A rail transit maintenance knowledge graph construction method, a terminal device, and a medium

Abstract

The application provides a rail transit maintenance knowledge graph construction method, a terminal device and a medium, comprising: acquiring structured data and unstructured data in the field of rail transit; performing block processing on the unstructured data to obtain a plurality of text blocks, and performing context feedback type information distillation operation on each text block by using a large language model to extract paragraph abstracts and keywords with confidence scores, and obtaining distilled text information; based on a predefined initial ontology layer, using a dynamic ontology-aware large language model extraction method to extract entity relationship triples from the distilled text information, and dynamically expanding the initial ontology layer during the extraction process; storing the extracted knowledge units in the form of triples into a graph database to complete the construction of the rail transit industry maintenance knowledge graph. The application can improve the accuracy of the rail transit maintenance knowledge graph.

Description

Technical Field

[0001] This invention relates to the fields of knowledge graph and artificial intelligence technology, specifically to a method for constructing a knowledge graph for rail transit maintenance, terminal equipment, and media. Background Technology

[0002] With the rapid development of the rail transit industry, the complexity of equipment inspection and maintenance is increasing. Knowledge graphs, as a structured knowledge representation method, can effectively integrate multi-source heterogeneous maintenance knowledge, providing important support for intelligent diagnosis, maintenance decision-making, and personnel training.

[0003] Currently, knowledge graph construction in the rail transit field typically involves extracting knowledge from structured and unstructured data, organizing the knowledge through a predefined ontology layer, and finally storing it in a graph database. However, existing methods suffer from significant technical bottlenecks: traditional text segmentation methods disrupt the inherent logical structure of documents, leading to semantic incompleteness during subsequent information extraction and affecting knowledge accuracy; predefined ontology layers struggle to cover all domain concepts and cannot adaptively expand when encountering undefined entities or relationships, resulting in incomplete knowledge extraction; information distillation and knowledge extraction are independent steps, and errors in the preceding steps can propagate and amplify to subsequent stages, impacting overall quality; furthermore, traditional methods lack effective feedback mechanisms, failing to dynamically optimize preprocessing based on extraction results during the construction process. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method, terminal equipment and medium for constructing a knowledge graph of rail transit maintenance, so as to improve the accuracy of the knowledge graph of rail transit maintenance.

[0005] In a first aspect, the present invention provides a method for constructing a knowledge graph for rail transit maintenance, the method comprising the following steps: Acquire structured and unstructured data in the rail transit sector; unstructured data includes production operation instructions and manufacturing instructions. Unstructured data is segmented to obtain multiple text blocks. A large language model is then used to perform context-feedback information distillation on each text block to extract paragraph summaries and keywords with confidence scores, thus obtaining distilled text information. Based on a predefined initial ontology layer, a dynamic ontology-aware large language model extraction method is adopted to extract entity relation triples from the distilled text information, and the initial ontology layer is dynamically expanded during the extraction process. The extracted knowledge units in the form of triples are stored in a graph database to complete the construction of a knowledge graph for maintenance in the rail transit industry.

[0006] Optional, context-feedback information distillation operations include: The semantic context of adjacent text blocks is dynamically obtained based on the sliding window mechanism; Generate paragraph summaries and keywords for the current text block; Calculate the semantic similarity between the summary of the current text block and the context of adjacent text blocks, and determine whether there is a semantic conflict based on a preset threshold; When a semantic conflict is detected, an iterative redistillation process is triggered until the conflict is resolved or the maximum number of iterations is reached. Output confidence scores for the finalized paragraph summaries and keywords.

[0007] Optionally, the sliding window mechanism uses a window size of N text blocks before and after the current text block, where N is a positive integer greater than or equal to 2.

[0008] Optionally, semantic similarity is the cosine similarity of the text embedding vectors.

[0009] Optionally, the method further includes an automatic verification step based on confidence scores, the automatic verification step including: For content with a confidence score lower than a preset threshold, a lightweight verification model finely tuned based on a rail transit terminology database is invoked for secondary processing. If the confidence score of the content after secondary processing is still lower than the preset threshold, it will be marked as an item requiring manual review.

[0010] Optional methods for extracting large language models with dynamic ontology awareness include: Based on the initial ontology layer, initial entity relation triples are extracted from unstructured data; During the extraction process, entities or relations not defined in the initial ontology layer are identified and recorded, and candidate triples are generated. A score is calculated for each candidate triplet based on a domain fit scoring model. Based on the score, the ontology layer expansion process is triggered, and candidate entities or relations with high fit are included in the initial ontology layer.

[0011] Optionally, the domain fit scoring model is calculated by fusing the frequency of candidate terms in the domain corpus and their co-occurrence with industry standard documents.

[0012] Optionally, the ontology layer extension process also includes: automatically generating definition suggestions for candidate entities or relationships to be extended, sorting them according to their domain suitability scores, and then pushing them to the expert review queue.

[0013] In a second aspect, the present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method.

[0014] Thirdly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0015] The present invention has at least the following beneficial effects: By performing context-feedback information distillation on each text block, the semantic fragmentation caused by text segmentation can be effectively solved, which is conducive to improving the accuracy of the rail transit maintenance knowledge graph. By adopting a dynamic ontology-aware large language model extraction method, entity relation triples are extracted from the distilled text information, and the initial ontology layer is dynamically expanded during the extraction process. This can adapt to the evolution of domain knowledge, improve knowledge coverage, and further improve the accuracy of the rail transit maintenance knowledge graph. Attached Figure Description

[0016] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0017] Figure 1 This is a flowchart of a method for constructing a knowledge graph for rail transit maintenance in one embodiment of this application; Figure 2 This is a schematic diagram of structured data in one embodiment of this application; Figure 3 This is a flowchart illustrating the operation of the intelligent question-answering method and digital human system in another embodiment of this application. Figure 4 This is a flowchart illustrating the operation of the digital human interaction front end in another embodiment of this application. Figure 5 This is a schematic diagram of the structure of a terminal device in another embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] To address the technical problem of low accuracy in rail transit maintenance knowledge graphs constructed using traditional methods, this invention provides a method, terminal device, and medium for constructing a rail transit maintenance knowledge graph. This method effectively solves the semantic fragmentation problem caused by text segmentation by performing context-feedback information distillation on each text block, thus improving the accuracy of the rail transit maintenance knowledge graph. Furthermore, it employs a dynamic ontology-aware large language model extraction method to extract entity relation triples from the distilled text information and dynamically expands the initial ontology layer during the extraction process. This adapts to the evolution of domain knowledge, improves knowledge coverage, and further enhances the accuracy of the rail transit maintenance knowledge graph.

[0020] Example 1 like Figure 1 As shown, the method for constructing a knowledge graph for rail transit maintenance provided by this invention includes the following steps: Step 11: Obtain structured and unstructured data in the field of rail transit.

[0021] In this embodiment of the invention, structured data can be extracted from the database of rail transit enterprises. This structured data includes equipment attribute tables (such as vehicle number, component model, and production date); maintenance work order tables (such as fault codes, maintenance measures, and work hour records); and standard specification tables (such as parameter thresholds in the "Technical Standard for Metro Vehicle Maintenance"). Figure 2 The document illustrates the planned repair, emergency repair, and minor repair records of an electric locomotive of a certain locomotive company in one embodiment.

[0022] Unstructured data includes production operation instructions (such as the "Daily Inspection Operation Manual for Metro Trains", which includes maintenance procedures and safety specifications) and manufacturing instructions (such as the "Bogie Manufacturing Process Instruction Manual", which includes component assembly procedures and technical parameters).

[0023] It should be noted that after obtaining structured and unstructured data, the data needs to be preprocessed. For example, the preprocessing operations include: (1) removing headers and footers and noise from scanned documents (using OpenCV image processing); identifying and separating table content (using the LayoutParser tool library) and converting the table data into structured records; deleting irrelevant characters (such as “□” and “※”) and retaining technical terms (such as “pantograph” and “brake disc”).

[0024] Step 12: Perform block processing on the unstructured data to obtain multiple text blocks, and use a large language model to perform context feedback information distillation on each text block to extract paragraph summaries and keywords with confidence scores, thereby obtaining distilled text information.

[0025] In this embodiment of the invention, the sliding window mechanism uses a window size of N text blocks before and after the current text block, where N is a positive integer greater than or equal to 2.

[0026] In this embodiment of the invention, the context feedback information distillation operation includes steps 12.1.1 to 12.1.5.

[0027] Step 12.1.1: Dynamically obtain the semantic context of adjacent text blocks of the current text block based on the sliding window mechanism.

[0028] Specifically, set a symmetrical window centered on the current text block, with a window size of K (usually K=2). Process the first... text blocks At that time, the system will automatically set the front and back sides. adjacent blocks (i.e., from) arrive The concatenation of text content creates a rich context, denoted as semantic context. .

[0029] Step 12.1.2: Generate paragraph summaries and keywords for the current text block.

[0030] To enhance the model's professionalism in the rail transit field, it is necessary to collect as many text fragments as possible (e.g., 1000) in this field, which should be annotated with summaries, keywords, and triplet knowledge by domain experts. Then, this data is used to perform supervised fine-tuning of the basic model to obtain the core model.

[0031] After obtaining the core model, a structured prompt word is constructed, instructing the core model to use the provided semantic context. Focus text blocks The text content is used to generate a concise paragraph summary and 3 to 5 core keywords.

[0032] Step 12.1.3: Calculate the semantic similarity between the summary of the current text block and the context of adjacent text blocks, and determine whether there is a semantic conflict based on a preset threshold.

[0033] Specifically, semantic similarity is the cosine similarity of the text embedding vectors. For example, calculating the current summary... Summary of its adjacent blocks , The cosine similarity between the two is used, and the minimum cosine similarity is taken as the consistency score. A similarity threshold is set (e.g., 0.75). If the consistency score is less than the similarity threshold, it is considered a semantic conflict.

[0034] Step 12.1.4: When a semantic conflict is detected, trigger the iterative redistillation process until the conflict is resolved or the maximum number of iterations is reached.

[0035] Specifically, at this point, the conflict feedback and the original context need to be reorganized into new prompts to correct the core model. This process can be iterated, up to a maximum of For example (This continues) until the conflict is resolved or the maximum number of iterations is reached.

[0036] Step 12.1.5: Output confidence scores for the finalized paragraph summaries and keywords.

[0037] Specifically, each distilled abstract and keyword is assigned a confidence score. This score is a weighted sum of the model's own probability and the degree of matching with domain terms.

[0038] Specifically, it can be calculated using formulas.

[0039] Receive confidence score ;in, This indicates that the model outputs the following when generating this content: The average probability reflects the model's own confidence level. This indicates the degree to which the generated content matches the pre-built "Rail Transit Maintenance Terminology Database". The more and more relevant the matched terms, the higher the score. These are the weighting coefficients, and .

[0040] Step 13: Based on the predefined initial ontology layer, the dynamic ontology-aware large language model extraction method is used to extract entity relation triples from the distilled text information, and the initial ontology layer is dynamically expanded during the extraction process.

[0041] In this embodiment of the invention, the dynamic ontology-aware large language model extraction method specifically includes steps 13.1.1 to 13.1.4.

[0042] Step 13.1.1: Based on the initial ontology layer, extract initial entity relation triples from unstructured data.

[0043] Specifically, a basic ontology architecture is pre-built in the graph database, defining core entity types (e.g., equipment, components, faults, operations, standards) and relation types (e.g., containing components, may cause, maintenance methods, conforms to standards). Then, extract prompt words, specifying the constraints as the predefined ontology. The text blocks that have undergone information distillation... The high-confidence keywords are input into the core model, which is then instructed to output a structured list of entity-relationship-entity triples.

[0044] Step 13.1.2: During the extraction process, identify and record entities or relations that are not defined in the initial ontology layer, and generate candidate triples.

[0045] In the triples output by the model, any triple containing an entity type or relation type that is not registered in the predefined ontology is identified and marked as a "candidate triple".

[0046] Step 13.1.3: Calculate a score for each candidate triple based on the domain fit scoring model.

[0047] Calculate a domain fit score for each candidate triple (especially its new relation or new entity). Specifically, this can be achieved using the following formula: The domain fit score is obtained; among which, This indicates the frequency of the candidate term in the domain's internal corpus. This indicates the strength of co-occurrence between the candidate term and sentences in industry-standard documents, which can be calculated using a pre-trained co-occurrence model. and Indicates the weighting coefficient; .

[0048] Step 13.1.4: Based on the score, trigger the ontology layer expansion process to include candidate entities or relations with high fit into the initial ontology layer.

[0049] Specifically, a high fit threshold is set (e.g., 0.8). When the score of a candidate triple exceeds this high fit threshold, the ontology expansion process is automatically triggered. High-scoring candidate terms and their definition suggestions automatically generated by the core model are sorted by score and enter the "expert review queue." Domain experts review these terms through a dedicated interface, and upon approval, the new terms (such as relation temperature tolerance ranges) and their definitions are automatically added to the ontology layer of the graph database.

[0050] In one feasible implementation, the ontology layer extension process further includes: automatically generating definition suggestions for candidate entities or relationships to be extended, sorting them according to their domain suitability scores, and then pushing them to the expert review queue.

[0051] Step 14: Store the extracted knowledge units in the form of triples into the graph database to complete the construction of the knowledge graph for maintenance in the rail transit industry.

[0052] Specifically, the final confirmed triplet data is persistently stored in the graph database using a batch import tool. Indexes are then created for entities and relations to support efficient and accurate graph traversal and complex queries.

[0053] Example 2 In this embodiment, after performing step 12.1.5 in embodiment 1, an automatic verification step based on confidence score is also included, which includes steps I to II.

[0054] Step I: For content with a confidence score lower than a preset threshold, a lightweight verification model based on a fine-tuned rail transit terminology database is invoked for secondary processing.

[0055] Specifically, the low-confidence content is further analyzed and corrected by a lightweight verification model. M_verify is trained to use standard terminology more readily. For example, a model with a larger number of parameters and faster inference speed is selected, and fine-tuned using a standard terminology library in the rail transit field (such as terminology definitions in TB / T and ISO relevant standards) to make it a "verification expert" of domain terminology, which is the lightweight verification model.

[0056] Step II: If the confidence score of the content after secondary processing is still lower than the preset threshold, then mark it as an item that requires manual review.

[0057] Specifically, if the new confidence score is still lower than the preset threshold after verification by the lightweight verification model, the content is marked as "requires manual review" and pushed to the expert review platform along with its context information.

[0058] Example 3 This embodiment provides an intelligent question-answering method and a digital human system. The system includes a digital human interactive front-end, an intent recognition module, a functional module 1 (for question-answering-assisted decision-making), a functional module 2 (front-end page operation), and a functional module 3 (report generation). Among them, functional module 1 relates to the rail transit maintenance knowledge graph construction method described in embodiments 1-2 above.

[0059] Specifically, the flowchart of the intelligent question-answering method and the digital human system is as follows: Figure 3 As shown, the user first enters the digital human module from the front-end page and inputs their needs via voice / text multimodal input. The digital human provides human-like services in the form of natural language dialogue. After obtaining the user's needs, the speech is converted into Chinese strings based on ASR technology. Then, based on a pre-trained large language model, the user's natural language input question is subjected to multi-dimensional intent determination, generating intent classification results including three major categories: "rail transit questions", "page operations", and "report generation".

[0060] If the user's intention is to inquire about rail transit issues, the system proceeds to the question-and-answer-assisted decision-making module. Once in the module, the rail transit maintenance knowledge graph construction method described in Examples 1-2 is utilized. The system employs entity extraction and vector retrieval to obtain relevant data support, and then generates a structured solution based on RAG technology, achieving end-to-end decision support from problem description to handling suggestions. If the user's intention is to perform a page operation, the system will perform the front-end page operation according to the user's intention, such as selecting a model, adding, deleting, querying, and modifying raw data, and redirecting to system pages. If the user's intent is to generate a report, the system initiates a collaborative workflow between the large model and the database. Data is retrieved by querying the business database, and then the large model generates the required report, such as a weekly or monthly report. Finally, after processing the three main categories of issues, the results are fed back to the digital human interaction front-end, which provides voice and text feedback to the user, while the digital human displays corresponding facial expressions and gestures.

[0061] In one feasible implementation, the digital human interaction front end consists of three parts: input, output, and digital human avatar. The input part can use voice input or text input; the output includes facial expression / action output, voice output, and text output. Voice input uses ASR technology, and voice output uses TTS technology. In this embodiment, the ASR algorithm uses FuASR developed by the DAMO Academy Speech Lab, and the voice output algorithm uses GPT-SoVITS; both algorithms are open source. The digital human avatar part can use AI-generated 2D or 3D avatars, or users can create their own 3D models according to specific needs. Users input a question via text or voice. If it's text input, the intent is directly passed to intent recognition and the corresponding question is solved. If it's voice input, the ASR algorithm converts the speech into text, then passes it to intent recognition and solves the corresponding question. After solving, the corresponding output characters are converted back to speech using a TTS algorithm, and then the digital human facial expression / action control algorithm generates corresponding facial expression / action videos based on the speech. Finally, the text and digital human avatar are displayed on a large screen, with the voice playing synchronously. The ASR algorithm, TTS algorithm, and digital human facial expression and motion generation involve a large amount of computation and require high-end hardware configuration. All of these processes are performed on the backend server. The frontend transmits data through API calls. The specific computing power of the server can be configured according to the specific number of paths the digital human is running on the frontend and the computing power of each path. Figure 4 A flowchart of the operation of the digital human interaction front end.

[0062] In one feasible implementation, a pre-trained large language model is used to perform multi-dimensional intent determination on the natural language questions input by the user. The specific implementation process includes: firstly, using a large language model (such as deepseek-671B) to receive the user's original question text and generate intent classification results including three major categories: "rail transit question", "page operation", and "report generation".

[0063] Example 4 like Figure 5 As shown, embodiments of the present invention provide a terminal device, such as... Figure 5 As shown, the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 5 The diagram shows only one processor, a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 executes the computer program D102 to implement the steps in any of the above method embodiments.

[0064] Specifically, when the processor D100 executes the computer program D102, it acquires structured and unstructured data in the rail transit field; it performs block processing on the unstructured data to obtain multiple text blocks, and uses a large language model to perform context-feedback information distillation on each text block to extract paragraph summaries and keywords with confidence scores, thus obtaining distilled text information; based on a predefined initial ontology layer, it uses a dynamic ontology-aware large language model extraction method to extract entity relation triples from the distilled text information, and dynamically expands the initial ontology layer during the extraction process; it stores the extracted triple-form knowledge units in a graph database to complete the construction of a maintenance knowledge graph for the rail transit industry. Among these methods, performing context-feedback information distillation on each text block effectively solves the semantic fragmentation problem caused by text segmentation, which is beneficial to improving the accuracy of the rail transit maintenance knowledge graph. The dynamic ontology-aware large language model extraction method extracts entity relation triples from the distilled text information and dynamically expands the initial ontology layer during the extraction process, which can adapt to the evolution of domain knowledge, improve knowledge coverage, and further improve the accuracy of the rail transit maintenance knowledge graph.

[0065] The processor D100 can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0066] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may be an external storage device of the terminal device D10, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device D10. Furthermore, the memory D101 may include both internal and external storage units of the terminal device D10. The memory D101 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory D101 can also be used to temporarily store data that has been output or will be output.

[0067] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0068] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments above.

[0069] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0070] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

Claims

1. A method for constructing a knowledge graph for rail transit maintenance, characterized in that, include: Acquire structured and unstructured data in the field of rail transit; wherein, the unstructured data includes production operation instructions and manufacturing instructions; The unstructured data is segmented to obtain multiple text blocks. A large language model is then used to perform context-feedback information distillation on each text block to extract paragraph summaries and keywords with confidence scores, resulting in distilled text information. The context-feedback information distillation operation includes: The semantic context of adjacent text blocks is dynamically obtained based on the sliding window mechanism; Generate a paragraph summary and keywords for the current text block; Calculate the semantic similarity between the summary of the current text block and the context of adjacent text blocks, and determine whether there is a semantic conflict based on a preset threshold; When a semantic conflict is detected, an iterative redistillation process is triggered until the conflict is resolved or the maximum number of iterations is reached. Output confidence scores for the finalized paragraph summaries and keywords; Based on a predefined initial ontology layer, a dynamic ontology-aware large language model extraction method is used to extract entity relation triples from the distilled text information, and the initial ontology layer is dynamically expanded during the extraction process; the dynamic ontology-aware large language model extraction method specifically includes: Based on the initial ontology layer, initial entity relation triples are extracted from the unstructured data; During the extraction process, entities or relations not defined in the initial ontology layer are identified and recorded, and candidate triples are generated. A score is calculated for each candidate triplet based on the domain fit scoring model. Based on the score, the ontology layer expansion process is triggered to include highly adaptable candidate entities or relationships into the initial ontology layer. The extracted knowledge units in the form of triples are stored in a graph database to complete the construction of a knowledge graph for maintenance in the rail transit industry.

2. The method for constructing a knowledge graph for rail transit maintenance according to claim 1, characterized in that, The sliding window mechanism uses a window size of N text blocks before and after the current text block, where N is a positive integer greater than or equal to 2.

3. The method for constructing a knowledge graph for rail transit maintenance according to claim 1, characterized in that, The semantic similarity is the cosine similarity of the text embedding vectors.

4. The method for constructing a knowledge graph for rail transit maintenance according to claim 1, characterized in that, The method further includes an automatic verification step based on confidence scores, the automatic verification step including: For content with a confidence score lower than a preset threshold, a lightweight verification model finely tuned based on a rail transit terminology database is invoked for secondary processing. If the confidence score of the content after secondary processing is still lower than the preset threshold, it will be marked as an item requiring manual review.

5. The method for constructing a knowledge graph for rail transit maintenance according to claim 1, characterized in that, The domain fit scoring model is calculated by fusing the frequency of candidate terms in the domain corpus and their co-occurrence with industry standard documents.

6. The method for constructing a knowledge graph for rail transit maintenance according to claim 1, characterized in that, The ontology layer extension process also includes: automatically generating definition suggestions for candidate entities or relationships to be extended, sorting them according to their domain suitability scores, and then pushing them to the expert review queue.

7. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 6.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 6.

Images