CN112525505A - Fault detection method, system and storage medium - Google Patents
Fault detection method, system and storage medium Download PDFInfo
- Publication number
- CN112525505A CN112525505A CN202011232566.9A CN202011232566A CN112525505A CN 112525505 A CN112525505 A CN 112525505A CN 202011232566 A CN202011232566 A CN 202011232566A CN 112525505 A CN112525505 A CN 112525505A
- Authority
- CN
- China
- Prior art keywords
- neural network
- characteristic parameter
- fan component
- current
- normal
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 56
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 52
- 238000013528 artificial neural network Methods 0.000 claims abstract description 48
- 230000002159 abnormal effect Effects 0.000 claims abstract description 47
- 238000012545 processing Methods 0.000 claims abstract description 9
- 239000013598 vector Substances 0.000 claims description 47
- 210000002569 neuron Anatomy 0.000 claims description 27
- 238000000034 method Methods 0.000 claims description 24
- 238000005096 rolling process Methods 0.000 claims description 19
- 238000010606 normalization Methods 0.000 claims 1
- 230000001133 acceleration Effects 0.000 description 17
- 238000011176 pooling Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 7
- 238000012549 training Methods 0.000 description 7
- 230000014509 gene expression Effects 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 230000002860 competitive effect Effects 0.000 description 4
- 238000003745 diagnosis Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000013139 quantization Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 238000012284 sample analysis method Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
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
-
- 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
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Acoustics & Sound (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The application provides a fault detection method, a system and a storage medium. The fault detection method comprises the steps of obtaining a current characteristic parameter value of the fan component in a current detection period; processing the current characteristic parameter value through a trained self-organizing neural network to obtain the running state of the fan component, wherein the running state is an abnormal running state representing abnormal running of the fan component or a normal running state representing normal running of the fan component; if the operation state is the abnormal operation state, classifying the current characteristic parameter value through a trained convolutional neural network to obtain a classification result, wherein the classification result is a normal operation type representing normal operation of the fan component or one of a plurality of fault types of the fan component. The accuracy of fault detection can be improved.
Description
Technical Field
The present application relates to the field of wind power, and in particular, to a fault detection method, system, and storage medium.
Background
Wind generators include many components, such as bearings. The bearing is an important part widely applied in various mechanical equipment, is also a part with a higher failure occurrence rate, and directly influences the operation condition of the whole equipment. According to different properties of measured signals, fault diagnosis methods of bearings can be divided into vibration methods, oil sample analysis methods, noise methods, acoustic emission methods and the like. Among them, the diagnosis method based on the vibration signal is a popular and effective detection method at home and abroad. However, some technologies cannot accurately locate the bearing fault when the bearing fault is located by using a vibration signal diagnosis method.
Disclosure of Invention
The application provides a fault detection method, a fault detection system and a storage medium, which can improve the accuracy of fault location.
The application provides a fault detection method for detecting the fault type of a fan component, the fault detection method comprises the following steps:
acquiring a current characteristic parameter value of the fan component in a current detection period;
processing the current characteristic parameter value through a trained self-organizing neural network to obtain the running state of the fan component, wherein the running state is an abnormal running state representing abnormal running of the fan component or a normal running state representing normal running of the fan component;
if the operation state is the abnormal operation state, classifying the current characteristic parameter value through a trained convolutional neural network to obtain a classification result, wherein the classification result is a normal operation type representing normal operation of the fan component or one of a plurality of fault types of the fan component.
The present application provides a fault detection system comprising one or more processors for implementing a fault detection method as described in any one of the above.
The present application provides a storage medium having stored thereon a program which, when executed by a processor, implements a method as described in any one of the above.
In some embodiments, the fault location method of the application processes the current characteristic parameter value of the fan component in the current detection period through the trained self-organizing neural network, so as to obtain the operation state of the fan component, and when the operation state of the fan is an abnormal operation state, classifies the current characteristic parameter value through the trained convolutional neural network, so as to obtain a classification result, and the fault type of the fan component can be determined through the classification result, so that the accuracy of the fault location of the fan component can be improved.
Drawings
FIG. 1 is a flow chart of a fault detection method provided by one embodiment of the present application;
fig. 2 is a block diagram of a fault detection system according to an embodiment of the present application.
Detailed Description
Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of the present specification. Rather, they are merely examples of apparatus and methods consistent with certain aspects of one or more embodiments of the specification, as detailed in the claims which follow.
