CN111476263B - Bearing defect identification method based on SDAE and improved GWO-SVM - Google Patents
Bearing defect identification method based on SDAE and improved GWO-SVM Download PDFInfo
- Publication number
- CN111476263B CN111476263B CN201911381284.2A CN201911381284A CN111476263B CN 111476263 B CN111476263 B CN 111476263B CN 201911381284 A CN201911381284 A CN 201911381284A CN 111476263 B CN111476263 B CN 111476263B
- Authority
- CN
- China
- Prior art keywords
- sdae
- gwo
- svm
- data
- training
- 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
- 230000007547 defect Effects 0.000 title claims abstract description 50
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000012549 training Methods 0.000 claims abstract description 43
- 238000000605 extraction Methods 0.000 claims abstract description 18
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 241000282461 Canis lupus Species 0.000 claims description 26
- 230000006870 function Effects 0.000 claims description 19
- 230000035772 mutation Effects 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 8
- 238000012360 testing method Methods 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 5
- 230000006872 improvement Effects 0.000 claims description 4
- 238000012544 monitoring process Methods 0.000 claims description 4
- 238000013075 data extraction Methods 0.000 claims description 3
- 238000000354 decomposition reaction Methods 0.000 claims description 3
- 230000005484 gravity Effects 0.000 claims description 3
- 230000009191 jumping Effects 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 2
- 238000013145 classification model Methods 0.000 abstract 1
- 238000003745 diagnosis Methods 0.000 description 6
- 238000005457 optimization Methods 0.000 description 5
- 238000013507 mapping Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 3
- 241000282421 Canidae Species 0.000 description 2
- 238000012512 characterization method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G06N3/04—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)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Acoustics & Sound (AREA)
- Image Analysis (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to a bearing defect identification method based on SDAE and an improved GWO-SVM, which comprises the following steps: 1. collecting vibration signals of the bearing under normal conditions and fault conditions, and preprocessing data; 2. constructing a four-layer initial stacking denoising self-coding SDAE, and training an initial stacking denoising self-coding network SDAE;3. establishing an improved GWO-SVM classifier model, extracting the deepest features of the SDAE, and training an improved GWO-SVM classifier; SDAE fine-tunes and retrains the improved GWO-SVM classifier until classification accuracy is met. The invention has the advantages that: according to the invention, the data characteristic extraction and the classification model are combined, so that the accuracy of recognition is improved; through the noise-removing self-coding training SDAE network, not only the characteristics of the original data are learned, but also the degraded characteristics after being destroyed are learned, so that the SDAE network has stronger generalization and robustness; the improved GWO method is better in terms of convergence speed and accuracy.
Description
Technical Field
The invention belongs to the field of fault identification, and particularly relates to a bearing defect identification method based on SDAE and an improved GWO-SVM.
Background
It is counted that among the types of faults that occur in the mechanical equipment, bearing faults account for 30% of the weight. If the fault occurs on large-scale ship equipment, serious safety accidents and economic losses can be caused; if the fault occurs on the aeroengine, the air parking can be caused, and the tragic of the death of the aircraft is destroyed. The fault defect of the bearing is timely found, reliable equipment defect information is provided for maintenance personnel, so that reasonable maintenance strategies are formulated, the accident occurrence probability can be reduced, and the service life and the production efficiency of the equipment are improved.
The traditional defect identification method and means play a good role in single system faults, but have poor diagnosis effect on composite faults and large equipment. The artificial intelligent diagnosis does not depend on a specific diagnosis object and a specific mathematical model, only trains a defect identification model through learning historical data, and then judges and positions the defect type by combining online data, thereby realizing online diagnosis of large equipment. Since intelligent fault diagnosis of complex systems often requires more abstraction, deeper networks are required.
At present, defect identification of equipment is mostly completed by monitoring parameters such as vibration signals related to the equipment and utilizing methods such as feature extraction, information fusion, pattern identification and the like. For example, fault diagnosis is achieved with wavelet transformation and support vector machines, and defect identification is achieved with kernel principal component analysis and support vector machines. However, the above two methods analyze only a single vibration signal, and the vibration signal is often an aliasing of a plurality of signals, and is greatly interfered by other signals, so that it is difficult to extract effective features. The method combines wavelet analysis and D-S evidence theory to realize defect identification, and combines wavelet packet, kernel principal component analysis and SVM to realize defect identification of a rotor and a bearing, and the two methods combine vibration signals and current signals to make up for the defects of a single signal, but the characteristic extraction process needs a great deal of priori knowledge, abundant signal processing theory and practical experience as support, and the characteristic extraction process and the defect identification process are completely isolated, so that the data characteristic extraction and model training are not connected.
In recent years, SVM has been increasingly applied to various classification and regression problems, and can solve the learning problem of small samples of equipment data and the uncertainty problem of evaluation results. However, the generalization ability of SVR is mainly determined by penalty coefficient C and kernel function parameter σ, and the selection of the two parameters affects the recognition rate. The intelligent optimization algorithm such as genetic algorithm and particle swarm algorithm and the improved algorithm thereof are common methods for optimizing SVM parameters at present.
