CN114881071A - Synchronous motor rotor winding turn-to-turn short circuit fault diagnosis method based on multi-source information - Google Patents

Abstract

The invention provides a multi-source information-based synchronous motor rotor winding turn-to-turn short circuit fault diagnosis method, which comprises the steps of collecting excitation current, rotor vibration and stator vibration signals of a synchronous motor working under normal and different rotor winding turn-to-turn short circuit working conditions in real time; converting the one-dimensional time sequence signal into a two-dimensional gray scale image by using a signal-image conversion method; respectively inputting two-dimensional gray images of an exciting current signal, rotor vibration and stator vibration into an optimal two-dimensional CNN model for training to obtain a fault diagnosis loss function and diagnosis accuracy of a single signal and determine basic trust distribution; and performing decision fusion on the fault diagnosis accuracy of the three evidence bodies by using a multi-source information fusion algorithm to realize final diagnosis. The method disclosed by the invention furthest retains the characteristics of the original signal, eliminates the influence of manually extracting the characteristics, and obviously improves the accuracy of diagnosing the turn-to-turn short circuit fault of the rotor winding of the synchronous generator.

Description

Synchronous motor rotor winding turn-to-turn short circuit fault diagnosis method based on multi-source information

Technical Field

The invention relates to the technical field of synchronous motor fault diagnosis, in particular to a synchronous motor rotor winding turn-to-turn short circuit fault diagnosis method based on multi-source information.

Background

The synchronous motor has a complex structure and a variable operating environment, so that faults occur frequently. The rotor winding turn-to-turn short circuit is one of the most common faults of the synchronous motor, when the synchronous motor has such faults, the excitation current is increased, the temperature of a short circuit point is increased, the vibration of a shaft system is increased, if the short circuit degree is expanded due to the fact that the short circuit point cannot be found and processed in time, the large shaft magnetization and the unit with one-point or two-point grounding fault of the rotor can be forced to stop, and the safe operation of a generator and the whole power grid is threatened. Therefore, the method for diagnosing the turn-to-turn short circuit fault of the rotor winding of the synchronous motor is researched, the fault diagnosis accuracy is improved, and the method has important practical significance on the operation and maintenance safety of a plant network.

The selection of the characteristic signals is the premise of realizing the diagnosis target, and domestic and foreign scholars propose to diagnose the turn-to-turn short circuit fault of the rotor winding by taking the shaft voltage, the stator circulating current, the exciting current, the expected electromagnetic power, the leakage flux and the unbalanced magnetic pull force as fault characteristic quantities. In recent years, the rapid development of deep learning provides a new idea for intelligent fault diagnosis. The deep learning can automatically extract the characteristics of the original signal, reduces the dependence on experience, and solves the problems that machine learning depends on expert experience, and is large in time consumption and poor in generalization capability. However, deep learning is mostly applied to motor bearing faults at present, the application in the aspect of motor rotor turn-to-turn short circuit fault diagnosis is less, and the diagnosis purpose is mostly realized by depending on single fault characteristics; the fault information reflected by the single signal is one-sided, and the acquisition of the single signal is easily affected by factors such as weak fault characteristics, sensor faults and variable environments, so that phenomena such as misjudgment and missed judgment occur. The D-S evidence theory is used as a main means for fusing uncertain information, and has been applied to motor fault diagnosis to a certain extent at present. Therefore, deep learning and a D-S evidence theory are combined, and the method has important significance for improving the accuracy of turn-to-turn short circuit fault diagnosis of the synchronous motor rotor.

Disclosure of Invention

The invention aims to provide a synchronous motor rotor winding turn-to-turn short circuit fault diagnosis method based on multi-source information, which furthest retains the characteristics of original signals, eliminates the influence of manually extracting the characteristics, avoids the phenomena of missing judgment, erroneous judgment and the like caused by the fault of a single sensor, and obviously improves the accuracy of the synchronous generator rotor winding turn-to-turn short circuit fault diagnosis.

