Knowledge-driven equipment predictive maintenance method and device

Abstract

The invention provides an equipment predictive maintenance method and device based on knowledge driving, and relates to the technical field of data processing. Feature extraction is carried out on working parameters of industrial equipment through a pre-constructed feature extraction model, and dimension reduction processing is carried out after data dynamic remodeling operation is carried out on target features; and carrying out classification identification on the dimension reduction feature representation by using a pre-constructed classifier model, and determining an equipment fault phenomenon indicated by the current working parameter of the industrial equipment. Wherein the feature extraction model carries out network parameter updating based on a feature nonlinear dependency relationship corresponding to a preset training sample set, a more complex feature relationship can be mined, a coding layer of the auto-encoder is obtained based on sparse optimization of a nonlinear coupling optimization strategy, and on the premise of enhancing an internal relationship between features, a more complex feature relationship can be mined. And the dimension reduction process is adaptively adjusted according to the data change and the mode, and the discrimination of feature representation is maximized, so that the potential fault phenomenon of the equipment under the current working parameters can be accurately captured.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a knowledge-driven equipment predictive maintenance method and device. Background Art

[0002] With the improvement of industrial automation and intelligence, the way of equipment maintenance has gradually changed from traditional periodic maintenance to condition-based maintenance, and then to predictive maintenance. Predictive maintenance monitors equipment status in real time, analyzes equipment data, and predicts equipment failures, so as to perform maintenance in advance, reduce unexpected downtime, reduce maintenance costs, and improve production efficiency.

[0003] Among them, actual industrial data often contains rich nonlinear structures. Traditional feature extraction methods are based on linear models, which cannot fully capture these complex nonlinear dependencies, resulting in limited feature representation capabilities. When processing high-dimensional, strongly coupled data generated by complex industrial equipment, it is easy to fall into local minima. As a result, the feature extraction process cannot fully explore the potential nonlinear dependencies within the data, which in turn affects the accuracy and robustness of feature representation. Moreover, when faced with dynamically changing industrial data, existing processing methods may lose important information, or the retained features may not be able to distinguish the equipment status under different working conditions, reducing the effect of subsequent classification and recognition. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a knowledge-driven equipment predictive maintenance method and device, which can process complex linear data of equipment operating parameters, accurately identify target features, and accurately capture and identify equipment faults based on current operating parameters.

[0005] In a first aspect, an embodiment of the present invention provides a knowledge-driven predictive maintenance method for equipment, the method comprising: obtaining working parameters of preset industrial equipment, performing data preprocessing on the working parameters, and generating a vector to be tested; inputting the vector to be tested into a pre-built feature extraction model, and extracting target features corresponding to the vector to be tested; wherein the feature extraction model updates network parameters based on the feature nonlinear dependency relationship corresponding to a preset training sample set; performing a data dynamic reshaping operation on the target features, and using a preset autoencoder to perform dimensionality reduction processing on the dynamically reshaped target features to generate a reduced-dimensionality feature representation; wherein the encoding layer of the autoencoder is obtained by sparse optimization based on a nonlinear coupling optimization strategy; and using a pre-built classifier model to classify and identify the reduced-dimensionality feature representation to determine the equipment failure phenomenon indicated by the current working parameters of the industrial equipment.

[0006] In combination with the first aspect, an embodiment of the present invention provides a first implementation method of the first aspect, wherein a step of classifying and identifying the reduced-dimensionality feature representation using a pre-built classifier model to determine the equipment failure phenomenon indicated by the current working parameters of the industrial equipment includes: inputting the reduced-dimensionality feature representation into the pre-built classifier model, and outputting the equipment failure phenomenon classification result; using a preset predictive maintenance knowledge graph system to diagnose the equipment failure phenomenon classification result, and determine the potential failure cause corresponding to the equipment failure phenomenon classification result.

[0007] In combination with the first aspect, an embodiment of the present invention provides a second implementation of the first aspect, wherein the step of dynamically reshaping the data of the target feature includes: analyzing the statistical characteristics corresponding to the target feature, calculating the data reshaping factor corresponding to the target feature; using a preset autoencoder to perform preliminary dimensionality reduction processing on the target feature, determining the loss function corresponding to the autoencoder, and determining the weight matrix reshaping factor corresponding to the loss function; calculating a reshaping matrix based on the data reshaping factor and the weight matrix reshaping factor; and using the reshaping matrix to dynamically reshape the data of the target feature.

[0008] In combination with the first aspect, an embodiment of the present invention provides a third implementation of the first aspect, wherein the step of sparsely optimizing the encoding layer of the autoencoder based on the nonlinear coupling optimization strategy includes: using a preset coupling degree constraint function to determine the nonlinear correlation between the features of a preset training sample set; based on the nonlinear correlation and the preset feature sparsity constraint, determining the preset loss function of the autoencoder; based on the loss function and the preset momentum factor, back-propagating and updating the weight matrix of the autoencoder to optimize the encoding layer of the autoencoder.

[0009] In combination with the first aspect, an embodiment of the present invention provides a fourth implementation of the first aspect, wherein a method for constructing a classifier model includes: inputting a preset training sample set into a preset support vector machine, and adjusting the support vector machine parameters of the support vector machine using a chaos theory algorithm; determining the Lagrange multipliers corresponding to the training sample set, and adjusting the parameters of the kernel function of the support vector machine based on the Lagrange multipliers; until the support vector machine meets the preset training conditions, a classifier model is constructed based on the support vector machine.

[0010] In combination with the first aspect, an embodiment of the present invention provides a fifth implementation of the first aspect, wherein the step of adjusting the support vector machine parameters of the support vector machine using a chaos theory algorithm includes: generating random perturbations using chaos theory; and adjusting the support vector machine parameters of the support vector machine based on the random perturbations using a preset non-convex optimization algorithm.

[0011] In combination with the first aspect, an embodiment of the present invention provides a sixth implementation of the first aspect, wherein a method for constructing a feature extraction model includes: obtaining a preset training sample set, performing nonlinear decomposition on the training sample set using a nonlinear feature dual decomposition algorithm to obtain multiple dual components; inputting the multiple dual components into a preset neural network for forward propagation, and determining the loss function corresponding to the neural network; updating the network parameters of the neural network based on the loss function and the characteristic nonlinear dependency relationship indicated by the dual components; until the neural network meets the preset iteration conditions, constructing a feature extraction model based on the neural network.

[0012] In combination with the first aspect, an embodiment of the present invention provides a seventh implementation method of the first aspect, wherein a method for constructing a preset training sample set includes: obtaining pre-stored working parameters of preset industrial production equipment, labeling the working parameters, and constructing an initial sample set; generating random noise corresponding to the initial sample set, and obtaining current operating data of the industrial production equipment; determining the coupling input corresponding to the industrial production equipment based on the random noise and the current operating data; and performing data expansion on the initial sample set based on the coupling input to construct a training sample set.

