CN114330413A - Fault type identification and positioning method for traction motor bearing - Google Patents
Fault type identification and positioning method for traction motor bearing Download PDFInfo
- Publication number
- CN114330413A CN114330413A CN202111409414.6A CN202111409414A CN114330413A CN 114330413 A CN114330413 A CN 114330413A CN 202111409414 A CN202111409414 A CN 202111409414A CN 114330413 A CN114330413 A CN 114330413A
- Authority
- CN
- China
- Prior art keywords
- layer
- data
- deep learning
- training
- network model
- 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
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000013135 deep learning Methods 0.000 claims abstract description 105
- 238000012549 training Methods 0.000 claims abstract description 87
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 238000012545 processing Methods 0.000 claims description 38
- 238000011176 pooling Methods 0.000 claims description 26
- 238000012360 testing method Methods 0.000 claims description 20
- 230000005540 biological transmission Effects 0.000 claims description 17
- 230000001133 acceleration Effects 0.000 claims description 14
- 238000003062 neural network model Methods 0.000 claims description 14
- 238000010606 normalization Methods 0.000 claims description 14
- 238000013527 convolutional neural network Methods 0.000 claims description 13
- 238000010276 construction Methods 0.000 claims description 10
- 238000012795 verification Methods 0.000 claims description 10
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 238000005520 cutting process Methods 0.000 claims description 5
- 230000006870 function Effects 0.000 claims description 5
- 230000006403 short-term memory Effects 0.000 claims description 5
- 238000011478 gradient descent method Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 238000012544 monitoring process Methods 0.000 claims description 3
- 230000003044 adaptive effect Effects 0.000 claims description 2
- 230000002457 bidirectional effect Effects 0.000 claims description 2
- 102100033620 Calponin-1 Human genes 0.000 claims 1
- 102100033591 Calponin-2 Human genes 0.000 claims 1
- 102100033592 Calponin-3 Human genes 0.000 claims 1
- 101000945318 Homo sapiens Calponin-1 Proteins 0.000 claims 1
- 101000945403 Homo sapiens Calponin-2 Proteins 0.000 claims 1
- 101000945410 Homo sapiens Calponin-3 Proteins 0.000 claims 1
- 230000015654 memory Effects 0.000 claims 1
- 238000010200 validation analysis Methods 0.000 claims 1
- 230000036541 health Effects 0.000 abstract description 3
- 238000001514 detection method Methods 0.000 abstract description 2
- 239000010410 layer Substances 0.000 description 85
- 238000012423 maintenance Methods 0.000 description 7
- 238000003745 diagnosis Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 239000002346 layers by function Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005299 abrasion Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000000306 component Substances 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000003449 preventive effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Abstract
The invention discloses a fault type identification and positioning method for a traction motor bearing, and relates to the field of bearing fault detection of traction motors. The method of the invention consists of two stages: in the first stage, a deep learning network model is built and trained. The method mainly comprises the steps of collecting and preprocessing various fault bearing vibration signals, establishing and configuring a deep learning network, carrying out multiple iterative training, checking the accuracy of the deep learning network and deriving the accuracy. And in the second stage, the trained deep learning network model is applied to the reasoning stage. The method comprises the steps of acquiring vibration signals of a bearing to be identified and positioned, preprocessing the vibration signals, and judging the vibration signals through a deep learning network model to obtain the fault type and the positioning result of the traction motor bearing. The method has higher accuracy in identifying and positioning the fault type, and provides support for the fault prediction and health management of the traction motor. The method is easy to realize the expansion of the new fault type and the positioning of the traction motor bearing of the rail transit vehicle.
Description
Technical Field
The invention relates to the field of bearing fault detection of traction motors, in particular to a fault type identification and positioning method of a traction motor bearing.
Background
The traction motor is one of key core components of the rail transit vehicle, is responsible for power output and completes conversion from electric energy to mechanical energy, and the health state of the traction motor is related to safe and stable operation of the whole vehicle. The traction motor is influenced by the operating environment and the equipment in actual work, and mechanical faults such as abrasion stripping and electric corrosion of the inner ring and the outer ring of the bearing, the retainer and the rolling body cannot be avoided. The system can monitor the state of the traction motor bearing, diagnose the fault, find the fault as soon as possible and position the fault, can provide guidance for further adopting corresponding control and maintenance strategies, and is an important link for realizing intelligent operation and maintenance of the rail transit traction electric transmission system.
