CN111967364B - Composite fault diagnosis method, device, electronic equipment and storage medium - Google Patents
Composite fault diagnosis method, device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN111967364B CN111967364B CN202010800302.2A CN202010800302A CN111967364B CN 111967364 B CN111967364 B CN 111967364B CN 202010800302 A CN202010800302 A CN 202010800302A CN 111967364 B CN111967364 B CN 111967364B
- Authority
- CN
- China
- Prior art keywords
- sample
- mll
- neural network
- network model
- value
- 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.)
- Active
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 48
- 238000000034 method Methods 0.000 title claims abstract description 39
- 239000002131 composite material Substances 0.000 title claims description 59
- 238000003062 neural network model Methods 0.000 claims abstract description 78
- 238000007781 pre-processing Methods 0.000 claims abstract description 25
- 210000002569 neuron Anatomy 0.000 claims description 92
- 238000004422 calculation algorithm Methods 0.000 claims description 44
- 238000012549 training Methods 0.000 claims description 39
- 238000012360 testing method Methods 0.000 claims description 38
- 230000006870 function Effects 0.000 claims description 34
- 230000006872 improvement Effects 0.000 claims description 15
- 238000013528 artificial neural network Methods 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 7
- 238000009833 condensation Methods 0.000 claims description 5
- 230000005494 condensation Effects 0.000 claims description 5
- 238000011156 evaluation Methods 0.000 claims description 5
- 150000001875 compounds Chemical class 0.000 abstract description 10
- 238000010586 diagram Methods 0.000 description 14
- 230000004913 activation Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 230000035945 sensitivity Effects 0.000 description 5
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008707 rearrangement Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- 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
- G01M13/02—Gearings; Transmission mechanisms
-
- 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
- G01M13/02—Gearings; Transmission mechanisms
- G01M13/021—Gearings
-
- 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
- G01M13/02—Gearings; Transmission mechanisms
- G01M13/028—Acoustic or vibration analysis
-
- 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
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Acoustics & Sound (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Signal Processing (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The embodiment of the application provides a compound fault diagnosis method, a device, electronic equipment and a storage medium, which are used for preprocessing a plurality of paths of working state signals of a rotary machine to be detected by acquiring the plurality of paths of working state signals of the rotary machine to be detected to obtain a characteristic signal data set, analyzing the characteristic signal data set according to a BP-MLL neural network model to determine the compound fault type of the rotary machine to be detected, so that the compound fault diagnosis of the rotary machine is realized, and the diagnosis accuracy of the compound fault is improved.
Description
Technical Field
The embodiment of the application relates to a mechanical fault diagnosis technology, in particular to a composite fault diagnosis method, a device, electronic equipment and a storage medium.
Background
Rotary machines play an important role in machine tools as power transmission devices in various machine tools, and fault location of the rotary machine tools is an important point in machine tool management and maintenance work. The composite fault is a concept opposite to a single fault, and the related algorithm of the current multi-junction neural network performs composite fault positioning due to the complex structure and working environment of the rotary machine.
In the prior art, vibration acceleration signals of a rotary machine under single fault and compound fault working conditions are collected, certain sample extraction parameters are set to intercept and extract a plurality of samples, for each sample, a single fault is given to a single label, a compound fault is given to a plurality of labels, then a sample set of the given label is randomly divided into a training set and a testing set according to a certain proportion, a Keras is utilized to build a deep one-dimensional convolutional neural network, an output layer activation function is set as a Sigmoid activation function, a cost function is set as a boundary loss function, vibration data of the training set is directly input into the one-dimensional convolutional neural network for training under the condition that the samples are not subjected to any preprocessing, an optimal model is selected through Grid Search, and the optimal model is applied to the testing set, so that fault state classification results are obtained.
In the prior art, although the composite fault can be diagnosed, the composite fault is distinguished only by a plurality of labels, and the association relation among the faults is not considered in the prior art, so that the problem of low accuracy exists when the composite fault diagnosis is performed on the rotary machine by adopting a model in the prior art.
Disclosure of Invention
The application provides a compound fault diagnosis method, a compound fault diagnosis device, electronic equipment and a storage medium, which are used for solving the problem of low diagnosis accuracy in the prior art.
In a first aspect, an embodiment of the present application provides a composite fault diagnosis method, including:
acquiring a multi-path working state signal of a rotary machine to be detected;
preprocessing the multipath working state signals to obtain a characteristic signal data set;
and analyzing the characteristic signal data set according to a back propagation multi-label learning BP-MLL neural network model, and determining the composite fault type of the rotary machine to be detected.
Optionally, the BP-MLL neural network model is trained according to a BP-MLL improvement algorithm for determining a distance function from a value of a set of labels associated with the sample and a value of a set of labels not associated with the sample:
wherein, i represents the sample number, k represents the sample number of the relevant tag, l represents the sample number of the irrelevant tag, +.>Output value representing output layer neuron of relevant tag location,/->Representing the output value of the uncorrelated tag location output layer neurons, Y i Representing a set of labels associated with a sample, +.>Representing a sample independent set of labels, |Y i I represents the value of the tag set associated with the sample, +.>A value representing a sample independent set of labels, f distance Representing a distance function.
