CN107590506A - A kind of complex device method for diagnosing faults of feature based processing - Google Patents
A kind of complex device method for diagnosing faults of feature based processing Download PDFInfo
- Publication number
- CN107590506A CN107590506A CN201710706030.8A CN201710706030A CN107590506A CN 107590506 A CN107590506 A CN 107590506A CN 201710706030 A CN201710706030 A CN 201710706030A CN 107590506 A CN107590506 A CN 107590506A
- Authority
- CN
- China
- Prior art keywords
- feature
- fault
- data
- value
- equipment
- 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.)
- Granted
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Test And Diagnosis Of Digital Computers (AREA)
Abstract
The present invention provides a kind of complex device method for diagnosing faults of feature based processing, and step is as follows：1st, action current signal is gathered in real time；2nd, section partition is carried out to action current curve；3rd, section partition is carried out to operation curve；4th, the high dimensional feature for building device action current curve represents data set；5th, data set, which carries out feature selecting, to be represented to high dimensional feature；6th, data set, which carries out feature extraction, to be represented to high dimensional feature；7th, division processing is carried out to character representation data set；8th, optimizing solution is carried out to SVM parameter；9th, SVM supervised learnings are carried out, obtain fault diagnosis model；10th, the accuracy rate of diagnosis of fault diagnosis model is verified using test set data；By above-mentioned steps, the complex device method for diagnosing faults of feature based processing can be realized, and fault diagnosis grader is realized by the SVM methods of parameter optimization, completes Fault Identification and the analysis of characteristic to equipment operating current curve.
Description
Technical Field
The invention provides a method for diagnosing faults of complex equipment based on feature processing, relates to the implementation of the method for diagnosing the faults of the complex equipment based on the feature processing, and belongs to the field of reliability and fault diagnosis of the complex equipment.
Background
The fault diagnosis is a comprehensive emerging science which searches for a fault source according to equipment operation state information and determines a corresponding decision. The failure diagnosis technology starts in the 60's of the 20 th century, and has made great progress through the development of more than 50 years. The content of the method is developed from fault diagnosis of simple equipment to fault diagnosis of complex systems, and nowadays, fault diagnosis technology is widely applied to various industrial departments. At present, fault diagnosis for complex equipment mainly involves two problems: firstly, the corresponding feature description of the complex equipment in different working modes has the difficulty that how to carry out concise and efficient feature combination on the working state of the equipment; secondly, the construction of a fault diagnosis model has the difficulty that how to design the adaptability and the fault tolerance of the diagnosis model enables the model to make intelligent decisions according to different working states of complex equipment and accurately judge and identify faults. In order to solve the two problems, the invention provides a complex equipment fault diagnosis method based on feature processing, which is characterized in that current signals of complex equipment are subjected to feature representation, selection and extraction through technologies such as fusion feature representation of intelligent partitions, fisher feature selection, principal Component Analysis (PCA) and the like to form feature data, and a Support Vector Machine (SVM) is used for supervised learning of the feature data to establish a fault diagnosis model for fault diagnosis of the complex equipment.
The method is realized by fusing relevant theories, methods and technologies such as fault analysis, fault processing and the like based on the characteristic processing and support vector machine technology, and achieves the purposes of improving the reliability, safety and usability of the complex equipment while improving the maintainability of the complex equipment.
Disclosure of Invention
Objects of the invention
In the research of the fault diagnosis of the complex equipment, the following problems exist:
● The input information for the failure diagnosis is ambiguous. Although more research has been directed to fault diagnosis classification methods, less discussion has been made regarding the effective characterization of faults. Although there are some methods of characterizing features at present, the basis is unclear;
● The machine learning method has long training time and is tedious to train. Although the optimization algorithm based on machine learning is various, the optimization effect of the machine learning method for fault diagnosis is poor;
● The research on fault judgment of complex equipment is more. The fault identification is in a neglected or auxiliary position in fault diagnosis, and the judgment and the identification cannot be simultaneously combined to carry out complete fault diagnosis on the complex equipment.
In order to solve the problems, the invention provides a complex equipment fault diagnosis method based on feature processing, which overcomes the defects of the prior art. The method comprises the steps of firstly carrying out fusion characteristic representation of intelligent partitions according to action current data of equipment. And then screening natural fusion features by using a Fisher feature selection method, reserving feature data beneficial to fault diagnosis, further extracting the screened features by using a PCA method on the basis, and representing the operating data of the equipment by using the least and most effective feature combination. And finally, constructing a fault diagnosis classifier model by adopting an Optimization Algorithm (GA-PSO Algorithm) and an SVM (support vector machine) method, wherein the Optimization Algorithm (GA) is combined with a Genetic Algorithm (GA) and a Particle Swarm Optimization (PSO) to realize fault diagnosis of the equipment characteristic data. The complex equipment fault diagnosis method based on machine learning can not only express effective characteristics of faults, but also judge and identify the faults by adopting an optimized diagnosis model, provides a new solution for the field of fault diagnosis, and innovates the existing fault diagnosis method.
(II) technical scheme
The invention relates to a complex equipment fault diagnosis method based on feature processing, which comprises the following steps:
step 1, acquiring an action current signal of complex equipment in real time according to a proper sampling interval, obtaining action curve data of the equipment, and analyzing a corresponding relation between an action current curve form and an equipment fault type;
step 2, performing section division on the action current curve of the equipment from the time domain angle, and extracting time characteristic information of action curve data of each section;
step 3, performing section division on the action curve of the equipment from a value domain angle, and extracting current value characteristic information of action curve data of each section;
step 4, fusing the time characteristic information and the current value characteristic information of the equipment action to construct a high-dimensional characteristic representation data set of the equipment action current curve;
step 5, performing feature selection on the high-dimensional feature representation data set by using a Fisher method, screening out features with the best representativeness and separability, and eliminating features which have ambiguous edges and are difficult to distinguish, so that the dimension of the feature representation set is reduced;
step 6, extracting the features of the high-dimensional feature expression data set by using a PCA method, eliminating redundant information and further reducing the dimension of the feature expression set;
step 7, dividing the feature representation data set to obtain SVM training sample set data and test sample set data;
step 8, optimizing and solving the parameters of the SVM by adopting a GA-PSO optimization algorithm on the basis of the training sample set;
step 9, performing SVM supervised learning by using the optimized parameters and the training set sample data to obtain a fault diagnosis model;
and step 10, verifying the diagnosis accuracy of the fault diagnosis model by using the test set data.
In step 1, the method for analyzing the relationship between the operating current curve form and the type of the equipment fault is as follows: analyzing the normal mode and the typical fault mode of the equipment, summarizing the fault type, the occurrence phenomenon and the generation reason of each fault mode, and searching the corresponding relation between the action current curve form of the equipment and the fault type of the turnout for subsequent fault diagnosis steps.
In step 2, the method for "segmenting the motion curve of the device from the time domain perspective" is as follows: analyzing current signal data of different actions by the equipment to obtain current value domain information of the actions, and correspondingly setting time division points for the time domain (x axis) of the action curve according to the current value domain information, so that the time domain of the action curve can be divided into different sections, and each time domain section represents the duration of different actions of the equipment.
In step 2, the method for extracting the time characteristic information of the motion curve data of each segment includes the following steps: the time is used as an independent variable to describe the physical quantity of the motion curve data, and the working state of each motion section of the equipment is reflected in the most basic and intuitive expression mode. The invention takes the maximum value, the minimum value, the average value, the variance value, the effective value, the mean square error, the peak value, the kurtosis value, the peak factor, the wave form factor, the pulse factor, the margin factor and the like as time domain characteristic parameters, and collects the 12 characteristic parameters in each section divided by the time domain.