In the prior art, operation and maintenance personnel trained professionally identify and position bearing fault types, and the operation and maintenance personnel can perform visual and tool measurement by combining vibration, electrical signals, field environment, design and production record files, operation and maintenance record files and the like of a traction motor and also can disassemble the motor to comprehensively judge the fault condition of the traction motor, but the inspection is usually used for regular preventive maintenance, and generally cannot realize real-time and online fault diagnosis; the method has strong dependence on professional knowledge and experience of operation and maintenance personnel, is easily influenced by external noise and the like, has more conditions of erroneous judgment or missed judgment, can only judge a plurality of limited types of faults generally, and has poor fault positioning capability. The method is a technical scheme of manually extracting features and additionally arranging a simple identification model, wherein the vibration signals of the traction motor bearing collected in practice are often influenced by factors such as multi-source excitation, response mutual coupling, strong noise and the like, so that proper features are difficult to extract and the process is complex; on the other hand, a simple identification model cannot set out implicit characteristics in data, and the defects of low accuracy in identification and positioning of the traction motor bearing fault types exist; in addition, the fault characteristics are often related to the rotating speed, so that the fault can be identified and positioned only at a fixed rotating speed, and the flexibility is poor.
Artificial intelligence is a strategic technology leading a new round of technological revolution, industrial revolution and social revolution. Deep learning is a very important research branch in the field of artificial intelligence, the deep learning automatically analyzes data and extracts intrinsic useful information by simulating a hierarchical abstract structure of a brain by using a multilayer neural network architecture, and fault identification and positioning of key equipment can be realized after a proper deep learning model is trained, so that the method is an effective data-driven fault diagnosis method. The deep learning network model shows excellent performances of high accuracy, strong generalization and good expandability in the fault identification and classification of the rotary machine. The vibration acceleration signal contains abundant state information, and is very suitable for fault diagnosis technology of rotating machinery such as a traction motor. Therefore, the deep learning network model can be trained by using the vibration acceleration signals of the traction motor and is used for identifying and positioning the fault type of the traction motor bearing.
Under the era background of continuous development of the current big data technology and the artificial intelligence technology, along with continuous development of rail transit equipment towards networking and intelligentization directions, the method can be used for continuously accumulating mass data of state monitoring and fault diagnosis of key sub-components, and has great potential and practical value in identifying and positioning the fault type of the bearing of the traction motor by utilizing a deep learning network model.
Disclosure of Invention
The invention provides a novel method for identifying and positioning the fault type of a traction motor bearing, aiming at solving the problems that the accuracy of manual identification is not high and the fault type of the traction motor bearing can only be detected at a fixed rotating speed after modeling in the prior art.
The invention is realized by the following technical scheme: a method for identifying and positioning a fault type of a traction motor bearing is provided, wherein a vibration acceleration signal of the traction motor bearing is used as a data source, and a deep learning network model is built to automatically analyze fault characteristics. According to the method, a deep learning network model is built by using a method of combining a convolutional neural network and a recursive neural network, valuable information in a vibration signal is automatically extracted for analysis, the identification and the positioning of the fault type are high in accuracy, and support is provided for the fault prediction and the health management of the traction motor. The method comprises two stages, wherein the first stage is a bearing fault type identification and positioning model construction and training stage, and the second stage is a bearing fault type identification and positioning model reasoning stage, as shown in figure 1. In the first stage, a deep learning network model is built and trained. The method mainly comprises the steps of collecting and preprocessing various fault bearing vibration signals, establishing and configuring a deep learning network, carrying out multiple iterative training, checking the accuracy of the deep learning network and deriving the accuracy. And in the second stage, the trained deep learning network model is applied to the reasoning stage. The method comprises the steps of acquiring vibration signals of a bearing to be identified and positioned, preprocessing the vibration signals, and judging the vibration signals through a deep learning network model to obtain the fault type and the positioning result of the traction motor bearing. The specific process is as follows:
the first stage is as follows: the method comprises a construction and training stage of a bearing fault type identification and positioning model, which is a deep learning network model construction and training stage, and comprises the following steps:
step 1: presetting traction motor bearing faults, wherein the preset number of the traction motor bearing faults is A;
step 2: the method comprises the steps that motor vibration signals are collected and preprocessed, data used for training a deep learning network model are traction motor vibration acceleration data collected by a vibration sensor, the sensor can be arranged at a transmission end or a non-transmission end of a traction motor according to specific conditions, and data in a T time range of corresponding faults are collected in order to achieve fault identification of different types and positions of traction motor bearings; then, preprocessing the vibration data of each fault: cutting original data to form B parts of data with a certain length, wherein the time interval T of each part is equal to T/B; secondly, marking the data, and corresponding the data and the label to be identified; thirdly, dividing the data into three parts of training data, verification data and test data: selecting part of data A multiplied by B multiplied by a% as training data of the deep learning network model, the other part of data A multiplied by B multiplied by B% as verification data of the deep learning network model, and the rest part of data A multiplied by B multiplied by (1-a% -B%) as test data of the deep learning network model;
and step 3: creating and configuring a deep learning network model:
when the deep learning network is established, a layer combination mode is adopted, so that the complexity of a deep learning network model is reduced, and the network debugging is more convenient; in practice, the network model is configured by adding or modifying layers according to specific situations so as to achieve the required accuracy, and the deep learning network model comprises the following steps: (1) a signal input layer for inputting the vibration acceleration data into the network model; (2) one or more convolution processing layers for extracting high-dimensional characteristic information of the vibration data; wherein each convolution processing layer comprises: (2.1) convolutional layer: characteristic information for learning and storing data; (2.2) batch normalization layer: the convolutional neural network is used for normalizing the output result of the convolutional layer and improving the training speed of the convolutional neural network; (2.3) Relu layer: realizing the nonlinear processing of data through a threshold function; (2.4) maximum pooling layer: realizing down-sampling of data; (3) one or more recursive processing layers for learning high-dimensional feature information processed by the convolutional processing layers, wherein each recursive processing layer comprises: (3.1) recursive layers: characteristic information used for learning and storing data, including long short term memory network layer LSTM and bidirectional long short term memory network BilTM; (3.2) discarding layers: for discarding part of the characteristic information; (4) a full link layer for integrating the information extracted by the recursive processing layer; (5) a Softmax layer for normalizing the output result of the full connection layer; (6) a signal output layer which outputs the data processed by the previous layers as the identification and positioning result of the bearing fault type of the traction motor; the network structure of the deep learning network model is shown in fig. 2.
