Distillation methods, apparatuses, devices, media, and products for aeroengines
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- AERO ENGINE ACAD OF CHINA
- Filing Date
- 2026-02-04
- Publication Date
- 2026-06-02
Smart Images
Figure CN122133752A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to a distillation method, apparatus, equipment, medium, and product for aero-engines. Background Technology
[0002] As the core power system of aviation equipment, the maintenance and support of aero-engines involves a wealth of professional knowledge, including engine structural principles, operating conditions, typical failure modes, and standardized maintenance procedures. This places high demands on information understanding, analysis, and decision-making capabilities. In recent years, the development of artificial intelligence models has accelerated rapidly, enabling them to perform various natural language processing tasks such as question-and-answer generation, text analysis, and logical reasoning. Therefore, applying them to the field of aero-engine maintenance is poised to become a future trend.
[0003] However, these AI models typically have a large number of parameters. If they are directly applied to specialized fields such as aircraft engine maintenance, the deployment of these models would require significant computing resources, making them difficult to deploy on edge devices with limited computing power. Moreover, in practical maintenance applications, the response speed and engineering controllability of these models are often insufficient to meet the demands of field use. Summary of the Invention
[0004] This disclosure is made in view of the above-mentioned problems and provides a distillation method, apparatus, equipment, medium and product for aircraft engines.
[0005] According to one aspect of this disclosure, a distillation method for an aircraft engine is provided, comprising:
[0006] Collect multi-source heterogeneous data in the field of aero-engine maintenance; Based on the actual business processes in the field of aero-engine maintenance, multi-source heterogeneous data is reconstructed in a task-oriented manner to obtain a distillation dataset in the form of command response. Distillation supervision signals are obtained by reasoning on the distillation dataset using a teacher model; wherein, the distillation supervision signals include at least one of the following: the reasoning result of the teacher model, the reasoning process, and the candidate result ranking information; The student model is trained on distillation based on the distillation supervision signal, and the trained student model is then applied to the field of aero-engine maintenance.
[0007] The beneficial effects of this solution are as follows: Addressing the highly specialized and engineering-constrained nature of aero-engine maintenance, this solution uses a distillation training method to transfer the domain knowledge representation and decision-making behavior of a teacher model with complex reasoning capabilities to a student model with smaller parameters. This significantly reduces the model parameter size and reasoning resource consumption while maintaining a relatively accurate understanding of professional knowledge such as engine structure, fault mechanisms, and maintenance procedures. This is beneficial for improving the model's response speed and engineering controllability in practical maintenance applications.
[0008] Furthermore, according to one aspect of the distillation method of this disclosure, inference is performed on the distillation dataset based on a teacher model to obtain distillation supervision signals, including: The distillation dataset is input into the teacher model, and black-box distillation is used to obtain the inference results output by the teacher model; where black-box distillation means that only the input and output of the teacher model are accessed. And / or, white-box distillation technology is used to obtain the reasoning process of the teacher model output; wherein, white-box distillation technology means supporting access to the teacher model's input, output, candidate result ranking information, and intermediate reasoning states; the reasoning process includes fault analysis, causal inference, and maintenance decision-making processes; Based on the reasoning results and / or the reasoning process and candidate result ranking information, a distillation supervision signal is obtained.
[0009] In one or more embodiments, the beneficial effects of this solution are as follows: by constructing multi-type distillation supervision signals for the field of aero-engine maintenance, and introducing a supervision form that includes reasoning results (standardized answer text), reasoning process (chain-like reasoning process), and candidate result ranking information, the student model can not only learn the output results of the teacher model, but also learn its reasoning path and decision-making tendency in complex fault diagnosis and maintenance decision-making scenarios, thereby improving the reasoning consistency, stability, and reliability of the student model in complex maintenance scenarios.