The gray wolf optimization algorithm (Grey Wolf Optimizer, GWO) was a group intelligent optimization algorithm proposed by university of griffish in australia mirjalli et al in 2014. The algorithm simulates the gray wolf group level system and predation behavior in the nature, and achieves the purpose of efficient optimization through the processes of group searching, surrounding, pursuing attack hunting objects and the like. Therefore, the method is suitable for SVM parameter optimization. However, the standard GWO algorithm has the characteristics of low post convergence rate, easy sinking into local optimum, low recognition accuracy and the like, so that the method needs to be improved.
Disclosure of Invention
The invention aims to provide a bearing defect identification method based on SDAE and an improved GWO-SVM, which can improve the convergence speed and the identification precision and solve the defect that the bearing is easy to fall into local optimum.
In order to solve the technical problems, the technical scheme of the invention is as follows: a bearing defect identification method based on SDAE and improved GWO-SVM is characterized in that: the method comprises the following steps:
step 1: collecting monitoring data of the bearing under normal conditions and different defect conditions, preprocessing the data, extracting and normalizing the processed data, and randomly dividing each type of preprocessed features into a training sample and a testing sample according to a certain proportion;
step 2: establishing a stacked denoising self-coding network SDAE with the network layer number of 4, wherein the stacked denoising self-coding network SDAE is used for extracting characteristics of training data and test data and training an initial stacked denoising self-coding network SDAE;
step 3: establishing an improved GWO-SVM classifier model, and taking the deepest data features extracted from the SDAE of the coding network through initial stacking denoising as GWO-SVM classifier input to train the classifier;
step 4: the stacked denoising self-coding network SDAE parameters are finely adjusted by using a back propagation BP algorithm, a gradient descent algorithm is applied to update weights, and the GWO-SVM classifier is retrained and improved until classification accuracy is met;
step 5: and obtaining a data extraction and defect identification integral model based on the SDAE and the improved GWO-SVM classifier according to the steps, and realizing deep feature extraction and defect identification of the bearing by using the model.
Further, the data in the step 1 is subjected to feature extraction, and the feature extraction comprises 13 time domain features, 4 frequency domain features and 5 time-frequency domain features which are extracted by empirical mode decomposition; the time domain features comprise 7 dimensionless time domain features of skewness, mean value, variance, peak value, minimum value, peak-to-peak value and mean square value, and 6 dimensionless time domain features of kurtosis, peak value factor, pulse factor, waveform factor, margin factor and skewness factor; the frequency domain features comprise mean square frequency, center of gravity frequency, frequency variance and standard frequency variance; the time-frequency domain features include the first 4 IMF energy indices and the total energy index.
Further, the structure of the 4-layer stacked denoising self-coding network SDAE in the step 2 from bottom to top is [44-22-11-5 ]; the training process of the stacked denoising self-coding network SDAE adopts layer-by-layer stacking learning, and after the unsupervised training of each denoising self-coder DAE is completed, the output of an implicit layer is used as the input of the next denoising self-coder DAE, namely the training of the first denoising self-coder DAE1 is completed; taking the data characteristic 1 of the hidden layer as the input of a second denoising self-encoder DAE2, and performing unsupervised training on the DAE2 to obtain the data characteristic 2; taking the data characteristic 2 as the input of a third denoising self-encoder DAE3, and performing unsupervised training on the DAE3 to obtain the data characteristic 3; taking the data characteristic 3 as the input of a third denoising self-encoder DAE4, and performing unsupervised training on the DAE4 to obtain the data characteristic 4; the feature data 4 would be input as a GWO-SVM classifier.
Further, in the stacked denoising self-coding network SDAE, an improved GWO-SVM classifier is added after the last feature representation layer of the SDAE, and the whole trained network can simultaneously realize the feature extraction and defect identification tasks of data.
Further, in the step 3, the improved GWO-SVM classifier model optimizes the SVM penalty factor C and the kernel function parameter σ by using an improved wolf algorithm GWO, and specifically comprises the following steps:
step 1: initializing parameters improving GWO, e.g. population size N, maximum number of iterations t max Initial value a of distance control parameter a initial And a termination value a final The method comprises the steps of carrying out a first treatment on the surface of the Initializing a, A and C; initializing the population to obtain N initial wolf positions X= [ X ] 1 ,X 2 ,…,X N ]Wherein X is i =(C i ,σ i );
Step 2: calculating fitness value of each individual of the population, sequencing the fitness values, and respectively recording that three individuals with optimal fitness are alpha, beta and delta, and the corresponding positions are X α ,X β ,X δ ;
Step 3: updating the positions of other wolf individuals in the population;
step 4: updating the values of the convergence factor a and the parameter A, C;
step 5: calculating individual fitness f (X) i (t)) and sorting the sirius population in descending order, differentially mutating the last 10% of the sirius individuals, and calculating the fitness f (X) of the individuals after mutation i ' (t)) and comparing the fitness of individuals before and after mutation, if f (X) i '(t))≥f(X i (t)) replacing the individual before mutation with the individual after mutation, iff(X i '(t))<f(X i (t)) retaining the pre-variant individual;
step 6: judging whether the maximum iteration times t are reached max If t is not reached max T=t+1 and returns to step 2; if so, stopping iteration and returning to the optimal individual;
step 7: and obtaining an optimal punishment factor C and an optimal kernel function parameter sigma according to the optimal gray wolf position, and performing defect identification by utilizing an improved GWO-SVM classifier.
