CN115859077A - Multi-feature fusion motor small sample fault diagnosis method under variable working conditions - Google Patents
Multi-feature fusion motor small sample fault diagnosis method under variable working conditions Download PDFInfo
- Publication number
- CN115859077A CN115859077A CN202211405623.8A CN202211405623A CN115859077A CN 115859077 A CN115859077 A CN 115859077A CN 202211405623 A CN202211405623 A CN 202211405623A CN 115859077 A CN115859077 A CN 115859077A
- Authority
- CN
- China
- Prior art keywords
- fault
- data
- data sample
- rotating speed
- sample
- 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
Abstract
The application discloses motor small sample fault diagnosis method of multi-feature fusion under variable working conditions, which comprises the following steps: obtaining vibration signal data and rotating speed data of a motor in a running state to obtain a data sample to be identified; acquiring historical fault vibration signal data of motors of the same model and corresponding rotating speed data of the motors, and performing data enhancement on the historical fault vibration signal data according to the rotating speed corresponding to the data sample to be identified to obtain a fault data sample set; constructing a data sample pair based on a data sample to be identified and a fault data sample set, inputting a motor fault identification model trained in advance to obtain the Euclidean distance of the sample to the characteristic, and determining the motor fault type according to the Euclidean distance of the sample to the characteristic; the motor fault recognition model is obtained by adopting a twin network and training by using a small sample-based meta-learning strategy.
Description
Technical Field
The application relates to the field of motor fault diagnosis and industrial artificial intelligence, in particular to a multi-feature fusion motor small sample fault diagnosis method under variable working conditions.
Background
The motor fault diagnosis is widely applied to a plurality of national economy key fields such as manufacturing industry, energy, rail transit and the like. The motor inevitably breaks down in the operation process, if the motor operates abnormally in the whole production system, the whole system can be stopped, the production efficiency is influenced and economic loss is caused, and serious safety accidents and personal safety are threatened. Therefore, the motor running state is monitored and diagnosed in real time, problems can be found in time, targeted maintenance work can be carried out, and serious loss is avoided.
One of the limitations in the development of the current motor fault diagnosis technology is that fault data acquisition is difficult, and the reasons for this condition are mainly two kinds: (1) The industrial system is not allowed to be in a failure state in actual production, and the adverse result is caused, especially for a key system; (2) Most electromechanical systems fail slowly and follow a degradation path, failure degradation of the system can take months or even years, and failure formation periods are long. The method has the advantages that the device fault data which can be acquired in the actual production scene is limited, so that the problems of poor fitting, weak generalization capability and the like of a diagnosis model are easily caused. Therefore, developing intelligent fault diagnosis research under small sample conditions is of great significance for processing practical engineering problems.
On the other hand, the motor is widely applied to various production scenes as a driving device, in the actual production process, the motor often involves the change of rotating speed, when the motor operates at different rotating speeds, signals generated by the device and characteristics of the signals can also change along with the change of the rotating speed, and due to the complexity of the operating environment of the device, the current diagnosis model is often only suitable for a single working condition, and the limitation is too large in the actual use, so that the establishment of the diagnosis model with strong generalization has important significance.
In recent years, deep learning is paid extensive attention by virtue of intelligent feature extraction and state recognition capability, and dependence on inefficient and incomplete manual analysis in traditional fault diagnosis is avoided. Although the neural network model can be used for extracting general features in signals in a self-adaptive manner, the neural network model does not have the fault mechanism characteristics, important fault information can be omitted, and a large number of training samples are needed to obtain a good feature extraction effect.
Disclosure of Invention
The application aims to provide a motor small sample fault diagnosis method based on multi-feature fusion under variable working conditions, which can be suitable for motor fault diagnosis under limited fault sample data and improve the accuracy of motor fault identification under the conditions of small samples and variable working conditions. The technical scheme adopted by the invention is as follows.
