Method and device for determining entity in fault information, processor and electronic equipment

Abstract

The invention discloses a method and a device for determining an entity in fault information, a processor and electronic equipment. The method comprises the steps that target fault information is recognized from initial fault information of an engine; the target fault information is identified by using a first entity identification model to obtain a first entity, and the first entity identification model is obtained by training a deep learning model by using a target fault information sample and a corresponding first entity sample; the first entity is key information used for representing that the engine is in a fault state in the target fault information; in response to the fact that the entity number of the first entity meets an entity number threshold value, detecting the quality of the first entity by using a quality detection model to obtain a quality detection result; based on the quality detection result, a second entity is determined, and the quality of the second entity is higher than that of the first entity. The technical problem of low determination accuracy of entities in fault information is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault information extraction, and in particular to a method, device, processor and electronic device for determining an entity in fault information. Background Art

[0002] At present, with the development of technology, the structure and function of the engine are constantly improved and optimized. In addition to installing standard parts and structural parts, more and more functional control modules are installed for engine control and monitoring. The data collected by these modules and sensors can be further analyzed and used to provide information for engine fault diagnosis. Among them, quality data, including engine production data, maintenance data, usage data, physical data and performance data, are important data in the process of engine use and maintenance, and are of great significance to engine fault diagnosis and fault location. In the process of engine fault diagnosis, fault information is the key to fault location and fault prevention, that is, fault information is an important reference for fault diagnosis, and the cause and location of the fault can be analyzed based on the fault information to locate the fault.

[0003] In the related art, the fault diagnosis expert system can analyze the fault information to achieve accurate diagnosis and location of the engine fault. However, the expert system needs to determine the type of fault information and the corresponding data type in advance, which brings difficulties to the construction of the expert system and requires the participation of high-level experts, making the expert system construction time long, costly, and difficult to maintain. Therefore, there is still a technical problem of low accuracy in determining entities in fault information.

[0004] Currently, no effective solution has been proposed for the technical problem of low accuracy in determining entities in the above-mentioned fault information. Summary of the invention

[0005] Embodiments of the present invention provide a method, device, processor, and electronic device for determining an entity in fault information, so as to at least solve the technical problem of low accuracy in determining an entity in fault information.

[0006] According to one aspect of an embodiment of the present invention, a method for determining entities in fault information is provided. The method may include: identifying target fault information from initial fault information of an engine, wherein the initial fault information is unstructured fault information that does not meet a fixed format, and the target fault information is structured fault information that meets a fixed format; using a first entity recognition model, identifying the target fault information to obtain a first entity, wherein the first entity recognition model is obtained by training a deep learning model using a target fault information sample and a corresponding first entity sample, and the first entity is key information in the target fault information for characterizing that the engine is in a fault state; in response to the entity quantity of the first entity meeting an entity quantity threshold, using a quality detection model, detecting the quality of the first entity to obtain a quality detection result, wherein the quality detection model is obtained by training a large language model using the first entity sample and the corresponding quality detection result sample; based on the quality detection result, determining a second entity, wherein the quality of the second entity is higher than the quality of the first entity.

[0007] Optionally, the entity quantity threshold includes a first entity quantity threshold, wherein, in response to the entity quantity of the first entity satisfying the entity quantity threshold, using a quality detection model, detecting the quality of the first entity to obtain a quality detection result, including: a first detection step, in response to the entity quantity being equal to the first entity quantity threshold, using a quality detection model to detect the quality of the first entity to obtain a quality detection result; a second detection step, in response to the entity quantity being greater than the first entity quantity threshold, using a quality detection model to analyze the first entity through few-sample learning and prompt words to determine the quality detection result; executing the first detection step for the first entity that satisfies the first entity quantity threshold and corresponds to the maximum quality detection result.

[0008] Optionally, the entity quantity threshold includes a second entity quantity threshold, the second entity quantity threshold is less than the first entity quantity threshold, wherein, in response to the entity quantity of the first entity satisfying the entity quantity threshold, using a quality detection model, the quality of the first entity is detected to obtain a quality detection result, including: in response to the entity quantity being equal to the second entity quantity threshold, inputting the target fault information into a second entity recognition model, using the second entity recognition model, through few-sample learning and prompt words, identifying the first entity from the target fault information, wherein the second entity recognition model is obtained by training a large language model using fault information samples and first entity samples; in response to the entity quantity of the first entity identified from the second entity recognition model being equal to the first entity quantity threshold, executing the first detection step; or, in response to the entity quantity being greater than the first entity quantity threshold, executing the second detection step.

[0009] Optionally, in response to the entity quantity of the first entity satisfying an entity quantity threshold, a quality detection model is used to detect the quality of the first entity to obtain a quality detection result, including: in response to the entity quantity satisfying the entity quantity threshold, a quality detection model is used to analyze the quality of the first entity through few-sample learning and prompt words to determine the quality detection result.

[0010] Optionally, determining the second entity based on the quality detection result includes: in response to the quality detection result being greater than or equal to a quality detection result threshold, determining the first entity as the second entity.

[0011] Optionally, the method also includes: an identification step, in response to the quality detection result being less than a quality detection result threshold, inputting the target fault information into a second entity recognition model, using the second entity recognition model, through few-sample learning and prompt words, identifying the first entity from the target fault information, wherein the second entity recognition model is obtained by training a large language model using fault information samples and first entity samples; inputting the first entity identified in the second entity recognition model into the quality detection model, and determining the quality detection result of the first entity identified from the second recognition model through few-sample learning and prompt words; in response to the quality detection result being less than the quality detection result threshold, repeating the identification step until the quality detection result is greater than or equal to the quality detection result threshold.

[0012] Optionally, identifying target fault information from initial fault information of the engine includes: acquiring engine quality data from the initial fault information; acquiring fault fields from the engine quality data, and aggregating the fault fields into target fault information.

[0013] Optionally, the method also includes: obtaining target fault information samples, and labeling key information in the target fault information samples to obtain first entity samples; respectively extracting target fault information samples and first entity samples with target proportions from target fault information samples and first entity samples of different fault types as training data sets for training deep learning models.

[0014] According to another aspect of an embodiment of the present invention, a device for determining entities in fault information is also provided. The device may include: a first identification unit, used to identify target fault information from initial fault information of an engine, wherein the initial fault information is unstructured fault information that does not meet a fixed format, and the target fault information is structured fault information that meets a fixed format; a second identification unit, used to identify the target fault information using a first entity recognition model to obtain a first entity, wherein the first entity recognition model is obtained by training a deep learning model using a target fault information sample and a corresponding first entity sample, and the first entity is key information in the target fault information for characterizing that the engine is in a fault state; a detection unit, used to detect the quality of the first entity using a quality detection model in response to the entity quantity of the first entity meeting an entity quantity threshold, and obtain a quality detection result, wherein the quality detection model is obtained by training a large language model using a first entity sample and a corresponding quality detection result sample; a determination unit, used to determine a second entity based on the quality detection result, wherein the quality of the second entity is higher than the quality of the first entity.

[0015] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is also provided. The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method for determining entities in fault information of an embodiment of the present invention.

[0016] According to another aspect of an embodiment of the present invention, a processor is provided, which is used to run a program, wherein the method for determining an entity in fault information of an embodiment of the present invention is executed when the program is run.

[0017] According to another aspect of an embodiment of the present invention, a computer program product is provided, which includes a computer program, and when the computer program is executed by a processor, the method for determining an entity in fault information of the embodiment of the present application is implemented.

