A CapsNet-based method for predicting unknown fault types
By improving the classification structure of CapsNet, using the transformation matrix and dynamic routing algorithm, combined with the hyperparameter threshold λ, the adaptability and interpretability problems of deep learning models in the prediction of unknown type faults is solved, and the accuracy of unknown type faults is achieved Identify and predict.
Patent Information
- Application Number
- CN202110756467.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-07-05
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2041-07-05
AI Technical Summary
Existing deep learning models have poor adaptability when predicting faults of unknown types, cannot rely on themselves to adjust model output, and have poor interpretability, limiting their prediction accuracy in noisy environments.
Using the unknown type fault prediction method based on CapsNet, the inherent classification structure of CapsNet is improved by determining the transformation matrix, dynamic routing algorithm and hyperparameter threshold λ, so that it can adaptively identify unknown type faults and judge through the regularization value of the eigenvector. Is it an unknown type of failure?
Effectively process multi-dimensional state perception signals of complex equipment, realize accurate perception of equipment failures, improve the model's adaptability and transfer learning ability, and have fault prediction capabilities when there are unknown types of faults, which improves the prediction accuracy.
Smart Images

Figure CN113642228B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of weapon equipment fault prediction, and more specifically, to a CapsNet-based unknown type fault prediction method, a computer device, and a computer-readable storage medium. Background Art
[0002] Fault prediction is a comprehensive technology that detects the operating status of equipment, improves equipment reliability, and ensures safe operation of equipment. It plays an important role in the modern industrial society. However, with the development of computer technology, the Internet of Things, wireless communications, e-commerce, and intelligent manufacturing, the real-time automatic collection of industrial equipment product status and operating condition data has caused the amount of data collection to grow exponentially. In addition, the data has big data characteristics such as large capacity, low value density, diversity, and timeliness. How to effectively process the multi-dimensional perception signals of complex equipment so that its value can be reflected and reasonably utilized is an issue that urgently needs to be considered and solved in the current field of fault prediction.
[0003] The fault prediction method based on deep learning (DL) can extract, select and classify fault features end-to-end, but due to the design defects of the method itself, its adaptive ability and transfer learning ability are poor. When an unknown type of fault occurs, the model output cannot be adaptively adjusted. It is necessary to collect data to retrain the model or use other components to learn to identify unknown types of faults. Research has found that Capsule Network (CapsNet), as one of the most exciting advances in the field of DL, has the potential to realize unknown type fault diagnosis and fault prediction by itself, but its inherent classification structure assumes that each data point belongs to a known class to varying degrees. Therefore, when a fault of an unknown type occurs, the CapsNet model can only classify it into an existing fault category, which limits its ability to predict unknown types of faults and seriously affects the prediction accuracy of the model in a noisy environment. Summary of the invention
[0004] The purpose of the present invention is to provide a CapsNet-based unknown type fault prediction method, a computer device and a computer-readable storage medium, which is dedicated to solving the problems that the existing DL model has poor adaptability, cannot rely on itself to achieve unknown type fault prediction and has poor interpretability.
[0005] In order to achieve the above object, the present invention adopts the following technical solutions:
[0006] In a first aspect, the present invention provides a CapsNet-based method for predicting unknown type faults, comprising:
[0007] S101, determine the conversion matrix, and the low-level capsule uses the conversion matrix to generate the corresponding high-level capsule prediction vector ;
[0008] S102, determining a dynamic routing algorithm, and integrating the prediction vector generated by the lower-level capsule into the higher-level capsule that agrees with it by using the dynamic routing algorithm;
[0009] S103: Generate a feature vector f of a high-level capsule using the prediction vector of the high-level capsule i , and select the eigenvector f i The maximum value f among the regularization values;
[0010] S104, determining a hyperparameter threshold λ, comparing the threshold λ with the feature vector value f, and when f is greater than or equal to the threshold λ, outputting the feature vector value f; when f is less than the threshold λ, it indicates that an unknown type vector or a fault occurs.