On the one hand, the application provides a motor small sample fault diagnosis method of multi-feature fusion under variable working conditions, including:
acquiring vibration signal data and rotating speed data of a motor in a running state, and taking the vibration signal data as a to-be-identified data sample;
acquiring historical fault vibration signal data of motors of the same model and corresponding rotating speed data of the motors, and performing data enhancement processing on the historical fault vibration signal data according to the rotating speed corresponding to the data sample to be identified to obtain a fault data sample set of the corresponding rotating speed;
taking the data sample to be identified and the fault data sample in the fault data sample set as a data sample pair;
inputting the data sample pair into a motor fault recognition model trained in advance to obtain the Euclidean distance of the data sample pair to the characteristics;
determining the motor fault type according to the Euclidean distance of the data sample pair characteristics;
the motor fault identification model is obtained by inputting training data sample pairs into a pre-constructed twin network model for training; the training data sample pair is obtained by the following steps:
acquiring historical fault vibration signal data and corresponding rotating speed data of motors of the same type, and performing data enhancement processing on the acquired historical fault vibration signal data to obtain at least one fault data sample set, wherein each fault data sample set corresponds to one rotating speed and comprises a plurality of fault data samples corresponding to a plurality of fault types;
randomly selecting a first preset number of fault types from any fault data sample set, and randomly extracting a second preset number of fault data samples for each selected fault type;
and dividing a support set and a query set from the extracted fault data samples, and constructing the training data sample pairs by using the fault data samples in the support set.
Optionally, performing data enhancement processing on the historical fault vibration signal data according to the rotating speed corresponding to the data sample to be identified, including:
and carrying out frequency domain migration processing on the historical fault vibration signal data according to a first rotating speed corresponding to the data sample to be identified and a second rotating speed corresponding to the historical fault vibration signal data to obtain a plurality of fault vibration signal data with the rotating speed changed to the first rotating speed.
Optionally, the frequency domain migration processing is performed according to the following formula:
where x (t) is the original vibration signal before frequency domain shift, x ω0 (t) represents the vibration signal corresponding to the designated rotation speed after the frequency domain migration, A represents the amplitude correction coefficient, delta N is the rotation speed difference between the rotation speed corresponding to the original vibration signal and the designated rotation speed, and t is timeAnd (3) removing the solvent.
Optionally, the pre-constructed twin network model sequentially includes a feature extraction module, a multi-feature fusion module, and a distance measurement module; the characteristic extraction module comprises two identical characteristic extraction networks which are respectively used for carrying out characteristic extraction on two data samples in the data sample pair; the characteristic extraction networks comprise deep characteristic extraction networks and prior characteristic extraction networks, and the characteristics extracted by the prior characteristic extraction networks comprise time domain characteristics, frequency domain characteristics and typical fault frequency characteristics of motor parts;
the multi-feature fusion module is used for carrying out feature fusion processing on features extracted by the deep feature extraction network and the prior feature extraction network;
the measurement distance module is used for calculating the Euclidean distance of the fused features of the two data samples in the data sample pair.
Optionally, the deep feature extraction network includes a convolutional neural network, the convolutional neural network includes at least 2 convolutional layers, and a convolution kernel of the first convolutional layer is larger than a convolution kernel of a later convolutional layer.
Furthermore, the convolutional neural network is composed of 4 one-dimensional convolutional layers, 4 maximum pooling layers and a full-connection layer, wherein one maximum pooling layer is connected behind each convolutional layer; a Flatten layer and a Dropout layer are connected behind the last layer of the maximum pooling layer, and then a full-connection layer is connected;
the activation function of the convolutional layer adopts a Relu function, and the activation function of the full-connection layer adopts a Sigmoid function.
Optionally, the time-domain feature of the vibration signal extracted by the prior feature extraction network includes at least one of a mean value, a variance, a standard deviation, a peak-to-peak value, a skewness, a root mean square, a peak index, a waveform index, a pulse index, a square root amplitude, an absolute average value, a margin index, and a kurtosis;
the frequency domain features comprise at least one of center of gravity frequency, mean square frequency, root mean square frequency, frequency variance and frequency standard deviation;
the fault characteristics comprise at least one of outer ring fault frequency, inner ring fault frequency, rolling body fault frequency and retainer fault frequency.
Optionally, the multi-feature fusion module performs feature fusion processing on features extracted by the deep feature extraction network and the prior feature extraction network, and the processing includes:
normalizing the characteristic value extracted by the prior characteristic extraction network;
and fusing the feature data after the normalization processing with the feature data Feat extracted by the deep feature extraction network to obtain a fused feature set Fea = { Feat, fea1, fea2, fea3}, wherein Fea1, fea2 and Fea3 respectively represent the time domain feature, the frequency domain feature and the fault feature after the normalization processing.