[0010] Furthermore, according to one aspect of the distillation method of this disclosure, distillation training is used to guide a student model to generate outputs similar to those of a teacher model under the same input; During the distillation training process, the decision-making tendency of the student model is guided by minimizing the difference between the output results of the teacher model and the student model; and, through the preference optimization mechanism, the student model selects the answer that conforms to the field of aero-engine maintenance from multiple candidate results.
[0011] Furthermore, according to one aspect of the distillation method of this disclosure, the instruction response form includes at least one of the following: a fault diagnosis question and answer task, a maintenance decision reasoning task, and a process explanation and specification description task for maintenance procedures.
[0012] Furthermore, according to one aspect of the distillation method of this disclosure, multi-source heterogeneous data is reconstructed in a task-oriented manner to obtain a distillation dataset in the form of an instruction response, including: Task-oriented reconstruction of multi-source heterogeneous data yields task samples in the form of instruction responses. The task samples are preprocessed to generate a distillation dataset.
[0013] Furthermore, according to one aspect of the distillation method of this disclosure, a student model that has completed distillation training is applied to the field of aircraft engine maintenance, including: The student model, having completed distillation training, is deployed on a computing platform and integrated with the aircraft engine maintenance business system.
[0014] In one or more embodiments, the beneficial effects of this solution are as follows: the student model obtained through distillation training can be deployed on the edge with limited computing power and deeply integrated with the aero-engine maintenance system, providing an efficient and engineering-applicable intelligent model for tasks such as aero-engine maintenance Q&A, fault auxiliary diagnosis and maintenance decision support, which is conducive to promoting the intelligent improvement of aero-engine maintenance and support capabilities.
[0015] According to another aspect of this disclosure, a distillation apparatus for an aircraft engine is provided, comprising: The data acquisition module is used to collect multi-source heterogeneous data in the field of aero-engine maintenance; The reconstruction module is used to perform task-oriented reconstruction of multi-source heterogeneous data based on actual business processes in the field of aero-engine maintenance, and obtain distilled datasets in the form of command responses. The teacher model inference module is used to perform inference on the distillation dataset through the teacher model to obtain distillation supervision signals; wherein, the distillation supervision signals include at least one of the teacher model's inference results, inference process, and candidate result ranking information; The distillation module is used to train student models based on distillation supervision signals, and then apply the trained student models to the field of aero-engine maintenance.
[0016] According to another aspect of this disclosure, a computer device is provided, including a memory, a processor, and a computer program stored in the memory, the processor executing the computer program to implement the method of one aspect above.
[0017] According to another aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the method of one aspect above.
[0018] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method of the above-described aspect.
[0019] It should be understood that both the foregoing general description and the following detailed description are exemplary and intended to provide further illustration of the claimed technology. Attached Figure Description
[0020] The above and other objects, features, and advantages of this disclosure will become more apparent from the more detailed description of the embodiments thereof in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this disclosure and form part of the specification. They are used together with the embodiments of this disclosure to explain the disclosure and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.
[0021] Figure 1 This is a system architecture diagram illustrating a distillation method for an aircraft engine according to an embodiment of the present disclosure.
[0022] Figure 2 This is a flowchart illustrating a distillation method according to an embodiment of the present disclosure.
[0023] Figure 3 This is a flowchart illustrating a distillation method for an aircraft engine according to an embodiment of the present disclosure.
[0024] Figure 4 This is a schematic diagram of a distillation apparatus for an aircraft engine according to an embodiment of the present disclosure.
[0025] Figure 5 This is a schematic diagram illustrating the structure of a computer device according to an embodiment of the present disclosure.
[0026] Figure 6 This is a schematic diagram illustrating a computer program product according to an embodiment of the present disclosure. Detailed Implementation
[0027] To make the objectives, technical solutions, and advantages of this disclosure more apparent, exemplary embodiments according to this disclosure will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments of this disclosure. It should be understood that this disclosure is not limited to the exemplary embodiments described herein.