[0013] In combination with the first aspect, an embodiment of the present invention provides an eighth implementation of the first aspect, wherein the step of performing data expansion on the initial sample set based on the coupled input to construct a training sample set includes: inputting the coupled input into a pre-trained generative adversarial network, and using the generative adversarial network to generate expanded samples corresponding to the coupled input; merging the expanded samples with the initial sample set to construct a training sample set; wherein the generative adversarial network performs exponential calculation on the output of the discriminator to update the generator.

[0014] In the second aspect, an embodiment of the present invention provides a knowledge-driven equipment predictive maintenance device, wherein the device includes: a data acquisition module, used to obtain the working parameters of preset industrial equipment, perform data preprocessing on the working parameters, and generate a vector to be tested; a feature extraction module, used to input the vector to be tested into a pre-built feature extraction model, and extract the target features corresponding to the vector to be tested; wherein the feature extraction model updates the network parameters based on the feature nonlinear dependency relationship corresponding to the preset training sample set; a data processing module, used to perform data dynamic reshaping operations on the target features, and use a preset autoencoder to reduce the dimension of the dynamically reshaped target features to generate a reduced-dimensional feature representation; wherein the encoding layer of the autoencoder is sparsely optimized based on a nonlinear coupling optimization strategy; an execution module, used to use a pre-built classifier model to classify and identify the reduced-dimensional feature representation, and determine the equipment failure phenomenon indicated by the current working parameters of the industrial equipment.

[0015] The embodiments of the present invention bring the following beneficial effects: the knowledge-driven predictive maintenance method and device provided by the present invention can mine more complex feature relationships in high-dimensional space and improve the accuracy of feature extraction. It can also adaptively adjust the dimensionality reduction process according to the changes and patterns of data on the premise of strengthening the intrinsic connection between features, effectively avoid information loss, maximize the discriminability of feature representation, and accurately capture the potential fault phenomenon of the equipment under the current working parameters.

[0016] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description and the drawings.

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A flowchart of a knowledge-driven equipment predictive maintenance method provided by an embodiment of the present invention; Figure 2 A flowchart of another knowledge-driven equipment predictive maintenance method provided by an embodiment of the present invention; Figure 3 A flow chart of a method for constructing a training sample set provided by an embodiment of the present invention; Figure 4 A schematic diagram of the structure of a knowledge-driven equipment predictive maintenance device provided by an embodiment of the present invention; Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the following embodiments of the present disclosure are described by specific specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.

[0021] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described in the present invention may be embodied in a wide variety of forms, and any specific structure and / or function described in the present invention is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described in the present invention may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described in the present invention may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described in the present invention may be used to implement this device and / or practice this method.

[0022] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The drawings only show components related to the present disclosure rather than being drawn according to the number, shape and size of components in actual implementation. The type, quantity and proportion of each component in actual implementation may be changed at will, and the component layout type may also be more complicated. In addition, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the aspects described can be practiced without these specific details.

[0023] The embodiment of the present invention provides a knowledge-driven equipment predictive maintenance method and device, which can process complex linear data of equipment operating parameters, accurately identify target features, and accurately capture and identify equipment faults based on current operating parameters.

[0024] For ease of understanding, a knowledge-driven predictive maintenance method for equipment provided by an embodiment of the present invention is first described in detail. Figure 1 A flowchart of a knowledge-driven equipment predictive maintenance method provided by an embodiment of the present invention is shown. Figure 1, the method comprises the following steps: Step S102, obtaining preset working parameters of industrial equipment, performing data preprocessing on the working parameters, and generating a vector to be measured.

[0025] The embodiment of the present invention collects data from industrial production equipment, including but not limited to mechanical equipment, electrical equipment, sensor equipment, etc., to perform predictive maintenance of equipment. Among them, data is collected through interfaces such as monitoring systems, sensor networks, and SCADA systems (monitoring control and data acquisition) related to the equipment. Among them, the current state of the equipment is determined by obtaining various working parameters monitored in real time by sensors on preset industrial equipment. Among them, the equipment status information monitored by the sensor can be uploaded to the central repository through the data acquisition system for data analysis. Historical data is extracted through the equipment's logs, maintenance records, or historical monitoring data, which are transmitted through the network and stored in a distributed database or cloud storage; data storage adopts a structured format, and all equipment operation data are stored in a unified format. Each piece of data consists of multiple attributes (such as temperature, pressure, vibration, etc.), and each attribute corresponds to a vector component. The data storage format is CSV format.

[0026] It should be noted that this embodiment is only used to illustrate a data format and type of the present invention. In actual applications, the attributes of the data are usually more than 10 attributes, and the number of data attributes may reach dozens or even hundreds. Specifically, it may include the extraction of time domain features and frequency domain features of the collected time series data, such as the average amplitude, amplitude variance, root mean square amplitude, peak ratio, kurtosis, skewness, maximum amplitude, minimum amplitude, main frequency, etc. of the vibration data.

[0027] Step S104: input the vector to be tested into a pre-built feature extraction model to extract target features corresponding to the vector to be tested.

[0028] The data features are extracted using a pre-trained feature extraction model, and important features that can reflect the changes in the device state are mined from the original data. The model will output a set of new feature vectors, which can more accurately describe the current state of the device. Among them, the model of the embodiment of the present invention is established based on a large number of historical samples. In the specific implementation, the model updates the network parameters based on the nonlinear dependency relationship of the features corresponding to the preset training sample set. In the training process of the traditional neural network, a symmetrical topological structure is adopted, that is, the connection mode between neurons is symmetrical. Although this structure is simple and easy to implement, it is easy to fall into a local minimum when processing complex nonlinear data, resulting in poor training effect. The embodiment of the present invention updates the network parameters based on the nonlinear dependency relationship of the features, and decomposes the data input to the neural network into multiple dual components by performing nonlinear decomposition on the data, so that the connection mode of each layer of the neural network is no longer simple and symmetrical, so that the network can mine more complex feature relationships in high-dimensional space, thereby improving the accuracy of feature extraction.

[0029] Step S106, dynamically reshape the target features, and use a preset autoencoder to reduce the dimension of the dynamically reshaped target features to generate a reduced-dimensional feature representation.

[0030] Traditional data dimensionality reduction processing maps input data to a low-dimensional space, but when processing complex and nonlinear data, it often fails to effectively extract the most representative features. The embodiment of the present invention first dynamically adjusts the input features according to the feature distribution of the input data, thereby adaptively adjusting its dimensionality reduction process according to the changes and patterns of the data, effectively avoiding information loss, and maximizing the discriminability of feature representation.

[0031] The present invention uses an autoencoder to perform data dimensionality reduction processing. The autoencoder is a neural network architecture that can reduce the dimension of the feature space by learning the compressed representation of the input data. Among them, the encoding layer of the autoencoder of the embodiment of the present invention is obtained by sparse optimization based on a nonlinear coupling optimization strategy. The nonlinear coupling optimization strategy is to deal with the interaction between multiple variables in a complex system. In one embodiment, by adopting the constraints of the coupling system, a certain nonlinear correlation can exist between each encoding neuron, thereby strengthening the intrinsic connection between the features. By optimizing the coupling degree, the neural network can better maintain the original information of the data during the dimensionality reduction process while eliminating noise interference.