And 4, step 4: training a deep learning network model:
first, the training parameters are selected: the method comprises the steps of specifying a solver, maximum training iteration times, minimum batch training size, learning rate and selecting a network training processor; the solver is a solver of a training network and comprises solvers of a random gradient descent method (SGDM), an adaptive moment estimation (ADAM) and a gradient square root (RMSProp) 3; one iteration means that the training algorithm completely passes through the whole training data set; the minimum batch training refers to a subset of a training data set processed on the processor at the same time; the learning rate is a parameter for controlling the training speed of the deep learning network; generally, the lower the learning rate, the higher the accuracy of the training result, but the longer the network training time. The network training processor generally comprises a Central Processing Unit (CPU), a Graphics Processing Unit (GPU) and 3 Field Programmable Gate Array (FPGA) processors; after the training parameters of the network are specified, training a deep learning network model by using a training data set and verification data and monitoring the training progress of the deep learning network model;
and 5: checking the accuracy of the deep learning network model and adjusting:
when the deep learning network model is trained, the accuracy of the verification data is not improved after a plurality of iterations, which means that the network model cannot be converged on a solution and cannot be improved; at the moment, test data are imported into the network model to obtain the accuracy of the test data; if the accuracy of the test data does not reach the expected accuracy, returning to the step 3 or the step 4, adjusting the configuration or the training parameters of the deep learning network model, trying to adjust the configuration or the training parameters of the deep learning network model, for example, trying to modify the deep learning network by methods of increasing a convolution processing layer, reducing the learning rate and the like, and performing iterative training again;
step 6: deriving a deep learning neural network model;
when the accuracy of the deep learning network model on the test data reaches an expected value, stopping the training of the deep learning network model and exporting;
and a second stage: and (3) reasoning stage of the bearing fault type identification and positioning model:
the method for identifying and positioning the fault type of the traction motor bearing in the deep learning network model reasoning phase generally comprises the following 3 steps:
step 1: motor vibration signal acquisition and pretreatment: firstly, acquiring vibration data of a traction motor to be detected in real time; then, preprocessing the vibration data; cutting data, wherein the time interval t1 of each piece of data is equal to t in the deep learning network model building and training stage;
step 2: importing data into a deep learning neural network model: importing data into the deep learning neural network model obtained in the step 6 of the first stage;
and step 3: obtaining a fault type identification and positioning result: the deep learning neural network model realizes the identification and positioning of the bearing fault of the traction motor with expected accuracy; each data can obtain a fault identification and positioning result; and finally, deploying the deep learning neural network model on edge end hardware or a cloud computing platform.
Compared with the prior art, the invention has the following beneficial effects: the invention provides a fault type identification and positioning method of a traction motor bearing, (1) the invention only adopts an original time domain signal of vibration acceleration data, does not need to carry out FFT (fast Fourier transform) equal frequency domain transformation calculation, and reduces the operation amount of a data acquisition device; (2) the invention adopts a deep learning network model formed by combining a plurality of convolutional neural networks and a plurality of recurrent neural networks to identify and position the key fault types of the bearing, the accuracy is greatly improved, and technical support can be provided for the fault diagnosis solution of the whole life cycle of the traction motor; (3) the deep learning network model is formed by combining different functional layers; the addition, deletion and configuration of the functional layer are very convenient, and the management, optimization and upgrading of the model are facilitated; (4) the method has strong expansibility, and if bearing faults to be identified and positioned need to be newly added, the identification and the positioning of the new fault type of the traction motor bearing are realized by acquiring corresponding fault vibration acceleration data and training a deep learning network model; and with the continuous increase of available data, more vibration data are used for deep learning network model training, which is beneficial to the improvement of accuracy and generalization.
