CN114993678A - Three-phase asynchronous motor bearing life prediction method - Google Patents
Three-phase asynchronous motor bearing life prediction method Download PDFInfo
- Publication number
- CN114993678A CN114993678A CN202210558858.4A CN202210558858A CN114993678A CN 114993678 A CN114993678 A CN 114993678A CN 202210558858 A CN202210558858 A CN 202210558858A CN 114993678 A CN114993678 A CN 114993678A
- Authority
- CN
- China
- Prior art keywords
- bearing
- life
- state
- asynchronous motor
- phase asynchronous
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/34—Testing dynamo-electric machines
- G01R31/343—Testing dynamo-electric machines in operation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
- G06F17/141—Discrete Fourier transforms
- G06F17/142—Fast Fourier transforms, e.g. using a Cooley-Tukey type algorithm
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Discrete Mathematics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Acoustics & Sound (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a three-phase asynchronous motor bearing life prediction method, which comprises the steps of firstly extracting root mean square characteristics from three-phase asynchronous motor full life cycle vibration data, and introducing a new bottom-up time sequence segmentation algorithm to segment a characteristic sequence into 3 states of a normal period, a degradation period and a degradation period; carrying out state information labeling on an amplitude sequence of the vibration signal after fast Fourier transform, inputting the amplitude sequence into a full convolution neural network of a newly added convolution layer, and extracting deep layer characteristics to obtain a pre-training model; the gradient of the pre-training model is used as a feature to participate in the target domain network training process together with the traditional pre-training model feature, so that a state recognition model is obtained; the method has the advantages that the bearing life prediction model is established by combining the state probability estimation method and the state recognition model, and experiments verify that the method does not need to establish health indexes and can realize the prediction of the residual life of the three-phase asynchronous motor bearing under different working conditions under the unsupervised condition.
Description
Technical Field
The invention relates to the technical field of electromechanical equipment fault monitoring, in particular to a method for predicting the service life of a bearing of a three-phase asynchronous motor
Background
The rolling bearing is one of easily damaged parts, and its remaining life (RUL) is closely relevant with equipment running state, carries out the shut down maintenance that the life prediction can avoid leading to because of the bearing became invalid, the casualties scheduling problem through carrying out the life prediction to rolling bearing, has the significance.
The machine learning method is also applied to fault diagnosis of the rotary machine, and the machine learning can extract features for classification without rich expert knowledge for judgment, so that the difficulty of diagnosis is reduced. Machine learning uses vibration signals as samples for input in fault diagnosis, and then performs classification by extracting features. The traditional fault classification algorithm and the machine learning algorithm have certain achievements in the field of fault identification, but the traditional fault detection algorithm depends on expert experience, meanwhile, the machine learning algorithm cannot well learn complex nonlinear relations in vibration signals, fault features cannot be automatically extracted, and different types of fault identification needs to design different types of feature extractors.
The deep learning method is greatly developed in the field of fault identification, deep features can be extracted from original signals, massive complex data are processed, representative features can be extracted in a self-adaptive mode based on the deep learning method, manual intervention is not needed, and compared with a traditional algorithm, the fault identification has higher accuracy. In the deep learning field, the convolutional neural network reduces the parameter quantity due to the convolution and pooling operation, and improves the speed and the accuracy of identification through local receptive fields (local perceptual fields) and shared weights (shared weights).
Although various rotary machine fault detection models are available and good experimental results are obtained, many challenges still exist in the field of rotary machine fault diagnosis. In order to obtain higher accuracy, the traditional deep learning fault detection model usually adopts a multilayer neural network superposition mode, so that the model complexity is too high. The model with the stacked multilayer networks has extremely high requirements on training equipment, and the fault detection time of the successfully trained model is too long, so that the method is not suitable for the problem of fault real-time diagnosis in an industrial scene. Secondly, a large number of high-quality samples are needed for the deeply learned fault diagnosis model, so that the model can be ensured not to be over-fitted. The existing model is tested under the condition of ideal sample quantity, and the actual industrial condition is that a fault sample is less than a normal sample, so that the data imbalance phenomenon generally exists in the industrial field.