Optionally, the distance measuring module is configured to measure an euclidean distance between feature vectors output by two feature extraction network modules in the twin network;
the determining of the motor fault type according to the Euclidean distance of the sample pair features comprises the following steps:
calculating the probability that two data samples in the data sample pair are the same according to the Euclidean distance;
calculating the sum of the sample identity probabilities between the data sample to be identified and all historical fault data samples under each fault type as a comprehensive identity probability;
and taking the fault type with the maximum comprehensive same probability as the fault type of the data sample to be identified.
In a second aspect, the application provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the method as introduced in the first aspect.
In a third aspect, the present application provides an electronic device, comprising:
one or more processors;
a memory configured to store one or more programs,
wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method introduced in the first aspect.
Advantageous effects
According to the motor fault diagnosis method, from the perspective of actual operation condition change conditions of the motor and generalization capability of the diagnosis model, frequency domain migration processing is carried out on an original fault sample according to motor mechanism characteristics, migration of motor state information from one working condition to other working conditions is achieved, and further motor signal data enhancement under the variable working conditions is achieved, so that accurate fault identification can be carried out on the motor with less fault sample data under the variable working conditions, the engineering application range of the diagnosis model is expanded, the problem that the identification precision of the transmission diagnosis model is low under the complex working conditions of variable rotating speed of equipment and the like is solved, and the accuracy of motor faults under the limited condition of the fault sample can be effectively improved.
Meanwhile, the application provides a twin neural network model based on the fusion of the wide-kernel convolutional neural network and the priori knowledge characteristics by combining the deep learning theory and the priori knowledge, and the general characteristics and the priori knowledge characteristics of the samples in the signals are obtained, and the signals are subjected to multi-characteristic fusion, so that the model can obtain the characteristic information which can represent the equipment state more under the condition of limited samples, the model training efficiency and the convergence speed are improved, and the problem that a large amount of fault sample data is needed in the existing intelligent fault diagnosis model training is solved
Drawings
FIG. 1 is a schematic flow chart showing an implementation process of a multi-feature fusion motor small sample fault diagnosis method under variable working conditions according to an embodiment of the present application
FIG. 2 is a schematic structural diagram of a small sample learning twin model in the method according to the embodiment of the present disclosure;
fig. 3 is a diagram illustrating a small sample learning strategy.
Detailed Description
The following further description is made in conjunction with the accompanying drawings and the specific embodiments.
Referring to fig. 1, an embodiment of the present application introduces a method for diagnosing a fault of a multi-feature fused small sample of a motor under variable working conditions, including:
acquiring vibration signal data and rotating speed data of a motor in a running state, and taking the vibration signal data as a to-be-identified data sample;
acquiring historical fault vibration signal data of motors of the same model and corresponding rotating speed data of the motors, and performing data enhancement processing on the historical fault vibration signal data according to the rotating speed corresponding to the data sample to be identified to obtain a fault data sample set of the corresponding rotating speed;
constructing a data sample pair based on the data sample to be identified and the fault data sample in the fault data sample set;
inputting the data sample pairs into a motor fault recognition model trained in advance to obtain characteristic Euclidean distances of the data sample pairs;
determining the motor fault type according to the characteristic Euclidean distance of the data sample pair;
the motor fault identification model is obtained by inputting training data sample pairs into a pre-constructed twin network model for training; the training data sample pair is obtained by the following steps:
acquiring historical fault vibration signal data and corresponding rotating speed data of motors of the same model, and performing data enhancement processing on the acquired historical fault vibration signal data to obtain at least one fault data sample set, wherein each fault data sample set corresponds to one rotating speed and comprises a plurality of fault data samples corresponding to a plurality of fault types;
randomly selecting a first preset number of fault types from any fault data sample set, and randomly extracting a second preset number of fault data samples for each selected fault type;
and dividing a support set and a query set from the extracted fault data samples, and constructing the training data sample pairs by using the fault data samples in the support set.
Referring to fig. 1 and 2, a motor fault recognition model provided in an embodiment of the present application is obtained by training a twin network model, where the twin network model sequentially includes a feature extraction module, a multi-feature fusion module, and a distance measurement module;
the characteristic extraction module comprises two same characteristic extraction networks which are respectively used for carrying out characteristic extraction on two data samples in the data sample pair; the characteristic extraction networks comprise deep characteristic extraction networks and prior characteristic extraction networks, wherein the deep characteristic extraction networks use deep convolutional networks, and the characteristics extracted by the prior characteristic extraction networks comprise time domain characteristics, frequency domain characteristics and typical fault frequency characteristics of motor parts;
the multi-feature fusion module is used for carrying out feature fusion processing on features extracted by the deep feature extraction network and the prior feature extraction network; the measurement distance module is used for calculating the Euclidean distance of the fused features of the two data samples in the data sample pair.