It should be noted that: in other embodiments, the steps of the corresponding methods are not necessarily performed in the order shown and described herein. In some other embodiments, the method may include more or fewer steps than those described herein. Moreover, a single step described in this specification may be broken down into multiple steps for description in other embodiments; multiple steps described in this specification may be combined into a single step in other embodiments.
The Self-organizing (competitive) neural network is a Self-organizing feature Map (SOM), which is a Self-organizing (competitive) neural network. In terms of network structure, it is generally a two-layer network composed of an input layer and a competition layer; the two layers of neurons realize bidirectional connection, and the network has no hidden layer. There is also a lateral connection between neurons in the competition layer. The basic idea behind the formation of competitive neural networks is that each neuron in the competitive layer of the network competes for an opportunity to respond to an input pattern, and finally only one neuron becomes the winner of the competition. This winning neuron then represents a classification of the input vector.
Convolutional Neural Networks (CNN) are a class of feed forward Neural Networks (fed forward Neural Networks) that contain convolution computations and have a deep structure, and are one of the representative algorithms for deep learning (deep learning). The convolutional neural network is composed of an input layer, an output layer and a plurality of hidden layers, wherein the hidden layers can be divided into a convolutional layer, a pooling layer, a RELU layer and a full-link layer. Traditionally, the input to a convolutional neural network is typically a picture, such as a classical MNIST dataset. In recent years, the convolutional neural network is also applied to 1-dimensional tensor data, and is applied to the fields of character exploration, time series, abnormality diagnosis and the like.
Fig. 1 is a flowchart of a fault detection method according to an embodiment of the present application. The fault detection method is used for detecting the fault type of the fan component. In the embodiments illustrated herein, the fan component comprises a rolling bearing of the fan. Wherein, the fan is a wind power generator (or called as a wind turbine). In other embodiments, the fan component may also include other components on the fan, which is not limited in this application. Referring to fig. 1, the fault detection method includes steps S11 through S13.
And step S11, acquiring the current characteristic parameter value of the fan component in the current detection period. In some embodiments, the fan component includes a plurality of characteristic parameters. When the operating states of the fan components are different, the values of at least some of the plurality of characteristic parameters are different. The operation state of the fan component comprises a normal operation state which indicates that the fan component operates normally and an abnormal operation state which indicates that the fan component operates abnormally.
In some embodiments, the time-domain vibration acceleration of the wind turbine component may be collected at different time points during the operation of the wind turbine component to obtain a plurality of time-domain vibration acceleration data. The plurality of time domain vibration acceleration data may be provided as a data set x. The characteristic parameters of the wind turbine component may include one or more of a standard deviation, a peak, a skewness, a kurtosis, a root mean square, a peak indicator, a margin indicator, a waveform indicator, a pulse indicator, and a kurtosis indicator of the plurality of time-domain vibratory acceleration data. Expressions (1) to (10) respectively give calculation formulas for calculating partial characteristic parameter values based on time-domain vibration acceleration data:
standard deviation of
Peak value xp＝max|x(i)| (2)
Skewness degree
Kurtosis
Root mean square
Peak index
Margin index
Waveform index
Pulse index
Kurtosis index
Wherein N is the Nth time point for collecting the time domain vibration acceleration, x is the time domain vibration acceleration data set, and xiIs the ith time domain vibration acceleration data, x in the time domain vibration acceleration data set(i)The ith data, x, of the time domain vibration acceleration data in the time domain vibration acceleration data set after sorting from big to small(m)The time domain vibration acceleration data in the time domain vibration acceleration data set is an average value.
In some embodiments, the current characteristic parameter value of the wind turbine component during the current detection period may be determined based on a plurality of time domain vibration acceleration data acquired at a plurality of time points within the current detection period.
In some embodiments, raw data representing the vibration acceleration of a fan component may be collected by a piezoelectric sensor disposed on the fan. The raw data may be a voltage value representing a vibration acceleration of the fan component. The piezoelectric sensor may further convert the voltage value to time domain vibration acceleration data.
And step S12, processing the current characteristic parameter value through the trained self-organizing neural network to obtain the running state of the fan component, wherein the running state is an abnormal running state representing abnormal running of the fan component or a normal running state representing normal running of the fan component. In some embodiments, after the ad hoc neural network processes the current characteristic parameter value, two signals (e.g., 0 and 1) may be output, where one signal (e.g., 1) indicates that the wind turbine component is operating abnormally, and the other signal (e.g., 0) indicates that the wind turbine component is operating normally.
