An intelligent electric control valve fault diagnosis method

Abstract

The application discloses a kind of intelligent electric control valve fault diagnosis methods, comprising: based on cloud oversampling is carried out data expansion, capture the uncertainty in fault data, construct new sample reflecting the inherent randomness of original data in cloud feature space;Optimize new sample, control the inherent random fluctuation brought by cloud oversampling;And adopt cloud membership function as fitness calculation function;Build random configuration network fault diagnosis model with parallel structure, respectively for predicting the number and category of fault;Combined with cloud model similarity calculation method, the potential correlation between single fault and concurrent fault is quantified;The output result is synthesized, and the final fault diagnosis result is obtained.The application effectively controls the random fluctuation brought by the inherent uncertainty of cloud model, thereby enhancing the adaptability to different data sets;Parallel network model is constructed to predict the number and category of fault, and the problem that single fault and concurrent fault are difficult to distinguish is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of valve fault diagnosis, and in particular to a method for diagnosing faults of intelligent electronically controlled valves. Background Art

[0002] With the significant increase in the complexity and autonomy of industrial systems, the requirements for safety and reliability have also increased accordingly. Effective and accurate fault diagnosis to avoid economic losses, production line downtime, and even casualties has become one of the most pressing challenges in the current industrial sector. Data-driven fault diagnosis methods can automatically discover potential failure modes from data and have therefore attracted widespread attention and application in the field of industrial fault diagnosis. However, establishing such high-precision fault diagnosis models requires two indispensable prerequisites: a large amount of fault data and samples covering a wide range of fault types. In real industrial scenarios, equipment typically operates normally most of the time, which significantly limits the collection of fault samples. Furthermore, due to the complexity of industrial systems, insufficient equipment informatization, and unstable testing environments, the collection of industrial fault data presents significant challenges. This results in a widespread problem of small sample sizes in practical applications, severely restricting the development and application of data-driven fault diagnosis. Furthermore, with the increasing complexity of industrial equipment, concurrent faults are becoming increasingly frequent. Concurrent faults often consist of multiple single faults occurring in parallel, inevitably exhibiting potential correlations. Distinguishing between single faults and concurrent faults based solely on sensor data is difficult, posing new challenges for accurate fault identification. Therefore, developing diagnostic methods that can accurately identify and handle concurrent faults is of great significance to improving the reliability and safety of industrial systems. Summary of the Invention

[0003] In order to solve the problems of lack of fault data and difficulty in distinguishing single faults from concurrent faults in industrial valve fault diagnosis, the present invention provides an intelligent electronically controlled valve fault diagnosis method, which not only effectively solves the data scarcity problem, but also improves the accuracy and comprehensiveness of fault diagnosis, providing an innovative solution for the intelligent maintenance of industrial valves.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A method for diagnosing faults of intelligent electronically controlled valves comprises the following steps:

[0006] S1: Data expansion based on cloud oversampling captures the uncertainty in fault data and constructs new samples in the cloud feature space that reflect the inherent randomness of the original data;

[0007] S2: Optimize the new samples to control the inherent random fluctuations caused by cloud oversampling; and use the cloud membership function as the fitness calculation function to enable a more comprehensive search in the solution space;

[0008] S3: Build a random configuration network fault diagnosis model with a parallel structure to predict the number and type of faults;

[0009] S4: Combined with the cloud model similarity calculation method, quantify the potential correlation between single faults and concurrent faults;

[0010] S5: Integrate the output results to obtain the final fault diagnosis result.

[0011] Compared with the prior art, the present invention has the following beneficial effects:

[0012] 1. This invention not only effectively captures the uncertainty in fault data but also constructs new samples in the cloud feature space that reflect the inherent randomness of the original data. It then uses a genetic algorithm to iteratively optimize this new sample, effectively controlling the random fluctuations caused by the inherent uncertainty of the cloud model and thus enhancing its adaptability to diverse data sets.

[0013] 2. In order to more accurately evaluate the quality of the solution, the present invention adopts the cloud model membership function as the fitness calculation function of the genetic algorithm, which effectively quantifies the randomness of the data and helps the genetic algorithm to perform a more comprehensive search in the complex solution space.

[0014] 3. The present invention proposes a dual-classification random configuration network fault diagnosis and constructs a parallel network model to predict the number and category of faults respectively, solving the problem of difficulty in distinguishing between single faults and concurrent faults.