[0018] In an embodiment of the present invention, if it is necessary to determine an entity with qualified quality from an unstructured initial fault information that does not meet a fixed format, the target fault information that is structured and meets a fixed format can be first analyzed from the initial fault information. The target fault information can be input into a pre-trained first entity recognition model to identify the target fault information and obtain a first entity. If the number of entities of the first entity meets the entity number threshold, the quality of the first entity can be detected using a quality detection model to obtain a quality detection result. Thus, according to the quality detection result, a second entity with higher quality can be determined. In this embodiment, through structured information processing, targeted training of deep learning models, entity number threshold screening, quality detection of large language models, and iterative optimization mechanisms, the method significantly improves the accuracy of determining entities in fault information, reduces misidentification and missed identification, and provides more reliable and efficient information support for fault analysis, diagnosis, and maintenance decisions. Through the close cooperation of these steps, even complex unstructured fault descriptions can be accurately and quickly converted into structured and high-quality entity information, thereby improving the efficiency and accuracy of the entire fault information processing process. Thereby, the technical effect of improving the accuracy of determining entities in fault information is achieved, and the technical problem of low accuracy of determining entities in fault information is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0020] Figure 1 is a flow chart of a method for determining an entity in fault information according to an embodiment of the present invention;

[0021] Figure 2 is a flow chart of a method for extracting fault information based on engine quality data according to an embodiment of the present invention;

[0022] Figure 3 It is a schematic diagram of a device for determining an entity in fault information according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only 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 should fall within the scope of protection of the present invention.

[0024] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0025] According to an embodiment of the present invention, an embodiment of a method for determining an entity in fault information is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0026] Figure 1 is a flow chart of a method for determining an entity in fault information according to an embodiment of the present invention. Figure 1 As shown, the method may include the following steps:

[0027] Step S102, identifying target fault information from initial fault information of the engine.

[0028] In the technical solution provided in the above step S102 of the present invention, the initial fault information is unstructured fault information that does not meet the fixed format, that is, the initial fault information is unstructured fault information. The target fault information is structured fault information that meets the fixed format, that is, the target fault information is a structured fault record. The initial fault information is usually in free text form, without a fixed format or structure, and may contain a lot of noise and useless information, such as non-professional descriptions of technicians, colloquial expressions, irrelevant environmental descriptions, etc., which are only used as examples and are not specifically limited. Since the information comes from different recording methods or sources, the formats are different, and there may be a lack of consistent labels or classifications, making direct processing and analysis difficult. The target fault information can be organized and stored according to a predefined structure or format, for example, using a table form, each column corresponds to a different attribute of the fault (such as fault location, fault form, fault description, etc.). Each fault information in the target fault information should conform to a specific template or format to facilitate computer system recognition and processing. Structured fault information usually includes fixed fields, such as fault code, fault description, maintenance time, etc., to ensure data consistency and readability.

[0029] In this embodiment, if it is necessary to identify an entity from the engine fault information, target fault information can be identified from the initial fault information of the engine.

[0030] Optionally, the unstructured fault information is converted into a structured fault record, thereby providing input data with a uniform format and clear information for the subsequent entity recognition step.

[0031] In the embodiment of the present invention, the unstructured initial fault information is converted into structured target fault information, which not only improves the readability and usability of the information, but also provides high-quality input data for the subsequent entity recognition step, which is the cornerstone for accurate and efficient extraction of fault information. The generation of structured fault information enables the entity recognition model to more accurately identify and classify entities, thereby improving the automation and efficiency of the entire fault analysis process.

[0032] Step S104: using the first entity recognition model, identify the target fault information to obtain a first entity.

[0033] In the technical solution provided in the above step S104 of the present invention, the first entity recognition model can be obtained by training a deep learning model using target fault information samples and corresponding first entity samples. The deep learning model may include a bidirectional transformer encoder (Bidirectional Encoder Representations from Transformers, referred to as BERT) and a conditional random field (Conditional Random Field, referred to as CRF) model. The first entity can be an extraction result corresponding to an entity extracted from the target fault information by the above-mentioned first entity recognition model. The first entity can be key information in the target fault information for characterizing that the engine is in a fault state.

[0034] In this embodiment, after the target fault information is identified from the initial fault information, the target fault information may be identified using a first entity recognition model to obtain a first entity.

[0035] Optionally, the first entity recognition model is used to perform deep learning analysis on the target fault information structured in the previous step to identify entities related to the fault.

[0036] Optionally, before performing entity recognition, the first entity recognition model needs to be fully trained by a deep learning model (such as BERT+CRF). This process usually involves the following steps: Use the target fault information sample and the corresponding known entity (first entity sample) as a training data set. These samples include structured fault records, each of which is labeled with the actual fault entity, such as the fault location, fault form, etc. Build a deep learning model that combines BERT and CRF. BERT is used to generate high-quality word vectors for fault descriptions to capture context; CRF is used to predict the label of each position in the sequence based on the word vector generated by BERT to determine the boundary and type of the entity. The BERT+CRF model is trained using the training data set, with the goal of minimizing the difference between the predicted label and the actual label, that is, optimizing the entity recognition accuracy of the model.

[0037] Optionally, once the model training is completed, the target fault information (i.e., structured fault records) can be input into the model to identify the entities therein. The output of the model is the first entity, which is the key information in the fault information used to characterize the engine in a faulty state, such as the fault location, fault form, fault description, etc. In the BERT part, the description of the input structured fault information is encoded to generate a context-sensitive word vector representation, which is the basis for the model to understand the text context. In the CRF part, on the BERT-encoded word vector, the CRF model predicts the label of each word and determines the boundary and type of the entity. CRF can consider the dependencies between labels in the sequence, thereby making globally appropriate label sequence predictions. The above model outputs a series of labels that correspond to the words in the input fault information. The labels can determine which words are combined to form entities and the types of entities.

[0038] In an embodiment of the present invention, by using a first entity recognition model that combines BERT and CRF, this step can efficiently and accurately identify key entities in fault information. BERT's contextual understanding ability ensures that the model can capture the semantic details in the fault description, while CRF's sequence modeling ability further improves the accuracy of entity recognition and ensures the correctness of entity boundaries and types. This process reduces the errors and inefficiencies of manual recognition and provides a reliable data foundation for subsequent quality inspection and optimization. The use of the first entity recognition model realizes the automatic extraction of entities from structured fault descriptions, which not only speeds up the processing speed of fault information, but also improves the accuracy of entity recognition. It is one of the key steps in improving the quality and efficiency of fault information processing in the present invention. Through the application of this model, complex information in unstructured data can be converted into structured and analyzable entity information, providing valuable data support for subsequent fault analysis, intelligent diagnosis and maintenance decisions.

[0039] Step S106 , in response to the entity quantity of the first entity satisfying the entity quantity threshold, using a quality detection model to detect the quality of the first entity to obtain a quality detection result.

[0040] In the technical solution of step S106 of the present invention, the entity quantity threshold may be a preset entity quantity limit of the first entity, for example, it may be preset to 0 or 1. The quality detection model may be obtained by training a large language model using the first entity sample and the corresponding quality detection result sample, that is, it may be a recognition quality detection model based on the large language model. The quality detection result may be a score or quality corresponding to the first entity.

[0041] In this embodiment, after the target fault information is identified by using the first entity recognition model and the first entity is obtained, it can be determined whether the entity quantity of the first entity meets the entity quantity threshold. If the entity quantity threshold is met, the quality of the first entity can be detected by using a pre-trained quality detection model to obtain a quality detection result.

