CN116127297A - Electromechanical device fault diagnosis method and system - Google Patents
Electromechanical device fault diagnosis method and system Download PDFInfo
- Publication number
- CN116127297A CN116127297A CN202310126200.0A CN202310126200A CN116127297A CN 116127297 A CN116127297 A CN 116127297A CN 202310126200 A CN202310126200 A CN 202310126200A CN 116127297 A CN116127297 A CN 116127297A
- Authority
- CN
- China
- Prior art keywords
- module
- data
- attention
- fault diagnosis
- self
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
Images
Classifications
-
- 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
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Selective Calling Equipment (AREA)
Abstract
The invention discloses a fault diagnosis method and system for electromechanical equipment, which can be applied to the technical field of equipment fault treatment. According to the invention, the plurality of encoder modules and decoder modules are arranged in the fault diagnosis network model, and the probability sparse self-attention sub-module and the self-attention distillation sub-module are respectively arranged in each self-attention module, so that when the fault diagnosis network model is trained through time sequence data, the characteristic information of the time sequence data can be extracted through the probability sparse self-attention sub-module, and the characteristic information is compressed through the self-attention distillation sub-module to obtain a characteristic sequence, so that the time complexity is reduced, the processing efficiency of long-sequence input data is effectively improved, and the scalability of the model is further expanded.
Description
Technical Field
The invention relates to the technical field of equipment fault processing, in particular to a method and a system for diagnosing faults of electromechanical equipment.
Background
In the related art, as the service time of the electromechanical device increases, the failure rate increases, which results in an increase in maintenance cost, repair period and miswork cost. At present, the fault diagnosis mode of the electromechanical equipment is to calculate the variance of each scale signal after the collected vibration signal is subjected to discrete wavelet transformation, and finally estimate the multi-scale slope characteristic based on the wavelet according to the slope of the logarithmic variance, thereby realizing the identification of fault types and the diagnosis of wear of the electromechanical equipment. However, this approach has high complexity of time-space computation for fault diagnosis of time series data and high memory requirements, thereby limiting scalability of the model when receiving long sequence inputs.
Disclosure of Invention
The present invention aims to solve at least one of the technical problems existing in the prior art. Therefore, the invention provides a fault diagnosis method and system for electromechanical equipment, which can effectively improve the processing efficiency of long-sequence input data and expand the scalability of a model.
In one aspect, an embodiment of the present invention provides a method for diagnosing a fault of an electromechanical device, including the following steps:
acquiring historical monitoring data of the electromechanical equipment as first time sequence data, wherein the first time sequence data comprises normal data and fault data of different types of the electromechanical equipment;
training a fault diagnosis network model through the first time sequence;
acquiring current monitoring data of the electromechanical equipment;
inputting the current monitoring data into the trained fault diagnosis network model to obtain a fault diagnosis result of the electromechanical equipment;
the fault diagnosis network model comprises a plurality of encoder modules and decoder modules;
the plurality of encoder modules are connected end to end, each encoder module comprises a plurality of self-attention modules, the plurality of self-attention modules are connected end to end, and each self-attention module comprises a probability sparse self-attention sub-module and a self-attention distillation sub-module; the probability sparse self-attention sub-module is used for extracting the characteristic information of the first time sequence data, and the self-attention distillation sub-module is used for compressing the characteristic information to obtain a characteristic sequence;
and the decoder module is used for predicting and obtaining a fault diagnosis result corresponding to the historical monitoring data according to the characteristic sequence.
In some embodiments, after the acquiring the historical monitoring data of the electromechanical device as the first time series data, the method further comprises the steps of:
normalizing the first time sequence data;
performing fast Fourier change on the history monitoring data after normalization processing to obtain frequency domain data;
taking half of the frequency domain data to perform inverse fourier transform.
In some embodiments, the fault diagnosis network model further comprises a multi-scale convolution module for convolving and then adding the first time series data and the corresponding time stamps.
In some embodiments, the self-attention distillation sub-module includes a convolution layer, an activation layer, and a max-pooling layer; the convolution layer is connected with the maximum pooling layer through the activation layer, and the step length of the maximum pooling layer is 2.
In some embodiments, the decoder module comprises a generative decoder.
In some embodiments, the training of the fault diagnosis network model by the first time sequence includes:
dividing the first time series data into a training set and a testing set;
and after the training set is input into the fault diagnosis network model for training, the test set is input into the trained fault diagnosis network model for testing the prediction effect.
