CN115856623A - Motor fault diagnosis method based on uniformity and kurtosis calculation - Google Patents
Motor fault diagnosis method based on uniformity and kurtosis calculation Download PDFInfo
- Publication number
- CN115856623A CN115856623A CN202211597697.6A CN202211597697A CN115856623A CN 115856623 A CN115856623 A CN 115856623A CN 202211597697 A CN202211597697 A CN 202211597697A CN 115856623 A CN115856623 A CN 115856623A
- Authority
- CN
- China
- Prior art keywords
- motor
- fault
- uniformity
- kurtosis
- algorithm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a motor fault diagnosis method based on uniformity and kurtosis calculation. The method comprises the following steps: through the analysis of the uniformity and the kurtosis of the transient starting current signal of the motor, parameters which are simpler and can effectively represent the motor fault are obtained. The method comprises the steps of predicting errors by initializing weights, selecting father weights according to the predicted errors, completing processing of two algorithms for the selected father weights, wherein one algorithm is an adaptive genetic algorithm, the other algorithm is a cuckoo search algorithm, and finally combining results obtained by the two algorithms to obtain optimized neural network weights and an optimal solution, so that a motor fault diagnosis model based on uniformity, kurtosis and AGA-CS is constructed, diagnosis of five faults of an induction motor is realized, and a good basis is provided for motor fault diagnosis. Compared with the prior art, the method has lower calculation complexity, shorter processing time and is suitable for online detection.
Description
Technical Field
The invention relates to the technical field of motor health state diagnosis, in particular to a motor fault diagnosis method based on uniformity and kurtosis calculation.
Background
The motor is the foundation of the power industry and the manufacturing industry, while the induction motor is always an important part of various electrical equipment and manufacturing processes, and has the main characteristics of low cost, good rigidity and quality, but the economic and technical consequences caused by the accidental shutdown of the induction motor due to the failure of the induction motor are serious. Designing a reliable early motor fault diagnosis method is a low cost operation because early detection of motor faults can prevent the machine from catastrophic damage.
Research into induction motor fault detection has attracted considerable interest over the past few years. A great deal of research has been done by many scholars to prevent catastrophic failure of the motor, but many of the existing methods are based on analysis of the motor's transient start-up current, which has the disadvantage of high mathematical complexity and high computational cost in its development. In addition, most of the existing fault detection methods are also only dedicated to detecting a single specific fault, and are generally based on monitoring and analyzing current and vibration signals, these detection methods and techniques are generally applicable to diagnosing faults of independent induction motors, and in the implementation process, most of the fault diagnosis methods also rely on complex mathematical algorithms, for example, some frequency domain processing methods need to bring time domain signals into the frequency domain, and need to return to the time domain after processing, which requires long execution time and calculation resources for signal processing, thereby increasing the difficulty in implementation.
Disclosure of Invention
The purpose of the invention is as follows: aiming at the problems of high complexity and large calculation cost of the traditional fault detection method and the current situation that a common method excessively concentrates on detecting a single specific fault, the invention provides a motor fault diagnosis method based on uniformity and kurtosis calculation, which has lower calculation complexity and shorter processing time and is suitable for online detection.
The technical scheme is as follows: a motor fault diagnosis method based on uniformity and kurtosis calculation comprises the following steps:
step 1: calculating the uniformity and the kurtosis of the transient starting current signal of the motor to obtain a fault characteristic parameter of the motor;
step 2: constructing a hybrid classifier model based on AGA-CS, training the hybrid classifier model by using uniformity and kurtosis characteristic parameters, predicting errors through initialized weights, selecting a father weight according to the predicted errors, finishing the processing of two algorithms for the selected father weight, wherein one algorithm is an adaptive genetic algorithm AGA, the other algorithm is a cuckoo search algorithm CS, and finally combining the results of the two methods to obtain an optimized neural network weight and an optimal solution, thereby constructing a motor fault diagnosis model based on uniformity, kurtosis calculation and AGA-CS;
and step 3: and realizing the fault diagnosis of the induction motor based on the uniformity, the kurtosis calculation and the AGA-CS motor fault diagnosis model.
Further, the motor fault characteristic parameters in the step 1 include the following:
setting four fault types, namely a motor fault 1BRB with one broken rotor bar, a motor fault 2BRB with two broken rotor bars, a motor fault BRB with damaged bearing outer ring and a motor fault UNB with unbalanced mechanical load;
(1) Rotor broken bar fault BRB
Using the relevant frequency parameter f BRB Indicating the presence of a BRB fault, as specified by the formula:
f BRB ＝(1-2kb)·f s ，k＝1，2，3，…
wherein k is an integer, f s Is the main frequency component of the power supply, b represents the motor slip, and the value range is 0 to 1;
(2) Bearing fault BRN
Frequency f associated with outer race 0 Determined by the following equation:
where n is the number of rolling elements, f r Is the rotational frequency of the shaft, BD is the ball diameter, PD is the bearing race diameter, and β is the angle between the rolling elements in the race;
(3) Mechanical unbalance fault UNB
The rotor is unbalanced, generating an unbalanced force u, which is given by:
u＝n×r
where m is the mass and r is the eccentricity.
When the imbalance forces fluctuate with the rotational speed and drag the rotor from the center of the stator to other locations, the mutual inductance between the stator and rotor circuits becomes non-uniform due to rotor imbalance, resulting in the induction of a frequency component f in the stator current unb Specifically, the formula is shown as follows:
f unb ＝f s [1±k(1-b)/p]，k＝1，2，3，…
wherein, f s Is the fundamental frequency of the current source, k is an integer, b is the induction motor slip, and p is the electrode logarithm.
Further, the uniformity H and the kurtosis K of the current signal of the motor extracted in the step 1 are extracted urt The characteristic parameters are as follows:
calculating uniformity and kurtosis parameters under each motor state, and normalizing the uniformity and the kurtosis parameters; performing statistical analysis on each state to obtain the average value mu and the standard deviation sigma of the uniformity parameter and the kurtosis parameter of current signals of a healthy motor, a motor with one rotor broken bar fault 1BRB, a motor with two rotor broken bar faults 2BRB, a motor with a bearing outer ring damage fault BRB and a motor with an unbalanced mechanical load fault UNB; the calculation steps are as follows:
(1) Calculating uniformity characteristic parameters
The uniformity parameter H is described by the following equation:
where p (i, j) is the element with position number (i, j) in the normalized GLCM. The uniformity parameter can be used as an index for motor fault detection and classification because in the fault current signal of the induction motor there are different frequency signatures associated with the fault and thus will change its uniformity, so the uniformity parameter will also change with it;
(2) Calculating kurtosis characteristic parameters
The kurtosis characteristic parameter, which may measure the deviation of the probability distribution, is a fourth order moment describing the shape of the signal probability distribution, and is calculated as follows:
wherein N is the number of samples, x i Is the original signal sample, i =1,2,3, \8230;, N, μ is the random event X = [ X 1 ，x 2 ，x 3 ，…，x N ]The mean value of (a), is the standard deviation.
Further, the specific steps of initializing weights and predicting errors in the step 2, and selecting the parent weights according to the predicted errors are as follows:
initializing the weights of the neural network in a random manner, and then predicting the error through the initialized weights, specifically according to the following formula:
E＝A V -H
where E is the error, H is the activation function, A V Is the actual value;
the activation function H can be obtained by:
wherein M is the number of neurons in the hidden layer, N is the number of neurons in the input layer,
is the jth weighting level between hidden and output, is>
Is the weight between the jth input layer and the jth hidden layer neuron, x i Is the ith input value;
finally, a parent weight is selected, and the parent weight is selected through error prediction based on the optimization purpose of reducing errors, so that the calculation errors are further reduced.
Further, the operation of completing the adaptive genetic algorithm AGA for the selected parent weight in step 2 specifically is:
according to the approaching condition of the individual adaptive value, the group average adaptive value and the maximum adaptive value, the cross rate and the variation rate are linearly adjusted, which specifically comprises the following steps:
(1) Rate of completion crossing m a The improvement of the cross rate is as follows:
in the formula, g b For greater fitness values in two crossed individuals, g max Is the maximum individual fitness value in the population, g avg Is the mean fitness value of all individuals in the population, a 1 、a 2 Is a random number between 0 and 1;
(2) Complete mutation rate m r Improvement and adaptive mutation of (2) and (3) the mutation rate m r The improvement is that:
wherein g is the individual fitness value of the mutation, g max Adaptation to the largest individual in a populationValue of (D) g avg Is the mean fitness value of all individuals in the population, a 4 Is a random number between 0 and 1, a 3 The settings were as follows:
where R is the random value between [0,1] that varies in each iteration, c and μ are the variation parameters; after the mutation process is executed, we will obtain a complete new generation, and then the AGA algorithm obtains a new optimization weight through the new generation, and repeatedly executes (1) and (2), thereby obtaining a set of new optimization weights based on AGA.
Further, the operation of completing the cuckoo algorithm CS for the selected parent weight in step 2 specifically includes:
during execution, mapping the positions of the nests of cuckoos as solutions in an algorithm population space, specifically comprising the following steps:
(1) Initializing, randomly generating N problem solutions, and recording an optimal solution;
let N e ＝{N e1 ，N e1 ，N e1 ，…，N em The algorithm completes the search in a random walking mode, and takes the advantages and disadvantages of the bird nest position as the fitness, and the concrete calculation is as follows:
wherein the content of the first and second substances,
represents the ith nest to be updated in the tth generation>
For the updated next generation nest, ρ is the step scaling factor, <>
The method is characterized in that the method is dot product operation, L (lambda) is a Levy random flight path, represents a hopping path of Levy flight, and has uncertain direction and length; the relationship between the continuous Levis flight jump path and time obeys Levis distribution, and a corresponding probability density function can be obtained by simplifying the distribution function and performing Fourier transform, specifically the following formula:
L(λ)～u＝t -λ ；1＜λ＜3
wherein λ represents a power coefficient;
(2) Step size evaluation, obtaining a flight jump path:
evaluating the obtained parent weight step size, and calculating the step size by using the following expression:
wherein S is Z Is the step size, α is the step size parameter, w t Is the current parent weight that is the current parent weight,
is the best solution so far, r is [0,1] from a standard normal distribution]An inter-random number, S being a step number, the value of S being obtained by simulating a lewy flight jump path using the formula:
wherein S represents a Lave flight jump path, corresponding to L (lambda), beta is a [1,2] parameter, u and v are normally distributed random numbers, and the distribution is shown in the following formula
Standard deviation of corresponding normal distribution is σ u And σ v The specific definition is shown as the following formula:
(3) Updating the solution position by utilizing the Laevir flight to obtain a new candidate solution position, and reserving a better solution by utilizing a greedy selection strategy;
(4) According to the discovery probability P a Discarding the partial solution and generating a new solution by a random walk mode so as to replace the discarded solution;
step length S according to each father weight Z The cuckoo individuals complete the flight process, and the search paths are randomly changed, so that the cuckoo can be switched from one area to another area, and the overall optimizing capacity of the cuckoo is enhanced; after the flying walk of the global exploration is finished, local search is finished by adopting a mode of preferring random walk based on the current solution; comparing the discovery probability P a And a threshold constant
If>
The bird nest is found, namely the current solution is a poor solution, then bird nest parameters are updated, the poor solution is eliminated, and a new candidate solution is generated, so that the best cuckoo individual is left; the bird nest location update formula is described by the following equation:
wherein the content of the first and second substances,
indicates that the bird nest is to be updated and stands>
For the updated bird nest, xi is a random number which satisfies uniform distribution within [0,1], and>
is any two of the t generationA bird nest;
(5) If the termination condition is met, outputting the best solution, otherwise, returning to the step (3) to continue iteration;
(6) Generating a new solution, creating a new optimization scheme, and completing the updating of the parent weight based on the CS algorithm, which is specifically as follows:
w t+1 ＝w t +S z
wherein, w t+1 Is a new weight, S z Is the step size, w t Is the current parent weight.
Further, the specific steps of implementing the fault diagnosis of the induction motor based on the uniformity, the kurtosis calculation and the motor fault diagnosis model of the AGA-CS hybrid classifier in the step 3 are as follows:
obtaining a motor transient starting current signal, then processing and carrying out analog-to-digital conversion on the motor transient starting current signal, obtaining expected characteristics including uniformity parameters and kurtosis parameters from the obtained discrete current signal, and taking the expected characteristics as the input of an AGA-CS (advanced glass-substrate learning-class) based hybrid classifier model; in a hybrid classifier based on AGA-CS, an adaptive genetic algorithm AGA and a cuckoo search algorithm CS are used in a hybrid mode, neural network weights are used as input, initialization weights are evaluated to be fitness, father weights are selected according to fitness values, then the selected father weights are processed by two algorithms, and finally results of the two methods are combined, so that optimized neural network weights and optimal solutions are obtained, fault diagnosis is performed on an induction motor by the aid of the optimized neural network weights and the father weights, and correct classification of five motor states is finally obtained, wherein the five states comprise a healthy motor, a motor with a rotor broken bar fault 1BRB, a motor with two rotor broken bar faults 2BRB, a motor with a bearing outer ring damage fault BRB and a motor with an unbalanced mechanical load fault UNB.
Has the advantages that:
on one hand, the motor fault characteristic parameter extraction based on the uniformity and the kurtosis parameters is completed, and the characteristic extraction is completed only by detecting the uniformity and the kurtosis of the single-phase transient starting current of the motor, so that the calculation complexity of an algorithm is effectively reduced, the processing time is shorter, the method is more suitable for on-line processing, and the method is a low-cost calculation technology; on the other hand, a hybrid classifier model based on AGA-CS is established, the father weight is selected by utilizing the error fitness value of the neural network, the selected father weight is processed through two algorithms, one is an Adaptive Genetic Algorithm (AGA), the other is a cuckoo search algorithm (CS), and finally the results of the two methods are combined, so that the optimized neural network weight and the optimal solution are obtained, and the classification capability of the algorithm is further improved; and finally, constructing a motor fault diagnosis model based on uniformity, kurtosis and AGA-CS, and realizing effective prediction of five faults of the induction motor, wherein the diagnosis objects are a healthy motor, a motor fault (1 BRB) with one rotor broken bar, a motor fault (2 BRB) with two rotor broken bars, a motor fault (BRB) with bearing outer ring damage and a motor fault (UNB) with unbalanced mechanical load.
The invention focuses on the problem of high-frequency motor faults and achieves the purpose of simultaneously detecting a plurality of specific faults, so the adaptability is stronger. Future work will focus on evaluating multi-sensing motor fault detection techniques in different scenarios, considering the combination with other fault conditions and signal inspection techniques to recover more signal features, thereby completing more accurate fault identification and classification.
Drawings
FIG. 1 is a Probability Density Function (PDF) of different motion states of a motor based on kurtosis;
FIG. 2 is a Probability Density Function (PDF) of different motion states of the motor based on uniformity;
FIG. 3 is a conventional neural network ANN structure;
FIG. 4 is an implementation of the AGA-CS hybrid classifier model;
FIG. 5 is an implementation of a motor fault diagnosis model based on uniformity, kurtosis, and AGA-CS;
fig. 6 is a graph of the average effect of different classification algorithms on motor fault diagnosis.
Detailed Description
In order to better explain the present invention and to facilitate understanding, the following detailed description of the technical solution of the present invention is provided. The following examples are illustrative of the present invention, and the present invention is not limited to the following examples.
The invention provides a motor fault diagnosis method based on uniformity and kurtosis calculation. The specific implementation process comprises the following steps:
step 1: and finishing the extraction of the motor fault parameters based on the uniformity and the kurtosis parameters. Based on the analysis of the uniformity and the kurtosis of the transient starting current signal of the motor, the characteristic parameters which are simpler and can effectively represent the motor fault are obtained.
Step (1) a: selecting a fault diagnosis object and type of the induction motor, collecting three-phase current signals, and obtaining a motor fault sample;
five motor states are set, namely a healthy motor, a motor fault 1BRB with one rotor bar, a motor fault 2BRB with two rotor bars, a motor fault BRB with bearing outer ring damage and a motor fault UNb with unbalanced mechanical load. The fault type is determined as follows:
(1) Rotor broken bar fault BRB
BRBs are the most common types of motor failure, typically caused by overload, thermal imbalance, electromagnetic forces, noise, vibration, environmental damage or manufacturing processes, etc., which are difficult to diagnose. When a rotor broken bar fault BRB occurs, the rotor is asymmetric, so that the induction motor generates rotor current of opposite sequence under the condition that the motor works under an abnormal operation condition, and further a unique component is introduced into a stator current frequency spectrum. At this time, the relevant frequency parameter f may be utilized BRB To indicate the existence of the BRB fault, specifically, the formula (1):
f BRB ＝(1-2kb)·f s ，k＝1，2，3，… (1)
wherein k is an integer, f s Is the main frequency component of the power supply, b represents the motor slip, and the value range is 0 to 1.
(2) Bearing fault BRN
When any part of the motor bearing fails, such as the inner race, the outer race or the rolling elements of the bearing, specific components in the vibration or current signal of the induction motor are caused, and the components are related to the power supply frequency and the mechanical system frequencyAre all relevant and the characteristic frequency associated with bearing failure depends on the geometry of the bearing and the machine speed. For example, when the outer ring is defective, the outer ring defect causes a specific pulse whenever the rolling elements come into contact with the defect. Theoretically, the frequency f associated with the outer race 0 Can be determined by the formula (2):
wherein n is the number of balls (rolling elements), f r Is the rotational frequency of the shaft, BD is the ball diameter, PD is the race diameter, and β is the angle between the rolling elements in the race.
(3) Mechanical unbalance fault UNB
When the induction motor has uneven distribution of mechanical load, if the center of mass is positioned outside the rotating shaft of the motor, the motor has unbalance failure. The defects of the manufacturing process are the main reasons for the generation of the unbalance faults of the rotor, and in addition, the heating expansion can also generate internal dislocation or shaft deviation to influence the rotor, thereby generating the unbalance faults of the rotor. When the weight distribution near the center of rotation of the rotor is not uniform, an imbalance occurs, resulting in an imbalance force u, which is given by equation (3):
u＝m×r (3)
where m is the mass and r is the eccentricity.
When the imbalance forces fluctuate with the rotational speed and drag the rotor from the center of the stator to other locations, the mutual inductance between the stator and rotor circuits becomes non-uniform due to rotor imbalance, resulting in the induction of a frequency component f in the stator current unb Specifically, as shown in formula (4):
f unb ＝f s [1±k(1-b)/p]，k＝1，2，3，… (4)
wherein f is s Is the fundamental frequency of the current source, k is an integer, b is the induction motor slip, and p is the electrode logarithm.
Step (1) b: uniformity H and kurtosis K of current signal of motor state are extracted urt A characteristic parameter.
Calculating uniformity H parameter and kurtosis parameter K under each motor state urt Normalizing the current signals, and performing statistical analysis on each state to obtain uniformity H parameter and kurtosis K parameter of current signals of a healthy motor, a motor with one rotor broken bar fault 1BRB, a motor with two rotor broken bar faults 2BRB, a motor with bearing outer ring damage fault BRB and a motor with unbalanced mechanical load fault UNB urt Mean value μ and standard deviation σ of.
The uniformity and kurtosis characteristic parameters are calculated as follows:
(1) Calculating uniformity characteristic parameter H
The uniformity parameter H is a texture attribute that estimates the gray level variation of pixels of an image, and is derived from a gray level co-occurrence matrix (GLCM) of the image to measure the compactness of the diagonally distributed elements in the GLCM, and the GLCM can display the number of occurrences of gray levels of each pixel at a predetermined geometric position by the gray level function of any other pixel, and the uniformity ranges from 0 to 1, and reaches a maximum value when the value of the diagonal element is 1, and the following formula can be used to describe the uniformity parameter H:
where p (i, j) is the element with position number (i, j) in the normalized GLCM. The uniformity parameter can be used as an index for motor fault detection and classification because in the fault current signal of an induction motor there are different frequency signatures associated with the fault and thus will change its uniformity, and the uniformity parameter will vary accordingly.
(2) Calculating kurtosis characteristic parameter K urt
Kurtosis parameter K urt Deviations of the probability distribution can be measured as fourth order moments describing the shape of the signal probability distribution, and small differences between distributions having different shapes can also be measured and thus can be used as an indication for motor fault diagnosis. If there is a high pulse fault component in the kurtosis parameter, the signal will have a sharp signal strength distribution, soThere are high peak values. The kurtosis parameter for random event x is calculated as follows:
wherein N is the number of samples, x i Is the original signal sample, i =1,2,3, \ 8230;, N, μ is the random event X = [ X = 1 ，x 2 ，x 3 ，…，x N ]The mean value of (a), is the standard deviation.
In specific implementation, the uniformity H parameter and the kurtosis K parameter under each motion condition are obtained by the formulas (5) and (6) respectively urt And normalized. Statistical analysis was performed for each motion state. Obtaining uniformity H parameter and kurtosis K parameter of current signals for healthy electric machines, electric machines with one rotor bar break fault 1BRB, electric machines with two rotor bar break faults 2BRB, electric machines with bearing outer ring damage fault BRB and electric machines with unbalanced mechanical load fault UNB urt The mean value μ and the standard deviation σ to indicate the probability density function PDF of the corresponding signal.
Fig. 1 and fig. 2 are probability density function PDFs based on uniformity and kurtosis characteristics, and it can be seen that there is a significant overlap between processing conditions, and finally, a classification operation for motor faults is completed through a classifier so as to accurately identify the fault type of the motor.
Step 2: and completing the construction of the AGA-CS-based hybrid classifier model. Predicting errors by initializing weights, selecting father weights according to the predicted errors, finishing the processing of two algorithms for the selected father weights, wherein one algorithm is an adaptive genetic algorithm AGA, the other algorithm is a cuckoo search algorithm CS, and finally combining the results of the two methods to obtain the optimized neural network weights and the optimal solution.
The artificial neural network ANN replicates neural structures and human brain functions, has great flexibility and adaptability, and generally consists of a single input layer and a single output layer. The structure of a conventional neural network ANN is shown in fig. 3.
In the neural network, the training process is very important, and the optimal weight can be obtained by utilizing a training algorithm. The common training algorithm is a back propagation algorithm, the convergence rate of the method is slow, the calculation amount is large, and the method can often fall into local optimum. Therefore, the invention develops a new neural network training method, namely a mixed classification method AGA-CS based on an adaptive genetic algorithm AGA and a cuckoo algorithm CS. The method comprises the steps of using an adaptive genetic algorithm AGA and a cuckoo search algorithm CS in a mixed mode, taking weights as input, evaluating initialization weights to be fitness, selecting father weights according to fitness values, processing the selected father weights through two algorithms, wherein one algorithm is the adaptive genetic algorithm AGA, the other algorithm is the cuckoo search algorithm CS, and finally combining results of the two algorithms to obtain optimized neural network weights and an optimal solution.
Fig. 4 is an implementation process of the AGA-CS hybrid classifier model, and in conjunction with fig. 4, the main implementation process of step 2 is as follows:
step (2) a: initializing weight, predicting error, and selecting parent weight through predicting error;
initializing the weights of the neural network in a random manner, and then predicting the error by initializing the weights, see formula (7):
E＝A V -H (7)
where E is the error, H is the activation function, A V Is the actual value.
The activation function H can be obtained by using the equation (8):
wherein M is the number of neurons in the hidden layer, N is the number of neurons in the input layer,
is the jth weighting level between hidden and output, is>
Is the weight between the ith input layer and the jth hidden layer neuron, x i Is the ith input value.
Finally, a parent weight is selected, and the parent weight is selected through error prediction based on the optimization purpose of reducing errors, so that the calculation errors are further reduced.
Step (2) b: completing the operation of an adaptive genetic algorithm AGA;
the crossing rate and the variation rate of the genetic algorithm GA directly influence the convergence rate of the algorithm. On one hand, the crossing rate is too high, good individual genes are difficult to iterate, and the speed of a new individual is slow, both of which can cause the algorithm to stop and to converge too early; on the other hand, the larger the variation rate is, the algorithm becomes a completely random search state, and whether a globally optimal result can be searched is not influenced by operations such as selection, intersection and variation, and on the contrary, the smaller the variation rate is, a new excellent individual is not easy to generate. Therefore, in the implementation of the standard GA, the targeted cross rate and variance rate should be found by using a cyclic trial and error for each optimization problem, but this is almost impossible to achieve, and eventually leads to the difficulty in finding the optimal parameters. Therefore, the invention adopts the adaptive genetic algorithm AGA, and carries out linear adjustment on the cross rate and the variation rate according to the approaching condition of the individual adaptive value, the group average adaptive value and the maximum adaptive value, thereby improving the convergence rate of the algorithm. The method comprises the following specific steps:
(1) Rate of completion crossing m a Improvement of
The crossover rate is performed for new offspring of the genetic algorithm, which are formed by the combination of parent chromosomes. On one hand, if the difference between the larger fitness of the two individuals to be crossed and the maximum fitness of the population is larger, the crossed individual is regarded as a non-high-quality individual, and the crossing rate needs to be increased appropriately, so that the probability of obtaining a high-quality sample through crossing is improved; on the other hand, if the difference between the optimal fitness and the average fitness of the population is large, the samples are scattered, which means that the fitness function does not enter an extreme value yet, and the crossing rate needs to be reduced appropriately, so that the scattered population is iterated as soon as possible. In addition, when the better fitness value among the crossing individuals is smaller than the population average fitness, the sample is regarded as a worse sample, and the crossing rate of the worse sample is not considered according to the population distribution and is set to be a constant value to reduce the calculation amount. According to the above principle, the cross rate is improved as follows:
in the formula, g b For greater fitness values in two crossed individuals, g max Is the maximum individual fitness value in the population, g avg Is the mean fitness value of all individuals in the population, a 1 、a 2 Is a random number between 0 and 1 and is usually set manually.
(2) Complete mutation Rate m r Improvement and adaptive mutation of
The purpose of adaptive mutation is to obtain complete new offspring. In genetic algorithms, the adaptive mutation rate is randomly selected. Similar to the crossover rate, the difference between the mutant individual and the average fitness needs to be considered and adjusted, so the mutation rate mr is improved as follows:
wherein g is the individual fitness value of the mutation, g max Is the maximum individual fitness value in the population, g avg Is the mean fitness value of all individuals in the population, a 4 Is a random number between 0 and 1, a 3 The settings were as follows:
where R is a random value between [0,1] that varies in each iteration, and c and μ are variation parameters (in practice, c = μ =2 is set). After the variation is executed, a complete new generation is obtained, next, the AGA algorithm obtains a new optimization weight value through the new generation, and the steps (1) and (2) are repeatedly executed, so that a group of new optimization weights based on the AGA is obtained.
Step (2) c: completing the operation of a cuckoo algorithm CS;
like all evolutionary algorithms, the cuckoo optimization algorithm CS performs iterative search in a manner of randomly generating populations. During execution, mapping the positions of the nests of cuckoos as solutions in an algorithm population space, specifically comprising the following steps:
(1) Initializing, randomly generating N problem solutions, and recording an optimal solution.
Let N e ＝{N e1 ，N e1 ，N e1 ，…，N em The algorithm finishes searching in a random walking mode, and takes the advantages and disadvantages of the bird nest position as fitness, and the specific calculation is as the following formula (12):
wherein the content of the first and second substances,
represents the ith nest to be updated of the tth generation and is then preserved>
For the updated next generation bird nest, ρ is the step scaling factor, and in practice, it is set to be ρ =0.01, and/or>
Is the dot product operation, and L (lambda) is a Lavie random flight path, which represents a hopping path of the Lavie flight, and the direction and the length of the hopping path are uncertain. The relationship between the continuous lewy flight jump path and the time obeys the lewy distribution, and the corresponding probability density function can be obtained by simplifying the distribution function and performing Fourier transform, specifically see formula (13):
L(λ)～u＝t -λ ；1＜λ＜3 (13)
wherein λ represents a power coefficient.
(2) And step size evaluation is carried out to obtain a flight jump path.
In this step, the parent weight step lengths obtained in step 3 are evaluated, and the step length is calculated using the following expression:
wherein S is Z Is the step size, α is the step size parameter (in the implementation, α = 0.02), w t Is the current parent weight of the current parent,
is the best solution so far, r is [0,1] from a standard normal distribution]An inter-random number, S being a step number, obtained by simulating a levey flight jump path using equation (15):
wherein S represents a Lavy flight jump path, corresponding to L (λ), and β is [1,2]]In practice, β =1.6, u and v are normally distributed random numbers, the distribution is shown in formula (16), and the standard deviation of the corresponding normal distribution is σ u And σ v Specifically defined as shown in formula (17):
(3) And updating the solution position by utilizing the Laevir flight to obtain a new candidate solution position, and reserving a better solution by utilizing a greedy selection strategy.
(4) According to the discovery probability P a The partial solutions are discarded and new solutions are generated by random walk means, replacing the discarded solutions.
Step length S according to each father weight Z And searching paths, wherein the cuckoo individuals complete the flight process, and the cuckoo is arranged because the searching paths are randomly changedBirds will switch from one area to another, increasing their overall optimization. And after the flying walk of the global exploration is finished, local search is finished by adopting a mode of preference random walk based on the current solution. Comparing the discovery probability P a And a threshold constant
If/or>
(in practice, set->
) If the bird nest is found, namely the current solution is a poor solution, the bird nest parameters are updated, the poor solution is eliminated, and a new candidate solution is generated, so that the best cuckoo individuals are left. The bird nest position updating formula is as shown in formula (18):
wherein the content of the first and second substances,
indicates that the bird nest is to be updated and stands>
For the updated bird nest, xi is a random number which satisfies uniform distribution within [0,1], and>
is any two bird nests of the t generation.
(5) If the termination condition is met, outputting the best solution, otherwise, returning to (3) and continuing the iteration.
(6) A new solution is generated.
In this step, a new optimization scheme is created, and the updating of the parent weight is completed based on the CS algorithm, which is specifically as follows:
w t+1 ＝w t +S z (19)
wherein, w t+1 Is a new weight, S z Is the step size, w t Is the current parent weight.
Step (2) d: merging the mixing algorithms;
and (3) combining the algorithm results of the step (2) b and the step (2) c, and considering the combined weight as an optimized weight to obtain the optimal solution.
Step (2) e: judging a termination standard;
and (3) taking the maximum weight number of the training neural network as a termination standard, judging whether a termination condition is reached, if not, turning to the error prediction of the step 1, and continuously executing optimization until the termination condition is reached so as to obtain an optimal solution.
And step 3: and constructing a motor fault diagnosis model based on uniformity, kurtosis and AGA-CS, and realizing correct prediction of the faults of the five induction motors.
Fig. 5 is a process for implementing a motor fault diagnosis model based on uniformity, kurtosis, and AGA-CS, in which a motor transient start current signal is obtained through a current clamp, and then analog-to-digital conversion is performed on the motor transient start current signal, a uniformity parameter and a kurtosis parameter are obtained from the obtained discrete current signal, the uniformity parameter and the kurtosis parameter are used as input of an AGA-CS hybrid classifier, an adaptive genetic algorithm AGA and a cuckoo search algorithm CS are used in a hybrid manner, and a neural network weight is used as input to evaluate an initialization weight as fitness, a parent weight is selected according to the fitness value, then the selected parent weight is processed by using two algorithms, and finally results of the two methods are combined to obtain optimized neural network weights and optimal solutions, thereby finally implementing correct classification of five motor states, including a healthy motor, a motor with one rotor bar fault 1BRB, a motor with two rotor bar faults 2BRB, a motor with a bearing outer ring damage fault BRB, and a motor with an unbalanced mechanical load fault UNB.
And 4, step 4: based on the technology, an MATLAB neural network tool box is selected as an implementation tool, an experiment platform is built, a motor fault diagnosis model based on uniformity, kurtosis and AGA-CS is built, and classification of five motor states is completed. The following test experiments were mainly completed:
(1) Motor fault diagnosis experiment based on uniformity, kurtosis and AGA-CS
The method comprises the steps of utilizing transient starting current signals of the motor to identify and classify the states of the motor, wherein five diagnosis objects are respectively a healthy motor, a motor fault 1BRB with one rotor broken bar, a motor fault 2BRB with two rotor broken bars, a motor fault BRB with bearing outer ring damage and a motor fault UNB with unbalanced mechanical load. The experimental object selects an induction motor with the model of WEG 00136APE48T, the working power supply of the induction motor is 220V alternating current, the frequency is 60Hz, 28 rotors are arranged, and the used mechanical load is a common alternating current generator which is equivalent to one fourth of the nominal load of the detected induction motor. In the specific implementation, three-phase current signals of the induction motor are acquired through an alternating current clamp, and a 16-bit analog-to-digital converter (ADS 7809) is used for carrying out A/D conversion on the acquired data. In the implementation process, a motor experimental object is ensured to be in a healthy state, and the 1BRB/2BRB with one and two rotor broken bar faults is generated by respectively drilling a hole and two holes on a rotor shaft of the experimental motor object, wherein the hole diameter is 8mm, in addition, the hole with 2mm is drilled on a motor bearing by using a drill bit, the motor bearing is artificially damaged, so that the bearing outer ring is damaged, the motor fault BRB with the damaged bearing outer ring is generated, and finally, the mechanical unbalance fault UNB is generated by adding mass in a pulley arm. In the experiment, 50 times of experiments are carried out for each state, the sampling frequency of each experiment is 1.5kHz, namely 4096 samples are obtained at the starting moment of the induction motor. Finally, the experimental results are obtained as shown in Table 1.
Table 1 motor fault detection and classification results based on uniformity and kurtosis characteristics
| State of the electric machine | Fault classification accuracy |
| Health state motor | 98％ |
| Motor fault 1BRB of one rotor |
95％ |
| Motor failure 2BRB with two rotor broken bars | 96％ |
| Motor fault BRB (bus brake bridge) caused by bearing |
95％ |
| Unbalanced mechanical load motor fault UNB | 94％ |
As can be seen from table 1, the proposed method can effectively identify and classify the states of the induction motors, and the fault classification accuracy of the motors in a healthy state, the motors with one rotor bar breakage fault 1BRB, the motors with two rotor bar breakage faults 2BRB, the motors with bearing outer ring damage faults BRB and the motors with unbalanced mechanical load faults UNb can reach 98%, 95%, 96%, 95% and 94%.
(2) Comparative experiments with different classifier models
In practice, experimental verification was performed on the hybrid classification method (AGA-CS) proposed by the present invention with the conventional BP algorithm, as well as using the Adaptive Genetic Algorithm (AGA), the Cuckoo algorithm (CS) and the Genetic Algorithm (GA) alone. After the model is trained, 30 groups of samples are randomly selected, fault samples are sent into the trained model for testing, in the specific implementation process, the states of five motor states including a healthy motor, a motor with one rotor broken bar fault 1BRB, a motor with two rotor broken bar faults 2BRB, a motor with bearing outer ring damage fault BRB and a motor with unbalanced mechanical load fault UNB need to be counted, finally, the recognition rates obtained by different methods are averaged, and the diagnosis result is shown in table 2.
TABLE 2 comparison of the diagnostic results of different classification algorithms for different states of the motor
Fig. 6 is the average effect of different classification algorithms on motor fault diagnosis. From the above chart, it can be seen that the motor fault diagnosis method based on the uniformity and kurtosis calculation and the AGA-CS mixed classification mode has the best prediction result, and the absolute value of the prediction error is obviously smaller than the effect of using only one traditional algorithm (AGA, CS, BP, GA). Because the training sample contains the fault information of the motor and the AGA-CS mixed classification model can obtain the characterization information of the fault more deeply, compared with the traditional algorithm, the AGA-CS classifier has more obvious advantages in the aspect of motor fault diagnosis, has higher diagnosis accuracy and can obtain the diagnosis effect far higher than that of the diagnosis effect obtained by singly using one classification algorithm.
The above embodiments are merely illustrative of the technical concepts and features of the present invention, and the purpose of the embodiments is to enable those skilled in the art to understand the contents of the present invention and implement the present invention, and not to limit the protection scope of the present invention. All equivalent changes and modifications made according to the spirit of the present invention should be covered in the protection scope of the present invention.
Claims (7)
1. A motor fault diagnosis method based on uniformity and kurtosis calculation is characterized by comprising the following steps:
step 1: calculating the uniformity and the kurtosis of the transient starting current signal of the motor to obtain a fault characteristic parameter of the motor;
step 2: constructing a hybrid classifier model based on AGA-CS, training the hybrid classifier model by using uniformity and kurtosis characteristic parameters, predicting errors by initialized weights, selecting a father weight according to the predicted errors, finishing the processing of two algorithms for the selected father weight, wherein one algorithm is an adaptive genetic algorithm AGA, the other algorithm is a cuckoo search algorithm CS, and finally combining the results of the two methods to obtain an optimized neural network weight and an artificial neural network ANN optimal solution, thereby constructing a motor fault diagnosis model based on uniformity, kurtosis calculation and AGA-CS;
and step 3: and realizing the fault diagnosis of the induction motor based on the uniformity, the kurtosis calculation and the AGA-CS motor fault diagnosis model.
2. The motor fault diagnosis method based on uniformity and kurtosis calculation according to claim 1, characterized in that the motor fault characteristic parameters in step 1 set five motor states, which are respectively a healthy motor, a motor fault 1BRB with one rotor bar, a motor fault 2BRB with two rotor bars, a motor fault BRB with bearing outer ring damage, and a motor fault un with unbalanced mechanical load; the fault types are four, and the determination basis of the fault types is as follows:
(1) Rotor broken bar fault BRB
Using the relevant frequency parameter f BRB Indicating the presence of a BRB fault, as specified by the formula:
f BRB ＝(1-2kb)·f s ，k＝1，2，3，…
wherein k is an integer, f s The main frequency component of the power supply, b represents the motor slip, and the value range of the motor slip is 0 to 1;
(2) Bearing fault BRN
Outer race relative frequency f 0 Determined by the following equation:
where n is the number of rolling elements, f r Is the rotational frequency of the shaft, BD is the ball diameter, PD is the bearing race diameter, and β is the angle between the rolling elements in the race;
(3) Mechanical unbalance Fault UNB
The rotor is unbalanced, generating an unbalanced force u, which is given by:
u＝m×r
where m is mass and r is eccentricity.
