Industrial operation and inspection large-scale model system and method and electronic equipment

Abstract

The invention provides an industrial operation inspection large model system and method and electronic equipment, and relates to the technical field of industrial operation inspection.The industrial operation inspection large model system comprises an intention recognition sub-model which recognizes an operation inspection problem based on an operation inspection inquiry statement and judges whether the operation inspection problem exists in a question and answer library or not; the first interaction sub-model is used for outputting an answer to the operation inspection question when the operation inspection question exists in the question and answer library; and the second interaction sub-model is used for searching and outputting an answer to the operation inspection question in a knowledge graph based on the operation inspection question when the operation inspection question does not exist in the question and answer library. Through the method, the operation inspection problem of the user can be efficiently and accurately solved, and the accuracy and objectivity of the session of the model are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial operation and maintenance, and in particular to an industrial operation and maintenance large model system, method and electronic equipment. Background Art

[0002] With the improvement of science and technology, artificial intelligence (AI) is widely used in various fields, that is, using large models to achieve human-computer interaction.

[0003] In the training process of large models, there are specific terms and expression habits in the vertical application field of industrial system operation and inspection (hereinafter referred to as industrial operation and inspection). Without special training, large models will lack knowledge of this field. At the same time, large models are often black box systems. Without special training, it is impossible to explain the decision-making process of the model, and thus it is impossible to give efficient and accurate answers.

[0004] In summary, for industrial operation and maintenance detection models, large models in existing technologies cannot provide efficient and accurate answers, and the objectivity of model conversations is not high. Summary of the invention

[0005] In view of this, an object of the present invention is to provide an industrial operation and inspection large model system, method and electronic equipment, in order to solve at least part of the above-mentioned problems.

[0006] In a first aspect, the present application provides an industrial operation and inspection large model system, characterized in that it includes: Intention recognition sub-model: Identify inspection questions based on inspection query statements and determine whether the inspection questions exist in the question and answer database; The first interactive sub-model is used to output answers to the operation inspection questions when the operation inspection questions exist in the question and answer database; The second interactive sub-model is used for the operation inspection question that does not exist in the question-answering database, and the operation inspection question is searched in the knowledge graph based on the operation inspection question, and the answer to the operation inspection question is output. The embodiment of the present invention provides an industrial operation inspection large model system, including: In the example provided above, a preset question-and-answer library is used. If the question raised by the user is in the question-and-answer library, the answer is directly output using the question-and-answer library. If the question raised by the user is not in the question-and-answer library, the knowledge graph is used for retrieval. The above method can efficiently and accurately answer the user's operation and inspection questions, thereby improving the objectivity and accuracy of the model's conversation.

[0007] One possible approach is to construct the first interaction sub-model in the following manner: Constructing an operation and inspection knowledge data set, pre-training a language autoregressive model based on the operation and inspection knowledge data set, and generating a domain base model, wherein the operation and inspection knowledge data set includes: equipment information in a target system, common faults of the target system, and at least one of a maintenance method of the target system; Extracting keywords from each paragraph in the operation and inspection knowledge data set, and constructing at least one first operation and inspection question-answer pair based on the keywords, each of the first operation and inspection question-answer pairs comprising: an operation and inspection question for a target system and a first output answer for the operation and inspection question, wherein the first output answer for the operation and inspection question comprises at least part of the keywords; The first operation inspection question and answer pair is verified to generate a second operation inspection question and answer pair, and the domain base model is fine-tuned based on the second operation inspection question and answer pair.

[0008] The question-and-answer database includes all the operation and inspection questions for the target system in the second operation and inspection question-and-answer pairs.

[0009] In one possible manner, the industrial operation and inspection large model system further includes: a first interactive sub-model adjustment module; The first interaction sub-model adjustment module is used to fine-tune the first interaction sub-model based on the third operation-inspection question-answer pair; The third operation-inspection question-answer pair is constructed in the following manner: Acquire first feedback information from a user, and determine a target problem based on the first feedback information; Obtain answers to the operation and inspection questions corresponding to the target operation and inspection questions; The third operation and inspection question and answer pair is constructed based on the target question and the operation and inspection question answer corresponding to the target operation and inspection question.

[0010] One possible approach is that the first interaction sub-model outputs a plurality of answers to the operation inspection question, and each of the answers to the operation inspection question is set with a first weight; First interactive sub-model: outputting the answers to the operation inspection questions in sequence according to the first weight corresponding to each answer to the operation inspection question.

[0011] One possible manner is that the first weight is positively correlated with the priority of the answer to the inspection question.

