Fault diagnosis method, processor and storage medium for centrifugal compressor

Abstract

The embodiment of the present invention provides a fault diagnosis method, processor and storage medium for a centrifugal compressor, which belongs to the technical field of centrifugal compressor fault detection. By collecting the operating data of the centrifugal compressor, the operating data is input into the Gray Wolf optimization algorithm to determine the target operating data, the operating data and the target operating data are input into the Kernel Entropy Component Analysis KECA algorithm to determine the divergence measure statistic, and it is determined whether the centrifugal compressor has a fault based on the divergence measure statistic. If a fault exists, the divergence measure statistic is input into the fault diagnosis classifier to determine the type of fault. The operating data of the centrifugal compressor is extracted by the Gray Wolf optimization algorithm, the target operating data with the highest fitness is selected and then the Kernel Entropy Component Analysis is performed, which improves the accuracy of the algorithm, and fault diagnosis is performed based on the extracted target operating data, which effectively reduces the amount of computation of the KECA algorithm and improves the efficiency of fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of centrifugal compressor fault detection, and in particular to a fault diagnosis method, a processor and a storage medium for a centrifugal compressor. Background Art

[0002] A centrifugal compressor is a high-speed rotating device commonly used in chemical production. Its high-speed rotation generates centrifugal force, causing the gas to expand and flow through the impeller of the air compressor. The flow rate and pressure of the gas flowing out of the impeller are correspondingly increased, thereby compressing the air. During the use of centrifugal air compressors, some faults often occur, affecting their working performance. In order to analyze the causes of centrifugal compressor failures, it is necessary to take effective measures to troubleshoot them and improve the reliability of centrifugal compressors.

[0003] In recent years, several artificial intelligence methods have been applied to centrifugal compressor fault diagnosis, including support vector machines, genetic algorithms, particle swarm optimization, and neural networks, achieving some success. Support vector machines (SVMs) can be used to diagnose centrifugal compressor faults. Their advantages include requiring minimal training data, strong generalization capabilities, and widespread application. However, the regularization coefficient is difficult to determine, and its calculation requires significant computation, significantly reducing the efficiency of the algorithm. For data processing, the commonly used principal component analysis (PCA) method makes datasets more accessible, but it has significant limitations in eigenvalue decomposition. For example, the transformation matrix must be square, which significantly inconveniences data selection. Summary of the Invention

[0004] In view of the above-mentioned deficiencies in the prior art, an object of the embodiments of the present invention is to provide a fault diagnosis method, a processor, and a storage medium for a centrifugal compressor.

[0005] In order to achieve the above object, the present invention provides a first aspect of a fault diagnosis method for a centrifugal compressor, comprising:

[0006] Collecting operating data of centrifugal compressors;

[0007] Inputting the operating data into the Gray Wolf optimization algorithm to determine the target operating data;

[0008] Input the operating data and target operating data into the Kernel Entropy Component Analysis (KECA) algorithm to determine the divergence measure statistic;

[0009] Determine whether a centrifugal compressor is faulty based on the divergence measure statistic;

[0010] If a fault exists, the divergence measure statistics are input into the fault diagnosis classifier to determine the type of fault.

[0011] In an embodiment of the present invention, the operating data is input into the Gray Wolf Optimization Algorithm to determine the target operating data, including:

[0012] Determine the location of prey for the gray wolf optimization algorithm;

[0013] Use the running data as the individual gray wolf in the gray wolf optimization algorithm;

[0014] Determine the fitness value of the individual gray wolf based on the position of the individual gray wolf;

[0015] According to the dichotomy method, target gray wolf individuals with fitness values ​​greater than the mean fitness value are extracted from all gray wolf individuals;

[0016] The displacement of the target gray wolf individual relative to the prey position is weighted according to the local weighted regression algorithm to determine the first target position of the target gray wolf individual;

[0017] determining second target positions of other gray wolf individuals among the gray wolf individuals except the target gray wolf individual based on the first target position;

[0018] The fitness value of the gray wolf individual is re-determined based on the first target position and the second target position until the number of iterations is equal to the preset iteration threshold, and the running data corresponding to a preset number of target gray wolf individuals are extracted as target running data according to the size of the fitness value.

[0019] In an embodiment of the present invention, determining the location of prey by the gray wolf optimization algorithm includes:

[0020] Obtain the operating data standard corresponding to the centrifugal compressor;

[0021] determining first operational data in the operational data that deviates from the operational data standard;

[0022] A prey location of the gray wolf optimization algorithm is determined based on the first run data.