In step 3, the method of "segmenting the motion curve of the device from the view point of the value range" is as follows: firstly, the action current curve is projected to a current value range (y axis), the precision of a projection interval is set according to the acquisition precision of the equipment action current signal, and the precision of the projection interval is set to be consistent with the acquisition precision in order to avoid the influence of the projection interval on a divided current value range. Then, zero is set for the scattered current value point with the number of points less than 5 points in the projection interval, and statistics are not counted. Finally, segmenting the current value range according to the current information of the action of the equipment;
in step 3, "extracting current value characteristic information of the operation curve data of each segment" is performed as follows: and according to the current value range segmentation analysis, extracting non-zero current regions of each section, and performing parameter representation on the regions by using statistical characteristics. The invention takes total number, mean value, range, mode, variance, standard deviation, median, average absolute error and the like as statistical characteristic parameters, and collects the 8 characteristic parameters for each section to represent the action current characteristics of the equipment.
In step 4, the "fusing the time characteristic information and the current value characteristic information of the device operation" is performed as follows: firstly, summarizing 12 items of feature information of each divided time domain section to form a multi-dimensional time domain feature vector. Then, summarizing 8 items of feature information of each divided current value domain section to form a multi-dimensional current value domain feature vector; and finally, fusing the two multi-dimensional feature vectors to serve as a high-dimensional feature representation set of the device action current curve.
The Fisher method described in step 5 refers to a classical two-class feature selection method, in which the ratio of the inter-class distance to the intra-class distance of each feature representation value is used as a Fisher criterion function, and a larger function value indicates that the discrimination performance of the feature for classification recognition is stronger. The Fisher criterion function of a single feature is used for sequencing the features and selecting the features with strong identification performance, so that a new low-dimensional feature representation set can be formed, and the purpose of feature dimension reduction is realized;
in step 5, "feature selection is performed on the high-dimensional feature representation set data by using Fisher method", which is performed as follows: firstly, using Fisher criterion two-class feature selection method to calculate criterion function value between every typical fault mode and normal mode, d-th dimension feature F of equipment action current curve of i-th fault mode d The Fisher criterion function of (a) is:
here, between classesDifference S B,d And within class variance S W,d Are respectively defined as:
S B,d ＝(m 1,d -m i,d ) 2 i＝1,2,…n
here, m i,d And σ i,d Are respectively characteristic F d Mean and standard deviation in class i. Then, determining a feature selection standard, wherein the invention adopts a 'over half selection' mode, namely taking half of the maximum value of the Fisher criterion function value of the fault mode as a standard, selecting the Fisher criterion function value of each dimension feature data if the Fisher criterion function value exceeds the standard, and discarding the Fisher criterion function value if the Fisher criterion function value is lower than the standard; finally, the selected feature data are combined together to form a set of feature representations that facilitate distinguishing between typical failure modes.
The "PCA method" in step 6 refers to a feature extraction method, which can be used for data dimension reduction and finding effective and important elements and structures in the data. The main purpose of the PCA method is to use less variables to explain most variables in the original data and ensure the minimum information loss of the original data;
in step 6, "feature extraction is performed on the high-dimensional feature representation data set by using the PCA method", which is performed as follows:
step 6.1 feature centralization. Subtracting the mean value of the dimension from the attribute data of each dimension of the high-dimension feature representation data set A to change the mean value of each dimension of the matrix B obtained after transformation into 0;
step 6.2, calculating a covariance matrix C of the matrix B;
6.3, calculating an eigenvalue and an eigenvector of the covariance matrix C;
step 6.4, sorting the eigenvalues according to the sequence from large to small, selecting the largest k of the eigenvalues, and then respectively taking the corresponding k eigenvectors as column vectors to form an eigenvector matrix;
and 6.5, projecting the sample points to the selected characteristic vectors to obtain a new k-dimensional data set. Wherein k is selected based on the ratio of the sum of the first k eigenvalues to the sum of all eigenvalues
It should be noted that, in the PCA method, the class information of the samples in the training set does not need to be considered during the solution, and the training samples are treated equally no matter which class they come from. So that after the PCA feature space transformation, the classification is not necessarily favorably affected, and may even be adversely affected. Therefore, the k value should be selected according to the requirement of practical problem, so that the new k-dimensional data after mapping contains original data information as much as possible, and information effective for classification is lost as little as possible.
The "dividing the feature expression data set" described in step 7 is performed as follows: the feature representation data set is divided into a training set and a test set according to the proportion of 3. The training set data is used for SVM modeling, and the test set data is used for testing the recognition accuracy of the degradation state recognition model. On the basis, the same quantity of characteristic data of each fault mode is selected to enter a training set and a testing set, so that the diagnosis accuracy, the false alarm rate and the false alarm rate can become indexes for judging the effectiveness degree of the fault diagnosis model. If the sample distribution is too uneven, the method is used for judging the meaning of the common indexes of the fault diagnosis model;
the "SVM" in step 7 refers to a support vector machine model, which is a classifier model based on statistical theory and can be used for pattern classification, linear and nonlinear regression analysis. The principle of the support vector machine is that a training sample is given, a classification hyperplane is established as a decision curved surface, the isolation edge between a positive case and a negative case of the sample is maximized, and the classification of the sample is completed.
The "GA" in step 8 refers to a genetic algorithm, which simulates the process of low-level to high-level evolution of living beings in nature by using the natural selection concept and genetic mechanism of darwinian biogenesis theory, and is a global random search algorithm. For a complex optimization problem, an optimal solution can be obtained only by selecting, crossing and mutating three genetic operators, and the algorithm is used for solving the optimization problem and is irrelevant to gradient information, and only an objective function is required to be calculated;
the PSO in the step 8 refers to a particle swarm optimization algorithm, which is an optimization algorithm based on a swarm intelligence theory, has the advantages of self-learning and learning to others, and can find an optimal solution in a few iteration times;
in step 8, "the optimization solution is performed on the parameters of the SVM by using the GA-PSO optimization algorithm," which is performed as follows: and (3) with GA as a basic framework, carrying out PSO optimization on the basis of the selected excellent individuals, and enhancing the local search capability of the algorithm on the basis of keeping good global search capability. From the perspective of a genetic algorithm, the excellent samples are optimized continuously, so that in one iteration, excellent individuals are optimized by two different algorithms, and a parent and a child make 'common progress'; from the perspective of the particle swarm, the information of the optimal individual is equivalently maintained, suboptimal individuals are selected to carry out crossover and inheritance operations, and less excellent individuals carry out mutation operations, so that on one hand, good memory in an iteration process is saved, the average fitness of the population is moved to a good direction, on the other hand, the diversity of the population is ensured, and the situation that the population falls into a local optimal value as early as possible is avoided. The algorithm can keep good memorability, so that the searching speed is accelerated, meanwhile, the population information is continuously enriched, and the unidirectional flow of the information is avoided;
in step 9, "SVM model learning using the optimized parameters and training set sample data" is performed as follows:
step 9.1, setting a known training set: t = { (x) 1 ,y 1 ),…,(x i ,y i )}∈(X×Y) l
Wherein x is i Is a feature vector, y i For the corresponding attribute value, x i ∈X∈R n ，y i ∈Y＝{-1,1}，i＝1,2,…,l；
Step 9.2, selecting a kernel function g (x, x') optimized by GA-PSO and a penalty parameter C, and constructing and solving an optimization problem:
so that
Thereby obtaining an optimal solution
Step 9.3 selection of alpha * A positive component of 0<α * <, C, and calculating the threshold b based on the values * ：
Step 9.4 constructs a decision function f (x):
step 9.5 outputs the class according to the value of the decision function f (x).
And obtaining a fault diagnosis model after SVM training is completed. If the equipment has a fault, the diagnosis model can position the fault through the current signal and identify the mode type of the fault.