[0015] 4. The present invention uses the cloud model similarity calculation method to quantify the potential correlation between single faults and concurrent faults, and combines the output decision module to make the final diagnostic decision, providing more reliable and comprehensive diagnostic results for concurrent faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 Flowchart for data expansion based on cloud oversampling;

[0018] Figure 2 Schematic diagram of the optimization process for new samples;

[0019] Figure 3 A graph of random configuration network structure with parallel structure;

[0020] Figure 4 This is a schematic diagram of cloud model similarity calculation;

[0021] Figure 5 The test confusion matrix of the diagnosis results of the embodiment. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0023] Example

[0024] A method for diagnosing faults of intelligent electronically controlled valves comprises the following steps:

[0025] S1: If Figure 1 As shown, data expansion is performed based on cloud oversampling to capture the uncertainty in the fault data and construct new samples in the cloud feature space that reflect the inherent randomness of the original data. Step S1 is as follows:

[0026] S11: There are n valve small sample failure data x i (i=1,2,...,n), using the reverse cloud generator to realize the conversion of quantitative data into qualitative concepts, the small sample fault data is converted into three numerical features: expectation (Ex), entropy (En) and hyperentropy (He), which reflect the uncertainty in the fault information, as follows:

[0027] Calculate the expected Ex from the small sample failure data of n valves:

[0028] Calculate the first-order central moment and find the entropy En:

[0029] Calculate the sample variance and find the excess entropy He:

[0030] S12: Map the numerical feature information extracted by the reverse cloud generator to the cloud feature space to generate a new sample set. This process effectively quantifies and reflects the randomness of the original fault data, enriches the diversity of the new data, and achieves the purpose of data enhancement.

[0031] S13: Generate a Gaussian random number En′ with En as the expectation and He as the standard deviation:

[0032] En′=NROM(En,He 2 );

[0033] Generate a Gaussian random number a with Ex as the expectation and En′ as the standard deviation j , then a j is a cloud droplet in the cloud feature space:

[0034] a j =NROM(Ex,En′ 2 );

[0035] Calculate the cloud membership function u corresponding to each cloud droplet j :

[0036]

[0037] S14: Repeat steps S11-S13 until N cloud droplets are generated. The expanded fault data set is X j (j=1,2,...,N).

[0038] S2: If Figure 2 As shown, the new samples are optimized to control the inherent random fluctuations caused by cloud oversampling; and the cloud membership function is used as the fitness calculation function to enable a more comprehensive search in the solution space; step S2 is as follows:

[0039] S21: Initialize the population; population initialization is a key step in the initial stage of the genetic algorithm. In the present invention, data expanded by cloud oversampling is used as the initial population. This not only quantifies the randomness inherent in the original data measurement, but also significantly improves the overall quality of the initial population and ensures its diversity, thereby laying a solid foundation for finding higher quality solutions.

[0040] S22: Determine the fitness function: Use cloud membership function As a fitness calculation function; fitness is a key indicator to measure the quality of individuals in a group and determines the efficiency and accuracy of the genetic algorithm in the optimization search process; using the cloud membership function as the fitness calculation function of the genetic algorithm effectively quantifies the randomness of the data, which helps the genetic algorithm to perform a more comprehensive search in a complex solution space;

[0041] S23: Selection: During the selection process, the genetic algorithm selects the next generation of individuals based on their fitness. Individuals with high fitness are selected and have a higher probability of passing their genes to the next generation. The roulette wheel selection method is used to select individuals based on the fitness ratio. Individual i, whose fitness corresponds to f i, then the probability of being selected is p i The calculation is as follows:

[0042]

[0043] Where u is the number of individuals in the population;

[0044] S24: Crossover: Partially exchange genes from the coding chromosomes of two parent individuals to generate two new individuals; thereby increasing the diversity of the population and improving the search ability of the genetic algorithm;

[0045] S25: Mutation: By randomly changing one or more gene values ​​on an individual chromosome; used to introduce new genetic diversity during the iterative process of the genetic algorithm; mutations usually occur with a low probability to avoid excessively destroying the excellent gene combinations established through crossover and selection operations.

[0046] S3: If Figure 3 As shown, a random configuration network fault diagnosis model with a parallel structure is constructed to predict the number and type of faults respectively; step S3 is as follows:

[0047] A random configuration network fault diagnosis model with a parallel structure, namely a counting-random configuration neural network and a label-random configuration neural network, is constructed to predict the number of faults and their categories respectively. During the model training process, 70% of the data is randomly selected to construct a training set, and the remaining 30% is used to form a test set. The output J of the counting-random configuration neural network model is obtained. C (X) and label - the output of the randomly configured neural network model f L (X), the specific steps are as follows:

[0048] The counting-random configuration neural network model is used to predict the number of faults in three categories: no fault, single fault, and concurrent faults. The specific network construction is as follows:

[0049] S31: Set the maximum number of random configurations T max =50, the maximum number of hidden layer nodes L 1max =20, initial residual e0, tolerance error ε=0.1, and random parameter configuration range [-λ max ,λ max ] and step size Δλ = 10;

[0050] S32: Random configuration T max Secondary input weight ω L and bias b L ;

[0051] S33: Determine candidate nodes under the constraints of the supervision mechanism. If the requirements of the supervision mechanism are not met, relax the contraction parameter r and return to the previous step.