In some embodiments, when the fault diagnosis network model is trained, the method further comprises the steps of:
and calculating a cross entropy loss function corresponding to the fault diagnosis network model.
In some embodiments, the cross entropy loss function is formulated as follows:
wherein y represents the true distribution,
representing the network output distribution, n represents the total number of categories.
In some embodiments, the first time series data is normalized by the following formula:
wherein x represents first time sequence data, y represents normalized data, max represents the maximum value of all the first time sequence data, and min represents the minimum value of all the first time sequence data.
In another aspect, an embodiment of the present invention provides an electromechanical device fault diagnosis system, including:
the first module is used for acquiring historical monitoring data of the electromechanical equipment as first time sequence data, wherein the first time sequence data comprises normal data and fault data of different types of the electromechanical equipment;
a second module for training a fault diagnosis network model by the first time sequence;
the third module is used for acquiring current monitoring data of the electromechanical equipment;
a fourth module, configured to input the current monitoring data to the trained fault diagnosis network model, to obtain a fault diagnosis result of the electromechanical device;
the fault diagnosis network model comprises a plurality of encoder modules and decoder modules;
the plurality of encoder modules are connected end to end, each encoder module comprises a plurality of self-attention modules, the plurality of self-attention modules are connected end to end, and each self-attention module comprises a probability sparse self-attention sub-module and a self-attention distillation sub-module; the probability sparse self-attention sub-module is used for extracting the characteristic information of the first time sequence data, and the self-attention distillation sub-module is used for compressing the characteristic information to obtain a characteristic sequence;
and the decoder module is used for predicting and obtaining a fault diagnosis result corresponding to the historical monitoring data according to the characteristic sequence.
The electromechanical equipment fault diagnosis method provided by the embodiment of the invention has the following beneficial effects:
according to the method, the device and the system, the plurality of encoder modules and decoder modules are arranged in the fault diagnosis network model, and the probability sparse self-attention sub-module and the self-attention distillation sub-module are respectively arranged in each self-attention module, so that when the fault diagnosis network model is trained through time sequence data, the characteristic information of the time sequence data can be extracted through the probability sparse self-attention sub-module, the characteristic information is compressed through the self-attention distillation sub-module to obtain the characteristic sequence, the time complexity is reduced, the processing efficiency of long-sequence input data is effectively improved, and the scalability of the model is further expanded.
Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
The invention is further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a flow chart of a method for diagnosing faults of an electromechanical device according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a network model for fault diagnosis according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a self-focusing distillation sub-module according to an embodiment of the present invention;
fig. 4 is an application flowchart of an electromechanical device fault diagnosis method according to an embodiment of the present invention.
Detailed Description
Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are illustrative only and are not to be construed as limiting the invention.
In the description of the present invention, the meaning of a number is one or more, the meaning of a number is two or more, and greater than, less than, exceeding, etc. are understood to exclude the present number, and the meaning of a number is understood to include the present number. The description of the first and second is for the purpose of distinguishing between technical features only and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated or implicitly indicating the precedence of the technical features indicated.
In the description of the present invention, unless explicitly defined otherwise, terms such as arrangement, connection, etc. should be construed broadly and the specific meaning of the terms in the present invention can be reasonably determined by a person skilled in the art in combination with the specific contents of the technical solution.
In the description of the present invention, a description of the terms "one embodiment," "some embodiments," "an exemplary embodiment," "an example," "a particular example," or "some examples," etc., means that a particular feature or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
Under the intelligent manufacturing background, the working condition of the electromechanical equipment is more complex, the running condition is more severe, and the running state monitoring and fault diagnosis of the equipment are more important. As the length of service of the electromechanical device increases, the failure rate increases, resulting in increased maintenance costs, repair cycles, and miswork costs. In order to solve the problem of production efficiency reduction and even shutdown caused by sudden equipment failure, it is imperative to study fault diagnosis of electromechanical equipment.