Optionally, the BP-MLL improvement algorithm is configured to pass an error function:
adjusting the weight value of the neural network, wherein E represents an error, I represents the logarithm of the sample data set, beta represents a regularization coefficient, and V td Representing the weight value, W, between the t-th hidden layer neuron and the d-th input layer neuron st The weight value between the s-th output layer neuron and the t-th hidden layer neuron is represented, N represents the number of the input layer neurons, M represents the number of the hidden layer neurons, and J represents the number of the output layer neurons.
Optionally, before analyzing the characteristic signal data set according to the back propagation multi-label learning BP-MLL neural network model and determining the composite fault type of the rotating machine to be detected, the method further includes:
acquiring a training sample set and a test sample set, wherein data in the training sample set and the test sample set are obtained by preprocessing a pre-acquired working state signal of a rotary machine;
training the training sample set by using a BP-MLL improved algorithm to obtain an initial neural network model;
and testing the initial neural network model according to a test sample set to obtain the BP-MLL neural network model.
Optionally, the preprocessing the multiple paths of working state signals to obtain a characteristic signal data set includes:
performing feature extraction on the time domain and the frequency domain of the multi-path working state signals to obtain a time-frequency domain feature parameter set of the multi-path working state signals;
and constructing the characteristic signal data set through a condensation degree evaluation algorithm according to the time-frequency domain characteristic parameter set.
In a second aspect, an embodiment of the present application provides a composite fault diagnosis apparatus, including:
the acquisition module is used for acquiring multi-path working state signals of the rotary machine to be detected;
the processing module is used for preprocessing the multipath working state signals to obtain a characteristic signal data set; analyzing the characteristic signal data set according to a back propagation multi-label learning BP-MLL neural network model, determining the composite fault type of the rotary machine to be detected, wherein the BP-MLL neural network model is trained according to a BP-MLL improved algorithm.
Optionally, the BP-MLL neural network model is trained according to a BP-MLL improvement algorithm for determining a distance function from a value of a set of labels associated with the sample and a value of a set of labels not associated with the sample:
wherein, i represents the sample number, k represents the sample number of the relevant tag, l represents the sample number of the irrelevant tag, +.>Output value representing output layer neuron of relevant tag location,/->Representing the output value of the uncorrelated tag location output layer neurons, Y i Representing a set of labels associated with a sample, +.>Representing a sample independent set of labels, |Y i I represents the value of the tag set associated with the sample, +.>A value representing a sample independent set of labels, f distance Representing a distance function;
the BP-MLL refinement algorithm is used to pass the error function:
the weight value of the neural network is adjusted, wherein,e represents error, I represents logarithm of sample data set, beta represents regularization coefficient, V td Representing the weight value, W, between the t-th hidden layer neuron and the d-th input layer neuron st The weight value between the s-th output layer neuron and the t-th hidden layer neuron is represented, N represents the number of the input layer neurons, M represents the number of the hidden layer neurons, and J represents the number of the output layer neurons.
Optionally, the acquiring module is further configured to acquire a training sample set and a test sample set, where data in the training sample set and the test sample set are obtained by preprocessing a pre-acquired working state signal of the rotating machine;
the processing module is further used for training the training sample set by utilizing a BP-MLL improved algorithm to obtain an initial neural network model; and testing the initial neural network model according to a test sample set to obtain the BP-MLL neural network model.
In a third aspect, an embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the composite fault diagnosis method according to the first aspect.
In a fourth aspect, an embodiment of the present application provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a composite fault diagnosis method as described in the first aspect above.
According to the composite fault diagnosis method, the device, the electronic equipment and the storage medium, the multi-path working state signals of the rotary machine to be detected are obtained, the multi-path working state signals are preprocessed to obtain the characteristic signal data set, the characteristic signal data set is analyzed according to the BP-MLL neural network model, the composite fault type of the rotary machine to be detected is determined, and the composite fault diagnosis of the rotary machine is realized.
Drawings
Fig. 1 is a flow chart of a composite fault diagnosis method according to a first embodiment of the present application;
fig. 2 is a schematic diagram of a three-layer neural network structure according to an embodiment of the present application;
FIG. 3 is a schematic logic flow diagram of a composite fault diagnosis according to an embodiment of the present application;
fig. 4 is a flow chart of a composite fault diagnosis method according to a second embodiment of the present application;
FIG. 5 is a schematic diagram of training logic of a BP-MLL neural network model according to a second embodiment of the application;
FIG. 6 is a diagram showing the variation of error values with the rotation provided by the third embodiment of the present application;
FIG. 7 is a schematic diagram of a classification result of a BP-MLL neural network model according to a third embodiment of the application;
fig. 8 is a schematic structural diagram of a composite fault diagnosis device according to a fourth embodiment of the present application;
fig. 9 is a schematic structural diagram of an electronic device according to a fifth embodiment of the present application.
Detailed Description
The application is described in further detail below with reference to the drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the application and are not limiting thereof. It should be further noted that, for convenience of description, only some, but not all of the structures related to the present application are shown in the drawings.
First, terms related to the embodiments of the present application will be explained:
rotating machinery: the rotary machine is usually composed of a rotating shaft and various disc-shaped parts fixed on the rotating shaft, such as a bearing, a gear box, an impeller and the like.
Composite fault: refers to two or more failures occurring simultaneously, such as ball bearing failure and missing teeth, inner bearing failure and tooth breakage in rotating machinery, etc.