[0028] As the core power system of aviation equipment, the maintenance and support of aero-engines involves a wealth of professional knowledge, including engine structural principles, operating conditions, typical failure modes, and standardized maintenance procedures. This places high demands on information understanding, analysis, and decision-making capabilities. In recent years, the development of artificial intelligence models has accelerated rapidly, enabling them to perform various natural language processing tasks such as question-and-answer generation, text analysis, and logical reasoning. Therefore, applying them to the field of aero-engine maintenance is poised to become a future trend.
[0029] However, these AI models typically have a large number of parameters. If they are directly applied to specialized fields such as aircraft engine maintenance, the deployment of these models would require significant computing resources, making them difficult to deploy on edge devices with limited computing power. Moreover, in practical maintenance applications, the response speed and engineering controllability of these models are often insufficient to meet the demands of field use.
[0030] Knowledge distillation technology can transfer knowledge and reasoning behaviors from high-capacity teacher models to student models with smaller parameter sizes, providing a feasible approach to solving the aforementioned problems. However, existing distillation methods are mostly designed for general tasks and lack system design specific to the aero-engine maintenance domain. They fail to fully reflect the specialization of engine maintenance knowledge and the complexity of the reasoning process, and there is still room for optimization in areas such as domain task modeling, distillation signal construction, and engineering deployment during the distillation process. Therefore, there is an urgent need for a knowledge distillation method specifically for the aero-engine maintenance domain. This method should fully utilize high-capacity teacher models to effectively transfer domain knowledge and reasoning capabilities to lightweight student models, thereby reducing model deployment costs while ensuring professionalism and reliability, and improving its usability and engineering adaptability in practical aero-engine maintenance applications.
[0031] The above description, with reference to the accompanying drawings, describes a distillation method, apparatus, device, medium, and product for aero-engines according to embodiments of the present disclosure. By transferring domain knowledge and reasoning behavior from a teacher model with strong natural language understanding and complex reasoning capabilities to a student model with a smaller parameter size, efficient injection and lightweight deployment of aero-engine maintenance domain knowledge are achieved without relying on large-scale domain retraining. This effectively solves the problems of large model size, high deployment cost, and insufficient engineering adaptability of existing models in professional fields such as aero-engine maintenance.
[0032] To facilitate understanding of this embodiment, a distillation method for aero-engines disclosed in this disclosure will first be described in detail. The execution entity of the distillation method for aero-engines provided in this disclosure is generally a computer device with certain computing capabilities. This computer device may include, for example, a terminal device, a server, or other processing devices. The terminal device may be a user equipment (UE), mobile device, user terminal, terminal, cellular phone, cordless phone, handheld device, computing device, vehicle-mounted device, wearable device, etc. In some possible implementations, the distillation method for aero-engines can be implemented by a processor calling computer-readable instructions stored in memory.
[0033] like Figure 1 The diagram shown is a system architecture diagram of a distillation method for aero-engines provided in this embodiment of the present disclosure, including a distillation dataset construction module 1, a teacher model inference module 2, a student model distillation module 3, and a model deployment module 4. The following is a detailed description of each module: Distillation Dataset Building Module 1: Used to build a distillation knowledge dataset in the field of aero-engine maintenance.
[0034] By collecting multi-source heterogeneous data in the field of aero-engine maintenance, and combining the business processes and engineering application requirements of aero-engine maintenance, the multi-source heterogeneous data is task-oriented modeled and command-oriented reconstructed to form a distillation dataset suitable for distillation training.
[0035] The field of aero-engine maintenance involves a large amount of complex and diverse professional technical data, including both structured data and a significant amount of unstructured text data. To ensure the effectiveness of distillation training, this embodiment first systematically organizes and task-orientedly models the data in the aero-engine maintenance field. Specifically, this includes the following steps: Step 1.1: Collect and organize multi-source heterogeneous data in the field of aircraft engine maintenance to form raw domain data.
[0036] Among them, multi-source heterogeneous data includes, but is not limited to: aero-engine structural design documents, system principle descriptions, maintenance technical manuals, historical fault case records, typical fault analysis reports, maintenance experience summary documents, flight and ground test operation data description documents, and maintenance and support related technical specifications.
[0037] The aforementioned multi-source heterogeneous data were standardized, version-labeled, and cleaned to form an original domain data set, providing a foundation for subsequent task-oriented reconstruction.
[0038] Step 1.2: Based on the aircraft engine diagnostic and maintenance process, the original domain data set is reconstructed in a task-oriented manner, and the unstructured or semi-structured original domain data is converted into task samples in the form of instruction responses that can be learned by the model.
[0039] Since the original domain data is mostly in text form, it is difficult to directly use it for model distillation training. This embodiment reconstructs the original data into task-oriented data based on the actual business process of aero-engine maintenance. Specifically, the original domain data is transformed into task samples in the form of command responses, for example: Troubleshooting Q&A Task: What could be the cause of an unusual noise when starting the engine? Maintenance decision-making reasoning task: For a certain fault, which maintenance plan should be prioritized? The task is to explain and standardize the maintenance procedures, such as the standardized procedures for replacing turbine blades. Step 1.3: Preprocess the task samples, including quality screening, noise reduction and data augmentation, to build a high-quality distillation dataset for knowledge distillation.
[0040] Specifically, quality control is performed on the task samples after task-oriented reconstruction, including removing data samples that are semantically incomplete, logically conflicting, or unrelated to aircraft engine maintenance; and merging duplicate task samples.
[0041] Meanwhile, the task samples are enhanced by methods such as synonymous question expansion, task condition variation, and question formulation reorganization to improve the diversity and coverage of the distillation dataset, and finally construct a high-quality distillation dataset for the field of aero-engine maintenance.
[0042] Teacher Model Inference Module 2: Used to infer the distillation dataset based on the teacher model and construct distillation supervision signals corresponding to the field of aero-engine maintenance.
[0043] By calling a teacher model with strong domain understanding and reasoning capabilities, reasoning is performed on task samples in the distillation dataset to obtain distillation supervision signals that reflect the teacher model's knowledge representation and decision-making behavior in the field of aero-engine maintenance.
[0044] Specifically, the following steps are included: Step 2.1: Selection and application of the teacher model.
[0045] A model with strong natural language understanding and complex reasoning capabilities was selected as the teacher model. The teacher model can generate professional and structured analysis results for problems in the field of aero-engine maintenance.
[0046] Step 2.2: Generate distillation monitoring signals based on black-box distillation technology or white-box distillation technology.
[0047] In the black-box distillation scenario, only the output text of the teacher model for the task samples in the distillation dataset is obtained, i.e., the inference results, which are used to characterize the knowledge representation of the teacher model on the aircraft engine maintenance problem. In the white-box distillation scenario, not only can the input text and output text of the teacher model be obtained, but also the reasoning process and internal reasoning information of the teacher model can be obtained, including the probability distribution information, candidate result ranking information, or intermediate reasoning state of the teacher model during the generation process.
[0048] Step 2.3: Construct multiple types of distillation supervision signals.
[0049] Based on the output of the teacher model, construct at least one of the following distillation supervision signals: Standardized natural language response text (i.e., reasoning results) for aircraft engine maintenance issues; chain-thinking reasoning text (i.e., reasoning process) containing fault analysis, causal inference, and maintenance decisions; candidate result ranking information or preference labeling information for multiple candidate results for the same question; the above distillation supervision signals can be used individually or in combination.
[0050] Student Model Distillation Module 3: Used to perform knowledge distillation training on student models with a smaller parameter size than the teacher model based on distillation supervision signals.
[0051] During the distillation training process, given the same aircraft engine maintenance instructions, the student model is guided to learn the output behavior, reasoning path and decision-making tendency of the teacher model, thereby approximating the performance of the teacher model in aircraft engine maintenance tasks with a limited number of parameters.
[0052] Specifically, it includes the following steps: Step 3.1: Initialize the student model.
[0053] The model with a smaller parameter size than the teacher model is selected as the student model. The student model is suitable for scenarios with limited computing power or for deployment on the edge.