[0023] In an embodiment of the present invention, the operating data and the target operating data are input into the Kernel Entropy Component Analysis (KECA) algorithm to determine the divergence measure statistic, including:

[0024] Map the running data to the preset feature space through the KECA algorithm;

[0025] The target running data is used as the projection direction of the KECA algorithm to determine the KECA projection vector after normalization;

[0026] Determining a projection of the operating data on a KECA transformation axis according to the operating data mapped to a preset feature space and a KECA projection vector;

[0027] Determines a divergence measure statistic based on the projection.

[0028] In an embodiment of the present invention, determining whether a centrifugal compressor has a fault according to a divergence measure statistic includes:

[0029] Whether the centrifugal compressor is faulty is determined based on whether the divergence measure statistic exceeds a predetermined value.

[0030] In an embodiment of the present invention, the fault diagnosis classifier includes multiple support vector data description (SVDD) fault models, and the steps of establishing the fault diagnosis classifier include:

[0031] Obtain historical fault data;

[0032] Classify historical fault data according to fault type;

[0033] Use historical fault data of the same fault type as a set of data training sets;

[0034] Input the data training set into the preset SVDD to-be-trained model to generate an SVDD fault model corresponding to the data training set;

[0035] Determine the SVDD fault models corresponding to all fault types respectively;

[0036] Each SVDD fault model is sorted based on the probability of occurrence of the input fault type to generate a fault diagnosis classifier.

[0037] In an embodiment of the present invention, the divergence measure statistic is input into a fault diagnosis classifier to determine the fault type, including:

[0038] Traversing the SVDD fault models in the fault diagnosis classifier based on the divergence measure statistic to determine the target SVDD fault model;

[0039] The fault type corresponding to the target SVDD fault model is taken as the fault type.

[0040] In an embodiment of the present invention, traversing the SVDD fault models in the fault diagnosis classifier based on the divergence measure statistic to determine the target SVDD fault model includes:

[0041] Determine the radius of the hypersphere in the SVDD fault model;

[0042] Convert the divergence measure statistic to the distance from the center of the hypersphere;

[0043] The sphere radius of all SVDD fault models in the distance-based traversal fault diagnosis classifier;

[0044] The SVDD fault model corresponding to the hypersphere whose sphere radius is smaller than the distance is taken as the target SVDD fault model.

[0045] A second aspect of the present invention provides a processor configured to implement the steps of the fault diagnosis method for a centrifugal compressor as described above when executing.

[0046] A third aspect of the present invention provides a storage medium having instructions stored thereon. When the instructions are executed by a processor, the processor executes the fault diagnosis method for a centrifugal compressor as described above.

[0047] Through the above technical solution, the operating data of the centrifugal compressor is collected and input into the Gray Wolf Optimization Algorithm to determine the target operating data. The operating data and the target operating data are input into the Kernel Entropy Component Analysis (KECA) algorithm to determine the divergence measure statistic. Based on the divergence measure statistic, it is determined whether the centrifugal compressor has a fault. If a fault exists, the divergence measure statistic is input into the fault diagnosis classifier to determine the type of fault. The operating data of the centrifugal compressor is extracted through the Gray Wolf Optimization Algorithm, and the target operating data with the highest fitness is selected for Kernel Entropy Component Analysis. This improves the disadvantage of the Gray Wolf Optimization Algorithm's excessively fast convergence speed and improves the accuracy of the algorithm. Fault diagnosis is performed based on the extracted target operating data, effectively reducing the computational complexity of the KECA algorithm and improving the efficiency of fault diagnosis.

[0048] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the following detailed description, they are used to explain the present invention but do not constitute a limitation of the present invention. In the accompanying drawings:

[0050] Figure 1 is a flow chart of a fault diagnosis method for a centrifugal compressor according to an embodiment of the present invention;

[0051] Figure 2 is a schematic diagram of an application process according to an embodiment of the present invention;

[0052] Figure 3 FIG. 4 is a schematic diagram of the result of the divergence measure statistic according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following describes the specific embodiments of the present invention in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention and are not intended to limit the present invention.

[0054] Figure 1 FIG. 1 is a flow chart of a method for diagnosing a fault of a centrifugal compressor according to an embodiment of the present invention. Figure 1As shown, in an embodiment of the present invention, a fault diagnosis method for a centrifugal compressor is provided. Taking the method applied to a processor as an example, the method may include the following steps:

[0055] Step S100, collecting operating data of the centrifugal compressor;

[0056] In this embodiment, it should be noted that the operating data includes real-time data corresponding to the operation of the centrifugal compressor. This operating data can be obtained via sensors installed on the centrifugal compressor and can reflect the operating status of the centrifugal compressor. The operating data may include gas generator gas temperature, gas turbine bellows temperature, box ventilation fan gas temperature, dry gas seal heater gas temperature, centrifugal compressor outlet temperature, seal gas exhaust pressure, box ventilation fan gas pressure, drive end and non-drive end seal ring gas pressure, centrifugal compressor outlet pressure, dry gas seal differential pressure, primary vent gas pressure, secondary vent gas pressure, centrifugal compressor bearing vibration, and centrifugal compressor shaft displacement.