The deep feature extraction network comprises a convolutional neural network, which is composed of four one-dimensional convolutional layers, four maximum pooling layers and a full-connection layer, wherein each convolutional layer is followed by one maximum pooling layer; the last maximum pooling layer is followed by a Flatten layer and a Dropout layer followed by a full-link layer. The model details are as follows: the first convolution layer uses wide-Kernel convolution with Kernel Size of 64 × 1 to extract features, the number of convolution kernels is 16, the second convolution layer, the third convolution layer and the fourth convolution layer use narrow-Kernel convolution with Kernel Size of 3*1 to obtain more detailed fault features, the number of convolution kernels is 32, 64 and 64 respectively, and the activation function of the convolution layers adopts a Relu function; the four maximum pooling layers adopt the pooling windows of 2*1 in size, and the number of corresponding convolution kernels is 16, 32, 64 and 64 respectively; the Dropout layer probability is set to 0.2 to alleviate the over-fitting problem, the number of neurons in the fully connected layer is 100, and the Sigmoid function is adopted as the activation function. The convolution process of a convolutional neural network is represented as:
in the formula (I), the compound is shown in the specification,
denotes the ith in the intermediate layer lThe output of the neuron is selected>
Represents the output value of the previous layer, is greater than or equal to>
Represents a convolution kernel, <' > based on>
Representing the deviation value of the current layer, representing convolution operation, and f (representing nonlinear activation function calculation);
the expression for the max pooling layer is:
Z l ＝f(W l *max(x l-1 )+b l )
in the formula, Z l Denotes the output of the l-th layer, W l Represents the weight, x, of the l-th layer l-1 Represents the output of the previous layer, b l Represents the bias of the l-th layer;
the activation function of the convolutional layer is expressed as:
Relu(x)＝max{0，x}
the activation function of the fully-connected layer is expressed as:
the method of embodiments of the present application specifically relates to the following:
1. pre-training of motor fault recognition models
1.1 historical failure data sample Collection
In the embodiment of the application, historical fault vibration signal data and corresponding rotating speed data of at least one motor of the same type are collected for motors of any type, data enhancement processing is carried out on the collected historical fault vibration signal data to obtain at least one fault data sample set, and each fault data sample set corresponds to one rotating speed respectively and comprises a plurality of fault data samples corresponding to multiple fault types.
And labeling the signal data acquired in the actual engineering according to the actual fault type to be used as a training label in the supervised training. In order to improve the recognition accuracy and the application range after model training, fault data sample sets corresponding to various rotating speed working conditions can be obtained for motors of various models, the fault type in each fault data sample set should cover the fault type possibly occurring in the motor as much as possible, and multiple data samples should be possible under each fault type. However, because the available fault data samples of the motor in practical application are very limited, the embodiment provides a data sample expansion method for motor variable conditions according to the operating characteristics of the motor to realize data enhancement.
1.2 data enhancement of Small samples
According to the embodiment of the application, the frequency domain migration processing is carried out on the original fault sample according to the mechanism characteristics of the motor, so that the motor signal data enhancement under the variable working condition is realized, and the realization mode is as follows: determining the rotating speed corresponding to the fault sample data of the obtained motor vibration acceleration signal and the target rotating speed of the variable working condition, calculating the rotating speed difference delta N of the two, further calculating the frequency difference, realizing the generation of new fault sample data by means of the frequency shift characteristic under Fourier transform, wherein the frequency spectrum of the newly generated signal is that the original signal x (t) is translated in the frequency spectrum F (omega) of the original signal x (t) 0 To F (omega-omega) 0 ) And obtaining vibration acceleration signal data of the motor under the working condition of the target rotating speed.
Specifically, frequency domain migration processing is performed on the existing fault sample according to the following formula, so that fault sample data under a target rotating speed is obtained, wherein the fault type of the fault sample is the same as that of the existing fault sample before frequency domain migration:
where x (t) is the original vibration signal before frequency domain shift, x ω0 (t) represents the vibration signal corresponding to the specified rotation speed after the frequency domain migration, and A represents the amplitude correction coefficient.