[0054] Step 3.2: Perform distillation training based on distillation supervision signals.
[0055] Distillation supervision signals generated by the teacher model are used to train the student model, enabling the student model to generate specialized outputs similar to those of the teacher model.
[0056] Step 3.3: Joint distillation based on distribution matching and preference optimization.
[0057] By minimizing the difference between the output distributions of the teacher model and the student model, the student model is guided to learn the decision-making tendency of the teacher model. Furthermore, the student model's response quality in complex aero-engine maintenance scenarios is improved through a preference optimization mechanism, making the student model more inclined to select responses that conform to aero-engine maintenance specifications and engineering experience among multiple candidate results.
[0058] Step 3.4: Evaluation and iterative optimization of distillation results.
[0059] The performance of the distilled student model in an aero-engine maintenance task was evaluated, including answer accuracy and reasoning consistency, and the distillation process was iteratively optimized based on the evaluation results.
[0060] Model Deployment Module 4: Used to deploy student models that have completed distillation training in aero-engine maintenance application scenarios.
[0061] The distilled student model is deployed to business systems related to aero-engine maintenance to support practical applications such as engine maintenance Q&A, fault diagnosis assistance, or maintenance decision support.
[0062] Specifically, the following steps are included: Step 4.1: Model Deployment.
[0063] The distilled student model will be deployed on a domestically produced computing platform or a general-purpose GPU platform to meet the requirements of intranet isolation and autonomous control in the context of aero-engine maintenance.
[0064] Step 4.2: Application system integration.
[0065] By integrating student models with the aero-engine maintenance business system, vertical applications such as aero-engine maintenance Q&A system, fault diagnosis system, and maintenance decision support system can be built.
[0066] Step 4.3: Support for actual maintenance scenarios.
[0067] In practical applications, based on the natural language descriptions input by maintenance personnel, the student model analyzes and outputs engine fault phenomena, possible causes, and maintenance suggestions, providing intelligent assistance to maintenance personnel.
[0068] like Figure 2 The diagram shown is an overall flow chart of the distillation process, including: S201: Obtain raw domain data in the field of aircraft engine maintenance.
[0069] S202: Construct the distillation dataset.
[0070] S203: Select and call the teacher model.
[0071] S204: Generate a distillation monitoring signal based on black-box distillation and white-box distillation.
[0072] S205: Distillation training of the student model based on distillation supervision signals.
[0073] S206: Apply the trained student model to the aircraft engine maintenance application system.
[0074] Based on the above embodiments, this embodiment also provides a distillation method for aircraft engines, such as... Figure 3 The diagram shows a flowchart of the distillation method, including steps S301-S304: S301: Collects multi-source heterogeneous data in the field of aircraft engine maintenance.
[0075] S302: Based on the actual business processes in the field of aero-engine maintenance, multi-source heterogeneous data is reconstructed in a task-oriented manner to obtain a distillation dataset in the form of command response.
[0076] S303: Obtain distillation supervision signals by reasoning on the distillation dataset using a teacher model; The distillation supervision signal includes at least one of the following: the inference result of the teacher model, the inference process, and the candidate result ranking information.
[0077] S304: The student model is trained by distillation based on the distillation supervision signal, and the trained student model is applied to the field of aero-engine maintenance.
[0078] In one or more embodiments, inference is performed on the distillation dataset based on a teacher model to obtain distillation supervision signals, including: The distillation dataset is input into the teacher model, and black-box distillation is used to obtain the inference results output by the teacher model; where black-box distillation means that only the input and output of the teacher model are accessed. And / or, white-box distillation technology is used to obtain the reasoning process of the teacher model output; wherein, white-box distillation technology means supporting access to the teacher model's input, output, candidate result ranking information, and intermediate reasoning states; the reasoning process includes fault analysis, causal inference, and maintenance decision-making processes; Based on the reasoning results and / or the reasoning process and candidate result ranking information, a distillation supervision signal is obtained.