[0057] Step S200, inputting the operating data into the Grey Wolf Optimization Algorithm to determine target operating data;

[0058] In this embodiment, it should be noted that the Grey Wolf Optimizer (GWO) algorithm is a swarm intelligence optimization algorithm. This algorithm is an optimization search method developed based on the activity of grey wolves hunting prey. It has the characteristics of strong convergence performance, few parameters, and easy implementation. The target operating data refers to one or more groups of operating data in the operating data that are most relevant to possible faults. The operating data usually includes more than a dozen groups of data. If fault diagnosis is performed directly on these dozen groups of data, a large amount of computing resources will be consumed, and the calculation will be inaccurate and inefficient. In this embodiment, the Grey Wolf Optimizer is used to pre-process the operating data to obtain the target operating data, thereby avoiding the huge computing workload caused by substituting all the operating data for fault diagnosis, and effectively improving the efficiency and accuracy of fault diagnosis.

[0059] Step S300 , inputting the operating data and the target operating data into a Kernel Entropy Component Analysis (KECA) algorithm to determine a divergence measure statistic;

[0060] In this embodiment, it should be noted that the divergence measure statistic (CS, Cauchy-Schwarz) can measure the "distance" between two probability density functions p1(x) and p2(x), which represents the similarity between the two probability densities. In this embodiment, the value of the divergence measure statistic is used to determine whether the centrifugal compressor corresponding to the operating data has failed. The KECA (Kernel Entropy Component Analysis) algorithm maps data from a low-dimensional space to a high-dimensional feature space through kernel mapping to solve the nonlinear problem of the data. Its specific calculation method is known to technicians in the field and will not be repeated here.

[0061] Specifically, the operating data and target operating data are input into the Kernel Entropy Component Analysis (KECA) algorithm to determine the divergence measure statistics, including:

[0062] Step a, mapping the running data to the preset feature space through the KECA algorithm;

[0063] Step b, using the target running data as the projection direction of the KECA algorithm to determine the KECA projection vector after normalization;

[0064] Step c, determining the projection of the operating data on the KECA transformation axis according to the operating data mapped to the preset feature space and the KECA projection vector;

[0065] Step d: determining a divergence measure statistic based on the projection.

[0066] In this embodiment, the preset feature space is an objective function pre-constructed by the operator.

[0067] Assume that the running data is the dataset X∈R N , the preset feature space is F, and the data is mapped to the feature space F through the KECA algorithm: {φ:R N →F, x→φ(x)};

[0068] Select the target running data as the KECA projection direction μ i Components of and standardized ||μ i || 2 The KECA projection vector after =1 is: where μ i is the projection direction of KECA;

[0069] Therefore, the projection of the operating data on the KECA transformation axis can be obtained as Among them, k σ (x j ,x) is the kernel function.

[0070] Introducing the metric D of the divergence measure statistic cs , and its calculation formula is:

[0071] Where p(x) is the probability density function of the sample.

[0072] Specifically, the processor maps the operating data to a preset feature space through the KECA algorithm, and uses the target operating data as the projection direction of the KECA algorithm to determine the KECA projection vector after standardization, thereby determining the projection of the operating data on the KECA transformation axis based on the operating data mapped to the preset feature space and the KECA projection vector, so as to determine the divergence measure statistic based on the projection.

[0073] Step S400, determining whether the centrifugal compressor is faulty based on the divergence measure statistic;

[0074] In this embodiment, it should be noted that the divergence measure statistic is used to reflect whether the operating data is within a normal range, and the divergence measure statistic obtained based on the operating data is used to determine whether the centrifugal compressor has a fault.

[0075] Specifically, determining whether a centrifugal compressor has a fault according to the divergence measure statistic includes:

[0076] In step e, it is determined whether the centrifugal compressor has a fault based on whether the divergence measure statistic exceeds a predetermined value.

[0077] In this embodiment, it should be noted that the predetermined value is a value determined by the operator based on experience or reference to a large amount of information, which can reflect the standard used to characterize that the divergence measure statistic is within a normal range. If the divergence measure statistic exceeds the predetermined value, it is determined that the current divergence measure statistic exceeds the normal range, and the centrifugal compressor associated with the operating data corresponding to the divergence measure statistic is faulty; if the divergence measure statistic does not exceed the predetermined value, it is determined that the current divergence measure statistic does not exceed the normal range, and the centrifugal compressor associated with the operating data corresponding to the divergence measure statistic is in normal working condition.

[0078] Step S500: If a fault exists, the divergence measure statistic is input into a fault diagnosis classifier to determine the type of fault.

[0079] In this embodiment, it should be noted that a centrifugal compressor may have a variety of faults. Determining that a centrifugal compressor has a fault does not directly determine the type of fault it belongs to. Therefore, further determination of the type of fault is required. A fault diagnosis classifier can be used to classify the centrifugal compressor faults and determine the type of fault.

[0080] Specifically, when the processor determines that a fault currently exists, the obtained divergence measure statistic is input into the fault classifier to determine the corresponding fault type.

[0081] The above-mentioned fault diagnosis method for centrifugal compressors collects the operating data of the centrifugal compressor, inputs the operating data into the Gray Wolf optimization algorithm to determine the target operating data, inputs the operating data and the target operating data into the Kernel Entropy Component Analysis (KECA) algorithm to determine the divergence measure statistic, and determines whether the centrifugal compressor has a fault based on the divergence measure statistic. If a fault exists, the divergence measure statistic is input into the fault diagnosis classifier to determine the type of fault. The operating data of the centrifugal compressor is extracted by the Gray Wolf optimization algorithm, and the target operating data with the highest fitness is selected and then subjected to the Kernel Entropy Component Analysis. This improves the disadvantage of the Gray Wolf optimization algorithm that the convergence speed is too fast, improves the accuracy of the algorithm, and performs fault diagnosis based on the extracted target operating data, effectively reducing the computational complexity of the KECA algorithm and improving the efficiency of fault diagnosis. The fault diagnosis classifier is used to determine the type of fault, thereby enhancing the accuracy of fault classification.

[0082] In one embodiment, the operating data is input into the Gray Wolf optimization algorithm to determine target operating data, including:

[0083] Step f, determining the prey location of the gray wolf optimization algorithm;

[0084] Step g, using the running data as the gray wolf individual of the gray wolf optimization algorithm;

[0085] Step h, determining the fitness value of the individual gray wolf based on the position of the individual gray wolf;

[0086] Step i, extracting target gray wolf individuals whose fitness values ​​are greater than the mean fitness values ​​from all gray wolf individuals according to the dichotomy method;

[0087] Step j, assigning a weight to the displacement of the target gray wolf individual relative to the prey position according to a local weighted regression algorithm to determine the first target position of the target gray wolf individual;

[0088] Step k, determining second target positions of other gray wolf individuals among the gray wolf individuals except the target gray wolf individual based on the first target position;

[0089] Step 1: Re-determine the fitness value of the gray wolf individual based on the first target position and the second target position until the number of iterations is equal to the preset iteration threshold, and extract the operating data corresponding to a preset number of target gray wolf individuals as target operating data according to the size of the fitness value.

[0090] In this embodiment, it should be noted that the Gray Wolf Optimization Algorithm is a swarm intelligence optimization algorithm and an optimization search method. Its specific calculation method is known to those skilled in the art and will not be described in detail here.

[0091] In this embodiment, the input operating data will also be preprocessed, the operating data will be normalized and formed into a matrix and input into the gray wolf optimization algorithm. The initial parameters of the gray wolf algorithm will also be set, specifically including the setting of parameters such as the initialization of the gray wolf population size, stopping conditions, and the maximum number of iterations. The initial parameters can be adjusted and determined according to actual conditions.

[0092] The fitness value of each wolf is determined based on its position. A dichotomy method is used to extract target wolf individuals whose fitness values ​​are greater than the mean fitness value from all wolf individuals. The influence of the higher fitness value on the other wolf individuals is considered, and a local weighted regression algorithm is used to assign a weight to the displacement of the target wolf individual relative to the prey. After the target wolf individual's first target position is determined, the remaining wolf individuals, excluding the target wolf individual, move toward the first target position. Based on this first target position, the second target position of the remaining wolf individuals, excluding the target wolf individual, is determined. After all wolf individuals have determined their corresponding target positions, the individual's fitness value is re-determined based on the first and second target positions. The Gray Wolf Optimization Algorithm involves multiple iterations to identify the optimal solution. In this embodiment, a preset iteration threshold is used to limit the number of iterations of the Gray Wolf Optimization Algorithm. When the number of iterations reaches the preset threshold, the iterations are terminated. The fitness values ​​of all gray wolf individuals at that point are determined and ranked from highest to lowest. Starting with the target gray wolf individual with the highest fitness value, a preset number of target gray wolf individuals are identified, and the operating data corresponding to these target gray wolf individuals is used as the target operating data. The preset number of target gray wolf individuals is determined based on data review and extensive practical experience and can be adjusted based on actual needs. For example, the preset number is 5.

[0093] In this example, the running data is used as the individual gray wolves in the gray wolf optimization algorithm, and the location of the prey in the gray wolf optimization algorithm is determined. The gray wolf algorithm is then used to filter the target running data from the running data. In order to mathematically model the social hierarchy of gray wolves when designing the gray wolf optimization algorithm, the three wolves with the highest fitness are named α, β, and δ. During the hunting process, the gray wolf's behavior of surrounding prey is defined as follows:

[0094] The distance formula between the gray wolf and the prey is: D = |C·X p (t)-X(t)|; where t represents the current iteration number, X p represents the position vector of the prey, X represents the position vector of the individual gray wolf, C is the random weight coefficient, C = 2r2;

[0095] The position update formula of the gray wolf individual is: X(t+1)=X p (t)-A·D; where A is the decreasing inertia weight,

[0096] In the above formula, r1 and r2 are both random numbers between [0, 1]. Based on the divergence of mathematical modeling, a random value of |A| > 1 or less than |A| < -1 can be used to force the separation of individual wolves from their prey, thereby enhancing the global search capability of the GWO algorithm. The random weight coefficient C is a random value between [0, 2], representing the random weight of the influence of the individual wolf's location on its prey, which promotes more random behavior of individual wolves and supports global search.

[0097] Gray wolves are able to identify the location of their prey and surround them. Once the wolf identifies the location of its prey, the mathematical model for tracking its prey is described as follows:

[0098]

[0099]

[0100]

[0101] Among them, D α 、D β 、D δ Respectively represent the distances between α, β, δ and other gray wolf individuals, X α 、X β 、X δ Represent the current positions of α, β, and δ respectively, C1, C2, and C3 represent random weight coefficients, and X represents the current position of the gray wolf individual.

[0102] In a group of gray wolves, each individual gray wolf needs to move closer to the three gray wolf individuals α, β, and δ with the highest fitness. For a gray wolf individual, the three directions of displacement it needs to move are:

[0103]

[0104]

[0105]

[0106] Among them, X1, X2, and X3 represent the three directional displacements that a gray wolf individual needs to move, and A1, A2, and A3 represent the decreasing inertia weights.

[0107] In traditional gray wolf optimization algorithms, the total displacement is calculated by averaging the displacements of the three most fit individuals. This average ignores the influence of these three wolves on the prey. In this embodiment, a binary search method is used to select all gray wolf individuals with fitness values ​​greater than the average as target individuals. A local weighted regression algorithm is used to assign a weight to the displacement of each target individual relative to the prey, enabling the target individuals to move more specifically toward the prey. The number of weighted target individuals is reduced as the number of iterations increases, reducing the computational complexity while also enhancing the wolf's local search capabilities.

[0108] In this embodiment, the number of target gray wolf individuals can be further limited. When the number of iterations gradually increases, the number of gray wolf individuals with weights is reduced. The target number of gray wolf individuals can be determined by the formula: Among them, w represents the target number of gray wolves, and n represents the number of iterations.

[0109] The formula for determining the weight is:

[0110] Among them, w represents the number of target gray wolf individuals, which controls the rate at which the training points decay as the distance from the query point x increases. The larger w is, the slower the decay, and vice versa.

[0111] The calculation formula of the local weighted regression algorithm includes: Among them, i represents the i-th target gray wolf individual, Indicates the position of the current gray wolf individual at the next moment.

[0112] The displacement of the gray wolf can be obtained as follows: Among them, i represents the i-th target gray wolf individual, k1, k2, k3...k i Represents the influence weight of the target gray wolf individual on the other gray wolf individuals in the gray wolf group.

[0113] The fitness value of the gray wolf individual is calculated based on the position of the gray wolf individual. When the number of iterations is equal to the preset iteration threshold, the running data corresponding to a preset number of target gray wolf individuals are extracted as the target running data according to the size of the fitness value.

[0114] Specifically, determining the prey location of the gray wolf optimization algorithm includes:

[0115] Step f1, obtaining the operating data standard corresponding to the centrifugal compressor;

[0116] Step f2, determining first operating data in the operating data that deviates from the operating data standard;

[0117] Step f3: determining the prey location of the gray wolf optimization algorithm based on the first operating data.

[0118] In this embodiment, it should be noted that the target operating data determined by the Gray Wolf Optimization Algorithm is the operating data close to the time of the fault occurrence. The Gray Wolf Optimization Algorithm converges toward the fault as its prey. However, since it is impossible to determine whether a fault has occurred or the specific fault type when the operating data is acquired, the algorithm uses the range of abnormal data in the operating data as the convergence direction and determines the fitness value of the Gray Wolf individuals corresponding to this range. The operating data standard includes the normal reference value of the operating data, and the first operating data includes the operating data that deviates from the operating data standard. This first operating data can be one or more pieces of data.

[0119] Specifically, after obtaining the operating data standard corresponding to the centrifugal compressor, the processor determines first operating data in the operating data that deviates from the operating data standard, so as to determine the prey position of the gray wolf optimization algorithm based on the first operating data.

[0120] In one embodiment, the fault diagnosis classifier includes multiple support vector data description (SVDD) fault models, and the steps of establishing the fault diagnosis classifier include:

[0121] Step m, obtaining historical fault data;

[0122] Step n, classifying historical fault data according to fault type;

[0123] Step o, taking historical fault data belonging to the same fault type as a set of data training sets;

[0124] Step p: input the data training set into the preset SVDD to-be-trained model to generate an SVDD fault model corresponding to the data training set;

[0125] Step q, determining the SVDD fault models corresponding to all fault types respectively;

[0126] In step r, each SVDD fault model is sorted based on the probability of occurrence of the input fault type to generate a fault diagnosis classifier.

[0127] In this embodiment, it should be noted that the SVDD (support vector data description) model can project the training data into a feature space and then construct a hypersphere in the feature space. This hypersphere should be as small as possible and enclose as much training data as possible. The shape of the hypersphere is determined by the radius of the center of the hypersphere. The specific calculation method is well known to those skilled in the art and will not be detailed here.

[0128] Historical fault data refers to the fault operation data of centrifugal compressors that have been identified and have a determined fault type. The historical fault data are classified according to the fault type, and the historical fault data belonging to the same fault type are used as a group of data training sets; the preset SVDD model to be trained is an initial SVDD fault model that has not been trained by the fault data. After being trained by the fault data, the preset SVDD model to be trained becomes an SVDD fault model that can identify the fault data. The SVDD fault model generated by training the data training set corresponding to the same type of historical fault data can identify whether the input data belongs to this type of fault type. Different SVDD training models are generated for different data training sets. After generating multiple different SVDD training models for different data training sets, the various SVDD training models are sorted in a certain order to generate the fault diagnosis classifier. In this embodiment, the input is sorted according to the probability of occurrence of the fault type, which is determined by experience or by consulting relevant field documents.

[0129] In one embodiment, the divergence measure statistics are input into the fault diagnosis classifier to determine the fault type, including:

[0130] Step s, traversing the SVDD fault models in the fault diagnosis classifier based on the divergence measure statistic to determine the target SVDD fault model;

[0131] In step t, the fault type corresponding to the target SVDD fault model is used as the fault type.

[0132] In this embodiment, it should be noted that the fault diagnosis classifier includes multiple SVDD fault models, each of which can identify a fault type. After determining the divergence measure statistic, all SVDD fault models in the fault diagnosis classifier are traversed, and the SVDD fault model that matches the divergence measure statistic is determined to be the target SVDD fault model. The fault type corresponding to the target SVDD fault model is then determined as the fault type corresponding to the centrifugal compressor. The SVDD fault model that matches the divergence measure statistic is an SVDD fault model whose hypersphere radius is smaller than the divergence measure statistic.

[0133] Specifically, the SVDD fault model in the fault diagnosis classifier is traversed based on the divergence measure statistic to determine the target SVDD fault model, including:

[0134] Step s1, determining the radius of the hypersphere in the SVDD fault model;

[0135] Step s2, converting the divergence measure statistic into the distance from the center of the hypersphere;

[0136] Step s3, traversing the sphere radius of all SVDD fault models in the fault diagnosis classifier based on the distance;

[0137] Step s4: The SVDD fault model corresponding to the hypersphere whose sphere radius is smaller than the distance is used as the target SVDD fault model.

[0138] In this embodiment, the hypersphere corresponding to the SVDD fault model includes a center and a radius. The divergence measure statistic is converted to the distance from the center of the hypersphere. The sphere radii of all SVDD fault models are traversed to determine the SVDD fault model whose radius is smaller than the distance from the hypersphere center. For example, the hypersphere that satisfies the SVDD fault model conditions is obtained by solving the following optimization problem.

[0139]

[0140] St.||φ(x i )|| 2 ≤R 2 +ζ i ,ζ i ≥0,i=1,...,N

[0141] Among them, a represents the center of the sphere, R represents the radius, φ(x i ) represents the Gaussian space mapping function. The parameter C is introduced to balance the size of the hypersphere and the normal sample misclassification rate. i represents the relaxation coefficient.

[0142] The original problem is transformed into:

[0143] f(z)=||za|| 2 ≤R 2

[0144]

[0145] Among them, the CS statistic is converted into the distance from the center of the sphere a ||za||, and the radius R is the control limit of the statistic. In the linear case, the process data can be mapped to a high-dimensional space, and the Gaussian kernel function K(x i ,x j ) to replace (x i ,x j ), the statistics and control limits can be reconstructed.

[0146] Specifically, the processor determines the radius of the hypersphere in the SVDD fault model to convert the divergence measure statistic into the distance from the center of the hypersphere, and then traverses the sphere radius of all SVDD fault models in the fault diagnosis classifier based on the distance, and takes the SVDD fault model corresponding to the hypersphere with a sphere radius smaller than the distance as the target SVDD fault model.

[0147] refer to Figure 2 In the overall process of the embodiment of the present invention, the processor obtains historical fault data from the centrifugal compressor system, trains the SVDD fault model to generate a fault diagnosis classifier, obtains real-time operating data, and filters the operating data through the Gray Wolf Optimization Algorithm, inputs the data into the KECA algorithm, determines the value of the divergence measure statistic corresponding to the operating data, and then performs fault diagnosis based on the divergence measure statistic, determines the category of the corresponding SVDD fault model, and determines the fault type corresponding to the centrifugal compressor.

[0148] In one embodiment, the diagnosis of a centrifugal compressor failure is illustrated using a dry gas seal failure as an example. Under normal operating conditions, the first-stage seal bears all or most of the load, while the second-stage seal bears no or only a small portion of the pressure drop. Process gas leaking through the primary seal is directed to the flare for combustion, while the remaining minimal process gas is directed through the secondary seal to a safe area for venting. In the event of a primary seal failure, the secondary seal can provide additional safety.

[0149] The main causes of dry gas seal damage are:

[0150] (1) Low sealing gas temperature causes liquid precipitation. Liquid substances with high viscosity exist between the dynamic and static rings. Under high speed, they will rub against the end faces of the dynamic and static rings to generate high temperature and cause damage. Therefore, the sealing gas needs to be kept dry; if the electric heater fails or the heating temperature is too low, the throttling temperature of the pressure regulating valve will drop, and the sealing gas temperature may be lower than the dew point, resulting in condensate liquid, which will enter the dry gas sealing device with the sealing gas and make it ineffective; cold weather causes light hydrocarbon components and water in natural gas to precipitate, causing ice blockage in the pipeline, which will also cause the dry gas seal to shut down due to low differential pressure alarm;

[0151] (2) Failure of the filtration system leads to contamination of the dry gas seal. The depth of the gas groove on the end face of the dry gas seal dynamic ring is only a few microns. Clean gas must be ensured between the end faces to form a stable gas film between the two end faces to completely separate the sealing end faces. Therefore, the dry gas seal has high requirements for gas quality and the filtration accuracy of the filter;

[0152] (3) The isolation air (secondary sealing air) provided by the instrument air system is not supplied normally, and the oil mist contaminates the seal;

[0153] (4) Improper maintenance can cause contamination of the dry gas seal. During maintenance, the dry gas seal pipeline needs to be disassembled. If the pipeline is not well protected or improperly installed, it will also cause contamination of the dry gas seal and lead to damage. During maintenance, the disassembled pipeline should be sealed. Before installation, the entire pipeline should be purged with nitrogen. If the interface is dirty, it should be cleaned with alcohol before installation.

[0154] 500 sets of historical fault data were collected, 360 of which were used as the training set to train the SVDD fault model. The remaining 140 sets were used as the test set and fed into the improved Grey Wolf optimization algorithm. The improved Grey Wolf optimization algorithm ranked the relevance of different data types to the fault type, as shown in Table 1. The three data sets with the highest relevance weights were selected: dry gas seal heater gas temperature, compressor bearing vibration, and secondary vent gas pressure.

[0155] Table 1

[0156]

[0157]

[0158] The three data selected by the Gray Wolf optimization algorithm: dry gas seal heater gas temperature, compressor bearing vibration, and secondary vent gas pressure are input into the nuclear entropy component analysis KECA algorithm to calculate the divergence measure statistic (CS) for fault diagnosis. The results are referenced. Figure 3 ,

[0159] The training results of the KECA algorithm for fault diagnosis are shown in Table 2:

[0160] Table 2

[0161] Number of training samples Number of failures Diagnosis results Accuracy 140 16 15 93.75％

[0162] In the prior art, when processing fault data, fault analysis is usually performed directly on all equipment operating data, which results in a large amount of computation and inaccurate analysis results. In an embodiment of the present invention, the operating data of a centrifugal compressor is collected and input into the Gray Wolf optimization algorithm to determine the target operating data. The operating data and the target operating data are then input into the Kernel Entropy Component Analysis (KECA) algorithm to determine the divergence measure statistic. Based on the divergence measure statistic, it is determined whether the centrifugal compressor is faulty. The Gray Wolf optimization algorithm is used to extract the operating data of the centrifugal compressor, and the target operating data with the highest fitness is selected for kernel entropy component analysis. This improves the Gray Wolf optimization algorithm's disadvantage of fast convergence speed and improves the accuracy of the algorithm. Fault diagnosis is performed based on the extracted target operating data, effectively reducing the computational load of the KECA algorithm and improving the efficiency of fault diagnosis.