According to the prior mechanical fault diagnosis practice, in the fault maintenance stage, corresponding maintenance personnel need to support through system connection, and judge whether the mechanical equipment has faults or not by means of feedback information received by the system. The judging of the fault type is a process of judging the fault type and the cause of the fault by means of the relevant feedback information on the basis of fault detection. After the fault type is correctly judged, the maintainer needs to further refine the type of the fault and the specific position of the fault according to the information obtained before and the past overhaul experience, so that preparation is made for fault maintenance. The fault maintenance is the process of processing the equipment after fault finding so as to enable the equipment to be reused, and is the most important step in the whole fault diagnosis task. The fault cause is found out in advance, and the personnel safety and the property safety can be ensured.
In the fault diagnosis process, yang Wucheng in the prior art utilizes a calculation order analysis method to resample a vibration signal angle domain, further utilizes Hilbert envelope demodulation to demodulate a low-frequency fault signal from a high-frequency carrier signal, further extracts fault characteristics, and finally achieves fault diagnosis by comparing fault characteristic coefficients. The main workflow of the method is to acquire vibration signals at fixed time intervals and acquire rotation speed information in real time, and then resample the vibration signals through a numerical interpolation algorithm according to rotation phases calculated by the tachometer pulse, so as to convert time domain vibration signals into angular domain vibration signals. And then carrying out Local Mean Decomposition (LMD) on the angle domain signals to obtain a plurality of product function components, and judging the fault type and damage degree of the electromechanical by analyzing the frequency spectrum of each product function component so as to carry out fault diagnosis. Although the calculation order tracking method gets rid of the limitation of hardware such as an analog filter and the like, saves cost and system complexity, the accuracy is seriously dependent on interpolation, and a tachometer is required to acquire speed information, so that the application of the method is limited.
Li et al in the prior art found that since the window size of the fast fourier transform is fixed, it is not adaptive to frequency. The wavelet transformation can well balance time resolution and frequency resolution through multi-resolution analysis by using a window function with a fixed area and a variable shape, and the essence of the wavelet transformation is that signals are filtered and weighted through a wavelet basis function, and the selection of the wavelet basis is a key for influencing the wavelet analysis effect. The main workflow of this method is to place an acceleration sensor over the motor as it runs to collect vibration information of the electromechanical device. It is expected that the vibration frequency of the failure damage for different electromechanical devices should also be different. And then, after the collected vibration signals are subjected to discrete wavelet transformation, calculating the variance of each scale signal, and finally estimating the multiscale slope characteristic based on the wavelet according to the slope of the logarithmic variance, thereby realizing the identification of fault types and the diagnosis of wear of electromechanical equipment. The diagnosis method has high time space calculation complexity for fault diagnosis of time sequence data and high memory requirement, and limits the scalability of the model when receiving long-sequence input.
Based on this, referring to fig. 1, an embodiment of the present invention provides a method for diagnosing an electromechanical device fault, where the method of the present embodiment may be applied to a cloud server, an edge server, or a processor corresponding to an electromechanical device fault diagnosis platform.
During application, the method of the embodiments of the present application includes, but is not limited to, the following steps:
step S110, historical monitoring data of the electromechanical device is obtained as first time sequence data.
In an embodiment of the present application, the first time sequence data includes normal data and different types of fault data of the electromechanical device. Specifically, the monitoring data of the present embodiment includes vibration data, current data, and rotational speed data of the electromechanical device. After a large amount of historical monitoring data is obtained, the embodiment performs normalization processing on the monitoring data to place partial data to influence the global effect. In the normalization processing, the processing result may be controlled to be within the [0,1] interval. Illustratively, let the original monitoring data be x, the maximum value of the data be max, the minimum value be min, the normalized output result be y, the normalization formula of this embodiment is shown in formula (1):
in this embodiment, after normalization processing is performed on the original monitoring data, fast fourier transform is performed on the historical monitoring data after normalization processing, so as to obtain frequency domain data. Illustratively, when the normalized historical monitoring data is x (t), ω is the harmonic order, the fourier series of the historical monitoring data x (t) may be as shown in formula (2):
the fourier transform after converting x (t) into complex form is shown in equation (3):
because the frequency domain data corresponding to the converted monitoring data has symmetry, the implementation carries out inverse transformation on the data obtained by the formula (3), and then takes half output x' (t) to train the model through the formula (4):
in the formula (4), L is the length of the original monitoring data.
In the embodiment of the application, after the historical monitoring data is collected, noise reduction processing can be performed on the historical monitoring data, so that noise is reduced from a data source, and the influence of the noise on model accuracy in the subsequent model training process can be reduced.
