Fault diagnosis method and device and electronic equipment

By combining the SPOT algorithm and a pre-trained support vector machine model with a physical attribute definition function, the problems of low accuracy and poor applicability of fault diagnosis methods are solved, achieving more accurate and universal fault diagnosis and reducing misjudgment and reliance on expert experience.

CN118940136APending Publication Date: 2024-11-12SHENZHEN ENVICOOL INFORMATION TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411060176.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-02
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

Existing fault diagnosis methods have low accuracy and lack universality. They rely on expert experience and classification algorithms, which carries the risk of misjudgment. They also have poor applicability and are complex and costly to implement.

Method used

The SPOT algorithm is used to calculate the feature threshold. Combined with a pre-trained support vector machine model, the feature data of the equipment under the same working conditions is obtained for classification. The physical attribute definition function is used to map the data as physical attribute values. The model is trained based on the annotation information to achieve automatic and dynamic adjustment of the threshold.

Benefits of technology

It improves the accuracy and universality of fault diagnosis, reduces reliance on expert experience, can adapt to changes in different equipment and environments, and reduces the risk of misjudgment and implementation complexity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940136A_ABST
    Figure CN118940136A_ABST
Patent Text Reader

Abstract

The invention belongs to the field of fault diagnosis, and discloses a fault diagnosis method and device and electronic equipment, and the method comprises the steps: obtaining a first feature data set, and calculating a corresponding first feature threshold value according to the first feature data set through a preset SPOT algorithm; and utilizing a pre-trained support vector machine model to classify each piece of first feature data based on the first feature threshold, and determining the fault category of each piece of equipment according to a classification result. According to the invention, by comparing the performance of different devices under the same working condition, the abnormal behavior can be recognized more objectively, the risk of misjudgment possibly caused by single device data is reduced, and the diagnosis accuracy is improved. Meanwhile, the transverse comparison does not depend on the characteristics of specific equipment, and the threshold value can be dynamically adjusted according to actual data, so that the diagnosis system can adapt to the changes of different equipment and environments and can be widely applied to different equipment and systems, and the universality of the fault diagnosis method is improved.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to the field of fault diagnosis, and in particular to a fault diagnosis method, device, electronic device and computer-readable storage medium. [Background technology]

[0002] In the traditional field of equipment fault diagnosis, it is usually necessary to rely on experts in related technical fields to combine their own experience and knowledge to set the fault thresholds of key performance indicators. This method can preliminarily judge the fault phenomenon of the equipment by observing the changing trends of multiple key performance indicators. However, this threshold setting that relies on expert experience has obvious limitations. First, different experts may have different judgment criteria, resulting in a lack of objectivity in this method; second, because the same threshold may not accurately reflect the fault status in different systems and equipment, this method is not universal.

[0003] Furthermore, although the use of advanced classification algorithms such as convolutional neural networks and extreme gradient boosting trees can improve the accuracy of fault diagnosis, these methods still have some problems. Relying solely on classification algorithms for fault diagnosis may bring the risk of misjudgment because they may not fully consider the specific characteristics and operating environment of the equipment. In addition, the applicability of these algorithms between different systems is also challenging, and they need to be adjusted and optimized for specific systems, which increases the complexity and cost of implementation.

[0004] Therefore, how to provide a fault diagnosis method that is both accurate and universal is a technical problem that needs to be solved in the art. [Summary of the invention]

[0005] In order to effectively solve the problem that the fault diagnosis method in the prior art has low accuracy and is not universal, the present invention provides a fault diagnosis method.

[0006] A fault diagnosis method, comprising:

[0007] Acquire a first characteristic data group, where the first characteristic data group is one or more groups of first characteristic data of a preset number of devices under the same working conditions;

[0008] Calculating corresponding first feature thresholds according to the first feature data group using a preset SPOT algorithm, wherein the number of the first feature thresholds is consistent with the number of the first feature data group;

[0009] A pre-trained support vector machine model is used to classify each of the first feature data based on the first feature threshold, and a fault category of each of the devices is determined according to the classification result.

[0010] Optionally, before using the pre-trained support vector machine model to classify each of the first feature data based on the first feature threshold, the method further includes:

[0011] Acquire a second characteristic data group, where the second characteristic data group is one or more groups of second characteristic data of a preset number of devices under the same working conditions;

[0012] Calculating corresponding second feature thresholds according to the second feature data group using the preset SPOT algorithm, wherein the number of the second feature thresholds is consistent with the number of the second feature data group;

[0013] Calculate the second physical property value corresponding to each second feature data by using a physical property definition function combined with the second feature threshold;

[0014] The labeling information corresponding to each second physical property value is received, and an initial support vector machine model is trained according to each second physical property value and the corresponding labeling information to obtain the pre-trained support vector machine model.