Further, the population initialization formula in the step 1 is as follows:
wherein the method comprises the steps ofAnd x is E [0,1 ]]。
Further, the population position update in the step 3 satisfies the following conditions:
wherein A and C are coefficient vectors, X i (t) is the current individual location, X i (t+1) is the individual position after iterative update.
Further, the update of a in the step 4 satisfies:
t is the current iteration number, random variable λ=0.01;
the coefficient vector a, C update satisfies:
A=2a·r 1 -a
C=2·r 2
wherein r is 1 And r 2 Is [0,1]Random numbers within.
Further, the differential variation formula in the step 5 is:
X'(t)=r[X α -X(t)]-r[X s (t)-X(t)],X s (t) wherein r is [0,1 ] for a randomly selected individual one of the wealth of the population]Is a random number of (a) in the memory.
Further, the fine tuning in the step 4 specifically includes the following steps:
step one: inputting training samples into an initial stacking denoising self-coding network SDAE which is pre-trained, extracting top-level features, and taking the top-level features as training samples for supporting an improved GWO-SVM classifier to obtain an SVM optimal penalty factor C and an optimal kernel function parameter sigma;
step two: fine tuning network parameters by using a back propagation BP algorithm, and updating weights by using a gradient descent algorithm;
step three: inputting training samples into a stack denoising self-coding network SDAE with fine adjustment completed, extracting top-level characteristics, retraining and improving a GWO-SVM classifier to obtain an optimal penalty factor C and an optimal kernel function parameter sigma of the SVM classifier;
step four: inputting the top-level features into the trained optimal SVM classifier, judging whether a termination condition is met, if so, finishing fine tuning, stopping iteration, finishing SDAE network and improved GWO-SVM classifier training, and if not, jumping to the second step.
The invention has the advantages that: according to the bearing defect identification method based on the SDAE and the improved GWO-SVM, the stacked denoising self-coding network SDAE is adopted to extract the characteristics of the data, so that the self-adaptive mining of high-dimensional deep fault characteristics is realized, the problems that the types of the faults of the bearing are multiple, the fault characteristics are difficult to extract and the like are solved, the SDAE not only learns the characteristics of the original data, but also learns the degraded characteristics after being damaged, so that the method has stronger generalization and robustness; the data characteristic extraction and defect recognition process are combined, the connection of the data characteristic extraction and defect recognition process is established, and the defects that a large amount of priori knowledge, abundant signal processing theory and practical experience are needed as supports in the characteristic extraction process are overcome; the GWO algorithm is improved, the convergence speed and the recognition accuracy of the algorithm are improved, the defect that the algorithm is easy to fall into local optimum is overcome, the SVM parameter is optimized by utilizing the improvement GWO, and the defect recognition accuracy is improved.
Drawings
The invention will be described in further detail with reference to the drawings and the detailed description.
FIG. 1 is a flow chart of a method for identifying bearing defects based on SDAE and modified GWO-SVM according to the present invention.
Fig. 2 is a schematic diagram of the operation of the first denoising self-encoder according to the present invention.
Fig. 3 is a block diagram of a stacked denoising self-encoder SDAE according to the present invention.
Fig. 4 is a flow chart of the stacked denoising self-encoder SDAE tuning according to the present invention.
FIG. 5 is a flowchart of a modified GWO-SVM according to the present invention.
Detailed Description
The following examples will provide those skilled in the art with a more complete understanding of the present invention and are not intended to limit the invention to the embodiments described.
Examples
The bearing defect identification method based on SDAE and improved GWO-SVM in the embodiment mainly comprises the following steps as shown in FIG. 1:
step 1: collecting monitoring data of the bearing under normal conditions and different defect conditions, preprocessing the data, and extracting features of the processed data, wherein the feature extraction comprises 13 time domain features, 4 frequency domain features and 5 time-frequency domain features extracted by Empirical Mode Decomposition (EMD); wherein the time domain features comprise 7 dimensional time domain features of skewness, mean, variance, peak value, minimum value, peak-to-peak value and mean square value (see table 1); kurtosis, peak factor, pulse factor, waveform factor, margin factor, skew factor 6 dimensionless time domain features (see table 2); the frequency domain features include mean square frequency, center of gravity frequency, frequency variance, standard frequency variance (see table 3); the time-frequency domain features include the first 4 IMF energy indices and the total energy index.