In order to achieve the purpose, the invention provides the following scheme:

a synchronous motor rotor winding turn-to-turn short circuit fault diagnosis method based on multi-source information comprises the following steps:

s1, acquiring excitation current, rotor vibration and stator vibration signals of the synchronous motor in real time under normal and various rotor winding turn-to-turn short circuit fault operation conditions by using a data acquisition device, and dividing a training set and a test set according to a set proportion;

s2, converting the one-dimensional time sequence signals of the exciting current, the rotor vibration and the stator vibration collected in the step S1 into a two-dimensional gray scale image by using a signal-image conversion method;

s3, determining the network structure and convolution kernel size of the two-dimensional CNN classification model, training by using a training set, adjusting the CNN network structure and the hyper-parameter according to the loss function and accuracy rate change condition in the iteration process, training again, repeating the above processes, and determining the optimal two-dimensional CNN model;

s4, respectively inputting the two-dimensional gray scale images of the exciting current signal, the rotor vibration and the stator vibration into an optimal two-dimensional CNN model to obtain a fault diagnosis loss function and fault diagnosis accuracy based on a single signal of the exciting current signal, the rotor vibration signal and the stator vibration signal;

s5, determining an identification frame theta, wherein A represents any subset of the identification frame, recording the identification frame as A belonging to theta, corresponding to different fault operation conditions of the synchronous motor, and respectively inputting excitation current, rotor vibration and stator vibration signals into an optimal two-dimensional CNN model to obtain fault diagnosis accuracy as basic trust distribution;

and S6, taking the excitation current, the rotor vibration and the stator vibration signals as three evidence bodies, and performing decision fusion on the fault diagnosis accuracy of the three evidence bodies by using a multi-source information fusion algorithm.

Further, in step S1, the multiple operating conditions of the rotor winding inter-turn short circuit fault include 3%, 6%, and 10% of the rotor winding inter-turn short circuit; the set proportion is 8: 2.

further, in step S1, the data acquisition device includes an excitation current signal detection device, a rotor vibration acceleration sensor, and a stator winding axial vibration acceleration sensor.

Further, in step S2, the signal-image conversion method is to randomly extract the length N from the original one-dimensional time domain signal ² The signals with the length of N are selected each time to fill the pixels of the image line by line and then normalization is carried out, thus obtaining the two-dimensional gray image of NxN.

Further, in step S3, the optimal two-dimensional CNN model specifically includes: an input layer, a convolutional layer CL, a pooling layer PL, a full-link layer and an output layer; the convolutional layer CL comprises a convolutional layer CL1, a convolutional layer CL2, a convolutional layer CL3 and a convolutional layer CL4, and the pooling layer PL comprises a pooling layer PL1, a pooling layer PL2, a pooling layer PL3 and a pooling layer PL 4;

the input layer inputs a gray level image with the size of 64 multiplied by 1;

the convolution kernel size of the convolution layer CL1 is set to 5 × 5, the step size Stride is set to 1, zero Padding is 'same', the feature map size is 64 × 64, the feature map depth is 16, and the relu function is selected as the activation function;

the pooling layer PL1 adopts a maximum pooling method, the width of a pooling area is 2 multiplied by 2, and the size of a characteristic diagram is 32 multiplied by 32;

the convolution kernel size of the convolution layer CL2 is set to 3 × 3, the step size Stride is set to 1, zero Padding is 'same', the feature map size is 32 × 32, the feature map depth is 16, and the relu function is selected as the activation function;

the pooling layer PL2 adopts a maximum pooling method, the width of a pooling area is 2 x 2, and the size of a characteristic diagram is 16 x 16;

the convolution kernel size of the convolution layer CL3 is set to 3 × 3, the step size Stride is set to 1, zero Padding is 'same', the feature map size is 16 × 16, the feature map depth is 64, and the relu function is selected as the activation function;

the pooling layer PL3 adopts a maximum pooling method, the width of a pooling area is 2 x 2, and the size of a characteristic diagram is 8 x 8;

the convolution kernel size of the convolution layer CL4 is set to 3 × 3, the step size Stride is set to 1, zero Padding is 'same', the feature map size is 8 × 8, the feature map depth is 128, and the relu function is selected as the activation function;

the pooling layer PL4 adopts a maximum pooling method, the width of a pooling area is 2 multiplied by 2, and the size of a characteristic diagram is 4 multiplied by 4;

the number of the neurons is set to 512 by the full connection layer, and a relu function is selected by an activation function;

the output layer adopts a Softmax classifier, the output class is 4, and the output layer corresponds to four operation working conditions of normal operation and rotor winding turn-to-turn short circuit of 3%, 6% and 10% respectively.