[0163] An embodiment of the present invention provides a processor configured to implement the fault diagnosis method for a centrifugal compressor as described in the above embodiment when executing.

[0164] An embodiment of the present invention provides a machine-readable storage medium having instructions stored thereon. When the instructions are executed by a processor, the processor is configured to execute the above-mentioned fault diagnosis method for a centrifugal compressor.

[0165] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0166] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0167] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0168] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0169] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0170] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0171] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0172] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0173] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A fault diagnosis method for a centrifugal compressor, characterized in that: include: Collecting operating data of centrifugal compressors; Inputting the operating data into a Gray Wolf optimization algorithm to determine target operating data; Inputting the operating data and the target operating data into a kernel entropy component analysis (KECA) algorithm to determine a divergence measure statistic; determining whether the centrifugal compressor is faulty based on the divergence measure statistic; If a fault exists, the divergence measure statistic is input into a fault diagnosis classifier to determine the type of fault; Inputting the operating data into the Gray Wolf optimization algorithm to determine target operating data includes: Obtaining an operating data standard corresponding to the centrifugal compressor; determining first operating data among the operating data that deviates from the operating data standard; determining a prey location of the gray wolf optimization algorithm based on the first operating data; Using the operating data as the gray wolf individual of the gray wolf optimization algorithm; Determining a fitness value of the gray wolf individual based on the position of the gray wolf individual; Extracting a target gray wolf individual whose fitness value is greater than the mean of the fitness values ​​from all gray wolf individuals according to a dichotomy method; Assigning a weight to the displacement of the target gray wolf individual relative to the prey position according to a local weighted regression algorithm to determine a first target position of the target gray wolf individual; Determine, based on the first target position, second target positions of other gray wolf individuals among the gray wolf individuals except the target gray wolf individual; The fitness value of the gray wolf individual is re-determined based on the first target position and the second target position until the number of iterations is equal to a preset iteration threshold, and the operating data corresponding to a preset number of target gray wolf individuals are extracted as target operating data according to the size of the fitness value.

2. The fault diagnosis method according to claim 1, characterized in that: Inputting the operating data and the target operating data into a kernel entropy component analysis (KECA) algorithm to determine a divergence measure statistic includes: Mapping the operating data to a preset feature space through the KECA algorithm; Using the target operating data as a projection direction of the KECA algorithm to determine a KECA projection vector after normalization; Determining a projection of the operating data on a KECA transformation axis according to the operating data mapped to a preset feature space and the KECA projection vector; A divergence measure statistic is determined based on the projections.

3. The fault diagnosis method according to claim 1, characterized in that: Determining whether the centrifugal compressor is faulty according to the divergence measure statistic includes: Whether the centrifugal compressor is faulty is determined based on whether the divergence measure statistic exceeds a predetermined value.

4. The fault diagnosis method according to claim 1, characterized in that: The fault diagnosis classifier includes a plurality of support vector data description (SVDD) fault models, and the steps of establishing the fault diagnosis classifier include: Obtain historical fault data; classifying the historical fault data according to fault types; Use historical fault data of the same fault type as a set of data training sets; Inputting the data training set into a preset SVDD to-be-trained model to generate an SVDD fault model corresponding to the data training set; Determine the SVDD fault models corresponding to all fault types respectively; The SVDD fault models are sorted based on the input fault type occurrence probability to generate the fault diagnosis classifier.

5. The fault diagnosis method according to claim 1, characterized in that: Inputting the divergence measure statistic into a fault diagnosis classifier to determine the fault type includes: Traversing the SVDD fault models in the fault diagnosis classifier based on the divergence measure statistic to determine a target SVDD fault model; The fault type corresponding to the target SVDD fault model is used as the fault category.

6. The fault diagnosis method according to claim 5, characterized in that: The traversing the SVDD fault models in the fault diagnosis classifier based on the divergence measure statistic to determine a target SVDD fault model includes: Determining a sphere radius of a hypersphere in the SVDD fault model; Converting the divergence measure statistic to a distance from the center of a hypersphere; Traversing the sphere radius of all SVDD fault models in the fault diagnosis classifier based on the distance; The SVDD fault model corresponding to the hypersphere whose sphere radius is smaller than the distance is used as the target SVDD fault model.

7. A processor, characterized in that: The method is configured to execute the fault diagnosis method for a centrifugal compressor according to any one of claims 1 to 6.

8. A storage medium having instructions stored thereon, characterized in that: When the instructions are executed by a processor, the processor executes the fault diagnosis method for a centrifugal compressor according to any one of claims 1 to 6.

Images