[0042] Optionally, the number of entities output by the first entity recognition model is checked by setting a threshold of the number of entities, and the recognition quality of these entities is further evaluated using a quality detection model.

[0043] Optionally, after the entity recognition model outputs the first entity, it is first necessary to check whether the number of entities outputted meets a preset entity number threshold. This threshold can be set based on the characteristics of the fault information and historical data statistics. For example, a fault record usually contains 1 fault location and 1 or more fault forms. Therefore, the threshold can be set to 1 or 2. If the number of entities output by the entity recognition model is 0, it may mean that the model has failed to identify any relevant fault entities. At this time, remedial measures need to be taken, such as trying to identify again through an entity recognition model based on a large language model. If the number of entities exceeds 1 but is lower than the threshold, the entity directly enters the optimizer for subsequent processing. However, if the number of entities is equal to or exceeds the threshold, a more detailed entity quality detection is required.

[0044] Optionally, when the number of entities of the first entity meets the entity number threshold, the next step is to evaluate the recognition quality of these entities using the quality detection model. The quality detection model receives the entities output by the first entity recognition model as input, and scores the recognition quality of each entity based on the training of the model and the understanding of the fault information. This scoring process may involve multiple dimensions, including the contextual consistency of the entity, the accuracy of the entity type, the certainty of the entity boundary, etc. If the entity recognition result is marked as "good", that is, the quality detection result shows that the entity recognition quality meets the expected standard, then these first entities will be confirmed as the final output entities. However, if the recognition result is marked as "poor", it means that the quality of the recognized entity is insufficient and further measures need to be taken to optimize the results.

[0045] In an embodiment of the present invention, the rationality and accuracy of the entity recognition results are ensured by setting the entity quantity threshold and applying the quality detection model. The check of the entity quantity threshold avoids the model from misreading or missing the fault information, while the quality detection model further improves the accuracy and reliability of the recognition results. This quality control mechanism can effectively reduce errors in later fault analysis and processing, and improve work efficiency and accuracy. Through the above method, the output of the entity recognition model can be effectively screened and optimized to ensure that the final entity information is of high quality and high reliability, providing a solid data foundation for subsequent fault analysis, pattern recognition, and intelligent diagnosis.

[0046] Step S108: determining the second entity based on the quality detection result.

[0047] In the technical solution provided in the above step S108 of the present invention, the quality of the second entity is higher than the quality of the first entity, that is, the second entity can be marked as "good" as a result and serve as the final output of the entity extraction method.

[0048] In this embodiment, after the quality detection result is used to detect the quality of the first entity and the quality detection result is obtained, the second entity may be determined based on the quality detection result.

[0049] Optionally, the above process is the quality control and result confirmation stage in the fault information extraction process based on engine quality data, and its main purpose is to screen and determine the final high-quality entity, i.e., the second entity, based on the output results of the quality detection model in the previous step.

[0050] Optionally, the quality detection model performs a quality assessment on the entity output by the first entity recognition model, and the obtained quality detection result can be a quality score or a good or bad mark. These results directly reflect whether the recognized entity is accurate, complete and can meet the requirements of subsequent fault analysis.

[0051] Optionally, the second entity is a set of entities that are screened out from the first entity and meet higher quality standards. Specifically, if the quality inspection result marks the entity as "good", it means that the quality of these entities has reached the preset standard and can be used as the final output entity, that is, the second entity. These entities have the following characteristics: Contextual consistency: The entity can accurately reflect its contextual meaning in the fault description without being separated from the text context. Entity type accuracy: The type of each entity (for example, fault location, fault form, etc.) is correctly identified without type confusion or error. Entity boundary certainty: The boundaries of the entity are accurately identified, and there are no missing or redundant parts of the entity.

[0052] Optionally, after the quality inspection model evaluates the recognition result of the first entity, if the result is marked as "good", these entities are directly used as the second entity and no longer need further processing. They are directly output and become the final result of the entity extraction method. This means that the entity information that has passed the two processes of the entity recognition model and the quality inspection model is reliable and can be directly used for subsequent tasks such as fault analysis, pattern recognition, and intelligent diagnosis. If the recognition result of the first entity is marked as "poor", that is, the quality inspection result shows that the quality of entity recognition is insufficient, then these entities will not be directly confirmed as the second entity. In this case, it is necessary to return to the entity recognition process and re-identify it through a more sophisticated model or algorithm until an entity result with a quality marked as "good" is obtained before it can be confirmed as the second entity output.

[0053] In the embodiment of the present invention, a strict quality control mechanism is used to ensure high-quality output of fault information extraction. This process effectively filters out low-quality or erroneous entity recognition results, avoids inaccurate data from entering the subsequent fault analysis process, and thus improves the work efficiency and analysis accuracy of the entire system. As a result, the second entity, as the final output, can provide a solid data foundation for intelligent diagnosis, prediction, and maintenance strategy formulation of faults. Through the above method, the accuracy and reliability of fault information can be ensured, which is of great significance for accurate fault location, rapid maintenance decision-making, and improving the overall operation and maintenance efficiency of the engine.

[0054] In the above steps S102 to S108 of the present application, if it is necessary to determine the qualified entities from the unstructured initial fault information that does not meet the fixed format, the structured target fault information that meets the fixed format can be first analyzed from the initial fault information. The target fault information can be input into the pre-trained first entity recognition model to identify the target fault information and obtain the first entity. If the number of entities of the first entity meets the entity number threshold, the quality of the first entity can be detected by the quality detection model to obtain the quality detection result. Thus, according to the quality detection result, a second entity with higher quality can be determined. In this embodiment, through structured information processing, targeted training of deep learning models, entity number threshold screening, quality detection of large language models and iterative optimization mechanism, the method significantly improves the accuracy of determining entities in fault information, reduces the situation of misidentification and missed identification, and provides more reliable and efficient information support for fault analysis, diagnosis and maintenance decisions. Through the close cooperation of these steps, even complex unstructured fault descriptions can be accurately and quickly converted into structured and high-quality entity information, thereby improving the efficiency and accuracy of the entire fault information processing process. Thereby, the technical effect of improving the accuracy of determining entities in fault information is achieved, and the technical problem of low accuracy of determining entities in fault information is solved.

[0055] The above method of this embodiment is further introduced below.

[0056] As an optional implementation method, the entity quantity threshold includes a first entity quantity threshold, wherein step S106, in response to the entity quantity of the first entity satisfying the entity quantity threshold, using a quality detection model to detect the quality of the first entity to obtain a quality detection result, includes: a first detection step, in response to the entity quantity being equal to the first entity quantity threshold, using a quality detection model to detect the quality of the first entity to obtain a quality detection result; a second detection step, in response to the entity quantity being greater than the first entity quantity threshold, using a quality detection model to analyze the first entity through few-sample learning and prompt words to determine the quality detection result; executing the first detection step for the first entity that satisfies the first entity quantity threshold and corresponds to the maximum quality detection result.

[0057] In this embodiment, in the process of detecting the quality of the first entity using the quality detection model, if the number of entities is equal to the first entity number threshold, the first detection step can be performed. In the first detection step, the quality detection model can be used to detect the quality of the first entity to obtain a quality detection result. If the number of entities is greater than the first entity number threshold, the second detection step can be performed. In the second detection step, the quality detection model can be used to analyze the first entity through few-sample learning and prompt words to determine the quality detection result. The first detection step is performed for the first entity that meets the first entity number threshold and corresponds to the maximum quality detection result. Among them, the entity number threshold may include the first entity number threshold. The first entity number threshold can be pre-set to 1.