In step 10, "verify the diagnosis accuracy of the fault diagnosis model using the test set data" is performed as follows: and verifying the model by a test set data injection verification mode, comparing the diagnosis result and the actual result of the fault mode to obtain the fault diagnosis accuracy, and further judging whether the diagnosis model can meet the requirements.
Through the steps, the complex equipment fault diagnosis method based on the characteristic processing can be realized, the complex equipment is taken as an object, and firstly, the fusion characteristic representation of the intelligent subareas is carried out on the working current signals according to the fault characteristics of different fault modes of the equipment. And then, screening the fusion features by using a Fisher feature selection method, reserving features beneficial to fault diagnosis, and further extracting the screened features by combining a PCA method. And finally, obtaining SVM optimization parameters by using a GA-PSO optimization algorithm combining GA and PSO, and realizing a fault diagnosis classifier by using a parameter optimization SVM method to finish fault identification and analysis of the characteristic number of the equipment working current curve.
(III) advantages
Compared with the prior art, the invention has the advantages that: at present, most fault diagnosis technologies related to complex equipment cannot effectively represent characteristics of data collected by the equipment, so that the problems of redundancy, omission and the like of characteristic data input into a diagnosis system are caused. Moreover, the training technology for the diagnosis data has long operation time, complicated steps and poor practical application effect. The invention can perform feature fusion-selection-extraction processing on the working data acquired by the equipment to obtain a data set which effectively represents the fault features of the equipment. On the basis of the data set, a fusion optimization algorithm of GA and PSO is adopted, optimization parameters of the fault diagnosis model are rapidly and efficiently obtained, and the fault diagnosis model is obtained in an SVM supervision learning mode.
Drawings
FIG. 1 is a general process flow of the present invention.
FIG. 2 is a value range characterization process of the present invention.
FIG. 3 is a genetic algorithm process flow of the present invention.
Fig. 4 is a particle swarm algorithm processing flow of the present invention.
FIG. 5 is a flow chart of genetic-particle swarm algorithm parameter optimization according to the present invention.
FIG. 6 is a flow chart of a fault diagnosis model of the production support vector machine of the present invention.
Fig. 7 is an overall implementation flow of the fault diagnosis method of the present invention.
The numbers, symbols and codes in the figures are explained as follows:
"Fisher" in FIG. 1,7 refers to the Fisher method for feature selection for high-dimensional representation of datasets;
"PCA" in FIG. 1,7 refers to a principal component analysis method for feature extraction of a high-dimensional representation data set;
"GA-PSO" in FIG. 1,7 refers to a method combining genetic algorithm and particle swarm optimization algorithm for optimization of support vector machine parameters;
"SVM" in FIGS. 1,3,4,7 refers to a support vector machine for generating a fault diagnosis model;
"PSO" in FIG. 5 refers to a particle swarm optimization method for optimization of support vector machine parameters.
Detailed Description
In order to make the technical problems, technical solutions and advantages of the present invention more apparent, the following detailed description is provided with reference to the accompanying drawings.
The invention provides a complex equipment fault diagnosis method based on feature processing and a support vector machine, which can be used for performing feature representation on action signal data of complex equipment to obtain a feature representation data set. On the basis of the feature representation data set, a GA-PSO algorithm and an SVM model are combined, and a fault diagnosis model is obtained through a supervised learning mechanism and is used for fault judgment and identification of equipment.
The invention relates to a complex equipment fault diagnosis method based on feature processing and a support vector machine, which comprises the following concrete implementation steps as shown in figure 1:
the method comprises the following steps: acquiring action current signals of complex equipment in real time according to a proper sampling interval to obtain action curve data of the equipment, analyzing the normal mode and the typical fault mode of the equipment, summarizing fault types, occurrence phenomena and reasons of occurrence of the fault modes, and finding out a corresponding relation between the action current curve form of the equipment and the fault type of a turnout;
step two: the method comprises the steps of carrying out section division on an action current curve of the equipment from the time domain angle, analyzing current signal data of different actions of the equipment to obtain current value domain information of the actions, and correspondingly setting time division points on the time domain (x axis) of the action curve according to the current value domain information, so that the time domain of the action curve can be divided into different sections, wherein each time domain section represents the duration of different actions of the equipment. After the division of the sections is finished, the time is used as an independent variable to describe the physical quantity of the action curve data, and the working state of each action section of the equipment is reflected in the most basic and intuitive expression mode. The invention takes the maximum value, the minimum value, the average value, the variance value, the effective value, the mean square error, the peak value, the kurtosis value, the peak value factor, the wave form factor, the pulse factor, the margin factor and the like as time domain characteristic parameters. The characteristic parameters are explained as follows:
| characteristic parameter | Detailed Description |
| Maximum value | Representing the distribution interval of current values |
| Minimum value | Representing the distribution interval of current values |
| Mean value of | Indicating the central tendency of the current value |
| Effective value | Average energy of reaction current value |
| Variance value | Describing the degree of smoothing of the current value |
| Mean square error | Describing the degree of smoothing of the current value |
| Peak to peak value | Indicating maximum fluctuation of current value |
| Kurtosis value | Indicating the degree of sensitivity to current surge signals |
| Crest factor | Indicator for indicating whether surge signal exists in current signal |
| Form factor | The fluctuation trend of the current signal is reflected and is independent of the amplitude |
| Pulse factor | Indicating the degree of sensitivity to current-surge-pulse-like signals |
| Margin factor | Indicating sensitivity of the current impulse signal, which parameter reduces variation in deviation |
The 12 characteristic parameters are collected in each section divided by the time domain, and the extraction work of the time characteristic information of the section action curve is completed;
step three: the operation flow of the segment division and the feature parameter extraction of the action curve of the device from the value domain perspective is shown in fig. 2. Firstly, the action current curve is projected to a current value range (y axis), the precision of a projection interval is set according to the acquisition precision of the equipment action current signal, and the precision of the projection interval is set to be consistent with the acquisition precision in order to avoid the influence of the projection interval on a divided current value range. Then, zero setting is carried out on scattered current value points with the number of points less than 5 points in the projection interval, statistics are not counted, and the current value domain is segmented according to the current value points of the action of the equipment; and finally, carrying out segmentation analysis according to the current value range, extracting non-zero current areas of each section, and carrying out parameter representation on the areas by using statistical characteristics. The invention takes total number, mean value, range, mode, variance, standard deviation, median, average absolute error and the like as statistical characteristic parameters, and the description of each characteristic parameter is as follows:
collecting the 8 characteristic parameters of each section to complete the extraction of the current value characteristic information of the section action curve;
step four: and fusing the time characteristic information and the current value characteristic information of the equipment action to construct a high-dimensional characteristic representation data set of the equipment action current curve. Firstly, summarizing 12 items of feature information of each divided time domain section to form a multi-dimensional time domain feature vector. Then, the 8 items of feature information of each divided current value domain section are summarized to form a multi-dimensional current value domain feature vector. Finally, fusing the two multi-dimensional feature vectors to serve as a high-dimensional feature representation set of the equipment action current curve;
step five: selecting the characteristics of the high-dimensional characteristic data set by using a Fisher methodThe most representative and separable characteristics are removed, and the characteristics which are ambiguous and difficult to distinguish are removed, so that the dimension of the characteristic expression set is reduced. The specific operation is as follows: firstly, using Fisher criterion two-class feature selection method to calculate criterion function value between every typical fault mode and normal mode, d-th dimension feature F of equipment action current curve of i-th fault mode d The Fisher criterion function of (a) is:
here, the between-class variance S B,d And within class variance S W,d Are respectively defined as:
S B,d ＝(m 1,d -m i,d ) 2 i＝1,2,…n
here, m i,d And σ i,d Are respectively characteristic F d Mean and standard deviation in class i. Then, determining a feature selection standard, and adopting a 'over half selection' mode by the invention, namely taking half of the maximum value of the Fisher criterion function value of the fault mode as a standard, selecting the Fisher criterion function value of each dimension feature data if the Fisher criterion function value exceeds the standard, and discarding the Fisher criterion function value if the Fisher criterion function value is lower than the standard. Finally, combining the selected feature data together to form a feature representation set which is beneficial to distinguishing typical failure modes;
step six: and performing feature extraction on the high-dimensional feature representation data set of the equipment by using a PCA (principal component analysis) method, eliminating redundant information and further reducing the dimension of the feature representation set. The specific operation is as follows:
step 6.1 feature centralization. Subtracting the mean value of the dimension from the attribute data of each dimension of the high-dimension feature representation data set A to change the mean value of each dimension of the matrix B obtained after transformation into 0;
step 6.2, calculating a covariance matrix C of the matrix B;
6.3, calculating an eigenvalue and an eigenvector of the covariance matrix C;
step 6.4, sorting the eigenvalues in the descending order, selecting the largest k eigenvectors, and then taking the corresponding k eigenvectors as column vectors to form an eigenvector matrix;
and 6.5, projecting the sample points to the selected characteristic vectors to obtain a new k-dimensional data set. Wherein k is selected according to the ratio of the sum of the first k characteristic values to the sum of all characteristic values.