Drawings
FIG. 1 is a schematic diagram of construction, training and reasoning of a deep learning network model.
Fig. 2 is a diagram of a deep learning network model architecture.
Fig. 3 is a diagram showing an example of a waveform of vibration data for each type of fault.
Fig. 4 is a diagram of a deep learning neural network architecture.
FIG. 5 is a test data confusion matrix map of a deep learning network model.
Detailed Description
The present invention is further illustrated by the following specific examples.
The method comprises two stages, wherein the first stage is a construction and training stage of a bearing fault type identification and positioning model, and the second stage is an inference stage of the bearing fault type identification and positioning model, and the specific process is as follows:
the first stage is as follows: the method comprises a construction and training stage of a bearing fault type identification and positioning model, which is a deep learning network model construction and training stage, and comprises the following steps:
step 1: presetting traction motor bearing faults, wherein the preset number of the traction motor bearing faults is A;
the type one to be identified and positioned is as follows: failure of the inner ring of the transmission end;
the type II to be identified and positioned is as follows: failure of the outer ring of the transmission end;
the type to be identified and positioned is three: failure of the transmission end retainer;
and the type to be identified and positioned is four: failure of the drive end roller;
type five to be identified and positioned: failure of the inner ring of the non-transmission end;
the type to be identified and positioned is six: failure of the outer ring of the non-transmission end;
type seven to be identified and located: failure of the non-drive end retainer;
eight types of identification and positioning are to be carried out: non-drive end roller failure.
Step 2: motor vibration signal acquisition and pretreatment: the data used for training the deep learning network model are traction motor vibration acceleration data collected by a vibration sensor, the traction motor is installed on a traction joint debugging system test bed, the motor operates at a random rotating speed, and the rotating speed range is 1000-4000 rpm; the acceleration sensor is arranged at the transmission end of the traction motor, the direction is vertical to the ground, the sampling rate is 25.6 kHz, and each type of fault respectively collects vibration acceleration data of 360 seconds;
the time length of each type of fault data is 360 seconds, the fault data is cut into 1800 parts in equal length, the length of each part of data is 0.2 second, and corresponding type fault labels are added to each part of data. The eight types of fault data together form 14400 labeled data sets serving as data sets of the deep learning neural network. 11520 parts of data (1440 parts of each type of fault data) are selected as training data of the network, and the other 1440 parts of data (180 parts of each type of fault data) are selected as verification data of the network. The remaining 1440 data (180 for each type of failure data) was used as test data for the network. The resulting vibration data waveform for a random set of each type of fault is shown in fig. 3.
And step 3: creating and configuring a deep learning network model:
in this embodiment, the created deep learning network model is composed of eleven different layers. The first layer is a signal Input layer (Input) for inputting a vibration signal into the network. The second to seventh layers are 6 convolution processing layers for identifying the characteristics of different faults. The second layer is a convolution layer 1, which is composed of a convolution layer (CNN 1), a batch normalization layer (BN 1), a Relu layer (Relu 1), and a maximum pooling layer (MaxPool 1), and the convolution kernel size of this layer is set to 16, and the pooling size is set to 2. The third layer is a convolution processing layer 2, which is composed of a convolution layer (CNN 2), a batch normalization layer (BN 2), a Relu layer (ReLU 2) and a maximum pooling layer (Maxpool 2), and the convolution kernel size of the layer is set to be 24, and the pooling size is set to be 2. The fourth layer is a convolution processing layer 3, which is composed of a convolution layer (CNN 3), a batch normalization layer (BN 3), a Relu layer (Relu 3), and a maximum pooling layer (MaxPool 3), and the convolution kernel size of this layer is set to 32, and the pooling size is set to 2. The fifth layer is a convolution processing layer 4, which is composed of a convolution layer (CNN 4), a batch normalization layer (BN 4), a Relu layer (ReLU 4) and a maximum pooling layer (Maxpool 4), and the filter size of the layer is set to be 48, and the pooling size is set to be 2. The sixth layer is a convolution processing layer 5, which is composed of a convolution layer (CNN 5), a batch normalization layer (BN 5), a Relu layer (Relu 5), and a maximum pooling layer (MaxPool 5), and the filter size of this layer is set to 64, and the pooling size is set to 2. The sixth layer is a convolution processing layer 6, which is composed of a convolution layer (CNN 6), a batch normalization layer (BN 6), a Relu layer (Relu 6), and a maximum pooling layer (MaxPool 6), and the filter size of the layer is set to 96, and the pooling size is set to 2.
And the eighth layer is a recursive processing layer and is used for learning the high-dimensional characteristic information processed by the convolution processing layer. Consisting of a recursive layer and a discard layer. Wherein the recursive layer is a long short term memory network Layer (LSTM) with 24 implicit elements, and the discard rate of the discard layer (Dropout) is set to 20%.