And for any rotating speed working condition of any type of motor, combining the original data collected in the corresponding engineering with the newly generated data to construct a fault data sample set. And combining fault data sample sets corresponding to various rotating speeds of motors of various types into a total training sample set.
1.3 feature extraction
Further, referring to fig. 2, the feature extraction module of the twin network model of this embodiment includes two identical feature extraction networks, which are respectively used for feature extraction on two data samples in the data sample pair; in the feature extraction network module, a deep layer feature extraction network extracts feature data Feat by using a deep layer convolution network, and the features extracted by the prior feature extraction network comprise a time domain feature Fea1, a frequency domain feature Fea2 and a typical fault frequency feature Fea3 of the motor component. Two feature extraction network modules of the feature extraction module have the same network structure and share weights. The feature extraction part is the combination of a convolutional neural network and a priori knowledge feature extraction module, and a calculation form of measuring distance is adopted as a classifier. The specific calculation method is as follows.
X in the above time domain features represents the original signal input; p (f) in the frequency domain features represents the power spectrum of the signal; r represents the motor bearing rotation speed; n represents the number of balls; d represents the diameter of the rolling body; d represents the pitch diameter of the bearing; α represents a rolling element contact angle.
1.4 Multi-feature fusion
Furthermore, when the multi-feature fusion module of the twin network model in the embodiment of the application performs feature fusion processing on features extracted by a deep feature extraction network and a prior feature extraction network, firstly, normalization processing is performed on feature values extracted by the prior feature extraction network; fusing the feature data after the normalization processing with feature data Feat extracted by a deep feature extraction network to obtain a fused feature set Fea = { Feat, fea1, fea2, fea3};
the normalization processing formula is as follows:
where x is the initial data to be normalized, x min Is the minimum value in the array, x max Is the maximum value in the array, x norm Is the result of normalization processing.
1.5 sample-to-distance metric
The distance measuring module is used for measuring the Euclidean distance between the feature vectors output by the two feature extraction network modules in the twin network, calculating the probability that the two data samples in the data sample pair are the same according to the Euclidean distance, and then determining the motor fault type according to the Euclidean distance of the features of the data sample pair.
The euclidean distance between the feature vectors is calculated as follows:
in the formula (f) 1 (. The) represents the processing of feature extraction and fusion networks,
representing the Euclidean distance between the feature vectors corresponding to the two data samples in the input data sample pair;
the probability that two data samples are identical is calculated as follows:
in the formula, FC (×) represents processing of the fully-connected layer network, and sigm (×) represents an activation function of the fully-connected layer network.
1.6 model training
The embodiment of the application provides a meta-learning strategy based on small samples, wherein a first preset number of fault categories are selected from a collected fault data sample set, and a second preset number of sample data are randomly selected for each selected fault category; and taking a third preset number of sample data in the selected second preset number of sample data as a meta-training support set, and taking the remaining sample data as a meta-training query set.
Will have the same or different categories
The set of sample pairs is used as the input of the model, the output of the model training is a probability distance used for judging whether two data samples in the sample pairs belong to the same class or different classes, and the output of the model is the probability p (x) that the two input samples are the same 1 ，y 1 )。
Referring to fig. 3, unlike the conventional classification method, the performance of the small sample learning of the present embodiment is tested by N samples K categories, as shown in fig. 3 (c). In the class K single sample test, a single test sample is classified and a support set S is given, the support set includes class K samples, each class sample has a different label y, as follows:
S＝{(x 1 ，y 1 )，...，(x K ，y K )}
samples in the support set have corresponding labels, and then the test samples are classified according to the most similar (shortest measurement distance) sample in the support set, as shown in the following formula:
in the N-sample K-class test, a support set of K different classes is given, each class having N samples (S) 1 ，...，S N ). The model then determines to which support set class (most similar to which support set synthesis) the test sample should belong, as shown in the following equation:
and training by using historical fault data under the working conditions of multiple motor models and multiple rotating speeds to obtain a trained motor fault recognition model.
The loss function in model training is selected as
It is a vector of length m containing a small number of tags. When/is>
And &>
When from the same fault class, take t j =1, otherwise t j =0, where j is the jth sample pair of the ith lot. The optimizer selects an Adam optimizer, calculates the self-adaptive learning rate of each parameter, and realizes the optimization of the network.