Table 1 dimensional time domain characteristic parameters
Table 2 dimensionless time domain characterization parameters
TABLE 3 frequency domain characterization parameters
And carrying out normalization processing on the data, setting labels for normal conditions and different defect conditions of the equipment, wherein each label corresponds to a respective defect state, and randomly dividing each type of preprocessed characteristics into a training sample and a test sample according to a certain proportion.
Step 2: and establishing a stacked denoising self-coding network SDAE with the network layer number of 4, wherein the stacked denoising self-coding network SDAE is used for extracting characteristics of training data and test data and training an initial stacked denoising self-coding network SDAE.
Step 3: an improved GWO-SVM classifier is built and the deepest data features extracted from the coding network SDAE through initial stacking denoising are used as GWO-SVM classifier inputs to train the classifier.
Step 4: and fine tuning network parameters by using a back propagation BP algorithm, updating weights by using a gradient descent algorithm, and retraining the improved GWO-SVM classifier until classification accuracy is met.
Step 5: and obtaining a data extraction and defect identification integral model based on the SDAE and the improved GWO-SVM classifier according to the steps, and realizing deep feature extraction and defect identification of the bearing by using the model.
Referring to fig. 2 and 3, the first denoising self-encoder is composed of an encoder, a decoder and an hidden layer; the denoising self-encoder maps x' to q D (x '|x) adding a random noise to the original input x, and adding noise q to the data set D to obtain a damaged sample x', wherein D is the data set, and q is the random noise; at this time, the encoder input is a corrupted sample, which is encoded by the encoding function f θ Mapping to the hidden layer y can be expressed as:
y 1 =f θ1 (x')=s(W 1 x'+b 1 ) (1)
where s is a mapping function, W 1 Is the coding weight of the first denoising self-encoder, b 1 Is the coding bias of the first denoising self-encoder, and theta 1 is the pre-trained coding parameter of the first denoising self-encoder; output y of first denoising self-encoder hidden layer 1 Reconstructing through a decoding function:
z 1 =g θ1 '(y 1 )=S'(W 1 'y+b 1 ') (2)
wherein S' is a mapping function, W 1 ' is the decoding weight of the first denoising self-encoder, b 1 'is the decoding bias of the first denoising auto-encoder, θ1' is the decoding parameter of the first denoising auto-encoder. For each input x i Mapping to a y 1 i Then get the reconstruction function z 1 i All parameters are optimized continuously, and the error of minimizing input and reconstruction decoding is obtained:
L(x i ,z 1 i )=||x i -z 1 i || 2 (3)
the working principles of the second denoising self-encoder, the third denoising self-encoder and the fourth denoising self-encoder are the same as those of the first denoising self-encoder.
In the embodiment, the structure of the 4-layer stacked denoising self-coding network SDAE in the step 2 from bottom to top is [44-22-11-5 ]; the training process of the stacked denoising self-coding network SDAE adopts layer-by-layer stacking learning, and after the unsupervised training of each denoising self-coder DAE is completed, the output of an implicit layer is used as the input of the next denoising self-coder DAE, namely the training of the first denoising self-coder DAE1 is completed; taking the data characteristic 1 of the hidden layer as the input of a second denoising self-encoder DAE2, and performing unsupervised training on the DAE2 to obtain the data characteristic 2; taking the data characteristic 2 as the input of a third denoising self-encoder DAE3, and performing unsupervised training on the DAE3 to obtain the data characteristic 3; taking the data characteristic 3 as the input of a third denoising self-encoder DAE4, and performing unsupervised training on the DAE4 to obtain the data characteristic 4; the data feature 4 would be input as a GWO-SVM classifier.
Referring to fig. 4, the stacked denoising self-coding network SDAE tuning process includes the following steps:
step one: inputting training samples into an initial stacking denoising self-coding network SDAE which is pre-trained, extracting top-level features, and taking the top-level features as training samples for supporting an improved GWO-SVM classifier to obtain an SVM optimal penalty factor C and an optimal kernel function parameter sigma;
step two: fine tuning network parameters by using a back propagation BP algorithm, and updating weights by using a gradient descent algorithm;
step three: inputting training samples into a stack denoising self-coding network SDAE with fine adjustment completed, extracting top-level characteristics, retraining and improving a GWO-SVM classifier to obtain an optimal penalty factor C and an optimal kernel function parameter sigma of the SVM classifier;
step four: inputting the top-level features into the trained optimal SVM classifier, judging whether a termination condition is met, if so, finishing fine tuning, stopping iteration, finishing SDAE network and improved GWO-SVM classifier training, and if not, jumping to the second step.
