Fault knowledge base generative update method, system, electronic device and medium

Abstract

The present invention provides a method, system, electronic device, and medium for generatively updating a fault knowledge base. The method includes: obtaining operating condition data of equipment in a system to be tested; inputting the operating condition data into a potential fault diagnosis model; when the potential fault diagnosis model identifies a fault, collecting real potential fault samples, fine-tuning the potential fault diagnosis model, and updating the fault knowledge base; wherein the potential fault diagnosis model is trained based on virtual samples of potential fault types; and the virtual samples of potential fault types are generated based on normalized multivariate representation vectors of potential fault types according to operating condition evolution patterns of known fault types. The method, system, electronic device, and medium for generatively updating a fault knowledge base provided by the present invention generate virtual samples of potential fault types using operating condition evolution patterns of known fault types, and then train a potential fault diagnosis model, identify faults, obtain real potential fault samples, and expand and update the fault knowledge base.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis and maintenance, and in particular to a fault knowledge base generative updating method, system, electronic equipment and medium. Background Art

[0002] Due to the complex structure, high integration of components, and interconnected coupling, failures in industrial equipment can cause significant economic losses. Therefore, effectively monitoring system variable deviations, building diagnostic models to determine whether a fault has occurred, and then determining the type, severity, and onset of the fault, and then locating and troubleshooting it, is a key technology for improving the reliability of industrial equipment. Existing fault diagnosis methods can be categorized as qualitative and quantitative. Qualitative diagnosis primarily relies on empirically derived diagnostic rules, while quantitative diagnosis relies on mechanisms and data to construct diagnostic models for fault diagnosis. Given the wide variety of fault types, establishing a unified fault knowledge base can categorize and store various diagnostic rules and models, providing data support for the operation and decision-making of fault diagnosis systems.

[0003] Industrial equipment failures can be caused by a variety of factors, including component degradation, improper operation, and environmental changes. Monitoring data for equipment in different operating states can also vary, making fault knowledge bases unable to identify potential unknown faults and unable to exhaustively characterize known faults under different operating conditions. This is especially true when fault data typically contains small sample sizes and imbalanced categories, which severely impacts the accuracy of diagnostic rules and models. This results in a low reliability of the fault knowledge base and prevents it from being effectively updated to meet the requirements for safe and stable operation of industrial systems.

[0004] At the same time, the application of fault knowledge bases in industrial enterprises is relatively rare, and the updating method is mainly post-fault updating. That is, after a new fault occurs, new rules and models are obtained through analysis and diagnosis and incorporated into the knowledge base. However, due to the lack of data and representation of potential faults, it is impossible to provide early warning to avoid economic losses caused by the fault.

[0005] Therefore, there is an urgent need for a fault knowledge base generative update method to solve the above problems. Summary of the Invention

[0006] The present invention provides a fault knowledge base generative update method, system, electronic device and medium to solve the technical problems in the prior art that the fault knowledge base cannot identify potential unknown faults, cannot exhaust the representations of known faults under different operating states, and cannot provide early warning.

[0007] In a first aspect, the present invention provides a method for generating and updating a fault knowledge base, comprising:

[0008] Obtain operating condition data of equipment in the system under test;

[0009] Inputting the operating condition data into a potential fault diagnosis model, and when the potential fault diagnosis model identifies a fault, collecting real potential fault samples, fine-tuning the potential fault diagnosis model, and updating a fault knowledge base;

[0010] Among them, the potential fault diagnosis model is obtained by training based on virtual samples of potential fault types; the virtual samples of potential fault types are generated based on the normalized multivariate representation vector of the potential fault type according to the working condition evolution pattern of the known fault type.

[0011] According to a method for generating and updating a fault knowledge base provided by the present invention, the process of acquiring virtual samples of potential fault types includes:

[0012] Selecting a normalized multivariate representation vector of potential fault types;

[0013] Inputting the normalized multivariate representation vector of the potential fault type into the generator model, learning the evolution pattern of the fault characteristics under the change of working conditions through the generator model, and generating virtual samples of the potential fault type;

[0014] The generator model is obtained through self-supervised adversarial training based on the concatenation of normalized multivariate representation vectors of known and potential fault types and pre-generated random noise of fixed dimension.

[0015] According to a fault knowledge base generative updating method provided by the present invention, the normalized multivariate characterization vector of the potential fault type is within the distribution range of the normalized multivariate characterization vector of the known fault type.

[0016] According to a fault knowledge base generative updating method provided by the present invention, the training process of the generator model includes:

[0017] According to the known fault types and corresponding operating condition variables in the fault knowledge base, a normalized multivariate representation vector of known and potential fault types is constructed;

[0018] Establishing a generator model, wherein the input of the generator model is the data obtained by concatenating the normalized representation vectors of the known and potential fault types with pre-generated random noise of fixed dimension, and the output of the generator model is virtual samples of the known and potential fault types;

[0019] Establishing a discriminator model, wherein the input of the discriminator model is the virtual samples of the known and potential fault types and the real samples in the fault knowledge base, and the output of the discriminator model is the binary authenticity information and type information of the samples;

[0020] Some fault type samples are extracted from the known fault type sample set as the supervision sample set, the known fault type sample set without the supervision sample set is used as the training set, and the supervision sample set is used as the verification set. The parameter sets of the generator model and the discriminator model are cyclically optimized through adversarial training until the discrimination accuracy of the discriminator model for the supervision sample set reaches the preset accuracy threshold, thereby obtaining a trained generator model.

[0021] A method for generating and updating a fault knowledge base according to the present invention further includes:

[0022] The real potential fault samples are added to the known fault sample set, and the generator model and the discriminator model are trained and updated.

[0023] According to a generative updating method for a fault knowledge base provided by the present invention, the normalized multivariate characterization vector is a multidimensional discrete description vector obtained by splitting the fault type by dimension according to the operating condition variable, and performing polynomial discretization and normalization based on the extreme values of the operating condition variable.

[0024] According to a fault knowledge base generative updating method provided by the present invention, fine-tuning the potential fault diagnosis model includes:

[0025] The feature extractor of the potential fault diagnosis model is fixed and the classification layer of the potential fault diagnosis model is fine-tuned to adapt to the distribution characteristics of real potential fault samples.

[0026] In a second aspect, the present invention further provides a fault knowledge base generative update system, comprising:

[0027] An acquisition module is used to obtain the working condition data of the equipment in the system under test;

[0028] an updating module, configured to input the operating condition data into a potential fault diagnosis model, collect real potential fault samples when the potential fault diagnosis model identifies a fault, fine-tune the potential fault diagnosis model, and update a fault knowledge base;

[0029] Among them, the potential fault diagnosis model is obtained by training based on virtual samples of potential fault types; the virtual samples of potential fault types are generated based on the normalized multivariate representation vector of the potential fault type according to the working condition evolution pattern of the known fault type.

[0030] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the above-mentioned fault knowledge base generation update methods when executing the program.

[0031] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any one of the above-mentioned methods for generating and updating the fault knowledge base.

[0032] The fault knowledge base generative update method, system, electronic device and medium provided by the present invention generate virtual samples of potential fault types through the working condition evolution pattern of known fault types, and then train to obtain a potential fault diagnosis model. The potential fault diagnosis model can accurately identify potential faults and obtain real potential fault samples. At the same time, the potential fault diagnosis model can also be fine-tuned through the real potential fault samples, thereby further improving the reliability of the potential fault diagnosis model, realizing the expansion and updating of the fault knowledge base, and providing data support for equipment fault warning and maintenance decision-making in industrial systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0034] Figure 1 This is a schematic diagram of the overall process of the fault knowledge base generation updating method provided by the present invention;

[0035] Figure 2 Detailed flowchart of the fault knowledge base generation updating method provided by the present invention;

[0036] Figure 3 This is a schematic diagram of the implementation principle of the fault knowledge base generation updating method provided by the present invention;

[0037] Figure 4 This is a schematic diagram of the structural architecture of the fault knowledge base generation updating system provided by the present invention;

[0038] Figure 5 It is a schematic diagram of the structural architecture of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0039] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0040] Attachment Figure 1 The present invention provides a method for generating and updating a fault knowledge base, including:

[0041] S110: Acquire operating condition data of the equipment in the system to be tested;

[0042] S120: Inputting the operating condition data into the potential fault diagnosis model. When the potential fault diagnosis model identifies a fault, collecting real potential fault samples, fine-tuning the potential fault diagnosis model, and updating the fault knowledge base;

[0043] Among them, the potential fault diagnosis model is obtained by training based on virtual samples of potential fault types; the virtual samples of potential fault types are generated based on the normalized multivariate representation vector of the potential fault type according to the working condition evolution pattern of the known fault type.

[0044] Specifically, the process of obtaining virtual samples of potential fault types includes:

[0045] Selecting a normalized multivariate representation vector of potential fault types;

[0046] Input the normalized multivariate representation vector of the potential fault type into the generator model to generate virtual samples of the potential fault type;

[0047] Among them, the generator model is obtained through self-supervised adversarial training based on the concatenation of normalized multivariate representation vectors of known and potential fault types and pre-generated random noise of fixed dimension.

[0048] In this embodiment, after the generator model is constructed, it needs to be trained based on the discriminator model for self-supervision adversarial training to be able to be used normally. Figure 2 The construction and training of the generator model and the discriminator model, the construction of the potential fault diagnosis model, and the updating process of the fault knowledge base are shown, which specifically include:

[0049] S210: Constructing a normalized multivariate representation vector, that is, constructing a normalized multivariate representation vector of known and potential fault types based on known fault types and corresponding operating condition variables in the fault knowledge base.

[0050] In this step, the normalized multivariate representation vector is a multidimensional discrete description vector obtained by splitting the fault type into dimensions based on the operating condition variables and performing polynomial discretization and normalization based on the extreme values of the operating condition variables. The polynomial parameter values used in polynomial discretization and normalization depend on the relationship between operating condition changes and sample changes. This can be identified by calculating the corresponding relationship between statistics such as variance.

[0051] This example uses the CWRU (Case Western Reserve University) bearing fault dataset as an example to illustrate the construction method of a normalized multivariate representation vector. The experimental equipment for the CWRU dataset includes a motor, torque sensor, decoder, power meter, and controller. Bearings with fault diameters of 0.007, 0.014, 0.021, and 0.028 inches (corresponding to inner race I, rolling element B, and outer race O) were tested under loads of 0, 1, 2, and 3 horsepower, respectively. Acceleration time series data were collected from the drive end, fan end, and base.

[0052] The inner race, rolling element, and outer race each represent three different fault types. The fault diameter and load level represent two-dimensional operating condition variables. Selecting a linear function as the polynomial and normalizing the operating condition variables to the range [0, 1] yields the corresponding normalized multivariate representation vector. For example, an inner race fault with a diameter of 0.014 inches and a load of 2 horsepower is represented as [I, 0.33, 0.66]. The remaining representations are similar and are not detailed here.

[0053] S220: Construct a generator model, that is, establish a generator model for fault knowledge expression, splice the normalized representation vectors of known and potential fault types with random noise of fixed dimension generated by a random sequence generator, input the spliced data into the generator model, and output virtual samples of known and potential fault types.

[0054] In this step, the generator model is built based on the neural network architecture, and the convolutional layer or the fully connected layer is selected as the basic layer. The random sequence generator generates a pseudo-random sequence with a standard Gaussian distribution. The dimension of the pseudo-random sequence is determined according to the dimension of the known fault type sample variable and the length of the sampling time window.

[0055] This example sets the time window length to 1024, concatenates the drive-end and fan-end data from the CWRU bearing fault dataset end-to-end as training samples, sets the output dimension of the pseudo-random sequence to 100, the operating condition variable dimension to 2, and constructs a fully connected deep network as the generator model. FC, BN, and LR are defined as the fully connected layer, batch normalization layer, and LeakyRelu activation layer, respectively. [FC, BN, LR] is defined as a block. The generator model consists of five blocks and one FC layer, and the feature dimension transformation sequence is [100+2, 128, 256, 512, 1024, 2048, 2048].

[0056] S230: Construct a discriminator model, that is, establish a discriminator model for authenticity identification and type classification, input virtual samples of known and potential fault types and real samples in the fault knowledge base into the discriminator model, apply neighbor smoothing constraints to the virtual samples of known and potential fault types in batches, and output binary authenticity information and type information of the samples.

[0057] In this step, the discriminator model is built based on the neural network architecture, and the convolutional layer or the fully connected layer is also selected as the basic layer. The output of the model includes three parts: feature layer, adversarial layer and classification layer, which correspond to the output of the last feature layer, binary true and false output and type output respectively; the nearest neighbor smoothing constraint is used to limit the difference in the neuron value of the last feature layer between the virtual samples and the nearest neighbor known samples in the training batch, so as to ensure that there is no large difference between the generated virtual samples and the adjacent samples, thereby improving the accuracy of the generated samples.

[0058] In this embodiment, DO represents the Dropout layer. The feature transformer architecture of the discriminator model can be expressed as [FC, LR, FC, DO, LR, FC, DO, LR]. The corresponding feature dimension transformation sequence is [2048, 256, 128, 64], and the feature layer is output. The model then bifurcates into two branches, one of which is followed by an FC layer and a Sigmund activation function to output binary true / false information, and the other is followed by an FC layer to output type information. A Softmax activation function can be added depending on whether the type can be determined in advance. A neighbor smoothing constraint is applied to the feature layer output, and adjacent known samples are extracted for all virtual samples in a batch to form sample pairs. For example, assuming [I, 0.33, 0] is a potential fault sample type and the rest are known sample types, the corresponding adjacent sample types are represented as [I, 0, 0.33] and [I, 0.66, 0.33]. All adjacent sample types in a batch are cyclically searched for matches with the virtual samples.

[0059] S240: Model self-supervised adversarial training, that is, extracting some fault type samples from the known fault type sample set as the supervised sample set, using the known fault type sample set without the supervised sample set as the training set, and using the supervised sample set as the verification set. The parameter sets of the generator model and the discriminator model are cyclically optimized through adversarial training until the discrimination accuracy of the discriminator model for the supervised sample set reaches the preset accuracy threshold, thereby obtaining the trained generator model and discriminator model.

[0060] In this step, due to the lack of actual samples of potential fault types, it is impossible to use them to measure the discriminator's training effectiveness. Therefore, some samples of known fault types are used as supervision samples, and the verification accuracy of the supervision samples is used as the indicator for stopping model training. During the training process, the mean absolute error, binary cross entropy, cross entropy, or mean squared error is used as the loss function for the feature layer, adversarial layer, and classification layer, respectively. The generator model and the discriminator model are trained sequentially until the verification accuracy of the supervision samples reaches the specified threshold.

[0061] S250: Generate potential fault samples and build a potential fault diagnosis model, that is, select the normalized multivariate representation vector of the potential fault type, input the trained generator model to generate virtual samples of the potential fault type, build the potential fault diagnosis model, and add it to the fault knowledge base.

[0062] In this step, the normalized multivariate representation vector of the potential fault type should generally be within the distribution range of the normalized multivariate representation vector of the known fault type.

[0063] For example, the known faulty bearing diameters are 0.007, 0.014, 0.021, and 0.028 inches, and the operating conditions are 0, 1, 2, and 3 horsepower, respectively. Therefore, in the process of selecting the normalized multivariate characterization vector of the potential fault type, the faulty bearing diameter and load horsepower should be within the extreme values of [0.007, 0.028] and [0, 3], thereby ensuring the reliability of generating potential fault type samples.

[0064] When the number of known fault types is small, random selection can lead to large errors. Therefore, fault types with the same operating condition variables as known fault types are preferentially selected as potential fault types. Virtual samples are generated and a potential fault diagnosis model is constructed. When the total number of potential fault types is bounded, the discriminator model can be used directly as the potential fault diagnosis model.

[0065] S260: Fine-tune the potential fault diagnosis model based on measured samples, that is, monitor the system operating status in real time. When the potential fault diagnosis model identifies a fault, collect real potential fault samples, fine-tune the potential fault diagnosis model, and update the fault knowledge base.

[0066] Specifically, after collecting real potential fault samples, the feature extractor of the potential fault diagnosis model is fixed and the classification layer of the model is fine-tuned to adapt to the distribution characteristics of the real samples.

[0067] In this embodiment, the discriminator model is copied as a potential fault diagnosis model. After collecting real potential fault samples, the parameters of the fixed feature converter remain unchanged, the parameters of the classification layer are fine-tuned to optimize the accuracy of the diagnosis model, and the original discriminator model parameters remain unchanged.

[0068] In order to make the generator and the discriminator in the above fault knowledge base generative update method more accurate and reliable, the following may also be included:

[0069] S270: Update the generator model and the discriminator model based on the real potential fault samples, that is, add the real potential fault samples to the known fault sample set, and train and update the generator model and the discriminator model.

[0070] In this step, only real potential fault samples are used as training sets, and optimization and debugging are performed based on the parameter sets of the generator model and the discriminator model.

[0071] Figure 3 The implementation principle of the generative updating method for the fault knowledge base provided by an embodiment of the present invention is shown. By mining the changes in fault operating conditions and statistical laws, a normalized multivariate representation vector describing known and potential fault operating condition types is constructed. Based on the neural network adversarial learning architecture, the evolution pattern of fault characteristics under changing operating conditions is learned through the generator model. At the same time, supervision samples are extracted and the model is trained through a self-supervised adversarial training method with neighbor smoothing constraints, thereby generating more accurate potential fault samples. Based on this, a new fault diagnosis model is constructed to achieve fault knowledge amplification. Finally, the diagnosis model, generator model, and discriminator model are updated according to the measured samples.

[0072] The fault knowledge base generative update method provided by the embodiment of the present invention is suitable for scenarios where there are many fault types and few samples, making it difficult to build a diagnostic model. It can generate potential fault type samples and build a diagnostic model based on the working condition evolution rules of existing fault types, which is of great significance for fault prediction and maintenance decision-making of industrial systems and equipment.

[0073] The following describes a fault knowledge base generation and updating system provided by the present invention. The fault knowledge base generation and updating system described below and the fault knowledge base generation and updating method described above can refer to each other.

[0074] Figure 4 The present invention provides a fault knowledge base generation and updating system, including:

[0075] An acquisition module 410 is used to acquire operating condition data of the equipment in the system under test;

[0076] An updating module 420 is configured to input operating condition data into a potential fault diagnosis model, collect real potential fault samples when the potential fault diagnosis model identifies a fault, fine-tune the potential fault diagnosis model, and update the fault knowledge base;

[0077] Among them, the potential fault diagnosis model is obtained by training based on virtual samples of potential fault types; the virtual samples of potential fault types are generated based on the normalized multivariate representation vector of the potential fault type according to the working condition evolution pattern of the known fault type.

[0078] In the update module 420, when obtaining virtual samples of potential fault types, it is first necessary to select the normalized multivariate representation vector of the potential fault type, and then input the normalized multivariate representation vector of the potential fault type into the generator model to generate virtual samples of the potential fault type; wherein the generator model is obtained through self-supervised adversarial training based on the splicing results of the normalized multivariate representation vectors of known and potential fault types and pre-generated random noise of fixed dimension.

[0079] In this embodiment, the normalized multivariate characterization vector of the potential fault type is within the distribution range of the normalized multivariate characterization vector of the known fault type.

[0080] Specifically, the training process of the generator model includes:

[0081] According to the known fault types and corresponding operating condition variables in the fault knowledge base, a normalized multivariate representation vector of known and potential fault types is constructed;

[0082] A generator model is established to concatenate the normalized representation vectors of known and potential fault types with pre-generated random noise of fixed dimensions. The concatenated data is used as the input of the generator model, and virtual samples of known and potential fault types are output.

[0083] Establish a discriminator model, take virtual samples of known and potential fault types and real samples in the fault knowledge base as input, apply nearest neighbor smoothing constraints to the virtual samples of known and potential fault types in batches, and output binary true and false information and type information of the samples;

[0084] Some fault type samples are extracted from the known fault type sample set as the supervision sample set, the known fault type sample set without the supervision sample set is used as the training set, and the supervision sample set is used as the verification set. The parameter sets of the generator model and the discriminator model are cyclically optimized through adversarial training until the discrimination accuracy of the discriminator model for the supervision sample set reaches the preset accuracy threshold, and a trained generator model is obtained.

[0085] More preferably, the fault knowledge base generation updating system provided by the embodiment of the present invention further includes:

[0086] The optimization module is used to add real potential fault samples to the known fault sample set, and train and update the generator model and discriminator model.

[0087] It can be understood that the normalized multivariate characterization vector mentioned in the embodiment of the present invention is a multidimensional discrete description vector obtained by splitting the fault type by dimension according to the operating condition variable, and performing polynomial discretization and normalization based on the extreme values of the operating condition variable.

[0088] In the updating module 420 , the feature extractor of the potential fault diagnosis model is fixed and the classification layer of the potential fault diagnosis model is fine-tuned to achieve model fine-tuning to adapt to the distribution characteristics of real potential fault samples.

[0089] The fault knowledge base generative update system provided by the embodiment of the present invention utilizes the working condition evolution pattern of known fault types to generate virtual samples of potential fault types, and then trains a potential fault diagnosis model. The potential fault diagnosis model can accurately identify faults and obtain real potential fault samples. At the same time, the potential fault diagnosis model can also be fine-tuned through the real potential fault samples, thereby further improving the reliability of the potential fault diagnosis model. The system can realize the expansion and updating of the fault knowledge base, and can provide data support for equipment fault warning and maintenance decision-making in industrial systems.

[0090] Figure 5 An example of a physical structure diagram of an electronic device is shown below. Figure 5 As shown, the electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communications bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other via the communications bus 540. The processor 510 may call the logic instructions in the memory 530 to execute a generative updating method for a fault knowledge base, which includes: obtaining operating condition data of the equipment in the system to be tested; inputting the operating condition data into a potential fault diagnosis model; when the potential fault diagnosis model identifies a fault, collecting real potential fault samples, fine-tuning the potential fault diagnosis model, and updating the fault knowledge base; wherein the potential fault diagnosis model is obtained by training based on virtual samples of potential fault types; the virtual samples of potential fault types are generated based on the normalized multivariate representation vector of the potential fault type according to the operating condition evolution pattern of the known fault type.

[0091] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0092] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the fault knowledge base generation update method provided by the above methods, the method including: obtaining operating condition data of the equipment in the system to be tested; inputting the operating condition data into a potential fault diagnosis model, when the potential fault diagnosis model identifies a fault, collecting real potential fault samples, fine-tuning the potential fault diagnosis model, and updating the fault knowledge base; wherein the potential fault diagnosis model is obtained by training based on virtual samples of potential fault types; the virtual samples of potential fault types are generated based on the normalized multivariate representation vector of the potential fault type according to the operating condition evolution pattern of the known fault type.

[0093] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the above-mentioned fault knowledge base generative update method provided above, the method comprising: obtaining operating condition data of equipment in the system to be tested; inputting the operating condition data into a potential fault diagnosis model, and when the potential fault diagnosis model identifies a fault, collecting real potential fault samples, fine-tuning the potential fault diagnosis model, and updating the fault knowledge base; wherein the potential fault diagnosis model is obtained by training based on virtual samples of potential fault types; the virtual samples of potential fault types are generated based on the normalized multivariate representation vector of the potential fault type according to the operating condition evolution pattern of the known fault type.

[0094] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0095] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for generating and updating a fault knowledge base, characterized in that: include: Obtain operating condition data of equipment in the system under test; Inputting the operating condition data into a potential fault diagnosis model, and when the potential fault diagnosis model identifies a fault, collecting real potential fault samples, fine-tuning the potential fault diagnosis model, and updating a fault knowledge base; The potential fault diagnosis model is obtained by training based on virtual samples of potential fault types; the virtual samples of potential fault types are generated based on normalized multivariate representation vectors of potential fault types according to the working condition evolution patterns of known fault types; The normalized multivariate characterization vector is a multidimensional discrete description vector obtained by splitting the fault type by dimension according to the operating condition variable and performing polynomial discretization and normalization according to the extreme values of the operating condition variable; The process of obtaining the virtual sample of the potential fault type includes: Selecting a normalized multivariate representation vector of potential fault types; Inputting the normalized multivariate representation vector of the potential fault type into a generator model to generate a virtual sample of the potential fault type; The generator model is obtained through self-supervised adversarial training based on the concatenation of normalized multivariate representation vectors of known and potential fault types and pre-generated random noise of fixed dimension.

2. A fault knowledge base generation updating method according to claim 1, characterized in that: The normalized multivariate characterization vector of the potential fault type is within the distribution range of the normalized multivariate characterization vector of the known fault type.

3. A fault knowledge base generation updating method according to claim 1, characterized in that: The training process of the generator model includes: According to the known fault types and corresponding operating condition variables in the fault knowledge base, a normalized multivariate representation vector of known and potential fault types is constructed; Establishing a generator model, wherein the input of the generator model is the data obtained by concatenating the normalized representation vectors of the known and potential fault types with pre-generated random noise of fixed dimension, and the output of the generator model is virtual samples of the known and potential fault types; Establishing a discriminator model, wherein the input of the discriminator model is the virtual samples of the known and potential fault types and the real samples in the fault knowledge base, and the output of the discriminator model is the binary authenticity information and type information of the samples; Some fault type samples are extracted from the known fault type sample set as the supervision sample set, the known fault type sample set without the supervision sample set is used as the training set, and the supervision sample set is used as the verification set. The parameter sets of the generator model and the discriminator model are cyclically optimized through adversarial training until the discrimination accuracy of the discriminator model for the supervision sample set reaches the preset accuracy threshold, thereby obtaining a trained generator model.

4. A fault knowledge base generation updating method according to claim 3, characterized in that: Also includes: The real potential fault samples are added to the known fault sample set, and the generator model and the discriminator model are trained and updated.

5. A fault knowledge base generation updating method according to claim 1, characterized in that: Fine-tuning the potential fault diagnosis model includes: The feature extractor of the potential fault diagnosis model is fixed and the classification layer of the potential fault diagnosis model is fine-tuned to adapt to the distribution characteristics of real potential fault samples.

6. A fault knowledge base generation update system, characterized in that: include: An acquisition module is used to obtain the working condition data of the equipment in the system under test; an updating module, configured to input the operating condition data into a potential fault diagnosis model, collect real potential fault samples when the potential fault diagnosis model identifies a fault, fine-tune the potential fault diagnosis model, and update a fault knowledge base; The potential fault diagnosis model is obtained by training based on virtual samples of potential fault types; the virtual samples of potential fault types are generated based on normalized multivariate representation vectors of potential fault types according to the working condition evolution patterns of known fault types; The normalized multivariate characterization vector is a multidimensional discrete description vector obtained by splitting the fault type by dimension according to the operating condition variable and performing polynomial discretization and normalization according to the extreme values of the operating condition variable; The process of obtaining the virtual sample of the potential fault type includes: Selecting a normalized multivariate representation vector of potential fault types; Inputting the normalized multivariate representation vector of the potential fault type into a generator model to generate a virtual sample of the potential fault type; The generator model is obtained through self-supervised adversarial training based on the concatenation of normalized multivariate representation vectors of known and potential fault types and pre-generated random noise of fixed dimension.

7. An electronic 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 steps of the fault knowledge base generation updating method according to any one of claims 1 to 5 are implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the fault knowledge base generation updating method according to any one of claims 1 to 5 are implemented.

Images