[0058] Optionally, the entity quantity threshold includes a first entity quantity threshold. This step particularly refines the quality inspection process after entity recognition, and adopts different processing strategies based on the difference in entity quantity to ensure the accuracy and reliability of the final output entity information - the second entity.

[0059] Optionally, the first entity quantity threshold is set to 1, which means that in most fault records, we expect to identify a main fault entity, such as a key fault location or fault form.

[0060] Optionally, in the first detection step, when the number of entities output by the entity recognition model is exactly equal to the first entity number threshold (ie, equal to 1), the system will directly enter the first detection step. In this step, the quality detection model performs quality detection on this uniquely identified entity. The purpose of quality detection is to verify whether the entity is accurate, complete and reasonable, that is, whether it truly represents the engine fault information. If the first entity passes the quality detection, that is, the quality detection result is marked as "good", then the entity will be confirmed as the second entity, directly output and used for subsequent fault analysis; if the detection result is marked as "bad", it is necessary to return to the entity recognition step for re-identification or correction.

[0061] Optionally, in the second detection step, when the number of entities output by the entity recognition model exceeds the first entity number threshold (i.e., greater than 1), the system will enter the second detection step. This usually occurs when the fault information contains multiple possible fault entities, or when a false positive occurs during the model recognition process. In the second detection step, the quality detection model will conduct a detailed analysis of multiple first entities through few-shot learning and cue word engineering to determine their quality. Few-shot learning allows the model to quickly learn and adapt to new tasks based on limited examples, while cue word engineering guides the model to focus on key information through specific text structures to improve recognition accuracy.

[0062] Optionally, the quality detection model can assign a quality score to each entity, which reflects the accuracy and relevance of the entity. The system retains the entity with the highest score and treats it as the final second entity. This is because, in the case of multi-entity recognition, the entities that are most likely to reflect real fault information are those with the highest quality scores. If the quality scores of multiple entities are similar, further manual review or model optimization may be required to improve the certainty of recognition.

[0063] In an embodiment of the present invention, by implementing the first and second detection steps, it is ensured that the quality control of the entity recognition results is more refined and flexible. The first detection step simplifies the process and directly performs quality inspection on a single entity, while the second detection step improves the decision-making efficiency and accuracy in the case of multiple entity recognition through few-sample learning and prompt word engineering. The above-mentioned differentiated quality detection strategy based on the number of entities can effectively cope with the complexity of different fault information, improve the degree of automation and accuracy of fault information processing, reduce errors and confusion that may be encountered in subsequent fault analysis, thereby improving the efficiency and reliability of the entire fault handling process. The second entity finally output as high-quality fault information can provide more accurate data support for fault pattern recognition, intelligent diagnosis, etc., and is an important part of the present invention to enhance the fault information processing capability.

[0064] As an optional implementation method, the entity quantity threshold includes a second entity quantity threshold, and the second entity quantity threshold is less than the first entity quantity threshold, wherein step S106, in response to the entity quantity of the first entity satisfying the entity quantity threshold, using the quality detection model, detects the quality of the first entity to obtain a quality detection result, including: in response to the entity quantity being equal to the second entity quantity threshold, inputting the target fault information into the second entity recognition model, and using the second entity recognition model to identify the first entity from the target fault information through few-sample learning and prompt words, wherein the second entity recognition model is obtained by training a large language model using fault information samples and first entity samples; in response to the entity quantity of the first entity identified from the second entity recognition model being equal to the first entity quantity threshold, executing the first detection step; or, in response to the entity quantity being greater than the first entity quantity threshold, executing the second detection step.

[0065] In this embodiment, in the process of detecting the quality of the first entity using the quality detection model, if the number of entities is equal to the second number of entities, the target fault information can be input into the second entity recognition model, and the second entity recognition model can be used to identify the first entity from the target fault information through few-sample learning and prompt words. If the number of entities of the first entity identified from the second entity recognition model is equal to the first entity number threshold, the first detection step is performed. If the number of entities of the first entity identified from the second entity recognition model is greater than the first entity number threshold, the second detection step can be performed. Among them, the entity number threshold may include a second entity number threshold. The second entity number threshold is less than the first entity number threshold. For example, if the first entity number threshold is 1, the second entity number threshold can be 0. The second entity recognition model can be obtained by training a large language model using fault information samples and first entity samples, that is, an entity recognition model based on a large language model.

[0066] Optionally, the entity quantity threshold is further subdivided, and a second entity quantity threshold is introduced, which is smaller than the first entity quantity threshold. This strategy aims to optimize the accuracy and efficiency of entity recognition in fault information.

[0067] Optionally, the second entity number threshold is usually set to 0, which means that in some cases, the entity recognition model may fail to recognize any entities from the fault information. This may be due to unclear expression of the fault information, insufficient model training, or there are indeed no clear entities to be identified in the fault information. When the number of entities output by the entity recognition model is equal to the second entity number threshold (that is, 0), it indicates that additional attempts are needed to identify the fault entity. At this time, the target fault information is input into the second entity recognition model, which is an entity recognition model based on a large language model and is trained by Few-shot Learning and prompt word engineering. Few-shot Learning allows the model to quickly learn and adapt on a small number of training samples, improving recognition capabilities when the amount of data is limited; prompt word engineering guides the model to more accurately recognize entities by constructing a specific text structure.

[0068] Optionally, the second entity recognition model attempts to identify the first entity from the target fault information by means of few-sample learning and prompt word engineering. The operation of this stage of the model is essentially trying to make up for the situation where the first entity recognition model fails to recognize any entity, and to identify the key entities in the fault information through a more intelligent and more scenario-adaptive processing method. Once the second entity recognition model successfully identifies the first entity, regardless of the number of entities, the quality of the entity needs to be further checked to determine whether it meets the expectations of the first entity number threshold. If the number of entities of the first entity identified from the second entity recognition model is equal to the first entity number threshold (usually 1), the first detection step is directly executed, and the quality detection model performs quality detection on this entity. If the number of entities is greater than the first entity number threshold, a more complex second detection step is executed, and the entities are further analyzed and screened by the quality detection model to determine the final second entity.

[0069] In an embodiment of the present invention, the above-mentioned dual entity recognition and quality detection process can effectively cope with the challenges of entity recognition in different situations. First, through the intervention of the second entity recognition model, the processing gap when the first entity recognition model fails to recognize the entity is solved, and the integrity and reliability of the recognition are improved. Secondly, through the refined second entity quantity threshold and the first entity quantity threshold, the recognition strategy can be adjusted more flexibly to ensure that high-quality second entities can be output under any circumstances. Finally, the second entity processed by this optimization process, as the final output of fault information extraction, can provide more accurate and detailed data support for fault analysis, pattern recognition and intelligent diagnosis, and significantly improve the efficiency and accuracy of fault handling. The above embodiment fully demonstrates the flexibility and efficiency of the present invention in processing complex engine quality data and fault information, and is one of the important strategies to improve the accuracy and practicality of entity recognition technology in practical applications.

[0070] As an optional implementation method, step S106, in response to the entity quantity of the first entity satisfying the entity quantity threshold, using the quality detection model, detects the quality of the first entity to obtain a quality detection result, including: in response to the entity quantity satisfying the entity quantity threshold, using the quality detection model, through few-sample learning and prompt words, analyzing the quality of the first entity to determine the quality detection result.