And the ninth layer to the eleventh layer are composed of a full connection layer (FC), a SoftMax function layer (SoftMax) and a signal Output layer (Output) and are used for outputting the deep learning network model as eight fault type identification and positioning results. The deep learning network model structure is shown in fig. 4.
And 4, step 4: training a deep learning network model:
a solver of the network is designated as a random gradient descent method (SGDM); setting the maximum training iteration number to be 20; a minimum batch training size of 64 is specified; setting the initial learning rate to be 0.02, and reducing the iterative learning rate by 10% after each 10 times of training; the training processor hardware is designated as a Graphics Processor (GPU). And importing the training data into a deep learning network model and training. With the progress of network training, the accuracy of the deep learning network on verification data is continuously improved, and meanwhile, the loss function is continuously reduced, which shows that the fault type identification and positioning capabilities of the deep learning network are stronger and stronger.
And 5: checking the accuracy of the deep learning network model and adjusting:
and importing the test data into the trained convolutional neural network. In this embodiment, the setting accuracy rate needs to be more than 98%. If the accuracy rate does not meet the requirement, the network can be adjusted by modifying the network model in the step 3 or modifying the training parameters in the step 4, for example, the deep learning network is modified by methods of increasing a convolution processing layer, reducing the learning rate and the like, and the iterative training is continued; and if the accuracy reaches the requirement, stopping training.
Step 6: deriving a deep learning neural network model:
and when the accuracy of the deep learning network model on the test data is higher than 98%, stopping training and deriving the deep learning network model. FIG. 5 is a confusion matrix of the deep learning network model for the test data, where the data on the diagonal of the confusion matrix is the number of correct identifications of various faults, and the others are misjudgment cases; the test set accuracy of the deep learning network model is 99.24%.
(2) Inference phase of model
Step 1: motor vibration signal acquisition and pretreatment:
firstly, acquiring vibration data of a traction motor to be measured in real time. The vibration data is then preprocessed. Cutting the data, wherein the time length of each data is 0.2 s;
step 2: importing data into a deep learning neural network model:
and importing the data into the eleven-layer deep learning neural network model obtained after training.
And step 3: obtaining a fault type identification and positioning result:
the deep learning neural network model realizes the identification and the positioning of the faults of the motor bearing of the traction motor with expected accuracy. Each data will get a fault identification and location result. The deep learning network model is deployed in an edge-end embedded ARM processor, and the identification and the positioning of the fault type of the traction motor bearing are realized with expected accuracy.
The scope of the invention is not limited to the above embodiments, and various modifications and changes may be made by those skilled in the art, and any modifications, improvements and equivalents within the spirit and principle of the invention should be included in the scope of the invention.
Claims (6)
1. A fault type identification and positioning method for a traction motor bearing is characterized by comprising the following steps: the method comprises two stages, wherein the first stage is a construction and training stage of a bearing fault type identification and positioning model, the second stage is an inference stage of the bearing fault type identification and positioning model, and the specific process is as follows:
the first stage is as follows: the method comprises a construction and training stage of a bearing fault type identification and positioning model, which is a deep learning network model construction and training stage, and comprises the following steps:
step 1: presetting traction motor bearing faults, wherein the preset number of the traction motor bearing faults is A;
step 2: the method comprises the steps that motor vibration signals are collected and preprocessed, data used for training a deep learning network model are traction motor vibration acceleration data collected by a vibration sensor, the sensor is arranged at a transmission end or a non-transmission end of a traction motor, and data in a T time range of corresponding faults are collected in order to realize fault identification of different types and positions of traction motor bearings; then, preprocessing the vibration data of each fault: cutting original data to form B parts of data with a certain length, wherein the time interval T of each part is equal to T/B; secondly, marking the data, and corresponding the data and the label to be identified; thirdly, dividing the data into three parts of training data, verification data and test data: selecting part of data A multiplied by B multiplied by a% as training data of the deep learning network model, the other part of data A multiplied by B multiplied by B% as verification data of the deep learning network model, and the rest part of data A multiplied by B multiplied by (1-a% -B%) as test data of the deep learning network model;
and step 3: creating and configuring a deep learning network model:
