Fault classification method, system, controller and storage medium
Patent Information
- Application Number
- CN202510564976.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-29
- Publication Date
- 2026-10-30
AI Technical Summary
[0003]本申请所要解决的技术问题在于，提供一种故障分类方法、故障分类系统及控制器和计算机可读存储介质，以解决现有故障分类中容易忽略特征判别性、易受到数据噪声干扰以及无法自适应不同故障类型等问题
[0055]本申请与现有技术相比具有明显的优点和有益效果。借由上述技术方案，本申请的故障分类方法、故障分类系统系统及控制器和计算机可读存储介质可以达到相当的技术进步性及实用性，并具有产业上的广泛利用价值，其至少具有下列优点：
Smart Images
Figure CN123012786A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of fault identification technology, specifically relating to fault classification methods, fault classification systems and controllers, and computer-readable storage media. Background Technology
[0002] When performing quality inspection on process data (such as bolt tightening data), some defect data may be found. By classifying this defect data, the cause of the defect can be quickly determined, which is beneficial for targeted prevention and process improvement. A pre-trained fault classification model is typically used to classify the defect data. A single model is usually used for training, such as an autoencoder (including encoder and decoder) or a random forest model. Traditional fault classification models rely on expert experience, making it difficult to handle complex fault types. Furthermore, using only an autoencoder can easily overlook feature discriminative power, and directly processing raw defect data using only a random forest model is susceptible to noise interference. Additionally, the static feature extraction method used during the training process makes the fault classification model unable to adapt to different fault types. Summary of the Invention
[0003] The technical problem to be solved by this application is to provide a fault classification method, a fault classification system and controller and a computer-readable storage medium to solve the problems of existing fault classification methods, such as easy neglect of feature discrimination, susceptibility to data noise interference and inability to adapt to different fault types.
[0004] To solve or improve the aforementioned technical problems to some extent, according to one aspect of this application, a fault classification method is provided, which applies a pre-trained fault classification system. The fault classification method includes:
[0005] The first fault data is preprocessed to obtain the target fault data;
[0006] The target fault data is input into the fault classification system to obtain the fault type corresponding to the first fault data;
[0007] The fault classification system includes an autoencoder and a random forest model. The autoencoder includes an encoder and a decoder. The training method for the fault classification system includes:
[0008] The encoder is controlled to extract multiple first automatically extracted features from the second fault data, which are used as multiple first target features.
[0009] The importance of a first feature of multiple first target features is evaluated using the random forest model, and the first weight of the multiple first target features is adjusted based on the importance of the first feature.
[0010] The decoder is controlled to construct reconstructed fault data based on multiple first target features after adjusting the first weight;
[0011] The encoder and the decoder are optimized based on the reconstruction error between the second fault data and the reconstructed fault data to complete the training of the fault classification system.
[0012] In some implementations, the training method for the fault classification system includes:
[0013] Receive multiple first manually extracted features;
[0014] Each of the first automatically extracted features and the first manually extracted features is assigned a second weight.
[0015] The first target feature is obtained by fusing multiple first automatically extracted features and multiple first manually extracted features that are assigned the second weight.
[0016] In some implementations, the step of evaluating the first feature importance of multiple first target features using the random forest model, and adjusting the first weights of the multiple first target features based on the first feature importance, includes:
[0017] Multiple first target features are input into the random forest model;
[0018] The importance of the first feature of the multiple first target features is determined based on the contribution of the first target features to the reduction of Gini impurity in each decision tree split node of the random forest model;
[0019] Based on the importance of the first feature corresponding to each of the multiple first target features, the first weights corresponding to the multiple first target features are determined respectively.
[0020] In some implementations, the step of determining the importance of the first features of the plurality of first target features based on their contribution to the reduction of Gini impurity at each split node of the decision tree in the random forest model includes:
[0021] Calculate the contribution of each of the first target features to the reduction of Gini impurity at each node of the random forest model;
[0022] The total Gini importance of each first target feature is obtained by summing the contribution values of each first target feature to the reduction of Gini impurity across all nodes of the random forest model.
[0023] The importance of the first feature for each first target feature is evaluated based on the total Gini importance.