Further, in step S5, determining an identification frame Θ, where a represents any subset of the identification frame, and is recorded as a ∈ Θ, and corresponding to different operating conditions of the synchronous motor, respectively inputting the excitation current, the rotor vibration, and the stator vibration signal into the optimal two-dimensional CNN model to train the obtained fault diagnosis accuracy as a basic trust distribution, specifically including:

firstly, determining an identification framework theta, wherein A represents any subset of the identification framework and is recorded as A belonging to theta and corresponds to different operation conditions of the synchronous generator;

the synchronous motor works under four operating conditions of normal operation and 3%, 6% and 10% of turn-to-turn short circuit of a rotor winding, and the operating conditions are respectively marked as A1, A2, A3 and A4;

the basic trust distribution function m is a set of slaves 2 ^Θ To [0,1]And satisfies the following conditions:

in the formula:

is an empty set; m (A) a basic trust distribution function called event A, representing the degree of trust of evidence for A; and respectively inputting excitation current, rotor vibration and stator vibration signals into the optimal two-dimensional CNN model to train to obtain the output probability of each state of the synchronous motor, namely the fault diagnosis accuracy of each state, and taking the output probability as basic trust distribution.

According to the specific embodiment provided by the invention, the invention discloses the following technical effects: the invention provides a multisource information-based synchronous motor rotor winding turn-to-turn short circuit fault diagnosis method, which comprises the steps of collecting excitation current, rotor vibration and stator vibration signals of a synchronous motor in normal and different short circuit fault states on line, converting a one-dimensional time sequence signal into a two-dimensional gray image by adopting a signal-image conversion method, inputting the processed image into a two-dimensional convolution neural network for training, and finally performing decision fusion on the training precision of three evidence bodies by utilizing D-S to realize final diagnosis; the signal-image conversion method used by the invention can retain more characteristics of the original signal, eliminate the influence of manually extracting the characteristics, reduce noise interference caused by external environment and improve the fault diagnosis accuracy; the identification degrees of different types of signals on turn-to-turn short circuit faults of the rotor winding of the synchronous motor are different, the evidence theory of multi-source electromechanical information is fused, the phenomena of misjudgment, missing judgment and the like caused by faults of a single sensor are avoided, and the fault diagnosis accuracy rate is further improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.

FIG. 1 is a schematic flow chart of a method for diagnosing turn-to-turn short circuit faults of a rotor winding of a synchronous motor based on multi-source information according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a signal-to-image conversion method according to an embodiment of the present invention;

FIG. 3 is a two-dimensional gray scale image of different signals under different states according to an embodiment of the present invention;

4a-4c are graphs of loss functions based on field current, rotor vibration, and stator vibration, respectively, according to embodiments of the present invention;

5a-5c are graphs of the accuracy of fault diagnosis based on the excitation current, rotor vibration, and stator vibration, respectively, according to embodiments of the present invention;

FIG. 6 is a comparison graph of the fault diagnosis accuracy before and after D-S fusion according to the embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The invention aims to provide a synchronous motor rotor winding turn-to-turn short circuit fault diagnosis method based on multi-source information, which furthest retains the characteristics of original signals, eliminates the influence of manually extracting the characteristics, avoids the phenomena of missing judgment, erroneous judgment and the like caused by the fault of a single sensor, and obviously improves the accuracy of the synchronous generator rotor winding turn-to-turn short circuit fault diagnosis.

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.

The synchronous motor is used as electromechanical coupling equipment, each part in the synchronous motor has a complex coupling relation, when the synchronous motor normally runs, the air gap magnetic field is uniformly and symmetrically distributed, and the vibration amplitude in the running process cannot exceed the standard allowance. When the rotor winding has a turn-to-turn short circuit fault, the effective turns of the excitation winding are reduced, so that the excitation current is increased, the distribution of an air gap magnetic field is uneven, and the stator and the rotor of the synchronous motor are subjected to unbalanced magnetic pull force, so that the unit generates abnormal vibration; on the other hand, the winding turn-to-turn short circuit also causes uneven heating of the generator rotor, which causes thermal bending and aggravation of rotor vibration. Therefore, the invention selects the exciting current, the rotor vibration and the stator vibration as the fault characteristic quantity of the turn-to-turn short circuit of the rotor winding to carry out fault diagnosis.