[0012] One possible way is that the knowledge graph corresponds to a knowledge graph topology, the knowledge graph topology includes: a node, and a plurality of edges, the node includes: at least one device node and at least one fault node; For the first target device, the second interaction sub-model uses the following relationship to output answers to the operation inspection questions; Outputting the answer to the operation inspection question based on the edge formed by the first target device node and the fault node, wherein the edge is used to determine the dependency relationship between the corresponding device node and the fault node; and / or Based on the edge formed by the second target device and the first target device node in the knowledge graph topology, the answer to the operation and inspection question is output, the second target device is an associated device of the first target device, and the edge is used to determine the dependency relationship between the corresponding two device nodes.

[0013] One possible way is that the second interaction sub-model adjustment module: obtains second feedback information; The device node and / or faulty node in the knowledge graph topology is modified based on the second feedback information.

[0014] One possible manner is that in the step of outputting the answer to the operation inspection question based on the edge formed by the first target device node and the fault node, the first target device node and the plurality of fault nodes respectively form a plurality of edges, and a plurality of answers to the operation inspection question are output correspondingly, and each answer to the operation inspection question is provided with a second weight. The second interaction sub-model: outputting the answer to the operation inspection question according to the second weight corresponding to each answer to the operation inspection question; and / or In the step of outputting the answer to the operation and inspection question based on the edge formed by the second target device and the first target device node in the knowledge graph topology, the first target device node respectively forms multiple edges with multiple second target devices, and correspondingly outputs multiple answers to the operation and inspection question, each of which is set with a second weight. The second interaction sub-model outputs the answers to the operation and inspection questions according to the third weight corresponding to each answer to the operation and inspection question.

[0015] In a second aspect, an industrial operation inspection method is characterized by comprising the following steps: Get the operation and inspection query statement; The operation and inspection query statement is input into the industrial operation and inspection large model system as described in the first aspect.

[0016] In a third aspect, an embodiment of the present application provides an electronic device, including: at least one processor; and at least one memory in communication with the processor, wherein: The memory stores program instructions that can be executed by the processor, and the processor can call the program instructions to execute the method as described in the second aspect above.

[0017] 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, claims and drawings.

[0018] 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

[0019] 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.

[0020] Figure 1 A structural diagram of an industrial operation and inspection large model system provided by an embodiment of the present invention; Figure 2 A structural diagram of another industrial operation and inspection large model system provided by an embodiment of the present invention; Figure 3 Another industrial operation and inspection large model system structure diagram provided by an embodiment of the present invention; Figure 4 A step diagram of an industrial operation and inspection method provided by an embodiment of the present invention; Figure 5 This is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] The following is a brief introduction to the terms involved in the embodiments of the present application: Operation and maintenance: The work of operating and repairing system equipment. It includes daily inspection, maintenance, care, testing, adjustment and repair of equipment to ensure the normal operation of the equipment.

[0023] At present, with the improvement of science and technology, artificial intelligence (AI) is widely used in various fields, that is, using large models to achieve human-computer interaction.

[0024] In the training process of large models, there are specific terms and expression habits in the vertical application field of industrial system operation and inspection (hereinafter referred to as industrial operation and inspection). Without special training, the large model will lack knowledge of this field. At the same time, large models are often black box systems. Without special training, the decision-making process of the model cannot be explained, which in turn affects the implementation of the application.

[0025] Based on this, an embodiment of the present invention provides a method that can efficiently and accurately answer users' operation and inspection questions, thereby improving the accuracy and objectivity of the model's conversation.

[0026] First, the various data provided in the embodiments of the present application will be described: Knowledge graph: It is essentially a vector database that is only queried when in use. Unlike ordinary vector databases, it is stored in the form of a graph, with "relationships" between nodes.

[0027] python-docx: Python library for creating, modifying, and extracting Microsoft Word documents (.docx files).

[0028] To facilitate understanding of this embodiment, an industrial operation and inspection large model system disclosed in an embodiment of the present invention is introduced in detail below.

[0029] Reference Figure 1 In the embodiment provided in the application, the industrial operation and inspection large model system specifically includes: an intention recognition sub-model, a first interaction sub-model and a second interaction sub-model.

[0030] In this embodiment, the intention recognition sub-model identifies the operation inspection question based on the operation inspection query statement and determines whether the operation inspection question exists in the question and answer database.