[0015] Optionally, the calculating a corresponding second feature threshold value according to the second feature data group using the preset SPOT algorithm includes:

[0016] The probability density of the second feature data group is determined as:

[0017]

[0018] Among them, X i is the second feature data set, f(X i ) is the probability density of the second feature data set, t is the initial comparison value of each window, x is a real number that satisfies any 1+γx>0, γ is a shape parameter, σ(t) is a scale parameter, and τ is a limit value of the distribution;

[0019] The maximum likelihood function of the second feature data group is determined according to the probability density as follows:

[0020]

[0021] The maximum likelihood function is used to determine the maximum log-likelihood function of the second feature data group as follows:

[0022]

[0023] Among them, N t is the number of abnormal feature data in the second feature data group;

[0024] Determining a predicted value of the shape parameter and a predicted value of the scale parameter according to the maximized log-likelihood function;

[0025] Calculating a feature threshold of the second feature data group according to the predicted value of the shape parameter and the predicted value of the scale parameter using a preset SPOT algorithm;

[0026]

[0027] Among them, z q is the second feature threshold of the second feature data group, including a second feature lower limit value and a second feature upper limit value, p is the initial probability value, and n is the total data volume.

[0028] Optionally, the calculating the second physical property value corresponding to each second feature data by using the physical property definition function in combination with the second feature threshold value includes:

[0029] Calculating a first preset quantile and a second preset quantile of the second feature data group;

[0030] Each of the second feature data is mapped to a corresponding second physical property value by using the physical property definition function in combination with the second feature lower limit value, the second feature upper limit value, the first preset quantile, and the second preset quantile:

[0031]

[0032] Among them, A(X i ) is the physical property value, l is the second characteristic lower limit value of the second characteristic data group, q1 is the first preset quantile, q2 is the second preset quantile, and u is the second characteristic upper limit value of the second characteristic data group.

[0033] Optionally, the training of an initial support vector machine model according to each of the second physical property values ​​and the corresponding annotation information to obtain the pre-trained support vector machine model includes:

[0034] The hyperplane linear equation for determining the initial support vector machine model is:

[0035] ωA(X i )+b=0

[0036] Among them, ω is the weight and b is the bias;

[0037] Determine the objective function of the initial support vector machine model according to the hyperplane linear equation:

[0038]

[0039] Among them, ω is the weight, b is the bias, ξi is a slack variable, parameter C is a penalty factor, y i To mark information;

[0040] The initial support vector machine model is trained according to each of the second physical property values ​​and the corresponding annotation information to obtain the weight and the bias:

[0041]

[0042] Among them, α i is the KKT conditional multiplier.

[0043] Optionally, the using a pre-trained support vector machine model to classify each of the first feature data based on the first feature threshold, and determining a fault category of each of the devices according to the classification result, includes:

[0044] Mapping each of the first characteristic data to a corresponding first physical property value using the physical property definition function;

[0045] Each of the first physical property values ​​is input into the hyperplane linear equation for calculation, and the fault category of each of the devices is determined according to the calculation result.

[0046] Optionally, before calculating the first preset quantile and the second preset quantile of the second feature data group, the method further includes:

[0047] An input modification instruction is received, and the first preset quantile and the second preset quantile are modified.

[0048] A fault diagnosis device, comprising:

[0049] A first acquisition module, used to acquire a first characteristic data group, where the first characteristic data group is one or more groups of first characteristic data of a preset number of devices under the same working conditions;

[0050] A first calculation module, configured to calculate corresponding first feature thresholds according to the first feature data group using a preset SPOT algorithm, wherein the number of the first feature thresholds is consistent with the number of the first feature data group;

[0051] A classification module is used to classify each of the first feature data based on the first feature threshold using a pre-trained support vector machine model, and determine a fault category of each of the devices according to the classification result.

[0052] An electronic device, comprising:

[0053] A processor and a memory, wherein the memory is used to store at least one instruction, and when the instruction is loaded and executed by the processor, the fault diagnosis method as described in any one of the above items is implemented.

[0054] A computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements any one of the above-mentioned fault diagnosis methods.

[0055] The fault diagnosis method provided by the embodiment of the present invention has at least the following beneficial effects:

[0056] The present invention obtains a first feature data group, calculates a corresponding first feature threshold value based on the first feature data group using a preset SPOT algorithm, and the number of first feature threshold values ​​is consistent with the number of first feature data groups; uses a pre-trained support vector machine model to classify each first feature data based on the first feature threshold value, and determines the fault category of each device according to the classification result. By comparing the performance of different devices under the same working conditions, the present invention can more objectively identify abnormal behaviors, reduce the risk of misjudgment that may be caused by single device data, and improve diagnostic accuracy. At the same time, through horizontal comparison, which is not dependent on the characteristics of a specific device, the threshold value can be dynamically adjusted according to the actual data, so that the diagnostic system can adapt to changes in different devices and environments, and can be widely used in different devices and systems, thereby improving the universality of the fault diagnosis method and reducing the dependence on expert experience, making the diagnostic process more objective.