2. Motor fault identification process in practical application
When the method is actually used for identifying the fault of the motor running state, the vibration acceleration signal of the motor in the running state and the corresponding rotating speed data thereof can be obtained in real time, and the vibration signal data is used as a data sample to be identified;
then historical fault vibration signal data of the same type of motor and corresponding rotating speed data are obtained, frequency domain migration is carried out on the historical fault vibration signal data according to the current real-time rotating speed of the motor, generation of new fault data samples under variable working conditions is achieved, and then an expanded fault data sample set corresponding to the current real-time rotating speed is obtained;
constructing a data sample pair based on the data sample to be identified and the fault data sample set, and respectively forming the data sample pair by the data sample to be identified and each sample data in the fault data sample set; inputting the obtained data sample pairs into a trained motor fault recognition model, extracting the characteristics of two data samples in each sample pair, and calculating to obtain the Euclidean distance between the characteristics of each sample pair to obtain the sample identity probability corresponding to each data sample pair;
then calculating the sum of the sample identity probabilities between the data sample to be identified and all historical fault data samples under each fault type as a comprehensive identity probability;
and taking the fault type with the maximum comprehensive same probability as the fault type of the data sample to be identified, so as to obtain the real-time fault identification results of the motor, including the fault-free fault or the fault and the fault type under the fault condition.
In another aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method for diagnosing a fault of a multi-feature fused small sample of an electric motor under variable operating conditions as described in embodiment 1 is implemented.
An embodiment of the present application further provides an electronic device, which includes:
one or more processors;
a memory configured to store one or more programs,
wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the aforementioned multi-feature fused motor small sample fault diagnosis method under variable operating conditions.
In summary, according to the embodiments, on one hand, from the perspective of the actual operating condition change condition of the motor and the generalization capability of the diagnostic model, the frequency domain migration processing is performed on the original fault sample according to the motor operating mechanism characteristics, so that the migration of the motor state information from one operating condition to other operating conditions is realized, the engineering application range of the diagnostic model is expanded, and the problem of low identification precision of the transmission diagnostic model under complex operating conditions such as equipment variable speed is solved;
on the other hand, the twin neural network model based on the fusion of the wide-kernel convolutional neural network and the priori knowledge characteristics is provided and used for fault diagnosis and identification under small samples, general characteristics extracted by the convolutional network are fused with the priori knowledge characteristics which are more pertinent and interpretable in engineering application, so that the model can obtain characteristic information which can represent the equipment state under the condition of limited samples, the model training efficiency and the convergence speed are improved, the problem that the existing intelligent fault diagnosis model needs a large amount of fault sample data is solved, and the accuracy of motor faults under the condition of limited fault samples is effectively improved by combining the data enhancement technology of frequency domain migration.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. 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, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While the present invention has been described with reference to the particular illustrative embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but is intended to cover various modifications, equivalent arrangements, and equivalents thereof, which may be made by those skilled in the art without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (11)
1. A multi-feature fusion motor small sample fault diagnosis method under variable working conditions is characterized by comprising the following steps:
acquiring vibration signal data and rotating speed data of a motor in a running state, and taking the vibration signal data as a to-be-identified data sample;
acquiring historical fault vibration signal data of motors of the same model and corresponding rotating speed data of the motors, and performing data enhancement processing on the historical fault vibration signal data according to the rotating speed corresponding to the data sample to be identified to obtain a fault data sample set of the corresponding rotating speed;
taking the data sample to be identified and the fault data sample in the fault data sample set as a data sample pair;
inputting the data sample pair into a motor fault recognition model trained in advance to obtain the Euclidean distance of the data sample pair to the characteristics;
determining the motor fault type according to the Euclidean distance of the data sample to the characteristic;
the motor fault identification model is obtained by inputting training data sample pairs into a pre-constructed twin network model for training; the training data sample pair is obtained by the following steps:
acquiring historical fault vibration signal data and corresponding rotating speed data of motors of the same model, and performing data enhancement processing on the acquired historical fault vibration signal data to obtain at least one fault data sample set, wherein each fault data sample set corresponds to one rotating speed and comprises a plurality of fault data samples corresponding to a plurality of fault types;
randomly selecting a first preset number of fault types from any fault data sample set, and randomly extracting a second preset number of fault data samples for each selected fault type;
and dividing a support set and a query set from the extracted fault data samples, and constructing the training data sample pairs by using the fault data samples in the support set.