Referring to fig. 5, the improved GWO-SVM classifier utilizes an improved gray wolf algorithm GWO to optimize an SVM penalty factor C and a kernel function parameter σ, and specifically comprises the following steps:
step 1: initializing parameters of improvement GWO, the gray wolf population scale N and the maximum iteration number t max Initial value a of distance control parameter a initial And a termination value a final The method comprises the steps of carrying out a first treatment on the surface of the Initializing parameters a, A and C; let i (i=1,) N]) The position of the Pechamus at the t-th iteration is X i (t) the gray wolf group can be divided into a first-neck wolf alpha, a second-neck wolf beta, a captain wolf delta and other individual wolves omega according to a hierarchical mechanism from high to low; all the individual wolves were initialized using the following formula:
wherein the method comprises the steps ofAnd x is E [0,1 ]]K is the maximum chaotic iteration step number, and N initial gray wolf positions X= [ X ] are obtained through initialization 1 ,X 2 ,…,X N ]Wherein X is i =(C i ,σ i ) The position of the gray wolf is the SVM penalty factor C and the kernel function parameter sigma which need to be optimized i 。
Step 2: calculating fitness value f (X i (t)) and sorting the individuals, and respectively recording that three individuals with optimal fitness are alpha, beta and delta, and the corresponding positions are X α ,X β ,X δ The corresponding fitness value is f (X α (t)),f(X β (t)),f(X δ (t))。
Step 3: updating the positions of other wolf individuals in the population and the position update satisfies:
where a and C are coefficient vectors.
Step 4: updating the values of the convergence factor a and the parameter A, C, the updating satisfying:
t is the current iteration number, random variable λ=0.01.
A=2a·r 1 -a (10)
C=2·r 2 (11)
Wherein r is 1 And r 2 Is [0,1]Random numbers within.
Step 5: calculating individual fitness f (X) i (t)) and ordering the wolf population in descending order, differential variation of the last 10% of wolf individuals:
X i '(t)=r[X α -X i (t)]-r[X s (t)-X i (t)] (12)
wherein X is s (t) is a single gray wolf selected randomly from the population, r is [0,1 ]]Is a random number of (a) in the memory. Calculating fitness f (X) of individuals after mutation i ' (t)) and comparing the fitness of individuals before and after mutation, if f (X) i '(t))≥f(X i (t)) replacing the individual before mutation with the individual after mutation, if f (X) i '(t))<f(X i (t)) and preserving the pre-variant individuals.
Step 6: judging whether the maximum iteration times t are reached max If t is not reached max T=t+1 and returns to step 2;if so, stopping iteration and returning to the optimal individual;
step 7: and obtaining an optimal punishment factor C and an optimal kernel function parameter sigma according to the optimal gray wolf position, and performing defect identification by utilizing an improved GWO-SVM classifier.
The foregoing has shown and described the basic principles and main features of the present invention and the advantages of the present invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, and that the above embodiments and descriptions are merely illustrative of the principles of the present invention, and various changes and modifications may be made without departing from the spirit and scope of the invention, which is defined in the appended claims. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (8)
1. A bearing defect identification method based on SDAE and an improved GWO-SVM is characterized in that: the method comprises the following steps:
step 1: collecting monitoring data of the bearing under normal conditions and different defect conditions, preprocessing the data, extracting and normalizing the processed data, and randomly dividing each type of preprocessed features into a training sample and a testing sample according to a certain proportion;
step 2: establishing a stacked denoising self-coding network SDAE with the network layer number of 4, wherein the stacked denoising self-coding network SDAE is used for extracting characteristics of training data and test data and training an initial stacked denoising self-coding network SDAE;
step 3: establishing an improved GWO-SVM classifier model, and taking the deepest data features extracted from the SDAE of the coding network through initial stacking denoising as GWO-SVM classifier input to train the classifier;
step 4: the stacked denoising self-coding network SDAE parameters are finely adjusted by using a back propagation BP algorithm, a gradient descent algorithm is applied to update weights, and the GWO-SVM classifier is retrained and improved until classification accuracy is met;
step 5: obtaining a data extraction and defect identification integral model based on the SDAE and the improved GWO-SVM classifier according to the steps, and realizing deep feature extraction and defect identification of the bearing by using the model;
in the step 3, the improved GWO-SVM classifier model optimizes the SVM penalty factor C and the kernel function parameters by using an improved gray wolf algorithm GWO, and the specific steps are as follows:
step 1: initializing parameters of the improvement GWO, such as population size N, maximum iteration number, initial value and termination value of the distance control parameter; initializing a, A and C; initializing a population to obtain N initial wolf positions, wherein;
step 2: calculating fitness values of each individual of the population, sequencing the fitness values, and respectively recording three individuals with optimal fitness as corresponding positions;