Step S120, training a fault diagnosis network model through the first time sequence.
In this embodiment of the present application, the first time sequence data may also be replaced by data obtained by halving the original monitoring data. In this embodiment, after acquiring the time series data for training, the embodiment may split the time series data into a training set and a test set, where the splitting ratio may be that all the collected data is split in a ratio of 8:2. And then inputting the training set into the fault diagnosis network model for training, and inputting the testing set into the trained fault diagnosis network model for testing the prediction effect.
In the embodiment of the application, as shown in fig. 2, the fault diagnosis network model includes a plurality of encoder modules and decoder modules; the plurality of encoder modules are in a stacked form end-to-end, and each encoder module includes a plurality of self-attention modules, the plurality of self-attention modules are in a stacked form end-to-end, and each self-attention module includes a probabilistic sparse self-attention sub-module and a self-attention distillation sub-module. The probability sparse self-attention sub-module is used for extracting the characteristic information of the first time sequence data, and the self-attention distillation sub-module is used for compressing the characteristic information to obtain a characteristic sequence; and the decoder module is used for predicting and obtaining fault diagnosis results corresponding to the historical monitoring data according to the feature sequences.
Specifically, the probability sparse self-attention sub-module of the embodiment can make each attention mechanism to optimize different characteristic parts of each time sequence data, so as to balance possible deviation generated by the same attention mechanism, thereby improving the efficiency of the model and reducing the space-time complexity. The self-attention distillation submodule can strengthen the weight of fault characteristics, shorten the input length of each layer and reduce the space complexity so as to reduce the calculation cost.
It can be understood that the probability sparse self-attention submodule of this embodiment has sparsity, the distribution is long-tail distribution, only a few dot products contribute to the attention of the main fault feature, and other dot products have very weak effects and can be ignored. Based on such results, ith query's attention (attracting the attention of the interrogator) was introduced by formula (5):
in the formula (5) of the present invention,
is the distribution of P, h i Representing query i ，l j Representing keys i . The element consists of an address key and a value, with the key i :query i To find the value of the corresponding item. Equation (5) represents the form of the attention coefficient of the ith query.
In the present embodiment, KL divergence (relative entropy) is used to evaluate the i-th query sparsity: the difference between the distribution of the attention distribution and the uniform distribution of the fault signature is evaluated. The larger the KL divergence, the larger the distance between the attention distribution and the uniform distribution, which indicates the fault characteristics, the larger the difference, which indicates that the unimodal, i.e., sparsity, is more likely. So compared to the results subject to uniform distribution (KL divergence between the two distributions), each key and the attention score of the first u domino queries are shown in the following equation (6):
equation (6) is to scale the input. Variables in equation (6)
Is a sparse matrix of the same shape as H, including only top-u queries, with the others being zero. H represents query, L represents key, Q represents value, wherein,
in this formula, u in top-u takes the value of u, u=clnlh. I.e. only top-u queries on the logarithmic level of LH are needed, thus ensuring that the final time complexity is reduced. Under the multi-headed view, this attention generates a different sparse "query-key pair" for each head, avoiding serious information loss in return.
It will be appreciated that the input to each encoder module in this embodiment is processed electromechanical device timing data and a corresponding time stamp. In this embodiment, before inputting the time series data and the time stamp into the encoder module, as shown in fig. 2, a convolution layer Convld is passed, so that the time series data and the time stamp can be convolved on the convolution layer, thereby obtaining a representation with a size of l×b, and then adding the convolution results of the two to be used as the input of the encoder module.
In this embodiment, after the probability sparse self-attention sub-module performs feature extraction on the convolved time series data, the output of each probability sparse self-attention sub-module is subjected to the operation of one self-attention distillation sub-module to reduce the time-space complexity of the fault feature extraction. Specifically, as shown in fig. 3, the self-attention distillation sub-module includes a convolution layer, an activation layer, and a max-pooling layer; the convolution layer is connected with the maximum pooling layer through the activation layer, and the step size of the maximum pooling layer is 2. The output of the probability sparse self-attention sub-module is firstly input into a Conv1d convolution layer, then passes through an ELU activation layer and a maximum pooling layer with the step length of 2, so that the original fault sequence is sequentially halved into 1/4L, and then is spliced with a Feature map with the length of 1/4L to be used as the output of the encoder module. The specific processing procedure is shown in the formula (7):
in the formula (7) of the present invention,
including multi-head probspare self-intent and key operations in intent block, convld represents a one-dimensional convolution operation over a time sequence and is activated by the ELU as a function of activation.