In the prior art, when the compound faults are diagnosed through the multi-label, only simple superposition of a plurality of faults is considered, but the association relation among the plurality of faults is not considered, so that the problem of low diagnosis accuracy exists in the prior art. Based on the technical problems existing in the prior art, the embodiment of the application provides a composite fault diagnosis method, which is based on a BP-MLL improved algorithm, a BP-MLL neural network model is constructed, and the correlation between labels is reserved by carrying out exponential operation on difference values between neurons marked on an output layer and neurons not marked on the output layer, so that the characteristics of multi-mark learning are captured, the diagnosis accuracy of composite faults is improved, and meanwhile, the distance between the actual output value of the neurons of the relevant labels and the actual output value of the neurons of the irrelevant labels is as large as possible by improving the error function in the original BP-MLL algorithm, so that the diagnosis accuracy of composite faults is further improved.
Example 1
Fig. 1 is a schematic flow chart of a composite fault diagnosis method according to an embodiment of the present application, where the method may be performed by an electronic device having a composite fault diagnosis function, such as a rotary machine management/control platform, a terminal device, etc. As shown in fig. 1, the method of this embodiment specifically includes the following steps:
s110, acquiring a multi-path working state signal of the rotary machine to be detected.
In this step, for example, the electronic device for performing the composite fault diagnosis may be connected to a multi-path sensor, where the multi-path sensor is used to measure working state signals of the rotating machine to be detected, such as a force signal, a moment signal, and a vibration signal, and it may be understood that the electronic device has a function of controlling the multi-path sensor, and may obtain a corresponding multi-path working state signal according to a requirement, for example, may obtain a multi-path working state signal of the rotating machine to be detected acquired by the multi-path sensor in real time, or may obtain a multi-path working state signal of the rotating machine acquired by the multi-path sensor when the rotating machine to be detected fails.
The rotary machine to be detected is a rotary machine which needs to be subjected to compound fault diagnosis.
The working state signals are parameters related to mechanical performance of the rotary machine, for example, the force signals, the moment signals and the vibration signals can be included in multiple paths of working state signals, and the force signals, the moment signals and the vibration signals can be multiple paths, for example, the multiple paths of working state signals can include 2 paths of force signals, 3 paths of moment signals and 2 paths of vibration signals, and the working state signals can be determined according to the structure and actual analysis requirements of the rotary machine.
In the embodiment, the composite fault diagnosis is performed by acquiring the multipath working state signals of the rotary machine to be detected, so that the composite fault diagnosis accuracy is improved.
S120, preprocessing the multipath working state signals to obtain a characteristic signal data set.
In the step, the input characteristic signal data set meeting the BP-MLL neural network model is obtained by preprocessing the multipath working state information acquired in the step S110.
Optionally, the preprocessing is performed on the multiple working state signals to obtain a characteristic signal data set, which may be implemented by the following steps:
A. and sequentially carrying out feature extraction on the time domain and the frequency domain of each path of working state signals to obtain a time-frequency domain feature parameter set of the multipath working state signals.
Feature extraction is a key factor affecting accuracy and efficiency of the model, and in this step, for example, when feature extraction is performed on the time domain and the frequency domain of each path of working state signal, the feature extraction may be performed by combining a fast fourier transform, a wavelet transform and a feature parameter calculation method.
B. And constructing a characteristic signal data set through a condensation degree evaluation algorithm according to the time-frequency domain characteristic parameter set.
In the step, the sensitivity factors of each time-frequency domain characteristic parameter are calculated firstly according to a condensation degree evaluation algorithm, and then the sensitivity factors are ordered, and the first d parameters with larger sensitivity factors are selected to construct a sensitivity vector, so that a characteristic signal data set is obtained.
Illustratively, the sensitivity factors are represented by η, and are ordered by ν=sort (η) to construct a feature signal dataset
S130, analyzing the characteristic signal data set according to the BP-MLL neural network model, and determining the composite fault type of the rotary machine to be detected.
In the step, the BP-MLL neural network model is a neural network model which is trained in advance according to a BP-MLL improved algorithm, and in the step, the characteristic signal data set obtained through preprocessing in the S120 is input into the BP-MLL neural network model, and the characteristic signal data set is analyzed through the BP-MLL neural network model to realize characteristic classification, so that the composite fault type of the rotary machine to be detected is determined.
The BP-MLL improvement algorithm is obtained by improving an error formula of the BP-MLL algorithm, specifically, a regular term is added into an original error formula of the BP-MLL algorithm, and a distance function of the original error formula is adjusted.
For example, fig. 2 is a schematic diagram of a three-layer neural network structure provided in the embodiment of the present application, and the principle of implementing feature classification by the BP-MLL improvement algorithm is described by taking the neural network structure shown in fig. 2 as an example, it can be understood that the technical scheme of the present application is also applicable when the hidden layer is two or more layers.
The three-layer neural network structure shown in FIG. 2 comprises an input layer composed of d neurons, an hidden layer composed of t neurons and an output layer composed of s neurons, wherein hollow circles and solid circles in the figure represent the neurons of each layer, and a 1 ,a 2 ,……,a d Corresponding to the input layer neuron, { a 1 ,a 2 ,……,a d The input value of the neurons of the input layer, b 1 ,……,b t Corresponding hidden layer neurons, { b 1 ,……,b t The output value of the hidden layer neuron, c 1 ,……,c s Corresponding to the output layer neuron, { c 1 ,……,c s The output value of the output layer neuron, a 0 、b 0 Corresponding to the bias neuron, a 0 Representing the value of the bias term of the hidden layer, b 0 Representing the value of the bias term of the output layer. Except bias neurons, all hidden layer neurons and input layer neurons are connected in pairs, and the weight value is V td Table type, e.g. V 12 Representing weight values between hidden layer neuron 1 and input layer neuron 2, all output layer neurons and hidden layer neurons being connected in pairs, the weight values being W st Watch type, e.g. W 12 Representing the weight value between the output layer neuron 1 and the hidden layer neuron 2.