2. The method according to claim 1, wherein performing data enhancement processing on the historical fault vibration signal data according to the rotating speed corresponding to the data sample to be identified comprises:
and carrying out frequency domain migration processing on the historical fault vibration signal data according to a first rotating speed corresponding to the data sample to be identified and a second rotating speed corresponding to the historical fault vibration signal data to obtain a plurality of fault vibration signal data from the rotating speed to the first rotating speed.
3. The method of claim 2, wherein the performing the frequency domain migration process is performed according to the following equation:
where x (t) is the original vibration signal before frequency domain shift, x ω0 (t) representing a vibration signal corresponding to a specified rotating speed after frequency domain migration, A representing an amplitude correction coefficient, deltaN being a rotating speed difference between the rotating speed corresponding to the original vibration signal and the specified rotating speed, and t being time.
4. The method according to claim 1, wherein the pre-constructed twin network model comprises a feature extraction module, a multi-feature fusion module and a metric distance module in sequence;
the characteristic extraction module comprises two identical characteristic extraction networks which are respectively used for carrying out characteristic extraction on two data samples in the data sample pair; wherein:
the characteristic extraction networks comprise deep characteristic extraction networks and prior characteristic extraction networks, and the characteristics extracted by the prior characteristic extraction networks comprise time domain characteristics, frequency domain characteristics and typical fault frequency characteristics of motor parts;
the multi-feature fusion module is used for carrying out feature fusion processing on features extracted by the deep feature extraction network and the prior feature extraction network;
the measurement distance module is used for calculating the Euclidean distance of the fused features of the two data samples in the data sample pair.
5. The method of claim 4, wherein the deep feature extraction network comprises a convolutional neural network comprising at least 2 convolutional layers, the convolutional kernel of the first convolutional layer being larger than the convolutional kernels of the following convolutional layers.
6. The method of claim 5, wherein the convolutional neural network is comprised of 4 one-dimensional convolutional layers, 4 max pooling layers, and one fully-connected layer, each convolutional layer followed by one max pooling layer;
a Flatten layer and a Dropout layer are connected behind the last layer of the maximum pooling layer, and then a full-connection layer is connected; the Relu function is adopted as the activation function of the convolutional layer, and the Sigmoid function is adopted as the activation function of the full connection layer.
7. The method of claim 4, wherein the time-domain features of the prior feature extraction network extracted vibration signal include at least one of a mean, a variance, a standard deviation, a peak-to-peak value, a skewness, a root mean square, a peak indicator, a waveform indicator, a pulse indicator, a square root amplitude, an absolute mean, a margin indicator, and a kurtosis;
the frequency domain features include at least one of a center of gravity frequency, a mean square frequency, a root mean square frequency, a frequency variance, and a frequency standard deviation;
the fault characteristics comprise at least one of outer ring fault frequency, inner ring fault frequency, rolling body fault frequency and retainer fault frequency.
8. The method according to claim 4, wherein the multi-feature fusion module performs feature fusion processing on features extracted by a deep feature extraction network and a prior feature extraction network, and comprises:
normalizing the characteristic value extracted by the prior characteristic extraction network;
fusing the feature data after normalization processing with the feature data Feat extracted by the deep feature extraction network to obtain a fused feature set Fea = { Feat, fea1, fea2, fea3}; wherein, the first and the second end of the pipe are connected with each other,
and Fea1, fea2 and Fea3 respectively represent the time domain characteristic, the frequency domain characteristic and the fault characteristic after normalization processing.
9. The method of claim 4, wherein the metric distance module is configured to measure Euclidean distance between feature vectors output by two feature extraction network modules in the twin network;
the determining of the motor fault type according to the Euclidean distance of the sample pair features comprises the following steps:
calculating the probability that two data samples in the data sample pair are the same according to the Euclidean distance;
calculating the sum of the sample identity probabilities between the data sample to be identified and all historical fault data samples under each fault type as a comprehensive identity probability;
and integrating the fault types with the same maximum probability as the fault types of the data sample to be identified.
10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1-9.