It should be noted that, in the PCA method, the class information of the samples in the training set does not need to be considered in the solving process, and the training samples are treated equally no matter which class the training samples come from. After the PCA feature space transformation, the classification is not necessarily favorably affected, and may even be adversely affected. Therefore, the k value should be selected more according to the requirement of practical problems, so that the mapped new k-dimensional data contains original data information as much as possible, and information effective for classification is lost as little as possible;
step seven: the feature representation data set is divided into a training set and a test set according to the proportion of 3. The training set data is used for SVM modeling, and the test set data is used for diagnosis accuracy verification of the fault diagnosis model. On the basis, each fault mode selects the same quantity to enter a training set and a testing set, so that the diagnosis accuracy, the false alarm rate and the false alarm rate become indexes capable of judging the effectiveness degree of the fault diagnosis model. If the samples are distributed too unevenly, the common indexes used for judging the fault diagnosis model lack significance;
step eight: on the basis of the training sample set, the GA-PSO algorithm is adopted to carry out optimization solution on the parameters of the SVM. The GA algorithm is a global random search algorithm, for a complex optimization problem, an optimized solution can be obtained only by selecting, crossing and mutating three genetic operators, the solution of the optimization problem is irrelevant to gradient information, only a target function is needed, and the processing flow of the GA is shown in figure 3; the PSO is an optimization method based on a group intelligence theory, the method has the advantages of self-learning and learning to others, an optimal solution can be found in a few iteration times, and the processing flow of the PSO is shown in figure 4. The optimization algorithm takes GA as a basic frame, carries out PSO optimization on the basis of selected excellent individuals, and aims to enhance the local search capability on the basis of keeping good global search capability. From the perspective of GA, the excellent samples are optimized continuously, so that in one iteration, excellent individuals are optimized twice by different algorithms, and the parent and the offspring are improved jointly; from the perspective of PSO, equivalently, the information of the optimal individual is maintained, suboptimal individuals are selected to perform cross and inheritance operations, and less excellent individuals perform mutation operations, so that on one hand, good memory in an iteration process is saved, the average fitness of the population is moved to a good direction, on the other hand, the diversity of the population is ensured, and the situation that the population falls into a local optimal value as soon as possible is avoided.
Step nine: the method comprises the following steps of carrying out SVM supervised learning by using parameters optimized by a GA-PSO algorithm and sample data of a training set to obtain a fault diagnosis model, wherein the specific flow is shown in FIG. 6, and the selected training set and a test set and data preprocessing in the graph are finished in the previous steps from step four to step seven. The step mainly comprises two parts of training a model and generating a fault diagnosis model, wherein the training process of the SVM is a supervised learning process taking the classification accuracy as an index, namely a process of searching an optimal classification surface. In training, the penalty function c and the kernel parameter g have decisive influence on the optimal performance of the SVM classifier, and the GA-PSO method is adopted in the step eight to optimize the two parameters, so that the method for carrying out SVM supervised learning by combining the optimized parameters and the sample data of the training set is as follows:
step 9.1, setting a known training set: t = { (x) 1 ,y 1 ),…,(x i ,y i )}∈(X×Y) l
Wherein x is i Is a feature vector, y i For corresponding attribute values, x i ∈X∈R n ，y i ∈Y＝{-1,1}，i＝1,2,…,l；
Step 9.2, selecting a kernel function g (x, x') optimized by GA-PSO and a penalty parameter C, and constructing and solving an optimization problem:
so that
Thereby obtaining an optimal solution
Step 9.3 selection of alpha * A positive component of 0<α * <, C, and calculating the threshold b based on the above * ：
Step 9.4 construct a decision function f (x):
step 9.5 outputs the class according to the value of the decision function f (x).
And obtaining a fault diagnosis model after SVM training is completed. If the equipment has a fault, the diagnosis model can position the fault through the current signal and identify the mode type of the fault. The "verification of the diagnosis result" in fig. 6 is completed in the following "step ten";
step ten: the invention uses the sample data in the test set to verify the diagnosis accuracy of the fault diagnosis model, and the invention adopts the mode of sample data injection in the test set to verify the model, obtains the diagnosis accuracy by comparing the diagnosis result and the actual result of the fault mode, and further judges whether the diagnosis model can meet the requirements.
Through the steps, a complex equipment fault diagnosis method based on feature processing can be realized, and the whole realization flow is shown in fig. 7. The method takes complex equipment as an object, and performs fusion characteristic representation of intelligent partitions on working current signals according to the fault characteristics of different fault modes of the equipment. And then screening the fusion characteristics by using a Fisher characteristic selection method, reserving the characteristics favorable for fault diagnosis, and further extracting the screened characteristics by combining a PCA method. And finally, obtaining SVM optimization parameters by using a GA-PSO optimization algorithm combining GA and PSO, and realizing a fault diagnosis classifier by using a parameter optimization SVM method to finish fault identification and analysis of the characteristic number of the working current curve of the equipment.
The invention has not been described in detail and is within the skill of the art.
The above description is only a partial embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are also included in the scope of the present invention.
Claims (11)
1. A method for diagnosing faults of complex equipment based on feature processing is characterized by comprising the following steps: the method comprises the following steps:
step 1, collecting action current signals of complex equipment in real time according to a proper sampling interval to obtain action curve data of the equipment, and analyzing a corresponding relation between an action current curve form and an equipment fault type;
step 2, performing section division on the action current curve of the equipment from the time domain angle, and extracting time characteristic information of action curve data of each section;
step 3, performing section division on the action curve of the equipment from a value domain angle, and extracting current value characteristic information of action curve data of each section;
step 4, fusing the time characteristic information and the current value characteristic information of the equipment action to construct a high-dimensional characteristic representation data set of the equipment action current curve;
step 5, performing feature selection on the high-dimensional feature representation data set by using a Fisher method, screening out features with the best representativeness and separability, and eliminating features with two distinct edges and difficult discrimination, thereby reducing the dimension of the feature representation set;
step 6, extracting the features of the high-dimensional feature expression data set by using a PCA method, eliminating redundant information and further reducing the dimension of the feature expression set;
step 7, dividing the feature representation data set to obtain SVM training sample set data and test sample set data;
step 8, optimizing and solving the parameters of the SVM by adopting a GA-PSO optimization algorithm on the basis of the training sample set;
step 9, performing SVM supervised learning by using the optimized parameters and the training set sample data to obtain a fault diagnosis model;
step 10, verifying the diagnosis accuracy of the fault diagnosis model by using the test set data;
by the steps, the complex equipment fault diagnosis method based on the feature processing can be realized, the method takes the complex equipment as an object, and firstly, fusion feature representation of intelligent partitions is carried out on working current signals according to the fault characteristics of different fault modes of the equipment; then, screening fusion features by using a Fisher feature selection method, reserving features beneficial to fault diagnosis, and further extracting the screened features by combining a PCA method; and finally, obtaining SVM optimization parameters by using a GA-PSO optimization algorithm combining GA and PSO, and realizing a fault diagnosis classifier by using a parameter optimization SVM method to finish fault identification and analysis of the characteristic number of the working current curve of the equipment.
2. The method for diagnosing the fault of the complex equipment based on the feature processing as claimed in claim 1, wherein:
the "analysis of the relationship between the operating current curve form and the type of the equipment fault" described in step 1 is performed as follows: and analyzing the normal mode and the typical fault mode of the equipment, summarizing the fault type, the occurrence phenomenon and the generation reason of each fault mode, and searching the corresponding relation between the action current curve form of the equipment and the fault type of the turnout for the subsequent fault diagnosis step.
3. The method for diagnosing the fault of the complex equipment based on the feature processing as claimed in claim 1, wherein:
in step 2, "segment division is performed on the motion current curve of the device from the time domain perspective, and the time characteristic information of the motion curve data of each segment is extracted", which is performed as follows:
the method for dividing the motion curve of the device from the time domain perspective specifically comprises the following steps: analyzing current signal data of different actions of equipment to obtain current value domain information of the actions, and correspondingly setting time division points for time domains, namely x-axes, of an action curve according to the current value domain information, so that the time domains of the action curve are divided into different sections, and each time domain section represents the duration of different actions of the equipment;
the specific method of "extracting the time characteristic information of the motion curve data of each segment" is as follows: describing physical quantity of motion curve data by using time as an independent variable, and reflecting the working state of each motion section of the equipment in a most basic and most intuitive expression mode; the method takes the maximum value, the minimum value, the average value, the variance value, the effective value, the mean square error, the peak value, the kurtosis value, the peak factor, the wave form factor, the pulse factor and the margin factor as time domain characteristic parameters, and collects the 12 characteristic parameters in each section divided by the time domain.
4. The method for diagnosing the fault of the complex equipment based on the feature processing as claimed in claim 1, wherein:
the "segment division of the motion curve of the device from the value range perspective" in step 3 is performed as follows: firstly, projecting an action current curve to a current value range, namely a y-axis, and setting the precision of a projection interval according to the acquisition precision of an equipment action current signal, wherein in order to avoid the influence of the projection interval on a divided current value range, the precision value of the projection interval is set to be consistent with the acquisition precision value; then, zero is set for the scattered current value points with the number of points less than 5 points in the projection interval, and statistics are not counted; finally, segmenting the current value range according to the current information of the action of the equipment;
the "extracting current value characteristic information of the operation curve data of each segment" described in step 3 is performed as follows: according to the current value domain segmentation analysis, extracting non-zero current areas of each section, and performing parameter representation on the areas by using statistical characteristics; the invention takes the total number, the mean value, the range, the mode, the variance, the standard deviation, the median and the average absolute error as statistical characteristic parameters, and collects the 8 characteristic parameters for each section to represent the action current characteristics of the equipment.
5. The method for diagnosing the fault of the complex equipment based on the feature processing as claimed in claim 1, wherein:
the "fusing the time characteristic information and the current value characteristic information of the device operation" described in step 4 is performed as follows: firstly, summarizing 12 items of characteristic information of each divided time domain section to form a multi-dimensional time domain characteristic vector; then, summarizing 8 items of feature information of each divided current value domain section to form a multi-dimensional current value domain feature vector; and finally, fusing the two multi-dimensional characteristic vectors to be used as a high-dimensional characteristic representation set of the equipment action current curve.
6. The method for diagnosing the fault of the complex equipment based on the feature processing as claimed in claim 1, wherein:
the Fisher method in step 5 is a classical two-class feature selection method, in which the ratio of the inter-class distance to the intra-class distance of each feature representation value is used as a Fisher criterion function, and the larger the function value is, the stronger the identification performance of the feature on classification identification is; the Fisher criterion function of a single feature is used for sequencing the features and selecting the features with strong identification performance, so that a new low-dimensional feature representation set can be formed, and the purpose of feature dimension reduction is realized;
the "feature selection of high-dimensional feature representation set data using Fisher's method" described in step 5 is performed as follows: firstly, using Fisher criterion two-class feature selection method to calculate criterion function value between every typical fault mode and normal mode, d-th dimension feature F of equipment action current curve of i-th fault mode d The Fisher criterion function of (a) is:
here, the between-class variance S B,d And within class variance S W,d Are respectively defined as:
S B,d ＝(m 1,d -m i,d ) 2 i＝1,2,…n
here, m i,d And σ i,d Are respectively characteristic F d Mean and standard deviation in class i; then, determining a feature selection standard, wherein the invention adopts a 'over half selection' mode, namely taking half of the maximum value of the Fisher criterion function value of the fault mode as a standard, selecting the Fisher criterion function value of each dimension feature data if the Fisher criterion function value exceeds the standard, and discarding the Fisher criterion function value if the Fisher criterion function value is lower than the standard; finally, the selected feature data are combined together to form a feature representation set which is beneficial to distinguishing typical failure modes.
7. The method for diagnosing the fault of the complex equipment based on the feature processing as claimed in claim 1, wherein:
the "PCA method" in step 6 is a feature extraction method, which can be used for data dimension reduction and finding effective and important elements and structures in data; the main purpose of the PCA method is to use less variables to explain most variables in the original data and ensure the minimum information loss of the original data;
the "feature extraction on high-dimensional feature representation data set using PCA method" described in step 6 is done as follows:
step 6.1, feature centralization; subtracting the mean value of the dimension from the attribute data of each dimension of the high-dimension feature representation data set A to change the mean value of each dimension of the matrix B obtained after transformation into 0;
step 6.2, calculating a covariance matrix C of the matrix B;
6.3, calculating an eigenvalue and an eigenvector of the covariance matrix C;
step 6.4, sorting the eigenvalues in the descending order, selecting the largest k eigenvectors, and then taking the corresponding k eigenvectors as column vectors to form an eigenvector matrix;
step 6.5, projecting the sample points to the selected characteristic vectors to obtain a new k-dimensional data set, wherein the selection basis of k is the ratio of the sum of the first k characteristic values to the sum of all the characteristic values;
it should be noted that, the PCA method does not need to consider the class information of the samples in the training set during the solution, and the training samples are treated equally no matter which class they come from; therefore, after the PCA feature space transformation, the classification is not necessarily influenced favorably, and even possibly influenced adversely; therefore, the k value should be selected according to the requirement of practical problems, so that the new k-dimensional data after mapping contains original data information as much as possible, and information effective for classification is lost as little as possible.
8. The method for diagnosing the fault of the complex equipment based on the feature processing as claimed in claim 1, wherein:
the "partition process for the feature representation data set" described in step 7 is performed as follows: dividing the feature representation data set into a training set and a testing set according to the proportion of 3; the training set data is used for SVM modeling, and the test set data is used for testing the recognition accuracy of the degradation state recognition model; on the basis, the same quantity of characteristic data of each fault mode is selected to enter a training set and a test set, so that the diagnosis accuracy, the false alarm rate and the false alarm rate can become indexes for judging the effectiveness degree of the fault diagnosis model; if the sample distribution is too uneven, the method is used for judging the meaning of the common indexes of the fault diagnosis model;
the "SVM" in step 7 refers to a support vector machine model, which is a classifier model provided on the basis of statistical theory and is used for pattern classification, linear and nonlinear regression analysis; the principle of the support vector machine is that a training sample is given, a classification hyperplane is established as a decision curved surface, the isolation edge between a positive case and a negative case of the sample is maximized, and the classification of the sample is completed.
