天津)智能科技有限责任公司 基于vmd与深度卷积神经网络的滚动轴承故障诊断方法
CN113112000A (zh) * 2021-03-29 2021-07-13 太原科技大学 一种多层次卷积神经网络模型的构建方法及应用
CN113469060A (zh) * 2021-07-02 2021-10-01 浙大城市学院 多传感器融合卷积神经网络航空发动机轴承故障诊断方法
CN114722965A (zh) * 2022-04-27 2022-07-08 烟台杰瑞石油装备技术有限公司 一种机械设备故障的确定方法及装置

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103761569A (zh) * 2013-12-20 2014-04-30 上海电机学院 一种风力发电机故障诊断方法及装置
US20200225655A1 (en) * 2016-05-09 2020-07-16 Strong Force Iot Portfolio 2016, Llc Methods, systems, kits and apparatuses for monitoring and managing industrial settings in an industrial internet of things data collection environment
CN109635677A (zh) * 2018-11-23 2019-04-16 华南理工大学 基于多标签分类卷积神经网络的复合故障诊断方法及装置
CN109612728A (zh) * 2018-12-20 2019-04-12 上海理工大学 基于振谱图和深度卷积神经网络的轴承故障诊断方法
CN112084974A (zh) * 2020-09-14 2020-12-15 北京工商大学 一种基于元学习的多标签滚动轴承故障诊断方法
CN112395968A (zh) * 2020-11-11 2021-02-23 北京京航计算通讯研究所 基于神经网络的机械旋转部件故障诊断方法及装置
CN112903294A (zh) * 2021-01-07 2021-06-04 泰华宏业(天津)智能科技有限责任公司 基于vmd与深度卷积神经网络的滚动轴承故障诊断方法
CN113112000A (zh) * 2021-03-29 2021-07-13 太原科技大学 一种多层次卷积神经网络模型的构建方法及应用
CN113469060A (zh) * 2021-07-02 2021-10-01 浙大城市学院 多传感器融合卷积神经网络航空发动机轴承故障诊断方法
CN114722965A (zh) * 2022-04-27 2022-07-08 烟台杰瑞石油装备技术有限公司 一种机械设备故障的确定方法及装置

Also Published As

Publication number Publication date
CN114722965A (zh) 2022-07-08

Similar Documents

Publication Publication Date Title
WO2023206860A1 (zh) 一种机械设备故障的确定方法及装置
CN111933188A (zh) 一种基于卷积神经网络的声音事件检测方法
CN112257263B (zh) 基于自注意力机制的设备剩余寿命预测系统
US11036981B1 (en) Data monitoring system
CN113125135A (zh) 旋转机械的故障诊断方法、存储介质及电子设备
Li et al. Cross-attribute adaptation networks: Distilling transferable features from multiple sampling-frequency source domains for fault diagnosis of wind turbine gearboxes
Qiao et al. Fault diagnosis method of rolling bearings based on VMD and MDSVM
CN115248757A (zh) 一种硬盘健康评估方法和存储设备
Li et al. Experimental investigation for fault diagnosis based on a hybrid approach using wavelet packet and support vector classification
Yang et al. A novel multiple feature-based engine knock detection system using sparse Bayesian extreme learning machine
Imani et al. A scalable spark-based fault diagnosis platform for gearbox fault diagnosis in wind farms
CN117009814A (zh) 一种基于多源信息融合的发动机故障诊断方法及系统
Ayati et al. Classification-based fuel injection fault detection of a trainset diesel engine using vibration signature analysis
Xiao et al. Fault diagnosis of gearbox based on wavelet packet transform and CLSPSO-BP
CN112052712A (zh) 一种电力设备状态监测与故障识别方法及系统
CN114048787A (zh) 一种基于Attention CNN模型的轴承故障实时智能诊断方法与系统
Song et al. Research on rolling bearing fault diagnosis method based on improved LMD and CMWPE
Verstraeten et al. Edge computing for advanced vibration signal processing
CN117116293A (zh) 一种复杂声场环境中机器设备故障诊断系统
CN115238833B (zh) 一种空气涡轮起动机轴承性能退化特征提取方法
CN114639391A (zh) 机械故障提示方法、装置、电子设备及存储介质
CN114202224A (zh) 用于检测生产环境中焊接质量的方法、设备、介质及程序产品
Wang et al. An online wear state identification method for axial piston pump key friction pair based on FSANN
US9800973B1 (en) Sound source estimation based on simulated sound sensor array responses
CN115529331A (zh) 配电网箱式设备故障监测方法、系统和计算机设备