In some embodiments, the trained self-organizing neural network is trained based on normal characteristic parameter samples of the wind turbine component in a normal operating state. The normal characteristic parameter sample is a set of characteristic parameter values of a plurality of characteristic parameters in a plurality of detection periods under the normal operation state of the fan component.
In some embodiments, before the current characteristic parameter value of the wind turbine component is processed by the trained self-organizing neural network, a normal characteristic parameter sample of the wind turbine component in a normal operation state may be obtained, and the normal characteristic parameter sample of the wind turbine component is input to the initial self-organizing neural network to train the initial self-organizing neural network, so as to obtain the trained self-organizing neural network. Specifically, the initial self-organizing neural network may include initialized initial neuron weight vectors with the same number as the number of the characteristic parameters, for example, assuming that the fan component includes N characteristic parameters, the initial self-organizing neural network may include initialized N initial neuron weight vectors. When the initial self-organizing neural network is trained, based on the input normal characteristic parameter sample, the self-organizing neural network can adjust the weight of each initial neuron weight vector, so that the trained self-organizing neural network comprises a plurality of normal neuron weight vectors corresponding to each characteristic parameter.
In some embodiments, the current characteristic parameter value of each characteristic parameter may be normalized to determine a current input vector corresponding to each current characteristic parameter value; and inputting each current input vector into the trained self-organizing neural network, so that the trained self-organizing neural network compares each current input vector with each normal neuron weight vector respectively to determine the minimum Euclidean distance between each current input vector and each normal neuron weight vector respectively, and thus, the current characteristic parameter value of the fan component is processed through the trained self-organizing neural network. Specifically, for a specific current input vector, the current input vector is compared with each normal neuron weight vector respectively to obtain a plurality of euclidean distances, and among the euclidean distances, the euclidean distance with the minimum distance is the minimum euclidean distance between the current input vector and the normal neuron weight vector. Meanwhile, the normal neuron weight vector with the minimum Euclidean distance from the current input vector is the normal neuron weight vector which is most matched with the current input vector. For example, assuming that the self-organizing neural network includes N (N is a positive integer greater than 1) normal neuron weight vectors, after the current input vector of the feature parameter 1 is input into the self-organizing neural network model, the current input vector is respectively compared with the normal neuron weight vectors 1, 2, … …, and N to obtain corresponding euclidean distances 1, 2, … …, and N, wherein, of the N euclidean distances, the euclidean distance with the minimum distance is the minimum euclidean distance between the current input vector of the feature parameter 1 and the normal neuron weight vectors. According to the rule, the minimum Euclidean distance between each other current input vector and the normal neuron weight vector can be determined in sequence.
In some embodiments, after determining the minimum euclidean distance between each current input vector and the normal neuron weight vector, the sum of the minimum euclidean distances corresponding to at least some of the current input vectors may be determined to be the sum of the current minimum euclidean distances. In the embodiment shown in fig. 1, the sum of the minimum euclidean distances is the sum of the minimum euclidean distances corresponding to all current input vectors.
In some embodiments, the operational status of the fan component may be determined based on the sum of the current minimum euclidean distances. The larger the sum of the current minimum Euclidean distances is, the farther the current input vector corresponding to the characteristic parameter value deviates from the normal neuron weight vector is, namely the farther the current operation state of the fan component deviates from the normal operation state is; and the smaller the sum of the current minimum Euclidean distances is, the closer the current input vector corresponding to the characteristic parameter value is to the weight vector of the normal neuron, namely, the closer the current operation state of the fan component is to the normal operation state. The sum of the current minimum euclidean distances may be defined as the minimum quantization error mqe (minimum quantization error). The calculation formula of the minimum quantization error MQE can be expressed as expression (11):
MQE＝||D-mBMU|| (11)
wherein D represents the current input vector corresponding to the characteristic parameter value; the mBMU represents the normal neuron weight vector that best matches the current input vector. The greater the MQE value, the greater the degree to which the current operating state of the fan component deviates from the normal state, and the greater the degree of damage. By tracking MQE values, the current operating state of the fan assembly may be monitored.
In some embodiments, the tendency of the value MQE to change may be monitored by an EWMA (empirical Weighted Moving-Average) control chart. Specifically, the EWMA statistic may be determined based on the sum of the current minimum euclidean distances and the sum of the minimum euclidean distances of the last detection period. The calculation formula for the EWMA statistic may be expressed as expression (12):
Zi＝λXi+(1-λ)Zi-1 (12)
wherein Z isiEWMA statistic, Z, representing the current detection periodi-1EWMA statistic, X, representing the last detection periodiRepresenting the sum of the current minimum euclidean distances (i.e., the MQE value for the current detection period), and λ represents the EWMA weight value.