11. An electronic device, comprising:
one or more processors;
a memory configured to store one or more programs,
wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of any of claims 1-9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211405623.8A CN115859077A (en) | 2022-11-10 | 2022-11-10 | Multi-feature fusion motor small sample fault diagnosis method under variable working conditions |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211405623.8A CN115859077A (en) | 2022-11-10 | 2022-11-10 | Multi-feature fusion motor small sample fault diagnosis method under variable working conditions |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115859077A true CN115859077A (en) | 2023-03-28 |
Family
ID=85663010
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211405623.8A Pending CN115859077A (en) | 2022-11-10 | 2022-11-10 | Multi-feature fusion motor small sample fault diagnosis method under variable working conditions |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115859077A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116699390A (en) * | 2023-04-20 | 2023-09-05 | 上海宇佑船舶科技有限公司 | Diesel engine set fault detection method and system |
| CN117404765A (en) * | 2023-12-14 | 2024-01-16 | 山东省人工智能研究院 | Air conditioner system fan fault diagnosis method and system under weak supervision condition and air conditioner |
| CN116699390B (en) * | 2023-04-20 | 2024-04-26 | 上海宇佑船舶科技有限公司 | Diesel engine set fault detection method and system |
-
2022
- 2022-11-10 CN CN202211405623.8A patent/CN115859077A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116699390A (en) * | 2023-04-20 | 2023-09-05 | 上海宇佑船舶科技有限公司 | Diesel engine set fault detection method and system |
| CN116699390B (en) * | 2023-04-20 | 2024-04-26 | 上海宇佑船舶科技有限公司 | Diesel engine set fault detection method and system |
| CN117404765A (en) * | 2023-12-14 | 2024-01-16 | 山东省人工智能研究院 | Air conditioner system fan fault diagnosis method and system under weak supervision condition and air conditioner |
| CN117404765B (en) * | 2023-12-14 | 2024-03-22 | 山东省人工智能研究院 | Air conditioner system fan fault diagnosis method and system under weak supervision condition and air conditioner |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Lei et al. | Fault diagnosis of wind turbine based on Long Short-term memory networks | |
| Zhang et al. | Machinery fault diagnosis with imbalanced data using deep generative adversarial networks | |
| Li et al. | Understanding and improving deep learning-based rolling bearing fault diagnosis with attention mechanism | |
| Han et al. | An adaptive spatiotemporal feature learning approach for fault diagnosis in complex systems | |
| Zhao et al. | Deep multi-scale convolutional transfer learning network: A novel method for intelligent fault diagnosis of rolling bearings under variable working conditions and domains | |
| CN111721535B (en) | Bearing fault detection method based on convolution multi-head self-attention mechanism | |
| CN111914883B (en) | Spindle bearing state evaluation method and device based on deep fusion network | |
| CN112763214B (en) | Rolling bearing fault diagnosis method based on multi-label zero-sample learning | |
| Miao et al. | A novel real-time fault diagnosis method for planetary gearbox using transferable hidden layer | |
| CN111459144A (en) | Airplane flight control system fault prediction method based on deep cycle neural network | |
| CN112308147A (en) | Rotating machinery fault diagnosis method based on integrated migration of multi-source domain anchor adapter | |
| CN116593157A (en) | Complex working condition gear fault diagnosis method based on matching element learning under small sample | |
| CN115358259A (en) | Self-learning-based unsupervised cross-working-condition bearing fault diagnosis method | |
| CN112504682A (en) | Chassis engine fault diagnosis method and system based on particle swarm optimization algorithm | |
| CN113188794A (en) | Gearbox fault diagnosis method and device based on improved PSO-BP neural network | |
| CN114091525A (en) | Rolling bearing degradation trend prediction method | |
| CN115859077A (en) | Multi-feature fusion motor small sample fault diagnosis method under variable working conditions | |
| Xu et al. | Cross-category mechanical fault diagnosis based on deep few-shot learning | |
| Di et al. | Fault diagnosis of rotating machinery based on domain adversarial training of neural networks | |
| CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
| Zhang et al. | Fault Diagnosis with Bidirectional Guided Convolutional Neural Networks Under Noisy Labels | |
| Techane et al. | Rotating machinery prognostics and application of machine learning algorithms: Use of deep learning with similarity index measure for health status prediction | |
| Guan et al. | Fault diagnosis of rolling bearing with imbalanced small sample scenarios | |
| Cheng et al. | A novel adversarial one-shot cross-domain network for machinery fault diagnosis with limited source data | |
| Ahsan et al. | Advanced Fault Diagnosis in Rotating Machines Using 2D Grayscale Images with Improved Deep Convolutional Neural Networks |
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 |