9. The method for diagnosing the fault of the complex equipment based on the feature processing as claimed in claim 1, wherein:
the "optimizing solution of the parameters of the SVM using the GA-PSO optimization algorithm" described in step 8 is performed as follows: with GA as a basic framework, PSO optimization is carried out on the basis of the selected excellent individuals, so that the algorithm enhances the local search capability on the basis of keeping good global search capability; from the perspective of a genetic algorithm, the excellent samples are optimized continuously, so that in one iteration, excellent individuals are optimized by two different algorithms, and a parent and a child make 'common progress'; from the perspective of the particle swarm, the information of the optimal individual is equivalently kept, suboptimal individuals are selected to carry out cross and genetic operations, and less excellent individuals carry out mutation operations, so that on one hand, good memory in an iteration process is saved, the average fitness of the population is moved to a good direction, on the other hand, the diversity of the population is ensured, and the situation that the optimal value is introduced into the local value as soon as possible is avoided; the algorithm can keep good memorability, so that the searching speed is accelerated, meanwhile, the population information is continuously enriched, and the unidirectional flow of the information is avoided.
10. The method for diagnosing the fault of the complex equipment based on the feature processing as claimed in claim 1, wherein:
the "SVM model learning Using optimized parameters and training set sample data" described in step 9 is done as follows:
step 9.1, setting a known training set: t = { (x) 1 ,y 1 ),…,(x i ,y i )}∈(X×Y) l
Wherein x is i Is a feature vector, y i For corresponding attribute values, x i ∈X∈R n ，y i ∈Y＝{-1,1}，i＝1,2,…,l；
Step 9.2, selecting a kernel function g (x, x') optimized by GA-PSO and a penalty parameter C, and constructing and solving an optimization problem:
so that
Thereby obtaining an optimal solution
Step 9.3 selecting alpha * A positive component of 0<α * <, C, and calculating the threshold b based on the values * ：
Step 9.4 construct a decision function f (x):
step 9.5, outputting the category according to the value of the decision function f (x);
obtaining a fault diagnosis model after SVM training is completed; if the equipment has faults, the diagnosis model can locate the faults through the current signals and identify the mode types of the faults.
11. The method for diagnosing the fault of the complex equipment based on the feature processing as claimed in claim 1, wherein:
the "use of test set data to verify the diagnostic accuracy of the fault diagnosis model" described in step 10 is performed as follows: and verifying the model by a test set data injection verification mode, comparing a diagnosis result and an actual result of the fault mode to obtain a fault diagnosis accuracy rate, and further judging whether the diagnosis model can meet the requirements.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710706030.8A CN107590506B (en) | 2017-08-17 | 2017-08-17 | A kind of complex device method for diagnosing faults of feature based processing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710706030.8A CN107590506B (en) | 2017-08-17 | 2017-08-17 | A kind of complex device method for diagnosing faults of feature based processing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107590506A true CN107590506A (en) | 2018-01-16 |
| CN107590506B CN107590506B (en) | 2018-06-15 |
Family
ID=61042214
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710706030.8A Active CN107590506B (en) | 2017-08-17 | 2017-08-17 | A kind of complex device method for diagnosing faults of feature based processing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107590506B (en) |
Cited By (37)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108345284A (en) * | 2018-03-06 | 2018-07-31 | 宁波大学 | A kind of quality dependent failure detection method becoming gauge block based on two |
| CN108549741A (en) * | 2018-03-07 | 2018-09-18 | 上海交通大学 | Main shaft seals of shield machine performance health evaluating method and system |
| CN108985279A (en) * | 2018-08-28 | 2018-12-11 | 上海仁童电子科技有限公司 | The method for diagnosing faults and device of double-unit traction controller waveform |
| CN109947080A (en) * | 2019-03-21 | 2019-06-28 | 北京明略软件系统有限公司 | A kind of method, apparatus of fault diagnosis, computer storage medium and terminal |
| CN110349289A (en) * | 2019-06-21 | 2019-10-18 | 东软集团股份有限公司 | Failure prediction method, device, storage medium and electronic equipment |
| CN110363208A (en) * | 2018-04-09 | 2019-10-22 | 南瑞集团有限公司 | A kind of multi-parameter fusion piler rail deformation detection and method for early warning |
| CN110441629A (en) * | 2019-07-29 | 2019-11-12 | 交控科技股份有限公司 | Method for diagnosing faults and device based on point machine action current curve |
| CN110503004A (en) * | 2019-07-29 | 2019-11-26 | 七彩安科智慧科技有限公司 | A kind of online method of discrimination of Switching Power Supply operating status |
| WO2019232866A1 (en) * | 2018-06-08 | 2019-12-12 | 平安科技（深圳）有限公司 | Human eye model training method, human eye recognition method, apparatus, device and medium |
| CN110579709A (en) * | 2019-08-30 | 2019-12-17 | 西南交通大学 | fault diagnosis method for proton exchange membrane fuel cell for tramcar |
| CN110674991A (en) * | 2019-09-25 | 2020-01-10 | 国家能源集团谏壁发电厂 | OCSVM (online charging management system VM) -based method for detecting abnormality of primary fan of thermal power plant |
| CN110674725A (en) * | 2019-09-20 | 2020-01-10 | 电子科技大学 | Equipment signal type identification method based on multi-dimensional feature vector combination of detection signals |
| CN110717301A (en) * | 2019-09-19 | 2020-01-21 | 中国石油大学(华东) | Flow unit information classification and identification method based on support vector machine algorithm |
| TWI684882B (en) * | 2018-12-21 | 2020-02-11 | 帆宣系統科技股份有限公司 | Automatic optimization method for generating fault characteristics |
| CN110824280A (en) * | 2019-10-08 | 2020-02-21 | 西南交通大学 | Diagnosis method and visualization method of turnout health state based on feature similarity |
| CN110866502A (en) * | 2019-11-19 | 2020-03-06 | 安徽工业大学 | Fault diagnosis method based on linear discriminant analysis and particle swarm optimization support vector machine |
| CN110927243A (en) * | 2019-12-09 | 2020-03-27 | 金陵科技学院 | Cold-rolled strip steel fault diagnosis optimization method based on PCA-Kmeans algorithm |
| CN111307453A (en) * | 2020-03-20 | 2020-06-19 | 朗斯顿科技(北京)有限公司 | Transmission system fault diagnosis method based on multi-information fusion |
| CN111474475A (en) * | 2020-03-22 | 2020-07-31 | 华南理工大学 | Motor fault diagnosis system and method |
| CN111641384A (en) * | 2020-04-28 | 2020-09-08 | 特变电工新疆新能源股份有限公司 | Photovoltaic power station string fault diagnosis method, device, equipment and readable storage medium |
| CN111667355A (en) * | 2020-05-29 | 2020-09-15 | 北京睿知图远科技有限公司 | Big data analysis method and system |