when the deep learning network is established, the complexity of a deep learning network model is reduced by adopting a layer combination mode; in practice, the network model is configured by adding or modifying layers according to specific situations so as to achieve the required accuracy, and the deep learning network model comprises the following steps: (1) a signal input layer for inputting the vibration acceleration data into the network model; (2) one or more convolution processing layers for extracting high-dimensional characteristic information of the vibration data; wherein each convolution processing layer comprises: (2.1) convolutional layer: characteristic information for learning and storing data; (2.2) batch normalization layer: the convolutional neural network is used for normalizing the output result of the convolutional layer and improving the training speed of the convolutional neural network; (2.3) Relu layer: realizing the nonlinear processing of data through a threshold function; (2.4) maximum pooling layer: realizing down-sampling of data; (3) one or more recursive processing layers for learning high-dimensional feature information processed by the convolutional processing layers, wherein each recursive processing layer comprises: (3.1) recursive layers: characteristic information used for learning and storing data, including long short term memory network layer LSTM and bidirectional long short term memory network BilTM; (3.2) discarding layers: for discarding part of the characteristic information; (4) a full link layer for integrating the information extracted by the recursive processing layer; (5) a Softmax layer for normalizing the output result of the full connection layer; (6) a signal output layer which outputs the data processed by the previous layers as the identification and positioning result of the bearing fault type of the traction motor;
and 4, step 4: training a deep learning network model:
first, the training parameters are selected: the method comprises the steps of specifying a solver, maximum training iteration times, minimum batch training size, learning rate and selecting a network training processor; the solver is a solver of a training network and comprises solvers of a random gradient descent method (SGDM), an adaptive moment estimation (ADAM) and a gradient square root (RMSProp) 3; one iteration means that the training algorithm completely passes through the whole training data set; the minimum batch training refers to a subset of a training data set processed on the processor at the same time; the learning rate is a parameter for controlling the training speed of the deep learning network; the network training processor generally comprises a Central Processing Unit (CPU), a Graphics Processing Unit (GPU) and 3 Field Programmable Gate Array (FPGA) processors; after the training parameters of the network are specified, training a deep learning network model by using a training data set and verification data and monitoring the training progress of the deep learning network model;
and 5: checking the accuracy of the deep learning network model and adjusting:
when the deep learning network model is trained, the accuracy of the verification data is not improved after a plurality of iterations, which means that the network model cannot be converged on a solution and cannot be improved; at the moment, test data are imported into the network model to obtain the accuracy of the test data; if the accuracy of the test data does not reach the expected accuracy, returning to the step 3 or the step 4, adjusting the configuration or the training parameters of the deep learning network model, and performing iterative training again;
step 6: deriving a deep learning neural network model;
when the accuracy of the deep learning network model on the test data reaches an expected value, stopping the training of the deep learning network model and exporting;
and a second stage: and (3) reasoning stage of the bearing fault type identification and positioning model:
the method for identifying and positioning the fault type of the traction motor bearing in the deep learning network model reasoning phase generally comprises the following 3 steps:
step 1: motor vibration signal acquisition and pretreatment: firstly, acquiring vibration data of a traction motor to be detected in real time; then, preprocessing the vibration data; cutting data, wherein the time interval t1 of each piece of data is equal to t in the deep learning network model building and training stage;
step 2: importing data into a deep learning neural network model: importing data into the deep learning neural network model obtained in the step 6 of the first stage;
and step 3: obtaining a fault type identification and positioning result: the deep learning neural network model realizes the identification and positioning of the bearing fault of the traction motor with expected accuracy; each data can obtain a fault identification and positioning result; and finally, deploying the deep learning neural network model on edge end hardware or a cloud computing platform.
2. The method for identifying and positioning the fault type of the traction motor bearing according to claim 1, wherein the method comprises the following steps: the preset traction motor bearing faults and the number in the step 1 of the first stage are as follows:
the type one to be identified and positioned is as follows: failure of the inner ring of the transmission end;
the type II to be identified and positioned is as follows: failure of the outer ring of the transmission end;
the type to be identified and positioned is three: failure of the transmission end retainer;
and the type to be identified and positioned is four: failure of the drive end roller;
type five to be identified and positioned: failure of the inner ring of the non-transmission end;
the type to be identified and positioned is six: failure of the outer ring of the non-transmission end;
type seven to be identified and located: failure of the non-drive end retainer;
eight types of identification and positioning are to be carried out: non-drive end roller failure.
3. The method for identifying and positioning the fault type of the traction motor bearing according to claim 1, wherein the method comprises the following steps: in the step 2 of the first stage, in the motor vibration signal acquisition and pretreatment, the motor runs at a random rotating speed, and the rotating speed range is 1000-4000 rpm; the acceleration sensor is arranged at the transmission end of the traction motor, the direction is vertical to the ground, the sampling rate is 25.6 kHz, and each type of fault respectively collects vibration acceleration data of 360 seconds;
the time length of each type of fault data is 360 seconds, the fault data is cut into 1800 parts in equal length, the length of each part of data is 0.2 second, and corresponding type fault labels are added to each part of data; 14400 data sets with labels are formed by the eight types of fault data and are used as data sets of the deep learning neural network; 11520 parts of data are selected as training data of the network, wherein 1440 parts of each type of fault data are selected; 1440 additional data was included as network validation data, 180 for each type of failure; the remaining 1440 data was used as test data for the network, 180 for each type of fault data.