[0071] In this embodiment, in the process of detecting the quality of the first entity using the quality detection model, if the number of entities meets the entity number threshold, the quality detection model can be used to analyze the quality of the first entity through few-sample learning and prompt words to determine the quality detection result.

[0072] Optionally, when the number of entities output by the entity recognition model meets the entity number threshold, this means that the number of recognized entities is within a preset reasonable range, neither too many nor too few. At this time, it is necessary to further verify whether the quality of these entities meets the expected standards.

[0073] Optionally, the quality detection model uses a strategy of few-shot learning and cue word engineering to analyze the quality of the first entity. Few-shot learning allows the model to learn quickly on a very small number of training samples, which is particularly useful in quality detection because high-quality entity samples may be relatively few, but the model needs to be able to quickly grasp the criteria for judging entity quality from these samples. Cue word engineering effectively guides the model to focus on key elements in entity recognition, such as entity boundaries, entity types, contextual relevance, etc., by designing specific text cues, thereby improving the accuracy of judging entity quality.

[0074] Optionally, for each first entity identified, the quality detection model performs a detailed analysis of the entity based on the entity quality standards learned during its training. The analysis involves multiple dimensions such as the entity's contextual consistency, entity type accuracy, and entity boundary certainty. Through few-shot learning and cue word engineering, the model can more accurately identify whether these attributes of the entity meet the standards of a high-quality entity, thereby giving a quality detection result.

[0075] Optionally, the quality inspection result is usually a quality score, or a good or bad mark. If the quality score of the entity reaches the set threshold, or is marked as "good", these entities are regarded as qualified second entities and can be directly used as the final output for subsequent fault analysis, intelligent diagnosis and other links. On the contrary, if the quality score of the entity is lower than the threshold, or is marked as "bad", it may be necessary to re-identify or manually review to ensure the accuracy and reliability of the fault information.

[0076] In the embodiment of the present invention, through the quality detection process of few-sample learning and prompt word engineering, efficient evaluation of entity recognition quality can be achieved with less training data. This mechanism not only reduces the dependence on a large amount of labeled data, but also improves the adaptability and flexibility of the model in entity recognition tasks. The second entity finally output, as a result of quality control, has higher accuracy and reliability, provides a solid data foundation for subsequent fault handling processes, and helps to improve the operating efficiency and decision-making quality of the overall system.

[0077] As an optional implementation manner, step S108, determining the second entity based on the quality detection result, includes: in response to the quality detection result being greater than or equal to a quality detection result threshold, determining the first entity as the second entity.

[0078] In this embodiment, in the process of determining the second entity based on the quality detection result, the first entity whose quality detection result is greater than the quality detection result threshold can be determined as the second entity. The quality detection result threshold can be used to measure the quality of the first entity. If the quality detection result is greater than or equal to the quality detection result threshold, the first entity is marked as "good".

[0079] Optionally, the process of determining the second entity based on the quality detection result, that is, those first entities that have been verified by the quality detection model and meet or exceed the preset quality detection result threshold will be directly confirmed as the second entity as the final output of fault information extraction.

[0080] Optionally, the quality detection result threshold is a pre-set standard for determining whether the quality of entity recognition reaches a desired level. When the first entity is evaluated by the quality detection model and the score of its quality detection result is greater than or equal to the threshold, it indicates that the quality of these entities is reliable and can accurately reflect the key features of the engine fault information.

[0081] Optionally, after the entity recognition model outputs the first entity, the quality detection model will evaluate the quality of these entities and give a quantitative quality detection result, which can be in the form of a score, representing the accuracy and completeness of entity recognition. If the evaluation results of one or more first entities are higher than or equal to the quality detection result threshold, then these entities will be regarded as qualified entities and will be directly confirmed as second entities as the final fault information extraction result.

[0082] In an embodiment of the present invention, by setting and applying a threshold value for quality detection results, high accuracy and reliability of fault information extraction are ensured. This mechanism can effectively filter out entities with low recognition quality, preventing low-quality data from entering the subsequent fault analysis and processing process, thereby improving the efficiency of the overall system and the accuracy of decision-making. As the final confirmed high-quality entity, the second entity can provide solid data support for accurate fault location, pattern recognition, and intelligent diagnosis, and help improve the automation level of engine maintenance and fault handling. In addition, this process also reflects strict control of entity recognition quality, promotes the standardization and normalization of fault information processing technology, and helps to establish a more efficient and accurate fault information management and analysis system.

[0083] In summary, by setting the threshold of quality inspection results, the determination condition of the second entity, that is, the first entity with qualified quality, is clarified. This mechanism not only ensures the accuracy of fault information extraction, but also reduces the need for manual intervention through automated processes, improves the efficiency and reliability of the fault handling process, and is an important part of fault information processing technology.

[0084] As an optional implementation mode, the method also includes: an identification step, in response to the quality detection result being less than a quality detection result threshold, inputting the target fault information into a second entity recognition model, using the second entity recognition model, through few-sample learning and prompt words, identifying the first entity from the target fault information, wherein the second entity recognition model is obtained by training a large language model using fault information samples and first entity samples; inputting the first entity identified in the second entity recognition model into the quality detection model, and determining the quality detection result of the first entity identified from the second recognition model through few-sample learning and prompt words; in response to the quality detection result being less than the quality detection result threshold, repeating the identification step until the quality detection result is greater than or equal to the quality detection result threshold.

[0085] In this embodiment, if the quality detection result is less than the quality detection result threshold, the identification step can be performed. In the identification step, the target fault information can be input into the second entity recognition model, and the second entity recognition model can be used to identify the first entity from the target fault information through few-sample learning and prompt words. The first entity identified in the second entity recognition model can be input into the quality detection model, and the quality detection result of the first entity identified from the second recognition model can be determined through few-sample learning and prompt words. If the quality detection result is less than the quality detection result threshold, the identification step can be repeated until the quality detection result is greater than or equal to the quality detection result threshold.

[0086] Optionally, the above method further improves the fault information extraction method based on engine quality data, and especially introduces a cyclic iterative identification and detection mechanism for situations where the quality inspection results do not meet expectations, to ensure that the second entity finally outputted meets high quality standards.

[0087] Optionally, after the first entity output by the initial entity recognition model is evaluated by the quality detection model, if its quality detection result is lower than a preset quality detection result threshold, it indicates that the quality of these entities is not sufficient to meet the needs of subsequent fault analysis. At this point, the system will enter a cyclic iterative recognition and detection process to improve the quality of entity recognition.

[0088] Optionally, in the recognition step, the target fault information will be input into the second entity recognition model. Compared with the first entity recognition model, the second entity recognition model focuses more on improving the accuracy of entity recognition through specific training methods, such as few-shot learning and prompt word engineering, especially when dealing with complex or specific types of fault information. Few-shot learning allows the model to adapt and learn quickly on limited examples, while prompt word engineering guides the model to more accurately identify the key features of the entity by constructing a specific text structure.

[0089] Optionally, for the first entity identified by the second entity recognition model, the system will use the quality detection model again to perform quality detection. This process also uses the method of few-shot learning and prompt word engineering to ensure that the model can accurately evaluate the quality of the entity based on existing knowledge and instructions. The quality detection result will be compared with the quality detection result threshold again to determine whether the output condition is met.

[0090] Optionally, if the quality detection result of the first entity identified from the second entity recognition model is still lower than the quality detection result threshold, the system will repeat the recognition step, that is, input the target fault information to the second entity recognition model again for entity recognition, and then perform quality detection on the recognition result. This cycle is iterated until the quality detection result of the entity identified from the second entity recognition model meets or exceeds the threshold, thereby ensuring that the second entity outputted in the end has high quality.