[0031] If the operation inspection question exists in the question and answer database, the first interactive sub-model outputs the answer to the operation inspection question. Otherwise, that is, the operation inspection question does not exist in the question and answer database, the second interactive sub-model searches the knowledge graph based on the operation inspection question and outputs the answer to the operation inspection question.

[0032] In some examples, the answers to the operation and inspection questions output by the first interactive sub-model only include maintenance solutions for solving equipment failures. In some examples, the answers to the operation and inspection questions output by the first interactive sub-model also include collected real-time data. In this example, in order to generate answers to the operation and inspection questions, the data collected by the measuring device is called. Taking the equipment as an example, the real-time data collected on the punching machine includes: current pressure value, current temperature value, current vibration frequency, etc.

[0033] What needs to be explained is that when querying the operation and inspection questions in the knowledge graph, you will not get a single answer, but also "relationships". The question and answer library here can be a list of questions, including questions and answers. The questions are used as matches, and the answers prepared in advance are output if the match is successful.

[0034] From this, we can see that the question and answer database corresponds to the set of all questions that can be answered by the aforementioned first interaction sub-model.

[0035] It should be noted that the industrial operation and inspection large-scale model system in this application can be specifically used for automobile stamping, welding, painting, assembly, and parts manufacturing.

[0036] For example, assuming that the operation and inspection problem is that device A makes an abnormal sound, and this problem exists in the question and answer database, at this time, the operation and inspection problem does not need to call the knowledge graph, and the first interactive sub-model outputs the answer to the operation and inspection problem, that is, the first interactive sub-model generates the fault cause of device A making an abnormal sound and the fault solution of device A making an abnormal sound.

[0037] For another example, suppose the operation and inspection problem is: how to solve the problem of A suddenly being prompted to overload during normal working state. This problem does not exist in the question and answer database. At this time, the operation and inspection problem cannot be generated through the first interactive sub-model, that is, the operation and inspection problem needs to call the knowledge graph. The second interactive sub-model searches in the knowledge graph based on the operation and inspection problem and outputs the answer to the operation and inspection problem.

[0038] In the example provided above, a preset question-and-answer library is used. If the question raised by the user is in the question-and-answer library, the answer is directly output using the question-and-answer library. If the question raised by the user is not in the question-and-answer library, the knowledge graph is used for retrieval. The above method can efficiently and accurately answer the user's operation and inspection questions, thereby improving the accuracy and objectivity of the model's conversation.

[0039] The following will explain how to construct each sub-model: The first interaction sub-model can be constructed in the following way: As a preferred embodiment, in order to construct the first interaction sub-model, an operation and inspection knowledge data set is first constructed. Then, a language autoregressive model is pre-trained based on the operation and inspection knowledge data set to generate a domain base model.

[0040] It should be noted that in the embodiments provided in the present application, the language autoregressive model can be a GPT model or a BERT model, which is not limited here, and the GPT model is preferably used.

[0041] Further, the industrial operation and maintenance data set includes: the industrial operation and maintenance data set includes: the operation and maintenance knowledge data set includes at least one of equipment information of the target system, common faults of the target system and maintenance methods of the target system.

[0042] Specifically, the target system field is the system field to be detected. Specifically, the target system in the present application can be any field of automobile stamping, welding, painting, assembly, and parts manufacturing.

[0043] For example, the stamping workshop that needs to be inspected uses PDF documents, academic papers, professional books, equipment operation manuals, actual inspection records, industry standards, etc. of various equipment on the stamping line as original data.

[0044] For example, the welding workshop that needs to be inspected and maintained uses PDF documents, academic papers, professional books, equipment operation manuals, actual inspection records, industry standards, etc. of various equipment on the welding line as original data.

[0045] For example, the operation and inspection needs to be carried out for painting, and the PDF documents, academic papers, professional books, equipment operation manuals, actual operation and inspection records, industry standards, etc. of various equipment on the painting line are used as original data.

[0046] For example, the final assembly workshop needs to undergo operation and inspection, and the PDF documents, academic papers, professional books, equipment operation manuals, actual operation and inspection records, industry standards, etc. of various equipment on the final assembly line are used as original data.

[0047] For example, the parts manufacturing workshop that needs to be inspected and maintained uses PDF documents, academic papers, professional books, equipment operation manuals, actual inspection records, industry standards, etc. of various equipment on the parts manufacturing line as original data.

[0048] Then the collected raw data is processed. Combined with the above, it can be seen that the format of raw data may be PDF (Portable Document Format) documents, Word files, Excel tables, image scans, HTML (HyperText Markup Language) web pages, etc. The data structure and representation of each format are different. In order to unify the data format of raw data.