4. The method for identifying and positioning the fault type of the traction motor bearing according to claim 1, wherein the method comprises the following steps: in step 3 of the first stage, the created deep learning network model consists of eleven different layers; the first layer is a signal Input layer and is used for inputting a vibration signal into a network; the second to seventh layers are 6 convolution processing layers and are used for identifying the characteristics of different faults; the second layer is a convolution processing layer 1 and consists of a convolution layer CNN1, a batch normalization layer BN1, a Relu layer ReLU1 and a maximum pooling layer MaxPool1, the convolution kernel size of the layer is set to be 16, and the pooling size is set to be 2; the third layer is a convolution processing layer 2 and consists of a convolution layer CNN2, a batch normalization layer BN2, a Relu layer ReLU2 and a maximum pooling layer Maxpool2, the convolution kernel size of the layer is set to be 24, and the pooling size is set to be 2; the fourth layer is a convolution processing layer 3, which consists of a convolution layer CNN3, a batch normalization layer BN3, a Relu layer ReLU3 and a maximum pooling layer MaxPool3, wherein the convolution kernel size of the layer is set to be 32, and the pooling size is set to be 2; the fifth layer is a convolution processing layer 4 and consists of a convolution layer CNN4, a batch normalization layer BN4, a Relu layer ReLU4 and a maximum pooling layer Maxpool4, the filter size of the layer is set to be 48, and the pooling size is set to be 2; the sixth layer is a convolution processing layer 5 and consists of a convolution layer CNN5, a batch normalization layer BN5, a Relu layer ReLU5 and a maximum pooling layer Maxpool5, the filter size of the layer is set to be 64, and the pooling size is set to be 2; the sixth layer is a convolution processing layer 6 and consists of a convolution layer CNN6, a batch normalization layer BN6, a Relu layer ReLU6 and a maximum pooling layer Maxpool6, the filter size of the layer is set to be 96, and the pooling size is set to be 2;
the eighth layer is a recursive processing layer and is used for learning the high-dimensional characteristic information processed by the convolutional processing layer; the device consists of a recursive layer and a discard layer; wherein, the recursion layer is a long-short term memory network layer LSTM containing 24 hidden units, and the discarding rate of the discarding layer Dropout is set as 20%;
and the ninth layer to the eleventh layer are composed of a full connection layer FC, a SoftMax function layer SoftMax and a signal Output layer Output and are used for outputting the deep learning network model as eight fault type identification and positioning results.
5. The method for identifying and positioning the fault type of the traction motor bearing according to claim 1, wherein the method comprises the following steps: in the step 4 of the first stage, a deep learning network model is trained, and a solver of a network is designated as a random gradient descent method SGDM; setting the maximum training iteration number to be 20; a minimum batch training size of 64 is specified; setting the initial learning rate to be 0.02, and reducing the iterative learning rate by 10% after each 10 times of training; designating training processor hardware as a Graphics Processing Unit (GPU); and importing the training data into a deep learning network model and training.
6. The method for identifying and positioning the fault type of the traction motor bearing according to claim 1, wherein the method comprises the following steps: in step 5 of the first stage, an expected accuracy of 98% was obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111409414.6A CN114330413A (en) | 2021-11-25 | 2021-11-25 | Fault type identification and positioning method for traction motor bearing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111409414.6A CN114330413A (en) | 2021-11-25 | 2021-11-25 | Fault type identification and positioning method for traction motor bearing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114330413A true CN114330413A (en) | 2022-04-12 |
Family
ID=81045894
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111409414.6A Pending CN114330413A (en) | 2021-11-25 | 2021-11-25 | Fault type identification and positioning method for traction motor bearing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114330413A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116167749A (en) * | 2023-04-23 | 2023-05-26 | 合肥荣叙科技有限公司 | Permanent magnet synchronous motor fault diagnosis method based on deep learning |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200210824A1 (en) * | 2018-12-28 | 2020-07-02 | Utopus Insights, Inc. | Scalable system and method for forecasting wind turbine failure with varying lead time windows |
| CN111538947A (en) * | 2020-05-18 | 2020-08-14 | 中车永济电机有限公司 | Method for constructing wind power generator bearing fault classification model |
| CN111597996A (en) * | 2020-05-18 | 2020-08-28 | 中车永济电机有限公司 | Method for constructing wind turbine generator bearing fault identification model based on deep learning |