step 3: updating the positions of other wolf individuals in the population;
step 4: updating the convergence factor and the value of the parameter A, C;
step 5: calculating individual fitness, sorting the gray wolf population in a descending order, carrying out differential mutation on the gray wolf individuals arranged at the rear 10%, calculating the fitness of the individuals after mutation, comparing the individual fitness before and after mutation, if the individuals before mutation are replaced by the gray wolf individuals after mutation, if the individuals before mutation are reserved;
step 6: judging whether the maximum iteration times are reached, if not, t=t+1, returning to the step 2, and if so, stopping iteration, and returning to the optimal individual;
step 7: obtaining an optimal punishment factor C and an optimal kernel function parameter according to the optimal gray wolf position, and performing defect identification by utilizing an improved GWO-SVM classifier;
the fine tuning in the step 4 comprises the following specific steps:
step one: inputting training samples into an initial stacking denoising self-coding network SDAE which is pre-trained, extracting top-level features as training samples of a support improvement GWO-SVM classifier, and obtaining an SVM optimal penalty factor C and optimal kernel function parameters;
step two: fine tuning network parameters by using a back propagation BP algorithm, and updating weights by using a gradient descent algorithm;
step three: inputting training samples into a stack denoising self-coding network SDAE with fine adjustment completed, extracting top-level characteristics, retraining and improving a GWO-SVM classifier to obtain an optimal penalty factor C and optimal kernel function parameters of the SVM classifier;
step four: inputting the top-level features into the trained optimal SVM classifier, judging whether a termination condition is met, if so, finishing fine tuning, stopping iteration, finishing SDAE network and improved GWO-SVM classifier training, and if not, jumping to the second step.
2. The method for identifying bearing defects based on SDAE and modified GWO-SVM according to claim 1, characterized in that: the data in the step 1 is subjected to feature extraction, and the feature extraction comprises 13 time domain features, 4 frequency domain features and 5 time-frequency domain features which are extracted by empirical mode decomposition; the time domain features comprise 7 dimensionless time domain features of skewness, mean value, variance, peak value, minimum value, peak-to-peak value and mean square value, and 6 dimensionless time domain features of kurtosis, peak value factor, pulse factor, waveform factor, margin factor and skewness factor; the frequency domain features comprise mean square frequency, center of gravity frequency, frequency variance and standard frequency variance; the time-frequency domain features include the first 4 IMF energy indices and the total energy index.
3. The method for identifying bearing defects based on SDAE and modified GWO-SVM according to claim 1, characterized in that: the structure of the 4-layer stacked denoising self-coding network SDAE in the step 2 from bottom to top is [44-22-11-5 ]; the training process of the stacked denoising self-coding network SDAE adopts layer-by-layer stacking learning, and after the unsupervised training of each denoising self-coder DAE is completed, the output of an implicit layer is used as the input of the next denoising self-coder DAE, namely the training of the first denoising self-coder DAE1 is completed; taking the data characteristic 1 of the hidden layer as the input of a second denoising self-encoder DAE2, and performing unsupervised training on the DAE2 to obtain the data characteristic 2; taking the data characteristic 2 as the input of a third denoising self-encoder DAE3, and performing unsupervised training on the DAE3 to obtain the data characteristic 3; taking the data characteristic 3 as the input of a third denoising self-encoder DAE4, and performing unsupervised training on the DAE4 to obtain the data characteristic 4; the feature data 4 would be input as a GWO-SVM classifier.
4. The method for identifying bearing defects based on SDAE and modified GWO-SVM according to claim 1, characterized in that: in the stacked denoising self-coding network SDAE, an improved GWO-SVM classifier is added after the last feature representation layer of the SDAE, and the whole trained network can simultaneously realize the feature extraction and defect identification tasks of data.
5. The method for identifying bearing defects based on SDAE and modified GWO-SVM according to claim 1, characterized in that: the population initialization formula in the step 1 is as follows:
wherein the method comprises the steps ofAnd x is E [0,1 ]]。
6. The method for identifying bearing defects based on SDAE and modified GWO-SVM according to claim 1, characterized in that: the updating of the population position in the step 3 meets the following conditions:
wherein A and C are coefficient vectors, X i (t) is the current individual location, X i (t+1) is the individual position after iterative update.
7. The method for identifying bearing defects based on SDAE and modified GWO-SVM according to claim 1, characterized in that: the update of a in the step 4 satisfies the following conditions:
t is the current iteration number, random variable λ=0.01;
the coefficient vector a, C update satisfies:
A=2a·r 1 -a
C=2·r 2
wherein r is 1 And r 2 Is [0,1]Random numbers within.