Assume that a dataset has I-pair multi-tag samples { (X) 1 ,Y 1 ),…,(X i ,Y i ) I=1, …, I represents a sample number, I is a positive integer of 1 or more, each X i Is a d-dimensional input characteristic vector and corresponds to an output vector Y i Is a set of labels, which is divided into X i Related tags and unrelated tags, using Y i Representation and X i A set of related tags, usingRepresenting a sample independent set of labels, and accordingly, |Y i I represents the value of the tag set associated with the sample, +.>A value representing a set of labels that is independent of the sample.
Each layer is completely connected with the next layer and has weight V td ,W st ](d=1, 2, …, N; t=1, 2, …, M; s=1, 2, …, J), wherein N, M, J respectively represents the numbers of the input layer neurons, hidden layer neurons and output layer neurons, N, M, J are positive integers greater than or equal to 1, and d, t, s are respectively the input layer neurons, hidden layer neuronsLayer neurons and output layer neuron sequence numbers.
(1) The BP-MLL refinement algorithm defines the distance function as follows:
wherein, output value representing output layer neuron of relevant tag location,/->Output value of output layer neuron representing uncorrelated label position, |Y i I represents the value of the tag set associated with the sample, +.>A value representing a sample independent set of labels, f distance Representing a distance function.
Representing when |Y i When =0 (indicating that all tags are uncorrelated with the sample, i.e. the case of empty tags), let +.>Get 1, when |Y i When l not equal to 0 (indicating that it is not an empty tag), let +.>Output value equal to output layer->
Indicating when->(indicating that all tags are related to the sample, i.e. the case of a full tag), let +.>Take-1, when->(indicating that it is not a full label), let +.>Output value equal to output layer->
It follows that if one sample does not have any tag label (empty tag), i.e. |Y i I=0, i.e. there are only uncorrelated labels, where the distance function is:
if all tags are marked in one sample, i.eI.e. all the labels are related (full label), the distance function is then:
ensuring the correlation between labels through a distance function to ensure that the output value of the neuron at the position of the relevant label is larger than that of the neuron at the position of the relevant labelOutput values of the relevant tag location neurons and by correlation with |Y i I and Y i I pairAnd->The value of (1) is subjected to classified discussion, so that the BP-MLL neural network model is also suitable for the conditions of empty labels and full labels, and the accuracy of the BP-MLL neural network model in diagnosing the composite fault is improved.
(2) The BP-MLL refinement algorithm defines the following error function:
wherein E represents an error;
beta represents the regularization coefficient and,the method is a regular term, and the regular term is used for balancing errors, so that the situations of infinite or infinitesimal error values and the like are avoided;
max(1,|Y i i) represents taking 1 and Y i The larger value in the i is used,the representation takes 1 and +.>Through max (1, |y) i I) and +.>The multiplied value is used as the denominator of the first term, so that the denominator of the first term in the error formula is not 0, the error formula can be suitable for common conditions, full labels and empty labels, and the applicability of the algorithm is improved.
In the BP-MLL improvement algorithm, the error is used forEquation 4 calculates an error value between the output layer output value and the expected value, and adjusts the weight V according to the error value td ,W st ]Until the error value meets the preset condition.
(3) Updating of weights
Order the(wherein c s Representing the output value of the s-th output layer neuron, sc s E is a weighted sum of the output values of the s output layer neurons i Error values representing i samples), h is the activation function, derived:
where h' is the derivative of the activation function, c l Representing the output value of the uncorrelated tag location neurons, c k Representing the output value of the associated tag location neuron.
In this embodiment, the weight value W of the neural network is calculated according to the following formula st Updating:
W st =W st +ΔW st equation 6
Wherein alpha is learning rate, b t Is the output value of the t-th hidden layer neuron.
V is also described as td Update mode and W of (2) st The updating manner of (c) is similar and will not be described in detail herein.
(3) Classifying the output result according to the following formula:
where ε represents the threshold value of the output value and res represents the marking result. The above formula indicates that when the output value of the output layer is greater than the threshold epsilon, it is marked as 1, and when the output value of the output layer is less than or equal to the threshold epsilon, it is marked as-1.
In this embodiment, by collecting the working state signal of the rotating machine and training the working state signal according to the BP-MLL improvement algorithm, a BP-MLL neural network model can be obtained, and when the characteristic signal dataset of the rotating machine to be detected is input into the BP-MLL neural network model, the diagnosis of the composite fault type of the rotating machine to be detected can be realized. Fig. 3 is a logic flow diagram of a composite fault diagnosis according to an embodiment of the present application.
In this embodiment, by acquiring multiple paths of working state signals of the rotating machine to be detected, preprocessing the multiple paths of working state signals to obtain a characteristic signal data set, analyzing the characteristic signal data set according to a BP-MLL neural network model, determining a composite fault type of the rotating machine to be detected, and realizing composite fault diagnosis of the rotating machine.