[0091] In the embodiment of the present invention, the above-mentioned iterative recognition and detection mechanism significantly improves the quality control level of entity recognition. It can not only handle the situation where the quality does not meet the standard in the initial recognition, but also continuously improve the accuracy and reliability of entity recognition through iterative optimization. The final confirmed second entity, as a high-quality output of fault information extraction, can more effectively support tasks such as fault analysis, pattern recognition and intelligent diagnosis, and improve the efficiency of the fault handling process and the accuracy of decision-making.

[0092] In summary. By introducing the second entity recognition model and the iterative quality detection mechanism, the problem of insufficient entity recognition quality is solved, ensuring the quality control and optimization of the fault information extraction process. This is an important step in achieving automation, precision and efficiency in fault processing technology, and reflects the flexibility and intelligence of the embodiment of the present invention in processing complex engine fault information.

[0093] As an optional implementation method, step S102, identifying target fault information from the initial fault information of the engine, includes: obtaining engine quality data from the initial fault information; obtaining fault fields from the engine quality data, and aggregating the fault fields into target fault information.

[0094] In this embodiment, in the process of identifying the target fault information from the initial fault information of the engine, the engine quality data can be obtained from the initial fault information, the fault field can be obtained from the engine quality data, and the fault field can be summarized as the target fault information.

[0095] Optionally, the process of extracting key, structured target fault information from complex initial fault information is the initial stage of the fault information extraction method, and its importance is self-evident.

[0096] Optionally, obtain engine-related quality data from the unstructured initial fault information. This usually involves preprocessing of text information, including but not limited to data cleaning, standardization and formatting to remove irrelevant information, correct errors and unify data formats to ensure smooth subsequent processing. Engine quality data may contain key information such as fault description, fault location, fault code, fault occurrence time, maintenance records, etc. The accuracy and completeness of this information is crucial for the extraction of fault information.

[0097] Optionally, specific fault fields can be further extracted from the acquired engine quality data. The above fields are the core components of fault information, such as fault location (which part or system has a problem), fault form (specific manifestation of the fault, such as wear, breakage, leakage, etc.), fault description (detailed information of the fault, including symptoms, occurrence environment, etc.) and maintenance measures (how to deal with the fault). The process of extracting fault fields is essentially to convert unstructured quality data into structured information that is easy to further process and analyze.

[0098] Optionally, after extracting the fault fields, the system needs to aggregate them into target fault information. Target fault information is a clearer and standardized fault description that integrates multiple fault fields into a complete fault record. This information structure not only contains a direct description of the fault, but may also include additional information such as fault classification, severity, frequency, etc., making the fault information richer and more comprehensive, and providing a solid foundation for subsequent fault analysis, pattern recognition, and intelligent diagnosis.

[0099] In the embodiment of the present invention, by obtaining key engine quality data from the initial fault information, and then extracting and summarizing specific fault fields, the present invention ensures the accuracy, completeness and reliability of fault information extraction. This preprocessing process converts unstructured fault text into structured, easy-to-understand and easy-to-analyze data, providing high-quality input for subsequent entity recognition, quality detection and optimization steps, thereby improving the efficiency and accuracy of the entire fault processing process.

[0100] As an optional implementation method, the method also includes: obtaining target fault information samples, and marking key information in the target fault information samples to obtain first entity samples; respectively extracting target fault information samples and first entity samples with target proportions from target fault information samples and first entity samples of different fault types as training data sets for training deep learning models.

[0101] In this embodiment, during the process of training the deep learning model, target fault information samples can be obtained, and key information in the target fault information samples can be annotated to obtain first entity samples. Target fault information samples and first entity samples with target proportions are extracted from target fault information samples and first entity samples of different fault types as training data sets for training the deep learning model.

[0102] Optionally, prepare high-quality training datasets for the deep learning model to ensure the performance and accuracy of the model in identifying and extracting fault information entities.

[0103] Optionally, representative target fault information samples are selected from a large amount of collected engine fault information. These samples should cover various types of faults that the engine may encounter, including but not limited to common, rare, complex, and simple problems. By ensuring the diversity and comprehensiveness of the samples, the generalization ability of the model can be improved, so that it can accurately identify and extract different types of fault information.

[0104] Optionally, the key information in the target fault information sample is manually labeled, which is a crucial step in building a training data set. During the labeling process, experts or labelers need to identify and mark the entities in the fault information, such as fault location, fault form, fault description, etc. These labeled entities will become the first entity samples. Manual labeling ensures the accuracy and quality of the data and is the basis for deep learning models to learn and understand fault information.

[0105] Optionally, after obtaining the labeled first entity samples, a specific proportion of data is extracted from these samples to construct a training data set for training the deep learning model. The selection of this target ratio needs to be determined according to the needs of model training and the size of the data set. Usually, a certain proportion of samples are extracted from each fault type to ensure the diversity and balance of the training data. For example, if the fault types include "wear", "breakage", "leakage", etc., then the same or appropriate proportion of samples should be extracted from each category to avoid the model being too partial to certain types, affecting the fairness and accuracy of recognition.

[0106] Optionally, use the constructed training dataset to train a deep learning model, such as the BERT+CRF model. Through repeated learning and adjustment, the model can learn to identify and extract entities from fault information. During the training process, the model will learn the characteristics, context associations, and recognition rules of entities based on labeled entity samples, so that it can accurately identify entities in unseen fault information.

[0107] In the embodiment of the present invention, the quality and diversity of the deep learning model training data are ensured through the above method, thereby improving the recognition accuracy and generalization ability of the model. Through the fine control of manual labeling and sample selection, the model can more effectively learn and grasp the key features of fault information, and can make reliable entity recognition and extraction even in the face of complex or rare fault conditions.

[0108] In summary, obtaining and labeling target fault information samples, as well as extracting a specific proportion of samples from them to construct a training data set, is the basic work before deep learning model training, which is crucial to improving the performance of the model. This process not only needs to ensure the representativeness of the data, but also needs to pay attention to the quality of the data. Through manual labeling and sample balance, it provides solid data support for model training, which is one of the key steps to achieve efficient and accurate identification and extraction of fault information.

[0109] In an embodiment of the present invention, if it is necessary to determine an entity with qualified quality from an unstructured initial fault information that does not meet a fixed format, the target fault information that is structured and meets a fixed format can be first analyzed from the initial fault information. The target fault information can be input into a pre-trained first entity recognition model to identify the target fault information and obtain a first entity. If the number of entities of the first entity meets the entity number threshold, the quality of the first entity can be detected using a quality detection model to obtain a quality detection result. Thus, according to the quality detection result, a second entity with higher quality can be determined. In this embodiment, through structured information processing, targeted training of deep learning models, entity number threshold screening, quality detection of large language models, and iterative optimization mechanisms, the method significantly improves the accuracy of determining entities in fault information, reduces misidentification and missed identification, and provides more reliable and efficient information support for fault analysis, diagnosis, and maintenance decisions. Through the close cooperation of these steps, even complex unstructured fault descriptions can be accurately and quickly converted into structured and high-quality entity information, thereby improving the efficiency and accuracy of the entire fault information processing process. Thereby, the technical effect of improving the accuracy of determining entities in fault information is achieved, and the technical problem of low accuracy of determining entities in fault information is solved.