In some embodiments, if EWMA statistic ZiAnd determining that the fan component operates abnormally when the fan component is lower than the lower control limit value or higher than the upper control limit value. Wherein the calculation formula of the upper control limit value may be expressed as expression (13):
wherein, UCL represents an upper control limit value, L represents a control limit coefficient, and lambda represents an EWMA weighted value.
The calculation formula of the lower control limit value may be expressed as expression (14):
wherein LCL represents a lower control limit value, L represents a control limit coefficient, and λ represents an EWMA weight value.
In some embodiments, the control limit factor L and the EWMA weight value λ may be obtained by looking up table 1.
Table 1
In this embodiment, λ ═ 0.1 and L ═ 2.7 can be selected.
In some embodiments, if EWMA statistic ZiBelow the lower control limit or above the upper control limit, the self-organizing neural network outputs a value (e.g., 0) representing the abnormal operation of the fan component; otherwise, the self-organizing neural network outputs a value (e.g., 1) indicating that the wind turbine component is operating properly. In some embodiments of the application, the change trend of MQE values is monitored through an EWMA control chart, the operation state of the fan component is judged according to whether the EWMA statistic is lower than a lower control limit value or higher than an upper control limit value, a signal indicating the abnormal operation of the fan component can be output in advance, for example, the signal indicating the abnormal operation of the fan component is output when the fan component has the trend of the abnormal operation state but does not actually operate abnormally, and a worker can prevent and treat related abnormal operation problems in advance. And for example, a signal indicating the abnormal operation of the fan component can be output at the early stage of the abnormal operation of the fan component, so that a worker can process the abnormal operation condition of the fan component in advance, and mechanical accidents caused by the abnormal operation of the fan component (such as a rolling bearing) can be prevented.
In other embodiments, the sum of the current euclidean distances (MQE value) may be compared to a distance threshold, and if the sum of the current euclidean distances (MQE value) is greater than or equal to the distance threshold, it may be determined that the fan component is operating abnormally. Thus, the arithmetic processing logic can be simplified.
In some embodiments of the present application, the trained self-organizing neural network is obtained by training based on normal characteristic parameter samples of the fan component in a normal operation state, and since the fan component is in the normal operation state for most of the time, the number of the normal characteristic parameter samples of the fan component in the normal operation state that can be obtained is large, and the training data volume is large enough, the trained self-organizing neural network can be accurate, so that the current characteristic parameter value is processed through the trained self-organizing neural network, and the obtained result representing the operation state of the fan component can be accurate.
And step S13, if the operation state is an abnormal operation state, classifying the current characteristic parameter values through the trained convolutional neural network to obtain a classification result, wherein the classification result is one of a normal operation type representing normal operation of the fan component or a plurality of fault types of the fan component.
In some embodiments, the trained convolutional neural network is trained based on normal characteristic parameter samples of the fan component in a normal operating state and abnormal characteristic parameter samples of the fan component in a plurality of fault operating states. The abnormal characteristic parameter samples can include a characteristic parameter sample 1 of the fan component running in the fault type 1, a characteristic parameter sample 2 and … … of the fan component running in the fault type 2, and a characteristic parameter sample N of the fan component running in the fault type N.
In some embodiments, before classifying the current characteristic parameter values through the trained convolutional neural network, normal characteristic parameter samples of the fan component in the normal operating state and abnormal characteristic parameter samples of the fan component in multiple fault operating states may be obtained. The collecting principle of the normal characteristic parameter sample and the abnormal characteristic parameter sample is similar, and reference may be specifically made to the related description about the collection of the normal characteristic parameter sample in step S12, which is not described herein again.
In some embodiments, after the characteristic parameter samples are obtained, sample labeling may be performed on the normal characteristic parameter samples and the abnormal characteristic parameter samples, for example, the normal characteristic parameter samples are labeled as 1; marking a characteristic parameter sample of the fan component operating in the fault type 1 as 2; … …, respectively; and marking the characteristic parameter sample of the fan component operating in the fault type N as N. In the embodiment shown in fig. 1, the fan component comprises a rolling bearing, and the failure types of the rolling bearing comprise inner ring failure, outer ring failure, cage failure and rolling body failure. The characteristic parameter samples can be classified into 5 types, namely a characteristic parameter sample (for example, labeled as 0) of a rolling bearing in a normal operating state, a characteristic parameter sample (for example, labeled as 1) of an inner ring failure of the rolling bearing, a characteristic parameter sample (for example, labeled as 2) of an outer ring failure of the rolling bearing, a characteristic parameter sample (for example, labeled as 3) of a cage failure of the rolling bearing, and a characteristic parameter sample (for example, labeled as 4) of a rolling body failure of the rolling bearing.