As shown in fig. 1, the method for diagnosing turn-to-turn short circuit fault of rotor winding of synchronous motor based on multi-source information provided by the invention comprises the following steps:

s1, acquiring excitation current, rotor vibration and stator vibration signals of the synchronous motor in real time by using a data acquisition device under four operating conditions of normal operation and 3%, 6% and 10% of rotor winding turn-to-turn short circuit, and performing the following steps: 2, dividing a training set and a test set; the data acquisition device comprises an exciting current signal detection device, a rotor vibration acceleration sensor and a stator winding axial vibration acceleration sensor, the stator winding axial vibration acceleration sensor has the characteristics of high sensitivity, strong anti-electromagnetic field interference capability, high resolution and the like, various sensors are mounted on the synchronous motor and connected with the analysis device, and the parameters of the synchronous motor are shown in table 1;

s2, converting the one-dimensional time sequence signals of the exciting current, the rotor vibration and the stator vibration collected in the step S1 into a two-dimensional gray scale image by using a signal-image conversion method;

s3, determining the network structure and convolution kernel size of the two-dimensional CNN classification model, training by using a training set, adjusting the CNN network structure and the hyper-parameters according to the loss function and the accuracy rate change condition in the iteration process, training again, repeating the above processes, determining the optimal two-dimensional CNN model, and checking the effect of the optimal two-dimensional CNN model through the training set;

s4, respectively inputting the two-dimensional gray scale images of the exciting current signal, the rotor vibration and the stator vibration into an optimal two-dimensional CNN model to obtain a fault diagnosis loss function and fault diagnosis accuracy based on a single signal of the exciting current, the rotor vibration and the stator vibration, wherein the loss function expression is shown in the specification;

in the formula, m is the size of the input small batch; j is a target class; p is a real label of the sample; q is a classification result output by the model;

s5, determining an identification frame theta, wherein A represents any subset of the identification frame, recording as A belonging to theta, corresponding to different fault operation conditions of the synchronous motor, and respectively inputting excitation current, rotor vibration and stator vibration signals into an optimal two-dimensional CNN model to obtain fault diagnosis accuracy as basic trust distribution;

and S6, taking the excitation current, the rotor vibration and the stator vibration signals as three evidence bodies, and performing decision fusion on the fault diagnosis accuracy of the three evidence bodies by using a multi-source information fusion algorithm.

TABLE 1 synchronous machine parameters

In step S2, the signal-image conversion method is to randomly extract the length N from the original one-dimensional time domain signal ² The signal of length N is selected each time to fill the pixels of the image line by line and then normalized to obtain an N × N two-dimensional gray image, and a schematic diagram of the signal-image conversion method is shown in fig. 2. Fig. 3 shows a grey scale image of different signals in different states.

Since the pixel value of the gray image is 0 to 255, the data of each pixel point needs to be normalized, and the expression is as follows:

wherein E (i), i ═ 1,2, …, N ² Representing the taking of each point in the intercepted one-dimensional signalA value; l (i, j); 1,2, …, N; j is 1,2, …, and N represents the pixel value of each point in the two-dimensional grayscale image.

In step S3, the two-dimensional CNN network structure corresponding to the optimal two-dimensional CNN model is shown in table 2, and specifically includes: an input layer, a convolutional layer CL, a pooling layer PL, a full-link layer and an output layer; the convolutional layer CL comprises a convolutional layer CL1, a convolutional layer CL2, a convolutional layer CL3 and a convolutional layer CL4, and the pooling layer PL comprises a pooling layer PL1, a pooling layer PL2, a pooling layer PL3 and a pooling layer PL 4;

the input layer inputs a gray level image with the size of 64 multiplied by 1;