| US20200380337A1 (en) * | 2019-05-30 | 2020-12-03 | Hongfujin Precision Electronics (Zhengzhou) Co., Ltd. | User terminal hardware detection system and method |
| CN112508242A (en) * | 2020-11-25 | 2021-03-16 | 中车永济电机有限公司 | Method for constructing bearing fault location and classification model of wind power generator |
| CN112989712A (en) * | 2021-04-27 | 2021-06-18 | 浙大城市学院 | Aeroengine fault diagnosis method based on 5G edge calculation and deep learning |
| CN113094978A (en) * | 2021-03-20 | 2021-07-09 | 北京化工大学 | Method for constructing model of residual life of motor bearing based on deep learning |
-
2021
- 2021-11-25 CN CN202111409414.6A patent/CN114330413A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200210824A1 (en) * | 2018-12-28 | 2020-07-02 | Utopus Insights, Inc. | Scalable system and method for forecasting wind turbine failure with varying lead time windows |
| US20200380337A1 (en) * | 2019-05-30 | 2020-12-03 | Hongfujin Precision Electronics (Zhengzhou) Co., Ltd. | User terminal hardware detection system and method |
| CN111538947A (en) * | 2020-05-18 | 2020-08-14 | 中车永济电机有限公司 | Method for constructing wind power generator bearing fault classification model |
| CN111597996A (en) * | 2020-05-18 | 2020-08-28 | 中车永济电机有限公司 | Method for constructing wind turbine generator bearing fault identification model based on deep learning |
| CN112508242A (en) * | 2020-11-25 | 2021-03-16 | 中车永济电机有限公司 | Method for constructing bearing fault location and classification model of wind power generator |
| CN113094978A (en) * | 2021-03-20 | 2021-07-09 | 北京化工大学 | Method for constructing model of residual life of motor bearing based on deep learning |
| CN112989712A (en) * | 2021-04-27 | 2021-06-18 | 浙大城市学院 | Aeroengine fault diagnosis method based on 5G edge calculation and deep learning |
Non-Patent Citations (2)
| Title |
|---|
| 许爱华;杜洋;袁涛;: "基于深度学习的电机轴承故障诊断研究", 组合机床与自动化加工技术, no. 03, 20 March 2020 (2020-03-20) * |
| 颉潭成;陈孟会;徐彦伟;李先锋;: "卷积神经网络在轴承故障识别的应用", 制造业自动化, no. 01, 25 January 2020 (2020-01-25) * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116167749A (en) * | 2023-04-23 | 2023-05-26 | 合肥荣叙科技有限公司 | Permanent magnet synchronous motor fault diagnosis method based on deep learning |
| CN116167749B (en) * | 2023-04-23 | 2023-07-18 | 合肥荣叙科技有限公司 | Permanent magnet synchronous motor fault diagnosis method based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111898634B (en) | Intelligent fault diagnosis method based on depth-to-reactance-domain self-adaption | |
| Lei et al. | Fault diagnosis of wind turbine based on Long Short-term memory networks | |
| Xiang et al. | Fault diagnosis of rolling bearing under fluctuating speed and variable load based on TCO spectrum and stacking auto-encoder | |
| Zhang et al. | Fault diagnosis and prognosis using wavelet packet decomposition, Fourier transform and artificial neural network | |
| CN110261109B (en) | Rolling bearing fault diagnosis method based on bidirectional memory cyclic neural network | |
| CN111947928B (en) | Multi-source information fusion bearing fault prediction system and method | |
| CN113469060A (en) | Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method | |
| CN108256556A (en) | Wind-driven generator group wheel box method for diagnosing faults based on depth belief network | |
| CN111597996B (en) | Method for constructing wind turbine generator bearing fault identification model based on deep learning | |
| Dou et al. | A rule-based intelligent method for fault diagnosis of rotating machinery | |
| CN107631867A (en) | A kind of rotating machinery fault intelligent method for classifying based on deep learning | |
| CN111538947B (en) | Method for constructing wind power generator bearing fault classification model | |
| CN111141517B (en) | Fan fault diagnosis method and system | |
| Zhang et al. | Early fault detection method of rolling bearing based on MCNN and GRU network with an attention mechanism | |
| CN113076834B (en) | Rotating machine fault information processing method, processing system, processing terminal, and medium | |
| CN113094978A (en) | Method for constructing model of residual life of motor bearing based on deep learning | |
| CN112305388B (en) | On-line monitoring and diagnosing method for insulation partial discharge faults of generator stator winding | |
| CN112508242A (en) | Method for constructing bearing fault location and classification model of wind power generator | |
| CN114021932A (en) | Energy efficiency evaluation and diagnosis method, system and medium for wind turbine generator | |
| Saeki et al. | Visual explanation of neural network based rotation machinery anomaly detection system | |
| Li et al. | Intelligent fault diagnosis of aeroengine sensors using improved pattern gradient spectrum entropy | |
| CN111795819A (en) | Gear box fault diagnosis method integrating vibration and current signal collaborative learning | |
| CN114330413A (en) | Fault type identification and positioning method for traction motor bearing | |
| CN115859077A (en) | Multi-feature fusion motor small sample fault diagnosis method under variable working conditions | |
| CN114320773B (en) | Wind turbine generator system fault early warning method based on power curve analysis and neural network |
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 |