Brief Description of the Drawings

[0057] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0058] Figure 1 A flowchart of a fault diagnosis method provided by an embodiment of the present invention;

[0059] Figure 2 A flowchart of another fault diagnosis method provided by an embodiment of the present invention;

[0060] Figure 3 A schematic diagram of a function of a physical property value provided by an embodiment of the present invention;

[0061] Figure 4 A schematic diagram of a fault diagnosis provided by an embodiment of the present invention;

[0062] Figure 5 A schematic diagram of the overall situation of a sample provided by an embodiment of the present invention;

[0063] Figure 6 Based on Figure 5 The overall situation of the samples shown is a schematic diagram of the hyperplane obtained through the support vector machine model;

[0064] Figure 7 Based on Figure 6 Schematic diagram of classification using the hyperplane shown;

[0065] Figure 8 A schematic diagram of the structure of a fault diagnosis device provided by an embodiment of the present invention. [Specific implementation method]

[0066] In order to better understand the technical solution of the present invention, the embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0067] It should be clear that the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0069] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0070] In the field of traditional equipment fault diagnosis, threshold setting that relies on expert experience has obvious limitations. Relying solely on classification algorithms for fault diagnosis may bring the risk of misjudgment because it may not fully consider the specific characteristics and operating environment of the equipment. In addition, the applicability of these algorithms between different systems is also challenging, and they need to be adjusted and optimized for specific systems, which increases the complexity and cost of implementation. Therefore, the present invention provides a method for fault diagnosis to solve the above problems.

[0071] Please refer to Figure 1 , is a flow chart of a fault diagnosis method provided by an embodiment of the present invention, comprising the following steps:

[0072] Step S01, obtaining a first characteristic data group, where the first characteristic data group is one or more groups of first characteristic data of a preset number of devices under the same working conditions.

[0073] In this embodiment, the first characteristic data group is one or more groups of key performance indicator data collected when monitoring a preset number of devices under specific working conditions. These data may include, but are not limited to, device temperature, evaporator pressure, condenser pressure, air conditioning compressor voltage, energy consumption, operating status, vibration data, noise level, wind speed, filter status, etc. The purpose of performing this step is to ensure that the collected data can represent the typical performance of the device under specific working conditions, thereby providing an accurate basis for subsequent data analysis and fault diagnosis. Through these data, the performance of the equipment can be compared horizontally, and any behavior that deviates from the normal operating mode can be identified, thereby providing a basis for early detection and diagnosis of faults.

[0074] In some embodiments, obtaining the first feature data set may be specifically achieved by executing the following steps:

[0075] 1) Determine the monitoring indicators: Determine the key performance indicators that need to be monitored based on the equipment type and operating requirements.

[0076] 2) Select equipment samples: Select a certain number of equipment as samples, and these equipment are set to operate under the same or similar working conditions.

[0077] 3) Data collection: Use sensors or data acquisition systems to collect equipment operation data in real time to ensure the accuracy and timeliness of the data.

[0078] 4) Data synchronization: Ensure that data collection of all devices is synchronized in time to facilitate subsequent horizontal comparative analysis.

[0079] 5) Storage and management: Store the processed data in a database and establish a data management system to facilitate subsequent access and analysis.

[0080] Step S02: Calculate corresponding first feature thresholds according to the first feature data set using a preset SPOT algorithm, wherein the number of the first feature thresholds is consistent with the number of the first feature data set.

[0081] In this embodiment, the SPOT algorithm (Streaming Peaks-Over-Threshold) is a statistical process for outlier detection, which determines the abnormal points in the data by analyzing the distribution of the data. The present invention uses the first characteristic data group of the SPOT algorithm to calculate the corresponding first characteristic threshold, the purpose of which is to determine the normal fluctuation range of the data, thereby identifying abnormal data points beyond this range, and determining that the abnormal data points indicate potential equipment failures. The present invention uses the SPOT algorithm to automatically set the threshold for fault diagnosis, so that the determined threshold is more scientific and accurate, reduces human errors, and helps to improve the accuracy of fault detection. At the same time, the SPOT algorithm can dynamically adjust the threshold according to the distribution of actual data, can adapt to different data sets and equipment states, and improve the adaptability of the method.

[0082] In some embodiments, the preset SPOT algorithm is used to calculate the corresponding first feature threshold value according to the first feature data group. Specifically, the SPOT algorithm is applied to analyze the data distribution of each performance indicator and estimate its probability distribution parameters, such as mean, standard deviation, etc. Based on these parameters, the quantiles of the first feature data group are calculated. For example, the 98th percentile of the normal distribution can be taken as the upper threshold and the 1st percentile as the lower threshold. These quantiles define the range of normal data fluctuations. If the data points monitored in real time exceed these preset thresholds, it may indicate that the device is abnormal.

[0083] In some embodiments, the first feature data group may be a single feature group or a plurality of feature groups, and correspondingly, the number of the first feature thresholds is consistent with the number of the first feature data groups.

[0084] Step S03: using a pre-trained support vector machine model to classify each of the first feature data based on the first feature threshold, and determining a fault category of each of the devices according to the classification result.