In some embodiments, the normal characteristic parameter samples and the abnormal characteristic parameter samples of the wind turbine component may be input to the initial convolutional neural network to train the initial convolutional neural network, so as to obtain a trained convolutional neural network. The normal characteristic parameter samples and the abnormal characteristic parameter samples input into the initial convolutional neural network are normal characteristic parameter samples and abnormal characteristic parameter samples for completing sample labeling. In a specific neural network training process, for example, in the embodiment of the present application, 80% of the feature parameter samples may be used as training data, and 20% of the feature parameter samples may be used as test data. In other embodiments, the training data and the test data may be adjusted according to the amount of the characteristic parameter samples, for example, to increase the ratio of the test data. This is not limited by the present application.
In some embodiments, the trained convolutional neural network may classify the current feature parameter value and output a corresponding classification result. Through the classification result, whether the fan component operates normally or a certain type of fault occurs can be determined. For example, in the embodiment shown in fig. 1, since the characteristic parameter samples are classified into 5 types, each type of characteristic parameter sample represents one operation state (i.e., a normal operation state and 4 failure operation states) of the rolling bearing. Therefore, the classification result output by the trained convolutional neural network can be a signal representing one of the operation states of the rolling bearing. For example, assuming that the self-organizing neural network in step S12 outputs an abnormal operation state indicating an abnormal operation of the rolling bearing and the convolutional neural network in step S13 outputs a result indicating a failure of the inner ring of the rolling bearing, it is possible to accurately locate a failure of the rolling bearing by the self-organizing neural network and the convolutional neural network.
In some embodiments, the trained convolutional neural network may have an inaccuracy problem due to the limited number of abnormal characteristic parameter samples of the fan component in the abnormal operation state, which can be acquired within a specific time period (for example, half a year). Therefore, the trained convolutional neural network can be corrected based on the trained self-organizing neural network in step S12 to improve the accuracy of the convolutional neural network, thereby further improving the accuracy of locating the fault of the wind turbine component.
Specifically, in some embodiments, if the operation state obtained by the trained self-organizing neural network is an abnormal operation state and the classification result obtained by the convolutional neural network is a normal operation type, in this case, it can be known from the above related explanation that since the result output by the self-organizing neural network is relatively accurate, it can be determined that the result output by the convolutional neural network is inaccurate, and the convolutional neural network needs to be trained again. Specifically, the convolutional neural network may be retrained using at least the current characteristic parameter value, the normal characteristic parameter sample, and the abnormal characteristic parameter sample. The current characteristic parameter values may be used as samples for retraining the convolutional neural network. As the operation data collected is gradually increased along with the increase of the operation time of the fan component, and the sample data volume is larger and larger, the convolutional neural network obtained by training by using the current characteristic parameter value, the normal characteristic parameter sample and the abnormal characteristic parameter sample is more accurate. In some embodiments, the convolutional neural network may be retrained using the current characteristic parameter value, the characteristic parameter value before the current detection period, the normal characteristic parameter samples, and the abnormal characteristic parameter samples. The characteristic parameter value before the current detection period can also be used as a sample for retraining the convolutional neural network. The characteristic parameter values before the current detection period are also used for training the convolutional neural network, the sample data size is larger, and the obtained convolutional neural network is more accurate.
Similarly, if the operation state obtained through the trained self-organizing neural network is a normal operation state and the classification result obtained through the trained convolutional neural network is one of a plurality of fault types, the convolutional neural network is retrained by at least using the current characteristic parameter value, the normal characteristic parameter sample and the abnormal characteristic parameter sample.