Example two
In the first embodiment, before the characteristic signal data set is analyzed according to the BP-MLL neural network model to determine the composite fault type of the rotating machine to be detected, the method further includes a process of constructing the BP-MLL neural network model, and a specific embodiment of the process of constructing the BP-MLL neural network model will be described below.
Fig. 4 is a flow chart of a composite fault diagnosis method provided in the second embodiment of the present application, as shown in fig. 4, in this embodiment, the constructing a BP-MLL neural network model includes:
s210, acquiring a training sample set and a test sample set.
In this step, the data in the training sample set and the test sample set are obtained by preprocessing the collected working state signals of the rotating machine in advance, and the preprocessing implementation flow is similar to the preprocessing of the multipath working state signals of the rotating machine to be detected in the first embodiment, and will not be described in detail here.
It will be appreciated that the rotary machine that generates the operating state signals of the training sample set and the test sample set may be the same as or different from the rotary machine to be tested, and that the operating state information of the rotary machine used to derive the training sample set and the test sample set is collected with the type of fault known.
S220, training the training sample set by utilizing a BP-MLL improved algorithm to obtain an initial neural network model.
In this step, the training sample set is illustratively trained using a BP-MLL improvement algorithm, and the obtaining of the initial neural network model may be achieved by:
A. determining the structure of a neural network, the number of neurons of each layer and an activation function, initializing the values of weight values and bias items connected among the neurons of each layer, and setting the updating step length, the rotation and the error threshold value of the weight;
B. executing a forward propagation process, calculating an output value of an output layer and a loss value between the output value and an actual value, namely an error, according to a formula 4 according to the input sample, a weight value connected between neurons of each layer and a value of a bias term, if the error is smaller than a preset threshold value, obtaining an initial neural network model, and stopping training; if the error is greater than or equal to the preset threshold, executing the step C;
C. and executing a back propagation process, and updating the weight value of the connection between the neurons of each layer according to a weight updating formula 6 and a weight updating formula 7.
D. Step B, C is repeated until a set maximum run is reached or the error between the output value of the output layer and the actual value is less than the error threshold.
S230, testing the initial neural network model according to the test sample set to obtain the BP-MLL neural network model.
In the step, a test sample set is input into an initial neural network model, the initial neural network model is tested, whether the output of the initial neural network model is effective classification is judged, if the test results are all effective classification, the initial neural network model is used as a BP-MLL neural network model, and if the test results cannot meet the requirement of effective classification, the initial neural network model is adjusted according to the test results, so that the BP-MLL neural network model is obtained.
Specifically, according to the weight value of the last round in S220, a forward propagation process is performed on the test sample set, the confidence level of the predicted label of the test sample set is obtained, and the label of the predicted multi-label set is obtained. Fig. 5 is a schematic diagram of training logic of a BP-MLL neural network model according to a second embodiment of the present application.
In this embodiment, the data in the training sample set and the test sample set are obtained by preprocessing the working state signal of the rotary machine acquired in advance, the training sample set is trained by using the BP-MLL improved algorithm to obtain an initial neural network model, the initial neural network model is tested according to the test sample set to obtain a BP-MLL neural network model, and the reliability and accuracy of the BP-MLL neural network model are effectively ensured by preprocessing, algorithm improvement and measurement, thereby being beneficial to improving the diagnosis accuracy and efficiency of the rotary machine composite fault.
Example III
The construction and classification effects of the BP-MLL neural network model will be described below by taking a composite failure diagnosis of a rotary machine having a bearing and a gear box as an example.
TABLE 1
Type(s) | Quantity of |
Normal (without fault) | 48 |
Ball bearing failure and missing teeth | 48 |
Outer bearing failure and missing teeth | 48 |
External bearing failure and tooth breakage | 48 |
Inner bearing failure and tooth breakage | 48 |
The training sample set is 240 sets of samples of 5 fault types (including 48 non-faulty) as shown in table 1. The test sample set is also another 240 sets of samples of 5 fault types.
And training an initial neural network model by adopting a BP-MLL improved algorithm on a training sample set, then checking the accuracy of fault detection on a test sample set, and adjusting the initial neural network model according to the condition of effective classification to obtain the BP-MLL neural network model.
By reading the data in a personal computer (8G RAMi5-7200U CPU@2.70GHz) is applied to the BP-MLL neural network model, the BP-MLL neural network model can successfully classify 5 kinds of faults shown in the table 1, the accuracy reaches 100 percent, and the total training time is 55.3102 seconds. Table 2 shows the configuration of the neural network structure in the BP-MLL neural network model. In Table 2, the learning rate alpha is an exponentially decaying learning rate, and the learning rate is an initial value alpha 0 And round correlation (epochs) are formulated as α=0.95 epochs *α 0 。
TABLE 2
Project | Configuration of |
Number of neurons in input layer | 32 |
Number of hidden layer neurons | 44 |
Number of neurons in output layer | 5 |
Learning rate (alpha) | 0.95 epochs *α 0 |
Round (epochs) | 100 |
Fig. 6 is a schematic diagram of the variation of the error value with the number of rounds according to the third embodiment of the present application, as can be seen from fig. 6, the error value becomes smaller with the number of rounds, and the error value becomes stable when the number of rounds reaches 20.