[0085] In this embodiment, the support vector machine (SVM) is a supervised learning algorithm, which is mainly used for classification and regression tasks. The purpose of the present invention using a pre-trained support vector machine model to classify each first feature data based on a first feature threshold is to identify and predict faults by analyzing the key performance indicators of the equipment. Specifically, the first feature threshold obtained in step S02 is used as input, and combined with the pre-trained SVM model, the actual operation data of the equipment is classified to determine whether it is in a normal state or has some kind of fault. For example, if the operating temperature of the air conditioner exceeds the preset high temperature threshold, the SVM model may classify it as an "overheating" fault category. In this way, the operating status of the equipment can be automatically and accurately identified, so that maintenance measures can be taken in time to avoid potential faults.

[0086] In some embodiments, pre-training of the support vector machine model may include the following steps:

[0087] 1) Data preprocessing: Preprocess the data, including feature scaling, decentralization, etc., to improve model performance and avoid numerical calculation problems.

[0088] 2) Select kernel function: SVM can use different kernel functions to handle nonlinear problems. Common kernel functions include linear kernel, polynomial kernel, radial basis function kernel, etc. The choice of kernel function depends on the characteristics of the specific problem and data.

[0089] 3) Set the regularization parameter: The regularization parameter C is an important hyperparameter in SVM, which controls the degree of penalty for misclassification. A larger C value can increase the model's fit to the training data, but may lead to overfitting; a smaller C value can make the model smoother, but may increase the risk of underfitting.

[0090] 4) Training model: Use the training data set to train the SVM model. During the training process, the model will learn the optimal weights and biases to correctly classify data points on the decision boundary. For linearly separable data, hard margin maximization can be used; for approximately linearly separable data, soft margin maximization can be used to introduce some classification errors, thereby improving the generalization ability of the model.

[0091] 5) Optimization and cross-validation: To select the best model parameters, techniques such as cross-validation can be used to evaluate the performance of different parameter combinations. By performing multiple iterations on the training set, the parameters that perform best on the validation set are found.

[0092] 6) Model evaluation: After training, use an independent test set to evaluate the performance of the SVM model. Common evaluation indicators include accuracy, precision, recall, and F1 score.

[0093] The present invention can effectively identify abnormal patterns from a large amount of equipment performance indicator data and classify the data into normal or different categories of fault states. For example, if the first characteristic data is the compressor current in the air-conditioning system, the normal fluctuation range of the current is determined by a preset threshold, and then the actual current value is classified using the SVM model. If the current value exceeds the normal range, the model can classify it as a "compressor overload" fault. This method can not only reduce the reliance on expert experience and improve the consistency and reliability of diagnosis, but also respond quickly and provide decision support for preventive maintenance of equipment, thereby reducing downtime and maintenance costs and improving the operating efficiency and life of equipment.

[0094] Based on the above technical solution, the present invention can more objectively identify abnormal behavior by comparing the performance of different devices under the same working conditions, reduce the risk of misjudgment caused by single device data, and improve the accuracy of diagnosis. At the same time, through horizontal comparison, it does not rely on the characteristics of specific devices. The threshold can be dynamically adjusted according to the actual data, so that the diagnostic system can adapt to the changes of different devices and environments, and can be widely used in different devices and systems, which improves the universality of the fault diagnosis method, reduces the dependence on expert experience, and makes the diagnosis process more objective.

[0095] On the basis of the above embodiments, in some embodiments, in order to further improve the accuracy of fault diagnosis, before executing step S03 to classify each of the first feature data based on the first feature threshold using the pre-trained support vector machine model, the initial support vector machine model may be trained to obtain the pre-trained support vector machine model. Figure 2 , is a flow chart of another fault diagnosis method provided by an embodiment of the present invention, comprising the following steps:

[0096] Step S11, obtaining a second characteristic data group, where the second characteristic data group is one or more groups of second characteristic data of a preset number of devices under the same working conditions.

[0097] In some embodiments, the second feature data group may be data with the same key performance indicators as the first feature data group. The initial support vector machine model is trained using the second feature data group with the same key performance indicators as the first feature data group to improve the classification accuracy of the pre-trained support vector machine model for the first feature data group.

[0098] Step S12: Calculate corresponding second feature thresholds according to the second feature data group using the preset SPOT algorithm, wherein the number of the second feature thresholds is consistent with the number of the second feature data group.

[0099] In some embodiments, as mentioned in step S12, the corresponding second feature threshold is calculated according to the second feature data set using the preset SPOT algorithm, which may specifically be:

[0100] The probability density of the second feature data group is determined as:

[0101]

[0102] Among them, X i is the second feature data set, f(X i ) is the probability density of the second feature data set, t is the initial comparison value of each window, x is a real number that satisfies any 1+γx>0, γ is a shape parameter, σ(t) is a scale parameter, and τ is a limit value of the distribution;

[0103] The maximum likelihood function of the second feature data group is determined according to the probability density as follows:

[0104]

[0105] The maximum likelihood function is used to determine the maximum log-likelihood function of the second feature data group as follows:

[0106]

[0107] Among them, N t is the number of abnormal feature data in the second feature data group;

[0108] Determining a predicted value of the shape parameter and a predicted value of the scale parameter according to the maximized log-likelihood function;

[0109] Calculating a feature threshold of the second feature data group according to the predicted value of the shape parameter and the predicted value of the scale parameter using a preset SPOT algorithm;

[0110]

[0111] Among them, z q is the second feature threshold of the second feature data group, including a second feature lower limit value and a second feature upper limit value, p is the initial probability value, and n is the total data volume.

[0112] Step S13: Calculate the second physical property value corresponding to each second feature data by using the physical property definition function in combination with the second feature threshold.

[0113] In this embodiment, the purpose is to map the original second feature data into attribute values ​​with physical meaning through the physical attribute definition function, so as to intuitively reflect the actual operating status and potential abnormal conditions of the equipment. Specifically, the physical attribute definition function will consider the thresholds of the second feature data group, such as the upper and lower limits, and possible other statistics (such as quantiles), and convert these data into physical properties that describe the equipment status, such as "too large", "normal" or "too small". Such a conversion is helpful for subsequent analysis and decision-making, so that the machine learning model can better understand and process the actual operating status of the equipment, thereby improving the accuracy and efficiency of fault diagnosis.

[0114] In some embodiments, as mentioned in step S13, the second physical property value corresponding to each second feature data is calculated by using the physical property definition function in combination with the second feature threshold, which may specifically be:

[0115] Calculating a first preset quantile and a second preset quantile of the second feature data group;

[0116] Each of the second feature data is mapped to a corresponding second physical property value by using the physical property definition function in combination with the second feature lower limit value, the second feature upper limit value, the first preset quantile, and the second preset quantile:

[0117]

[0118] Among them, A(X i ) is the physical property value, l is the second characteristic lower limit value of the second characteristic data group, q1 is the first preset quantile, q2 is the second preset quantile, and u is the second characteristic upper limit value of the second characteristic data group.

[0119] In some embodiments, please refer to Figure 3 , is a schematic diagram of a function of a physical property value provided by an embodiment of the present invention. Figure 3 As shown, the qualitative description and quantitative score of the key performance indicators can be combined to achieve a two-way conversion of the qualitative description and quantitative score of the key performance indicators. That is, the second physical attribute value of the second feature data can be converted into the corresponding qualitative description:

[0120]

[0121] Among them, Y i A qualitative description of a physical property value.

[0122] Further, such as Figure 3As shown, the embodiment of the present invention, based on the fuzzy membership function, combines the quadratic parabolic distribution to partially correct the undersized and oversized values ​​therein, so as to further improve the accuracy of model recognition.

[0123] In some embodiments, in order to capture the distribution characteristics of the data in the second feature data set, the first preset quantile may be the 25% quantile, and the second preset quantile may be the 75% quantile. The user may also modify the first preset quantile and the second preset quantile according to personal choice or industry standards. For example, before calculating the first preset quantile and the second preset quantile of the second feature data set, the method further includes:

[0124] An input modification instruction is received, and the first preset quantile and the second preset quantile are modified.

[0125] Based on the above technical solution, the embodiment of the present invention captures the distribution characteristics of the data by determining specific quantiles in the second feature data group. Then, through the preset physical property definition function, these quantiles are combined with the second feature threshold, and each second feature data is evaluated and classified, and a physical property value is assigned to it. This value can be a qualitative description (such as "normal", "high", "low") or a quantitative score, so that it can be used in subsequent analysis or model training to enhance the model's understanding and classification capabilities of the device status.

[0126] Step S14, receiving the labeling information corresponding to each second physical property value, and training the initial support vector machine model according to each second physical property value and the corresponding labeling information to obtain the pre-trained support vector machine model.

[0127] In this embodiment, the initial support vector machine model learns to distinguish different equipment states by analyzing each physical attribute value and its corresponding actual label (i.e., fault category or state). In this way, the support vector machine model can recognize and understand the relationship between different physical attribute values ​​and equipment failures, so that new data can be accurately classified and predicted after training is completed. This method makes the support vector machine model more accurate and reliable in practical applications in the field of equipment fault diagnosis.

[0128] In some embodiments, as mentioned in step S14, the initial support vector machine model is trained according to each of the second physical property values ​​and the corresponding annotation information to obtain the pre-trained support vector machine model, which may specifically be:

[0129] The hyperplane linear equation for determining the initial support vector machine model is:

[0130] ωA(X i )+b=0

[0131] Among them, ω is the weight and b is the bias;

[0132] Determine the objective function of the initial support vector machine model according to the hyperplane linear equation:

[0133]

[0134] Among them, ω is the weight, b is the bias, ξ i is a slack variable, parameter C is a penalty factor, y i To mark information;

[0135] The initial support vector machine model is trained according to each of the second physical property values ​​and the corresponding annotation information to obtain the weight and the bias:

[0136]

[0137] Among them, α i is the KKT conditional multiplier.

[0138] In some embodiments, after determining the objective function of the initial support vector machine model according to the hyperplane linear equation, the constraints can be introduced into the objective function by means of Karush-Kuhn-Tucker (KKT) multipliers α and β, and the optimization objective can be converted into a dual problem:

[0139]

[0140] Then, based on the relaxed complementary conditions of the KKT conditions, we obtain the expressions of the above weights and biases.

[0141] Step S01, obtaining a first characteristic data group, where the first characteristic data group is one or more groups of first characteristic data of a preset number of devices under the same working conditions.

[0142] Step S02: Calculate corresponding first feature thresholds according to the first feature data set using a preset SPOT algorithm, wherein the number of the first feature thresholds is consistent with the number of the first feature data set.

[0143] In some embodiments, as mentioned in step S02, the corresponding first feature threshold is calculated according to the first feature data group using a preset SPOT algorithm, which may be specifically the same as the method of calculating the second feature threshold mentioned in step S12, and will not be repeated here.

[0144] Step S03: using a pre-trained support vector machine model to classify each of the first feature data based on the first feature threshold, and determining a fault category of each of the devices according to the classification result.

[0145] Based on the above embodiments, in some embodiments, as mentioned in step S03, each of the first feature data is classified based on the first feature threshold using a pre-trained support vector machine model, and the fault category of each of the devices is determined according to the classification result, which may specifically be:

[0146] Mapping each of the first characteristic data to a corresponding first physical property value using the physical property definition function;

[0147] Each of the first physical property values ​​is input into the hyperplane linear equation for calculation, and the fault category of each of the devices is determined according to the calculation result.

[0148] In this embodiment, by using a physical property definition function to convert the first feature data into a corresponding first physical property value, the actual operating status of the device can be more intuitively reflected. Then, the first physical property value is used as input and applied to the hyperplane linear equation of the support vector machine model for calculation. The fault category of each device is determined based on the output result of the support vector machine model, that is, the position of the data point in the hyperplane. If the data point is on one side of the hyperplane, it may be classified as a "normal" state; on the other side, it may be classified as a specific "fault" state. This method combines the intuitiveness of physical properties and the classification ability of the support vector machine model, improving the accuracy and efficiency of fault diagnosis.

[0149] Based on the above technical solution, the embodiment of the present invention introduces a second feature data group and uses the SPOT algorithm to adaptively determine the feature threshold, combines the physical attribute definition function to map to the physical attribute value, and then trains the support vector machine model through the annotation information, thereby enhancing the model's ability to identify the device state and improving the accuracy and robustness of fault diagnosis. This adaptive adjustment mechanism ensures that the model can be dynamically adjusted according to the distribution characteristics of the actual data, so as to better adapt to the unevenness of the data, reduce misjudgment, and optimize the performance of fault prediction, especially in the face of changing working conditions and complex equipment behaviors, it can achieve more accurate and reliable diagnosis results.

[0150] Please refer to Figure 4 , is a schematic diagram of a fault diagnosis provided by an embodiment of the present invention. Figure 4 As shown in the figure, taking a certain fault of a certain device as an example, the operating data of 100 devices with the same working conditions are compared horizontally. This example mainly monitors the operating status of the device, and the influencing factors are variables 1 and variable 2. The same working condition data of the normal operation of the device is taken as the training sample, and the time window is slid with the number of devices 100 as the time window, and the key performance indicators of variable 1 of these 100 devices are qualitatively described.

[0151] Preliminary calculations show that the upper limit u = 35 and the lower limit l = 15. At the same time, as the amount of data increases in the later period, u and l can be updated in real time. Combined with the current training set, q1 = 21 and q2 = 29 can be obtained. Therefore, combining the above physical properties to define the function, we can get:

[0152]

[0153] Similarly, we can calculate the u, l, q1 and q2 of variable 2 and the corresponding physical property value A(X i ). Then use the above qualitative description formula to accurately classify the data and divide the data samples into faulty and normal. i ) data set is imported into the model, and the support vector machine model will automatically classify normal samples and abnormal samples, thereby realizing real-time fault diagnosis of the equipment.

[0154] Please refer to Figure 5 , Figure 6 and Figure 7 , Figure 5 A schematic diagram of the overall situation of a sample provided by an embodiment of the present invention; Figure 6 Based on Figure 5 The overall situation of the samples shown is a schematic diagram of the hyperplane obtained through the support vector machine model; Figure 7 Based on Figure 6 Schematic diagram of the hyperplane shown for classification.

[0155] Based on the above technical scheme, the present invention can complete the fault diagnosis of the equipment according to the above steps. This embodiment only describes the situation of dividing the two categories of fault and normal under a certain working condition. In some embodiments, more data features can be used for multi-classification and diagnosis and positioning of faults, such as fault 1, fault 2, fault 3 and normal.

[0156] Please refer to Figure 8 , is a schematic diagram of the structure of a fault diagnosis device provided by an embodiment of the present invention, the fault diagnosis device comprising:

[0157] A first acquisition module 100 is used to acquire a first characteristic data group, where the first characteristic data group is one or more groups of first characteristic data of a preset number of devices under the same working conditions;

[0158] A first calculation module 200, configured to calculate corresponding first feature thresholds according to the first feature data group using a preset SPOT algorithm, wherein the number of the first feature thresholds is consistent with the number of the first feature data group;

[0159] The classification module 300 is used to classify each of the first feature data based on the first feature threshold using a pre-trained support vector machine model, and determine the fault category of each of the devices according to the classification result.

[0160] On the basis of the above embodiment, in a specific embodiment, the fault diagnosis device may further include:

[0161] A second acquisition module, used to acquire a second characteristic data group, where the second characteristic data group is one or more groups of second characteristic data of a preset number of devices under the same working conditions;

[0162] A second calculation module, used for calculating corresponding second feature thresholds according to the second feature data group by using the preset SPOT algorithm, wherein the number of the second feature thresholds is consistent with the number of the second feature data group;

[0163] A third calculation module, configured to calculate a second physical property value corresponding to each second feature data by using a physical property definition function in combination with the second feature threshold;

[0164] The training module is used to receive the labeling information corresponding to each second physical property value, and train the initial support vector machine model according to each second physical property value and the corresponding labeling information to obtain the pre-trained support vector machine model.

[0165] Based on the above embodiment, in a specific embodiment, the second calculation module is specifically used for:

[0166] The probability density of the second feature data group is determined as:

[0167]

[0168] Among them, X i is the second feature data set, f(X i ) is the probability density of the second feature data set, t is the initial comparison value of each window, x is a real number that satisfies any 1+γx>0, γ is a shape parameter, σ(t) is a scale parameter, and τ is a limit value of the distribution;

[0169] The maximum likelihood function of the second feature data group is determined according to the probability density as follows:

[0170]

[0171] The maximum likelihood function is used to determine the maximum log-likelihood function of the second feature data group as follows:

[0172]

[0173] Among them, N t is the number of abnormal feature data in the second feature data group;

[0174] Determining a predicted value of the shape parameter and a predicted value of the scale parameter according to the maximized log-likelihood function;

[0175] Calculating a feature threshold of the second feature data group according to the predicted value of the shape parameter and the predicted value of the scale parameter using a preset SPOT algorithm;

[0176]

[0177] Among them, z q is the second feature threshold of the second feature data group, including a second feature lower limit value and a second feature upper limit value, p is the initial probability value, and n is the total data volume.

[0178] Based on the above embodiment, in a specific embodiment, the third calculation module is specifically used for:

[0179] Calculating a first preset quantile and a second preset quantile of the second feature data group;

[0180] Each of the second feature data is mapped to a corresponding second physical property value by using the physical property definition function in combination with the second feature lower limit value, the second feature upper limit value, the first preset quantile, and the second preset quantile:

[0181]

[0182] Among them, A(X i ) is the physical property value, l is the second characteristic lower limit value of the second characteristic data group, q1 is the first preset quantile, q2 is the second preset quantile, and u is the second characteristic upper limit value of the second characteristic data group.

[0183] Based on the above embodiment, in a specific embodiment, the training module is specifically used for:

[0184] The hyperplane linear equation for determining the initial support vector machine model is:

[0185] ωA(X i )+b=0

[0186] Among them, ω is the weight and b is the bias;

[0187] Determine the objective function of the initial support vector machine model according to the hyperplane linear equation:

[0188]

[0189] Among them, ω is the weight, b is the bias, ξ i is a slack variable, parameter C is a penalty factor, y i To mark information;

[0190] The initial support vector machine model is trained according to each of the second physical property values ​​and the corresponding annotation information to obtain the weight and the bias:

[0191]

[0192] Among them, α i is the KKT conditional multiplier.

[0193] Based on the above embodiment, in a specific embodiment, the classification module 300 is specifically used for:

[0194] Mapping each of the first characteristic data to a corresponding first physical property value using the physical property definition function;

[0195] Each of the first physical property values ​​is input into the hyperplane linear equation for calculation, and the fault category of each of the devices is determined according to the calculation result.

[0196] Based on the above embodiment, in a specific embodiment, the third calculation module is further used for:

[0197] An input modification instruction is received, and the first preset quantile and the second preset quantile are modified.

[0198] This embodiment provides an electronic device, including a processor and a memory, the memory is used to store at least one instruction, and when the instruction is loaded and executed by the processor, the above-mentioned fault diagnosis method is implemented. Its execution method and beneficial effects are similar and will not be repeated here.

[0199] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the above-mentioned fault diagnosis method is implemented. The execution method and beneficial effects thereof are similar and will not be described in detail here.

[0200] It should be noted that although the above describes the various steps in a specific order, it does not mean that the steps must be executed in the above specific order. In fact, some of these steps can be executed concurrently or even in a different order as long as the required functions can be achieved.

[0201] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for fault diagnosis, It is characterized in that include: Acquire a first characteristic data group, where the first characteristic data group is one or more groups of first characteristic data of a preset number of devices under the same working conditions; Calculating corresponding first feature thresholds according to the first feature data group using a preset SPOT algorithm, wherein the number of the first feature thresholds is consistent with the number of the first feature data group; A pre-trained support vector machine model is used to classify each of the first feature data based on the first feature threshold, and a fault category of each of the devices is determined according to the classification result.

2. The method according to claim 1, It is characterized in that Before classifying each of the first feature data based on the first feature threshold using the pre-trained support vector machine model, the method further includes: Acquire a second characteristic data group, where the second characteristic data group is one or more groups of second characteristic data of a preset number of devices under the same working conditions; Calculating corresponding second feature thresholds according to the second feature data group using the preset SPOT algorithm, wherein the number of the second feature thresholds is consistent with the number of the second feature data group; Calculate the second physical property value corresponding to each second feature data by using a physical property definition function combined with the second feature threshold; The labeling information corresponding to each second physical property value is received, and an initial support vector machine model is trained according to each second physical property value and the corresponding labeling information to obtain the pre-trained support vector machine model.

3. The method according to claim 2, It is characterized in that The using the preset SPOT algorithm to calculate the corresponding second feature threshold according to the second feature data group includes: The probability density of the second feature data group is determined as: Among them, X i is the second feature data set, f(X i ) is the probability density of the second feature data set, t is the initial comparison value of each window, x is a real number that satisfies any 1+γx>0, γ is a shape parameter, σ(t) is a scale parameter, and τ is a limit value of the distribution; The maximum likelihood function of the second feature data group is determined according to the probability density as follows: The maximum likelihood function is used to determine the maximum log-likelihood function of the second feature data group as follows: Among them, N t is the number of abnormal feature data in the second feature data group; Determining a predicted value of the shape parameter and a predicted value of the scale parameter according to the maximized log-likelihood function; Calculating a feature threshold of the second feature data group according to the predicted value of the shape parameter and the predicted value of the scale parameter using a preset SPOT algorithm; Among them, z q is the second feature threshold of the second feature data group, including a second feature lower limit value and a second feature upper limit value, p is the initial probability value, and n is the total data volume.

4. The method according to claim 3, It is characterized in that The calculating the second physical property value corresponding to each second feature data by using the physical property definition function in combination with the second feature threshold comprises: Calculating a first preset quantile and a second preset quantile of the second feature data group; Each of the second feature data is mapped to a corresponding second physical property value by using the physical property definition function in combination with the second feature lower limit value, the second feature upper limit value, the first preset quantile, and the second preset quantile: Among them, A(X i ) is the physical property value, l is the second characteristic lower limit value of the second characteristic data set, q 1 is the first preset quantile, q 2 is the second preset quantile, and u is the second feature upper limit value of the second feature data group.

5. The method according to claim 4, It is characterized in that The step of training the initial support vector machine model according to each of the second physical property values ​​and the corresponding annotation information to obtain the pre-trained support vector machine model includes: The hyperplane linear equation for determining the initial support vector machine model is: ωA(X i )+b=0 Among them, ω is the weight and b is the bias; Determine the objective function of the initial support vector machine model according to the hyperplane linear equation: Among them, ω is the weight, b is the bias, ξ i is a slack variable, parameter C is a penalty factor, y i To mark information; The initial support vector machine model is trained according to each of the second physical property values ​​and the corresponding annotation information to obtain the weight and the bias: Among them, α i is the KKT conditional multiplier.

6. The method according to claim 5, It is characterized in that The method of using the pre-trained support vector machine model to classify each of the first feature data based on the first feature threshold, and determining the fault category of each of the devices according to the classification result, includes: Mapping each of the first characteristic data to a corresponding first physical property value using the physical property definition function; Each of the first physical property values ​​is input into the hyperplane linear equation for calculation, and the fault category of each of the devices is determined according to the calculation result.

7. The method according to claim 4, It is characterized in that Before calculating the first preset quantile and the second preset quantile of the second feature data group, the method further includes: An input modification instruction is received, and the first preset quantile and the second preset quantile are modified.

8. A fault diagnosis device, It is characterized in that include: A first acquisition module, used to acquire a first characteristic data group, where the first characteristic data group is one or more groups of first characteristic data of a preset number of devices under the same working conditions; A first calculation module, configured to calculate corresponding first feature thresholds according to the first feature data group using a preset SPOT algorithm, wherein the number of the first feature thresholds is consistent with the number of the first feature data group; A classification module is used to classify each of the first feature data based on the first feature threshold using a pre-trained support vector machine model, and determine a fault category of each of the devices according to the classification result.

9. An electronic device, It is characterized in that include: A processor and a memory, wherein the memory is used to store at least one instruction, and when the instruction is loaded and executed by the processor, the fault diagnosis method as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the fault diagnosis method according to any one of claims 1 to 7 is implemented.

Citation Information

Patent Citations

  • Electric power system fault reporting system and method based on heterogeneous networking

    CN118100415A

  • Photovoltaic power generation fault prediction and diagnosis system and method based on data driving

    CN118232832A