Disclosure of Invention
Therefore, in order to solve the above disadvantages, the present invention provides a method for predicting a bearing life of a three-phase asynchronous motor, which can monitor the aging degree of a bearing set on line and predict the service life without offline detection by a maintenance person.
The invention is realized in such a way that a health monitoring method of a bearing group comprises the following steps,
s1, obtaining vibration information, and arranging a plurality of sensors in different directions at the front end and the rear end of each bearing respectively; standardizing the data to eliminate data difference;
s2, solving time domain RMS characteristics of the original vibration signal source domain data of the bearing;
s3, dividing RMS characteristics of the bearing source domain data from bottom to top into a normal period, a degradation period and a decay period;
s4, respectively carrying out fast Fourier transform on the original vibration signals of the bearing in the source domain and the bearing in the target domain to obtain a frequency domain amplitude sequence;
s5, correspondingly marking the classified bearing degradation state categories on the amplitude sequence subjected to fast Fourier transform;
s6, inputting the amplitude sequence (with a label) obtained after the fast Fourier transform into the FCN network of the newly added convolutional layer for pre-training;
s7, setting a model hyper-parameter;
s8, transmitting the parameters of the pre-training model except the classification layer to a target domain network, and respectively passing through a plurality of connected convolution layers, a pooling layer and an activation layer;
s9, training network parameters;
s10, determining whether the loop parameter converges, if it does not converge, returning to step S9, if it does, continuing to step S11;
s11, importing a test set of the original vibration signal into a training model, finely adjusting model parameters, and predicting the residual life by using a state probability estimation method;
the invention relates to a method for segmenting RMS (root mean square) characteristics of bearing source domain data from top to bottom, which comprises the following steps of:
(1)for i(i<T)
(2) creating an initial approximation set Seg _ TS
(3)for i(i<Seg_TS)
(4) Calculating the merging error merge _ cost (i) of Seg _ TS (i) and Seg _ TS (i +1)
(5) while fitting cost minimum < segment threshold
(6) Let p be the minimum of the fitting cost
(7) Merging Seg _ TS (p) and Seg _ TS (p +1) to redefine Seg _ TS (p)
(8) Delete Seg _ TS (p +1), update
(9) Calculating the fitting cost of Seg _ TS (p) and Seg _ TS (p +1)
(10) Calculating the fitting cost of Seg _ TS (p-1) and Seg _ TS (p), defined as Seg _ TS (p-1)
The state probability estimation method is a method for predicting the residual service life of the current state according to the probability that the time sequence belongs to different degradation states and the residual service life corresponding to each degradation state determined by historical data (training samples) according to the probability that the time sequence belongs to different degradation states obtained by a state classifier. The calculation process of the residual life at a certain time comprises the following steps:
Du i indicating the dwell time of the device at the i (i-1, 2 … C) th state, derived from historical data (training samples)K represents the current state dwell time coefficient for adjusting the remaining life prediction accuracy, C represents the state class RUL i Representing the remaining life of the device in the i-th degraded state, derived from the training sample P (X) t I) stands for the sequence X t Probability of being in the i-th degraded state.
The invention has the following advantages:
according to the method, the degradation degree of the motor bearing can be monitored, maintenance personnel are not needed, offline detection can be performed, and meanwhile fault pre-diagnosis is performed on the bearing group based on the health assessment result, so that test risks or accidents are avoided.