In the embodiment of the present application, the decoder module sets an input as shown in formula (8) for the algorithm to generate a long sequence of outputs:
in equation (8), the masked multi-head position is applied to the computation of probspark self-position to prevent each position from noticing the next position, thereby avoiding autoregressive. Finally, a fully connected layer obtains the final output, the output dimension of which depends on whether univariate prediction or multivariate prediction is being performed. Concat is the full connectivity layer.
In this formula, the first half is a time series of the collected electromechanical device input, defined as "Start token", and the second half is the electromechanical device failure part that needs to be predicted, where the gradient is filled with 0. And then the sequence passes through a mask probability sparse self-care layer, and finally passes through a full connection layer to obtain a final fault classification result. The generated decoder used by us can output all prediction results at one time, and time-consuming dynamic decoding does not exist.
In this embodiment, the model parameter states are estimated by the cross entropy loss function while the fault diagnosis network model is being trained. Wherein the cross entropy loss function is as shown in formula (9):
in equation (9), for a single sample,y represents the true distribution of the light,
representing the network output distribution, n represents the total number of categories.
Specifically, the cross entropy loss function is a good choice for fault diagnosis classification, and meanwhile, the loss function is used when the acquired electromechanical equipment sample data set is unbalanced, so that the effect is very good, and the robustness is relatively strong.
Step S130, current monitoring data of the electromechanical equipment are obtained;
and step 140, inputting the current monitoring data into the trained fault diagnosis network model to obtain a fault diagnosis result of the electromechanical equipment.
In this embodiment, after the training of the fault diagnosis network model is completed, the fault diagnosis network model is applied to an actual fault diagnosis process, so that diagnosis efficiency and diagnosis accuracy can be improved.
In some embodiments, when training the fault diagnosis network model, as shown in fig. 4, the following steps may be included, but are not limited to:
collecting monitoring data of electromechanical equipment;
normalizing the collected monitoring data;
performing Fast Fourier Transform (FFT) on the normalized monitoring data, and converting the normalized monitoring data to obtain corresponding frequency domain data;
dividing the frequency domain data into a training set and a testing set;
training the fault diagnosis network model through the training set, and obtaining a classification result of data in the training set;
judging whether the epoch is finished according to the classification result, if the epoch is not finished, continuing to train the fault diagnosis network model through the training set, otherwise, determining that the model training is finished, inputting the testing set into the trained fault diagnosis network model, acquiring the fault classification result corresponding to the testing set, and outputting the classification result in a visual mode.
In summary, according to the electromechanical device fault diagnosis method provided by the embodiment of the application, the special processing of the graph structure is converted into the time sequence data, the feature information is extracted through the probability sparse self-attention module, the feature compression is performed through the operation of the self-attention distillation module, and the encoder module improves the robustness of the algorithm by stacking the two modules. A series of long sequence inputs are then received by the decoder module and the predicted target position is filled with 0's, and the final step generates a predicted output by the masked attention layer. Based on these processes, the present embodiment can realize the following advantageous effects:
first, feature extraction and noise reduction of long time series are solved: the local time sequence features are converted into frequency domain data, so that the nodes can learn the local features through a plurality of surrounding nodes, noise problems caused by dependence on fewer time sequence features are different, and feature extraction is realized through the encoder with sparse probability and self-attention.
Second, a low complexity attention model is implemented: the encoder module provided by the embodiment is used for carrying out long-sequence prediction, a probability sparse self-attention mechanism and a self-attention distillation mechanism are designed in the encoder module, and the space-time complexity is reduced. And meanwhile, the generating type decoder is used to improve algorithm efficiency. The accuracy is improved while the complexity is low, and an efficient fault diagnosis model is realized.