[0049] In some scenarios, the data format of the original data is a PDF document, and OCR (Optical Character Recognition) and PDF parsing tools are used to extract text content.

[0050] In some scenarios, the raw data is in the form of Word files or Excel spreadsheets, which are processed using tools such as python-docx and openpyxl.

[0051] In some scenarios, the data format of the original data is HTML format data. In this embodiment, tool libraries such as BeautifulSoup and lxml are used to parse it, remove HTML tags and redundant format information, and convert it into plain text.

[0052] After the above processing, all data from different sources are integrated into plain text files and saved in csv format. Finally, a text format dataset of stamping workshop operation and inspection knowledge is obtained, namely the aforementioned industrial operation and inspection dataset. The domain base model can be generated by pre-training the language autoregressive model of the above industrial operation and inspection dataset.

[0053] Specifically, taking the GPT model as an example, the next word or token is predicted through context. By inputting the previously sorted text in the field of stamping workshop operation and maintenance in an unsupervised manner, the model will learn information such as equipment names, operating steps, common faults, and maintenance methods.

[0054] However, the pre-trained system cannot give realistic answers based on the habits and professional terms of the operators. Therefore, the model needs to be retrained. The specific methods are as follows: First, the keywords of each paragraph in the operation and inspection knowledge data set are read and extracted in turn. Specifically, each paragraph of the text format data set of the aforementioned stamping workshop operation and inspection knowledge is read paragraph by paragraph, and the keywords of each paragraph are extracted. Then, based on the aforementioned keywords, the questions corresponding to the keywords are summarized, that is, at least one first operation and inspection question and answer pair is constructed based on the keywords.

[0055] Furthermore, the first operation and inspection question and answer pair includes: an operation and inspection question for the target system and a first output answer for the operation and inspection question, and the first output answer for the operation and inspection question includes at least some keywords.

[0056] Specifically, the first output answer to the operation and inspection problem includes: at least one of a cause of the operation and inspection problem, a fault handling method for the operation and inspection problem, and a maintenance strategy for the operation and inspection problem.

[0057] After the first operation and inspection question and answer pair is constructed, the first operation and inspection question and answer pair is verified. Specifically, the first question and answer pair can be manually reviewed, and the wrong and poor quality question and answer pairs can be deleted and modified to improve the quality of the data set. The second operation and inspection question and answer pair is thus generated, and the domain base model is fine-tuned based on the second operation and inspection question and answer pair.

[0058] It should be noted that the difference between the first question-answer pair and the second question-answer pair is that the second question-answer pair is the result of manual correction of the first question-answer pair.

[0059] For example, assuming that the first operation and inspection question and answer pair includes A, B, C and D, A, B, C and D are first manually reviewed and verified to generate , , and , , , and This is the second question and answer pair, using , , and Fine-tuning the domain base model completes the training of the first interaction sub-model.

[0060] Continuing with the above example, taking the stamping workshop as an example, the domain base model can learn the equipment name, operation steps, common faults and maintenance methods of each equipment in the stamping workshop. Assuming that the operation and maintenance problem is: equipment A has an overheating problem, the first interactive operation and maintenance sub-model outputs the cause and maintenance method of the overheating of equipment A.

[0061] Through the above method, the pre-trained model can be supervised and fine-tuned, so that the interactive model can give practical answers based on the habits and professional terms of the operation and maintenance personnel. The large model can be used to give efficient and accurate answers in the operation and maintenance field, thereby improving the objectivity of the model conversation.

[0062] In combination with the above, it can be known that the above question and answer library includes all the operation and inspection questions for the target system in the second operation and inspection question and answer pairs.

[0063] The intent recognition sub-model can be constructed in the following way: As a preferred embodiment, since the aforementioned domain base model already has the basic knowledge of the field, the aforementioned second operation inspection question and answer can be used to perform binary classification training on the domain base model to generate an intent recognition sub-model.

[0064] Specifically, in this embodiment, the question in the second inspection question and answer pair is input into the domain base model, and the model is trained for binary classification, so that the output of the intention recognition sub-model is a probability, which indicates whether the inspection question input by the user needs to call the knowledge graph.

[0065] Through the above method, the aforementioned intention recognition sub-model can be generated to judge the operation and inspection questions raised by the user to determine whether the operation and inspection questions need to call the knowledge graph.

[0066] Reference Figure 2 In some embodiments, in order to continue to optimize the aforementioned first interaction sub-model, the aforementioned industrial operation and inspection large model system also includes a first interaction sub-model adjustment module, which is specifically used to fine-tune the first interaction sub-model based on the third operation and inspection question and answer pair.

[0067] For the third operation and inspection question and answer pair, it specifically refers to the question and answer pair that needs to be optimized.

[0068] As a possible implementation method, the third operation-inspection question-answer pair may be constructed in the following manner: First, obtain the user's first feedback information, determine the target question based on the first feedback information, then obtain the operation and inspection question answer corresponding to the target operation and inspection question, and finally, construct a third operation and inspection question and answer pair based on the target question and the operation and inspection question answer corresponding to the target operation and inspection question.

[0069] Specifically, the target problem is the problem that needs to be corrected. For example, suppose the operation and maintenance problem input by the user is: device A overheated. At this time, the first interactive model outputs the answer to the operation and maintenance problem that device A overheated, and uses a pop-up dialog box to ask the user whether he is satisfied with the output answer. If the user outputs that he is not satisfied, then the overheating of device A is regarded as the target problem.

[0070] Then, the output answer to the overheating of device A can be modified manually, and the aforementioned third operation and inspection question and answer pair is generated in the above manner.

[0071] Through the above method, the aforementioned first interaction sub-model can be modified in time based on the user's actual first feedback information, thereby improving the accuracy and objectivity of the model output.

[0072] It should be noted that the second operation and inspection question and answer pair is used to fine-tune the domain base model to make the answers output by the first interactive sub-model more objective, while the third operation and inspection question and answer pair is mainly constructed based on real-time feedback from users and is used to correct operation and inspection answers that do not conform to reality.

[0073] In some embodiments, the first interactive sub-model outputs multiple answers to the operation and inspection questions. In order to enhance the objectivity of the answers to the operation and inspection questions output by the first interactive sub-model, avoid engineering personnel blindly checking the operation and inspection answers, and improve the targeted troubleshooting of operation and inspection questions, as a preferred embodiment, each of the answers to the operation and inspection questions is set with a first weight, and the first interactive sub-model outputs the answers to the operation and inspection questions in sequence according to the first weight corresponding to each of the answers to the operation and inspection questions.

[0074] Exemplarily, the answer to the operation and inspection problem output by the first interaction sub-model includes multiple causes of the operation and inspection problem, and each cause of the operation and inspection problem is set with a weight.

[0075] In this example, the first interaction sub-model outputs the causes of the operation and inspection problem in sequence according to the weight corresponding to each cause of the operation and inspection problem.

[0076] For example, the maintenance problem output by the user is: device A is overheating. In this situation, there may be multiple reasons causing device A to overheat. At this time, a weight is set for each reason causing device A to overheat, and the reason for device A overheating is output based on the set weight. Specifically, the weight can be output according to the probability of generating the maintenance reason.

[0077] In another exemplary embodiment, the answer to the operation and inspection problem output by the first interaction sub-model includes multiple fault handling methods for the operation and inspection problem, and each fault handling method for the operation and inspection problem is set with a weight.

[0078] In this example, the first interaction sub-model outputs the fault handling methods for the operation and inspection problems in sequence according to the weight corresponding to each fault handling method for the operation and inspection problems.

[0079] For example, the operation and maintenance problem output by the user is: device A is overheating. In this scenario, there are multiple solutions to deal with the overheating of device A according to the different causes of the overheating of device A. At this time, for each fault handling method for the operation and maintenance problem, the fault handling method for the overheating of device A is output according to the set weight.

[0080] In another exemplary embodiment, the answer to the operation and inspection problem output by the first interaction sub-model includes multiple maintenance strategies for the operation and inspection problem, and each maintenance strategy for the operation and inspection problem is set with a weight.

[0081] In this scenario, the first interactive sub-model outputs each maintenance strategy for the operation and inspection problem in turn according to the weight corresponding to each maintenance strategy for the operation and inspection problem.

[0082] It should be noted that in the above examples, the answers to the operation and inspection questions are all generated through the first interaction sub-model, and thus the above weights can all be regarded as the first weights.

[0083] In this embodiment, the first weight may be set specifically according to the priority of the answer to the operation inspection question. In other words, the first weight is positively correlated with the priority of the answer to the operation inspection question.

[0084] Specifically, the priority can be generated by technical personnel based on maintenance record statistics. Assuming that device A overheats, the causes of the operation and maintenance problem are a, b, and c. According to the maintenance history data, the probability that a causes device A to overheat is 70%, the probability that b causes device A to overheat is 20%, and the probability that c causes device A to overheat is 10%. The answers to the operation and maintenance problems are output in order of a, b, and c.

[0085] Through the above method, by setting weights to output answers to operation and maintenance questions, users can troubleshoot equipment in the order of output, improve the responsiveness to user needs, and make the output answers more meaningful for reference. Technical personnel can troubleshoot the system according to the order of output, thereby improving operation and maintenance efficiency.

[0086] The following will explain the second interaction sub-model mentioned above: In the embodiments provided in the present application, for the second interactive sub-model, a knowledge graph is specifically used for retrieval. In the embodiments provided in the present application, the aforementioned knowledge graph corresponds to a knowledge graph topology, and the knowledge graph topology includes: nodes and several edges, and the nodes include: at least one equipment node and at least one fault node. Specifically, continuing to take the stamping workshop as an example, the PDF documents, academic papers, professional books, equipment operation manuals, actual operation and inspection records, industry standards, etc. of various types of equipment on the pressing line are used as original data, and the information extraction algorithm is used to automatically extract key information such as equipment, fault type, and maintenance method from the documents. The extracted entities are used as nodes in the knowledge graph. Further, the nodes are divided into equipment nodes and fault nodes. Equipment nodes include equipment names, models, functional descriptions, etc., and fault nodes include common problems and their specific manifestations. In the knowledge graph, the extracted entities are used as nodes, and there will be mutual connections between nodes, which is called the relationship or edge between two nodes.

[0087] More specifically, with respect to the relationship between the device node and the fault node, the relationship between each device and its common fault type is established based on the maintenance records, so that the relationship between the device node and the fault node can be determined.

[0088] Regarding the relationship between device nodes, the order relationship between devices can be established based on physical connections or functional dependencies in the production process. For example, in a stamping workshop, the physical connection between the loader and the next-order molding equipment can construct the equipment relationship between the device node A corresponding to the loader and the molding equipment node B.

[0089] For a specific device, namely the first target device, the second interaction sub-model uses the following relationship to output the answer to the operation inspection question: In some embodiments, the answer to the operation and inspection question may be output based on the edge formed by the first target device node and the fault node.

[0090] In this embodiment, the dependency relationship between the corresponding device node and the faulty node is determined.

[0091] Assuming that device node A has a dependency relationship with faulty node B, when device A fails, the cause of the failure of device A can be output based on the dependency relationship between node A and node B.

[0092] In some other embodiments, in order to identify and analyze the possible impact on associated devices (such as devices B and C) when the first target device (such as device A) fails, the answer to the first inspection question is output based on the edge formed by the second target device, the first target device node A and the faulty node in the knowledge graph topology.

[0093] In this embodiment, the answer to the operation and inspection question includes: the cause of the failure of the operation and inspection question and the operation and inspection solution thereof provided by the second target device, the second target device being an associated device of the first target device, For example, taking a stamping workshop as an example, a stamping workshop has three devices A, B, and C. If A fails, the impact of A's associated devices on A can be determined through the relationship between the nodes of B and C devices and the node of A device.

[0094] Furthermore, assume that the three devices A, B, and C each have a set of nodes a, b, and c, and node a is connected to nodes b and c through edges. If node a is found, since nodes b and c are associated with node a, the answer to the inspection question is output through the edges between node a and node b, and the edges between node a and node c.

[0095] Through the above method, a complete knowledge graph can be constructed. When a device fails, not only can the current device be troubleshooted, but also the associated devices of the current device can be troubleshooted, making the results output by the second interactive sub-model more objective.

[0096] In order to realize the iteration and modification of the second interactive subsystem, as a preferred embodiment, the second interactive sub-model adjustment module first obtains the second feedback information, and then modifies the device nodes and / or faulty nodes in the knowledge graph topology based on the second feedback results.

[0097] Exemplarily, it is assumed that for each operation inspection question, the user will provide feedback on the generated answer to the operation inspection question, thereby generating second feedback information.

[0098] Through the second feedback information, the device nodes and / or faulty nodes can be modified, and then the knowledge graph can be iterated, so that the results output by the second interactive sub-model are more objective.

[0099] It should be noted that the difference between the first feedback information and the second feedback information is only the difference in context. The first feedback information provides feedback on the output result of the first interaction sub-model, while the second feedback information provides feedback on the output result of the second interaction sub-model.

[0100] In some embodiments, the first target device node and the multiple fault nodes respectively form multiple edges, and correspondingly output multiple answers to the operation and inspection questions, and each answer to the operation and inspection question is set with a second weight.

[0101] In this embodiment, the second interaction sub-model outputs answers to the operation inspection questions according to the second weight corresponding to each answer to the operation inspection question.

[0102] In some further embodiments, the first target device node forms multiple edges with multiple of the second target devices, and correspondingly outputs multiple answers to the operation and inspection questions. Each answer to the operation and inspection question is set with a third weight, and the second interaction sub-model outputs the answer to the operation and inspection question based on the weight corresponding to each answer to the operation and inspection question.

[0103] The second weight and the third weight may be set in accordance with the priority of the answer to each operation and inspection question. Specifically, the second weight and the third weight are positively correlated with the priority of the answer to the operation and inspection question.

[0104] Exemplarily, the priority can be set according to the probability of the cause of the fault. Exemplarily, assuming that a certain fault occurs in device A, it may be caused by water entering device A or power failure of device A. The probability of the fault occurring due to water entering device A is 60%, and the probability of the fault occurring due to power failure of device A is 40%. At this time, the output of the fault of water entering device A is prioritized, and then the output of the power failure of device A is given.

[0105] For another example, assume that a certain fault occurs in device A, which may be caused by water entering device B and power failure of device C. The probability of the fault occurring due to water entering device B is 60%, and the probability of the fault occurring due to power failure of device C is 40%. At this time, the output of the fault of water entering device B is prioritized, and then the output of the power failure of device C is given.

[0106] It should be noted that the first weight, the second weight and the third weight are only different in environment and scenario. Specifically, the weight corresponding to the answer to the operation and inspection question output by the first interactive sub-model is the first weight, the weight corresponding to the answer to the operation and inspection question output based on the edge formed by the first target device node and the fault node in the knowledge graph topology is the second weight, and the weight corresponding to the answer to the operation and inspection question output based on the edge formed by the second target device and the first target device node in the knowledge graph topology is the third weight.

[0107] Through the above method, the answers to operation and inspection questions can be output according to the probability of fault occurrence. Technical personnel can troubleshoot the system according to the output order, thereby improving the efficiency of operation and inspection.

[0108] Based on the above embodiments, the present application also provides a method for business operation inspection, which specifically includes the following steps: Reference Figure 4 , S1: Get the operation and inspection query statement.

[0109] For operation and inspection query statements, users can use external input devices to input query statements, for example, how to solve the problem of press overload: S2: Input the operation and inspection query statement into the industrial operation and inspection large model system.

[0110] For the industrial operation and maintenance large model system, please refer to the above for details. To avoid repetition, we will not go into details here.

[0111] In this embodiment, a preset question and answer library is used. If the question raised by the user is in the question and answer library, the answer is directly output using the question and answer library. If the question raised by the user is not in the question and answer library, the knowledge graph is used for retrieval. The above method can efficiently and accurately answer the user's operation and inspection questions, thereby improving the accuracy and objectivity of the model's conversation.

[0112] The flow charts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the system, method and computer program product according to multiple embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0113] Based on the above embodiments, the present application also provides an electronic device, Figure 5 A block diagram of an exemplary electronic device suitable for implementing an exemplary embodiment of the present invention is shown. Figure 5 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0114] like Figure 5 As shown, the electronic device is in the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: one or more processors 410, a memory 430, and a communication bus 440 connecting different system components (including the memory 430 and the processing unit 410).

[0115] The communication bus 440 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor or a local bus using any of a variety of bus structures. For example, these architectures include but are not limited to Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus, and Peripheral Component Interconnection (PCI) bus.

[0116] Electronic devices typically include a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device, including volatile and non-volatile media, removable and non-removable media.

[0117] The memory 430 may include a computer system readable medium in the form of a volatile memory, such as a random access memory (RAM) and / or a cache memory. The electronic device may further include other removable / non-removable, volatile / non-volatile computer system storage media. The memory 430 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of each embodiment of the present invention.

[0118] The processor 410 executes various functional applications and data processing by running the programs stored in the memory 430, such as implementing the embodiments of the present invention. Figure 4 The illustrated embodiment provides an industrial operation and inspection method.

[0119] The above describes specific embodiments of the present invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments 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 are also possible or may be advantageous.

[0120] In the description of the embodiments of the present invention, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiments of the present invention. In the embodiments of the present invention, the schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in the embodiments of the present invention and the features of the different embodiments or examples, without contradiction.

[0121] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the embodiments of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0122] The above description is only a preferred embodiment of the embodiment of the present invention and is not intended to limit the embodiment of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiment of the present invention should be included in the scope of protection of the embodiment of the present invention.

Claims

1. An industrial operation and inspection large model system, characterized in that: include: Intention recognition sub-model: Identify inspection questions based on inspection query statements and determine whether the inspection questions exist in the question and answer database; The first interactive sub-model is used to output answers to the operation inspection questions when the operation inspection questions exist in the question and answer database; The second interactive sub-model: when the operation inspection question does not exist in the question-answer database, the operation inspection question is searched in the knowledge graph and the answer to the operation inspection question is output; The first interaction sub-model is constructed in the following manner: Constructing an operation and inspection knowledge data set, pre-training a language autoregressive model based on the operation and inspection knowledge data set, and generating a domain base model, wherein the operation and inspection knowledge data set includes: equipment information in a target system, common faults of the target system, and at least one of a maintenance method of the target system; Extracting keywords from each paragraph in the operation and inspection knowledge data set, and constructing at least one first operation and inspection question-answer pair based on the keywords, each of the first operation and inspection question-answer pairs comprising: an operation and inspection question for a target system and a first output answer for the operation and inspection question, wherein the first output answer for the operation and inspection question comprises at least part of the keywords; Verifying the first operation and inspection question and answer pair, generating a second operation and inspection question and answer pair, and fine-tuning the domain base model based on the second operation and inspection question and answer pair; The question-and-answer database includes all the operation and inspection questions for the target system in the second operation and inspection question-and-answer pairs.

2. The system according to claim 1, characterized in that The industrial operation and inspection large model system also includes: a first interactive sub-model adjustment module; The first interaction sub-model adjustment module is used to fine-tune the first interaction sub-model based on the third operation-inspection question-answer pair; The third operation-inspection question-answer pair is constructed in the following manner: Acquire first feedback information from a user, and determine a target problem based on the first feedback information; Obtain answers to the operation and inspection questions corresponding to the target operation and inspection questions; The third operation and inspection question and answer pair is constructed based on the target question and the operation and inspection question answer corresponding to the target operation and inspection question.

3. The system according to claim 1, characterized in that The first interaction sub-model outputs a plurality of answers to the operation inspection question, and each of the answers to the operation inspection question is set with a first weight; First interactive sub-model: outputting the answers to the operation inspection questions in sequence according to the first weight corresponding to each answer to the operation inspection question.

4. The system according to claim 3, characterized in that The first weight is positively correlated with the priority of the answer to the inspection question.

5. The system according to any one of claims 1 to 4, characterized in that: The knowledge graph corresponds to a knowledge graph topology, and the knowledge graph topology includes: a node, and a plurality of edges, and the node includes: at least one device node and at least one fault node; For the first target device, the second interaction sub-model uses the following relationship to output answers to the operation inspection questions; Outputting the answer to the operation inspection question based on the edge formed by the first target device node and the fault node, wherein the edge is used to determine the dependency relationship between the corresponding device node and the fault node; and / or Based on the edge formed by the second target device and the first target device node in the knowledge graph topology, the answer to the operation and inspection question is output, the second target device is an associated device of the first target device, and the edge is used to determine the dependency relationship between the corresponding two device nodes.

6. The system according to claim 5, characterized in that The second interaction sub-model adjustment module: obtaining second feedback information; The device node and / or faulty node in the knowledge graph topology is modified based on the second feedback information.

7. The system according to claim 5, characterized in that In the step of outputting the answer to the operation inspection question based on the edge formed by the first target device node and the fault node, the first target device node and the plurality of fault nodes respectively form a plurality of edges, and a plurality of answers to the operation inspection question are outputted correspondingly, and each answer to the operation inspection question is provided with a second weight, The second interaction sub-model: outputting the answer to the operation inspection question according to the second weight corresponding to each answer to the operation inspection question; and / or In the step of outputting the answer to the operation and inspection question based on the edge formed by the second target device and the first target device node in the knowledge graph topology, the first target device node respectively forms multiple edges with multiple second target devices, and correspondingly outputs multiple answers to the operation and inspection question, each of which is set with a second weight. The second interaction sub-model outputs the answers to the operation and inspection questions according to the third weight corresponding to each answer to the operation and inspection question.

8. An industrial operation inspection method, characterized in that: The specific steps include: Get the operation and inspection query statement; The operation and inspection query statement is input into the industrial operation and inspection large model system according to any one of claims 1 to 7.

9. An electronic device, characterized in that: include: at least one processor; as well as at least one memory in communication with the processor, wherein: The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the method according to claim 8.

Images