[0110] The technical solution of the embodiment of the present invention is illustrated below in conjunction with preferred implementation modes.

[0111] In a related technology, an entity recognition method is proposed, which verifies the entity recognition results of the entity recognition model, and when the verification results are abnormal, updates the entity dictionary and corrects the entity recognition results to improve the recognition effect of named entities. This method can reduce the error rate of entity recognition, but the verification and correction process relies on model training and entity dictionary updates, and the degree of manual participation in the recognition process is relatively high.

[0112] In another related technology, a method for identifying assembly quality knowledge entities based on Bert-Crf is proposed. Complex product assembly quality entities are identified based on the Bert-Crf model, and complex product assembly quality entities in natural language texts are automatically extracted and classified. This technical solution can improve the accuracy of complex product assembly quality entity recognition, but for the trained Bert-Crf model, its output results lack an effective subsequent evaluation mechanism to ensure its generation quality.

[0113] However, the above methods fail to fully demonstrate the flexibility and reusability of the methods in entity recognition tasks, and lack an effective evaluation mechanism to ensure the quality of the recognition results. Therefore, there is still a technical problem of low accuracy in determining entities in fault information. Therefore, a more flexible and complete technical method is needed to strengthen entity recognition technology, especially in the field of quality data processing, to provide more reliable technical guarantees.

[0114] The embodiment of the present invention proposes a method for extracting fault information based on engine quality data, and the engine quality data contains various types of fault information, such as fault location, fault form, fault description, processing process, etc. In the current workflow, the recording and entry of fault information mainly rely on manual operation, which is inefficient and lacks unified specifications, resulting in uneven quality of entered data, poor data accuracy and readability, and inconvenience for relevant personnel in retrieving and using these data, thereby affecting the efficiency of fault analysis and processing and reducing overall work efficiency. The method systematically and logically combs fault information through the technology of combining large language models and small models. The method can greatly improve the accuracy and response speed of fault processing, thereby optimizing the entire maintenance process, improving the reliability of the system and the efficiency of maintenance work. The fault information processed by the method can provide support for subsequent fault mode recognition, intelligent diagnosis and other analysis work in a standardized and clear manner. Thereby achieving the technical effect of improving the accuracy of determining entities in fault information, and solving the technical problem of low accuracy of determining entities in fault information.

[0115] The embodiments of the present invention are further described below.

[0116] In an embodiment of the present invention, Figure 2 is a flow chart of a method for extracting fault information based on engine quality data according to an embodiment of the present invention. Figure 2 As shown, the method may include the following steps:

[0117] Step S201, obtaining engine quality data from unstructured fault information, selecting corresponding fields, and arranging them into structured fault records.

[0118] In this embodiment, the scope of processing data is determined, engine quality data is obtained from unstructured fault information, and fields such as fault location, fault form, fault description, and processing process are selected to be organized into structured fault records as data for input to the entity recognition generator. Each piece of data is a fault record, and the goal of this method is to extract N entities representing real fault information from each fault record.

[0119] Step S202, enter the entity recognition generator, construct a data set to train the Bert+CRF model in a stratified sampling manner, perform entity recognition on the unlabeled data after training, and use the recognition result as the generator output.

[0120] In this embodiment, the entity recognition generator is entered, and stratified sampling is used to extract 20% of the data from each faulty part category in the structured input data, and N entities to be identified are manually labeled to construct a training data set. The Bert+CRF model is trained, and after the training is completed, entity recognition is performed on the unlabeled fault record data to obtain the extraction result as the output of the entity recognition generator.

[0121] Step S203, enter the entity recognition detector, detect the received generator results based on the quantity rule, and detect the entity quantity for each entity in each fault record.

[0122] In this embodiment, an entity recognition detector is entered to evaluate the entity recognition results output by the generator, and the received generator results are detected based on the quantity rule. It is known that each fault record contains N entities to be identified representing real fault information. First, for each entity in each fault information, detection is performed according to the number of entities output by the generator.

[0123] Step S204, if the output number is 1, enter the entity recognition optimizer; if the output number is 0, enter the entity recognition model based on the large model, obtain the recognition result given by the large model, and then enter the entity recognition optimizer; if the output number is greater than 1, enter the recognition quality detection model based on the large model, retain the recognition results with high scores and enter the entity recognition optimizer.

[0124] In this embodiment, (1) if the output number is 1, it enters the entity recognition optimizer for further processing;

[0125] (2) If the output number is 0, it means that no entity is recognized in the generator, then the entity recognition model based on the large language model is entered. This model guides the large language model to extract the specified entity from the fault information through Few-shot Learning and prompt word engineering. Obtain the recognition result given by the large language model and then enter (1);

[0126] (3) If the number of outputs is greater than 1, the recognition quality detection model based on the large language model is entered. This model scores the received entity recognition results based on the understanding of the fault information through few-shot learning and prompt word engineering, and retains the recognition result with the largest score to enter (1).

[0127] Step S205, enter the entity recognition optimizer, and use the large model score to judge the quality of the recognition result.

[0128] In this embodiment, the entity recognition optimizer is entered, and the optimizer includes a recognition quality detection model based on a large language model. The scoring threshold of the model is set according to actual needs to judge the quality of the recognition result. If the recognition result score reaches or exceeds the set threshold, the result is marked as "good" and used as the final output of the entity extraction method. On the contrary, if the recognition result score is lower than the threshold, the result is marked as "bad" and input into the entity recognition model based on the large language model for reprocessing and re-scoring until the result is marked as "good".

[0129] In this embodiment, the entity recognition model and the recognition quality detection model based on the large language model control the large language model through few-shot learning and prompt word engineering, and perform entity recognition related tasks based on the understanding of fault information. The fault information extraction method based on engine quality data integrates the entity recognition generator, entity recognition detector, and entity recognition optimizer to achieve a systematic combing of quality data with a reliable, efficient, and comprehensive process, greatly improving the accuracy of fault handling.

[0130] In an embodiment of the present invention, if it is necessary to determine an entity with qualified quality from an unstructured initial fault information that does not meet a fixed format, the target fault information that is structured and meets a fixed format can be first analyzed from the initial fault information. The target fault information can be input into a pre-trained first entity recognition model to identify the target fault information and obtain a first entity. If the number of entities of the first entity meets the entity number threshold, the quality of the first entity can be detected using a quality detection model to obtain a quality detection result. Thus, according to the quality detection result, a second entity with higher quality can be determined. In this embodiment, through structured information processing, targeted training of deep learning models, entity number threshold screening, quality detection of large language models, and iterative optimization mechanisms, the method significantly improves the accuracy of determining entities in fault information, reduces misidentification and missed identification, and provides more reliable and efficient information support for fault analysis, diagnosis, and maintenance decisions. Through the close cooperation of these steps, even complex unstructured fault descriptions can be accurately and quickly converted into structured and high-quality entity information, thereby improving the efficiency and accuracy of the entire fault information processing process. Thereby, the technical effect of improving the accuracy of determining entities in fault information is achieved, and the technical problem of low accuracy of determining entities in fault information is solved.

[0131] According to an embodiment of the present invention, a device for determining an entity in fault information is also provided. It should be noted that the device for determining an entity in fault information can be used to execute the method for determining an entity in fault information in the above embodiment.

[0132] Figure 3 FIG. 1 is a schematic diagram of a device for determining an entity in fault information according to an embodiment of the present invention. Figure 3 As shown, the apparatus 300 for determining an entity in the fault information may include: a first identification unit 302 , a second identification unit 304 , a detection unit 306 and a determination unit 308 .

[0133] The first identification unit 302 is used to identify target fault information from initial fault information of the engine, wherein the initial fault information is unstructured fault information that does not meet a fixed format, and the target fault information is structured fault information that meets a fixed format.

[0134] The second recognition unit 304 is used to use the first entity recognition model to identify the target fault information and obtain the first entity, wherein the first entity recognition model is obtained by training a deep learning model using target fault information samples and corresponding first entity samples, and the first entity is key information in the target fault information for characterizing that the engine is in a fault state.

[0135] The detection unit 306 is used to detect the quality of the first entity using a quality detection model in response to the entity quantity of the first entity satisfying the entity quantity threshold, and obtain a quality detection result, wherein the quality detection model is obtained by training a large language model using the first entity sample and the corresponding quality detection result sample.

[0136] The determining unit 308 is configured to determine a second entity based on the quality detection result, wherein the quality of the second entity is higher than the quality of the first entity.

[0137] In an embodiment of the present invention, target fault information is identified from initial fault information of the engine by a first identification unit 302; the target fault information is identified by a second identification unit 304 using a first entity recognition model to obtain a first entity; in response to the entity quantity of the first entity satisfying an entity quantity threshold, the detection unit 306 detects the quality of the first entity using a quality detection model to obtain a quality detection result; the determination unit 308 determines the second entity based on the quality detection result, thereby solving the technical problem of low determination accuracy of entities in the fault information and achieving the technical effect of improving the determination accuracy of entities in the fault information.

[0138] According to an embodiment of the present invention, a computer-readable storage medium is further provided. The storage medium includes a stored program, wherein the program executes the method for determining an entity in fault information in the above embodiment.

[0139] According to an embodiment of the present invention, a processor is further provided. The processor is used to run a program. When the program is run, the method for determining an entity in the fault information in the above embodiment is executed.

[0140] The embodiment of the present application further provides a computer program product. Optionally, in this embodiment, the computer program product may include a computer program, and when the computer program is executed by a processor, the method for determining an entity in the fault information of the embodiment of the present application is implemented.

[0141] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0142] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0143] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0144] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0145] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, Read-Only Memory (ROM), Random Access Memory (RAM), mobile hard disk, magnetic disk or optical disk, etc., various media that can store program codes.

[0146] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for determining an entity in fault information, characterized in that: include: Identifying target fault information from initial fault information of the engine, wherein the initial fault information is unstructured fault information that does not meet a fixed format, and the target fault information is structured fault information that meets the fixed format; Using a first entity recognition model, the target fault information is recognized to obtain a first entity, wherein the first entity recognition model is obtained by training a deep learning model using target fault information samples and corresponding first entity samples, and the first entity is key information in the target fault information for characterizing that the engine is in a fault state; In response to the entity quantity of the first entity satisfying an entity quantity threshold, using a quality detection model to detect the quality of the first entity to obtain a quality detection result, wherein the quality detection model is obtained by training a large language model using the first entity sample and the corresponding quality detection result sample; Based on the quality detection result, a second entity is determined, wherein the quality of the second entity is higher than the quality of the first entity.

2. The method according to claim 1, characterized in that The entity quantity threshold includes a first entity quantity threshold, wherein, in response to the entity quantity of the first entity satisfying the entity quantity threshold, using a quality detection model to detect the quality of the first entity to obtain a quality detection result, including: A first detection step, in response to the number of entities being equal to the first entity number threshold, using a quality detection model to detect the quality of the first entity to obtain the quality detection result; In a second detection step, in response to the number of entities being greater than the first entity number threshold, the first entity is analyzed using the quality detection model through few-sample learning and prompt words to determine the quality detection result; for the first entity that meets the first entity number threshold and the maximum quality detection result corresponds to the first entity, the first detection step is executed.

3. The method according to claim 2, characterized in that The entity quantity threshold includes a second entity quantity threshold, and the second entity quantity threshold is less than the first entity quantity threshold, wherein, in response to the entity quantity of the first entity meeting the entity quantity threshold, using a quality detection model, the quality of the first entity is detected to obtain a quality detection result, including: In response to the number of entities being equal to the second entity number threshold, inputting the target fault information into a second entity recognition model, and using the second entity recognition model to recognize the first entity from the target fault information through few-sample learning and prompt words, wherein the second entity recognition model is obtained by training the large language model using the fault information sample and the first entity sample; In response to the entity quantity of the first entity identified from the second entity recognition model being equal to the first entity quantity threshold, the first detection step is performed; or, in response to the entity quantity being greater than the first entity quantity threshold, the second detection step is performed.

4. The method according to claim 2, characterized in that: In response to the entity quantity of the first entity meeting the entity quantity threshold, using a quality detection model to detect the quality of the first entity to obtain a quality detection result, including: In response to the entity quantity satisfying the entity quantity threshold, the quality of the first entity is analyzed by using the quality detection model through few-sample learning and prompt words to determine the quality detection result.

5. The method according to claim 4, characterized in that Determining a second entity based on the quality detection result includes: In response to the quality detection result being greater than or equal to a quality detection result threshold, the first entity is determined to be the second entity.

6. The method according to claim 5, characterized in that The method further comprises: an identification step, in response to the quality detection result being less than the quality detection result, inputting the target fault information into a second entity recognition model, and using the second entity recognition model to identify the first entity from the target fault information through the few-sample learning and the prompt word, wherein the second entity recognition model is obtained by training the large language model using the fault information sample and the first entity sample; Inputting the first entity identified from the second entity recognition model into the quality detection model, and determining the quality detection result of the first entity identified from the second recognition model through the few-sample learning and the prompt word; In response to the quality detection result being less than the quality detection result threshold, the identifying step is repeatedly performed until the quality detection result is greater than or equal to the quality detection result threshold.

7. The method according to claim 1, characterized in that Identify target fault information from the initial fault information of the engine, including: Acquiring engine quality data from the initial fault information; Fault fields are acquired from the engine quality data, and the fault fields are aggregated into the target fault information.

8. The method according to claim 1, characterized in that The method further comprises: Acquire the target fault information sample, and mark the key information in the target fault information sample to obtain the first entity sample; From the target fault information samples and the first entity samples of different fault types, target proportions of the target fault information samples and the first entity samples are respectively extracted as training data sets for training the deep learning model.

9. A device for determining an entity in fault information, characterized in that: include: A first identification unit is used to identify target fault information from initial fault information of the engine, wherein the initial fault information is unstructured fault information that does not meet the fixed format, and the target fault information is structured fault information that meets the fixed format; A second recognition unit is used to use a first entity recognition model to recognize the target fault information to obtain a first entity, wherein the first entity recognition model is obtained by training a deep learning model using target fault information samples and corresponding first entity samples, and the first entity is key information in the target fault information for characterizing that the engine is in a fault state; a detection unit, configured to detect the quality of the first entity using a quality detection model in response to the entity quantity of the first entity satisfying an entity quantity threshold, and obtain a quality detection result, wherein the quality detection model is obtained by training a large language model using the first entity sample and the corresponding quality detection result sample; A determining unit is configured to determine a second entity based on the quality detection result, wherein the quality of the second entity is higher than the quality of the first entity.

10. A processor, characterized in that: The processor is used to run a program, wherein the program, when run by the processor, executes the method for determining an entity in fault information as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method for determining an entity in fault information according to any one of claims 1 to 8.

12. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the method for determining an entity in fault information as claimed in any one of claims 1 to 8.

Images