| CN112015603A (en) * | 2019-05-30 | 2020-12-01 | 鸿富锦精密电子(郑州)有限公司 | User terminal hardware detection method, device, computer device and storage medium |
| CN112084910A (en) * | 2020-08-28 | 2020-12-15 | 北京旋极信息技术股份有限公司 | Fault diagnosis method and system |
| CN112633345A (en) * | 2020-12-17 | 2021-04-09 | 鄂尔多斯应用技术学院 | Method for establishing coal mining machine digital dictionary based on multi-view method |
| CN112650200A (en) * | 2020-12-25 | 2021-04-13 | 中能融安(北京)科技有限公司 | Equipment fault diagnosis method and device |
| CN113049749A (en) * | 2021-03-12 | 2021-06-29 | 中国农业大学 | Electronic nose detection system |
| CN113361324A (en) * | 2021-04-25 | 2021-09-07 | 杭州玖欣物联科技有限公司 | Motor current anomaly detection method based on lstm |
| CN113484662A (en) * | 2021-06-02 | 2021-10-08 | 青岛鼎信通讯股份有限公司 | Fault indicator starting wave recording method |
| CN113537519A (en) * | 2020-04-10 | 2021-10-22 | 北京京东乾石科技有限公司 | Method and device for identifying abnormal equipment |
| CN113641990A (en) * | 2021-06-21 | 2021-11-12 | 上海电力大学 | Intrusion detection method based on multi-innovation extended Kalman filtering |
| WO2022012083A1 (en) * | 2020-07-16 | 2022-01-20 | 中广核核电运营有限公司 | Venturi tube fault correction method and apparatus, computer device and storage medium |
| CN114021602A (en) * | 2021-10-14 | 2022-02-08 | 北京和利时系统集成有限公司 | Method and system for processing data in point switch fault diagnosis model |
| CN114021671A (en) * | 2022-01-05 | 2022-02-08 | 中国电力科学研究院有限公司 | Active power distribution network online fault processing method and system |
| CN116881700A (en) * | 2023-07-19 | 2023-10-13 | 中国人民解放军总参谋部第六十研究所 | Fault diagnosis method based on random walk and machine learning |
| CN117871994A (en) * | 2023-12-22 | 2024-04-12 | 湖南奕坤科技有限公司 | Rapid fault detection method and system for PLC (programmable logic controller) electric cabinet |
| CN117932322A (en) * | 2024-03-22 | 2024-04-26 | 山东冠县鑫恒祥面业有限公司 | Flour equipment fault diagnosis method and system |
| CN118332477A (en) * | 2024-06-13 | 2024-07-12 | 中国人民解放军海军航空大学 | Equipment fault detection model generation method based on state data analysis |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060195248A1 (en) * | 2005-02-14 | 2006-08-31 | Honeywell International, Inc. | Fault detection system and method for turbine engine fuel systems |
| CN101814114A (en) * | 2010-01-07 | 2010-08-25 | 北京航空航天大学 | Method for performing fault diagnosis by using model conversion |
| WO2012009804A1 (en) * | 2010-07-23 | 2012-01-26 | Corporation De L'ecole Polytechnique | Tool and method for fault detection of devices by condition based maintenance |
| CN104506137A (en) * | 2014-11-26 | 2015-04-08 | 北京金鸿泰科技有限公司 | Equipment fault diagnosis method and apparatus |
| CN105550700A (en) * | 2015-12-08 | 2016-05-04 | 国网山东省电力公司电力科学研究院 | Time series data cleaning method based on correlation analysis and principal component analysis |
| CN106844194A (en) * | 2016-12-21 | 2017-06-13 | 北京航空航天大学 | A kind of construction method of multi-level software fault diagnosis expert system |
-
2017
- 2017-08-17 CN CN201710706030.8A patent/CN107590506B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060195248A1 (en) * | 2005-02-14 | 2006-08-31 | Honeywell International, Inc. | Fault detection system and method for turbine engine fuel systems |
| CN101814114A (en) * | 2010-01-07 | 2010-08-25 | 北京航空航天大学 | Method for performing fault diagnosis by using model conversion |
| WO2012009804A1 (en) * | 2010-07-23 | 2012-01-26 | Corporation De L'ecole Polytechnique | Tool and method for fault detection of devices by condition based maintenance |
| CN104506137A (en) * | 2014-11-26 | 2015-04-08 | 北京金鸿泰科技有限公司 | Equipment fault diagnosis method and apparatus |
| CN105550700A (en) * | 2015-12-08 | 2016-05-04 | 国网山东省电力公司电力科学研究院 | Time series data cleaning method based on correlation analysis and principal component analysis |
| CN106844194A (en) * | 2016-12-21 | 2017-06-13 | 北京航空航天大学 | A kind of construction method of multi-level software fault diagnosis expert system |
Non-Patent Citations (5)
| Title |
|---|
| YANG SHUNKUN 等: "Model-based robustness testing for avionics-embedded software", 《中国航空学报(英文版)》 * |
| 何鑫 等: "基于FMEA／FTA的嵌入式软件故障诊断模型与应用", 《计算机测量与控制》 * |
| 卢海涛 等: "综合航空电子系统故障诊断与健康管理技术发展", 《电光与控制》 * |
| 程宇佳: "基于核方法的高速铁路道岔故障诊断", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 陈友东 等: "基于CBR的数控设备故障诊断系统知识表示", 《北京航空航天大学学报》 * |
Cited By (49)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108345284A (en) * | 2018-03-06 | 2018-07-31 | 宁波大学 | A kind of quality dependent failure detection method becoming gauge block based on two |
| CN108345284B (en) * | 2018-03-06 | 2020-06-16 | 宁波大学 | Quality-related fault detection method based on two variable blocks |
| CN108549741A (en) * | 2018-03-07 | 2018-09-18 | 上海交通大学 | Main shaft seals of shield machine performance health evaluating method and system |
| CN110363208A (en) * | 2018-04-09 | 2019-10-22 | 南瑞集团有限公司 | A kind of multi-parameter fusion piler rail deformation detection and method for early warning |
| CN110363208B (en) * | 2018-04-09 | 2022-08-12 | 南瑞集团有限公司 | Multi-parameter fusion stacker track fault detection and early warning method |
| WO2019232866A1 (en) * | 2018-06-08 | 2019-12-12 | 平安科技（深圳）有限公司 | Human eye model training method, human eye recognition method, apparatus, device and medium |
| CN108985279A (en) * | 2018-08-28 | 2018-12-11 | 上海仁童电子科技有限公司 | The method for diagnosing faults and device of double-unit traction controller waveform |
| TWI684882B (en) * | 2018-12-21 | 2020-02-11 | 帆宣系統科技股份有限公司 | Automatic optimization method for generating fault characteristics |
| CN109947080A (en) * | 2019-03-21 | 2019-06-28 | 北京明略软件系统有限公司 | A kind of method, apparatus of fault diagnosis, computer storage medium and terminal |
| CN112015603A (en) * | 2019-05-30 | 2020-12-01 | 鸿富锦精密电子(郑州)有限公司 | User terminal hardware detection method, device, computer device and storage medium |
| CN110349289A (en) * | 2019-06-21 | 2019-10-18 | 东软集团股份有限公司 | Failure prediction method, device, storage medium and electronic equipment |
| CN110503004B (en) * | 2019-07-29 | 2022-03-22 | 七彩安科智慧科技有限公司 | On-line judging method for operating state of switching power supply |
| CN110503004A (en) * | 2019-07-29 | 2019-11-26 | 七彩安科智慧科技有限公司 | A kind of online method of discrimination of Switching Power Supply operating status |
| CN110441629B (en) * | 2019-07-29 | 2022-02-15 | 交控科技股份有限公司 | Fault diagnosis method and device based on turnout switch machine operating current curve |
| CN110441629A (en) * | 2019-07-29 | 2019-11-12 | 交控科技股份有限公司 | Method for diagnosing faults and device based on point machine action current curve |
| CN110579709A (en) * | 2019-08-30 | 2019-12-17 | 西南交通大学 | fault diagnosis method for proton exchange membrane fuel cell for tramcar |
| CN110717301A (en) * | 2019-09-19 | 2020-01-21 | 中国石油大学(华东) | Flow unit information classification and identification method based on support vector machine algorithm |
| CN110674725B (en) * | 2019-09-20 | 2022-06-03 | 电子科技大学 | Equipment signal type identification method based on multi-dimensional feature vector combination of detection signals |