[0032] Step S108, using a pre-built classifier model to classify and identify the dimensionality reduction feature representation, and determine the equipment failure phenomenon indicated by the current working parameters of the industrial equipment.

[0033] After the collected working parameters are processed as above, they can be classified and identified by a pre-built classifier model (such as a support vector machine, random forest, etc.), and whether the current working parameters of the equipment indicate a certain specific fault phenomenon can be determined based on the classification results. In one embodiment, the dimension reduction feature representation can be input into the pre-built classifier model, and the equipment fault phenomenon classification result can be output, and the classification categories are different equipment fault phenomena, such as "hydraulic pressure is too high", "knife chain is not running properly", etc.

[0034] Furthermore, the preset predictive maintenance knowledge graph system can be used to diagnose the equipment fault phenomenon classification results and determine the potential fault causes corresponding to the equipment fault phenomenon classification results. The diagnosis methods include: 1-fault cause reasoning: inferring the potential fault cause from the input fault phenomenon, for example, if the fault phenomenon is "hydraulic pressure is too high", the system may infer that "the hydraulic pump output pressure is adjusted too high"; 2-concurrent fault reasoning: analyzing the association between multiple fault phenomena, for example, "the knife chain is not running in place" may cause "machine tool automatic alarm"; 3-operation-induced fault reasoning: certain operations may directly or indirectly cause fault phenomena, for example, "starting the hydraulic system" may cause "hydraulic pressure is too high"; 4-alarm information association: the system can associate the fault phenomenon with the alarm information to help users identify the meaning of the fault alarm code. If the current knowledge base cannot provide accurate diagnosis or solutions, the system will automatically crawl the network to obtain the latest solution and add it to the knowledge base, thereby achieving self-learning and continuous optimization of the knowledge base.

[0035] Among them, the embodiment of the present invention determines the equipment failure phenomenon and possible causes by constructing an equipment predictive maintenance knowledge graph system. The user inputs equipment-related parameters (such as failure phenomenon, fault code, abnormal parameters, fault description, etc.), and the system infers the cause of the failure and repair plan through the knowledge graph, and provides real-time diagnosis.

[0036] Specifically, the equipment predictive maintenance knowledge graph system includes the following functions: 1-Fault diagnosis function: After the user provides fault information, the system uses the knowledge graph to infer and give specific fault causes and repair plans; 2-Question and answer function: The system answers users' questions about fault causes, operating errors, alarm information, etc. The four types of questions supported include: phenomena caused by a certain fault cause, errors encountered when performing a certain operation, common faults in a certain part, and the meaning of a certain fault alarm.

[0037] In one embodiment, the above-mentioned knowledge graph can be constructed based on the ontology model, and the key entities and relationships are stored in the form of triples, each of which is composed of "subject-predicate-object". Among them, the entity nodes are: 1-fault phenomenon: the specific manifestation of the equipment when it fails, which is the key entry for fault diagnosis; 2-fault cause: the root cause of the fault phenomenon; 3-operation steps: the steps performed by the user during the operation of the equipment; 4-alarm information: the alarm code or information displayed when the equipment fails; 5-fault location: the component that fails in the equipment; 6-solution: the fault repair method reported by the user. The relationship nodes are: 1-operation-induced phenomenon: a certain operation may cause a specific fault phenomenon; 2-phenomenon-induced cause: the relationship between a certain fault phenomenon and the cause behind it; 3-phenomenon concurrent relationship: the co-occurrence relationship between different fault phenomena; 4-phenomenon-location relationship: the association between the fault phenomenon and the specific part or component; 5-phenomenon-alarm information relationship: the association between the fault phenomenon and the alarm information. 6-fault cause and solution: the relationship between the fault cause and the repair measure. Among them, the above-mentioned fault phenomenon is obtained by predicting the processed data through the classifier model of the embodiment of the present invention.

[0038] Furthermore, based on the above embodiment, the embodiment of the present invention also provides another knowledge-driven predictive maintenance method for equipment. Figure 2 A flowchart of an embodiment of the present invention is shown, referring to Figure 2 , including the following steps: Step S202, obtaining preset working parameters of industrial equipment, performing data preprocessing on the working parameters, and generating a vector to be measured.

[0039] Step S204: input the vector to be tested into a pre-built feature extraction model to extract target features corresponding to the vector to be tested.

[0040] The embodiment of the present invention mainly describes the construction method of the feature extraction model. The present invention uses a 5-layer fully connected neural network for feature extraction, and uses the following construction method to solve the problem that some neural network structures in the prior art may encounter gradient vanishing, gradient explosion, or fall into a local optimal solution, affecting the stability of training and the performance of the model. Among them, the steps of updating the network parameters based on the nonlinear dependency relationship of the features corresponding to the preset training sample set to construct the feature extraction model are as follows: 1) Obtain a preset training sample set, and use the nonlinear feature dual decomposition algorithm to perform nonlinear decomposition on the training sample set to obtain multiple dual components.

[0041] Among them, the embodiment of the present invention uses a nonlinear feature dual decomposition algorithm to preprocess the data input to the neural network, and decomposes the data into multiple dual components by nonlinearly decomposing the data, providing a richer feature input for the neural network. After nonlinear mapping, the input data of each layer is passed to the next layer for further processing. Specifically, the nonlinear feature dual decomposition performs nonlinear decomposition of the input data by local decomposition, which is expressed as:

[0042] Where Ke is the number of components of the dual decomposition, is the weight coefficient of the kth component, is the transformation matrix of the corresponding component. Preferably, Ke is set to 3.

[0043] Furthermore, the nonlinear feature dual decomposition is separately supervised and trained, which is constrained by the regularization loss function to avoid overly complex feature mapping and ensure that the model can be generalized to prevent overfitting. The calculation method of the regularization loss function is expressed as:

[0044] In the formula, Representation Matrix The Frobenius norm of reflects the complexity of the model parameters; is the Frobenius norm.

[0045] 2) Input multiple dual components into the preset neural network for forward propagation, and determine the loss function corresponding to the neural network.

[0046] Suppose the data input to the neural network is , n c is the dimension of the input features, Represents the dimensional space, the weight matrix of the neural network is , m c is the number of neurons in the hidden layer of the network. In order to handle the nonlinear mapping of each layer, the initialization method of the neural network is expressed as:

[0047] In the formula, is the initial weight of the neural network; It means the mean is 0 and the variance is Normal distribution of is a normal distribution; is the variance of the neural network parameter initialization. Preferably, Set to 0.01.

[0048] The input data is linearly transformed through the weight matrix and bias term of each layer of the neural network, and then nonlinearly mapped through the activation function. The forward propagation method is expressed as:

[0049]

[0050] In the formula, is the output of the l-1th layer of the neural network, is the weight matrix of the lth layer of the neural network, is the bias term of the lth layer of the neural network, is the Sigmoid activation function; is the linear output of the lth layer of the neural network; is the nonlinear output of the lth layer of the neural network.

[0051] 3) Update the network parameters of the neural network based on the loss function and the characteristic nonlinear dependencies indicated by the dual component.

[0052] In this embodiment of the present invention, after each gradient calculation of the loss function, the weight of the network is updated according to the optimization algorithm, and the updating method is expressed as follows:

[0053]

[0054] In the formula, It is the parameter update operation; For the neural network Bias gradient of the layer; is the regularization coefficient of the neural network; is L2 regularization. Among them, . Among them, the embodiment of the present invention calculates the gradient of the loss function of the neural network through the back propagation algorithm, and updates the weight of the network according to the gradient. The traditional neural network uses the gradient descent method to update the weight. When calculating the gradient of the loss function, the present invention uses the back propagation algorithm to transfer errors to each layer of the neural network. The calculation method of the partial derivative of the loss function of the neural network with respect to the weight is expressed as:

[0055] Similarly, the calculation method of the partial derivative of the loss function of the neural network with respect to the bias is expressed as:

[0056] In the formula, is the output of the l-1th layer of the neural network; is the nonlinear output of the lth layer of the neural network; is the linear output of the lth layer of the neural network.

[0057] Furthermore, when considering the error term transmission, in addition to considering the single gradient information, the nonlinear dependency between features is also used to calculate the partial derivative of the neural network loss function with respect to the nonlinear output, which is expressed as:

[0058] In the formula, It is the partial derivative of the loss function of the neural network with respect to the final output layer (i.e., the lsth layer); is the nonlinear output of the kth layer of the neural network; is the linear output of the kth layer of the neural network.

[0059] Among them, in order to improve the training stability of the model and avoid falling into local minima when tuning the neural network parameters. When the input data is forward propagated through each layer in the neural network, the present invention adopts the gradient optimization strategy of asymmetric topological stable points, accelerates convergence and improves stability through historical gradient information, and the calculation method of the weight parameter update amount of the neural network is expressed as:

[0060]

[0061] In the formula, is the weight update amount of the lth layer of the neural network; is the weight update amount of the l-1th layer of the neural network; is the learning rate of the neural network, is the weight gradient of the lth layer of the neural network; is the weight gradient of the i-th layer of the neural network; is the loss function of the neural network; It is a dynamically adjusted parameter of the neural network, which controls the influence of the weight update of the previous layer of the neural network; is the sum of the squares of historical gradients, indicating the magnitude of gradient changes. By weighted averaging the gradients, the model can obtain a smoother update process at different levels, avoiding the problems of local minima and oscillation; ls is the total number of layers of the neural network, which is preset to 5; is a small constant to prevent division by zero. Preferably, Set to 0.001, Set to 0.2.

[0062] 4) Until the neural network meets the preset iteration conditions, a feature extraction model is built based on the neural network.

[0063] Repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0064] In summary, the present invention adopts a neural network based on nonlinear feature dual decomposition, decomposes data into multiple dual components through nonlinear decomposition, improves the ability of neural network to mine complex feature relationships in high-dimensional space, solves the problem that traditional neural networks are prone to fall into local minima resulting in poor training effects, and can improve the accuracy of feature extraction.

[0065] Furthermore, data reshaping is performed through the following steps S206-S212.

[0066] Step S206, analyzing the statistical characteristics corresponding to the target feature, and calculating the data reshaping factor corresponding to the target feature.

[0067] Step S208, using a preset autoencoder to perform preliminary dimensionality reduction processing on the target feature, determine the loss function corresponding to the autoencoder, and determine the weight matrix reshaping factor corresponding to the loss function.

[0068] Step S210, calculating a reshaping matrix based on the data reshaping factor and the weight matrix reshaping factor.

[0069] Step S212, dynamically reshape the target feature data using the reshaping matrix.

[0070] In a specific implementation, the embodiment of the present invention combines data reshaping with an autoencoder. The data reshaping operation can also be used to train the autoencoder to adjust the structure and feature dimensions of the encoding layer according to the current characteristics of the data, ensuring that the network can be optimized for a specific data set in each training cycle, avoiding the overfitting or underfitting phenomenon that may exist in the static dimensionality reduction structure of the traditional autoencoder.

[0071] Specifically, suppose the new representation of the input data after dynamic reshaping is , the reshaping process is expressed as:

[0072] In the formula, To dynamically reshape the matrix, its function is to dynamically adjust the input features according to the feature distribution of the input data; is the reshaped input data; It is a feedback signal that reflects the optimization effect of the current dimensionality reduction path.

[0073] Among them, the embodiment of the present invention uses a reshaping matrix to reshape the data. The reshaping matrix consists of two parts, which not only depends on the characteristics of the original data, but also is dynamically updated according to the current state of the network (autoencoder). That is, the data characteristics are determined by inputting the data into the autoencoder, and the state of the autoencoder and the data characteristics are comprehensively calculated to determine the corresponding reshaped data. It is expressed as:

[0074] In the formula, is a reshaping factor based on the data input to the autoencoder, which determines the main distribution of the data; is a reshaping factor based on the encoder weight matrix to adjust the feature representation after dimensionality reduction.

[0075] The reshaping factor based on the data input to the autoencoder determines the reshaping weight of each feature by analyzing the statistical characteristics of the input data (such as mean, variance), and the calculation method is expressed as:

[0076] In the formula, For input data The mean of is the first dynamic adjustment factor for the reshaped matrix; is the second dynamic adjustment factor for the reshaped matrix; It is the second-order derivative of the loss function of the autoencoder with respect to the input data, which measures the sensitivity of the input data change to the loss, thereby adjusting the reshaping factor; is the loss function of the autoencoder. Preferably, α r Set to 0.1, β r Set to 0.3.

[0077] Furthermore, the reshaping factor of the encoder-based weight matrix is ​​dynamically adjusted according to the change of the current weight through the network training process, and the calculation method is expressed as:

[0078] In the formula, is the gradient of the loss function of the autoencoder to the network weight, indicating the influence of the current weight on the reshaped matrix; is the increment of the autoencoder, indicating the rate at which the weight changes during training; is the third dynamic adjustment factor for the reshaping matrix; A fourth dynamic adjustment factor for the reshaped matrix.

[0079] Furthermore, the above feedback signal reflects the optimization effect of the current dimensionality reduction path. When the model is overfitted or underfitted on certain samples, the network will automatically reconfigure its dimensionality reduction structure by introducing a nonlinear coupling feedback mechanism. The configuration process can be determined based on the training of the autoencoder. By adjusting the parameters to adapt to the actual distribution of the data, it not only makes the training process more flexible, but also effectively avoids the problems of oversimplification or information loss in traditional dimensionality reduction algorithms. Specifically, the feedback signal dynamically adjusts the training path (dimensionality reduction path) according to the weight matrix of the next iteration of the autoencoder. The calculation method is expressed as:

[0080] In the formula, Gradient of the loss function with respect to the reshaped matrix; is a feedback adjustment factor used to balance the contribution of gradient update and feedback signal; is the feedback signal; is the weight of the feedback signal. Preferably, Set to 1, Set to 0.05.

[0081] In summary, based on the above-mentioned dynamic reshaping operation, the autoencoder can perform data dimensionality reduction by analyzing the distribution characteristics and complexity of the input data. It can solve the overfitting or underfitting phenomenon that may exist in the static dimensionality reduction structure of the traditional autoencoder, effectively avoid information loss, and maximize the discriminability of feature representation.

[0082] Step S214, using a preset autoencoder to perform dimensionality reduction processing on the dynamically reshaped target features to generate a reduced-dimensional feature representation.

[0083] Furthermore, the steps of sparsely optimizing the encoding layer of the autoencoder based on the nonlinear coupling optimization strategy to train the autoencoder are described: 1) The nonlinear correlation between the features of the preset training sample set is determined using the preset coupling constraint function.

[0084] The nonlinear coupling constraint is to optimize the nonlinear correlation between feature representations. A complex coupling relationship is constructed through high-order interaction terms. The calculation method of the coupling constraint function is expressed as follows:

[0085] In the formula, is the i-th element of the low-dimensional feature representation, is the jth element of the low-dimensional feature representation; Kas is the dimension of the low-dimensional feature representation; Features and The coupling weight between them characterizes the nonlinear dependence between them; is the bandwidth parameter in the Gaussian kernel function, which controls the range of interaction between features. Preferably, Set to 0.1.

[0086] 2) Based on the nonlinear association and the preset feature sparsity constraint, determine the loss function of the preset autoencoder.

[0087] The training process first initializes the parameters of the autoencoder. In one embodiment, the parameters of the autoencoder are initialized randomly and obey a normal distribution with a mean of 0 and a variance of the unit matrix. , N is the number of samples after expansion, and D is the dimension of the data features input to the autoencoder.

[0088] The autoencoder network consists of two parts: the encoder and the decoder. The encoder maps the input data to a low-dimensional feature space, and the decoder remaps the low-dimensional features back to the original space. The mapping method of the encoder is expressed as:

[0089] In the formula, is the mapping function of the encoder, is the weight matrix of the encoder, Z r is a low-dimensional feature representation. Preferably, the encoder mapping function uses a ReLU activation function.

[0090] The loss function calculation method of the autoencoder is expressed as:

[0091] In the formula, is the loss function of the autoencoder. is the L2 norm; Characterize the reconstruction error, which measures the difference between the feature map after dimensionality reduction and the original data; is the mapping function of the decoder; is the L1 norm.

[0092] The sparsity constraints of the representation features make the feature representation as sparse as possible and reduce redundancy; is a coupling constraint function that aims to maximize the nonlinear correlation between features. is the weight coefficient of the reconstruction error, is the weight coefficient of the sparsity constraint, is the weight coefficient of the coupling constraint. Preferably, Set to 0.3, Set to 0.3.

[0093] 3) Based on the loss function and the preset momentum factor, the weight matrix of the autoencoder is back-propagated and updated to optimize the encoding layer of the autoencoder.

[0094] The autoencoder calculates the reconstruction error through forward propagation and updates the parameters through back propagation. Different from the traditional autoencoder, the present invention dynamically adjusts the optimization strategy in each round of iteration to ensure that the network can automatically select the optimal dimensionality reduction path according to the actual performance of the data. The weight update method of the autoencoder is expressed as:

[0095] In the formula, is the weight matrix of the autoencoder at the tth iteration; is the weight matrix of the autoencoder at the t+1th iteration; is the learning rate of the autoencoder; is the gradient of the loss function of the autoencoder with respect to the weight matrix; is the momentum factor of the autoencoder, which is used to control the smoothness of the gradient update; is the second-order gradient of the loss function with respect to the weight matrix. The second-order gradient term can converge more efficiently and avoid the oscillation or local minimum problem that may occur in traditional gradient descent. Preferably, Set to 0.01, Set to 0.3. In summary, the coding layer sparse optimization combined with the nonlinear coupling optimization strategy is achieved.

[0096] The weight coefficient of the reconstruction error in the embodiment of the present invention adopts an adaptive adjustment mechanism, and dynamically adjusts the weight of the reconstruction error during the training process, so that the contribution of different training stages can be balanced, thereby effectively avoiding overfitting or underfitting. The adjustment method is expressed as:

[0097] In the formula, is the variance of the loss function, indicating the volatility of the current loss function between different training cycles. Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is to reach the preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0098] Step S216, using a pre-built classifier model to classify and identify the dimensionality reduction feature representation, and determine the equipment failure phenomenon indicated by the current working parameters of the industrial equipment.

[0099] The present invention adopts a support vector machine algorithm to classify the data after dimensionality reduction. Different from the traditional support vector machine algorithm, the embodiment of the present invention introduces non-convex optimization to solve complex optimization problems to build a classifier model. By optimizing when the objective function is not a convex function, the problem of difficulty in finding a global optimal solution when there are multiple local extreme points is solved. Among them, the classification algorithm in the prior art cannot effectively apply non-convex optimization. Even if an extreme learning machine is used to map the input data to a nonlinear high-dimensional space, the algorithm may still be attracted by "shallow" extreme points during the optimization process. The non-convexity of the objective function still exists. The complex shape of the non-convex objective function will not disappear due to the transformation of the feature space. There may still be multiple local optimal solutions, and the true global optimal solution cannot be found, resulting in an insufficiently optimal decision boundary and limited classification accuracy. To solve the above problems, the present invention adopts a search mechanism based on chaos theory to explore more possible solution spaces in the non-convex optimization process, thereby finding a better decision boundary and improving the classification accuracy. The construction method of the classifier model of the embodiment of the present invention is as follows: 1) Input the preset training sample set into the preset support vector machine, and use the chaos theory algorithm to adjust the support vector machine parameters of the support vector machine.

[0100] Among them, the embodiment of the present invention uses chaos theory to generate random disturbances; and uses a preset non-convex optimization algorithm to adjust the support vector machine parameters of the support vector machine based on the random disturbances.

[0101] First, using the initial value sensitivity of chaos theory, the search process is started from multiple different initial points to ensure that different areas of the solution space are covered and to avoid the algorithm falling into the local optimal solution, which can be expressed as:

[0102]

[0103] In the formula, is the value of the chaotic variable at the tth iteration; is the chaotic variable value at the t+1th iteration; is the parameter of the chaotic map. Preferably, a Values ​​that show chaotic behavior, such as μ u =4; is the weight vector of the support vector machine at the tth iteration; is the initial weight vector of the support vector machine; is the chaotic learning rate. Preferably, Set to 0.01.

[0104] Furthermore, the role of random perturbations generated based on chaos theory is to increase the diversity of the search process. Chaos theory is used to increase the randomness and exploration of the optimization process, thereby avoiding falling into a local optimal solution. The calculation method is expressed as:

[0105] In the formula, is a coefficient that controls the magnitude of the disturbance effect. Preferably, Set to 0.001.

[0106] On each search path, a non-convex optimization algorithm is executed to adjust the decision boundary of the support vector machine. The optimization direction and step size are adjusted according to the classification error and the nonlinearity of the kernel function output. The non-convex optimization method of adjusting the support vector machine parameters through the gradient descent method is expressed as:

[0107]

[0108] In the formula, is the weight vector of the support vector machine at the t+1th iteration; is the step size for non-convex optimization, is the gradient of the objective function of the support vector machine with respect to the weights, is the label of the ith data sample, is the feature vector of the i-th data sample; represents a random disturbance generated based on chaos theory. Preferably, Set to 0.01.

[0109] The embodiment of the present invention also initializes the parameters of the support vector machine for training, including selecting a suitable kernel function type and initializing regularization parameters, which are expressed as:

[0110]

[0111] In the formula, is the radial basis function kernel, which is used to calculate two data samples and The kernel function value between ; is the first data sample input to the support vector machine; is a second data sample input to the support vector machine; is the width parameter of the kernel function; L u is the objective function of the support vector machine, w u is the weight vector of the support vector machine; represents the L2 norm; is a slack variable used to handle the situation of incomplete separability; is the regularization parameter, which controls the penalty intensity of the error term; N is the number of samples in the training sample set.

[0112] 2) Determine the Lagrange multipliers corresponding to the training sample set, and adjust the parameters of the kernel function of the support vector machine based on the Lagrange multipliers.

[0113] Among them, the parameters of the kernel function can be dynamically adjusted based on the data characteristics obtained by the above feature dimensionality reduction model to better adapt to the distribution characteristics of the data. The dynamic adjustment of the width parameter of the kernel function is achieved by minimizing the following error, which is expressed as:

[0114]

[0115] In the formula, is the width parameter of the kernel function for the t+1th iteration; is the width parameter of the kernel function of the tth iteration; is the kernel function width parameter adjustment rate, is the Lagrange multiplier of the ith data sample; is the Lagrange multiplier of the jth data sample; E total is the objective error function of the support vector machine; is the label of the i-th data sample; is the label of the jth data sample.

[0116] 3) Until the support vector machine meets the preset training conditions, a classifier model is built based on the support vector machine.

[0117] Repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0118] The embodiment of the present invention adopts a support vector machine algorithm with a search mechanism based on chaos theory for classification, explores more possible solution spaces in a non-convex optimization process, finds a better decision boundary, and solves the problem that the traditional support vector machine algorithm may fall into a local optimal solution in a non-convex optimization process.

[0119] Furthermore, the feature extraction model, autoencoder, and classifier model of the embodiment of the present invention are trained respectively by corresponding training sample sets, wherein the samples after feature extraction can be used as the training sample set of the autoencoder, and the data after dimensionality reduction processing can be further used as the training sample set of the classifier model. In actual industrial environments, the acquisition of high-quality fault data is often accompanied by high costs, such as the need to install a large number of sensors and accumulate data for a long time. In addition, due to the low frequency of equipment failures under normal operating conditions, the fault samples used to train the model are relatively scarce. This not only limits the generalization ability of the model, but also makes it difficult for the model accuracy to meet actual needs, and also leads to the problem of imbalanced training data. The collection, acquisition, labeling, and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor generalization ability of the model, while affecting the accuracy of the model; in order to solve this problem, the present invention uses a data expansion algorithm based on an adaptive coupling field to expand the data to construct a training sample set. The generation process not only relies on random noise, but also combines the context of equipment status information, and can more accurately generate samples with different fault categories, thereby improving the diversity and coverage of generated samples. Among them, Figure 3 The flowchart of the method for constructing the training sample set is shown in FIG. Figure 3 , the method comprises the following steps: Step S10, obtaining pre-stored working parameters of preset industrial production equipment, marking the working parameters, and constructing an initial sample set.

[0120] The working parameters can refer to the above embodiments and will not be described in detail here. Further, the collected data is labeled. In one embodiment, manual labeling can be used, and the labeled categories include different equipment failure phenomena, such as "hydraulic pressure is too high" and "knife chain is not running properly".

[0121] Step S20, generating random noise corresponding to the initial sample set, and obtaining current operating data of the industrial production equipment.

[0122] Step S30, determining the coupling input corresponding to the industrial production equipment based on the random noise and the current operation data.

[0123] The embodiment of the present invention expands the initial sample set data to construct a training sample set, wherein the input of the expansion algorithm of the embodiment of the present invention is not only a random noise vector, but also adjusted in combination with the state information of the device. During each training, the current operating data of the device (such as vibration, temperature, etc.) is used as the coupling field condition to adjust the input noise, so that the data expansion algorithm not only generates the form of the data, but also generates virtual samples with specific attributes (such as fault type, fault degree, etc.) according to the specific device fault state. The input of the data expansion algorithm is expressed as:

[0124] In the formula, For the adaptive coupled field driving function, a coupled input is generated based on a weighted combination of device state information and noise to generate a new input feature; is the input feature of the generator combined with the adaptive coupled field; Z c is a random noise vector; S c is the device status information (including vibration, temperature and other device parameters); α c is the first adaptive adjustment coefficient, β c is the second adaptive adjustment coefficient. The first adaptive adjustment coefficient and the second adaptive adjustment coefficient are used to adjust the weights of noise and device status information.

[0125] The first adaptive adjustment coefficient and the second adaptive adjustment coefficient follow the adaptive optimization mechanism of error back propagation, and the update rule is:

[0126]

[0127] In the formula, The loss function used to characterize the preset data augmentation algorithm; is the learning rate of the adjustment coefficient, which indicates the step size of the adjustment coefficient update; is the first adaptive adjustment coefficient of the t+1th iteration; is the second adaptive adjustment coefficient of the t+1th iteration; is the first adaptive adjustment coefficient of the t+1th iteration; is the first adaptive adjustment coefficient of the t+1th iteration; is the symbol of partial derivative.

[0128] Step S40: perform data expansion on the initial sample set based on the coupled input to construct a training sample set.

[0129] Furthermore, data expansion is performed on the above-mentioned coupled input to construct a training sample set of an embodiment of the present invention, which can improve sample diversity and coverage, solve the problem of imbalanced training data caused by high cost of actual equipment fault data collection and scarcity of fault states, as well as the problem of poor model generalization ability and accuracy impact caused by insufficient training samples.

[0130] In one embodiment, the data augmentation algorithm can use a generative adversarial network. A loss function that can be used to characterize the generator of a generative adversarial network. Corresponding to the generative adversarial network, after obtaining the above-mentioned coupled input, the coupled input is fed into the pre-trained generative adversarial network, and the generative adversarial network is used to generate an expanded sample corresponding to the coupled input; the expanded sample is merged with the initial sample set to construct a training sample set. The generative adversarial network updates the generator based on exponential calculation of the output of the discriminator.

[0131] In the training process of the generative adversarial network, the parameters of the generative adversarial network are first initialized, and the generator is set to G c , the discriminator is D c , the weight of the generator is , the weight of the discriminator is , the bias of the generator is , the bias of the discriminator is , the parameters of the generator and the discriminator are initialized to values ​​on the normal distribution, and the normal distribution is a distribution with a mean of 0 and a variance of the unit matrix.

[0132] Among them, the goal of the discriminator is to identify whether the input data is real data. Specifically, the discriminator accepts two types of data: one is the real fault data collected from the equipment, and the other is the false data generated by the generator; the discriminator optimizes its parameters so that it can accurately distinguish these two types of data. The loss function of the discriminator is expressed as:

[0133] In the formula, preal represents the distribution of real data, Xc is the real data, The data generated by the generator, is the discriminator function; To obey a specific distribution; Express expectations.

[0134] After the discriminator is updated, the generator is optimized. The goal of the generator is to deceive the discriminator so that the discriminator cannot accurately determine which data is generated. Moreover, the optimization goal of the generator is to maximize the probability that the discriminator determines that the generated data is real data. The loss function of the generator is expressed as:

[0135] Where P noise represents the distribution of noise; P state Represents the distribution of device states; is a generator function.

[0136] Different from the prior art, in the back propagation of the generator, the updating rule of the embodiment of the present invention is based on the feedback information of the discriminator, so that the weight update of the generator has higher flexibility and avoids the gradient disappearance or oscillation problems that may occur in traditional methods. The updating method of the generator weight is expressed as:

[0137] In the formula, is the learning rate of the generative adversarial network, is the rate constant; is the weight of the generator at the tth iteration training; is the weight of the generator during the t+1th iteration training. is an exponential function.

[0138] Among them, the weight update of the conventional generator is usually linear. When the quality of the samples generated by the generator is very high, the output of the discriminator is close to 1, resulting in a gradient close to 0, and the gradient disappearance problem occurs. The present invention adopts an exponential function to make the weight update have nonlinear characteristics and be more flexible. The generator dynamically adjusts the weight according to the feedback of the discriminator. When the quality of the samples generated by the generator is high (close to the "real"), the weight update amplitude is small; when the quality of the samples generated by the generator is low (far away from the "real"), the weight update amplitude is large. Among them, through the adjustment of the exponential function, even when the discriminator's discrimination probability is close to 1, a certain gradient can be maintained, ensuring that the weight update is smooth and bounded, avoiding drastic weight changes, avoiding gradient disappearance, and helping to improve the stability of training. Therefore, the weight update rule of the present invention is smoother and more flexible, can better cope with fluctuations in the training process, and improve the training stability of the entire generative adversarial network.

[0139] Further, the above steps are repeated until the preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is to reach the preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times. After the data expansion model training is completed, the number of samples is increased using the trained data expansion model. In one embodiment, assuming that the original collected samples are 800, and the data expansion model is expanded to generate 200 samples, the expanded data set contains 1000 samples.

[0140] Furthermore, based on the above method embodiment, the embodiment of the present invention also provides a knowledge-driven equipment predictive maintenance device, Figure 4 The schematic diagram of the structure corresponding to the embodiment of the present invention is shown. Figure 4, the device includes: a data acquisition module 100, used to obtain the preset working parameters of industrial equipment, perform data preprocessing on the working parameters, and generate a vector to be tested; a feature extraction module 200, used to input the vector to be tested into a pre-built feature extraction model, and extract the target feature corresponding to the vector to be tested; wherein the feature extraction model updates the network parameters based on the nonlinear dependency relationship of the features corresponding to the preset training sample set; a data processing module 300, used to perform data dynamic reshaping operation on the target features, and use a preset autoencoder to reduce the dimension of the dynamically reshaped target features to generate a reduced-dimensional feature representation; wherein the encoding layer of the autoencoder is obtained by sparse optimization based on a nonlinear coupling optimization strategy; an execution module 400, used to classify and identify the reduced-dimensional feature representation using a pre-built classifier model, and determine the equipment fault phenomenon indicated by the current working parameters of the industrial equipment. The knowledge-driven equipment predictive maintenance device provided in the embodiment of the present invention has the same technical features as the knowledge-driven equipment predictive maintenance method provided in the above embodiment, so it can also solve the same technical problems and achieve the same technical effects.

[0141] The execution module 400 is also used to input the dimension reduction feature representation into the pre-built classifier model, and output the equipment fault phenomenon classification result; use the preset predictive maintenance knowledge graph system to diagnose the equipment fault phenomenon classification result, and determine the potential fault cause corresponding to the equipment fault phenomenon classification result. The data processing module 300 is also used to analyze the statistical characteristics corresponding to the target feature, calculate the data reshaping factor corresponding to the target feature; use the preset autoencoder to perform preliminary dimension reduction processing on the target feature, determine the loss function corresponding to the autoencoder, and determine the weight matrix reshaping factor corresponding to the loss function; calculate the reshaping matrix based on the data reshaping factor and the weight matrix reshaping factor; and use the reshaping matrix to dynamically reshape the data of the target feature. The data processing module 300 is also used to determine the nonlinear association between the features of the preset training sample set using the preset coupling degree constraint function; determine the preset loss function of the autoencoder based on the nonlinear association and the preset feature sparsity constraint; and back-propagate and update the weight matrix of the autoencoder based on the loss function and the preset momentum factor to optimize the encoding layer of the autoencoder.

[0142] The execution module 400 is also used to input the preset training sample set into the preset support vector machine, and use the chaos theory algorithm to adjust the support vector machine parameters of the support vector machine; determine the Lagrange multiplier corresponding to the training sample set, and adjust the parameters of the kernel function of the support vector machine based on the Lagrange multiplier; until the support vector machine meets the preset training conditions, a classifier model is constructed based on the support vector machine. The execution module 400 is also used to generate random perturbations using chaos theory; and use the preset non-convex optimization algorithm to adjust the support vector machine parameters of the support vector machine based on random perturbations. Further, the feature extraction module 200 is also used to obtain a preset training sample set, and use the nonlinear feature dual decomposition algorithm to perform nonlinear decomposition on the training sample set to obtain multiple dual components; input the multiple dual components into a preset neural network for forward propagation, and determine the loss function corresponding to the neural network; based on the loss function and the characteristic nonlinear dependency relationship indicated by the dual component, the network parameters of the neural network are updated; until the neural network meets the preset iteration conditions, a feature extraction model is constructed based on the neural network.

[0143] The device also includes a construction module, which is used to obtain the pre-stored working parameters of the preset industrial production equipment, annotate the working parameters, and construct an initial sample set; generate random noise corresponding to the initial sample set, and obtain the current operating data of the industrial production equipment; determine the coupling input corresponding to the industrial production equipment based on the random noise and the current operating data; perform data expansion on the initial sample set based on the coupling input to construct a training sample set. The above-mentioned construction module is also used to input the coupling input into a pre-trained generative adversarial network, use the generative adversarial network to generate expanded samples corresponding to the coupling input; merge the expanded samples with the initial sample set to construct a training sample set; wherein the generative adversarial network is based on exponential calculation of the output of the discriminator to update the generator.

[0144] An embodiment of 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 executes the computer program to implement the above Figures 1 to 3 The embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to perform the above Figures 1 to 3 The embodiment of the present invention also provides a structural diagram of an electronic device, such as Figure 5 FIG. 5 is a schematic diagram of the structure of the electronic device, wherein the electronic device includes a processor 51 and a memory 50, the memory 50 stores computer executable instructions that can be executed by the processor 51, and the processor 51 executes the computer executable instructions to implement the above Figures 1 to 3 The method shown. Figure 5 In the illustrated embodiment, the electronic device further includes a bus 52 and a communication interface 53, wherein the processor 51, the communication interface 53 and the memory 50 are connected via the bus 52. The memory 50 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 53 (which may be wired or wireless), and the Internet, a wide area network, a local area network, a metropolitan area network, etc. may be used. The bus 52 may be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. It may also be an AMBA (Advanced Microcontroller Bus Architecture) bus, where AMBA defines three types of buses, including an APB (Advanced Peripheral Bus) bus, an AHB (Advanced High-performance Bus) bus, and an AXI (Advanced Xtensible Interface) bus. The bus 52 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0145] The processor 51 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 51. The above processor 51 can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to execute, or the hardware and software modules in the decoding processor are combined and executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor 51 reads the information in the memory and combines its hardware to complete the above Figures 1 to 3 Any of the methods shown.

[0146] A computer program product of a knowledge-driven equipment predictive maintenance method and apparatus provided in an embodiment of the present invention includes a computer-readable storage medium storing a program code, and the instructions included in the program code can be used to execute the method described in the previous method embodiment. The specific implementation can refer to the method embodiment, which will not be repeated here. A person skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the system described above can refer to the corresponding process in the above method embodiment, which will not be repeated here. In addition, in the description of the embodiment of the present invention, unless otherwise clearly specified and limited, the terms "installation", "connection" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood in specific circumstances. If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present invention, in essence, 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, which is stored in a storage medium and includes several instructions for 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: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, and other media that can store program codes. In the description of the present invention, it should be noted that the orientation or position relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. is based on the orientation or position relationship shown in the accompanying drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.Finally, it should be noted that the above embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above embodiments, those skilled in the art should understand that any person skilled in the art can still modify the technical solutions recorded in the above embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A knowledge-driven equipment predictive maintenance method, characterized in that: The method comprises: Obtaining preset working parameters of industrial equipment, performing data preprocessing on the working parameters, and generating a vector to be measured; Input the vector to be tested into a pre-built feature extraction model to extract the target feature corresponding to the vector to be tested; wherein the feature extraction model updates the network parameters based on the feature non-linear dependency relationship corresponding to the preset training sample set; Performing a data dynamic reshaping operation on the target feature, and using a preset autoencoder to perform dimensionality reduction processing on the dynamically reshaped target feature to generate a reduced-dimensional feature representation; wherein the encoding layer of the autoencoder is obtained by sparse optimization based on a nonlinear coupling optimization strategy; The reduced-dimensional feature representation is classified and identified using a pre-built classifier model to determine the equipment failure phenomenon indicated by the current working parameters of the industrial equipment.

2. The method according to claim 1, characterized in that The step of classifying and identifying the dimension reduction feature representation using a pre-built classifier model to determine the equipment fault phenomenon indicated by the current working parameters of the industrial equipment includes: Inputting the dimension-reduced feature representation into a pre-built classifier model, and outputting a classification result of the equipment fault phenomenon; The preset predictive maintenance knowledge graph system is used to diagnose the equipment failure phenomenon classification results to determine the potential failure causes corresponding to the equipment failure phenomenon classification results.

3. The method according to claim 1, characterized in that The step of dynamically reshaping the target feature includes: Analyze the statistical characteristics corresponding to the target feature, and calculate the data reshaping factor corresponding to the target feature; Performing preliminary dimensionality reduction processing on the target feature using a preset autoencoder, determining a loss function corresponding to the autoencoder, and determining a weight matrix reshaping factor corresponding to the loss function; Calculate a reshaping matrix based on the data reshaping factor and the weight matrix reshaping factor; The target feature is dynamically reshaped using the reshaping matrix.

4. The method according to claim 1, characterized in that: The steps of sparsely optimizing the encoding layer of the autoencoder based on the nonlinear coupling optimization strategy include: Determine the nonlinear correlation between the features of the preset training sample set using a preset coupling degree constraint function; Determining a preset loss function of the autoencoder based on the nonlinear association and the preset feature sparsity constraint; Based on the loss function and the preset momentum factor, the weight matrix of the autoencoder is back-propagated and updated to optimize the encoding layer of the autoencoder.

5. The method according to claim 1, characterized in that The method for constructing the classifier model comprises: Inputting a preset training sample set into a preset support vector machine, and adjusting the support vector machine parameters of the support vector machine using a chaos theory algorithm; Determining the Lagrangian multiplier corresponding to the training sample set, and adjusting the parameters of the kernel function of the support vector machine based on the Lagrangian multiplier; Until the support vector machine meets the preset training conditions, a classifier model is constructed based on the support vector machine.

6. The method according to claim 5, characterized in that The step of adjusting the support vector machine parameters of the support vector machine by using a chaos theory algorithm comprises: Generate random disturbances using chaos theory; A preset non-convex optimization algorithm is used to adjust support vector machine parameters of the support vector machine based on the random disturbance.

7. The method according to claim 1, characterized in that The method for constructing the feature extraction model comprises: Obtaining a preset training sample set, and performing nonlinear decomposition on the training sample set using a nonlinear feature dual decomposition algorithm to obtain a plurality of dual components; Inputting the plurality of dual components into a preset neural network for forward propagation, and determining a loss function corresponding to the neural network; Based on the loss function and the characteristic nonlinear dependency indicated by the dual component, updating the network parameters of the neural network; Until the neural network meets the preset iteration condition, a feature extraction model is constructed based on the neural network.

8. The method according to any one of claims 1 to 7, characterized in that: The method for constructing a preset training sample set includes: Obtaining pre-stored working parameters of preset industrial production equipment, marking the working parameters, and constructing an initial sample set; Generate random noise corresponding to the initial sample set, and obtain current operating data of the industrial production equipment; Determining a coupling input corresponding to the industrial production equipment based on the random noise and the current operating data; The initial sample set is expanded based on the coupled input to construct a training sample set.

9. The method according to claim 8, characterized in that The step of performing data expansion on the initial sample set based on the coupled input to construct a training sample set includes: Inputting the coupled input into a pre-trained generative adversarial network, and using the generative adversarial network to generate an expanded sample corresponding to the coupled input; Merging the expanded sample with the initial sample set to construct a training sample set; The generative adversarial network updates the generator based on exponential calculation of the output of the discriminator.

10. A knowledge-driven predictive maintenance device for equipment, characterized in that: The device comprises: A data acquisition module, used to acquire the working parameters of preset industrial equipment, perform data preprocessing on the working parameters, and generate a vector to be measured; A feature extraction module, used to input the vector to be tested into a pre-built feature extraction model to extract the target feature corresponding to the vector to be tested; wherein the feature extraction model updates the network parameters based on the feature non-linear dependency relationship corresponding to the preset training sample set; A data processing module is used to perform a data dynamic reshaping operation on the target feature, and use a preset autoencoder to reduce the dimension of the dynamically reshaped target feature to generate a reduced-dimensional feature representation; wherein the encoding layer of the autoencoder is obtained by sparse optimization based on a nonlinear coupling optimization strategy; The execution module is used to classify and identify the dimension reduction feature representation using a pre-built classifier model to determine the equipment failure phenomenon indicated by the current working parameters of the industrial equipment.

Images