the convolution kernel size of the convolution layer CL1 is set to 5 × 5, the step size Stride is set to 1, zero Padding is 'same', the feature map size is 64 × 64, the feature map depth is 16, and the relu function is selected as the activation function;

the pooling layer PL1 adopts a maximum pooling method, the width of a pooling area is 2 multiplied by 2, and the size of a characteristic diagram is 32 multiplied by 32;

the convolution kernel size of the convolution layer CL2 is set to 3 × 3, the step size Stride is set to 1, zero Padding is 'same', the feature map size is 32 × 32, the feature map depth is 16, and the relu function is selected as the activation function;

the pooling layer PL2 adopts a maximum pooling method, the width of a pooling area is 2 x 2, and the size of a characteristic diagram is 16 x 16;

the convolution kernel size of the convolution layer CL3 is set to 3 × 3, the step size Stride is set to 1, zero Padding is 'same', the feature map size is 16 × 16, the feature map depth is 64, and the relu function is selected as the activation function;

the pooling layer PL3 adopts a maximum pooling method, the width of a pooling area is 2 x 2, and the size of a characteristic diagram is 8 x 8;

the convolution kernel size of the convolution layer CL4 is set to 3 × 3, the step size Stride is set to 1, zero Padding is 'same', the feature map size is 8 × 8, the feature map depth is 128, and the relu function is selected as the activation function;

the pooling layer PL4 adopts a maximum pooling method, the width of a pooling area is 2 x 2, and the size of a characteristic diagram is 4 x 4;

the number of the neurons is set to 512 by the full connection layer, and a relu function is selected by an activation function;

the output layer adopts a Softmax classifier, the output class is 4, and the output layer corresponds to four operation working conditions of normal operation and rotor winding turn-to-turn short circuit of 3%, 6% and 10% respectively.

TABLE 2 two-dimensional CNN network architecture

The step S5 is to mainly calculate a basic probability distribution function, and specifically includes:

firstly, determining an identification framework theta, wherein A represents any subset of the identification framework and is recorded as A belonging to theta and corresponds to different operation conditions of the synchronous generator;

the synchronous motor works under four operating conditions of normal operation and 3%, 6% and 10% of turn-to-turn short circuit of a rotor winding, and the operating conditions are respectively marked as A1, A2, A3 and A4;

the basic trust distribution function m is a set of slaves 2 ^Θ To [0,1]And satisfies the following conditions:

in the formula:

is an empty set; m (A) a basic trust distribution function called event A, which represents the degree of trust of evidence for A; and respectively inputting excitation current, rotor vibration and stator vibration signals into the optimal two-dimensional CNN model to train to obtain the output probability of each state of the synchronous motor, namely the fault diagnosis accuracy of each state, and taking the output probability as basic trust distribution.

In step S6, the multi-source information fusion algorithm specifically includes: different data for the same problemSince the source of a problem is different in the mass function, it is necessary to recombine a plurality of mass functions to obtain a new probability function, and to perform the synthesis on the probability function according to a synthesis rule

Identifying a finite number of mass functions, m, on a framework ₁ ，m ₂ ，...，m _n The combination rule is as follows:

in the formula, S is a conflict factor, reflects the conflict degree between evidences, and ensures that S is less than 1, namely, turn-to-turn short circuit fault characteristics cannot be mutually exclusive. The expression is

After obtaining a plurality of basic probability distribution functions of event pairs, fusion can be performed only if the event pairs are verified to meet specific rules, wherein the specific rules are as follows:

is provided with

When the following conditions are satisfied:

is provided with

Then A is ₁ Is the judgment result. The above equation indicates that the difference between the classified object type and the basic probability distribution function of other types needs to be larger than the threshold ε ₁ (ii) a The fundamental probability distribution function of uncertainty needs to be less than a threshold epsilon ₂ (ii) a The judged target category needs to be larger than the uncertaintyAnd (4) rate.

Fusing two-dimensional CNN output results of exciting current, rotor vibration and stator vibration, and setting epsilon ₁ 、ε ₂ The value of (2) is 0.1.

Fig. 4a-4c show loss function graphs based on field current, rotor vibration, and stator vibration, where the loss function values of the field current, rotor vibration, and stator vibration signals all decrease with increasing number of iterations, and converge at 160, 260, and 170 iterations, respectively.