[0052] S34: Select the best parameters from all candidate nodes At the same time, the output matrix of the hidden layer is constructed The output weights of the counting network model are:

[0053]

[0054] in, is the hidden layer output matrix, is the hidden node output, H L1 The generalized inverse of Y C is the output matrix after the sample label is one-hot encoded. The output of the counting network model is:

[0055] f C (X) = h C (X)·R C

[0056] S35: The final prediction result of the number of faults is:

[0057]

[0058] S36: The construction of the label-random configuration neural network model is consistent with the counting-random configuration neural network model, and its output weight is:

[0059]

[0060] The output of the label network model is then:

[0061] f L (X) = h L (X)·β L .

[0062] S4: As Figure 4 As shown, the cloud model similarity calculation method is combined to quantify the potential correlation between a single fault and concurrent faults; step S4 is as follows:

[0063] This paper uses a cloud model similarity calculation method to accurately quantify the intrinsic connection between single faults and concurrent faults. Drawing on the principle of calculating triangle area, the similarity between two cloud models is evaluated by calculating the overlapping area. Let C1 and C2 be the two cloud models corresponding to a single fault and concurrent faults, respectively, with Ex1, En1, He1 and Ex2, En2, He2 being the corresponding cloud feature values. The intersection points of the two cloud models are p1 and p2, respectively. The specific steps for calculating cloud similarity are as follows:

[0064] S41: Calculate the boundary values inf of C1 and C2 based on the 3En criterion * , inf ** , sup * and sup ** ;

[0065] inf * = Ex * -3En *

[0066] inf ** = Ex ** -3En **

[0067] sup * = Ex * +3En *

[0068] sup ** = Ex ** +3En **

[0069] S42: Calculate the areas S1 and S2 of the triangles formed by C1 and C2 and the coordinate axes, respectively, according to the boundary values;

[0070]

[0071] S43: Calculate the overlapping area S3 of C1 and C2;

[0072]

[0073] S44: Calculate the similarity

[0074] S5: Integrate the output results to obtain the final fault diagnosis result; step S5 is specifically as follows:

[0075] S51: Receive the output matrix f L (X) generated by the label-random configuration neural network model, and combine it with the similarity matrix S OD , introduce the integration coefficient ρ, and optimize the weight distribution of the two matrices to find an optimal solution to make the final diagnosis result optimal. The final output matrix is:

[0076] f out (X) = ρf L (X) + (1-ρ)S OD

[0077] S52: After integrating the diagnosis results of the label-random configuration neural network model, select the output under the decision function to make f out(X) The largest top J labels are used as the final output of fault diagnosis, and the output decision function is:

[0078]

[0079] like Figure 5 As shown in the figure, the present invention selected four types of single faults, including a fault-free state, and three types of concurrent faults as case studies for verification. The experimental results, visualized through a confusion matrix, confirmed the model's high accuracy in fault diagnosis, achieving a final diagnostic accuracy of 95.42%.

Claims

1. A method for diagnosing faults of intelligent electronically controlled valves, characterized in that: The steps include: S1: Data expansion based on cloud oversampling captures the uncertainty in fault data and constructs new samples in the cloud feature space that reflect the inherent randomness of the original data; S2: Optimize the new samples to control the inherent random fluctuations caused by cloud oversampling; and use the cloud membership function as the fitness calculation function to enable a more comprehensive search in the solution space; S3: Build a random configuration network fault diagnosis model with a parallel structure to predict the number and type of faults; S4: Combined with the cloud model similarity calculation method, quantify the potential correlation between single faults and concurrent faults; S5: Integrate the output results to obtain the final fault diagnosis result.