[0011] In a specific embodiment, before step S101, the method further includes:
[0012] Input the CapsNet model to extract fault features and further generate low-level capsules.
[0013] In a specific embodiment,
[0014] The parameters of the conversion matrix are obtained during the training process of the CapsNet model. The number of parameters of the conversion matrix is: K·H·n·c, where K represents the number of low-level capsules, H represents the number of high-level capsules, n represents the dimension of the low-level capsule output vector, and c represents the dimension of the high-level capsule output vector.
[0015] In a specific embodiment,
[0016] The prediction vector of high-level capsule j is the feature vector μ of capsule i in the lower layer. i Multiply by the transformation matrix W ij obtained;
[0017] Among them, μ i is the feature vector of the lower-level capsule i, with a dimension of n×1; W ij Represents the transformation matrix of the low-level capsule i relative to the high-level capsule j, with a dimension of c×n; The dimension of is c×1.
[0018] In a specific embodiment, the S102 includes
[0019] The high-level capsule receives multiple prediction vectors from the low-level capsules. The dynamic routing algorithm measures the consistency between the predictions of the low-level capsules. When multiple predictions are consistent, the corresponding high-level capsule will be activated, and then all prediction vectors μ i The iterative dynamic routing process is weighted and sent to the high-level capsule.
[0020] In a specific embodiment,
[0021] The feature vector f i The weight is the coupling coefficient c i|j Indicates that c i|j The value of is updated through an iterative dynamic routing algorithm.
[0022] In a specific embodiment,
[0023] The upper limit of the threshold λ is f i Minimum among the regularization values.
[0024] In a second aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method provided in the first aspect of the present application.
[0025] In a third aspect, the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method provided in the first aspect of the present application is implemented.
[0026] The beneficial effects of the present invention are as follows:
[0027] The present invention provides a CapsNet-based method for predicting unknown type faults, which can effectively process multi-dimensional state perception signals of complex equipment and realize accurate perception of equipment faults. The improved CapsNet of the present invention has adaptive capabilities and transfer learning capabilities, and can adaptively adjust the model and identify unknown type faults when unknown type faults occur. On the premise that CapsNet has good fault prediction potential, the present invention introduces a threshold λ to change its inherent classification structure, so that it has fault prediction capabilities when unknown type faults occur. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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.
[0029] Figure 1 A flow chart of a CapsNet-based method for predicting unknown type faults according to an embodiment of the present invention is shown.
[0030] Figure 2 A schematic diagram showing the effect of a conversion matrix according to an embodiment of the present invention is shown.
[0031] Figure 3 A schematic diagram of a dynamic routing algorithm according to an embodiment of the present invention is shown.
[0032] Figure 4 A flowchart of training a CapsNet model to determine a threshold λ according to an embodiment of the present invention is shown.
[0033] Figure 5 A CapsNet fault classifier diagram after verification and improvement according to an embodiment of the present invention is shown.
[0034] Figure 6 A schematic diagram of the structure of a computer device suitable for implementing an embodiment of the present application is shown. DETAILED DESCRIPTION
[0035] In order to make the technical solutions and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below in conjunction with the accompanying drawings. Similar components in the accompanying drawings are represented by the same reference numerals. Those skilled in the art should understand that the content described below is illustrative rather than restrictive, and should not be used to limit the scope of protection of the present invention.
[0036] The first embodiment, as Figure 1 As shown, an embodiment of the present invention discloses a CapsNet-based method for predicting unknown type faults, including: inputting a CapsNet model, extracting fault features, and further generating low-level capsules;
[0037] S101, determine the conversion matrix, and the low-level capsule uses the conversion matrix to generate the corresponding high-level capsule prediction vector
[0038] In a specific embodiment, the conversion matrix is an important structure unique to the CapsNet model. It can predict the existence and posture of corresponding high-level features based on the existence and posture of low-level features. It is an important theoretical basis for CapsNet to achieve fault prediction on its own. The parameters of the conversion matrix are obtained during the training process of the CapsNet model. The number of parameters of the conversion matrix is: K·H·n·c, where K represents the number of low-level capsules, H represents the number of high-level capsules, n represents the dimension of the low-level capsule output vector, and c represents the dimension of the high-level capsule output vector.
[0039] In this embodiment, the prediction vector of the high-level capsule j is is the feature vector μ of capsule i in the lower layer. i Multiply by the transformation matrix W ij Obtained; Among them, μ i is the feature vector of the lower-level capsule i, with a dimension of n×1; W ijRepresents the transformation matrix of the low-level capsule i relative to the high-level capsule j, with a dimension of c×n; The dimension of is c×1.
[0040] S102, determining a dynamic routing algorithm, and integrating the prediction vector generated by the lower-level capsule into the higher-level capsule that agrees with it by using the dynamic routing algorithm;
[0041] In a specific embodiment, S102 specifically includes: the high-level capsule receives multiple prediction vectors from the low-level capsule, and the dynamic routing algorithm measures the consistency between the predictions of the low-level capsules. When multiple predictions are consistent, the corresponding high-level capsule will be activated, and then all prediction vectors μ i The iterative dynamic routing process is weighted and sent to the high-level capsule.
[0042] S103: Generate a feature vector f of a high-level capsule using the prediction vector of the high-level capsule i , and select the eigenvector f i The maximum value f among the regularized values; where the eigenvector f i The maximum value f among the regularized values can also be called the eigenvector value f.
[0043] In a specific embodiment, the feature vector f i The weight is the coupling coefficient c i|j Indicates that c i|j The value of is updated through an iterative dynamic routing algorithm, so that each low-level prediction vector Most of the information is sent to appropriate high-level capsules for feature fusion to form larger and more complex high-level features.
[0044] S104, determine the hyperparameter threshold λ, compare the threshold λ with the feature vector value f, and when f is greater than or equal to the threshold λ, output the feature vector value f; when f is less than the threshold λ, it indicates that an unknown type vector or fault has occurred. In this example, when the threshold λ is selected in an appropriate range, the model can perfectly distinguish between known and unknown fault categories. After multiple model performance verifications, the upper limit of the threshold λ is f i Minimum among the regularization values.
[0045] In this embodiment, the method of the present invention is described in detail with reference to the accompanying drawings.
[0046] Step 1: Determine the conversion matrix. When using CapsNet to predict unknown fault types, you first need to determine the conversion matrix. In this way, the prediction vector of the corresponding high-level capsule can be obtained by multiplying the low-level capsule by the conversion matrix. The parameters of the conversion matrix are obtained during the training process of the CapsNet model. The conversion matrix works as follows Figure 2 As shown, μ iis the feature vector of the lower-level capsule i, with a dimension of n×1; W ij Represents the transformation matrix of the low-level capsule i relative to the high-level capsule j, with a dimension of c×n; the prediction vector of the high-level capsule j is the feature vector μ of capsule i in the lower layer. i Multiply by the transformation matrix W ij Got it, The dimension is c × 1. The figure describes the face posture predicted by the mouth posture detected by the lower-level capsule, that is, the transformation matrix predicts the existence and posture of the face (whole, high-level feature) based on the existence and posture of the mouth (part, low-level feature) detected by the lower-level capsule. The posture is represented by the vector parameter, and the existence probability is represented by the vector size.
[0047] Step 2: Determine the dynamic routing algorithm. The high-level capsule receives multiple prediction vectors from the low-level. The dynamic routing algorithm measures the consistency between the predictions of the low-level capsules. When multiple predictions are consistent, the corresponding high-level capsule will be activated, and then all prediction vectors After the iterative dynamic routing process, it is weighted and sent to the high-level capsule to form the feature vector of the high-level capsule. The weight of the feature vector uses the coupling coefficient c i|j Indicates that c i|j The value of is updated through an iterative dynamic routing algorithm, so that each low-level prediction vector Most of the information is sent to appropriate high-level capsules for feature fusion to form larger and more complex high-level features.
[0048] Specifically, the iterative process of the dynamic routing algorithm is as follows: Figure 3 As shown: where r represents the number of iterations of dynamic routing; l represents the lth capsule layer; k represents the number of high-level capsules; the coupling coefficient c i|j It is calculated by the iterative dynamic routing process and represents the degree of membership of the entity detected by the lower-level capsule i to the higher-level capsule j. It is the intermediate value b ij After the soft max function is applied, the sum of the coupling coefficients of each low-level capsule relative to all capsules is 1, and c i|j Do not save, the model will be recalculated each time it is run; j It is the result of weighted addition and fusion of the prediction vector of the lower layer by the higher layer capsule j; v j Represents the output of the comprehensive prediction of high-level capsule j after squash nonlinear compression (the length of the vector represents the probability of entity existence, and the direction represents the instantiation parameter, so the direction is kept unchanged and the length is compressed to no more than 1). The 7th line indicates that the dot product similarity of each low-level capsule relative to the output of each high-level capsule is calculated, and the coupling coefficient of each low-level prediction vector is updated according to the size of the dot product, so that the information of each low-level capsule is sent to the high-level capsule that agrees with it.
[0049] c ij =exp(b ij ) / ∑ k exp(b ik ) (1)
[0050] v j =(||s j || / 1+||s j || 2 )·(s j / ||s j ||) (2)
[0051] Depend on Figure 3 It can be seen that at the beginning, the information of each prediction vector is evenly sent to each high-level capsule to estimate the unknown high-level feature vector. Because there is a similarity metric, the hypothesis of the current high-level feature is used to update the weight of the prediction vector, and the weighted prediction vector is used to estimate the unknown high-level feature vector in the next dynamic routing process. In this way, after a cyclic iterative process, multiple similar prediction vectors have larger weighted values, and they are added and fused to form a new high-level feature vector. The two crucial functions used in this process complete the normalization of the coupling coefficient and the nonlinear compression of the high-level feature vector:
[0052] (1) Since the prediction of each low-level capsule belongs to at most one high-level capsule, the soft max function is used to normalize the coupling coefficients so that the sum of the coupling coefficients of the prediction vectors belonging to the same low-level capsule is less than 1, and the relatively large coupling coefficients in the iterative process are quickly approached to 1 during the normalization process, accelerating the convergence process;
[0053] (2) Since the length of each capsule feature vector represents the existence probability of instances of that class, the squash function is used to make the total length of each high-level feature vector less than 1.
[0054] Since the soft max function in the dynamic routing algorithm makes the sum of the membership of a single data point in all classes equal to 1, this avoids the trivial solution of all memberships being 0, but it assumes that each data point belongs to the class represented by the high-level capsule to varying degrees. Moreover, the existing CaspNet model structure can only use existing high-level capsules as classifiers, and must output a category as the final classification result. Therefore, when the input is a category that is not included in the CapsNet model, it will still output a known fault class as the classification result of the unknown type of fault. Based on this, this patent improves the classifier structure of the CaspNet model so that it can effectively process multi-dimensional state perception signals of complex equipment, achieve accurate perception of equipment faults, and adaptively adjust the model to identify unknown types of faults when unknown types of faults occur.
[0055] Step 3: Determine the hyperparameter threshold λ. The value of the threshold λ can be determined by training the CapsNet model. Figure 4 As shown. Considering that the feature matrix is equivalent to a mapping template of a known fault category, when the feature belongs to a known class, the feature vector has a larger eigenvalue, and the dynamic routing algorithm will send most of the information of its multiple similar prediction vectors to the high-level capsules of that class. Therefore, when the input is a known class, its feature vector has a larger regularization value. When the input is an unknown class, the prediction vector value of the feature vector is relatively small after passing through the transformation matrix, and the dynamic routing process cannot find a similar class, so the prediction vector value will be scattered and sent to multiple high-level capsules. Therefore, when the input is an unknown class, the feature vector has a smaller regularization value. Based on the above analysis, the regularization value f output by the model is i The minimum value among them is taken as the upper limit of the threshold λ.
[0056] Step 4: Improve the classifier structure. In order to achieve unknown type fault prediction, a threshold λ is introduced in the process of fault feature classification in the last layer of CapsNet capsule. Its value has a great influence on the prediction performance of the model. The improved CapsNet fault classifier is shown in Figure 2. Figure 5 As shown in the figure. The input of the model is 10 ordered feature vectors for classification. Each feature vector is L2 regularized to obtain 10 ordered scalars. By comparing the values of the 10 ordered scalars, the largest scalar is selected and defined as f. f is compared with the threshold λ. If the feature vector value f is greater than or equal to the threshold λ, it indicates that no unknown type of fault occurs; if the feature vector value f is less than the threshold λ, it indicates that an unknown type of fault occurs.
[0057] The present invention provides a CapsNet-based method for predicting unknown type faults, which can effectively process multi-dimensional state perception signals of complex equipment and realize accurate perception of equipment faults. The improved CapsNet of the present invention has adaptive capabilities and transfer learning capabilities, and can adaptively adjust the model and identify unknown type faults when unknown type faults occur. On the premise that CapsNet has good fault prediction potential, the present invention introduces a threshold λ to change its inherent classification structure, so that it has fault prediction capabilities when unknown type faults occur.
[0058] Second embodiment
[0059] Figure 6 A schematic structural diagram of a computer device provided in another embodiment of the present application is shown. Figure 6 The computer device 50 shown is only an example and should not limit the functions and scope of use of the embodiments of the present application. Figure 6 As shown, the computer device 50 is in the form of a general-purpose computing device. The components of the computer device 50 may include, but are not limited to: one or more processors or processing units 500, a system memory 516, and a bus 501 connecting different system components (including the system memory 516 and the processing unit 500).
[0060] Bus 501 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus structures. For example, these architectures include but are not limited to Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus and Peripheral Component Interconnect (PCI) bus.
[0061] The computer device 50 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computer device 50, including volatile and non-volatile media, removable and non-removable media.
[0062] System memory 516 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 504 and / or cache memory 506. Computer device 50 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 508 may be used to read and write non-removable, non-volatile magnetic media ( Figure 6 not shown, usually called a "hard drive"). Although Figure 6Not shown in the figure, a disk drive for reading and writing a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 501 via one or more data medium interfaces. The memory 516 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the first embodiment.
[0063] A program / utility 510 having a set (at least one) of program modules 512 may be stored, for example, in a memory 516, such program modules 512 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. Program modules 512 generally perform the functions and / or methods of the embodiments described herein.
[0064] The computer device 50 may also communicate with one or more external devices 70 (e.g., keyboard, pointing device, display 60, etc.), one or more devices that enable a user to interact with the computer device 50, and / or any device that enables the computer device 50 to communicate with one or more other computing devices (e.g., network card, modem, etc.). Such communication may be performed through an input / output (I / O) interface 502. Furthermore, the computer device 50 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 514. Figure 6 As shown, the network adapter 514 communicates with other modules of the computer device 50 via the bus 501. It should be understood that although Figure 6 Not shown, other hardware and / or software modules may be used in conjunction with computer device 50, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.
[0065] The processor unit 500 executes various functional applications and data processing by running programs stored in the system memory 516, such as implementing a CapsNet-based unknown type fault prediction method provided in the first embodiment of the present application.
[0066] In response to the current existing problems, this application develops a computer device suitable for CapsNet's unknown type fault prediction method, which can effectively process multi-dimensional state perception signals of complex equipment and achieve accurate perception of equipment faults.
[0067] Third embodiment
[0068] Another embodiment of the present application provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method provided in the above-mentioned embodiment 1 is implemented. In practical applications, the computer-readable storage medium can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium.
[0069] The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, device, or device.
[0070] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.
[0071] The program code included in the computer-readable medium can be transmitted with any appropriate medium, including but not limited to wireless, electric wire, optical cable, RF, etc., or any suitable combination of the above. The computer program code for performing the operation of the present application can be written with one or more programming languages or their combination, and the programming language includes object-oriented programming languages-such as Java, Smalltalk, C++, and also includes conventional procedural programming languages-such as "C" language or similar programming languages. The program code can be executed completely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer, partially on the remote computer, or completely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer by any type of network-including local area network (LAN) or wide area network (WAN), or, can be connected to an external computer (for example, using an Internet service provider to connect to the Internet).
[0072] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not limitations on the implementation methods of the present invention. For ordinary technicians in the relevant field, other different forms of changes or modifications can be made based on the above description. It is impossible to list all the implementation methods here. All obvious changes or modifications derived from the technical solution of the present invention are still within the protection scope of the present invention.
Claims
1. A CapsNet-based method for predicting unknown type faults, characterized in that: include: S101, determine the conversion matrix, and the low-level capsule uses the conversion matrix to generate the corresponding high-level capsule prediction vector S102, determining a dynamic routing algorithm, and integrating the prediction vector generated by the lower-level capsule into the higher-level capsule that agrees with it by using the dynamic routing algorithm; S103: Generate a feature vector f of a high-level capsule using the prediction vector of the high-level capsule i The input of the CapsNet model is 10 ordered feature vectors for classification. Each feature vector is regularized to obtain 10 ordered scalars. By comparing the values of the 10 ordered scalars, the largest scalar is selected and defined as the feature vector value f; S104, improve the fault classifier structure of CapsNet, introduce a hyperparameter threshold λ in the process of implementing fault feature classification in the last layer of capsules of CapsNet, and determine the hyperparameter threshold λ, wherein the hyperparameter threshold λ is determined by training the CapsNet model, λ=min(f0,f1,...,f n ),f0,f1,f n is the feature vector value obtained according to the dynamic routing algorithm during the training process; the threshold λ is compared with the feature vector value f, and when f is greater than or equal to the threshold λ, the feature vector value f is output; when f is less than the threshold λ, it indicates that an unknown type vector or a fault occurs; The upper limit of the threshold λ is f i Minimum among the regularization values.
2. The method according to claim 1, characterized in that Before step S101, the following steps are also included: Input the CapsNet model to extract fault features and further generate low-level capsules.
3. The method according to claim 1, characterized in that The parameters of the conversion matrix are obtained during the training process of the CapsNet model. The number of parameters of the conversion matrix is: K·H·n·c, where K represents the number of low-level capsules, H represents the number of high-level capsules, n represents the dimension of the low-level capsule output vector, and c represents the dimension of the high-level capsule output vector.
4. The method according to claim 3, characterized in that The prediction vector of high-level capsule j is the feature vector μ of capsule i in the lower layer. i Multiply by the transformation matrix W ij obtained; Among them, μ i is the feature vector of the lower-level capsule i, with a dimension of n×1; W ij Represents the transformation matrix of the low-level capsule i relative to the high-level capsule j, with a dimension of c×n; The dimension of is c×1.
5. The method according to claim 1, characterized in that The S102 includes the high-level capsule receiving multiple prediction vectors from the low-level capsule, and the dynamic routing algorithm measuring the consistency between the predictions of the low-level capsules. When multiple predictions are consistent, the corresponding high-level capsule will be activated, and then all prediction vectors μ i The iterative dynamic routing process is weighted and sent to the high-level capsule.
6. The method according to claim 1, characterized in that The feature vector f i The weight is the coupling coefficient c i|j Indicates that c i|j The value of is updated through an iterative dynamic routing algorithm.
7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.
8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 6 is implemented.
Citation Information
Patent Citations
Fault detection and mitigation in hybrid drive system
CN102725619A
Fault classification method based on bidirectional long-short-term memory unit and capsule network
CN111046961A
Bearing sub-health identification method for improving capsule network optimization hierarchical convolution
CN111626361A
Capsule network rolling bearing fault diagnosis system based on gating circulation unit
CN111832663A
Image classification method based on gradient maintenance, storage medium and electronic equipment
CN112348118A