Fig. 5a-5c show the accuracy of fault diagnosis based on exciting current, rotor vibration and stator vibration, and the accuracy of all three signals increases with the number of iterations and tends to be stable. The fault diagnosis accuracy rate based on the exciting current tends to be stable when the iteration is carried out for 180 times, and the final value is 93.61%; the fault diagnosis accuracy based on the rotor vibration tends to be stable at 340 iterations with a final value of 92.06%, and the accuracy of the stator vibration tends to be stable at 175 iterations and finally reaches 85.98%. Therefore, the fault diagnosis accuracy of different sensors in the same model is different, namely, the fault information reflected by a single sensor is one-sided, and the method has limitation.

Fig. 6 is a comparison graph of accuracy of fault diagnosis before and after D-S fusion, the accuracy of single signals such as exciting current, rotor vibration, stator vibration and the like to the inter-turn short circuit fault diagnosis of the rotor winding of the synchronous motor is 93.61%, 92.06% and 85.98%, and after the three evidence bodies are fused by D-S, the diagnosis accuracy is 98.25%, which is remarkably improved compared with a single sensor, which indicates that the fault diagnosis method provided by the invention is effective.

According to the multi-source information-based synchronous motor rotor winding turn-to-turn short circuit fault diagnosis method, firstly, the used signal-image conversion method can retain more characteristics of original signals, eliminates the influence of manually extracted characteristics, reduces noise interference caused by external environment, and improves fault diagnosis accuracy; secondly, the identification degrees of different types of characteristic information on the turn-to-turn short circuit fault of the rotor winding of the synchronous motor are different, the evidence theory of multi-source electromechanical information is fused, the phenomena of misjudgment, missing judgment and the like caused by the fault of a single sensor are avoided, and the fault diagnosis accuracy rate is further improved.

The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.

Claims (6)

1. A synchronous motor rotor winding turn-to-turn short circuit fault diagnosis method based on multi-source information is characterized by comprising the following steps:

s1, acquiring excitation current, rotor vibration and stator vibration signals of the synchronous motor in real time under normal and various rotor winding turn-to-turn short circuit fault operation conditions by using a data acquisition device, and dividing a training set and a test set according to a set proportion;

s2, converting the one-dimensional time sequence signals of the exciting current, the rotor vibration and the stator vibration collected in the step S1 into a two-dimensional gray scale image by using a signal-image conversion method;

s3, determining the network structure and convolution kernel size of the two-dimensional CNN classification model, training by using a training set, adjusting the CNN network structure and the hyper-parameter according to the loss function and accuracy rate change condition in the iteration process, training again, repeating the above processes, and determining the optimal two-dimensional CNN model;

s4, respectively inputting the two-dimensional gray scale images of the exciting current signal, the rotor vibration and the stator vibration into an optimal two-dimensional CNN model to obtain a fault diagnosis loss function and fault diagnosis accuracy based on a single signal of the exciting current signal, the rotor vibration signal and the stator vibration signal;

s5, determining an identification frame theta, wherein A represents any subset of the identification frame, recording as A belonging to theta, corresponding to different operation conditions of the synchronous motor, and respectively inputting excitation current, rotor vibration and stator vibration signals into an optimal two-dimensional CNN model to obtain fault diagnosis accuracy as basic trust distribution;

and S6, taking the excitation current, the rotor vibration and the stator vibration signals as three evidence bodies, and performing decision fusion on the fault diagnosis accuracy of the three evidence bodies by using a multi-source information fusion algorithm.

2. The multi-source information-based synchronous motor rotor winding turn-to-turn short circuit fault diagnosis method according to claim 1, wherein in the step S1, the plurality of rotor winding turn-to-turn short circuit fault operation conditions include 3%, 6% and 10% of rotor winding turn-to-turn short circuit; the set proportion is 8: 2.

3. the multi-source information-based synchronous motor rotor winding turn-to-turn short circuit fault diagnosis method according to claim 1, wherein in the step S1, the data acquisition device comprises an excitation current signal detection device, a rotor vibration acceleration sensor and a stator winding axial vibration acceleration sensor.

4. The multi-source information-based diagnosis method for the turn-to-turn short circuit fault of the rotor winding of the synchronous motor according to the claim 1, wherein in the step S2, the signal-image conversion method is to randomly extract the length N from the original one-dimensional time domain signal ² The signals with the length of N are selected each time to fill the pixels of the image line by line and then normalization is carried out, thus obtaining the two-dimensional gray image of NxN.

5. The multi-source information-based synchronous motor rotor winding turn-to-turn short circuit fault diagnosis method according to claim 2, wherein in the step S3, the optimal two-dimensional CNN model specifically includes: an input layer, a convolutional layer CL, a pooling layer PL, a full-link layer and an output layer; the convolutional layer CL comprises a convolutional layer CL1, a convolutional layer CL2, a convolutional layer CL3 and a convolutional layer CL4, and the pooling layer PL comprises a pooling layer PL1, a pooling layer PL2, a pooling layer PL3 and a pooling layer PL 4;

the input layer inputs a gray level image with the size of 64 multiplied by 1;

the convolution kernel size of the convolution layer CL1 is set to 5 × 5, the step size Stride is set to 1, zero Padding is 'same', the feature map size is 64 × 64, the feature map depth is 16, and the relu function is selected as the activation function;

the pooling layer PL1 adopts a maximum pooling method, the width of a pooling area is 2 multiplied by 2, and the size of a characteristic diagram is 32 multiplied by 32;

the convolution kernel size of the convolution layer CL2 is set to 3 × 3, the step size Stride is set to 1, zero Padding is 'same', the feature map size is 32 × 32, the feature map depth is 16, and the relu function is selected as the activation function;

the pooling layer PL2 adopts a maximum pooling method, the width of a pooling area is 2 x 2, and the size of a characteristic diagram is 16 x 16;

the convolution kernel size of the convolution layer CL3 is set to 3 × 3, the step size Stride is set to 1, zero Padding is 'same', the feature map size is 16 × 16, the feature map depth is 64, and the relu function is selected as the activation function;

the pooling layer PL3 adopts a maximum pooling method, the width of a pooling area is 2 x 2, and the size of a characteristic diagram is 8 x 8;

the convolution kernel size of the convolution layer CL4 is set to 3 × 3, the step size Stride is set to 1, zero Padding is 'same', the feature map size is 8 × 8, the feature map depth is 128, and the relu function is selected as the activation function;

the pooling layer PL4 adopts a maximum pooling method, the width of a pooling area is 2 x 2, and the size of a characteristic diagram is 4 x 4;

the number of the neurons is set to 512 by the full connection layer, and a relu function is selected by an activation function;

the output layer adopts a Softmax classifier, the output class is 4, and the output layer corresponds to four operation working conditions of normal operation and rotor winding turn-to-turn short circuit of 3%, 6% and 10% respectively.

6. The method for diagnosing the turn-to-turn short circuit fault of the rotor winding of the synchronous motor based on the multi-source information as claimed in claim 2, wherein the step S5 is to determine an identification frame Θ, a represents any subset of the identification frame, and is recorded as a ∈ Θ, and corresponding to different operating conditions of the synchronous motor, the method for diagnosing the fault of the synchronous motor based on the multi-source information includes the steps of respectively inputting excitation current, rotor vibration and stator vibration signals into an optimal two-dimensional CNN model to obtain fault diagnosis accuracy as basic trust distribution, and specifically includes:

firstly, determining an identification framework theta, wherein A represents any subset of the identification framework and is recorded as A belonging to theta and corresponds to different operation conditions of the synchronous generator;

the synchronous motor works under four operating conditions of normal operation and 3%, 6% and 10% of turn-to-turn short circuit of a rotor winding, and the operating conditions are respectively marked as A1, A2, A3 and A4;

the basic trust distribution function m is a set of slaves 2 ^Θ To [0,1]And satisfies the following conditions:

in the formula:

is an empty set; m (A) a basic trust distribution function called event A, representing the degree of trust of evidence for A; and respectively inputting excitation current, rotor vibration and stator vibration signals into the optimal two-dimensional CNN model to train to obtain the output probability of each state of the synchronous motor, namely the fault diagnosis accuracy of each state, and taking the output probability as basic trust distribution.

Images