Drawings
FIG. 1 is a schematic flow chart of the present invention
FIG. 2 is a schematic representation of the FCN flow scheme of the present invention
Detailed Description
Step 1, obtaining vibration information, and arranging a plurality of sensors in different directions at the front end and the rear end of each bearing respectively; standardizing the data to eliminate data difference;
Step 3, solving time domain RMS characteristics of the original vibration signal source domain data of the bearing;
step 4, dividing RMS characteristics of the bearing source domain data from bottom to top into a normal period, a degradation period and a decay period;
step 5, respectively performing fast Fourier transform on the original vibration signals of the bearing in the source domain and the bearing in the target domain to obtain a frequency domain amplitude sequence;
step 6, inputting the amplitude sequence (with a label) obtained after the fast Fourier transform into the FCN of the newly added convolutional layer for pre-training;
the FCN network of the newly added convolutional layer sequentially comprises a convolutional layer 1, a BN, a relu, a convolutional layer 2, a BN, a relu, a convolutional layer 3, a BN, a relu, a convolutional layer 4 (newly added), a BN, a relu, a global pooling layer and a softmax layer which are connected.
Step 7, setting model hyper-parameters;
step 8, transmitting the parameters of the pre-training model except the classification layer to a target domain network, and respectively passing through a plurality of connected convolution layers, a pooling layer and an activation layer;
step 9, training network parameters;
step 10, whether the loop parameter converges or not, if no, returning to step S9, and if yes, continuing to step S11;
step 11, importing a test set of an original vibration signal into a training model, finely adjusting model parameters, and predicting the residual life by using a state probability estimation method;
wherein the expression of the probability is:
the formula for calculating the residual life of the bearing is as follows:
Du i k represents a current state residence time coefficient for adjusting the remaining life prediction accuracy, and C represents a state class i Representing the remaining life of the device in the i-th degraded state, derived from the training sample P (X) t I) stands for the sequence X t Probability of being in the i-th degraded state.
The prediction error formula for reflecting the performance of the model residual life prediction is as follows:
wherein, actril is a true value, and RUL is a predicted value.
The method comprises the following steps of segmenting the RMS characteristics of bearing source domain data from top to bottom, wherein the method comprises the following steps:
(11)for i(i<T)
(12) creating an initial approximation set Seg _ TS
(13)for i(i<Seg_TS)
(14) Calculating the merging error merge _ cost (i) of Seg _ TS (i) and Seg _ TS (i +1)
(15) while fitting cost minimum < segment threshold
(16) Let p be the minimum of the fitting cost
(17) Merging Seg _ TS (p) and Seg _ TS (p +1) to redefine Seg _ TS (p)
(18) Delete Seg _ TS (p +1), update
(19) Calculating the fitting cost of Seg _ TS (p) and Seg _ TS (p +1)
(20) Calculating the fitting cost of Seg _ TS (p-1) and Seg _ TS (p), defined as Seg _ TS (p-1)
Using classification results P i And the residence time Du corresponding to each degradation state, so as to obtain the RUL corresponding to the three degradation states. Du i And k is the ratio of the number of the test sample data points to the number of the training sample data points, and the RUL corresponding to each degradation state of the bearing is obtained and is used as the calculation basis of the residual service life of the test bearing.
Calculating the remaining life: and weighting the obtained result according to the proportion of the data points of the single training set in the data points of the whole training set to obtain the final residual life of the bearing. The remaining life of the test bearing was calculated in the same manner.
The method can monitor the service life of the motor bearing, does not need maintenance personnel, can perform off-line detection, and timely inform workers to reasonably arrange replacement and shutdown of the unit, thereby avoiding unnecessary economic loss. And meanwhile, based on the health assessment result, the service life of the motor bearing is pre-diagnosed, so that test risks or accidents are avoided.
Claims (9)
1. A three-phase asynchronous motor bearing life prediction method is characterized in that: comprises the following steps of (a) preparing a solution,
s1, obtaining vibration information of a three-phase asynchronous motor bearing under a certain working condition, obtaining a bearing vibration signal with a full life under a certain working condition, regarding the bearing vibration signal as a source domain (with a label), taking a bearing vibration signal with a non-full life as a target domain (without a label), solving time domain RMS (root mean square) characteristics of original vibration signal source domain data of the bearing and carrying out normalization processing, and then respectively carrying out fast Fourier transform on the original vibration signal of the bearing in the source domain and the target domain to obtain a frequency domain amplitude sequence.