Fig. 7 is a schematic diagram of classification results of a BP-MLL neural network model according to a third embodiment of the present application, and fig. 7 is a graph obtained by reducing an output result of the BP-MLL neural network model from 5 dimensions to 3 dimensions using a t-distributed random neighborhood embedding (t-distributed stochastic neighbor embedding, t-SNE) algorithm. As can be seen from fig. 7, the BP-MLL neural network model can be used to successfully classify the composite fault.
Example IV
Fig. 8 is a schematic structural diagram of a composite fault diagnosis apparatus according to a fourth embodiment of the present application, and as shown in fig. 8, a composite fault diagnosis apparatus 10 according to the present embodiment includes:
an acquisition module 11 and a processing module 12.
The acquisition module 11 is used for acquiring multiple paths of working state signals of the rotary machine to be detected;
the processing module 12 is used for preprocessing the multi-path working state signals to obtain a characteristic signal data set; analyzing the characteristic signal data set according to a back propagation multi-label learning BP-MLL neural network model, determining the composite fault type of the rotary machine to be detected, wherein the BP-MLL neural network model is trained according to a BP-MLL improved algorithm.
Optionally, the BP-MLL neural network model is trained according to a BP-MLL improvement algorithm for determining a distance function from a value of a set of labels associated with the sample and a value of a set of labels not associated with the sample:
wherein, i represents the sample number, k represents the sample number of the relevant tag, l represents the sample number of the irrelevant tag, +.>Output value representing output layer neuron of relevant tag location,/->Representing the output value of the uncorrelated tag location output layer neurons, Y i Representing a set of labels associated with a sample, +.>Representing a sample independent set of labels, |Y i I represents the value of the tag set associated with the sample, +.>A value representing a sample independent set of labels, f distance Representing a distance function;
the BP-MLL refinement algorithm is used to pass the error function:
adjusting the weight value of the neural network, wherein E represents an error, I represents the logarithm of the sample data set, beta represents a regularization coefficient, and V td Representing the weight value, W, between the t-th hidden layer neuron and the d-th input layer neuron st The weight value between the s-th output layer neuron and the t-th hidden layer neuron is represented, N represents the number of the input layer neurons, M represents the number of the hidden layer neurons, and J represents the number of the output layer neurons.
Optionally, the acquiring module 11 is further configured to acquire a training sample set and a test sample set, where data in the training sample set and the test sample set are obtained by preprocessing a pre-acquired working state signal of the rotating machine;
the processing module 12 is further configured to train the training sample set by using a BP-MLL improvement algorithm to obtain an initial neural network model; and testing the initial neural network model according to a test sample set to obtain the BP-MLL neural network model.
Optionally, the processing module 12 is specifically configured to:
performing feature extraction on the time domain and the frequency domain of the multi-path working state signals to obtain a time-frequency domain feature parameter set of the multi-path working state signals;
and constructing the characteristic signal data set through a condensation degree evaluation algorithm according to the time-frequency domain characteristic parameter set.
The composite fault diagnosis device provided by the embodiment of the application can execute the composite fault diagnosis method provided by any method embodiment of the application, and has the corresponding functional modules and beneficial effects of the execution method. The implementation principle and technical effect of the present embodiment are similar to those of the above method embodiment, and are not described here again.
Example five
Fig. 9 is a schematic structural diagram of an electronic device according to a fifth embodiment of the present application, and as shown in fig. 9, the electronic device 20 includes a memory 21, a processor 22, and a computer program stored in the memory and capable of running on the processor; the number of processors 22 of electronic device 20 may be one or more, one processor 22 being taken as an example in fig. 9; the processor 22, the memory 21 in the electronic device 20 may be connected by a bus or otherwise, in fig. 9 by way of example.
The memory 21 is a computer-readable storage medium that can be used to store software programs, computer-executable programs, and modules, such as program instructions/modules corresponding to the acquisition module 11 and the processing module 12 in the embodiment of the present application. The processor 22 executes various functional applications of the device/terminal/server and data processing by running software programs, instructions and modules stored in the memory 21, i.e., implements the above-described composite fault diagnosis method.
The memory 21 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, at least one application program required for functions; the storage data area may store data created according to the use of the terminal, etc. In addition, memory 21 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device. In some examples, memory 21 may further include memory remotely located relative to processor 22, which may be connected to the device/terminal/server via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
Example six
Embodiment D of the present application also provides a computer-readable storage medium having stored thereon a computer program for performing a composite fault diagnosis method when executed by a computer processor, the method comprising:
acquiring a multi-path working state signal of a rotary machine to be detected;
preprocessing the multipath working state signals to obtain a characteristic signal data set;
analyzing the characteristic signal data set according to a back propagation multi-label learning BP-MLL neural network model, determining the composite fault type of the rotary machine to be detected, wherein the BP-MLL neural network model is trained according to a BP-MLL improved algorithm.
Of course, the computer program of the computer readable storage medium according to the embodiment of the present application is not limited to the method operations described above, and may also perform the related operations in the composite fault diagnosis method according to any embodiment of the present application.
From the above description of embodiments, it will be clear to a person skilled in the art that the present application may be implemented by means of software and necessary general purpose hardware, but of course also by means of hardware, although in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as a floppy disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a FLASH Memory (FLASH), a hard disk or an optical disk of a computer, etc., and include several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method according to the embodiments of the present application.