8. The method for identifying bearing defects based on SDAE and modified GWO-SVM according to claim 1, characterized in that: the differential variation formula in the step 5 is as follows: x' (t) =r [ X ] α -X(t)]-r[X s (t)-X(t)],X s (t) wherein r is [0,1 ] for a randomly selected individual one of the wealth of the population]Is a random number of (a) in the memory.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911381284.2A CN111476263B (en) | 2019-12-27 | 2019-12-27 | Bearing defect identification method based on SDAE and improved GWO-SVM |
PCT/CN2020/071757 WO2021128510A1 (en) | 2019-12-27 | 2020-01-13 | Bearing defect identification method based on sdae and improved gwo-svm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911381284.2A CN111476263B (en) | 2019-12-27 | 2019-12-27 | Bearing defect identification method based on SDAE and improved GWO-SVM |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111476263A CN111476263A (en) | 2020-07-31 |
CN111476263B true CN111476263B (en) | 2024-02-09 |
Family
ID=71746262
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911381284.2A Active CN111476263B (en) | 2019-12-27 | 2019-12-27 | Bearing defect identification method based on SDAE and improved GWO-SVM |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN111476263B (en) |
WO (1) | WO2021128510A1 (en) |
Families Citing this family (29)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP6990634B2 (en) * | 2018-08-21 | 2022-02-03 | 株式会社日立ハイテク | State prediction device and semiconductor manufacturing device |
CN111917785B (en) * | 2020-08-06 | 2022-07-15 | 重庆邮电大学 | Industrial internet security situation prediction method based on DE-GWO-SVR |
CN112101445B (en) * | 2020-09-09 | 2023-11-28 | 浙江大学 | Continuous casting billet subcutaneous slag inclusion defect real-time forecasting method based on supervision neural network |
CN112329173B (en) * | 2020-11-10 | 2022-09-20 | 重庆大学 | Parameter optimization method for rotor-bearing bending-torsion coupling system |
CN112836604A (en) * | 2021-01-22 | 2021-05-25 | 合肥工业大学 | Rolling bearing fault diagnosis and classification method, system and equipment based on VMD-SSAE and storage medium thereof |
CN113593657B (en) * | 2021-07-26 | 2024-01-09 | 燕山大学 | Cement free calcium soft measurement system with quality target as guiding semi-supervised learning |
CN114118130B (en) * | 2021-09-28 | 2024-04-05 | 西安交通大学 | Transformer partial discharge mode identification method, system, medium and equipment |
CN113987936B (en) * | 2021-10-27 | 2022-04-08 | 中国人民解放军32801部队 | Equipment test resource overall allocation method based on chaotic genetic algorithm |
CN113987949B (en) * | 2021-11-03 | 2024-05-07 | 燕山大学 | Data-driven-based plate strip steel deformation resistance prediction method |
CN114112390B (en) * | 2021-11-23 | 2024-03-22 | 哈尔滨工程大学 | Nonlinear complex system early fault diagnosis method |
CN114040358B (en) * | 2021-12-06 | 2024-04-23 | 大连大学 | High-stability clustering method for unmanned aerial vehicle cluster network |
CN114354188A (en) * | 2022-01-10 | 2022-04-15 | 郑州大学 | Rotary machine fault diagnosis method based on fully-adaptive noise set empirical mode decomposition |
CN114584924A (en) * | 2022-02-28 | 2022-06-03 | 长沙融创智胜电子科技有限公司 | Intelligent unattended sensor system and target identification method |
CN115099885A (en) * | 2022-03-31 | 2022-09-23 | 日日顺供应链科技股份有限公司 | Commodity matching recommendation method and system |
CN114897004B (en) * | 2022-04-15 | 2023-05-02 | 成都理工大学 | Trapezoidal accumulation nuclear pulse identification method based on deep learning transducer model |
CN114781450B (en) * | 2022-04-24 | 2024-03-29 | 华东交通大学 | Train rolling bearing state identification method based on parameter optimization MOMEDA-MIA-CNN |
CN114781577A (en) * | 2022-05-06 | 2022-07-22 | 安徽理工大学 | Buck circuit fault diagnosis method based on VMD-DCNN-SVM |
CN114764579A (en) * | 2022-05-12 | 2022-07-19 | 西北工业大学 | Denoising feature extraction method for ship radiation noise signal |
CN114897023B (en) * | 2022-05-19 | 2024-03-29 | 西安交通大学 | Underwater sound target identification method based on extraction of sensitive difference features of underwater sound targets |
CN115082780B (en) * | 2022-05-25 | 2024-03-19 | 西北工业大学 | Multi-source heterogeneous image change detection method based on incremental difference learning network |
CN115685881B (en) * | 2022-11-07 | 2023-06-02 | 北京科技大学 | Low-stress high-precision arc additive process control method based on computational intelligence |
CN115876476B (en) * | 2023-02-16 | 2023-05-16 | 山东科技大学 | Rolling bearing fault diagnosis method, system, computer device and storage medium |
CN116499748B (en) * | 2023-06-27 | 2023-08-29 | 昆明理工大学 | Bearing fault diagnosis method and system based on improved SMOTE and classifier |
CN116563311B (en) * | 2023-07-04 | 2023-09-08 | 新乡市食品药品检验所 | Meat product detection method and device based on image segmentation |