S2, dividing the RMS characteristics of the bearing source domain data from bottom to top into a normal period, a degradation period and a decay period respectively, correspondingly marking the classified bearing degradation state categories on the amplitude sequence after the fast Fourier transform
S3, utilizing the FCN of the newly added convolutional layer to have the capability of better mining high-dimensional data characteristics and the characteristic of stronger robustness in space and time, and carrying out characteristic extraction on the frequency domain amplitude sequences of the bearing source domain data and the target domain data under the variable working condition to obtain the deep layer characteristics of the bearing.
S4, inputting the amplitude sequence (with a label) obtained after fast Fourier transform into the FCN of the newly added convolutional layer for pre-training to obtain a pre-training model containing gradient characteristics, transferring the parameters of the pre-training model except the classification layer to a target domain network, initializing the parameters of the softmax layer by using target domain data (without the label) and re-training to complete the parameter transfer process, and establishing a multi-state classification model of the bearing through multiple iteration optimization searching to realize the state identification of the bearing under an unsupervised condition to obtain a multi-classification result (probability).
S5, use stateEstimating, using the multi-state identification, the probability that the bearing belongs to each degradation state and the remaining useful life corresponding to each degradation state as determined from historical life cycle data, calculating the remaining useful life of the bearing using the multi-state identification J To reflect how good the model predicts the remaining life.
2. The method for predicting the bearing life of the three-phase asynchronous motor according to claim 1, wherein the method comprises the following steps: the data is normalized in step S1 using the following formula,
wherein the content of the first and second substances,and M i,j,k Vibration data of the jth position and the kth direction of the ith bearing before and after normalization respectively i,j,k And σ i,j,k Respectively, the mean value and the standard deviation of the kth direction of the jth position of the ith bearing.
3. The method for predicting the bearing life of the three-phase asynchronous motor according to claim 1, wherein the method comprises the following steps: in the step S2, the RMS characteristics of the bearing source domain data are segmented, first, time series data points with a length of T are connected pairwise, the time series data points are divided into T/2 initial segments which are not overlapped, and fitting costs (fitting errors) of adjacent segments to be combined are calculated.
4. The method for predicting the bearing life of the three-phase asynchronous motor according to claim 1, wherein the method comprises the following steps: the FCN network of the newly added convolutional layer in step S3 sequentially includes a convolutional layer 1, BN, relu, a convolutional layer 2, BN, relu, a convolutional layer 3, BN, relu, a convolutional layer 4 (newly added), BN, relu, a global pooling layer, and a softmax layer, which are connected.
5. The method for predicting the bearing life of the three-phase asynchronous motor according to claim 2, wherein the method comprises the following steps: at least one of the first convolution layer, the second convolution layer, the third convolution layer and a sub-convolution layer in the convolution feature fusion layer comprises a cascaded convolution operation module, an activation function module and a BN layer.
6. The method for predicting the bearing life of the three-phase asynchronous motor according to claim 1, characterized by comprising the following steps: the expression of the multi-classification result (probability) in step S4 is:
wherein: y is j Denotes the jth layer output, softmax denotes the activation function, P i Indicates the probability of class i, y i Representing a certain class of network output and n representing the total number of classes of network output.
7. The method for predicting the bearing life of the three-phase asynchronous motor according to claim 1, characterized by comprising the following steps: the formula for calculating the residual life of the bearing in step S5 is as follows:
Du i representing settings derived from historical data (training samples)K represents a current state dwell time coefficient for adjusting the remaining life prediction accuracy, C represents a state class, RUL, and k represents a state class i Represents the remaining life of the device in the i-th degraded state derived from the training sample P (X) t I) stands for the sequence X t Probability of being in the i-th degraded state.
8. The method for predicting the bearing life of the three-phase asynchronous motor according to claim 1, characterized by comprising the following steps: the prediction error formula used in step S5 to reflect the performance of the model with respect to remaining life prediction is:
wherein, actril is a true value, and RUL is a predicted value.
9. The method for predicting the bearing life of the three-phase asynchronous motor according to claim 1, wherein the method comprises the following steps: the state probability estimation method in step S5 is a method for obtaining the remaining life of the current state by obtaining the probability of the time series belonging to different degradation states obtained by the state classifier and the remaining service life corresponding to each degradation state determined by the historical data (training samples).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210558858.4A CN114993678A (en) | 2022-05-21 | 2022-05-21 | Three-phase asynchronous motor bearing life prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210558858.4A CN114993678A (en) | 2022-05-21 | 2022-05-21 | Three-phase asynchronous motor bearing life prediction method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114993678A true CN114993678A (en) | 2022-09-02 |
Family
ID=83027049
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210558858.4A Pending CN114993678A (en) | 2022-05-21 | 2022-05-21 | Three-phase asynchronous motor bearing life prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114993678A (en) |
-
2022
- 2022-05-21 CN CN202210558858.4A patent/CN114993678A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Yang et al. | Remaining useful life prediction based on a double-convolutional neural network architecture | |
CN110598736B (en) | Power equipment infrared image fault positioning, identifying and predicting method | |
CN109934130A (en) | The in-orbit real-time fault diagnosis method of satellite failure and system based on deep learning | |
CN109146246B (en) | Fault detection method based on automatic encoder and Bayesian network | |
CN113343633A (en) | Thermal runaway fault classification and risk prediction method and system for power lithium battery | |
CN111505424A (en) | Large experimental device power equipment fault diagnosis method based on deep convolutional neural network | |
CN114048568B (en) | Rotary machine fault diagnosis method based on multisource migration fusion shrinkage framework | |
CN111460728A (en) | Method and device for predicting residual life of industrial equipment, storage medium and equipment | |
CN111459144A (en) | Airplane flight control system fault prediction method based on deep cycle neural network | |
CN114676742A (en) | Power grid abnormal electricity utilization detection method based on attention mechanism and residual error network | |
CN112507479B (en) | Oil drilling machine health state assessment method based on manifold learning and softmax | |
CN115329986A (en) | Wind turbine generator anomaly detection and positioning method based on interpretable graph neural network | |
CN115859077A (en) | Multi-feature fusion motor small sample fault diagnosis method under variable working conditions | |
CN117375237B (en) | Substation operation and maintenance method and system based on digital twin technology | |
CN114118225A (en) | Method, system, electronic device and storage medium for predicting remaining life of generator | |
CN114091504A (en) | Rotary machine small sample fault diagnosis method based on generation countermeasure network | |
CN115358259A (en) | Self-learning-based unsupervised cross-working-condition bearing fault diagnosis method | |
CN116842379A (en) | Mechanical bearing residual service life prediction method based on DRSN-CS and BiGRU+MLP models | |
CN114879628A (en) | Multi-mode industrial process fault diagnosis method based on local maximum mean difference resistance | |
CN116894215B (en) | Gear box fault diagnosis method based on semi-supervised dynamic graph attention | |
Li et al. | Knowledge enhanced ensemble method for remaining useful life prediction under variable working conditions | |
CN117009916A (en) | Actuator fault diagnosis method based on multi-sensor information fusion and transfer learning | |
CN113469013B (en) | Motor fault prediction method and system based on transfer learning and time sequence | |
CN114993678A (en) | Three-phase asynchronous motor bearing life prediction method | |
CN116204781A (en) | Rotary machine fault migration diagnosis method and system |
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 |