[0024] In some implementations, the formula for calculating the Gini impurity of each of the first target features at each node is as follows:
[0025]
[0026] In the formula, P k It is the probability that a sample in node n belongs to fault category k, where K is the number of fault categories;
[0027] The formula for calculating the Gini importance of each of the first target features in each decision tree of the random forest model is as follows:
[0028]
[0029] In the formula, GiniImp(F j (,t) represents the first objective feature F in the t-th decision tree. j The importance of the gin, T t It is the set of all nodes in the t-th decision tree; |n samples | is the number of samples in node n; Gini(n) is the Gini impurity of node n; Gini(n) left ) and Gini(n right ) are the Gini impurities of the left and right child nodes, respectively; |n left | and |n right | These represent the number of samples for the left and right child nodes, respectively; |T tsamples | is the total number of samples in the t-th decision tree;
[0030] The formula for calculating the total Gini importance of each of the first target features is as follows:
[0031]
[0032] In the formula, w j The first target feature F j Total Gini Importance, GiniImp(F j (,t) represents the first objective feature F in the t-th decision tree. j The Gini importance is given by T, where T is the number of decision trees in the random forest model.
[0033] In some embodiments, the fault classification system training method further includes:
[0034] An attention mechanism is introduced into the autoencoder to assign a third weight to each feature in the second fault data.
[0035] Multiple first automatically extracted features are extracted based on the third weight.
[0036] In some implementations, the step of inputting the target fault data into the fault classification system to obtain the fault type corresponding to the first fault data includes:
[0037] The encoder extracts multiple second automatically extracted features from the target fault data, which serve as multiple second target features.
[0038] Multiple second target features are input into the random forest model, and the fault type corresponding to the first fault data is obtained based on the voting mechanism of multiple decision trees in the random forest model.
[0039] In some implementations, the fault classification method includes:
[0040] Receive multiple second manually extracted features;
[0041] A fourth weight is assigned to each of the second automatically extracted features and the second manually extracted features;
[0042] The second target feature is obtained by fusing multiple automatically extracted features and multiple manually extracted features that are assigned the fourth weight.
[0043] In some implementations, the step of obtaining the fault type corresponding to the first fault data through the voting mechanism of multiple decision trees based on the random forest model includes:
[0044] The importance of the second features corresponding to multiple second target features is evaluated using the random forest model.
[0045] Based on the importance of the second feature, a fifth weight is assigned to each of the second target features;
[0046] The fault type corresponding to the first fault data is determined based on multiple second target features assigned the fifth weight.
[0047] According to one embodiment of this application, a fault classification system is provided, wherein the fault classification system is configured to preprocess first fault data to obtain target fault data and classify the faults to obtain the fault type corresponding to the first fault data;
[0048] The fault classification system includes an autoencoder and a random forest model, wherein the autoencoder includes an encoder and a decoder;
[0049] The encoder is configured to extract multiple first automatically extracted features from the second fault data as multiple first target features;
[0050] The random forest model evaluates the importance of a first feature among multiple first target features, and adjusts the first weights of the multiple first target features based on the importance of the first feature;
[0051] The decoder constructs reconstructed fault data based on multiple first target features after adjusting the first weights;
[0052] The encoder and the decoder are further configured to optimize the encoder and the decoder based on the reconstruction error between the second fault data and the reconstructed fault data, thereby completing the training of the fault classification system.
[0053] According to one embodiment of this application, a controller is provided, which includes a memory and a processor. The memory stores a computer program that, when executed by the processor, can implement the steps of the fault classification method described in any of the above embodiments.
[0054] According to one embodiment of this application, a computer-readable storage medium is provided for storing a computer program that, when executed by a computer or processor, implements the steps of the fault classification method described in any of the above embodiments.
[0055] This application has significant advantages and beneficial effects compared with the prior art. Through the above technical solution, the fault classification method, fault classification system, controller, and computer-readable storage medium of this application achieve considerable technological advancement and practicality, and have broad industrial application value. It possesses at least the following advantages:
[0056] This application introduces a feature importance feedback mechanism based on random forests to dynamically optimize the features extracted by the autoencoder. This means that feature selection and optimization are automatic, independent of human experience. Through real-time feedback and dynamic adjustment of features, the autoencoder can automatically learn the features most discriminative for fault identification and adapt to more complex fault modes. An attention mechanism is introduced into the autoencoder, enabling the model to automatically learn which features are most important in fault classification. The weight of each feature is dynamically adjusted according to its contribution to the classification task, allowing the model to focus more on key features and improve the accuracy of fault identification. Combining deep features extracted by the autoencoder (i.e., automatically extracted features) with manually extracted features enhances the diversity and richness of features. By assigning different weights to each feature, the complementarity of multimodal data can be maximized, further improving the fault classification performance of the system.
[0057] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description
[0058] Figure 1 This is a schematic block diagram of a fault classification system according to an embodiment of this application;
[0059] Figure 2 This is a flowchart illustrating a training method for a fault classification system according to an embodiment of this application.
[0060] Figure 3 This is a flowchart illustrating a fault classification method according to an embodiment of this application;
[0061] Figure 4 This is a schematic block diagram of the controller according to an embodiment of this application. Detailed Implementation
[0062] 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.
[0063] 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,” “the,” and “the” as used in one or more embodiments of this specification and in the appended claims are also intended to include the stated forms unless the context clearly indicates otherwise.
[0064] It should be understood that although the terms first, second, third, fourth, 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."
[0065] To further illustrate the technical means and effects adopted by this application to achieve the intended purpose, the following detailed description, in conjunction with the accompanying drawings and preferred embodiments, explains the specific implementation methods and effects of the fault classification method, fault classification system and controller and readable storage medium proposed in this application.
[0066] This application proposes a fault classification method and a fault classification system. The fault classification method is applied to the fault classification system. By introducing a feature importance feedback mechanism of random forest, the fault classification system dynamically optimizes the features extracted by the encoder, realizing automatic feature selection and optimization without relying on human experience. Through real-time feedback and dynamic adjustment of the extracted features, the autoencoder (including encoder and decoder) can autonomously learn the features with the most discriminative power for fault classification, so as to adapt to more complex fault forms.
[0067] In one embodiment, such as Figure 1 As shown, the fault classification system of this application includes an autoencoder and a random forest model, wherein the autoencoder includes an encoder and a decoder.
[0068] In this embodiment, such as Figure 1 and Figure 2 As shown, the method for pre-training this fault classification system includes:
[0069] Step S10: Control the encoder to extract multiple first automatically extracted features from the second fault data as multiple first target features.
[0070] In this step, such as Figure 1 As shown, the pre-collected second fault data X (i.e., the original fault data) is input into the encoder. The encoder extracts multiple first automatically extracted features from the second fault data through its multi-layer neural network, which serve as multiple first target features for fault classification.
[0071] Specifically, the second fault data X is the original high-dimensional data, such as sensor data, image data, workpiece tightening data, etc. The encoder can extract low-dimensional potential features (i.e., the first automatically extracted features) from this original high-dimensional data, such as contour, color, texture and other features extracted from image data, and average torque, maximum torque, and torque curve slope extracted from workpiece tightening data.
[0072] In one embodiment, multiple first manually extracted features extracted based on the second fault data are weighted and fused with multiple first automatically extracted features extracted in step S10. This effectively enhances the diversity and richness of features in fault classification and maximizes the use of the complementarity of multimodal data to improve the performance of the fault classification system.
[0073] Specifically, methods for weighted fusion of multi-source features include:
[0074] Step 100: Receive multiple first manually extracted features.
[0075] In this step, based on expert experience, multiple first manually extracted features are extracted from the second fault data, and the extracted first manually extracted features are input into the fault classification system.
[0076] Step 102: Assign second weights to the multiple first automatically extracted features and the multiple first manually extracted features respectively.
[0077] In this step, depending on the actual situation, the second weights assigned to the multiple first automatically extracted features and the multiple first manually extracted features may be different, or the same second weight may be assigned to the multiple first automatically extracted features and the multiple first manually extracted features.
[0078] Step 104: Merge multiple first automatically extracted features and multiple first manually extracted features, which are assigned a second weight, as the first target feature.
[0079] In this step, after assigning second weights to multiple first automatically extracted features and multiple first manually extracted features, the multiple first automatically extracted features and multiple first manually extracted features are fused based on the assigned second weights. The fused multiple first automatically extracted features and multiple first manually extracted features are used as multiple first target features for fault classification. The specific fusion method is shown in the following formula:
[0080] F'=W_F×F+W_H×H
[0081] In the above formula, F' is the first target feature, F is the first automatically extracted feature, W_F is the second weight assigned to the first automatically extracted feature, H is the first manually extracted feature, and W_H is the second weight assigned to the first manually extracted feature.
[0082] Step S20: Evaluate the importance of the first features of multiple first target features using a random forest model, and adjust the first weights of the multiple first target features based on the importance of the first features.
[0083] In this step, to ensure that the extracted first target features can effectively support the subsequent fault classification task, a feature importance feedback mechanism based on a random forest model is introduced. A pre-trained random forest model is used to evaluate the first feature importance of the first target features extracted by the encoder, and this first feature importance is fed back to the autoencoder to dynamically adjust the first weights of multiple first target features, thereby optimizing the direction of feature learning.
[0084] In one embodiment, step S20 includes:
[0085] Step 200: Input multiple first target features into the random forest model.
[0086] In this step, the first target feature input into the random forest model can be multiple first automatically extracted features extracted in step S10, or multiple first automatically extracted features and multiple first manually extracted features that are fused and given a second weight in step 104.
[0087] Step 202: Determine the first feature importance of multiple first target features based on their contribution to the reduction of Gini impurity in each decision tree split node of the random forest model.
[0088] In this step, the importance of multiple first-target features (i.e., first-feature importance) is measured by calculating the contribution of multiple first-target features to the reduction of Gini impurity in each split node of the decision tree of the random forest model.
[0089] In one embodiment, step 202 includes:
[0090] Step 2020: Calculate the contribution of each first target feature to the reduction of Gini impurity at each node of the random forest model.
[0091] In this step, the formula for calculating the Gini impurity of each of the first target features at each node is as follows:
[0092]
[0093] In the formula, P k Let K be the probability that a sample in node n belongs to fault category k, where K is the number of fault categories. In a random forest model, when a node splits, if the first objective feature is used to make the decision, this first objective feature will reduce the Gini impurity of the node. Therefore, the contribution value of the first objective feature can be determined by the reduction of the node's Gini impurity.
[0094] Step 2022: Sum the contribution values of each first target feature to the reduction of Gini impurity in all nodes of the random forest model to obtain the total Gini importance of each first target feature.
[0095] In this step, the formula for calculating the Gini importance of each first target feature in each decision tree of the random forest model is as follows:
[0096]
[0097] In the formula, GiniImp(F j (,t) represents the first objective feature F in the t-th decision tree. j The importance of the gin, T t It is the set of all nodes in the t-th decision tree; |n samples| is the number of samples in node n; Gini(n) is the Gini impurity of node n; Gini(n) left ) and Gini(n right ) are the Gini impurities of the left and right child nodes, respectively; |n left | and |n right | These represent the number of samples for the left and right child nodes, respectively; |T tsamples | is the total number of samples in the t-th decision tree;
[0098] The formula for calculating the total Gini importance of each first target feature is as follows:
[0099]
[0100] In the formula, w j The first target feature F j Total Gini Importance, GiniImp(F j (,t) represents the first objective feature F in the t-th decision tree. j The Gini importance is given by T, where T is the number of decision trees in the random forest model.
[0101] Step 2024: Evaluate the first feature importance of each first target feature based on the total Gini importance.
[0102] In this step, after calculating the total Gini importance of each first target feature, the importance of its first feature can be evaluated based on the total Gini importance of each first target feature, which can effectively measure the magnitude of the influence of the first target feature on decision-making in the random forest model.
[0103] Step 204: Determine the first weights corresponding to the multiple first target features based on the importance of the first features corresponding to the multiple first target features.
[0104] In this step, after determining the importance of each of the multiple first target features, a first weight can be determined for each first target feature based on its importance. This first weight can then be used to adjust the weights of the features extracted by the encoder, thereby optimizing the learning direction of feature extraction.
[0105] Step S30: Control the decoder to construct reconstructed fault data based on multiple first target features after adjusting the first weight.
[0106] In this step, after adjusting the first weights of multiple first target features, the decoder constructs reconstructed fault data based on the multiple first target features and their corresponding first weights.
[0107] Step S40: Optimize the encoder and decoder based on the reconstruction error between the second fault data and the reconstructed fault data to complete the training of the fault classification system.
[0108] In this step, the second fault data X (i.e., the original fault data) input to the encoder is first calculated, along with the reconstructed fault data constructed by the decoder. Reconstruction error between That is, the second fault data X and the reconstructed fault data The difference between them. Then, based on the calculated reconstruction error... The encoder and decoder are trained to optimize them. During training, the reconstruction error is used as a reference. To optimize the weights of the encoder and decoder, and reduce the difference between the second faulty data X and the reconstructed faulty data. Reconstruction error between Even if the decoder can reconstruct the second fault data as much as possible using the first target feature extracted from the second fault data.
[0109] In one embodiment, an attention mechanism is introduced into the autoencoder (i.e., encoder and decoder) so that the fault classification system can automatically learn which features are most important in fault classification. The weight of each feature is dynamically adjusted according to its contribution to the fault classification task, thereby enabling the fault classification system to focus more on key features and improve the accuracy of fault classification.
[0110] Specifically, an attention mechanism is introduced into the autoencoder, that is, an attention mechanism is added to each layer of the autoencoder. Based on this attention mechanism, a third weight is assigned to each feature in the second fault data. The third weights for each feature can be different, or partially the same and partially different. After assigning a third weight to each feature in the second fault data, multiple first automatically extracted features can be extracted from the second fault data according to the magnitude of the third weight.
[0111] like Figure 3 As shown, the fault classification method of this application includes:
[0112] Step S50: Preprocess the first fault data to obtain the target fault data.
[0113] In this step, the first fault data usually contains noise points that are useless for fault classification or may affect fault classification. Therefore, before inputting the first fault data into the fault classification system, the first fault data is preprocessed to remove noise points, and the processed first fault data is then input into the fault classification model as the target fault data.
[0114] Taking the tightening defect curve (i.e., the first fault data) collected during the bolt tightening process as an example, a mean filtering technique is used to smooth the tightening defect curve. Specifically, multiple pixel values around a pixel in the tightening defect curve are selected, the average value of these multiple pixel values is calculated, and the calculated average value replaces the selected pixel in the tightening defect curve. Based on this method, each pixel in the tightening defect curve is traversed, thereby achieving a smoothing operation on the tightening defect curve.
[0115] Step S60: Input the target fault data into the fault classification system to obtain the fault type corresponding to the first fault data.
[0116] In this step, after the target fault data is input into the pre-trained fault classification system, the fault classification system can automatically classify the first fault data and finally output the fault type corresponding to the first fault data.
[0117] In one embodiment, the fault classification system performs fault classification by including the following steps:
[0118] Step 600: The encoder extracts multiple second automatically extracted features from the target fault data, which serve as multiple second target features.
[0119] In this step, after the target fault data is input into the fault classification system, the encoder of the fault classification system will extract multiple second automatically extracted features from the target fault data and use these multiple second automatically extracted features as multiple second target features.
[0120] In one embodiment, multiple second manually extracted features extracted manually based on the first fault data are weighted and fused with multiple second automatically extracted features extracted in step 600. This effectively enhances the diversity and richness of features in fault classification and maximizes the use of the complementarity of multimodal data to improve the accuracy of fault classification.
[0121] Specifically, the weighted fusion method of the second manually extracted features and the second automatically extracted features includes:
[0122] Step 6000: Receive multiple second manually extracted features.
[0123] In this step, based on expert experience, multiple second manually extracted features are extracted from the first fault data, and these extracted features are input into the fault classification system.
[0124] Step 6002: Assign fourth weights to the multiple second automatically extracted features and the multiple second manually extracted features respectively.
[0125] In this step, depending on the actual situation, the fourth weights assigned to multiple second automatically extracted features and multiple second manually extracted features may be different, or the same fourth weights may be assigned to multiple second automatically extracted features and multiple second manually extracted features.
[0126] Step 6004: Merge multiple second automatically extracted features and multiple second manually extracted features, which are assigned a fourth weight, as the second target feature.
[0127] In this step, after assigning a fourth weight to multiple second automatically extracted features and multiple second manually extracted features, the multiple second automatically extracted features and multiple second manually extracted features are fused based on the assigned fourth weight, and the fused multiple second automatically extracted features and multiple second manually extracted features are used as multiple second target features for fault classification.
[0128] Step 602: Input multiple second target features into the random forest model, and obtain the fault type corresponding to the first fault data based on the voting mechanism of multiple decision trees in the random forest model.
[0129] In this step, the random forest model of the fault classification system outputs the classification result through a voting mechanism that integrates multiple decision trees. The random forest model can provide the importance of each secondary objective feature in the classification decision, thereby improving the accuracy of fault classification.
[0130] Specifically, step 602 includes: evaluating the importance of the second features corresponding to multiple second target features through a random forest model, assigning fifth weights to the multiple second target features based on the importance of the second features, and determining the fault type corresponding to the first fault data based on the multiple second target features assigned fifth weights.
[0131] The evaluation of feature importance has been described in detail in the above embodiments, and will not be repeated here.
[0132] Another embodiment of this application provides a fault classification system, which is configured to preprocess first fault data to obtain target fault data and classify the faults to obtain the fault type corresponding to the first fault data.
[0133] The fault classification of the first fault data using a fault classification system has been described in detail in the embodiments of the above fault classification method, and will not be repeated here.
[0134] like Figure 1 As shown, the fault classification system includes an autoencoder and a random forest model, where the autoencoder includes an encoder and a decoder.
[0135] The encoder is configured to extract multiple first automatically extracted features from the second fault data as multiple first target features; the random forest model evaluates the first feature importance of the multiple first target features and adjusts the first weights of the multiple first target features based on the first feature importance; the decoder constructs reconstructed fault data based on the multiple first target features after adjusting the first weights; the encoder and decoder are also configured to optimize the encoder and decoder according to the reconstruction error between the original fault data and the reconstructed fault data to complete the training of the fault classification system.
[0136] The pre-training of the fault classification system has been described in detail in the embodiments of the training method of the fault classification system described above, and will not be repeated here.
[0137] According to another embodiment of this application, a controller is provided. For example... Figure 4 As shown, the components of the controller 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 the database 450 is used to store data.
[0138] The controller 400 also includes an access device 440 that enables the controller 400 to communicate via one or more networks 460. Examples of these networks 460 include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations 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., Network Interface Controller (NIC)), such as an IEEE 802.11 Wireless Local Area Networks (WLAN) wireless interface, a Wi-MAX (Worldwide Interoperability for Microwave Access) 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.
[0139] In one embodiment, the aforementioned components of the controller 400 and Figure 4 Other components, not shown, can also be connected to each other, for example, via bus 430. It should be understood that... Figure 4 The block diagram of the controller 400 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.
[0140] Controller 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. Controller 400 can also be a mobile or stationary server.
[0141] The processor 420 is used to execute the following computer-executable instructions to implement the steps of the fault classification method described above.
[0142] The above is a schematic diagram of a controller according to this application. It should be noted that the technical solution of this controller and the technical solution of the above-mentioned fault classification method belong to the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the above-mentioned fault classification method.
[0143] Another embodiment of this application provides a computer-readable storage medium storing computer instructions that, when executed by a processor, are used to implement the steps of the fault classification method described above.
[0144] 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 and the technical solution of the above-described fault classification method belong to the same concept. 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 fault classification method.
[0145] The foregoing has described specific embodiments of this application. 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 results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0146] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.
[0147] 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 this specification is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this specification. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.
[0148] 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.
[0149] 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 content of this specification. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of this specification, 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 classification method, characterized in that, The fault classification method, which applies a pre-trained fault classification system, includes: The first fault data is preprocessed to obtain the target fault data; The target fault data is input into the fault classification system to obtain the fault type corresponding to the first fault data; The fault classification system includes an autoencoder and a random forest model. The autoencoder includes an encoder and a decoder. The training method for the fault classification system includes: The encoder is controlled to extract multiple first automatically extracted features from the second fault data, which are used as multiple first target features. The importance of a first feature of multiple first target features is evaluated using the random forest model, and the first weight of the multiple first target features is adjusted based on the importance of the first feature. The decoder is controlled to construct reconstructed fault data based on multiple first target features after adjusting the first weight; The encoder and the decoder are optimized based on the reconstruction error between the second fault data and the reconstructed fault data to complete the training of the fault classification system.
2. The fault classification method according to claim 1, characterized in that, The training method for the fault classification system also includes: Receive multiple first manually extracted features; Each of the first automatically extracted features and the first manually extracted features is assigned a second weight. The first target feature is obtained by fusing multiple first automatically extracted features and multiple first manually extracted features that are assigned the second weight.
3. The fault classification method according to claim 1 or 2, characterized in that, The step of evaluating the first feature importance of multiple first target features using the random forest model, and adjusting the first weights of the multiple first target features based on the first feature importance, includes: Multiple first target features are input into the random forest model; The importance of the first feature of the multiple first target features is determined based on the contribution of the first target features to the reduction of Gini impurity in each decision tree split node of the random forest model; Based on the importance of the first feature corresponding to each of the multiple first target features, the first weights corresponding to the multiple first target features are determined respectively.
4. The fault classification method according to claim 3, characterized in that, The step of determining the importance of the first feature of the multiple first target features based on their contribution to the reduction of Gini impurity in each decision tree split node of the random forest model includes: Calculate the contribution of each of the first target features to the reduction of Gini impurity at each node of the random forest model; The total Gini importance of each first target feature is obtained by summing the contribution values of each first target feature to the reduction of Gini impurity across all nodes of the random forest model. The importance of the first feature for each first target feature is evaluated based on the total Gini importance.
5. The fault classification method according to claim 4, characterized in that, The formula for calculating the Gini impurity of each of the first target features at each node is as follows: In the formula, P k It is the probability that a sample in node n belongs to fault category k, where K is the number of fault categories; The formula for calculating the Gini importance of each of the first target features in each decision tree of the random forest model is as follows: In the formula, GiniImp(F j (,t) represents the first objective feature F in the t-th decision tree. j The importance of the gin, T t It is the set of all nodes in the t-th decision tree; |n samples | is the number of samples in node n; Gini(n) is the Gini impurity of node n; Gini(n) left ) and Gini(n right ) are the Gini impurities of the left and right child nodes, respectively; |n left | and |n right | These represent the number of samples for the left and right child nodes, respectively; |T tsamples | is the total number of samples in the t-th decision tree; The formula for calculating the total Gini importance of each of the first target features is as follows: In the formula, w j The first target feature F j Total Gini Importance, GiniImp(F j (,t) represents the first objective feature F in the t-th decision tree. j The Gini importance is given by T, where T is the number of decision trees in the random forest model.
6. The fault classification method according to claim 1, characterized in that, The training method for the fault classification system also includes: An attention mechanism is introduced into the autoencoder to assign a third weight to each feature in the second fault data. Multiple first automatically extracted features are extracted based on the third weight.
7. The fault classification method according to claim 1, characterized in that, The step of inputting the target fault data into the fault classification system to obtain the fault type corresponding to the first fault data includes: The encoder extracts multiple second automatically extracted features from the target fault data, which serve as multiple second target features. Multiple second target features are input into the random forest model, and the fault type corresponding to the first fault data is obtained based on the voting mechanism of multiple decision trees in the random forest model.
8. The fault classification method according to claim 7, characterized in that, The fault classification method also includes: Receive multiple second manually extracted features; A fourth weight is assigned to each of the second automatically extracted features and the second manually extracted features; The second target feature is obtained by fusing multiple automatically extracted features and multiple manually extracted features that are assigned the fourth weight.
9. The fault classification method according to claim 7, characterized in that, The step of obtaining the fault type corresponding to the first fault data through the voting mechanism of multiple decision trees based on the random forest model includes: The importance of the second features corresponding to multiple second target features is evaluated using the random forest model. Based on the importance of the second feature, a fifth weight is assigned to each of the second target features; The fault type corresponding to the first fault data is determined based on multiple second target features assigned the fifth weight.
10. A fault classification system, characterized in that, The fault classification system is configured to preprocess the first fault data to obtain target fault data and then classify the faults to obtain the fault type corresponding to the first fault data. The fault classification system includes an autoencoder and a random forest model, wherein the autoencoder includes an encoder and a decoder; The encoder is configured to extract multiple first automatically extracted features from the second fault data as multiple first target features; The random forest model evaluates the importance of a first feature among multiple first target features, and adjusts the first weights of the multiple first target features based on the importance of the first feature; The decoder constructs reconstructed fault data based on multiple first target features after adjusting the first weights; The encoder and the decoder are further configured to optimize the encoder and the decoder based on the reconstruction error between the second fault data and the reconstructed fault data, thereby completing the training of the fault classification system.
11. A controller comprising a memory and a processor, the memory storing a computer program that, when executed by the processor, can implement the fault classification method according to any one of claims 1-9.
12. A computer-readable storage medium for storing a computer program that, when executed by a computer or processor, implements the steps of the fault classification method according to any one of claims 1-9.