2. The intelligent electronically controlled valve fault diagnosis method according to claim 1, characterized in that: Step S1 is specifically as follows: S11: There are n valve small sample failure data x i (i=1,2,...,n), using the reverse cloud generator to realize the conversion of quantitative data into qualitative concepts, the small sample fault data is converted into three numerical features: expectation (Ex), entropy (En) and hyperentropy (He), which reflect the uncertainty in the fault information, as follows: Calculate the expected Ex from the small sample failure data of n valves: Calculate the first-order central moment and find the entropy En: Calculate the sample variance and find the excess entropy He: S12: Map the numerical feature information extracted by the reverse cloud generator to the cloud feature space to generate a new sample set; S13: Generate a Gaussian random number En′ with En as the expectation and He as the standard deviation: En′=NROM(En,He 2 , Generate a Gaussian random number a with Ex as the expectation and En′ as the standard deviation j , then a j is a cloud droplet in the cloud feature space: yes j =NROM(Ex,En′ 2 ); Calculate the cloud membership function u corresponding to each cloud droplet j : S14: Repeat steps S11-S13 until N cloud droplets are generated. The expanded fault data set is X j (j=1,2,...,N).

3. The intelligent electronically controlled valve fault diagnosis method according to claim 1, characterized in that: Step S2 is specifically as follows: S21: Initialize the population; S22: Determine the fitness function: Use cloud membership function As a fitness calculation function; S23: Selection: Roulette wheel selection method is used to select according to the fitness ratio; Individual i, whose fitness corresponds to f i , then the probability of being selected is p i The calculation is as follows: Where u is the number of individuals in the population; S24: Crossover: Partial exchange of genes from the coding chromosomes of two parent individuals to generate two new individuals; S25: Mutation: By randomly changing the value of one or more genes on an individual chromosome.

4. The intelligent electronically controlled valve fault diagnosis method according to claim 1, characterized in that: Step S3 is as follows: A random configuration network fault diagnosis model with a parallel structure, namely a counting-random configuration neural network and a label-random configuration neural network, is constructed to predict the number of faults and their categories respectively. During the model training process, 70% of the data is randomly selected to construct a training set, and the remaining 30% is used to form a test set. The output J of the counting-random configuration neural network model is obtained. C (X) and label - the output of the randomly configured neural network model f L (X), the specific steps are as follows: The counting-random configuration neural network model is used to predict the number of faults in three categories: no fault, single fault, and concurrent faults. The specific network construction is as follows: S31: Set the maximum number of random configurations T max =50, the maximum number of hidden layer nodes L 1max =20, initial residual e0, tolerance error ε=0.1, and random parameter configuration range [-λ max ,λ max ] and step size Δλ = 10; S32: Random configuration T max Secondary input weight ω L and bias b L ; S33: Determine candidate nodes under the constraints of the supervision mechanism. If the requirements of the supervision mechanism are not met, relax the contraction parameter r and return to the previous step. S34: Select the best parameters from all candidate nodes At the same time, the output matrix of the hidden layer is constructed The output weights of the counting network model are: in, is the hidden layer output matrix, is the hidden node output, for The generalized inverse of Y C is the output matrix after the sample label is one-hot encoded. The output of the counting network model is: f C (X)=h C (X)·R C S35: The final prediction result of the number of faults is: S36: The construction of the label-random configuration neural network model is consistent with the counting-random configuration neural network model, and its output weight is: The output of the label network model is then: f L (X)=h L (X)·β L 。 5. The intelligent electronically controlled valve fault diagnosis method according to claim 1, characterized in that: Step S4 is specifically as follows: Assume that C1 and C2 are two cloud models corresponding to a single fault and concurrent faults, respectively, where Ex1, En1, He1 and Ex2, En2, He2 are the corresponding cloud feature values; the intersection points of the two cloud models are p1 and p2 respectively. The specific steps for calculating cloud similarity are as follows: S41: Calculate the boundary values ​​of C1 and C2 based on the 3En criterion inf * 、inf ** 、sup * and sup ** ; inf * =Ex * -3En * inf ** =Ex ** -3En ** sup * =Ex * +3En * sup ** =Ex ** +3En ** S42: Calculate the areas S1 and S2 of the triangles formed by C1 and C2 and the coordinate axes respectively according to the boundary values; S43: Calculate the overlapping area S3 of C1 and C2; S44: Calculate similarity 6. The intelligent electronically controlled valve fault diagnosis method according to claim 1, characterized in that: Step S5 is specifically as follows: S51: Receive label-randomly configure the output matrix f generated by the neural network model L (X), and the similarity matrix S OD Combined with the integration coefficient ρ, the weights of the two matrices are optimized and distributed to find an optimal solution so that the final diagnosis result reaches the best state. The final output matrix is: f out (X)=ρf L (X)+(1-p)S OD S52: After comprehensive counting-random configuration of the neural network model’s diagnostic results, select the decision function that can output f out (X) The largest top J labels are used as the final output of fault diagnosis, and the output decision function is:

Images