It should be noted that, in the above-mentioned embodiments of the search apparatus, each unit and module included are only divided according to the functional logic, but not limited to the above-mentioned division, as long as the corresponding functions can be implemented; in addition, the specific names of the functional units are also only for distinguishing from each other, and are not used to limit the protection scope of the present application.
Note that the above is only a preferred embodiment of the present application and the technical principle applied. It will be understood by those skilled in the art that the present application is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the application. Therefore, while the application has been described in connection with the above embodiments, the application is not limited to the embodiments, but may be embodied in many other equivalent forms without departing from the spirit or scope of the application, which is set forth in the following claims.
Claims (7)
1. A composite fault diagnosis method, comprising:
acquiring a multi-path working state signal of a rotary machine to be detected;
preprocessing the multipath working state signals to obtain a characteristic signal data set;
analyzing the characteristic signal data set according to a back propagation multi-label learning BP-MLL neural network model, and determining the composite fault type of the rotary machine to be detected;
the BP-MLL neural network model is trained according to a BP-MLL improvement algorithm, and the BP-MLL improvement algorithm is used for determining a distance function according to the value of a label set related to a sample and the value of a label set unrelated to the sample:
wherein,i represents the sample number, k represents the sample number of the relevant tag, l represents the sample number of the irrelevant tag, +.>Output value representing output layer neuron of relevant tag location,/->Representing the output value of the uncorrelated tag location output layer neurons, Y i Representing a set of labels associated with a sample, +.>Representing a sample independent set of labels, |Y i I represents the value of the tag set associated with the sample, +.>A value representing a sample independent set of labels, f distance Representing a distance function;
the BP-MLL refinement algorithm is used to pass the error function:
adjusting the weight value of the neural network, wherein E represents an error, I represents the logarithm of the sample data set, beta represents a regularization coefficient, and V td Representing the weight value, W, between the t-th hidden layer neuron and the d-th input layer neuron st The weight value between the s-th output layer neuron and the t-th hidden layer neuron is represented, N represents the number of the input layer neurons, M represents the number of the hidden layer neurons, and J represents the number of the output layer neurons.
2. The method of claim 1, wherein the analyzing the characteristic signal data set according to a back propagation multi-tag learning BP-MLL neural network model, prior to determining the composite fault type of the rotating machine to be detected, further comprises:
acquiring a training sample set and a test sample set, wherein data in the training sample set and the test sample set are obtained by preprocessing a pre-acquired working state signal of a rotary machine;
training the training sample set by using a BP-MLL improved algorithm to obtain an initial neural network model;
and testing the initial neural network model according to a test sample set to obtain the BP-MLL neural network model.
3. The method of claim 1, wherein preprocessing the multiple operating state signals to obtain a characteristic signal data set comprises:
performing feature extraction on the time domain and the frequency domain of the multi-path working state signals to obtain a time-frequency domain feature parameter set of the multi-path working state signals;
and constructing the characteristic signal data set through a condensation degree evaluation algorithm according to the time-frequency domain characteristic parameter set.
4. A composite fault diagnosis apparatus, comprising:
the acquisition module is used for acquiring multi-path working state signals of the rotary machine to be detected;
the processing module is used for preprocessing the multipath working state signals to obtain a characteristic signal data set; analyzing the characteristic signal data set according to a back propagation multi-label learning BP-MLL neural network model, and determining the composite fault type of the rotary machine to be detected;
the BP-MLL neural network model is trained according to a BP-MLL improvement algorithm, and the BP-MLL improvement algorithm is used for determining a distance function according to the value of a label set related to a sample and the value of a label set unrelated to the sample:
wherein,i represents the sample number, k represents the sample number of the relevant tag, l represents the sample number of the irrelevant tag, +.>Output value representing output layer neuron of relevant tag location,/->Representing the output value of the uncorrelated tag location output layer neurons, Y i Representing a set of labels associated with a sample, +.>Representing a sample independent set of labels, |Y i I represents the value of the tag set associated with the sample, +.>A value representing a sample independent set of labels, f distance Representing a distance function;
the BP-MLL refinement algorithm is used to pass the error function:
adjusting the weight value of the neural network, wherein E represents an error, I represents the logarithm of the sample data set, beta represents a regularization coefficient, and V td Representing the weight value, W, between the t-th hidden layer neuron and the d-th input layer neuron st The weight value between the s-th output layer neuron and the t-th hidden layer neuron is represented, N represents the number of the input layer neurons, M represents the number of the hidden layer neurons, and J represents the number of the output layer neurons.
5. The apparatus of claim 4, wherein the device comprises a plurality of sensors,
the acquisition module is further used for acquiring a training sample set and a test sample set, wherein data in the training sample set and the test sample set are obtained by preprocessing a pre-acquired working state signal of the rotary machine;
the processing module is further used for training the training sample set by utilizing a BP-MLL improved algorithm to obtain an initial neural network model; and testing the initial neural network model according to a test sample set to obtain the BP-MLL neural network model.
6. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the composite fault diagnosis method of any of claims 1-3 when the program is executed by the processor.