The embodiment of the invention provides an electromechanical equipment fault diagnosis system, which comprises:
the first module is used for acquiring historical monitoring data of the electromechanical equipment as first time sequence data, wherein the first time sequence data comprises normal data and fault data of different types of the electromechanical equipment;
a second module for training a fault diagnosis network model by the first time sequence;
the third module is used for acquiring current monitoring data of the electromechanical equipment;
a fourth module, configured to input current monitoring data to the trained fault diagnosis network model, to obtain a fault diagnosis result of the electromechanical device;
the fault diagnosis network model comprises a plurality of encoder modules and decoder modules;
the plurality of encoder modules are connected end to end, each encoder module comprises a plurality of self-attention modules, the plurality of self-attention modules are connected end to end, and each self-attention module comprises a probability sparse self-attention sub-module and a self-attention distillation sub-module; the probability sparse self-attention sub-module is used for extracting the characteristic information of the first time sequence data, and the self-attention distillation sub-module is used for compressing the characteristic information to obtain a characteristic sequence;
and the decoder module is used for predicting and obtaining a fault diagnosis result corresponding to the historical monitoring data according to the characteristic sequence.
The content of the method embodiment of the invention is suitable for the system embodiment, the specific function of the system embodiment is the same as that of the method embodiment, and the achieved beneficial effects are the same as those of the method.
The embodiments of the present invention have been described in detail with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of one of ordinary skill in the art without departing from the spirit of the present invention. Furthermore, embodiments of the invention and features of the embodiments may be combined with each other without conflict.
Claims (10)
1. A method for diagnosing a fault in an electromechanical device, comprising the steps of:
acquiring historical monitoring data of the electromechanical equipment as first time sequence data, wherein the first time sequence data comprises normal data and fault data of different types of the electromechanical equipment;
training a fault diagnosis network model through the first time sequence;
acquiring current monitoring data of the electromechanical equipment;
inputting the current monitoring data into the trained fault diagnosis network model to obtain a fault diagnosis result of the electromechanical equipment;
the fault diagnosis network model comprises a plurality of encoder modules and decoder modules;
the plurality of encoder modules are connected end to end, each encoder module comprises a plurality of self-attention modules, the plurality of self-attention modules are connected end to end, and each self-attention module comprises a probability sparse self-attention sub-module and a self-attention distillation sub-module; the probability sparse self-attention sub-module is used for extracting the characteristic information of the first time sequence data, and the self-attention distillation sub-module is used for compressing the characteristic information to obtain a characteristic sequence;
and the decoder module is used for predicting and obtaining a fault diagnosis result corresponding to the historical monitoring data according to the characteristic sequence.
2. The method for diagnosing a malfunction of an electromechanical device according to claim 1, wherein after the acquisition of the historical monitoring data of the electromechanical device as the first time series data, the method further comprises the steps of:
normalizing the first time sequence data;
performing fast Fourier change on the history monitoring data after normalization processing to obtain frequency domain data;
taking half of the frequency domain data to perform inverse fourier transform.
3. The method of claim 1, wherein the network model further comprises a multi-scale convolution module for convolving and adding the first time series data and the corresponding time stamps.
4. The method of claim 1, wherein the self-attention distillation sub-module comprises a convolution layer, an activation layer, and a max-pooling layer; the convolution layer is connected with the maximum pooling layer through the activation layer, and the step length of the maximum pooling layer is 2.
5. The method of claim 1, wherein the decoder module comprises a generative decoder.
6. The method of claim 1, wherein training a fault diagnosis network model by the first time sequence comprises:
dividing the first time series data into a training set and a testing set;
and after the training set is input into the fault diagnosis network model for training, the test set is input into the trained fault diagnosis network model for testing the prediction effect.
7. The method for diagnosing a fault in an electromechanical device according to claim 6, wherein when the fault diagnosis network model is trained, the method further comprises the steps of:
and calculating a cross entropy loss function corresponding to the fault diagnosis network model.
8. The method of claim 7, wherein the cross entropy loss function is formulated as:
wherein y represents the true distribution,
representing the network output distribution, n represents the total number of categories.
9. The method of claim 2, wherein the normalizing operation is performed on the first time-series data by the following formula:
wherein x represents first time sequence data, y represents normalized data, max represents the maximum value of all the first time sequence data, and min represents the minimum value of all the first time sequence data.
10. An electromechanical device fault diagnosis system, comprising:
the first module is used for acquiring historical monitoring data of the electromechanical equipment as first time sequence data, wherein the first time sequence data comprises normal data and fault data of different types of the electromechanical equipment;
a second module for training a fault diagnosis network model by the first time sequence;
the third module is used for acquiring current monitoring data of the electromechanical equipment;
a fourth module, configured to input the current monitoring data to the trained fault diagnosis network model, to obtain a fault diagnosis result of the electromechanical device;
the fault diagnosis network model comprises a plurality of encoder modules and decoder modules;
the plurality of encoder modules are connected end to end, each encoder module comprises a plurality of self-attention modules, the plurality of self-attention modules are connected end to end, and each self-attention module comprises a probability sparse self-attention sub-module and a self-attention distillation sub-module; the probability sparse self-attention sub-module is used for extracting the characteristic information of the first time sequence data, and the self-attention distillation sub-module is used for compressing the characteristic information to obtain a characteristic sequence;
and the decoder module is used for predicting and obtaining a fault diagnosis result corresponding to the historical monitoring data according to the characteristic sequence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310126200.0A CN116127297A (en) | 2023-02-15 | 2023-02-15 | Electromechanical device fault diagnosis method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310126200.0A CN116127297A (en) | 2023-02-15 | 2023-02-15 | Electromechanical device fault diagnosis method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116127297A true CN116127297A (en) | 2023-05-16 |
Family
ID=86302681
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310126200.0A Pending CN116127297A (en) | 2023-02-15 | 2023-02-15 | Electromechanical device fault diagnosis method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116127297A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116893924A (en) * | 2023-09-11 | 2023-10-17 | 江西南昌济生制药有限责任公司 | Equipment fault processing method, device, electronic equipment and storage medium |
-
2023
- 2023-02-15 CN CN202310126200.0A patent/CN116127297A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116893924A (en) * | 2023-09-11 | 2023-10-17 | 江西南昌济生制药有限责任公司 | Equipment fault processing method, device, electronic equipment and storage medium |
| CN116893924B (en) * | 2023-09-11 | 2023-12-01 | 江西南昌济生制药有限责任公司 | Equipment fault processing method, device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109977920B (en) | Water turbine set fault diagnosis method based on time-frequency spectrogram and convolutional neural network | |
| CN105115594B (en) | Gear-box vibration signal fault signature extracting method based on Wavelet Entropy and information fusion | |
| CN109829402A (en) | Different operating condition lower bearing degree of injury diagnostic methods based on GS-SVM | |
| CN105275833A (en) | CEEMD (Complementary Empirical Mode Decomposition)-STFT (Short-Time Fourier Transform) time-frequency information entropy and multi-SVM (Support Vector Machine) based fault diagnosis method for centrifugal pump | |
| CN104807534B (en) | Equipment eigentone self study recognition methods based on on-line vibration data | |
| CN110398647B (en) | Transformer state monitoring method | |
| JP2003526859A (en) | Decompose and model complex signals | |
| CN110501631A (en) | A kind of online intermittent fault detection and diagnostic method | |
| CN113158984B (en) | Bearing fault diagnosis method based on complex Morlet wavelet and lightweight convolution network | |
| CN111170103A (en) | Equipment fault identification method | |
| CN113139278A (en) | Deep learning-based bearing residual life prediction method and system | |
| CN116127297A (en) | Electromechanical device fault diagnosis method and system | |
| CN112633098A (en) | Fault diagnosis method and system for rotary machine and storage medium | |
| CN110133500A (en) | Motor on-line monitoring and failure omen diagnostic system and method based on multi-layer framework | |
| CN114034481A (en) | Fault diagnosis system and method for rolling mill gearbox | |
| CN111678699B (en) | Early fault monitoring and diagnosing method and system for rolling bearing | |
| CN113391207A (en) | Motor fault detection method, medium and system | |
| Lu et al. | Early fault warning and identification in condition monitoring of bearing via wavelet packet decomposition coupled with graph | |
| CN114330430A (en) | Elevator fault judgment method and system based on big data characteristic analysis | |
| Mubaraali et al. | Intelligent fault diagnosis in microprocessor systems for vibration analysis in roller bearings in whirlpool turbine generators real time processor applications | |
| CN113758709A (en) | Rolling bearing fault diagnosis method and system combining edge calculation and deep learning | |
| CN117473414A (en) | Bearing fault position identification method based on low-noise time-frequency image | |
| CN110222390B (en) | Gear crack identification method based on wavelet neural network | |
| CN110044619B (en) | Multi-fault feature identification method based on sparse multi-cycle group lasso | |
| CN113822565B (en) | Method for graded and refined analysis of time-frequency characteristics of fan monitoring data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |