Equipment fault prediction method and device, storage medium and electronic equipment

Abstract

The invention provides an equipment fault prediction method and device, a storage medium and electronic equipment. The method comprises the following steps: acquiring a text feature vector for performing problem description on an abnormal condition of equipment; matching the text feature vector from a preset question and answer feature vector database to determine whether the question description is the question description related to the equipment fault; if the problem description is determined to be the problem description related to the fault, respectively obtaining a system fault prediction result and a system solution, a networking fault prediction result and a networking solution, and a map prediction fault result and a map solution of the question and answer knowledge base; and obtaining a system fault prediction result and a system solution, a networking fault prediction result and a networking solution, and a map prediction fault result and a map solution of the question and answer knowledge base by using the hybrid expert model, and obtaining an optimal fault prediction result and a corresponding optimal solution. Therefore, the fault prediction result and the fault solution are efficiently and accurately determined, and the operation and maintenance efficiency of the equipment is improved.

Description

Technical Field

[0001] The present application relates to the technical field of large language models, and in particular to a method, device, storage medium and electronic device for predicting equipment failure. Background Art

[0002] In the daily operation and maintenance of certificate-making equipment, fault diagnosis is the core task to ensure system stability and business continuity. Operation and maintenance work has always been a complex and arduous task in the IT industry. As the scale and complexity of the system increase, the challenges faced by operation and maintenance personnel are also increasing. In operation and maintenance work, logs are the main basis for troubleshooting. However, as the complexity of the system increases, the amount of logs increases dramatically, and how to quickly find the fault point has become a major challenge. Traditional log analysis methods often rely on manual screening, which is not only time-consuming but also easy to miss important information. Even if the fault can be located, the complexity of the log information is often difficult to understand. Operation and maintenance personnel need to interpret a large number of technical terms and error messages, which is not only time-consuming but also easy to lead to misinterpretation. The solution to equipment operation failure is generally in the enterprise knowledge base and needs to be synchronized to operation and maintenance engineers in various provinces and cities on a regular basis, resulting in untimely knowledge synchronization. Therefore, how to provide a method that can quickly detect equipment failures or anomalies, efficiently and accurately assist operation and maintenance personnel to learn about equipment problems and provide targeted solutions, and facilitate operation and maintenance personnel to improve operation and maintenance efficiency has become a technical problem that needs to be solved in the industry. Summary of the invention

[0003] In view of this, embodiments of the present application provide a method, apparatus, storage medium and electronic device for predicting equipment failure, so as to at least or partially solve the above problems.

[0004] In a first aspect, an embodiment of the present application provides a method for predicting equipment failure, comprising:

[0005] Obtain a text feature vector describing the abnormal condition of the device;

[0006] Matching the text feature vector from a preset question-answer feature vector database to determine whether the problem description is a problem description related to the equipment failure;

[0007] If it is determined to be a fault-related problem description, the text features are matched in the fault knowledge question and answer database using a search enhancement method to obtain system fault prediction results and system solutions from the question and answer knowledge database; based on the text feature vector, network fault prediction results and network solutions are obtained through network query; based on the text feature vector, combined with the knowledge graph constructed for the equipment fault, the graph prediction fault results and graph solutions are obtained;

[0008] By using a hybrid expert model, based on the system fault prediction results and system solutions, the networking fault prediction results and networking solutions, and the graph prediction fault results and graph solutions, the optimal fault prediction results and the corresponding optimal solutions are obtained.

[0009] Optionally, in an embodiment of the present application, the method further comprises: obtaining a text feature vector describing a problem of an abnormal condition of the device;

[0010] Matching the text feature vector from a preset question-answer feature vector database to determine whether the problem description is a problem description related to the equipment failure;

[0011] If it is determined to be a fault-related problem description, the text features are matched in the fault knowledge question and answer database using a search enhancement method to obtain system fault prediction results and system solutions from the question and answer knowledge database; based on the text feature vector, network fault prediction results and network solutions are obtained through network query; based on the text feature vector, combined with the knowledge graph constructed for the equipment fault, the graph prediction fault results and graph solutions are obtained;

[0012] By using a hybrid expert model, based on the system fault prediction results and system solutions, the networking fault prediction results and networking solutions, and the graph prediction fault results and graph solutions, the optimal fault prediction results and the corresponding optimal solutions are obtained.

[0013] Optionally, in an embodiment of the present application, the method of using retrieval enhancement to perform feature matching on the text features in the fault knowledge question and answer base to obtain the system fault prediction results and system solutions of the question and answer knowledge base includes:

[0014] Determine the fault knowledge question and answer base, wherein the fault knowledge question and answer base includes at least one of a local fault knowledge question and answer base and an external fault knowledge question and answer base;

[0015] Using an efficient retrieval algorithm to generate a query request according to the text features to select relevant documents from the selected fault knowledge question and answer library;

[0016] Extracting information from the relevant documents to extract key information of the relevant documents, wherein the key information includes one or more of entity information, entity relationship information, and context information;

[0017] Using a large language model, the key information is integrated to generate fault prediction information and predicted fault resolution information;

[0018] The fault prediction information and the predicted fault solution information are subjected to at least one text input processing of hallucination suppression, noise information processing, document screening and error correction to generate a system fault prediction result and a system solution of a question-and-answer knowledge base.

[0019] Optionally, in an embodiment of the present application, before acquiring the graph prediction fault result and graph solution based on the text feature vector and the knowledge graph constructed for the equipment fault, the method further includes: constructing a knowledge graph for the equipment fault;

[0020] Determine the equipment operation and maintenance information consisting of the fault description, solution and fault cause of the equipment from the open world data;

[0021] Converting the operation and maintenance information into structured semantic block data in a predefined format;

[0022] Performing incremental entity extraction and incremental relationship extraction on the structured semantic block data to obtain unique entity information in each semantic block and determine relationship information between the entities;

[0023] Integrate the acquired unique entity information and the relationship information into a preset graph database, and combine the device-related knowledge graph subgraph retrieved from the existing knowledge graph to generate a specific fault knowledge graph and a specific device knowledge graph, wherein the specific device knowledge graph contains sequential data during the operation of the specific device;

[0024] Perform node and edge clustering on the generated specific fault knowledge graph to create a fault aggregation knowledge graph;

[0025] The relevant specific fault graph and the time information about the sequential data in the specific knowledge graph are combined to generate a personalized knowledge graph for the device, and the personalized knowledge graph is determined as the knowledge graph for the device fault.

[0026] Optionally, in an embodiment of the present application, the text feature vector is combined with a knowledge graph constructed for the device failure to obtain the graph prediction failure result and graph solution, including: according to the text feature vector, dual attention enhancement and graph neural network are used to predict the failure of the device based on the personalized knowledge graph to obtain the graph prediction failure result and graph solution.

[0027] Optionally, in an embodiment of the present application, the method further comprises: periodically acquiring a device operation knowledge base;

[0028] The trained large language model is used to extract key information in the equipment operation knowledge base, where the key information includes at least one of equipment problem description, fault cause description, and fault solution.

[0029] According to the extracted key information, the knowledge graph or fault knowledge base of the equipment fault is updated.

[0030] Optionally, in an embodiment of the present application, after obtaining the optimal fault prediction result and the corresponding optimal solution, the method further includes:

[0031] According to the optimal fault prediction result, performing abnormality detection on the device;

[0032] According to the result of the abnormality detection, an early warning prompt is given to the device.

[0033] Optionally, in an embodiment of the present application, performing abnormality detection on the device according to the optimal fault prediction result includes:

[0034] creating a digital twin simulated device for the device;

[0035] Performing data fusion based on simulated operation data obtained through working state simulation of the simulated device, actual monitoring data obtained through monitoring the working state of the device, and statistical data obtained through statistics on the working states of devices of the same type, to obtain fused data;

[0036] Constructing a neural network model for predicting health indicators and remaining life of the equipment for the equipment according to the fused data;

[0037] Using the health indicators and the equipment remaining life prediction neural network model, combined with the optimal fault prediction result, the digital twin simulation equipment is performance evaluated to determine the abnormality detection result through the result of the performance evaluation.

[0038] Optionally, in an embodiment of the present application, the step of creating a digital twin simulation device for the device on the server includes:

[0039] Collecting multi-source data of the device;

[0040] Based on the multi-source data, a digital twin simulation device of the device is created using a preset geometric structure model, a data analysis model and a physical model.

[0041] In a second aspect, based on the device failure prediction method described in the first aspect of the present application, an embodiment of the present application further provides a device for predicting device failure, including:

[0042] An extraction module is used to obtain a text feature vector describing the abnormal condition of the equipment;

[0043] A matching module, used to match the text feature vector from a preset question-answer feature vector database to determine whether the problem description is a problem description related to the equipment failure;

[0044] A retrieval module is used to, if it is determined to be a fault-related problem description, use a retrieval enhancement method to perform feature matching on the text features in the fault knowledge question and answer database to obtain system fault prediction results and system solutions from the question and answer knowledge database; obtain network fault prediction results and network solutions through network query according to the text feature vector; obtain the graph prediction fault results and graph solutions in combination with the knowledge graph constructed for the equipment fault according to the text feature vector;

[0045] A generation module is used to use a hybrid expert model to obtain the optimal fault prediction result and the corresponding optimal solution based on the system fault prediction result and system solution, the networking fault prediction result and networking solution, and the graph prediction fault result and graph solution.

[0046] In a third aspect, an embodiment of the present application further provides a computer storage medium having computer executable instructions stored thereon, and when the computer executable instructions are executed, any one of the device failure prediction methods described in the first aspect of the embodiment of the present application is performed.

[0047] In a fourth aspect, an embodiment of the present application further provides an electronic device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus;

[0048] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute any one of the device failure prediction methods described in the first aspect of the embodiment of the present application.

[0049] The present application provides a method, device, storage medium and electronic device for predicting equipment failure, obtaining a text feature vector for describing the abnormal condition of the equipment; matching the text feature vector from a preset question-and-answer feature vector database to determine whether the problem description is a problem description related to the equipment failure; if it is determined to be a problem description related to the failure, using a retrieval enhancement method, feature matching the text feature in the fault knowledge question-and-answer database to obtain the system failure prediction result and system solution of the question-and-answer knowledge database; obtaining the network failure prediction result and network solution by means of network query according to the text feature vector; obtaining the graph prediction failure result and graph solution according to the text feature vector in combination with the knowledge graph constructed for the equipment failure; using a hybrid expert model, based on the system failure prediction result and system solution, the network failure prediction result and network solution, and the graph prediction failure result and graph solution, obtaining the optimal prediction result of the failure and the corresponding optimal solution. The prediction method of equipment failure in the embodiment of the present application can efficiently provide users with more accurate fault prediction results, and give the optimal solution in a targeted manner, significantly improving the user experience, and effectively improving the operation and maintenance efficiency of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0051] Figure 1 A schematic diagram of a workflow of a method for predicting equipment failure provided in an embodiment of the present application;

[0052] Figure 2 A schematic diagram of the structure of a device for predicting equipment failure provided in an embodiment of the present application;

[0053] Figure 3 A structural schematic diagram of an electronic device is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the embodiments of the present application should fall within the scope of protection of the embodiments of the present application.

[0055] It should be understood that the various steps described in the method implementation of the present application can be performed in different orders and / or performed in parallel. In addition, the method implementation may include additional steps and / or omit the steps shown. The scope of the present application is not limited in this respect.

[0056] Embodiment 1

[0057] The present application embodiment provides a method for predicting equipment failure, such as Figure 1 As shown, Figure 1 A schematic diagram of a workflow of a method for predicting equipment failure provided in an embodiment of the present application includes:

[0058] Step S101, obtaining a text feature vector describing a problem of an abnormal condition of a device. In an embodiment of the present application, the problem description is the text content of a question asked by a user to the system based on the actual abnormal condition of the device.

[0059] Optionally, in an implementation of the embodiment of the present application, obtaining a text feature vector for describing the problem of the abnormal condition of the device includes: preprocessing the text content of the problem description to obtain the preprocessed text, extracting the text feature vector from the preprocessed text to obtain the text feature vector, wherein the preprocessing includes converting the problem description into text data in a structured preset format that conforms to system recognition. This allows the system to efficiently and accurately identify the problem description content, thereby accurately extracting the text feature vector.

[0060] Step S102: Match the text feature vector from a preset question-answer feature vector database to determine whether the question description is a question description related to equipment failure. In this stage, the embodiment of the present application determines whether the user's current system question is a question description related to equipment failure by matching the features of the question description. The process is simple and easy to implement, and the judgment is highly accurate.

[0061] Step S103: If it is determined to be a problem description related to a fault, a retrieval enhancement method is used to perform feature matching on the text features in the fault knowledge question and answer library to obtain system fault prediction results and system solutions from the question and answer knowledge library; based on the text feature vector, network fault prediction results and network solutions are obtained through network query; based on the text feature vector, combined with the knowledge graph constructed for the equipment fault, the graph prediction fault results and graph solutions are obtained.

[0062] Optionally, in an implementation of the embodiment of the present application, a retrieval enhancement method is used to perform feature matching on the text features in a fault knowledge question and answer library to obtain system fault prediction results and system solutions from the question and answer knowledge library, including: determining the fault knowledge question and answer library, the fault knowledge question and answer library includes at least one of a local fault knowledge question and answer library or an external fault knowledge question and answer library; using an efficient retrieval algorithm to generate a query request based on the text features to select relevant documents from the selected fault knowledge question and answer library; extracting information from the relevant documents to extract key information of the relevant documents, the key information including one or more of entity information (i.e., fault description, solution or fault cause information), entity relationship information (i.e., fault description-fault cause relationship, fault cause-solution relationship) and context information; using a large language model to fuse the key information to generate fault prediction information and predicted fault solution information; performing at least one text processing method of hallucination suppression and noise information processing on the fault prediction information and the predicted fault solution information to generate system fault prediction results and system solutions from the question and answer knowledge library. In this stage of the present application embodiment, a variety of knowledge bases that can more comprehensively cover the operation field of the equipment, such as equipment fault descriptions, fault solutions or equipment upgrade information, are selected to make the process of matching the descriptions of the problems raised by users more reliable and comprehensive, thereby ensuring the accuracy and comprehensiveness of the text content generated for user feedback and improving user satisfaction. Furthermore, when checking and querying at this stage, an efficient retrieval algorithm is used, such as

[0063] Algorithms such as BM25 (Best Matching) or vector search are used to generate query requests to perform related information retrieval queries on the selected knowledge question and answer base, which can improve the accuracy and efficiency of retrieval queries. Information fusion can obtain more accurate and comprehensive fault prediction information and corresponding solutions. In the embodiment of the present application, hallucination suppression refers to checking whether the generated text is consistent with the known facts, for example, by comparing with the information in the knowledge base, or using a special fact-checking model to determine the accuracy of the generated text content. At the same time, constraints can also be configured for the generated text that is inconsistent with the facts, wherein the constraints can be constructed based on language rules, domain knowledge or prior text content, and the embodiment of the present application is not limited here. Noise processing refers to removing the document content with low relevance or pseudo-relevance in the retrieved documents mainly through the two text processing mechanisms of document screening and error correction. At the same time, the corresponding text correction processing algorithm is designed by using the text content analysis method, and the documents containing erroneous information are corrected by using other reliable data through verification or interactive supplementation based on the extracted context information. Thereby improving the accuracy of the feedback document.

[0064] Optionally, in one implementation of the embodiment of the present application, when generating system fault prediction results and system solutions of a question-and-answer knowledge base, the method also includes: text optimization of the system fault prediction results and system solutions, and the text optimization process performs text filtering, text compression and reordering on the data processing method to determine that the text content of the finally generated fault prediction results and system solutions has better accuracy.

[0065] Step S104, using a hybrid expert model, based on the system fault prediction results and system solutions, the networking fault prediction results and networking solutions, and the graph prediction fault results and graph solutions, obtain the optimal fault prediction results and the corresponding optimal solutions. The hybrid expert model (MoE, full name Mixture of Experts) is a deep learning technology that directly combines multiple models (these models are called "experts") to speed up the convergence of the model and obtain better prediction performance. This model design strategy is particularly important in large models, as it can solve some problems faced by large models during training. In this stage of the embodiment of the present application, a hybrid expert model is used to combine the results of various fault predictions and the corresponding solutions to obtain the optimal fault prediction results and the corresponding optimal solutions for feedback, which are efficient and accurate.

[0066] Optionally, in one embodiment of the present application, after obtaining the optimal prediction result of the fault and the corresponding optimal solution, the method further includes: determining the abnormal detection result of the device based on the optimal prediction result of the fault and in combination with the operating data of the device; and providing an early warning prompt according to the result of the abnormal detection. Specifically, this method allows operation and maintenance personnel to promptly discover equipment failures in order to adjust and optimize equipment operating cost parameters. For example, for printing equipment, equipment operating parameters can be optimized by generating targeted early warning prompts. According to the environment in which the equipment is deployed and the type of certificate, anti-counterfeiting material, type of consumables, processing and production processes such as thermal transfer, laser etching + color UV, etc., a personalized parameter setting scheme for early warning prompts is generated, including calibration of the operating temperature and sensor position of the printing equipment, setting the operating temperature of the laminating machine, laminating speed, speed of the conveying station, and other related parameter adjustments. The embodiments of the present application will not be repeated here. In addition to providing targeted content feedback to users regarding the problem description, the limitations at this stage of the embodiment of the present application can further combine the fault prediction results for anomaly detection and issue early warnings based on the results of the anomaly detection, so that the system can provide feedback to the user on the identified faults and solutions while promptly notifying the user of possible abnormal conditions and operating risks of the equipment, further ensuring stable operation of the equipment and improving user satisfaction.

[0067] Optionally, in an implementation of the embodiment of the present application, before obtaining the graph prediction fault result and graph solution based on the text feature vector and the knowledge graph constructed for the equipment fault, the method also includes: determining the equipment operation and maintenance information consisting of the fault description, solution and fault cause of the equipment from open world data; converting the operation and maintenance information into structured semantic block data in a predefined format; performing incremental entity extraction and incremental relationship extraction on the structured semantic block data to obtain unique entity information in each semantic block and determine the relationship information between the entities; and converting the acquired The unique entity information and the relationship information are integrated into a preset graph database, combined with the device-related knowledge graph subgraph retrieved from the existing knowledge graph, to generate a specific fault knowledge graph and a specific device knowledge graph, wherein the specific device knowledge graph contains sequential data during the operation of the specific device; perform node and edge clustering on the generated specific fault knowledge graph to create a fault aggregation knowledge graph; merge the relevant specific fault graph and the time information about the sequential data in the specific knowledge graph to generate a personalized knowledge graph for the device, and determine the personalized knowledge graph as the knowledge graph for the device fault. In an embodiment of the present application, the open world data includes system local data and Internet data, etc., to ensure the multi-source and comprehensiveness of the acquired data. In this way, the embodiment of the present application here can effectively ensure the richness of information covered by the knowledge graph of the constructed equipment fault, avoid the inclusion of semantically mixed information and conflicting information, and facilitate users to intuitively understand the relationship between complex equipment operation and maintenance knowledge points.

[0068] Optionally, in one embodiment of the present application, according to the text feature vector, combined with the knowledge graph constructed for the equipment failure, the graph prediction fault result and the graph solution are obtained, including: according to the text feature vector, using double attention enhancement (BAT) and graph neural network (GNN), based on the personalized knowledge graph, to predict the fault of the device to obtain the graph prediction fault result and the graph solution, so as to improve the accuracy of the acquired graph prediction fault result and graph solution.

[0069] Specifically, in the implementation of the embodiment of the present application, the specific process of constructing the knowledge graph can be implemented through the following four functional modules: document parsing module, incremental entity extraction module, incremental relationship extraction module and graph integration module. The constructed knowledge graph can be expressed as G = {E, R, F, T}, where E represents the entity set of the extracted device, R is the extracted entity relationship set, F represents the set of corresponding fault solutions, and T represents the time when the event occurs, which is composed of different types of relationships between entities.

[0070] In the process of constructing a knowledge graph, the document parsing module is used to take documents about device usage instructions such as solutions to device operation failures, operation log data, device operation and maintenance manuals, and user manuals as inputs. The large language model is used to extract semantic block entity information from these input documents for the text feature vectors of these inputs, so as to obtain a set of entity information containing 4-tuples (e, r, f, t). Among them, e represents the device entity, and the entity includes at least the device name, fault name, and actual fault phenomenon; r represents the relationship between entities, f represents the relevant fault solution, and t represents the time of the fault occurrence event. When i < j, t i <t j . The constructed knowledge graph G of the device at time t t can be expressed as:

[0071] G t ={(e i ,r i ,f i ,t i ):e i ∈E, r i ∈R, f i ∈F, t i <t}

[0072] When constructing the knowledge graph, the document parsing module can also pre-filter the quality of the extracted semantic block data in advance to remove useless information in the document, so as to ensure that the text feature data of the constructed knowledge graph meets the format and content requirements for constructing the knowledge graph.

[0073] Incremental entity extraction module: used to iteratively extract the obtained semantic blocks using a preset incremental entity matcher and extract global document entity information. Specifically, in the first step, the large language model is used to extract entities from the first semantic block D to form a global entity set E. According to the constraint condition C, the large language model is constrained to extract entities representing a unique concept to avoid extracting semantic mixed entities. For subsequent input semantic block data, the entities extracted from it are used as local entities E d . In the second step, the extracted local entity E d is matched with the global entities in the local global entity set E. If a corresponding local entity E i is matched in E, it will be added to the matching set E d匹配 , and this E d匹配 represents an incremental entity set temporarily created when extracting entities from these subsequent input semantic block data. If no match is found, the cosine similarity metric with a predefined threshold is used to search for similar entities in E. If no match is found, the local entity is added to E d匹配Otherwise, add the most matching global entity E based on the maximum similarity algorithm i ′. The third step is to combine E and E d匹配 Unify to update the global entity set E. Repeat this process for each semantic block to generate a more comprehensive global entity set E.

[0074] The incremental relationship extraction module is used to use the incremental relationship matcher to extract global document relations with each semantic block using the global document entity E as context. It uses the same method as the entity matcher. When the global entity is provided as context, for entities that are not clearly present in the semantic block, the large language model is used to extract the relationship between the semantic blocks, and the entity relationship obtained in the previous step is incrementally updated to obtain a more comprehensive global entity relationship;

[0075] The graph integration module stores the global semantic entities and global semantic relationships obtained in the above steps into the graph database, and constructs a knowledge graph based on the graph database. Preferably, in this stage of the present embodiment, a graph neural network (GNN) model based on a dual attention enhancement mechanism is used to construct a knowledge graph, which combines the characteristics of the graph attention network and heterogeneous graphs. The specific process is as follows:

[0076] Attention calculation method: The graph is represented as {E, R, F, T}. In the lth layer, a neural network model is used to generate the entity embedding representation Y l , the initial input is Y 0 .D l is the feature dimension of the lth layer, N = |E|.

[0077] The update function of each layer is:

[0078] Y l (f) = A(∑ e∈E At l (e,r,f)·Msg l (e,r,f))

[0079] A is the activation function, such as Leaky-ReLU

[0080] Atn l (e,r,f): attention weight, used to indicate the influence of equipment,fault phenomenon entity e on solution f;

[0081] Msg l (e,r,f): message vector, containing the information of the device entity;

[0082] The dual attention enhancement mechanism of the neural network model represents the input features of each layer as a combination of multiple subspaces, represented as Y l =[Yl,1 ,...,Y l,H ]∈R N×Dl , where H is the number of attention heads, in this paper H = 2, D l is the feature dimension of the lth layer, D l is the feature dimension of the lth layer, N = |E|.

[0083] The following formula is used to calculate the attention score S of each attention head:

[0084]

[0085] W l [e] T Represents the key vector of the entity, W l represents a relation-specific trainable weight matrix; Q l [f] is the query vector of the solution; is the relationship matrix between entities and solutions;

[0086] In order to better perform time-based modeling to facilitate equipment failure prediction and remaining life assessment, the present embodiment introduces time information into the model at this stage. Specifically, a recurrent neural network (RNN) is used to model time changes. in, is the temporal embedding at time t;

[0087] In order to capture the changes of the knowledge graph over time, the embodiment of the present application uses a recurrent neural network (RNN) at this stage to update the temporal embeddings and structural embeddings of entities and relationships.

[0088] Furthermore, in this stage of the present application, probability modeling can also be used to fully update the knowledge graph through the learning of the dynamic knowledge graph. Specifically, probability modeling defines a probability model for the occurrence of an event:

[0089]

[0090] in, Represents a set of new events occurring at time t. Decompose the conditional probability into two parts: structure and time:

[0091]

[0092] Then, the constructed knowledge graph is iteratively updated using the probabilistic modeling, and the loss function is set to define a loss function that at least contains time and structure information:

[0093]

[0094] μ 1 and μ 2 It is a preset adjustable parameter;

[0095] The embodiment of the present application constructs the final knowledge graph in the above-mentioned manner, which can better ensure that the final constructed knowledge graph has more comprehensive equipment entity information and relationship information, thereby further improving the accuracy of the output graph prediction fault results and graph solutions.

[0096] Optionally, in one embodiment of the present application, the method further includes: regularly acquiring a device operation knowledge base; using a trained large language model to extract key information in the device operation knowledge base, wherein the key information includes at least one of a device problem description, a fault cause description, and a fault solution. Based on the extracted key information, the knowledge graph of device faults or the fault knowledge base is updated to improve the novelty and accuracy of the system's feedback or solutions to device faults.

[0097] Specifically, in an optional implementation of an embodiment of the present application, according to the optimal prediction result of the fault, the device is detected for abnormality, including: creating a digital twin simulation device for the device; performing data fusion according to the simulated operation data obtained by the simulation of the working state of the simulation device, the actual monitoring data obtained by monitoring the working state of the device, and the statistical data obtained by statistically analyzing the working state of the same type of equipment; constructing a health indicator and a neural network model for predicting the remaining life of the equipment according to the fused data; using the health indicator and the neural network model for predicting the remaining life of the equipment, combined with the optimal prediction result of the fault, the digital twin simulation device is evaluated for performance, so as to determine the abnormal detection result according to the result of the performance evaluation. In the embodiment of the present application, the digital twin simulation device is used as the monitoring target for the system to monitor the abnormality of the equipment, and there is no need to connect the system to the actual equipment, thereby achieving the effect of low-cost monitoring of the health status of the equipment. In the monitoring effect, there is no need to perform cumbersome processes such as shutdown detection on the actual equipment. While ensuring the operating efficiency of the equipment, the health evaluation of the equipment performance is more accurate.

[0098] Optionally, in an actual application scenario of an embodiment of the present application, a digital twin simulation device for the device is created, including: collecting multi-source data of the device, and creating a digital twin simulation device for the device based on the multi-source data using a preset geometric structure model, a data analysis model, and a physical model. In an embodiment of the present application, the geometric model can accurately construct the shape of each component of the digital twin model corresponding to the device based on the device shape data such as the point cloud data of the collected device, and dynamically update the corresponding shape of the created digital twin simulation device according to the change of the shape of each component caused by the operation status of the device. The data analysis model can calibrate and update the operation status of the device through incremental learning or continuous learning. At the same time, the physical model can be used to discover new relevant physical knowledge data that affects the operation status of the device, and update the operation environment of the device or new fault sources. At this stage of the embodiment of the present application, at least these three models are used to create a digital twin simulation device model to achieve a high-assurance mapping between real-world assets and simulation simulations, and a more accurate prediction of the actual abnormal situation of the device through abnormal monitoring of the digital twin simulation device model is achieved.

[0099] The present application provides a method for predicting equipment failure, obtaining a text feature vector for describing the abnormal condition of the equipment; matching the text feature vector from a preset question-and-answer feature vector database to determine whether the problem description is a problem description related to the equipment failure; if it is determined to be a problem description related to the failure, using a retrieval enhancement method, feature matching the text feature in the fault knowledge question-and-answer database to obtain the system failure prediction result and system solution of the question-and-answer knowledge database; obtaining the network failure prediction result and network solution through network query according to the text feature vector; obtaining the graph prediction failure result and graph solution according to the text feature vector in combination with the knowledge graph constructed for the equipment failure; using a hybrid expert model, based on the system failure prediction result and system solution, the network failure prediction result and network solution, and the graph prediction failure result and graph solution, obtaining the optimal failure prediction result and the corresponding optimal solution. The prediction method of equipment failure in the embodiment of the present application can efficiently provide users with more accurate fault prediction results, and give the optimal solution in a targeted manner, significantly improving the user experience, and effectively improving the operation and maintenance efficiency of the equipment.

[0100] Embodiment 2

[0101] Based on the device failure prediction method provided in the first embodiment of the present application, the present embodiment also provides a device failure prediction apparatus, such as Figure 2 As shown, Figure 2A schematic diagram of the structure of a device 20 for predicting equipment failure provided in an embodiment of the present application, wherein the device 20 for predicting equipment failure includes:

[0102] Extraction module 201, used to obtain a text feature vector describing the abnormal condition of the device;

[0103] A matching module 202 is used to match the text feature vector from a preset question-answer feature vector database to determine whether the problem description is a problem description related to equipment failure;

[0104] The retrieval module 203 is used to, if it is determined to be a problem description related to the fault, use a retrieval enhancement method to perform feature matching on the text features in the fault knowledge question and answer database to obtain the system fault prediction results and system solutions of the question and answer knowledge database; obtain the network fault prediction results and network solutions by means of network query according to the text feature vector; obtain the graph prediction fault results and graph solutions according to the text feature vector in combination with the knowledge graph constructed for the equipment fault;

[0105] The generation module 204 is used to use the hybrid expert model to obtain the optimal fault prediction result and the corresponding optimal solution based on the system fault prediction result and system solution, the networking fault prediction result and networking solution, and the graph prediction fault result and graph solution.

[0106] Optionally, in an implementation of an embodiment of the present application, the retrieval module 203 is also used to: determine the fault knowledge question and answer base, the fault knowledge question and answer base includes at least one of a local fault knowledge question and answer base or an external fault knowledge question and answer base; use an efficient retrieval algorithm to generate a query request based on the text features to select relevant documents from the selected fault knowledge question and answer base; extract information from the relevant documents to extract key information of the relevant documents, the key information including one or more of entity information, entity relationship information and context information; use a large language model to fuse the key information to generate fault prediction information and predicted fault solution information; perform at least one text input processing of hallucination suppression, noise information processing, document screening and error correction on the fault prediction information and the predicted fault solution information to generate a system fault prediction result and system solution of the question and answer knowledge base.

[0107] Optionally, in an implementation method of the embodiment of the present application, the device 20 also includes a construction module (not shown in the accompanying drawings), which is used to determine the equipment operation and maintenance information consisting of the fault description, solution and fault cause of the equipment from the open world data before obtaining the fault result predicted by the graph and the graph solution based on the text feature vector and the knowledge graph constructed for the equipment fault; convert the operation and maintenance information into structured semantic block data in a predefined format; perform incremental entity extraction and incremental relationship extraction on the structured semantic block data to obtain unique entity information in each semantic block, and determine the relationship between the entities. ; integrating the acquired unique entity information and the acquired relationship information into a preset graph database, combining the device-related knowledge graph subgraph retrieved from the existing knowledge graph, generating a specific fault knowledge graph and a specific device knowledge graph, wherein the specific device knowledge graph contains sequential data during the operation of the specific device; performing node and edge clustering on the generated specific fault knowledge graph to create a fault aggregation knowledge graph; merging the time information about sequential data in the related specific fault graph and the specific knowledge graph to generate a personalized knowledge graph for the device, and determining the personalized knowledge graph as the knowledge graph for the device fault.

[0108] Optionally, in an implementation method of the embodiment of the present application, the retrieval module 203 is also used to: according to the text feature vector, use dual attention enhancement and graph neural network to perform fault prediction on the device based on the personalized knowledge graph to obtain the graph prediction fault result and graph solution.

[0109] Optionally, in an implementation method of the embodiment of the present application, the device 20 also includes an update module (not shown in the accompanying drawings), which is used to regularly obtain a device operation knowledge base; use a trained large language model to extract key information in the device operation knowledge base, and the key information includes at least one of a device problem description, a fault cause description, and a fault solution; based on the extracted key information, the personalized knowledge graph or the fault knowledge base is updated.

[0110] Optionally, in an implementation method of the embodiment of the present application, the device 20 also includes a detection module (not shown in the drawings), which is used to perform abnormality detection on the equipment according to the optimal fault prediction result after obtaining the optimal fault prediction result and the corresponding optimal solution; and to issue an early warning prompt to the equipment according to the result of the abnormality detection.

[0111] Optionally, in an implementation of the embodiment of the present application, the detection module is further configured to:

[0112] creating a digital twin simulated device for the device;

[0113] Performing data fusion based on simulated operation data obtained through working state simulation of the simulated device, actual monitoring data obtained through monitoring the working state of the device, and statistical data obtained through statistics on the working states of devices of the same type, to obtain fused data;

[0114] Constructing a neural network model for predicting health indicators and remaining life of the equipment for the equipment according to the fused data;

[0115] Using the health indicators and the equipment remaining life prediction neural network model, combined with the optimal fault prediction result, the digital twin simulation equipment is performance evaluated to determine the abnormality detection result through the result of the performance evaluation.

[0116] Optionally, in an implementation of the embodiment of the present application, the detection module is further used to collect multi-source data of the device;

[0117] Based on the multi-source data, a digital twin simulation device of the device is created using a preset geometric structure model, a data analysis model and a physical model.

[0118] The present application provides a prediction device for equipment failure, which includes an extraction module for obtaining a text feature vector for describing a problem of an abnormal condition of an equipment; a matching module for matching the text feature vector from a preset question-and-answer feature vector database to determine whether the problem description is a problem description related to the equipment failure; a retrieval module for matching the text feature in a fault knowledge question-and-answer database using a retrieval enhancement method to obtain a system failure prediction result and a system solution from the question-and-answer knowledge database if the problem description is determined to be a failure-related problem description; and a network failure prediction result and a network solution are obtained through network query based on the text feature vector; and a graph prediction failure result and a graph solution are obtained based on the text feature vector and a knowledge graph constructed for the equipment failure. A generation module is set to use a hybrid expert model to obtain an optimal failure prediction result and a corresponding optimal solution based on the system failure prediction result and the system solution, the network failure prediction result and the network solution, and the graph prediction failure result and the graph solution. This equipment failure prediction device can efficiently provide users with relatively accurate fault prediction results and give targeted optimal solutions, which significantly improves the user experience and effectively improves the operation and maintenance efficiency of the equipment. The device has a simple structure and is easy to implement.

[0119] Embodiment 3

[0120] An embodiment of the present application further provides a storage medium on which a computer program is stored. When the program is executed by a processor, any one of the device failure prediction methods described in the first embodiment of the present application is implemented.

[0121] Embodiment 4:

[0122] The present application also provides an electronic device, such as Figure 3 As shown, Figure 3 A schematic diagram of the structure of an electronic device 30 is provided for an embodiment of the present application. The electronic device 30 includes:

[0123] One or more processors 301, communication interface 302, memory 303 and communication bus 304, the processor 301, memory 303 and communication interface 302 communicate with each other via the communication bus 304;

[0124] Memory 303, used to store one or more programs;

[0125] When the one or more programs are executed by the one or more processors 301, the one or more processors 301 implement any one of the device failure prediction methods described in Embodiment 1 of the present application.

[0126] So far, the application has described specific embodiments of the subject matter. In some cases, the actions recorded in the claims can be performed in different orders and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing can be advantageous.

[0127] In the 1990s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system layer on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0128] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

[0129] The system layers, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0130] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0131] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0132] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, system layers or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0133] The present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific transactions or implement specific abstract data types. The present application may also be practiced in distributed computing environments where transactions are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0134] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system layer embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0135] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for predicting equipment failure, characterized in that: include: Obtain a text feature vector describing the abnormal condition of the device; Matching the text feature vector from a preset question-answer feature vector database to determine whether the problem description is a problem description related to the equipment failure; If it is determined to be a fault-related problem description, the text features are matched in the fault knowledge question and answer database using a search enhancement method to obtain system fault prediction results and system solutions from the question and answer knowledge database; based on the text feature vector, network fault prediction results and network solutions are obtained through network query; based on the text feature vector, combined with the knowledge graph constructed for the equipment fault, the graph prediction fault results and graph solutions are obtained; By using a hybrid expert model, based on the system fault prediction results and system solutions, the networking fault prediction results and networking solutions, and the graph prediction fault results and graph solutions, the optimal fault prediction results and the corresponding optimal solutions are obtained.

2. The method for predicting equipment failure according to claim 1, characterized in that: The method of using retrieval enhancement to perform feature matching on the text features in the fault knowledge question and answer database to obtain system fault prediction results and system solutions from the question and answer knowledge database includes: Determine the fault knowledge question and answer base, wherein the fault knowledge question and answer base includes at least one of a local fault knowledge question and answer base and an external fault knowledge question and answer base; Using an efficient retrieval algorithm to generate a query request according to the text features to select relevant documents from the selected fault knowledge question and answer library; Extracting information from the relevant documents to extract key information of the relevant documents, wherein the key information includes one or more of entity information, entity relationship information, and context information; Using a large language model, the key information is integrated to generate fault prediction information and predicted fault resolution information; The fault prediction information and the predicted fault solution information are subjected to at least one text input processing of hallucination suppression, noise information processing, document screening and error correction to generate a system fault prediction result and system solution of the question-answering knowledge base [HZTC-HXJKJD241005][HS2411223CCN].

3. The method for predicting equipment failure according to claim 1, characterized in that: Before obtaining the graph prediction result and graph solution of the equipment fault in combination with the knowledge graph constructed according to the text feature vector, the method further includes: Determine the equipment operation and maintenance information consisting of the fault description, solution and fault cause of the equipment from the open world data; Converting the operation and maintenance information into structured semantic block data in a predefined format; Performing incremental entity extraction and incremental relationship extraction on the structured semantic block data to obtain unique entity information in each semantic block and determine relationship information between the entities; Integrate the acquired unique entity information and the relationship information into a preset graph database, and combine the device-related knowledge graph subgraph retrieved from the existing knowledge graph to generate a specific fault knowledge graph and a specific device knowledge graph, wherein the specific device knowledge graph contains sequential data during the operation of the specific device; Perform node and edge clustering on the generated specific fault knowledge graph to create a fault aggregation knowledge graph; The relevant specific fault graph and the time information about the sequential data in the specific knowledge graph are combined to generate a personalized knowledge graph for the device, and the personalized knowledge graph is determined as the knowledge graph for the device fault.

4. The method for predicting equipment failure according to claim 3, characterized in that: The step of obtaining the graph prediction fault result and graph solution based on the text feature vector and the knowledge graph constructed for the equipment fault includes: According to the text feature vector, dual attention enhancement and graph neural network are used to predict the fault of the device based on the personalized knowledge graph to obtain the graph prediction fault result and graph solution.

5. The method for predicting equipment failure according to claim 1, characterized in that: The method further comprises: regularly acquiring a new equipment operation knowledge base; [HZTC-HXJKJD241005][HS2411223CCN] uses a trained large language model to extract key information from the equipment operation knowledge base, where the key information includes at least one of the equipment problem description, fault cause description, and fault solution. According to the extracted key information, the knowledge graph or fault knowledge base of the equipment fault is updated.

6. The method for predicting equipment failure according to claim 1, characterized in that: After obtaining the optimal fault prediction result and the corresponding optimal solution, the method further includes: According to the optimal fault prediction result, performing abnormality detection on the device; According to the result of the abnormality detection, an early warning prompt is given to the device.

7. The method for predicting equipment failure according to claim 6, characterized in that ,According to the optimal fault prediction result, performing abnormality detection on the device includes: creating a digital twin simulated device for the device; Performing data fusion based on simulated operation data obtained through working state simulation of the simulated device, actual monitoring data obtained through monitoring the working state of the device, and statistical data obtained through statistics on the working states of devices of the same type, to obtain fused data; Constructing a neural network model for predicting health indicators and remaining life of the equipment for the equipment according to the fused data; Using the health indicators and the equipment remaining life prediction neural network model, combined with the optimal fault prediction result, the digital twin simulation equipment is performance evaluated to determine the abnormality detection result through the result of the performance evaluation.

8. The method for predicting equipment failure according to claim 7, characterized in that: The step of creating a digital twin simulation device for the device on the server includes: Collecting multi-source data of the device; Based on the multi-source data, a digital twin simulation device of the device is created using a preset geometric structure model, a data analysis model and a physical model.

9. A device for predicting equipment failure, characterized in that: include: [HZTC-HXJKJD241005][HS2411223CCN] Extraction module, used to obtain text feature vectors describing the abnormal condition of the equipment; A matching module, used to match the text feature vector from a preset question-answer feature vector database to determine whether the problem description is a problem description related to the equipment failure; A retrieval module is used to, if it is determined to be a fault-related problem description, use a retrieval enhancement method to perform feature matching on the text features in the fault knowledge question and answer database to obtain system fault prediction results and system solutions from the question and answer knowledge database; obtain network fault prediction results and network solutions through network query according to the text feature vector; obtain the graph prediction fault results and graph solutions in combination with the knowledge graph constructed for the equipment fault according to the text feature vector; A generation module is used to use a hybrid expert model to obtain the optimal fault prediction result and the corresponding optimal solution based on the system fault prediction result and system solution, the networking fault prediction result and networking solution, and the graph prediction fault result and graph solution.

10. A computer storage medium, characterized in that: The computer storage medium stores computer executable instructions, and when the computer executable instructions are executed, the method for predicting equipment failure according to any one of claims 1 to 8 is executed.

Images