When the imbalance forces fluctuate with the speed of rotation and drag the rotor from the center of the stator to other locations, the mutual inductance between the stator and rotor circuits becomes non-uniform due to the rotor imbalance, resulting in a frequency component f being induced in the stator current unb Specifically, the formula is shown as follows:
f unb ＝f s [1±k(1-b)/p]，k＝1，2，3，…
wherein f is s Is the fundamental frequency of the current source, k is an integer, b is the induction motor slip, and p is the electrode logarithm.
3. The motor fault diagnosis method based on uniformity and kurtosis calculation of claim 1, wherein the uniformity H and kurtosis K of the motor current signal extracted in step 1 are urt Characteristic parameters are as follows:
calculating uniformity and kurtosis parameters under each motor state, and normalizing the uniformity and the kurtosis parameters; performing statistical analysis on each state to obtain the average value mu and the standard deviation sigma of the uniformity parameter and the kurtosis parameter of current signals of a healthy motor, a motor with one rotor broken bar fault 1BRB, a motor with two rotor broken bar faults 2BRB, a motor with a bearing outer ring damage fault BRB and a motor with an unbalanced mechanical load fault UNB; the calculation steps are as follows:
(1) Calculating uniformity characteristic parameters
The uniformity parameter H is described by the following equation:
where p (i, j) is the element with position number (i, j) in the normalized GLCM. The uniformity parameter can be used as an index for motor fault detection and classification because in the fault current signal of the induction motor there are different frequency signatures associated with the fault and thus will change its uniformity, so the uniformity parameter will also change with it;
(2) Calculating kurtosis characteristic parameters
The kurtosis characteristic parameter may measure the deviation of the probability distribution, which is the fourth moment describing the shape of the signal probability distribution, and is calculated as follows:
wherein N is the number of samples, x i Is the original signal sample, i =1,2,3, \8230;, N, μ is the random event X = [ X 1 ，x 2 ，x 3 ，…，x N ]The mean value of (a), is the standard deviation.
4. The motor fault diagnosis method based on uniformity and kurtosis calculation as claimed in claim 1, wherein in step 2, weight initialization, error prediction, and selection of a parent weight through prediction error are as follows:
initializing the weights of the neural network in a random manner, and then predicting the error through the initialized weights, specifically according to the following formula:
E＝A V -H
where E is the error, H is the activation function, A V Is the actual value;
the activation function H can be obtained by:
wherein M is the number of neurons in the hidden layer, N is the number of neurons in the input layer,
is the jth weighting level between hidden and output, is>
Is the weight between the ith input layer and the jth hidden layer neuron, x i Is the ith input value;
finally, a parent weight is selected, and the parent weight is selected through error prediction based on the optimization purpose of reducing errors, so that the calculation errors are further reduced.
5. The motor fault diagnosis method based on uniformity and kurtosis calculation as claimed in claim 4, wherein the operation of completing the adaptive genetic algorithm AGA for the selected parent weight in step 2 is specifically:
according to the approaching condition of the individual adaptive value, the group average adaptive value and the maximum adaptive value, the cross rate and the variation rate are linearly adjusted, which specifically comprises the following steps:
(1) Rate of completion crossing m a The improvement of the cross rate is as follows:
in the formula, g b Is the greater fitness value, g, in two intersecting individuals max Is the maximum individual fitness value in the population, g avg Is the mean fitness value of all individuals in the population, a 1 、a 2 Is a random number between 0 and 1;
(2) Complete mutation Rate m r Improvement and adaptive mutation of (2) and (3) the mutation rate m r The improvement is that:
wherein g is the individual fitness value of the mutation, g max Is the maximum individual fitness value in the population, g avg Is the mean fitness value of all individuals in the population, a 4 Is a random number between 0 and 1, a 3 The settings were as follows:
wherein R is a random value between [0,1] that varies in each iteration, c and μ are variation parameters; after the mutation process is executed, we will obtain a complete new generation, and then the AGA algorithm obtains a new optimization weight through the new generation, and repeatedly executes (1) and (2), thereby obtaining a set of new optimization weights based on AGA.
6. The motor fault diagnosis method based on uniformity and kurtosis calculation of claim 4, wherein the operation of completing cuckoo algorithm CS on the selected father weight in step 2 is specifically:
during execution, mapping the positions of the nests of cuckoos as solutions in an algorithm population space, specifically comprising the following steps:
(1) Initializing, randomly generating N problem solutions, and recording an optimal solution;
let N e ＝{N e1 ，N e1 ，N e1 ，…，N em The algorithm completes the search in a random walking mode, and takes the advantages and disadvantages of the bird nest position as the fitness, and the concrete calculation is as follows:
wherein the content of the first and second substances,
represents the ith nest to be updated of the tth generation and is then preserved>
For the updated next generation nest, ρ is the step scaling factor, <>
The method is characterized in that the method is dot product operation, L (lambda) is a Levy random flight path, represents a hopping path of Levy flight, and has uncertain direction and length; the relationship between the continuous Levis flight jump path and time obeys Levis distribution, and a corresponding probability density function can be obtained by simplifying the distribution function and performing Fourier transform, specifically the following formula:
L(λ)～u＝t -λ ；1＜λ＜3
wherein λ represents a power coefficient;
(2) Step size evaluation, obtaining a flight jump path:
evaluating each obtained parent weight step size and calculating the step size using the following expression:
wherein S is Z Is the step size, α is the step size parameter, w t Is the current parent weight that is the current parent weight,
is the best solution so far, r is [0,1] from a standard normal distribution]An inter-random number, S being a step number, the value of S being obtained by simulating a levey flight jump path using the following equation:
wherein S represents a Lave flight jump path, corresponding to L (lambda), beta is a [1,2] parameter, u and v are normally distributed random numbers, and the distribution is shown in the following formula
Standard deviation of corresponding normal distribution is σ u And σ v The specific definition is shown as the following formula:
(3) Updating the solution position by utilizing the Laevir flight to obtain a new candidate solution position, and reserving a better solution by utilizing a greedy selection strategy;
(4) According to the discovery probability P a Discarding the partial solution and generating a new solution by a random walk mode so as to replace the discarded solution;
step length S according to each father weight z The cuckoo individuals complete the flight process, and the search paths are randomly changed, so that the cuckoo can be switched from one area to another area, and the overall optimizing capacity of the cuckoo is enhanced; after the flying walk of the global exploration is finished, local search is finished by adopting a mode of preference random walk based on the current solution; comparing the discovery probability P a And a threshold constant
If/or>
The bird nest is found, namely the current solution is a poor solution, then bird nest parameters are updated, the poor solution is eliminated, and a new candidate solution is generated, so that the best cuckoo individual is left; the bird nest location update formula is described by the following equation:
wherein the content of the first and second substances,
indicates a bird nest to be updated, and>
for the updated bird nest, xi is [0, 1) and satisfies the uniformly distributed random number>
Is any two bird nests of the t generation;
(5) If the termination condition is met, outputting the best solution, otherwise, returning to the step (3) to continue iteration;
(6) Generating a new solution, creating a new optimization scheme, and completing the updating of the parent weight based on the CS algorithm, which is specifically as follows:
w t+1 ＝w t +S z
wherein, w t+1 Is a new weight, S z Is the step size, w t Is the current parent weight.