7. A computer readable storage medium having stored thereon a computer program, characterized in that the program, when executed by a processor, implements a composite fault diagnosis method according to any of claims 1-3.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010800302.2A CN111967364B (en) | 2020-08-11 | 2020-08-11 | Composite fault diagnosis method, device, electronic equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010800302.2A CN111967364B (en) | 2020-08-11 | 2020-08-11 | Composite fault diagnosis method, device, electronic equipment and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111967364A CN111967364A (en) | 2020-11-20 |
CN111967364B true CN111967364B (en) | 2023-11-21 |
Family
ID=73365453
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010800302.2A Active CN111967364B (en) | 2020-08-11 | 2020-08-11 | Composite fault diagnosis method, device, electronic equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111967364B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113361557A (en) * | 2020-12-21 | 2021-09-07 | 南京仁智网络科技有限公司 | Training method of neural network for underground coal mine fire extinguishing control based on vibration data |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2001001344A2 (en) * | 1999-06-26 | 2001-01-04 | Axeon Limited | Neural network for performing real-time channel equalisation |
CN108664924A (en) * | 2018-05-10 | 2018-10-16 | 东南大学 | A kind of multi-tag object identification method based on convolutional neural networks |
CN109635677A (en) * | 2018-11-23 | 2019-04-16 | 华南理工大学 | Combined failure diagnostic method and device based on multi-tag classification convolutional neural networks |
CN109765054A (en) * | 2019-01-22 | 2019-05-17 | 上海海事大学 | A kind of Fault Diagnosis of Roller Bearings |
CN110907176A (en) * | 2019-09-30 | 2020-03-24 | 合肥工业大学 | Wasserstein distance-based fault diagnosis method for deep countermeasure migration network |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160086185A1 (en) * | 2014-10-15 | 2016-03-24 | Brighterion, Inc. | Method of alerting all financial channels about risk in real-time |
-
2020
- 2020-08-11 CN CN202010800302.2A patent/CN111967364B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2001001344A2 (en) * | 1999-06-26 | 2001-01-04 | Axeon Limited | Neural network for performing real-time channel equalisation |
CN108664924A (en) * | 2018-05-10 | 2018-10-16 | 东南大学 | A kind of multi-tag object identification method based on convolutional neural networks |
CN109635677A (en) * | 2018-11-23 | 2019-04-16 | 华南理工大学 | Combined failure diagnostic method and device based on multi-tag classification convolutional neural networks |
CN109765054A (en) * | 2019-01-22 | 2019-05-17 | 上海海事大学 | A kind of Fault Diagnosis of Roller Bearings |
CN110907176A (en) * | 2019-09-30 | 2020-03-24 | 合肥工业大学 | Wasserstein distance-based fault diagnosis method for deep countermeasure migration network |
Non-Patent Citations (2)
Title |
---|
An Improved Simultaneous Fault Diagnosis Method based on Cohesion Evaluation and BP-MLL for Rotating Machinery;Yixuan Zhang 等;《2020 IEEE 29th International Symposium on Industrial Electronics (ISIE)》;摘要、第II至IV节 * |
基于径向基神经网络和正则化极限学习机的多标签学习模型;单东 等;《模式识别与人工智能》;833-840 * |
Also Published As
Publication number | Publication date |
---|---|
CN111967364A (en) | 2020-11-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Sugumaran et al. | Fault diagnosis of roller bearing using fuzzy classifier and histogram features with focus on automatic rule learning | |
CN109655259A (en) | Combined failure diagnostic method and device based on depth decoupling convolutional neural networks | |
CN113111820B (en) | Rotary part fault diagnosis method and device based on improved CNN and relation module | |
CN110113226A (en) | A kind of method and device of detection device exception | |
CN116450399B (en) | Fault diagnosis and root cause positioning method for micro service system | |
RU2011129298A (en) | IDENTIFICATION OF FAILURES IN THE AIRCRAFT ENGINE | |
CN112729834B (en) | Bearing fault diagnosis method, device and system | |
CN108919059A (en) | A kind of electric network failure diagnosis method, apparatus, equipment and readable storage medium storing program for executing | |
CN111538311B (en) | Flexible multi-state self-adaptive early warning method and device for mechanical equipment based on data mining | |
CN112465124B (en) | Twin depth space-time neural network model acquisition/fault diagnosis method and device | |
CN116304604B (en) | Multivariate time series data anomaly detection and model training method and system | |
CN114266289A (en) | Complex equipment health state assessment method | |
CN111967364B (en) | Composite fault diagnosis method, device, electronic equipment and storage medium | |
CN115374842A (en) | Mechanical bearing fault classification method and system based on generation countermeasure network | |
CN110766100A (en) | Bearing fault diagnosis model construction method, bearing fault diagnosis method and electronic equipment | |
CN117149486A (en) | Alarm and root cause positioning method, model training method, device, equipment and medium | |
CN110808947A (en) | Automatic vulnerability quantitative evaluation method and system | |
CN114186593A (en) | Bearing fault identification method, device, equipment and storage medium under noise condition | |
CN114385451A (en) | Fault root cause analysis method | |
CN114528906A (en) | Fault diagnosis method, device, equipment and medium for rotary machine | |
CN113487141A (en) | Bearing state evaluation method based on multi-source information clustering fusion | |
CN107884362B (en) | Method for rapidly detecting spandex content in cotton, polyester and spandex blended fabric | |
CN115658371B (en) | Diagnosis algorithm quantitative recommendation method based on case learning and diagnosability analysis | |
Lyu et al. | A Novel Fault Diagnosis Method Based on Feature Fusion and Model Agnostic Meta-Learning | |
CN109474445B (en) | Distributed system root fault positioning method and device |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |