Fault detection, diagnosis and verification imbalance sample generation method

Abstract

The embodiment of the invention provides a fault detection, diagnosis and verification imbalance sample generation method, and the method comprises the steps: screening initial samples stored in a database according to the project demands of a target project, and obtaining an intermediate sample related to the target project; sampling the intermediate samples according to a preset sample size to generate at least two sample sets; obtaining a mapping relationship between the fault mode related to the target project and the project parameter value of the target project, and determining a fault mode equivalent set corresponding to each sample set according to the mapping relationship; and determining a sufficiency metric value of the corresponding sample set according to the fault mode equivalent set, and determining a target sample related to the target item according to the sufficiency metric value.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of computer technology, and in particular to a method for fault detection, diagnosis and verification of imbalanced sample occurrence. Background Technology

[0002] Fault detection and diagnosis models can provide advance estimates of equipment health and defects based on various information data, enabling timely measures to be taken before catastrophic damage or failure occurs. In addition, fault detection and diagnosis models can also guide the formulation of equipment maintenance plans, thereby extending the lifespan of equipment.

[0003] When using fault detection and diagnosis models to design or apply fault diagnosis and prediction technologies for typical fault modes in a system, the challenge lies in addressing a large number of diverse system fault diagnosis and prediction needs. Different systems typically exhibit varying diagnostic and prediction capabilities, and these capabilities also differ significantly under different engineering conditions, including varying operating conditions, fault modes, and data conditions. Therefore, quickly determining suitable fault diagnosis and prediction samples based on existing engineering information and user requirements is extremely difficult. In other words, how to select and obtain suitable verification samples to ensure the actual performance of the fault detection and diagnosis model becomes a pressing issue. Summary of the Invention

[0004] In view of this, embodiments of this specification provide a method for fault detection, diagnosis, and verification of imbalanced sample generation. One or more embodiments of this specification also relate to a fault detection, diagnosis, and verification apparatus for imbalanced sample generation, a computing device, a computer-readable storage medium, and a computer program, to address the technical deficiencies existing in the prior art.

[0005] According to a first aspect of the embodiments of this specification, a method for fault detection, diagnosis, and verification of imbalanced sample occurrence is provided, comprising: Based on the project requirements of the target project, the initial samples stored in the database are filtered to obtain intermediate samples related to the target project; The intermediate samples are sampled according to a preset sample size to generate at least two sample sets; Obtain the mapping relationship between the failure modes related to the target project and the project parameter values ​​of the target project, and determine the equivalent set of failure modes corresponding to each sample set based on the mapping relationship; Based on the fault mode equivalence set, a sufficiency metric is determined for the corresponding sample set, and target samples related to the target project are determined based on the sufficiency metric.

[0006] In one optional implementation, the step of determining a sufficiency metric for the corresponding sample set based on the failure mode equivalence set, and determining target samples related to the target project based on the sufficiency metric, includes: Determine the number of fault modes included in the fault mode equivalence set; Determine the ratio between the number of failure modes and the total number of failure modes related to the target project, and use the ratio as a sufficiency measure of the sample set corresponding to the equivalent set of failure modes; Determine the target sample set with the largest sufficiency metric value in each sample set, and identify the samples contained in the target sample set as target samples related to the target project.

[0007] In one alternative implementation, after obtaining intermediate samples related to the target project, the method further includes: Determine the target sample size for the target samples related to the target project based on the project requirements; If the sample size of the intermediate samples meets the requirements based on the target sample size, the step of sampling the intermediate samples according to the preset sample size to generate at least two sample sets is performed.

[0008] In an optional implementation, the fault detection, diagnosis, and verification method for imbalanced sample occurrence further includes: If it is determined that the sample size of the intermediate sample is insufficient, fault data generated by performing forward diffusion noise injection on the initial sample is obtained, and the target component variable corresponding to the fault data is determined. The fault data and the target condition variable are input into the trained sample generation model for processing to generate intermediate samples to be screened. The sample generation model includes a generator network, a discriminator network, and a degradation feature reconstruction network. Accordingly, the step of sampling the intermediate samples according to a preset sample size to generate at least two sample sets includes: The intermediate samples and the intermediate samples to be screened are sampled according to the preset sample size to generate at least two sample sets.

[0009] In one optional implementation, before sampling the intermediate samples according to a preset sample size, the method further includes: Determine the failure mode corresponding to each intermediate sample, and extract the failure modes that meet the consistency index from each failure mode to obtain the set of failure modes to be verified. Determine the weight corresponding to each fault mode in the set of fault modes to be verified, and obtain intermediate sub-samples based on the weights from the intermediate samples; Accordingly, the sampling of the intermediate samples according to the preset sample size includes: The intermediate subsamples are sampled according to the preset sample size.

[0010] In one optional implementation, the step of extracting fault modes that meet the consistency index from each fault mode to obtain a set of fault modes to be verified includes: Divide each fault mode into sets to generate at least two initial fault mode sets; Determine the hierarchical fault data corresponding to each fault mode in the target initial fault mode set, wherein each hierarchical fault data consists of fault modes at different levels, and the target initial fault mode set is each of the at least two initial fault mode sets; A judgment matrix is ​​constructed based on the hierarchical fault data, and the fault modes that meet the consistency index in each fault mode contained in the hierarchical fault data are extracted according to the judgment matrix to obtain a set of fault modes to be verified.

[0011] In one optional implementation, sampling the intermediate samples according to a preset sample size includes: The intermediate samples are sampled using a random sampling algorithm that takes into account the failure rate and according to a preset sample size.

[0012] According to a second aspect of the embodiments of this specification, a fault detection, diagnosis, and verification apparatus for generating imbalanced samples is provided, comprising: The filtering module is configured to filter the initial samples stored in the database according to the project requirements of the target project to obtain intermediate samples related to the target project. The sampling module is configured to sample the intermediate samples according to a preset sample size to generate at least two sample sets; The acquisition module is configured to acquire the mapping relationship between the fault modes related to the target project and the project parameter values ​​of the target project, and determine the fault mode equivalent set corresponding to each sample set based on the mapping relationship; The determination module is configured to determine the sufficiency metric of the corresponding sample set based on the fault mode equivalence set, and to determine the target samples related to the target project based on the sufficiency metric.

[0013] According to a third aspect of the embodiments of this specification, a computing device is provided, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement any of the steps of the fault detection, diagnosis and verification method for the occurrence of imbalanced samples.

[0014] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores computer-executable instructions, which, when executed by a processor, implement the steps of any of the fault detection, diagnosis, and verification methods for the occurrence of imbalanced samples.

[0015] According to a fifth aspect of the embodiments of this specification, a computer program is provided, wherein when the computer program is executed in a computer, the computer is instructed to perform the steps of the above-described fault detection, diagnosis, and verification method for the occurrence of imbalanced samples.

[0016] This embodiment of the specification filters initial samples stored in a database according to the project requirements of the target project to obtain intermediate samples related to the target project; it then samples the intermediate samples according to a preset sample size to generate at least two sample sets; it obtains the mapping relationship between the fault modes related to the target project and the project parameter values ​​of the target project, and determines the fault mode equivalence set corresponding to each sample set based on the mapping relationship; it determines the sufficiency metric of the corresponding sample set based on the fault mode equivalence set, and determines the target samples related to the target project based on the sufficiency metric. By filtering samples according to the project requirements of different target projects to obtain qualified target samples, it is beneficial to ensure the accuracy and applicability of the generated samples, and also to ensure the accuracy of the detection results obtained by fault detection and diagnosis through the target samples. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for detecting, diagnosing, and verifying imbalanced sample occurrence, provided in one embodiment of this specification. Figure 2 This is a schematic diagram of a fault detection, diagnosis and verification system for unbalanced sample generation provided in one embodiment of this specification; Figure 3 This is a schematic diagram of the structure of a fault detection, diagnosis and verification unbalanced sample generation device provided in one embodiment of this specification; Figure 4 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0018] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0019] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0020] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0021] This specification provides a method for fault detection, diagnosis, and verification of imbalanced sample generation. It also relates to a device for fault detection, diagnosis, and verification of imbalanced sample generation, a computing device, a computer-readable storage medium, and a computer program, which will be described in detail in the following embodiments.

[0022] Figure 1 A flowchart is shown for a method for detecting, diagnosing, and verifying imbalanced sample occurrence according to an embodiment of this specification, specifically including the following steps.

[0023] Step 102: According to the project requirements of the target project, filter the initial samples stored in the database to obtain intermediate samples related to the target project.

[0024] Specifically, the target project can be a verification system for mechanical and electronic products in multiple fields. The project requirements can be the number and type of target samples needed to detect and diagnose typical failure modes of the target project.

[0025] Based on this, in the process of screening and obtaining the required target samples, the initial samples stored in the database can be screened according to the project requirements of the target project to obtain intermediate samples related to the target project.

[0026] In practical applications, an in-depth analysis of the initial samples stored in the database can be performed first to extract useful information and features. Specifically, when determining the samples needed for detecting and diagnosing typical failure modes in verification systems for multi-domain mechanical and electronic products, the basis for sample selection must first be clarified. Based on a summary of the basic process of verification system failure detection and diagnosis and the theoretical methods and experimental operations involved in each step, an in-depth analysis of the sample library for detecting and diagnosing typical failure modes of verification systems was conducted. A set of sample attribute information was selected to adequately characterize the applicability of comprehensive test samples for various verification systems, serving as the basis for determining the target samples.

[0027] In the embodiments of this specification, the data imbalance problem refers to a classification problem in which the number of samples in one or more classes is far greater than that in other classes. Data imbalance leads to a decline in the performance of classification algorithms, especially for classes with fewer samples, where the algorithm struggles to identify them correctly. Therefore, handling imbalanced data has become an important research direction in classification tasks.

[0028] Based on this, the specific sample meaning analysis in the embodiments of this specification includes: (1) Unbalance parameter.

[0029] These parameters are used to quantify the degree of imbalance among samples of different classes in a dataset. Common imbalance parameters include: Imbalance ratio: The ratio of the number of normal samples to the number of faulty samples.

[0030] Gini index: A statistic that measures the degree of imbalance in a dataset.

[0031] Matthews correlation coefficient: Considering true positives, false positives, true negatives, and false negatives, it is used to evaluate classifier performance.

[0032] (2) The normal data and the fault data are unbalanced.

[0033] This refers to a situation in which the number of samples in a dataset under normal operating conditions is far greater than the number of samples under fault conditions. For example, in equipment fault diagnosis, there is a large amount of detection data under normal conditions, while the detection data under various fault conditions is relatively scarce, resulting in a sample imbalance problem. When samples are imbalanced, the diagnostic results tend to favor the state with more sample data, leading to misdiagnosis of the equipment status.

[0034] (3) Unbalanced operating conditions.

[0035] This involves the distribution of samples under different operating conditions, such as temperature, humidity, and pressure. Imbalanced operating conditions may result in an overabundance of data under some conditions and an underabundance of data under others, affecting the model's adaptability to various operating conditions.

[0036] (4) Fault mode imbalance.

[0037] This refers to an imbalance in the distribution of different failure modes among the samples. For example, some failure modes may be more common and therefore have more samples in the dataset, while other failure modes have fewer. This imbalance can cause the model to degrade in performance when predicting less common failure modes.

[0038] After constructing the sample attribute information set of the fault detection and diagnosis samples of the verification system, the initial samples stored in the database can be screened based on this information set to obtain samples that meet the conditions. These samples can be used as intermediate samples. Alternatively, to ensure the accuracy of the screening results, the samples can be further screened to obtain intermediate samples.

[0039] Specifically, the aforementioned samples will be further screened, which can be done based on factors such as data completeness, data relevance, data timeliness, data anomalies, and data consistency.

[0040] Data integrity involves checking whether the data is complete and free of missing or corrupted values. For example, if a fault sample lacks crucial sensor readings, it may be unsuitable for fault diagnosis. Data relevance assesses the correlation between the data and the type of fault. For example, if a database contains temperature readings of a machine during normal operation, this data may be useful for diagnosing overheating faults. Data timeliness determines whether the data is up-to-date and relevant to current system configuration and operating conditions. For example, older fault data may not be applicable to newer versions of software or hardware. Data anomaly identification identifies outliers in the data that may indicate a fault. For example, if a sensor typically returns values ​​between 0 and 100 but suddenly shows readings of -10 or 110, this could be a sign of a fault. Data consistency checks whether the data is consistent with other data sources. For example, if two different sensors measure the same physical quantity but their readings differ significantly, this indicates a problem with one of the sensors.

[0041] In one alternative implementation, after obtaining intermediate samples related to the target project, the method further includes: Determine the target sample size for the target samples related to the target project based on the project requirements; If the sample size of the intermediate samples meets the requirements based on the target sample size, the step of sampling the intermediate samples according to the preset sample size to generate at least two sample sets is performed.

[0042] Specifically, after obtaining intermediate samples through the aforementioned screening method, it is necessary to determine whether the number of intermediate samples meets the requirements. If not, a certain number of samples need to be generated so that the sum of the number of generated samples and the number of intermediate samples meets the requirements.

[0043] In the embodiments of this specification, based on the intensional analysis, a sample size determination algorithm is used to evaluate whether the sample size of the intermediate samples meets the requirements. The algorithm considers the following factors: Fault type diversity: Ensure sufficient samples for each fault type; Feature coverage: Ensure that the selected samples can cover all changes in key features; Model complexity: More complex models may require more data to avoid overfitting.

[0044] Based on this, the embodiments of this specification employ a sample size determination algorithm based on sequential sampling inspection theory to evaluate whether the sample size of the intermediate sample meets the requirements. The specific steps are as follows: Step 1: Based on the system-level index constraints and risk constraints of the fault detection and diagnosis system, calculate the sequential verification chart parameters s, h1, and h2, and draw the relative accuracy sequential verification chart.

[0045] Step 2: Randomly select fault modes Fault samples.

[0046] Step 3: Based on fault mode F i Based on the indicator requirements and risk requirements, determine the failure mode F. i Number of samples N in the degradation process i1 When fault mode F cannot be obtained i When determining the required index and risk values, the sample size N for the degradation process can be determined using fault diagnosis system-level index constraints and risk constraints. i1 .

[0047] Step 4: For fault mode F i The relative accuracy of the obtained sample sequence is denoted as ,remember Fault mode F i The j-th relative accuracy sample value: Step 5: Let ,calculate: Test sample points Plotted on the sequential test chart.

[0048] Step Six: Based on the test sample points Given its position in the sequential test graph, make a decision: when When it is in the acceptor domain, the conclusion is "accepted"; when When it is located in the rejection region, the conclusion is "reject"; when If the test is located in the continued testing area, the test will continue.

[0049] Step 7: If you need to continue the experiment after completing Step 6, repeat Steps 2 through 6 until you get a "receive" or "reject" conclusion.

[0050] Step 104: Sampling of the intermediate samples according to the preset sample size to generate at least two sample sets.

[0051] In one optional implementation, sampling the intermediate samples according to a preset sample size includes: The intermediate samples are sampled using a random sampling algorithm that takes into account the failure rate and according to a preset sample size.

[0052] Specifically, sampling can be performed using a random sampling algorithm that takes into account the failure rate. A failure rate can be set first, and random sampling can be performed based on the failure rate. For example, if the failure rate is set to 0.8, then 80% of the failure data and 20% of the normal data can be randomly selected from the intermediate sample.

[0053] After extraction is complete, the extracted samples can be divided into at least two sample sets.

[0054] In one optional implementation, if it is determined that the sample size of the intermediate sample is insufficient, fault data generated by performing forward diffusion noise injection on the initial sample is obtained, and the target component variable corresponding to the fault data is determined. The fault data and the target condition variable are input into the trained sample generation model for processing to generate intermediate samples to be screened. The sample generation model includes a generator network, a discriminator network, and a degradation feature reconstruction network. Accordingly, the step of sampling the intermediate samples according to a preset sample size to generate at least two sample sets includes: The intermediate samples and the intermediate samples to be screened are sampled according to the preset sample size to generate at least two sample sets.

[0055] Specifically, after obtaining intermediate samples through the aforementioned screening process, if it is determined that the number of intermediate samples is insufficient, a certain number of samples need to be generated so that the sum of the number of generated samples and the number of intermediate samples meets the requirement. Specifically, this involves acquiring fault data generated by forward diffusion noise injection onto the initial samples, determining the target condition variable corresponding to the fault data, and inputting the fault data and the target condition variable into a trained sample generation model for processing to generate intermediate samples to be screened. The sample generation model includes a generator network, a discriminator network, and a degradation feature reconstruction network. Random sampling is performed on the multiple intermediate samples and the intermediate samples to be screened according to the target sample size to generate target samples related to the target project. These target samples can be used for fault detection and diagnosis of the target project.

[0056] In the embodiments described in this specification, the sample generation model includes a generator network, a discriminator network, and a degenerate feature reconstruction network.

[0057] Specifically, in the data preparation stage, it is mainly used to extract conditional variables of known failure sample degradation trends and to construct a training sample set composed of failure samples and conditional variables.

[0058] In the degradation trend condition variable extraction stage, the fault samples with known fault levels, i.e., historical fault data, are sent to the degradation trend condition variable extraction module. From each fault sample, k degradation features that can characterize the degree of equipment fault are extracted, i.e., k data carrying fault information are extracted and spliced ​​together to form the degradation trend condition variable, i.e., the initial condition variable.

[0059] As mentioned earlier, the initial condition variable can be expressed as: in, It can be composed of time-domain / frequency-domain / time-frequency-domain features (such as RMS, kurtosis, dominant frequency energy, envelope spectrum amplitude, etc.) and used to characterize the degree of degradation, the direction of degradation and the rate of change.

[0060] In practical applications, the extracted k features that characterize the degree of equipment failure can be the mean, variance, and standard deviation of the equipment failure degree. Taking k = 3, failure samples 1 and 2, and the extracted features being the mean, variance, and standard deviation as an example, the mean, variance, and standard deviation of the equipment failure degree are extracted from failure samples 1 and 2 respectively. The three features extracted from failure sample 1 are concatenated to generate the initial condition variable for failure sample 1; the three features extracted from failure sample 2 are also concatenated to generate the initial condition variable for failure sample 2.

[0061] After obtaining the initial condition variables, forward diffusion noise injection is performed on the target historical fault data according to the preset diffusion step number to generate target fault data: Assuming the target historical fault data (initial sample) is The forward diffusion noise is injected according to the preset diffusion step number to obtain... : in, , These are the diffusion scheduling parameters (determined by a preset noise schedule).

[0062] Let the number of diffusion steps be ,but .

[0063] The target fault data, initial condition variables, and preset diffusion steps are then concatenated to generate training samples.

[0064] In practical applications, for a sample set with a known fault severity (historical fault data, also known as fault samples) After the degradation trend condition variable extraction module, feature data carrying fault information is calculated for each sample. Then, each historical fault data is compared with its respective... The fault sample pairs are combined to form the training dataset. ,in k were extracted by different methods The conditional variable composed of the values ​​(second degenerate characteristic). .

[0065] During the model training phase, forward propagation is first performed using fault samples generated in the data preparation phase to calculate the generated samples, interpolated samples, and the discrimination results of the discriminator network. The discriminator network's loss is calculated based on its discrimination results for fault samples and generated samples. The generator network's loss is calculated based on its discrimination results for generated samples. The conditional mutual information maximization loss between the generator network and the Q network is calculated based on the reconstruction result of the degradation trend conditional variable contained in the generated samples by the degradation feature reconstruction network Q (i.e., the first degradation feature). Based on these loss terms, the discriminator network, generator network, and Q network are sequentially optimized and updated until the loss converges, completing the model training and obtaining the trained sample generation model.

[0066] Specifically, after generating fault samples, the fault samples are input into the generator network for processing to obtain generated samples under the control of condition variable c. .

[0067] Let the generator (denoising network) be The input is The output is a prediction of the noise: Depend on An estimate of the original sample can be obtained: Then, linear interpolation is performed on the fault samples and the generated samples to obtain the interpolated samples. ,in, , is a random number that is uniformly distributed between 0 and 1.

[0068] Next, the fault samples, generated samples, and interpolated samples are input into the discriminator network for processing to generate corresponding discrimination results. Based on the first discrimination result corresponding to the fault sample, the second discrimination result corresponding to the generated sample, and the interpolated sample, the first loss value corresponding to the discriminator network is calculated; based on the second discrimination result corresponding to the generated sample, the second loss value of the generator network is calculated.

[0069] The first loss value of the discriminator network is calculated as follows: The second loss value of the generator network is calculated as follows: For fault sample generation tasks that include conditional information, the more information about the input conditional variables the generated fault samples contain, the better; that is, the degradation information contained in the input conditional variables should be preserved to the greatest extent possible in the generated samples. Mutual information can measure the degree of dependence between two random variables: Therefore, consider using the mutual information between the input condition variables and the generated samples. This is used to measure the degree of preservation of degraded information and is defined as conditional mutual information. Therefore, in addition to the loss values ​​of the generator and discriminator networks, it is necessary to optimize this conditional mutual information to its maximum during training. However, directly... Performing the calculations is very difficult, so an auxiliary distribution is constructed. This provides a lower bound for conditional mutual information: At this point, the objective of maximizing conditional mutual information is to maximize... At this point, a network Q is constructed to predict the mean and standard deviation of the condition variable. Assuming the condition variable follows a normal distribution, for a mean of... Standard deviation is For a random variable, its probability density function is: To calculate Part 1 It means to calculate ,Right now Assume the mean and standard deviation of the condition variable output by network Q are respectively... and The objective of maximizing conditional mutual information can be written as: Since the distribution of the condition variable c is fixed, Equivalent to: If not considered The impact of this further optimizes the target. Equivalent to: Therefore, the output of network Q is a vector with the same dimension as the condition variable c.

[0070] In actual training, besides calculating the loss values ​​of the generator and discriminator networks in the Conditional Temporal Deep Generative Adversarial Network (CTGAN) module, the generated samples can also be input into the degradation feature reconstruction network. The degradation feature reconstruction network reconstructs the degradation features of the generated samples to generate the first degradation feature. Based on the first degradation feature, the conditional mutual information maximization loss value between the generator network and the degradation feature reconstruction network is calculated. The specific processing method is as follows: Introducing degradation trends to rebuild networks This allows the degenerate semantics of the generated samples to revert to the condition variable: Additionally, diffusion uniformity can be calculated using standard noise to predict MSE: After calculating the aforementioned loss value, we can obtain: Total loss of generator (denoising network): Reconstructing network loss (can be shared with generator) ): Discriminator loss is .

[0071] Based on this, the model parameters of the sample generation model to be trained are adjusted according to each loss value to obtain the trained sample generation model. Specifically, according to... , , The generator network, discriminator network, and degradation trend reconstruction network are trained separately to obtain the trained sample generation model G. .

[0072] The fault data and the target condition variable can then be input into the trained sample generation model, and the generator network can process the fault data and the target condition variable to generate corresponding intermediate samples to be screened.

[0073] Specifically, the sample generation model generates intermediate samples to be screened, corresponding to the sample generation stage under unknown fault levels. This stage first samples fault samples and then concatenates a degradation trend condition variable (target condition variable) under the target fault level onto these samples. The fault samples and the target condition variable are then input into the trained sample generation model, where the generator network processes them to obtain the generator network's output. This output represents the fault sample corresponding to the input degradation trend condition variable for the fault level. By interpolating the degradation trend condition variables extracted from existing fault samples, the degradation trend condition variable under the unknown fault level is calculated, thus completing the interpolation generation of missing fault samples under the unknown fault level.

[0074] In practical applications, for a known set of condition variables Interpolation is performed to obtain the conditional variable representing the unknown degree of the target fault. For example, under linear interpolation: in, Let be the set of condition variables corresponding to the a-th sample out of n initial samples; Let b be the set of condition variables corresponding to the b-th sample out of n initial samples.

[0075] Then, the sample generation is gradually inverted from pure noise, that is, the sample generation stage starts with high noise initialization: right Stepwise noise reduction: The final generated sample is: In addition, by interpolating the second degradation feature, the degradation feature corresponding to the target fault level can be obtained, and the target fault level can be preset.

[0076] Accordingly, intermediate samples are sampled according to a preset sample size to generate at least two sample sets. Specifically, intermediate samples and intermediate samples to be screened are sampled according to a preset sample size to generate at least two sample sets.

[0077] In one optional implementation, before sampling the intermediate samples according to a preset sample size, the method further includes: Determine the failure mode corresponding to each intermediate sample, and extract the failure modes that meet the consistency index from each failure mode to obtain the set of failure modes to be verified. Determine the weight corresponding to each fault mode in the set of fault modes to be verified, and obtain intermediate sub-samples based on the weights from the intermediate samples; Accordingly, the sampling of the intermediate samples according to the preset sample size includes: The intermediate subsamples are sampled according to the preset sample size.

[0078] Furthermore, the step of extracting the failure modes that meet the consistency index from each failure mode to obtain a set of failure modes to be verified includes: Divide each fault mode into sets to generate at least two initial fault mode sets; Determine the hierarchical fault data corresponding to each fault mode in the target initial fault mode set, wherein each hierarchical fault data consists of fault modes at different levels, and the target initial fault mode set is each of the at least two initial fault mode sets; A judgment matrix is ​​constructed based on the hierarchical fault data, and the fault modes that meet the consistency index in each fault mode contained in the hierarchical fault data are extracted according to the judgment matrix to obtain a set of fault modes to be verified.

[0079] Specifically, due to the complex and uncertain relationships among the various characteristics of the evaluation indicators, these indicators may be incomparable, and each indicator may have a different degree of influence on the evaluation results. Therefore, the Analytic Hierarchy Process (AHP) is introduced to analyze the fault sample set and the fault mode set, thereby obtaining the set of fault modes to be verified.

[0080] In the embodiments of this specification, each fault mode is first divided into sets to generate at least two initial fault mode sets, that is, at least two replaceable units are generated.

[0081] Targeting the replaceable units of the fault mode set, where K is the number of replaceable units, and each replaceable unit U i By N i It consists of several failure modes.

[0082] Assuming replaceable unit U i The failure mode set is , For the fault mode set F i The corresponding hierarchical fault data. Hierarchical fault data refers to data that organizes and classifies fault modes according to a certain hierarchical structure. For example, fault modes can be divided into broad categories such as "functional faults," "hardware faults," and "software faults," and then further subdivided within each category. Specifically, all fault modes that may affect the unit can be identified through historical data, expert knowledge, fault reports, and system analysis, thus forming the hierarchical fault data corresponding to the unit's fault mode set.

[0083] Furthermore, after determining the hierarchical fault data corresponding to the fault mode set, the specific implementation process of the fault mode random extraction algorithm based on hierarchical analysis is as follows: Step 1: Construct the judgment matrix.

[0084] Statistical sets The number of fault modes k i and this k i Each failure mode is extracted and combined into a set. and order Construct a judgment matrix.

[0085] Where P is the judgment matrix. These are the eigenvalues ​​of the matrix.

[0086] The judgment matrix P is composed as follows: The goal is to identify failure modes that meet consistency metrics within a set of failure modes. , ( ) represents each fault mode in the fault mode set. express for The relative importance values. Then the judgment matrix P can be written as: In the established indicator system, indicators (failure modes) at the same level are comparable. However, within the same level, each indicator has a different weight. Considering attributes such as experimental risk, accuracy, and confidence level, the 1-9 scaling method shown in Table 1 is used to measure the relative merits of the judgment objects when constructing the judgment matrix. This scaling method is simple, clear, and easy to use.

[0087] Table 1 Scale Description Step 2: Solve the characteristic equation.

[0088] In the Analytic Hierarchy Process (AHP), the decision matrix P is used to represent the relative importance of each factor (failure mode). The eigenvector corresponding to the largest eigenvalue (eigenvalue) of the decision matrix P is obtained by solving the characteristic equation det(P−λI)=0, where I is the identity matrix.

[0089] Step 3: Perform a consistency check.

[0090] Specifically, the largest eigenvalue (eigenvalue) of the judgment matrix P is usually used for consistency testing. The consistency index can be determined based on the largest eigenvalue, and the consistency ratio can be calculated based on the index value. If the largest eigenvalue is close to the order n of the matrix (i.e., λmax≈n), the consistency of the judgment matrix is ​​considered to be good.

[0091] The test formula is as follows: In the formula, CR is the consistency ratio of the judgment matrix; CI is the general consistency index of the judgment matrix; and RI is the average random consistency index of the judgment matrix. The RI values ​​of judgment matrices of order 1 to 9 are shown in the table.

[0092] Table 2. Values ​​of the average random consistency index RI When the CR of the judgment matrix P is less than 0.1 or When CI=0, P is considered to have satisfactory consistency; otherwise, the elements in P need to be adjusted to ensure satisfactory consistency.

[0093] Given that the consistency ratio meets the condition, a set of failure modes to be verified is constructed based on the failure modes associated with the judgment matrix. Specifically, failure modes that meet the consistency index are extracted from the failure mode set to construct the set of failure modes to be verified, F. Fi .

[0094] Specifically, after obtaining the set of fault modes to be verified, the weight corresponding to each fault mode in the set of fault modes to be verified can be determined, and intermediate sub-samples can be obtained by filtering the intermediate samples based on the weights.

[0095] In the Analytic Hierarchy Process (AHP), the relative weight of each factor (failure mode) can be calculated using the judgment matrix, thereby ranking their importance. Specifically, the eigenvector corresponding to the largest eigenvalue can be normalized, and the result can be used as the relative weight of each factor to rank their importance, i.e., weight allocation.

[0096] In the embodiments of this specification, indicators (failure modes) at the same level are comparable, but each indicator at the same level has a different weight. Furthermore, these indicators include both positive indicators ("expecting large values") and negative indicators ("expecting small values"). Indicators belonging to the "expecting large values" category include: minimum sample size requirement, critical function failure coverage, functional failure coverage, test activation rate, testability rate, test utilization rate, injection cost consumption rate (by default, costs are incurred whenever a failure injection test is performed), and failure injection rate. Indicators belonging to the "expecting small values" category include: failure undiagnosable rate, severe failure coverage, and injection hazard rate. Moreover, considering the test costs and the number of failure modes allocated, the unit failure allocation rate and complex unit allocation rate are mutually constrained by the injection failure consumption rate. Therefore, as long as the required number of failure samples is met, these are also set as positive indicators in the embodiments of this specification. To better apply the analytic hierarchy process to sample sampling, negative indicators in the embodiments of this specification are reversed; that is, a calculation reference value is set for each negative indicator, i.e., the reference value equals 1 minus the actual true value.

[0097] Once a unified set of judgment indicators for decision-making is established, the weights of the analytic hierarchy process (AHP) can be determined. Methods for calculating the AHP weight vector W include the geometric mean, arithmetic mean, eigenvector method, and least squares method.

[0098] Once the consistency check and weight determination of the failure mode set are completed, a better failure mode set can be determined based on the weight distribution of the hierarchical analysis.

[0099] Specifically, after sorting each fault mode in the set of fault modes to be verified in descending order of weight, a certain number of fault modes at the top of the sorting can be identified as target fault modes, and the corresponding samples to be verified can be identified. Then, the samples to be verified are used as intermediate subsamples.

[0100] Accordingly, intermediate samples are sampled according to a preset sample size, specifically, intermediate subsamples are sampled according to a preset sample size.

[0101] Step 106: Obtain the mapping relationship between the fault modes related to the target project and the project parameter values ​​of the target project, and determine the fault mode equivalent set corresponding to each sample set based on the mapping relationship.

[0102] Specifically, after generating at least two sample sets, the sufficiency of the sample parameters of the target project contained in each sample set can be tested. If the test passes, the sample is identified as a target sample related to the target project. This target sample can be used for fault detection and diagnosis of the target project.

[0103] Because constructing imbalanced samples may disrupt the original sample set's coverage of fault conditions, the generated samples need to be sufficiency tested to ensure they can cover all possible fault conditions.

[0104] For test validation experiments, the sample size is fixed and the sample set is generated by sampling. Therefore, it is necessary to measure the sufficiency of the generated fault detection and diagnosis sample set and select the optimal sample set for subsequent experiments.

[0105] The sufficiency of validation samples for data-driven fault detection and diagnosis models is categorized into two types: fault distribution coverage sufficiency measures and data distribution coverage sufficiency measures. Fault distribution coverage sufficiency measures include fault importance coverage sufficiency measures and fault occurrence probability coverage sufficiency measures. Data distribution coverage sufficiency measures include time-domain feature distribution coverage sufficiency measures, frequency-domain feature distribution coverage sufficiency measures, statistical domain feature distribution coverage sufficiency measures, and comprehensive data distribution coverage sufficiency measures.

[0106] For measuring the adequacy of fault importance coverage, firstly, considering that fault importance is usually represented by ordered discrete levels, and the difference between adjacent levels is much smaller than the difference across levels, the absolute difference is used as the point-to-point distance in local matching to accurately reflect the inherent hierarchical relationship of fault importance. Secondly, to address inconsistencies between two sequences in terms of time scale or event density, Dynamic Time Warping (DTW) is used for flexible alignment. DTW constructs an optimal nonlinear mapping path, allowing sequences to be moderately stretched or compressed on the time axis, thereby capturing potential similar evolution patterns and avoiding the underestimation of similarity due to small offsets.

[0107] By dividing the cumulative distance of DTW by the actual alignment path length instead of directly using it, the average point-by-point difference can be obtained. This operation effectively isolates the influence of sequence length on the similarity score, thus achieving a fair assessment of coverage adequacy. Furthermore, by dividing this average difference by the theoretical maximum span of the fault importance level for normalization, the final similarity can be defined as: in, The DTW average of fault importance between the overall population and the final generated sample set. The highest fault importance in the overall fault sample population. It represents the minimum fault importance in the overall fault sample population.

[0108] The range of values ​​for fault importance coverage adequacy is [0,1]. The closer the value is to 1, the more representative the sample is of the population in terms of fault importance, and the more adequate the coverage of critical faults. Conversely, there may be a risk of missing high-importance faults.

[0109] For the adequacy measure of fault occurrence probability coverage, this embodiment first uses DTW (Dynamic Data Wave) to flexibly align two probability sequences of unequal lengths to tolerate asynchrony caused by differences in modeling granularity or sampling frequency. Then, the mean absolute error (MAE) on the optimal alignment path is calculated and normalized relative to the maximum possible deviation in the probability space to form the adequacy measure of fault occurrence probability coverage. This measure not only effectively captures the overall trend consistency and local peak matching degree of the probability distribution, but also significantly reduces the interference of sequence length on the evaluation results by using average error instead of cumulative error, thus fairly reflecting the completeness and adequacy of the sample set's coverage of the baseline fault occurrence probability pattern. The closer the measure is to 1, the higher the fault probability distribution of the sample set is to the overall distribution, indicating sufficient coverage. Conversely, a lower measure indicates problems such as systematic underestimation, omission of high-risk faults, or false alarms of low-risk events. The mathematical definition of fault occurrence probability coverage adequacy is: in, The DTW average of the failure occurrence probability between the overall population and the final generated sample set. This represents the highest probability of failure occurring in the overall failure sample population. It represents the minimum probability of failure in the overall failure sample population.

[0110] The time-domain feature distribution coverage adequacy measure is an index that quantifies the representativeness of a sampled signal subset to the overall distribution in a multi-dimensional time-domain feature space, with a value range of [0,1]. This method first extracts 10 time-domain features from each signal segment, including mean, variance, peak value, peak-to-peak value, skewness, kurtosis, waveform factor, impulse factor, margin factor, and zero-crossing rate. Then, each feature dimension is standardized to 0-1, and the empirical probability distributions of the population and subset are estimated separately, and their Jensen-Shannon (JS) distance is calculated. By normalizing the JS distance and inversely mapping it to a coverage score, the average of all dimensions is finally taken to obtain the comprehensive coverage adequacy measure. The index constructed by this method possesses good mathematical properties such as symmetry, boundedness, and sensitivity to distribution differences. The closer the value is to 1, the better the subset's coverage of the overall characteristics in the time-domain feature distribution. Its mathematical expression is: in Features The JS distance can be calculated using the following formula: For the first sample in the overall sample set The probability distribution of dimensional features This represents the probability distribution for extracting features from a subset. .

[0111] For a measure of the sufficiency of frequency domain feature distribution coverage, in the frequency domain, for the Fourier-Bessel domain spectrum of the signal... The following method is used to determine the local maxima in the spectrum. Frequency band boundary Assume the number of local maxima found is ,but: At that time, keep the previous one The maximum points are used as the frequency band boundaries; When resetting parameters ,make For each sample, calculate the energy proportion of each frequency band, and take the dominant frequency band (the band with the highest energy) as its frequency domain category. Statistically analyze the frequency domain distribution of the sample proportions for each category. Based on the statistical results, the JS distance between the overall distribution and the fault sample set distribution is calculated. This is mapped to the sufficiency of frequency domain feature distribution coverage. Its mathematical expression is: To measure the sufficiency of statistical domain feature distribution coverage, statistical features are first extracted from the statistical domain: higher-order moment statistics such as 3rd-order skewness and 4th-order kurtosis, autocorrelation function decay rate, and information entropy. Within the statistical domain formed by these statistical features, the JS distance between the overall distribution and the fault sample set distribution is calculated. This is mapped to the sufficiency of coverage of statistical domain feature distributions. Its mathematical representation is: To comprehensively evaluate the representativeness of a subset of signals to the overall signal distribution across different feature space dimensions, a comprehensive data distribution coverage adequacy measure is proposed. This measure integrates the distribution coverage of features in the time, frequency, and statistical domains, providing a unified quantitative standard to evaluate the statistical representativeness of a sampled signal set. Specifically, the comprehensive data distribution coverage adequacy measure combines key features extracted from each feature domain and constructs its own distribution coverage adequacy measure based on these features, ultimately obtaining a comprehensive score through arithmetic averaging. The mathematical expression is as follows: In the embodiments of this specification, to perform sufficiency measurement on at least two generated sample sets, it is necessary to first obtain the mapping relationship between the failure modes related to the target project and the project parameter values ​​of the target project, and then determine the equivalent set of failure modes corresponding to each sample set based on the mapping relationship.

[0112] When determining the mapping relationship between failure modes and project parameter values, firstly, determine the mapping relationship based on the design data of the target project. The elements included are: a set of failure modes (FU), a set of characteristic (project parameter) values ​​(AU), and a mapping relationship (PU). Then, the equivalent set of failure modes for each sample set is determined based on the PU. In this context, "design documents" refers to the project specification of the target project. The failure mode set (FU), the characteristic value set (AU), and the mapping relationship (PU) are key elements for conducting sample sufficiency testing. FU refers to the set of all possible failure modes the target project may encounter. AU refers to the set of all possible values ​​for the product characteristics, i.e., the project parameters of the target project. These characteristics may include performance parameters, environmental conditions, operating conditions, etc. PU refers to the mapping relationship between the failure modes in the failure mode set FU and the characteristic values ​​in the characteristic value set AU. It describes how a specific failure mode is caused by a specific characteristic value. PU is the bridge connecting failure modes and characteristic values; through it, we can understand which characteristic changes may lead to which failure modes. The failure mode equivalence set refers to different failure modes that lead to the same failure performance under specific conditions. These failure modes are considered equivalent in analysis and testing, and their impact on the system is similar.

[0113] Based on this, after obtaining the mapping relationship PU between failure modes and project parameter values, since it directly relates failure modes and project parameter values, the equivalent set of failure modes corresponding to each sample set is determined according to the mapping relationship. Specifically, the mapping relationship PU can be analyzed in detail to understand how different project parameter values ​​affect different failure modes, identify which failure modes will lead to the same system effect under specific conditions, and determine the equivalent set of failure modes corresponding to each sample set based on the result. This equivalent set of failure modes contains information such as the corresponding failure mode description, characteristic values, and system impact.

[0114] To address the practical engineering challenges of "limited sample set, complex failure modes, and multidimensional coverage" in test validation experiments, this specification's embodiments introduce a multi-index joint evaluation and statistical confidence constraint mechanism on the basis of the original sample sufficiency measurement and optimization process, forming a sample sufficiency testing method with stability constraints.

[0115] This method, without altering the original sample generation and sufficiency measurement calculation model, improves the reliability and engineering interpretability of the sample set optimization results by analyzing the statistical characteristics of the sufficiency measurement results. The specific implementation process is as follows: (1) Sample sufficiency test and selection.

[0116] 1) Determine the set of elements for the test object Determine the set of failure modes based on product design data. , characteristic value set and the mapping relationship between the two. And based on this, construct an equivalent set of failure modes for the product.

[0117] 2) Construct multiple sets of candidate fault sample sets Given a fixed sample size, a random sampling method that considers the failure rate distribution characteristics is used to generate multiple candidate failure sample sets: 3) Calculate the multidimensional sufficiency measure index For each sample set, calculate one or more of the following sufficiency metrics: unit coverage sufficiency metric, functional coverage sufficiency metric, test coverage sufficiency metric, and comprehensive coverage sufficiency metric.

[0118] Record No. The sample set in the th case... The measurement results under each sufficiency indicator are as follows: 4) Construct a multi-indicator joint sufficiency evaluation vector The results of various sufficiency measures are uniformly represented as a sufficiency vector of the sample set: This vector can simultaneously reflect the sufficiency of the sample set under different coverage dimensions, avoiding the bias caused by a single indicator.

[0119] (2) Optimization method for sample sufficiency based on statistical stability.

[0120] Given that the sample set is obtained through random sampling, its sufficiency measurement results exhibit random fluctuations. To improve the stability of the optimization results, this specification introduces a repeated sampling and statistical confidence constraint mechanism in its embodiments. The specific steps are as follows: 1) Statistical analysis of repeated sampling and sufficiency measurement For each candidate sample set By performing multiple independent samplings or rearrangements under the same sampling rules, the statistical distribution of its sufficiency measure is obtained, and its expected value is calculated: If necessary, variance or confidence interval can be calculated simultaneously to measure the stability of the sufficiency assessment of the sample set.

[0121] 2) Introducing a comprehensive sufficiency index with stability constraints Based on the existing sufficiency measure, a stability constraint is introduced to construct a comprehensive evaluation index for the sample set: in, The weights of different sufficiency indicators can be determined by engineering experience or the aforementioned analytic hierarchy process.

[0122] 3) Sample set optimization rules From all candidate sample sets, select the sample set that satisfies the statistical stability requirement and has the largest comprehensive adequacy index: When multiple sample sets When values ​​are similar, prioritize the sample set with smaller variance or better coverage balance.

[0123] (3) Final sample set analysis and remedial mechanism Sufficiency analysis and remedial suggestions: for the final selected sample set The adequacy measurement results are analyzed to identify weak coverage areas at the unit, functional, or test levels, providing a basis for subsequent permissible remedial measures (such as targeted resampling and critical fault annotation).

[0124] Step 108: Determine the sufficiency metric of the corresponding sample set based on the fault mode equivalence set, and determine the target samples related to the target project based on the sufficiency metric.

[0125] In one optional implementation, the step of determining a sufficiency metric for the corresponding sample set based on the failure mode equivalence set, and determining target samples related to the target project based on the sufficiency metric, includes: Determine the number of fault modes included in the fault mode equivalence set; Determine the ratio between the number of failure modes and the total number of failure modes related to the target project, and use the ratio as a sufficiency measure of the sample set corresponding to the equivalent set of failure modes; Determine the target sample set with the largest sufficiency metric value in each sample set, and identify the samples contained in the target sample set as target samples related to the target project.

[0126] Specifically, the purpose of performing a sufficiency test on the generated sample set is to check the fault conditions it can cover, and the coverage rate is one of the most intuitive sufficiency measures. It measures the proportion of fault modes in the sample set that cover all possible fault modes.

[0127] Therefore, in the embodiments of this specification, the number of failure modes contained in each failure mode equivalence set can be determined first, then the ratio between the number of each failure mode and the total number of failure modes related to the target project can be calculated, and this ratio can be determined as the sufficiency measure of the sample set corresponding to the failure mode equivalence set; finally, the sample contained in the target sample set with the largest sufficiency measure in each sample set can be determined as the target sample related to the target project.

[0128] This embodiment of the specification filters initial samples stored in a database according to the project requirements of the target project to obtain intermediate samples related to the target project; it then samples the intermediate samples according to a preset sample size to generate at least two sample sets; it obtains the mapping relationship between the fault modes related to the target project and the project parameter values ​​of the target project, and determines the fault mode equivalence set corresponding to each sample set based on the mapping relationship; it determines the sufficiency metric of the corresponding sample set based on the fault mode equivalence set, and determines the target samples related to the target project based on the sufficiency metric. By filtering samples according to the project requirements of different target projects to obtain qualified target samples, it is beneficial to ensure the accuracy and applicability of the generated samples, and also to ensure the accuracy of the detection results obtained by fault detection and diagnosis through the target samples.

[0129] A schematic diagram of a fault detection, diagnosis, and verification system for imbalanced sample generation provided in the embodiments of this specification is shown below. Figure 2 As shown.

[0130] The database stores different types of data, including: historical unprocessed fault data, current unprocessed fault data, historical valid fault samples, current valid fault samples, historical sample sets to be evaluated, current sample sets to be evaluated, historical best sample sets, and best sample sets.

[0131] The process begins by performing an intensional analysis on historical and current unprocessed fault data stored in the database to obtain historically valid fault samples and currently valid fault samples. Then, data discrimination criteria are used to filter these samples, resulting in historical and current evaluation sample sets. Next, a sample size determination algorithm is used to determine if the sample sizes of these sets meet the requirements. If not, sample generation continues until the required sample size is met. If yes, multiple random sampling algorithms are used to sample the historical and current evaluation sample sets, and the sampling results are subjected to a sample sufficiency test. The sufficiency test results are then used to determine if the sample sufficiency meets the requirements. If not, the process returns to the intensional analysis until the sample sufficiency meets the requirements. If yes, the sampling results are determined as the historical optimal sample set and the optimal sample set and stored in the database.

[0132] In addition, the generated historical best sample set and best sample set can also be transmitted to the host machine through various communication interfaces such as CAN (Controller Area Network), I2C (Integrated Circuit Bus), RS485 (a serial communication protocol), USB (Universal Serial Bus), Ethernet interface, and expansion interface (used to connect and expand the system so as to communicate with different devices and systems) to realize the complete process of sample generation.

[0133] Corresponding to the above method embodiments, this specification also provides embodiments of a fault detection, diagnosis, and verification unbalanced sample generation device. Figure 3 A schematic diagram of a fault detection, diagnosis, and verification unbalanced sample generation device according to one embodiment of this specification is shown. Figure 3 As shown, the device includes: The filtering module 302 is configured to filter the initial samples stored in the database according to the project requirements of the target project to obtain intermediate samples related to the target project. The sampling module 304 is configured to sample the intermediate samples according to a preset sample size to generate at least two sample sets; The acquisition module 306 is configured to acquire the mapping relationship between the fault modes related to the target project and the project parameter values ​​of the target project, and determine the fault mode equivalent set corresponding to each sample set based on the mapping relationship; The determination module 308 is configured to determine the sufficiency metric of the corresponding sample set based on the fault mode equivalence set, and to determine the target sample related to the target project based on the sufficiency metric.

[0134] In an optional implementation, the determining module 308 is further configured to: Determine the number of fault modes included in the fault mode equivalence set; Determine the ratio between the number of failure modes and the total number of failure modes related to the target project, and use the ratio as a sufficiency measure of the sample set corresponding to the equivalent set of failure modes; Determine the target sample set with the largest sufficiency metric value in each sample set, and identify the samples contained in the target sample set as target samples related to the target project.

[0135] In one optional implementation, the fault detection, diagnosis, and verification imbalanced sample generation device further includes a processing module configured to: Determine the target sample size for the target samples related to the target project based on the project requirements; If the sample size of the intermediate samples meets the requirements based on the target sample size, the step of sampling the intermediate samples according to the preset sample size to generate at least two sample sets is performed.

[0136] In an optional implementation, the processing module is further configured to: If it is determined that the sample size of the intermediate sample is insufficient, fault data generated by performing forward diffusion noise injection on the initial sample is obtained, and the target component variable corresponding to the fault data is determined. The fault data and the target condition variable are input into the trained sample generation model for processing to generate intermediate samples to be screened. The sample generation model includes a generator network, a discriminator network, and a degradation feature reconstruction network. Accordingly, the sampling module 304 is also configured to: The intermediate samples and the intermediate samples to be screened are sampled according to the preset sample size to generate at least two sample sets.

[0137] In an optional implementation, the processing module is further configured to: Determine the failure mode corresponding to each intermediate sample, and extract the failure modes that meet the consistency index from each failure mode to obtain the set of failure modes to be verified. Determine the weight corresponding to each fault mode in the set of fault modes to be verified, and obtain intermediate sub-samples based on the weights from the intermediate samples; Accordingly, the sampling module 304 is also configured to: The intermediate subsamples are sampled according to the preset sample size.

[0138] In an optional implementation, the processing module is further configured to: Divide each fault mode into sets to generate at least two initial fault mode sets; Determine the hierarchical fault data corresponding to each fault mode in the target initial fault mode set, wherein each hierarchical fault data consists of fault modes at different levels, and the target initial fault mode set is each of the at least two initial fault mode sets; A judgment matrix is ​​constructed based on the hierarchical fault data, and the fault modes that meet the consistency index in each fault mode contained in the hierarchical fault data are extracted according to the judgment matrix to obtain a set of fault modes to be verified.

[0139] In an optional implementation, the sampling module is further configured to: The intermediate samples are sampled using a random sampling algorithm that takes into account the failure rate and according to a preset sample size.

[0140] The above is a schematic scheme of a fault detection, diagnosis, and verification imbalanced sample generation device according to this embodiment. It should be noted that the technical solution of this fault detection, diagnosis, and verification imbalanced sample generation device belongs to the same concept as the technical solution of the aforementioned fault detection, diagnosis, and verification imbalanced sample generation method. Details not described in detail in the technical solution of the fault detection, diagnosis, and verification imbalanced sample generation device can be found in the description of the technical solution of the aforementioned fault detection, diagnosis, and verification imbalanced sample generation method.

[0141] Figure 4 A structural block diagram of a computing device 400 according to one embodiment of this specification is shown. The components of the computing device 400 include, but are not limited to, a memory 410 and a processor 420. The processor 420 is connected to the memory 410 via a bus 430, and a database 450 is used to store data.

[0142] The computing device 400 also includes an access device 440, which enables the computing device 400 to communicate via one or more networks 460. Examples of these networks include a Public Switched Telephone Network (PSTN), a Local Area Network (LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device 440 may include one or more of any type of wired or wireless network interface (e.g., a Network Interface Card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) interface, a Wi-MAX interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.

[0143] In one embodiment of this specification, the aforementioned components of the computing device 400 and Figure 4 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 4 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0144] The computing device 400 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or PCs. The computing device 400 can also be a mobile or stationary server.

[0145] The processor 420 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the above-described method for fault detection, diagnosis, and verification of imbalanced sample occurrence.

[0146] The above is a schematic representation of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the above-described method for fault detection, diagnosis, and verification of imbalanced sample generation belong to the same concept. Details not described in detail in the technical solution of the computing device can be found in the description of the technical solution of the above-described method for fault detection, diagnosis, and verification of imbalanced sample generation.

[0147] An embodiment of this specification also provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the above-described method for fault detection, diagnosis, and verification of imbalanced sample occurrence.

[0148] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium belongs to the same concept as the technical solution of the above-described method for fault detection, diagnosis, and verification of imbalanced sample generation. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the above-described method for fault detection, diagnosis, and verification of imbalanced sample generation.

[0149] An embodiment of this specification also provides a computer program, wherein when the computer program is executed in a computer, the computer is instructed to perform the steps of the above-described fault detection, diagnosis, and verification method for the occurrence of imbalanced samples.

[0150] The above is an illustrative scheme of a computer program according to this embodiment. It should be noted that the technical solution of this computer program belongs to the same concept as the technical solution of the above-described method for fault detection, diagnosis, and verification of imbalanced sample generation. For details not described in detail in the technical solution of the computer program, please refer to the description of the technical solution of the above-described method for fault detection, diagnosis, and verification of imbalanced sample generation.

[0151] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0152] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added to or subtracted according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0153] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0154] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0155] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A fault detection diagnosis verification unbalanced sample generation method, comprising: screening initial samples stored in a database according to project requirements of a target project to obtain intermediate samples related to the target project; sampling the intermediate samples according to a preset sample quantity to generate at least two sample sets; obtaining a mapping relationship between fault modes related to the target project and project parameter values of the target project, and determining a fault mode equivalent set corresponding to each sample set according to the mapping relationship; determining a sufficiency metric value of the corresponding sample set according to the fault mode equivalent set, and determining a target sample related to the target project according to the sufficiency metric value.

2. The fault detection diagnosis verification unbalanced sample generation method according to claim 1, wherein the determination of the sufficiency metric value of the corresponding sample set according to the fault mode equivalent set, and the determination of the target sample related to the target project according to the sufficiency metric value, comprises: determining the number of fault modes contained in the fault mode equivalent set; determining a ratio between the number of fault modes and a total number of fault modes related to the target project, and determining the ratio as the sufficiency metric value of the sample set corresponding to the fault mode equivalent set; determining a target sample set with the largest sufficiency metric value in each sample set, and determining samples contained in the target sample set as the target sample related to the target project.

3. The fault detection diagnosis verification unbalanced sample generation method according to claim 1, wherein after the obtaining of the intermediate samples related to the target project, the method further comprises: determining a target sample quantity of the target sample related to the target project according to the project requirements; in a case where the sample quantity of the intermediate samples determined according to the target sample quantity meets the requirements, performing the sampling of the intermediate samples according to the preset sample quantity to generate at least two sample sets.

4. The fault detection diagnosis verification unbalanced sample generation method according to claim 1 or 3, further comprising: in a case where it is determined that the sample quantity of the intermediate samples is insufficient, obtaining fault data generated by performing forward diffusion noise injection on the initial samples, and determining a target condition variable corresponding to the fault data; inputting the fault data and the target condition variable into a trained sample generation model for processing to generate intermediate samples to be screened, the sample generation model comprising a generator network, a discriminator network, and a degradation feature reconstruction network; correspondingly, the sampling of the intermediate samples according to the preset sample quantity to generate at least two sample sets comprises: sampling the intermediate samples and the intermediate samples to be screened according to the preset sample quantity to generate at least two sample sets.

5. The fault detection diagnosis verification unbalanced sample generation method according to claim 1 or 3, wherein before the sampling of the intermediate samples according to the preset sample quantity, the method further comprises: determining a fault mode corresponding to each intermediate sample, and extracting fault modes satisfying a consistency index in each fault mode to obtain a fault mode set to be verified. determine a weight corresponding to each fault mode in the to-be-verified fault mode set, and filter the intermediate sample based on the weight to obtain an intermediate sub-sample; Accordingly, the sampling of the intermediate sample according to the preset sample size comprises: sampling the intermediate sub-sample according to the preset sample size.

6. The method of claim 5, wherein the extracting of the fault modes that meet the consistency index from each fault mode comprises: performing set division on each fault mode to generate at least two initial fault mode sets; determining a hierarchical fault data corresponding to each fault mode in a target initial fault mode set, wherein each hierarchical fault data is composed of fault modes of different levels, and the target initial fault mode set is each of the at least two initial fault mode sets; constructing a judgment matrix based on the hierarchical fault data, and extracting the fault modes that meet the consistency index from each fault mode included in the hierarchical fault data according to the judgment matrix to obtain the to-be-verified fault mode set.

7. The method of claim 1, wherein the sampling of the intermediate sample according to the preset sample size comprises: sampling the intermediate sample according to the preset sample size by considering a random sampling algorithm of fault rate.

8. An apparatus for generating an unbalanced sample for fault detection and diagnosis verification, comprising: a filtering module configured to filter an initial sample stored in a database according to a project requirement of a target project to obtain an intermediate sample related to the target project; a sampling module configured to sample the intermediate sample according to a preset sample size to generate at least two sample sets; an obtaining module configured to obtain a mapping relationship between fault modes related to the target project and project parameter values of the target project, and determine a fault mode equivalence set corresponding to each sample set according to the mapping relationship; a determining module configured to determine a sufficiency measurement value of the corresponding sample set according to the fault mode equivalence set, and determine a target sample related to the target project according to the sufficiency measurement value.

9. A computing device, comprising: a memory and a processor; the memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions, which, when executed by the processor, implement the steps of the method for generating an unbalanced sample for fault detection and diagnosis verification according to any one of claims 1 to 7.

10. A computer readable storage medium storing computer executable instructions, which, when executed by a processor, implement the steps of the method for generating an unbalanced sample for fault detection and diagnosis verification according to any one of claims 1 to 7.

Images