7. The method as claimed in claim 1, wherein the step 3 of implementing the fault diagnosis of the induction motor based on the motor fault diagnosis model of the uniformity, kurtosis calculation and AGA-CS hybrid classifier comprises the following specific steps:
obtaining a motor transient starting current signal, then processing and carrying out analog-to-digital conversion on the motor transient starting current signal, obtaining expected characteristics including uniformity parameters and kurtosis parameters from the obtained discrete current signal, and taking the expected characteristics as the input of an AGA-CS (advanced glass-substrate learning-class) based hybrid classifier model; in a hybrid classifier based on an AGA-CS (accelerated aging-learning) algorithm, an adaptive genetic algorithm AGA and a cuckoo search algorithm CS are used in a hybrid mode, neural network weights are used as input, initialization weights are evaluated to be fitness, father weights are selected according to fitness values, then the selected father weights are processed by two algorithms, and finally results of the two methods are combined, so that optimized neural network weights and an optimal solution are obtained, the induction motor fault is diagnosed by the aid of the father weights, and five motor states are correctly classified, wherein the five motor states comprise a healthy motor, a motor with a rotor broken bar fault 1BRB, a motor with two rotor broken bar faults 2BRB, a motor with a bearing outer ring damage fault BRB and a motor with an unbalanced mechanical load fault UNB.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211597697.6A CN115856623A (en) | 2022-12-13 | 2022-12-13 | Motor fault diagnosis method based on uniformity and kurtosis calculation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211597697.6A CN115856623A (en) | 2022-12-13 | 2022-12-13 | Motor fault diagnosis method based on uniformity and kurtosis calculation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115856623A true CN115856623A (en) | 2023-03-28 |
Family
ID=85672475
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211597697.6A Pending CN115856623A (en) | 2022-12-13 | 2022-12-13 | Motor fault diagnosis method based on uniformity and kurtosis calculation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115856623A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117407824A (en) * | 2023-12-14 | 2024-01-16 | 四川蜀能电科能源技术有限公司 | Health detection method, equipment and medium of power time synchronization device |
| CN117708574A (en) * | 2024-02-02 | 2024-03-15 | 江苏南高智能装备创新中心有限公司 | CNN variable-speed rolling bearing fault diagnosis method embedded with physical information |
-
2022
- 2022-12-13 CN CN202211597697.6A patent/CN115856623A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117407824A (en) * | 2023-12-14 | 2024-01-16 | 四川蜀能电科能源技术有限公司 | Health detection method, equipment and medium of power time synchronization device |
| CN117407824B (en) * | 2023-12-14 | 2024-02-27 | 四川蜀能电科能源技术有限公司 | Health detection method, equipment and medium of power time synchronization device |
| CN117708574A (en) * | 2024-02-02 | 2024-03-15 | 江苏南高智能装备创新中心有限公司 | CNN variable-speed rolling bearing fault diagnosis method embedded with physical information |
| CN117708574B (en) * | 2024-02-02 | 2024-04-12 | 江苏南高智能装备创新中心有限公司 | CNN variable-speed rolling bearing fault diagnosis method embedded with physical information |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109829402B (en) | GS-SVM-based bearing damage degree diagnosis method under different working conditions | |
| CN115856623A (en) | Motor fault diagnosis method based on uniformity and kurtosis calculation | |
| CN110276416B (en) | Rolling bearing fault prediction method | |
| Samanta et al. | Artificial neural networks and genetic algorithm for bearing fault detection | |
| US6606580B1 (en) | Fault diagnosis | |
| Subrahmanyam et al. | Using neural networks for the diagnosis of localized defects in ball bearings | |
| Chen et al. | A meta-learning method for electric machine bearing fault diagnosis under varying working conditions with limited data | |
| CN110070060B (en) | Fault diagnosis method for bearing equipment | |
| CN110376522B (en) | Motor fault diagnosis method of data fusion deep learning network | |
| CN112629863A (en) | Bearing fault diagnosis method for dynamic joint distribution alignment network under variable working conditions | |
| CN111832353B (en) | Steam turbine rotor fault diagnosis method based on EMD and BA optimization SVM | |
| CN111562108A (en) | Rolling bearing intelligent fault diagnosis method based on CNN and FCMC | |
| CN112257530B (en) | Rolling bearing fault diagnosis method based on blind signal separation and support vector machine | |
| WO2020090767A1 (en) | Abnormality diagnostic device, abnormality diagnostic method, and program | |
| Júnior et al. | The design of multiple linear regression models using a genetic algorithm to diagnose initial short-circuit faults in 3-phase induction motors | |
| CN107831024A (en) | Fan vibration malfunction diagnostic method based on multiple spot vibration signal characteristics value | |
| CN112308038A (en) | Mechanical equipment fault signal identification method based on classroom type generation confrontation network model | |
| CN107101829A (en) | A kind of intelligent diagnosing method of aero-engine structure class failure | |
| CN113408068A (en) | Random forest classification machine pump fault diagnosis method and device | |
| CN111428318A (en) | Rolling bearing degradation trend prediction method based on whale algorithm optimization | |
| CN112465124A (en) | Twin depth space-time neural network model acquisition/fault diagnosis method and device | |
| CN116595317A (en) | Motor fault diagnosis method based on orthogonal stator current GA analysis and SFO-PNN | |
| CN112818601A (en) | Hydroelectric generating set health assessment method based on GA-BP neural network and error statistical analysis | |
| CN115034137A (en) | RVM and degradation model-based two-stage hybrid prediction method for residual life of bearing | |
| CN115587290A (en) | Aero-engine fault diagnosis method based on variational self-coding generation countermeasure network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |