Spatial hierarchical semantic construction and equipment fault diagnosis method

Abstract

The invention discloses a spatial hierarchy semantic construction and equipment fault diagnosis method, and relates to the technical field of industrial equipment fault diagnosis, and the method comprises the steps: obtaining the features of all equipment in a to-be-diagnosed system; the features are vectors obtained after feature extraction is carried out on the operation data and the environment data; inputting the characteristics of each device in the to-be-diagnosed system into the fault diagnosis model to obtain a predicted value of a fault category in the to-be-diagnosed system; the fault diagnosis model is determined based on a fault known category data set, a fault known category spatial hierarchy semantic set, a sample feature generation model and a multi-layer perceptron; the label is an actual value of a fault category, the spatial hierarchy semantics is a vector formed by description attribute values of the fault, and the spatial hierarchy semantics comprises k layers of semantics. According to the invention, the accuracy of equipment fault diagnosis is improved.

Description

Technical Field

[0001] The present application relates to the technical field of industrial equipment fault diagnosis, and in particular to a spatial hierarchical semantic construction and equipment fault diagnosis method. Background Art

[0002] With the rapid development of artificial intelligence technology, deep learning has shown excellent performance in many fields. Traditional supervised learning methods require a large amount of labeled data to train models. However, in practical applications, it is often very difficult and expensive to obtain large-scale labeled data. Zero-shot learning methods are proposed to solve these problems. Zero-shot methods use the semantic information of categories to transfer knowledge learned from seen categories to unseen categories and identify new categories without additional labeled data. Generalized zero-shot learning extends traditional zero-shot learning and can simultaneously identify seen and unseen categories, which is more in line with practical application scenarios.

[0003] In the field of equipment fault diagnosis, especially in industrial equipment fault diagnosis, there is often a lack of samples for model training for new or rare faults, which limits the application of traditional supervised learning methods. Generalized zero-shot learning methods provide a new solution. Generalized zero-shot equipment fault diagnosis uses text semantics and attributes to describe faults, and learns the association between fault features and fault descriptions and attributes through a large number of existing equipment fault samples, thereby inferring and identifying the types of unseen equipment faults.

[0004] Different equipment failures exist in different systems, equipment, and equipment components. The characteristics of these spatial hierarchical distributions are of great significance for describing failures. The spatial hierarchical semantics of equipment failures can reflect the associations between different equipment failures at various spatial levels. It has a good constraint guidance effect on generating unseen equipment failure sample features in the generalized zero-shot equipment failure diagnosis method, and can improve the accuracy of generalized zero-shot equipment failure diagnosis. However, traditional generalized zero-shot equipment failure diagnosis methods often use the experience of experts and equipment operation and maintenance records to construct the description information and attribute characteristics of equipment failures, ignoring the characteristics of the spatial hierarchical distribution of equipment failures, resulting in low accuracy of equipment fault diagnosis. Summary of the invention

[0005] The purpose of this application is to provide a spatial hierarchical semantic construction and equipment fault diagnosis method to solve the problem of low accuracy of equipment fault diagnosis.

[0006] To achieve the above objectives, this application provides the following solutions.

[0007] The present application provides a spatial hierarchical semantic construction and equipment fault diagnosis method, including: Acquire the characteristics of each device in the system to be diagnosed; the characteristics are vectors obtained by extracting the characteristics of the operation data and the environmental data; The characteristics of each device in the system to be diagnosed are input into the fault diagnosis model to obtain the predicted value of the fault category in the system to be diagnosed; the fault diagnosis model is determined based on a data set of seen fault categories, a spatial hierarchical semantic set of seen fault categories, a sample feature generation model and a multi-layer perceptron; the seen fault category data set includes: original sample features and labels of multiple seen category faults, and the seen fault category spatial hierarchical semantic set includes: spatial hierarchical semantics of multiple seen category faults in the seen fault category data set; the label is the actual value of the fault category, the spatial hierarchical semantics is a vector composed of descriptive attribute values ​​of the fault, and the spatial hierarchical semantics includes k layers of semantics, k>1.

[0008] Optionally, the process of determining the fault diagnosis model includes: Obtain a training data set of seen fault categories and a spatial level semantic set of seen fault categories; Initializing parameters of the sample feature generation model; Based on the total loss function, the sample feature generation model is trained using the fault category training data set and the fault category space level semantic set to obtain a trained sample feature generation model; The trained sample feature generation model is used to randomly generate sample features and labels corresponding to multiple unseen fault category features, and a training data set of unseen fault categories is obtained; Merging the training data set of the seen fault category and the training data set of the unseen fault category to obtain a training data set for equipment fault diagnosis; Constructing an initial network of the fault diagnosis model based on a multi-layer perceptron; The initial network is trained using the equipment fault diagnosis training data set to obtain the fault diagnosis model.

[0009] Optionally, the sample feature generation model includes: a conditional variational autoencoder module, a conditional generative adversarial network module and a hierarchical semantic information alignment module; The conditional variational autoencoder module includes: an encoder and a decoder; the encoder is implemented by a multi-layer perceptron, and the decoder is composed of a multi-layer perceptron composed of a k-layer fully connected network and an activation function; The conditional generative adversarial network module includes: a generator and a discriminator; the generator reuses the decoder of the conditional variational autoencoder module, and the discriminator is implemented by a multi-layer perceptron; The hierarchical semantic information alignment module includes: k layers of semantic information alignment layers.

[0010] Optionally, based on the total loss function, the sample feature generation model is trained using the training dataset of the seen fault categories and the hierarchical semantic set of the seen fault category space, and a trained sample feature generation model is obtained, including: Based on the total loss function, the sample feature generation model is trained in multiple rounds using the training dataset of the seen fault categories and the hierarchical semantic set of the seen fault category space, and a trained sample feature generation model is obtained; wherein, the training process of any current round includes: The original sample features of each seen fault category in the training dataset of the seen fault categories and the hierarchical semantic of each seen fault category in the hierarchical semantic set of the seen fault category space are respectively input into the initial sample feature generation model in the current round, and the mean, variance, generated sample features, true / false degree scores, and embedding vectors corresponding to each seen fault category in the current round are obtained; wherein, when the current round is the initial round, the initial sample feature generation model in the current round is the initialized sample feature generation model, and when the current round is not the initial round, the initial sample feature generation model in the current round is the target sample feature generation model in the previous round; According to the mean, variance, generated sample features, true / false degree scores, and embedding vectors corresponding to all seen fault categories in the current round, calculate the total loss value in the current round; Determine whether the stop condition is satisfied; the stop condition is that the total loss value in the current round is less than the preset loss value or the current round reaches the preset number of training rounds; If so, determine the initial sample feature generation model in the current round as the trained sample feature generation model; If not, update the parameters of the initial sample feature generation model in the current round to obtain the target sample feature generation model in the current round, update the current round to the next round, and continue until the stop condition is satisfied to obtain the trained sample feature generation model.

[0011] Optionally, the original sample features of each seen fault category in the training dataset of the seen fault categories and the hierarchical semantic of each seen fault category in the hierarchical semantic set of the seen fault category space are respectively input into the initial sample feature generation model in the current round, and the mean, variance, generated sample features, true / false degree scores, and embedding vectors corresponding to each seen fault category in the current round are obtained, including: Determine any seen fault category in the training dataset of the seen fault categories as the current seen fault category; Input the original sample features and hierarchical semantic of the current seen fault category into the encoder of the initial sample feature generation model in the current round to obtain the mean and variance corresponding to the current seen fault category in the current round; Sampling is performed based on the mean and variance of each current round to obtain the noise vector of the current round corresponding to the currently seen category fault; Input the noise vector and spatial level semantics of the current round corresponding to the currently seen category fault into the decoder in the initial sample feature generation model of the current round, and obtain the generated sample features of the current round corresponding to the currently seen category fault; the generated sample features include k layers of features; Input the original sample features of the currently seen category fault, the generated sample features in the current round, and the spatial level semantics into the discriminator in the initial sample feature generation model in the current round, and obtain the true or false degree score of the discriminator output in the current round corresponding to the currently seen category fault; The generated sample features and spatial hierarchical semantics of the current round of the currently seen category fault are input into the hierarchical semantic information alignment module in the initial sample feature generation model of the current round to obtain the embedding vector of the current round corresponding to the currently seen category fault.

[0012] Optionally, the total loss function includes: .

[0013] ; ; ; ; in, represents the total loss value; Represents the loss value of the conditional variational autoencoder module; represents the loss weight of the conditional generative adversarial network module; Represents the loss value of the conditional generation adversarial network module; Represents the loss weight of the hierarchical semantic information alignment module; Represents the loss value of the hierarchical semantic information alignment module; Represents the total number of seen category faults in the training dataset of seen categories of faults; represents the KL information divergence; represents normal distribution; represents the mean value corresponding to the i-th category fault; represents the variance corresponding to the i-th seen category fault; Represents the original sample features of the i-th seen category fault; Represents the k-th layer feature in the generated sample features corresponding to the i-th seen category fault; express and The Euclidean distance between represents the loss of the gradient penalty of the discriminator; It means that the original sample features and spatial level semantics of the i-th seen category fault are input into the discriminator, and the discriminator outputs the true or false degree score; It means that the generated sample features and spatial level semantics of the i-th seen category fault are input into the discriminator, and the discriminator outputs the true or false degree score; Represents the gradient penalty coefficient; Represents the function gradient operation; express and Random interpolation samples between ; represents a multilayer perceptron; Represents the gradient modulus of the discriminator at the random interpolation sample; It indicates that the generated sample features corresponding to the i-th seen category fault are input into the hierarchical semantic information alignment module, and the j-th layer embedding vector output by the hierarchical semantic information alignment module; It indicates that the spatial hierarchical semantics corresponding to the i-th seen category fault is input into the hierarchical semantic information alignment module, and the j-th layer embedding vector output by the hierarchical semantic information alignment module; express and The Euclidean distance between .

[0014] Optionally, a trained sample feature generation model is used to randomly generate a plurality of generated sample features and labels corresponding to features of unseen fault categories, to obtain a training data set of unseen fault categories, including: Multiple noise vectors are obtained by random sampling according to the standard normal distribution, and labels and spatial level semantics of multiple groups of unseen fault categories are obtained by random sampling in unseen fault categories; The noise vector and spatial level semantics of each unseen fault category group are respectively input into the decoder of the trained sample feature generation model to obtain the corresponding generated sample features; The fault unseen category training data set is constructed according to the generated sample features and labels corresponding to each fault unseen category group.

[0015] Optionally, the initial network is trained using the equipment fault diagnosis training data set to obtain the fault diagnosis model, including: The generated sample features corresponding to each unseen category group of faults in the equipment fault diagnosis training data set and the original sample features of each seen category fault are used as input, and the corresponding labels are used as output to train the initial network to obtain the fault diagnosis model.

[0016] According to the specific embodiments provided by the present application, the present application discloses the following technical effects: the present application discloses a method for spatial hierarchical semantic construction and equipment fault diagnosis. First, the characteristics of each device in the system to be diagnosed are obtained; the characteristics are vectors after feature extraction of operation data and environmental data; then, the characteristics of each device in the system to be diagnosed are input into the fault diagnosis model to obtain the predicted value of the fault category in the system to be diagnosed; the fault diagnosis model is determined based on a fault category data set, a fault category spatial hierarchical semantic set, a sample feature generation model and a multilayer perceptron; the fault category data set includes: original sample features and labels of multiple seen category faults, and the fault category spatial hierarchical semantic set includes: spatial hierarchical semantics of multiple seen category faults in the fault category data set; the label is the actual value of the fault category, the spatial hierarchical semantics is a vector composed of descriptive attribute values ​​of the fault, and the spatial hierarchical semantics includes k layers of semantics. The present application uses a fault diagnosis model determined by a seen fault category data set, a seen fault category spatial hierarchical semantic set, a sample feature generation model, and a multi-layer perceptron to perform equipment fault diagnosis. The seen fault category spatial hierarchical semantic set includes spatial hierarchical semantics related to equipment faults. The spatial hierarchical semantics are used to drive the generation of generated sample features corresponding to the unseen fault category group, thereby improving the accuracy of generalized zero-sample equipment fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A flowchart of a method for constructing spatial hierarchical semantics and diagnosing equipment faults provided in one embodiment of the present application.

[0019] Figure 2 Generate a model structure diagram for sample features.

[0020] Figure 3 Schematic diagram of the decoder (generator) structure.

[0021] Figure 4 Schematic diagram of the hierarchical semantic information alignment module structure.

[0022] Figure 5 Schematic diagram of the spatial level semantics of the Tennessee-Eastman dataset. DETAILED DESCRIPTION

[0023] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.

[0024] The purpose of the present application is to provide a method for constructing spatial hierarchical semantics and diagnosing equipment faults, aiming to improve the accuracy of equipment fault diagnosis.

[0025] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0026] In an exemplary embodiment, as Figure 1 shown, a method for constructing spatial hierarchical semantics and diagnosing equipment faults is provided, including Step 1 - Step 2.

[0027] Step 1: Obtain the features of each device in the system to be diagnosed; the features are vectors obtained by extracting features from operation data and environmental data.

[0028] Specifically, the operation data and environmental data are adjusted according to the actual situation. The operation data includes, but is not limited to, pressure, flow rate, current, and power. The environmental data includes, but is not limited to, humidity and temperature.

[0029] Step 2: Input the features of each device in the system to be diagnosed into the fault diagnosis model to obtain the predicted values of the fault categories in the system to be diagnosed.

[0030] Among them, the fault diagnosis model is determined based on the fault seen category data set, the fault seen category spatial hierarchical semantics set, the sample feature generation model, and the multi-layer perceptron; the fault seen category data set includes: the original sample features and labels of multiple seen category faults, and the fault seen category spatial hierarchical semantics set includes: the spatial hierarchical semantics of multiple seen category faults in the fault seen category data set; the label is the actual value of the fault category, and the spatial hierarchical semantics is a vector composed of the descriptive attribute values of the fault. The spatial hierarchical semantics includes k layers of semantics, where k > 1.

[0031] As an optional implementation manner, the determination process of the fault diagnosis model includes Step 21 - Step 27.

[0032] Step 21: Obtain the fault seen category training data set and the fault seen category spatial hierarchical semantics set.

[0033] Specifically, the fault seen category training data set is expressed as: 。

[0034] Among them, represents the original sample feature of the i-th seen-class fault in the seen-class fault training dataset for faults, , represents the total number of seen-class faults in the seen-class fault training dataset for faults; represents the label of the i-th seen-class fault in the seen-class fault training dataset for faults.

[0035] The spatial hierarchical semantics is a vector composed of the description attribute values from top to bottom based on the system, equipment, components, and fault attribute information where the fault is located in the industrial scenario. The seen-class fault spatial hierarchical semantics set is expressed as: 。

[0036] Among them, represents the spatial hierarchical semantics of the i-th seen-class fault in the seen-class fault spatial hierarchical semantics set, , represents a real number, represents the number of seen-class faults of the i-th seen-class fault in the seen-class fault spatial hierarchical semantics set, represents the dimension of the spatial hierarchical semantics of the i-th seen-class fault in the seen-class fault spatial hierarchical semantics set; , represents the j-th layer semantics of the spatial hierarchical semantics of the i-th seen-class fault in the seen-class fault spatial hierarchical semantics set, 。

[0037] Step 22: Initialize the parameters of the sample feature generation model.

[0038] As an optional implementation manner, as Figure 2-Figure 4 shown, the sample feature generation model includes: a conditional variational autoencoder module, a conditional generative adversarial network module, and a hierarchical semantic information alignment module.

[0039] The conditional variational autoencoder module includes: an encoder and a decoder; the encoder is implemented using a multi-layer perceptron, and the decoder is composed of a multi-layer perceptron consisting of a k-layer fully connected network and an activation function.

[0040] The conditional generative adversarial network module includes: a generator and a discriminator; the generator reuses the decoder of the conditional variational autoencoder module, and the discriminator is implemented using a multi-layer perceptron.

[0041] The hierarchical semantic information alignment module includes: k layers of semantic information alignment layers.

[0042] Specifically, each semantic information alignment layer includes: multiple fully connected layers and multiple activation layers.

[0043] Step 23: Based on the total loss function, the sample feature generation model is trained using the fault category training data set and the fault category space level semantic set to obtain a trained sample feature generation model.

[0044] As an optional implementation, step 23 includes step 231.

[0045] Step 231: Based on the total loss function, the sample feature generation model is trained for multiple rounds using the fault category training data set and the fault category space level semantic set to obtain a trained sample feature generation model; wherein, any current round of training process includes steps 2311-2315.

[0046] Step 2311: input the original sample features of each seen category fault in the seen category training data set and the spatial hierarchical semantics of each seen category fault in the seen category spatial hierarchical semantic set into the initial sample feature generation model of the current round respectively, and obtain the mean, variance, generated sample features, true or false degree score and embedding vector of each seen category fault in the current round; wherein, when the current round is the initial round, the initial sample feature generation model of the current round is the initialized sample feature generation model, and when the current round is a non-initial round, the initial sample feature generation model of the current round is the target sample feature generation model of the previous round.

[0047] As an optional implementation, step 2311 includes steps 23111 to 23116.

[0048] Step 23111: Determine any seen category fault in the seen category training data set as the current seen category fault.

[0049] Step 23112: Input the original sample features and spatial hierarchical semantics of the currently seen category fault into the encoder in the initial sample feature generation model in the current round, and obtain the mean and variance in the current round corresponding to the currently seen category fault.

[0050] Specifically, the original sample features of the i-th seen category fault in the seen category training data set are and the corresponding spatial level semantics Input into the encoder to get the mean of the corresponding normal distribution and variance The expression is as follows.

[0051] .

[0052] .

[0053] in, Representing spatial level semantics The result of vector concatenation of the semantics of all layers in ; Represents a vector concatenation operation; represents the encoder; Represents the original sample features of the i-th seen category fault.

[0054] Step 23113: Sampling is performed based on the mean and variance in each current round to obtain the noise vector in the current round corresponding to the currently seen category fault.

[0055] Specifically, according to the mean and variance Sampling is performed to obtain the noise vector The expression is as follows.

[0056] .

[0057] in, Indicates sampling operation; Represents a normal distribution.

[0058] Step 23114: Input the noise vector and spatial level semantics of the current round corresponding to the currently seen category fault into the decoder in the initial sample feature generation model of the current round to obtain the generated sample features of the current round corresponding to the currently seen category fault; the generated sample features include k layers of features.

[0059] Specifically, the noise vector and layer 1 semantics Input to the first layer of the decoder ( ) Get the first layer feature of the generated sample feature corresponding to the i-th seen category fault ;Will and layer 2 semantics Input to the second layer of the decoder ( ) Get the second layer features in the generated sample features corresponding to the i-th seen category fault , and so on, the k-1th layer feature of the generated sample feature corresponding to the i-th seen category fault and the k-th semantic Input decoder layer k ( ) Get the k-th layer feature in the generated sample feature corresponding to the i-th seen category fault The expression is as follows.

[0060] .

[0061] in, Represents the j-th layer feature in the generated sample features corresponding to the i-th seen category fault; represents the jth layer of the decoder; Represents the j-1th layer feature in the generated sample features corresponding to the i-th seen category fault.

[0062] Step 23115: Input the original sample features of the currently seen category fault, the generated sample features in the current round, and the spatial hierarchical semantics into the discriminator in the initial sample feature generation model in the current round, and obtain the truth or falsehood score of the discriminator output in the current round corresponding to the currently seen category fault.

[0063] Specifically, the original sample features of the i-th seen category fault in the seen category training data set are and the corresponding spatial level semantics The result of vector concatenation of all semantic layers in the semantics Input to the discriminator to get the true or false degree score , the k-th layer feature in the generated sample feature corresponding to the i-th seen category fault generated by the generator and Input the discriminator to get the true or false degree score The expression is as follows.

[0064] .

[0065] .

[0066] in, represents the discriminator.

[0067] Step 23116: Input the generated sample features and spatial hierarchical semantics of the current round of the currently seen category fault into the hierarchical semantic information alignment module in the initial sample feature generation model of the current round to obtain the embedding vector of the current round corresponding to the currently seen category fault.

[0068] Specifically, the sample features will be generated and Input to the hierarchical semantic information alignment module, and then embed it to obtain the embedding vector in the alignment space. and .in, Indicates the generated sample features corresponding to the i-th seen category fault Input to the hierarchical semantic information alignment module, the j-th layer embedding vector output by the hierarchical semantic information alignment module; Indicates the spatial level semantics corresponding to the i-th seen category fault Input to the hierarchical semantic information alignment module, and the hierarchical semantic information alignment module outputs the j-th layer embedding vector.

[0069] Step 2312: Calculate the total loss value in the current round based on the mean, variance, generated sample features, true or false degree scores and embedding vectors in the current round corresponding to all seen category faults.

[0070] As an optional implementation, the total loss function includes: .

[0071] .

[0072] .

[0073] .

[0074] .

[0075] in, represents the total loss value; Represents the loss value of the conditional variational autoencoder module; represents the loss weight of the conditional generative adversarial network module; Represents the loss value of the conditional generation adversarial network module; Represents the loss weight of the hierarchical semantic information alignment module; Represents the loss value of the hierarchical semantic information alignment module; Represents the total number of seen category faults in the training dataset of seen categories of faults; represents the KL information divergence; represents normal distribution; represents the mean value corresponding to the i-th category fault; represents the variance corresponding to the i-th seen category fault; Represents the original sample features of the i-th seen category fault; Represents the k-th layer feature in the generated sample features corresponding to the i-th seen category fault; express and The Euclidean distance between represents the loss of the gradient penalty of the discriminator; It means that the original sample features and spatial level semantics of the i-th seen category fault are input into the discriminator, and the discriminator outputs the true or false degree score; It means that the generated sample features and spatial level semantics of the i-th seen category fault are input into the discriminator, and the discriminator outputs the true or false degree score; Represents the gradient penalty coefficient; Represents the function gradient operation; express and Random interpolation samples between ; represents a multilayer perceptron; Represents the gradient modulus of the discriminator at the random interpolation sample; It indicates that the generated sample features corresponding to the i-th seen category fault are input into the hierarchical semantic information alignment module, and the j-th layer embedding vector output by the hierarchical semantic information alignment module; It indicates that the spatial hierarchical semantics corresponding to the i-th seen category fault is input into the hierarchical semantic information alignment module, and the j-th layer embedding vector output by the hierarchical semantic information alignment module; express and The Euclidean distance between .

[0076] Step 2313: Determine whether the stopping condition is met; the stopping condition is that the total loss value in the current round is less than the preset loss value or the current round reaches the preset training round.

[0077] Step 2314: If yes, the initial sample feature generation model in the current round is determined as the trained sample feature generation model.

[0078] Step 2315: If not, update the parameters of the initial sample feature generation model in the current round to obtain the target sample feature generation model in the current round, and update the current round to the next round until the stopping condition is met to obtain a trained sample feature generation model.

[0079] Step 24: Use the trained sample feature generation model to randomly generate sample features and labels corresponding to multiple unseen fault category features to obtain an unseen fault category training data set.

[0080] As an optional implementation, step 24 includes steps 241 to 243.

[0081] Step 241: randomly sampling according to standard normal distribution to obtain multiple noise vectors, and randomly sampling in unseen fault categories to obtain labels and spatial level semantics of multiple unseen fault category groups.

[0082] Specifically, multiple noise vectors are obtained by random sampling according to the standard normal distribution , randomly sample the unseen fault categories to obtain labels for multiple unseen fault category groups and spatial level semantics .

[0083] in, represents the noise vector corresponding to the g-th fault unseen category group in the fault unseen category training data set, , Represents the number of fault-unseen category groups in the fault-unseen category training dataset; Represents the label of the g-th fault unseen category group in the fault unseen category training dataset; Represents the spatial hierarchical semantics of the g-th fault unseen category group in the fault unseen category training dataset.

[0084] Step 242: Input the noise vector and spatial level semantics of each unseen fault category group into the decoder in the trained sample feature generation model respectively to obtain the corresponding generated sample features.

[0085] Specifically, the noise vector of each fault unseen category group is and spatial level semantics Input them into the decoder of the trained sample feature generation model respectively to obtain the corresponding generated sample features , Represents the generated sample features of the g-th fault unseen category group in the fault unseen category training dataset.

[0086] Step 243: construct a training data set of unseen fault categories according to the generated sample features and labels corresponding to each unseen fault category group.

[0087] Specifically, the fault category training dataset It is expressed as: .

[0088] Step 25: Merge the training data set of the seen fault category and the training data set of the unseen fault category to obtain the equipment fault diagnosis training data set.

[0089] Specifically, the fault category training data set and training datasets for unseen fault categories Merge to obtain the equipment fault diagnosis training data set .

[0090] Step 26: Construct the initial network of the fault diagnosis model based on the multi-layer perceptron.

[0091] Step 27: Use the equipment fault diagnosis training data set to train the initial network to obtain a fault diagnosis model.

[0092] As an optional implementation, step 27 includes step 271.

[0093] Step 271: Using the generated sample features corresponding to each unseen category group of faults in the equipment fault diagnosis training data set and the original sample features of each seen category fault as input and the corresponding labels as output, the initial network is trained to obtain a fault diagnosis model.

[0094] Further, after step 27, the method further includes: testing the fault-seen category data set The diagnostic accuracy of the tested equipment failure categories , in the fault unseen category test dataset The diagnostic accuracy of the above test equipment failure was not found , and calculate the harmonic mean accuracy .

[0095] in, represents the original sample features of the qth fault-seen category group in the fault-seen category test dataset, , represents the number of fault-seen category groups in the fault-seen category test data set; Represents the label of the qth fault category group in the fault category test dataset; represents the original sample features of the qth fault-unseen category group in the fault-unseen category test dataset, , represents the number of the qth fault unseen category group in the fault unseen category test data set; Represents the labels of the fault-unseen class groups in the fault-unseen class test dataset.

[0096] In order to explore the method of the present application, the method of the present application was also used to perform fault diagnosis on the Tennessee-Eastman simulation platform, and the specific contents are as follows.

[0097] The Tennessee-Eastman simulation platform does not involve the division of different subsystems, so the construction of spatial hierarchical semantics starts from the equipment level. The specific locations where the fault occurs in the Tennessee-Eastman simulation platform include reactors, condensers, and various feed pipelines. Based on this, the relevant first-level spatial semantics of the fault, that is, the equipment level, can be constructed. The fault description in the Tennessee-Eastman simulation platform does not involve specific equipment components, so the second-level spatial semantics is constructed based on the relevant variables involved in the fault. The fault description in the Tennessee-Eastman simulation platform also includes detailed disturbance types for each fault, so the third-level spatial semantics is constructed based on the disturbance type in the fault description, such as Figure 5 As shown, k for the Tennessee-Eastman simulation platform is 3.

[0098] The Tennessee-Eastman dataset contains 15 different types of faults. Each type of fault sample is obtained under 24-hour simulation, and the total number of fault samples for each type is 480. In order to perform generalized zero-sample fault diagnosis, 12 types of faults are regarded as seen fault categories, and 3 types of faults are regarded as unseen fault categories. A total of 5 different grouping strategies are used for division, as shown in Table 1.

[0099] The Tennessee-Eastman dataset was used for performance verification. The method of this application and five other methods were used to perform generalized zero-shot fault diagnosis tasks on five groups of fault datasets with different partitions. Their performance was compared and the accuracy was tested using seen categories. , Unseen category test accuracy And their harmonic mean accuracy As the evaluation index, the results are shown in Table 2. Compared with other methods, the method of the present application has the highest harmonic mean accuracy in groups A, D and E, which is 11 percentage points higher than the second best method in group A, 8.7 percentage points higher than the second best method in group D, and 17.5 percentage points higher than the second best method in group E. The effect of the method of the present application is average in group B, and the effect of the method of the present application ranks second in group C. Overall, the method of the present application has a significant improvement in effect.

[0100] Table 1 Grouping strategy table

[0101] Table 2 Performance of different methods for generalized zero-shot fault diagnosis on the Tennessee-Eastman dataset

[0102] Among them, the ALE algorithm comes from the paper Z. Akata, F. Perronnin, Z. Harchaoui, and C. Schmid, "Label-embedding for image classification," IEEE Trans. Pattern Anal.Mach. Intell., vol. 38, no. 7, pp. 1425–1438, Jul. 2016.; the CS algorithm comes from the paper Y. Le Cacheux, H. Le Borgne, and M. Crucianu, "From classical to generalized zero-shot learning: A simple adaptation process," in Proc. 25th Int. Conf. MultiMedia Model., 2019, pp. 465–477; the CEWGAN algorithm comes from the paper D. Mandal et al., "Out-of-distribution detection for generalized zero-shot action recognition," in Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit., 2019, pp. 9985–9993; the VSG algorithm comes from the paper C. Geng, L. Tao, and S. Chen, “Guided CNN for generalized zero-shot and open-set recognition using visual and semantic prototypes,” Pattern Recognit., vol. 102, pp. 1–10, 2020; the AE-M2 algorithm comes from the paper S. Bhattacharjee, D. Mandal, and S. Biswas, “Autoencoder based novelty detection for generalized zero shot learning,” in Proc. IEEE Int. Conf. Image Process., 2019, pp. 3646–3650.

[0103] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0104] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0105] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A spatial hierarchical semantic construction and equipment fault diagnosis method, characterized in that: The spatial level semantic construction and equipment fault diagnosis method comprises: Acquire the characteristics of each device in the system to be diagnosed; the characteristics are vectors obtained by extracting the characteristics of the operation data and the environmental data; The characteristics of each device in the system to be diagnosed are input into the fault diagnosis model to obtain the predicted value of the fault category in the system to be diagnosed; the fault diagnosis model is determined based on a data set of seen fault categories, a spatial hierarchical semantic set of seen fault categories, a sample feature generation model and a multi-layer perceptron; the seen fault category data set includes: original sample features and labels of multiple seen category faults, and the seen fault category spatial hierarchical semantic set includes: spatial hierarchical semantics of multiple seen category faults in the seen fault category data set; the label is the actual value of the fault category, the spatial hierarchical semantics is a vector composed of descriptive attribute values ​​of the fault, and the spatial hierarchical semantics includes k layers of semantics, k>1.

2. The spatial hierarchical semantic construction and equipment fault diagnosis method according to claim 1 is characterized in that: The process of determining the fault diagnosis model includes: Obtain a training data set of seen fault categories and a spatial level semantic set of seen fault categories; Initializing parameters of the sample feature generation model; Based on the total loss function, the sample feature generation model is trained using the fault category training data set and the fault category space level semantic set to obtain a trained sample feature generation model; The trained sample feature generation model is used to randomly generate sample features and labels corresponding to multiple unseen fault category features, and a training data set of unseen fault categories is obtained; Merging the training data set of the seen fault category and the training data set of the unseen fault category to obtain a training data set for equipment fault diagnosis; Constructing an initial network of the fault diagnosis model based on a multi-layer perceptron; The initial network is trained using the equipment fault diagnosis training data set to obtain the fault diagnosis model.

3. The spatial hierarchical semantic construction and equipment fault diagnosis method according to claim 2 is characterized in that: The sample feature generation model includes: a conditional variational autoencoder module, a conditional generative adversarial network module and a hierarchical semantic information alignment module; The conditional variational autoencoder module includes: an encoder and a decoder; the encoder is implemented by a multi-layer perceptron, and the decoder is composed of a multi-layer perceptron composed of a k-layer fully connected network and an activation function; The conditional generative adversarial network module includes: a generator and a discriminator; the generator reuses the decoder of the conditional variational autoencoder module, and the discriminator is implemented by a multi-layer perceptron; The hierarchical semantic information alignment module includes: k layers of semantic information alignment layers.

4. The spatial hierarchical semantic construction and equipment fault diagnosis method according to claim 3 is characterized in that: Based on the total loss function, the sample feature generation model is trained using the fault category training data set and the fault category space level semantic set to obtain a trained sample feature generation model, including: Based on the total loss function, the sample feature generation model is trained for multiple rounds using the fault category training data set and the fault category space level semantic set to obtain a trained sample feature generation model; wherein any current round of training process includes: The original sample features of each seen category fault in the seen category training data set and the spatial hierarchical semantics of each seen category fault in the seen category spatial hierarchical semantic set are respectively input into the initial sample feature generation model in the current round, and the mean, variance, generated sample features, true or false degree score and embedded vector corresponding to each seen category fault in the current round are obtained; wherein, when the current round is the initial round, the initial sample feature generation model in the current round is the initialized sample feature generation model, and when the current round is the non-initial round, the initial sample feature generation model in the current round is the target sample feature generation model in the previous round; Calculate the total loss value in the current round based on the mean, variance, generated sample features, true or false degree scores, and embedding vectors corresponding to all seen category faults in the current round; Determine whether the stopping condition is met; the stopping condition is that the total loss value in the current round is less than the preset loss value or the current round reaches the preset training round; If yes, the initial sample feature generation model in the current round is determined as the trained sample feature generation model; If not, update the parameters of the initial sample feature generation model in the current round to obtain the target sample feature generation model in the current round, and update the current round to the next round until the stopping condition is met to obtain the trained sample feature generation model.

5. The spatial hierarchical semantic construction and equipment fault diagnosis method according to claim 4 is characterized in that: The original sample features of each seen category fault in the seen category training data set and the spatial hierarchical semantics of each seen category fault in the seen category spatial hierarchical semantic set are respectively input into the initial sample feature generation model in the current round, and the mean, variance, generated sample features, true or false degree score and embedded vector corresponding to each seen category fault in the current round are obtained, including: Determine any seen category fault in the seen category training data set as a currently seen category fault; Input the original sample features and spatial level semantics of the currently seen category fault into the encoder in the initial sample feature generation model in the current round, and obtain the mean and variance in the current round corresponding to the currently seen category fault; Sampling is performed based on the mean and variance of each current round to obtain the noise vector of the current round corresponding to the currently seen category fault; Input the noise vector and spatial level semantics of the current round corresponding to the currently seen category fault into the decoder in the initial sample feature generation model of the current round, and obtain the generated sample features of the current round corresponding to the currently seen category fault; the generated sample features include k layers of features; Input the original sample features of the currently seen category fault, the generated sample features in the current round, and the spatial level semantics into the discriminator in the initial sample feature generation model in the current round, and obtain the true or false degree score of the discriminator output in the current round corresponding to the currently seen category fault; The generated sample features and spatial hierarchical semantics of the current round of the currently seen category fault are input into the hierarchical semantic information alignment module in the initial sample feature generation model of the current round to obtain the embedding vector of the current round corresponding to the currently seen category fault.

6. The spatial hierarchical semantic construction and equipment fault diagnosis method according to claim 5 is characterized in that: The total loss function includes: 。 ； ； ； ； in, represents the total loss value; Represents the loss value of the conditional variational autoencoder module; represents the loss weight of the conditional generative adversarial network module; Represents the loss value of the conditional generation adversarial network module; Represents the loss weight of the hierarchical semantic information alignment module; Represents the loss value of the hierarchical semantic information alignment module; Represents the total number of seen category faults in the training dataset of seen categories of faults; represents the KL information divergence; represents normal distribution; represents the mean value corresponding to the i-th category fault; represents the variance corresponding to the i-th seen category fault; Represents the original sample features of the i-th seen category fault; Represents the k-th layer feature in the generated sample features corresponding to the i-th seen category fault; express and The Euclidean distance between represents the loss of the gradient penalty of the discriminator; It means that the original sample features and spatial level semantics of the i-th seen category fault are input into the discriminator, and the discriminator outputs the true or false degree score; It means that the generated sample features and spatial level semantics of the i-th seen category fault are input into the discriminator, and the discriminator outputs the true or false degree score; Represents the gradient penalty coefficient; Represents the function gradient operation; express and Random interpolation samples between ; represents a multilayer perceptron; Represents the gradient modulus of the discriminator at the random interpolation sample; It indicates that the generated sample features corresponding to the i-th seen category fault are input into the hierarchical semantic information alignment module, and the j-th layer embedding vector output by the hierarchical semantic information alignment module; It indicates that the spatial hierarchical semantics corresponding to the i-th seen category fault is input into the hierarchical semantic information alignment module, and the j-th layer embedding vector output by the hierarchical semantic information alignment module; express and The Euclidean distance between .

7. The method for constructing spatial hierarchical semantics and diagnosing equipment faults according to claim 5, characterized in that: The trained sample feature generation model is used to randomly generate sample features and labels corresponding to multiple unseen fault category features, and the unseen fault category training data set is obtained, including: Multiple noise vectors are obtained by random sampling according to the standard normal distribution, and labels and spatial level semantics of multiple groups of unseen fault categories are obtained by random sampling in unseen fault categories; The noise vector and spatial level semantics of each unseen fault category group are respectively input into the decoder of the trained sample feature generation model to obtain the corresponding generated sample features; The fault unseen category training data set is constructed according to the generated sample features and labels corresponding to each fault unseen category group.

8. The method for constructing spatial hierarchical semantics and diagnosing equipment faults according to claim 2, characterized in that: The initial network is trained using the equipment fault diagnosis training data set to obtain the fault diagnosis model, including: The generated sample features corresponding to each unseen category group of faults in the equipment fault diagnosis training data set and the original sample features of each seen category fault are used as input, and the corresponding labels are used as output to train the initial network to obtain the fault diagnosis model.

Images