CN116738868B (en) * | 2023-08-16 | 2023-11-21 | 青岛中德智能技术研究院 | Rolling bearing residual life prediction method |
CN117129565A (en) * | 2023-08-23 | 2023-11-28 | 广西大学 | Concrete filled steel tube void knocking force detection method based on energy ratio and GWO-SVM |
CN117094064B (en) * | 2023-10-19 | 2024-03-01 | 西南交通大学 | Method, device, equipment and storage medium for calculating layout parameters of components |
CN117635245A (en) * | 2023-11-30 | 2024-03-01 | 广东电力交易中心有限责任公司 | Power price prediction method and system based on multilevel frequency domain decomposition and IBM H optimization DELM |
CN117725394B (en) * | 2024-02-18 | 2024-05-07 | 浙江浙能技术研究院有限公司 | Wind power plant broadband oscillation identification method based on hierarchical embedded modal decomposition |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1050443A (en) * | 1989-09-22 | 1991-04-03 | 中国新型建筑材料工业杭州设计研究院 | In-line automatic tester of adfects on float glass |
CN102983890A (en) * | 2012-11-14 | 2013-03-20 | 北京小米科技有限责任公司 | Method and device of pairing connection of equipment |
CN106682688A (en) * | 2016-12-16 | 2017-05-17 | 华南理工大学 | Pile-up noise reduction own coding network bearing fault diagnosis method based on particle swarm optimization |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107957551A (en) * | 2017-12-12 | 2018-04-24 | 南京信息工程大学 | Stacking noise reduction own coding Method of Motor Fault Diagnosis based on vibration and current signal |
CN108344574B (en) * | 2018-04-28 | 2019-09-10 | 湖南科技大学 | A kind of Wind turbines Method for Bearing Fault Diagnosis based on depth joint adaptation network |
-
2019
- 2019-12-27 CN CN201911381284.2A patent/CN111476263B/en active Active
-
2020
- 2020-01-13 WO PCT/CN2020/071757 patent/WO2021128510A1/en active Application Filing
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1050443A (en) * | 1989-09-22 | 1991-04-03 | 中国新型建筑材料工业杭州设计研究院 | In-line automatic tester of adfects on float glass |
CN102983890A (en) * | 2012-11-14 | 2013-03-20 | 北京小米科技有限责任公司 | Method and device of pairing connection of equipment |
CN106682688A (en) * | 2016-12-16 | 2017-05-17 | 华南理工大学 | Pile-up noise reduction own coding network bearing fault diagnosis method based on particle swarm optimization |
Also Published As
Publication number | Publication date |
---|---|
CN111476263A (en) | 2020-07-31 |
WO2021128510A1 (en) | 2021-07-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111476263B (en) | Bearing defect identification method based on SDAE and improved GWO-SVM | |
CN109102005B (en) | Small sample deep learning method based on shallow model knowledge migration | |
CN111680614B (en) | Abnormal behavior detection method based on video monitoring | |
CN110751108B (en) | Subway distributed vibration signal similarity determination method | |
CN106124212A (en) | Based on sparse coding device and the Fault Diagnosis of Roller Bearings of support vector machine | |
Zhang et al. | Deep sparse autoencoder for feature extraction and diagnosis of locomotive adhesion status | |
CN112257341B (en) | Customized product performance prediction method based on heterogeneous data difference compensation fusion | |
CN106529574B (en) | Image classification method based on sparse autocoder and support vector machines | |
CN111832647A (en) | Abnormal flow detection system and method | |
CN109147817B (en) | Denoising frequency characteristic extraction method based on variation-limited Boltzmann machine | |
CN111222638B (en) | Neural network-based network anomaly detection method and device | |
CN111353373A (en) | Correlation alignment domain adaptive fault diagnosis method | |
CN111523603A (en) | Ship power equipment fault identification method | |
CN110009030A (en) | Sewage treatment method for diagnosing faults based on stacking meta learning strategy | |
CN112650204B (en) | Intelligent track unmanned vehicle fault gene identification method and system | |
CN110991471B (en) | Fault diagnosis method for high-speed train traction system | |
CN110363230A (en) | Stacking integrated sewage handling failure diagnostic method based on weighting base classifier | |
CN107153837A (en) | Depth combination K means and PSO clustering method | |
CN115587290A (en) | Aero-engine fault diagnosis method based on variational self-coding generation countermeasure network | |
CN113159139B (en) | Damage state diagnosis method based on improved acoustic emission density clustering | |
CN114880538A (en) | Attribute graph community detection method based on self-supervision | |
CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
CN111178897B (en) | Cost-sensitive dynamic clustering method for fast feature learning on unbalanced data | |
CN113869451A (en) | Rolling bearing fault diagnosis method under variable working conditions based on improved JGSA algorithm | |
CN114037001A (en) | Mechanical pump small sample fault diagnosis method based on WGAN-GP-C and metric learning |
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 |