In some embodiments, the convolutional neural network of the present application comprises a one-dimensional convolutional neural network (i.e., 1D CNN). The one-dimensional convolutional neural network comprises an input layer, a convolutional layer, a pooling layer, a full-link layer and an output layer. Convolutional layers consist of filters (filters) and kernels (kernels) that have small receptive fields, derived to the full depth. The filter dimension of a one-dimensional convolutional neural network may be an integer of 8, such as 16, 32, 64, 128, 256; the kernel kernels dimension of a one-dimensional convolutional neural network may be 3, 5, 7. The relevant parameters of the convolutional layer can be modified according to the actual situation, which is not limited in this application. The convolutional layer may embed a RELU layer as an activation function, where the activation function may be expressed as expression (15):
activation function max (0, x) (15)
In some embodiments, the pooling layer may include a maximum pooling layer with a step number of 2. And the full connection layer performs full connection on the data output by the convolution layer and the pooling layer to determine a classification result. And the output layer outputs the classification result. The output layer may be set as a softmax layer with the neuron number being the classification number.
In some embodiments, a deep stack structure may be used, and repeated input of layers to the pooling layer may be used to form a stack structure, which may be effective to improve accuracy.
Table 2 is a schematic structural diagram of a one-dimensional convolutional neural network provided in an embodiment of the present application.
TABLE 2
| Layer(s) | Output shape | Parameter(s) |
| 1-dimensional convolutional layer | (None，4092，32) | 192 |
| 1-dimensional maximum pooling layer | (None，2046，32) | 0 |
| 1-dimensional global max pooling layer | (None,32) | 0 |
| Output layer | (None，4) | 132 |
In some embodiments of the application, the trained self-organizing neural network is used for processing the current characteristic parameter value of the fan component in the current detection period, the running state of the fan component can be obtained, when the running state of the fan is in an abnormal running state, the trained convolutional neural network is used for classifying the current characteristic parameter value, a classification result can be obtained, the fault type of the fan component can be determined according to the classification result, and then the accuracy of fan component fault positioning can be improved.
Fig. 2 is a block diagram of a fault detection system 10 according to an embodiment of the present application.
The fault detection system 10 includes one or more processors 100 for implementing the fault detection method described above. In some embodiments, the fault detection system 10 may include a storage medium 109, which may store a program that may be called by the processor 100, which may include a non-volatile storage medium. In some embodiments, the fault detection system 10 may include a memory 108 and an interface 107. In some embodiments, the fault detection system 10 may also include other hardware depending on the application.
The storage medium 109 of the embodiment of the present application stores thereon a program for implementing the pledge monitoring method as described above when executed by the processor 100.
This application may take the form of a computer program product embodied on one or more storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having program code embodied therein. Storage media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media include, but are not limited to: phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technologies, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic tape storage or other magnetic storage devices, or any other non-transmission medium, may be used to store information that may be accessed by a computing device.
The above description is only a preferred embodiment of the present disclosure, and should not be taken as limiting the present disclosure, and any modifications, equivalents, improvements, etc. made within the spirit and principle of the present disclosure should be included in the scope of the present disclosure.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
Claims (11)
1. A fault detection method for detecting the fault type of a fan component is characterized by comprising the following steps:
acquiring a current characteristic parameter value of the fan component in a current detection period;
processing the current characteristic parameter value through a trained self-organizing neural network to obtain the running state of the fan component, wherein the running state is an abnormal running state representing abnormal running of the fan component or a normal running state representing normal running of the fan component;
if the operation state is the abnormal operation state, classifying the current characteristic parameter value through a trained convolutional neural network to obtain a classification result, wherein the classification result is a normal operation type representing normal operation of the fan component or one of a plurality of fault types of the fan component.
2. The method of claim 1, wherein the wind turbine component includes a plurality of feature parameters, the trained self-organizing neural network includes a plurality of normal neuron weight vectors corresponding to the feature parameters, and the obtaining the operating state of the wind turbine component by processing the current feature parameter value of the wind turbine component through the trained self-organizing neural network includes:
respectively carrying out normalization processing on the current characteristic parameter values of the characteristic parameters to determine current input vectors corresponding to the current characteristic parameter values;
inputting each current input vector into the trained self-organizing neural network, so that the trained self-organizing neural network compares each current input vector with each normal neuron weight vector respectively to determine the minimum Euclidean distance between each current input vector and the normal neuron weight vector respectively;
determining the sum of the minimum Euclidean distances corresponding to at least part of the current input vectors, wherein the sum is the sum of the current minimum Euclidean distances;
and determining the operation state of the fan component based on the sum of the current minimum Euclidean distances.
3. The fault detection method of claim 2, wherein said determining an operational status of said wind turbine component based on a sum of said current minimum euclidean distances comprises:
determining an EWMA statistic based on the sum of the current minimum Euclidean distances and the sum of the minimum Euclidean distances of the last detection period;
and if the EWMA statistic is lower than the lower control limit value or higher than the upper control limit value, determining that the fan component is abnormal in operation.