| CN110674725A (en) * | 2019-09-20 | 2020-01-10 | 电子科技大学 | Equipment signal type identification method based on multi-dimensional feature vector combination of detection signals |
| CN110674991A (en) * | 2019-09-25 | 2020-01-10 | 国家能源集团谏壁发电厂 | OCSVM (online charging management system VM) -based method for detecting abnormality of primary fan of thermal power plant |
| CN110824280B (en) * | 2019-10-08 | 2021-08-17 | 西南交通大学 | Diagnosis method and visualization method of turnout health state based on feature similarity |
| CN110824280A (en) * | 2019-10-08 | 2020-02-21 | 西南交通大学 | Diagnosis method and visualization method of turnout health state based on feature similarity |
| CN110866502B (en) * | 2019-11-19 | 2023-02-14 | 安徽工业大学 | Fault diagnosis method based on linear discriminant analysis and particle swarm optimization support vector machine |
| CN110866502A (en) * | 2019-11-19 | 2020-03-06 | 安徽工业大学 | Fault diagnosis method based on linear discriminant analysis and particle swarm optimization support vector machine |
| CN110927243A (en) * | 2019-12-09 | 2020-03-27 | 金陵科技学院 | Cold-rolled strip steel fault diagnosis optimization method based on PCA-Kmeans algorithm |
| CN111307453A (en) * | 2020-03-20 | 2020-06-19 | 朗斯顿科技(北京)有限公司 | Transmission system fault diagnosis method based on multi-information fusion |
| CN111307453B (en) * | 2020-03-20 | 2021-11-12 | 朗斯顿科技(北京)有限公司 | Transmission system fault diagnosis method based on multi-information fusion |
| CN111474475B (en) * | 2020-03-22 | 2021-06-08 | 华南理工大学 | Motor fault diagnosis system and method |
| CN111474475A (en) * | 2020-03-22 | 2020-07-31 | 华南理工大学 | Motor fault diagnosis system and method |
| CN113537519B (en) * | 2020-04-10 | 2024-05-24 | 北京京东乾石科技有限公司 | Method and device for identifying abnormal equipment |
| CN113537519A (en) * | 2020-04-10 | 2021-10-22 | 北京京东乾石科技有限公司 | Method and device for identifying abnormal equipment |
| CN111641384A (en) * | 2020-04-28 | 2020-09-08 | 特变电工新疆新能源股份有限公司 | Photovoltaic power station string fault diagnosis method, device, equipment and readable storage medium |
| CN111667355A (en) * | 2020-05-29 | 2020-09-15 | 北京睿知图远科技有限公司 | Big data analysis method and system |
| WO2022012083A1 (en) * | 2020-07-16 | 2022-01-20 | 中广核核电运营有限公司 | Venturi tube fault correction method and apparatus, computer device and storage medium |
| CN112084910A (en) * | 2020-08-28 | 2020-12-15 | 北京旋极信息技术股份有限公司 | Fault diagnosis method and system |
| CN112633345A (en) * | 2020-12-17 | 2021-04-09 | 鄂尔多斯应用技术学院 | Method for establishing coal mining machine digital dictionary based on multi-view method |
| CN112650200A (en) * | 2020-12-25 | 2021-04-13 | 中能融安(北京)科技有限公司 | Equipment fault diagnosis method and device |
| CN113049749A (en) * | 2021-03-12 | 2021-06-29 | 中国农业大学 | Electronic nose detection system |
| CN113361324A (en) * | 2021-04-25 | 2021-09-07 | 杭州玖欣物联科技有限公司 | Motor current anomaly detection method based on lstm |
| CN113484662A (en) * | 2021-06-02 | 2021-10-08 | 青岛鼎信通讯股份有限公司 | Fault indicator starting wave recording method |
| CN113641990A (en) * | 2021-06-21 | 2021-11-12 | 上海电力大学 | Intrusion detection method based on multi-innovation extended Kalman filtering |
| CN114021602A (en) * | 2021-10-14 | 2022-02-08 | 北京和利时系统集成有限公司 | Method and system for processing data in point switch fault diagnosis model |
| CN114021671B (en) * | 2022-01-05 | 2022-04-05 | 中国电力科学研究院有限公司 | Active power distribution network online fault processing method and system |
| CN114021671A (en) * | 2022-01-05 | 2022-02-08 | 中国电力科学研究院有限公司 | Active power distribution network online fault processing method and system |
| CN116881700A (en) * | 2023-07-19 | 2023-10-13 | 中国人民解放军总参谋部第六十研究所 | Fault diagnosis method based on random walk and machine learning |
| CN117871994A (en) * | 2023-12-22 | 2024-04-12 | 湖南奕坤科技有限公司 | Rapid fault detection method and system for PLC (programmable logic controller) electric cabinet |
| CN117932322A (en) * | 2024-03-22 | 2024-04-26 | 山东冠县鑫恒祥面业有限公司 | Flour equipment fault diagnosis method and system |
| CN117932322B (en) * | 2024-03-22 | 2024-06-04 | 山东冠县鑫恒祥面业有限公司 | Flour equipment fault diagnosis method and system |
| CN118332477A (en) * | 2024-06-13 | 2024-07-12 | 中国人民解放军海军航空大学 | Equipment fault detection model generation method based on state data analysis |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107590506B (en) | 2018-06-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107590506A (en) | A kind of complex device method for diagnosing faults of feature based processing | |
| He et al. | An end-to-end steel surface defect detection approach via fusing multiple hierarchical features | |
| CN111753985B (en) | Image deep learning model testing method and device based on neuron coverage rate | |
| CN105243276B (en) | Building seismic damage analysis method | |
| US6052481A (en) | Automatic method for scoring and clustering prototypes of handwritten stroke-based data | |
| CN107133651B (en) | The functional magnetic resonance imaging data classification method of subgraph is differentiated based on super-network | |
| CN111046931A (en) | Turnout fault diagnosis method based on random forest | |
| Alsmadi et al. | A hybrid memetic algorithm with back-propagation classifier for fish classification based on robust features extraction from PLGF and shape measurements | |
| CN106202952A (en) | A kind of Parkinson disease diagnostic method based on machine learning | |
| CN105528504A (en) | Rolling bearing health condition evaluation method based on CFOA-MKHSVM | |
| CN112364756B (en) | GIS fault classification method based on longicorn stigma and random forest | |
| CN106056059B (en) | The face identification method of multi-direction SLGS feature description and performance cloud Weighted Fusion | |
| CN106482967A (en) | A kind of Cost Sensitive Support Vector Machines locomotive wheel detecting system and method | |
| CN113376516A (en) | Medium-voltage vacuum circuit breaker operation fault self-diagnosis and early-warning method based on deep learning | |
| CN102682089A (en) | Method for data dimensionality reduction by identifying random neighbourhood embedding analyses | |
| CN110263867A (en) | A kind of rail defects and failures classification method | |
| Kamada et al. | An object detection by using adaptive structural learning of deep belief network | |
| CN107016377A (en) | Recognition of face optimization method based on SGASEN algorithms | |
| Yuan et al. | Improved random forest classification approach based on hybrid clustering selection | |
| CN114970643A (en) | High-speed electric spindle fault identification method based on UMAP dimension reduction algorithm | |
| CN106529470A (en) | Gesture recognition method based on multistage depth convolution neural network | |
| CN110097067B (en) | Weak supervision fine-grained image classification method based on layer-feed feature transformation | |
| CN113159139B (en) | Damage state diagnosis method based on improved acoustic emission density clustering | |
| CN114443338A (en) | Sparse negative sample-oriented anomaly detection method, model construction method and device | |
| Saligkaras et al. | Seeking the truth beyond the data. An unsupervised machine learning approach |
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 |