4. The fault detection method of claim 2, wherein said determining an operational status of said wind turbine component based on a sum of said current euclidean distances comprises: and if the sum of the current Euclidean distances is larger than or equal to the distance threshold value, determining that the fan component operates abnormally.
5. The fault detection method of claim 1, wherein prior to processing the current characteristic parameter values of the wind turbine component through the trained self-organizing neural network to obtain the operational status of the wind turbine component, the fault detection method further comprises:
acquiring a normal characteristic parameter sample of the fan component in a normal running state;
and inputting the normal characteristic parameter sample of the fan component into an initial self-organizing neural network so as to train the initial self-organizing neural network and obtain a trained self-organizing neural network.
6. The fault detection method of claim 5, wherein the fault detection method further comprises:
acquiring abnormal characteristic parameter samples of the fan component in various fault running states;
and inputting the normal characteristic parameter sample and the abnormal characteristic parameter sample of the fan component into an initial convolutional neural network so as to train the initial convolutional neural network to obtain a trained convolutional neural network.
7. The fault detection method of claim 5, wherein the fault detection method further comprises:
and if the running state obtained through the trained self-organizing neural network is the abnormal running state and the classification result obtained through the convolutional neural network is the normal running type, retraining the convolutional neural network by at least utilizing the current characteristic parameter value, the normal characteristic parameter sample and the abnormal characteristic parameter sample.
8. The fault detection method of claim 5, wherein the fault detection method further comprises:
and if the running state obtained through the trained self-organizing neural network is the normal running state and the classification result obtained through the trained convolutional neural network is one of the fault types, retraining the convolutional neural network by at least utilizing the current characteristic parameter value, the normal characteristic parameter sample and the abnormal characteristic parameter sample.
9. The fault detection method of claim 1, wherein the convolutional neural network comprises a one-dimensional convolutional neural network; and/or
The fan component comprises a rolling bearing, and the fault types of the rolling bearing comprise an inner ring fault, an outer ring fault, a retainer fault and a rolling body fault.
10. A fault detection system comprising one or more processors configured to implement the fault detection method of any one of claims 1-9.
11. A storage medium, having stored thereon a program which, when executed by a processor, implements the fault detection method according to any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011232566.9A CN112525505A (en) | 2020-11-06 | 2020-11-06 | Fault detection method, system and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011232566.9A CN112525505A (en) | 2020-11-06 | 2020-11-06 | Fault detection method, system and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112525505A true CN112525505A (en) | 2021-03-19 |
Family
ID=74979944
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011232566.9A Pending CN112525505A (en) | 2020-11-06 | 2020-11-06 | Fault detection method, system and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112525505A (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107631867A (en) * | 2017-09-07 | 2018-01-26 | 天津工业大学 | A kind of rotating machinery fault intelligent method for classifying based on deep learning |
| CN108398268A (en) * | 2018-03-15 | 2018-08-14 | 哈尔滨工业大学 | A kind of bearing performance degradation assessment method based on stacking denoising self-encoding encoder and Self-organizing Maps |
| CN109443769A (en) * | 2018-10-24 | 2019-03-08 | 中车株洲电力机车有限公司 | Detection method, system, device and the readable storage medium storing program for executing of motor bearings state |
| CN109765054A (en) * | 2019-01-22 | 2019-05-17 | 上海海事大学 | A kind of Fault Diagnosis of Roller Bearings |
| CN110410282A (en) * | 2019-07-24 | 2019-11-05 | 河北工业大学 | Wind turbines health status on-line monitoring and method for diagnosing faults based on SOM-MQE and SFCM |
| CN110544184A (en) * | 2019-09-24 | 2019-12-06 | 东北电力大学 | fan performance analysis method based on operation parameters |
| US20200131003A1 (en) * | 2017-10-20 | 2020-04-30 | China University Of Mining And Technology | Multiple-state health monitoring apparatus and monitoring method for critical components in hoisting system |
| CN111140986A (en) * | 2019-12-23 | 2020-05-12 | 珠海格力电器股份有限公司 | Operating state detection method and device of air conditioning system, storage medium and air conditioner |
| CN111753891A (en) * | 2020-06-11 | 2020-10-09 | 燕山大学 | Rolling bearing fault diagnosis method based on unsupervised feature learning |
-
2020
- 2020-11-06 CN CN202011232566.9A patent/CN112525505A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107631867A (en) * | 2017-09-07 | 2018-01-26 | 天津工业大学 | A kind of rotating machinery fault intelligent method for classifying based on deep learning |
| US20200131003A1 (en) * | 2017-10-20 | 2020-04-30 | China University Of Mining And Technology | Multiple-state health monitoring apparatus and monitoring method for critical components in hoisting system |
| CN108398268A (en) * | 2018-03-15 | 2018-08-14 | 哈尔滨工业大学 | A kind of bearing performance degradation assessment method based on stacking denoising self-encoding encoder and Self-organizing Maps |
| CN109443769A (en) * | 2018-10-24 | 2019-03-08 | 中车株洲电力机车有限公司 | Detection method, system, device and the readable storage medium storing program for executing of motor bearings state |
| CN109765054A (en) * | 2019-01-22 | 2019-05-17 | 上海海事大学 | A kind of Fault Diagnosis of Roller Bearings |
| CN110410282A (en) * | 2019-07-24 | 2019-11-05 | 河北工业大学 | Wind turbines health status on-line monitoring and method for diagnosing faults based on SOM-MQE and SFCM |
| CN110544184A (en) * | 2019-09-24 | 2019-12-06 | 东北电力大学 | fan performance analysis method based on operation parameters |
| CN111140986A (en) * | 2019-12-23 | 2020-05-12 | 珠海格力电器股份有限公司 | Operating state detection method and device of air conditioning system, storage medium and air conditioner |
| CN111753891A (en) * | 2020-06-11 | 2020-10-09 | 燕山大学 | Rolling bearing fault diagnosis method based on unsupervised feature learning |
Non-Patent Citations (2)
| Title |
|---|
| 张全德，陈果等: "《基于自组织神经网络的滚动轴承状态评估方法》", 《中国机械工程》 * |
| 艾延延，王克明，沙云东: "《航空发动机状态监测与故障诊断技术》", 北京理工大学出版社 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110276416B (en) | Rolling bearing fault prediction method | |
| CN109918752B (en) | Mechanical fault diagnosis method, equipment and medium based on migration convolution neural network | |
| CN111458142B (en) | Sliding bearing fault diagnosis method based on generation of countermeasure network and convolutional neural network | |
| Samanta et al. | Artificial neural networks and genetic algorithm for bearing fault detection | |
| Jack et al. | Genetic algorithms for feature selection in machine condition monitoring with vibration signals | |
| Malhi et al. | PCA-based feature selection scheme for machine defect classification | |
| CN110940523B (en) | Unsupervised domain adaptive fault diagnosis method | |
| Yang et al. | ART–KOHONEN neural network for fault diagnosis of rotating machinery | |
| Yang et al. | Bearing fault automatic classification based on deep learning | |
| CN111597651B (en) | Rolling bearing performance degradation evaluation method based on HWPSO-SVDD model | |
| Zhou et al. | Remaining Useful Life Prediction and Fault Diagnosis of Rolling Bearings Based on Short‐Time Fourier Transform and Convolutional Neural Network | |
| CN206504869U (en) | A kind of rolling bearing fault diagnosis device | |
| CN111353373A (en) | Correlation alignment domain adaptive fault diagnosis method | |
| CN108919059A (en) | A kind of electric network failure diagnosis method, apparatus, equipment and readable storage medium storing program for executing | |
| Lee et al. | Unsupervised anomaly detection of the gas turbine operation via convolutional auto-encoder | |
| CN114184367B (en) | Rotary mechanical equipment fault diagnosis method, device, equipment and readable storage medium | |
| CN112765890A (en) | Dynamic domain adaptive network-based multi-working-condition rotating machine residual life prediction method | |
| CN114091525A (en) | Rolling bearing degradation trend prediction method | |
| CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
| CN110956331A (en) | Method, system and device for predicting operation state of digital factory | |
| CN116881841A (en) | Hybrid model fault diagnosis method based on F1-score multistage decision analysis | |
| Saufi et al. | Machinery fault diagnosis based on a modified hybrid deep sparse autoencoder using a raw vibration time-series signal | |
| Zhou et al. | Degradation State Recognition of Rolling Bearing Based on K‐Means and CNN Algorithm | |
| Yadav et al. | Variable dropout one-dimensional CNN for vibration-based shaft unbalance detection in industrial machinery | |
| CN112525505A (en) | Fault detection method, system and storage medium |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210319 |