[0079] In one or more embodiments, distillation training is used to instruct a student model to generate outputs similar to those of a teacher model under the same inputs; During the distillation training process, the decision-making tendency of the student model is guided by minimizing the difference between the output results of the teacher model and the student model; and, through the preference optimization mechanism, the student model selects the answer that conforms to the field of aero-engine maintenance from multiple candidate results.
[0080] In one or more embodiments, the instruction response format includes at least one of the following: a fault diagnosis question-and-answer task, a maintenance decision reasoning task, and a process explanation and specification description task for maintenance procedures. For example: Troubleshooting Q&A Task: What could be the cause of an unusual noise when starting the engine? Maintenance decision-making reasoning task: For a certain fault, which maintenance plan should be prioritized? The task is to explain and standardize the maintenance procedures, such as the standardized procedures for replacing turbine blades. In one or more embodiments, multi-source heterogeneous data is reconstructed in a task-oriented manner to obtain a distillation dataset in the form of an instruction response, including: Task-oriented reconstruction of multi-source heterogeneous data yields task samples in the form of instruction responses. The task samples are preprocessed to generate a distillation dataset.
[0081] In one or more embodiments, the student model that has completed distillation training is applied to the field of aero-engine maintenance, including: The student model, having completed distillation training, is deployed on a computing platform and integrated with the aircraft engine maintenance business system.
[0082] According to another aspect of the embodiments of this disclosure, a distillation apparatus for an aircraft engine is provided, such as... Figure 4 As shown, the device includes: The acquisition module 401 is used to acquire multi-source heterogeneous data in the field of aero-engine maintenance. Reconstruction module 402 is used to perform task-oriented reconstruction of multi-source heterogeneous data based on actual business processes in the field of aero-engine maintenance, and obtain distillation datasets in the form of instruction responses. The teacher model inference module 403 is used to perform inference on the distillation dataset through the teacher model to obtain distillation supervision signals; wherein, the distillation supervision signals include at least one of the teacher model's inference results, inference process, and candidate result ranking information; Distillation module 404 is used to perform distillation training on the student model based on distillation supervision signals, and then apply the student model that has completed distillation training to the field of aero-engine maintenance.
[0083] The distillation apparatus for aircraft engines provided in this disclosure and the distillation method provided in this disclosure are based on the same inventive concept and have the same beneficial effects as the methods used, operated or implemented therein.
[0084] This disclosure also provides a computer device for performing the distillation method described above. Please refer to... Figure 5It illustrates a schematic diagram of a computer device provided by some embodiments of this disclosure. For example... Figure 5 As shown, the computer device 5 includes: a processor 500, a memory 501, a bus 502, and a communication interface 503. The processor 500, the communication interface 503, and the memory 501 are connected via the bus 502. The memory 501 stores a computer program that can run on the processor 500. When the processor 500 runs the computer program, it executes the distillation method provided in any of the foregoing embodiments of this disclosure.
[0085] The memory 501 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this device network element and at least one other network element is achieved through at least one communication interface 503 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc.
[0086] Bus 502 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. The memory 501 is used to store programs. After receiving an execution instruction, the processor 500 executes the program. The distillation method disclosed in any of the foregoing embodiments of this disclosure can be applied to the processor 500, or implemented by the processor 500.
[0087] The processor 500 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 500 or by instructions in software form. The processor 500 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPTA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this disclosure. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this disclosure can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 501. The processor 500 reads the information in memory 501 and, in conjunction with its hardware, completes the steps of the above method.
[0088] The computer equipment and distillation method provided in this disclosure are based on the same inventive concept and have the same beneficial effects as the methods they employ, operate, or implement.
[0089] This disclosure also provides a computer-readable storage medium corresponding to the distillation method provided in the foregoing embodiments. The computer-readable storage medium is an optical disc, on which a computer program (i.e., a computer program product) is stored. When the computer program is run by a processor, it executes the distillation method provided in any of the foregoing embodiments.
[0090] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be elaborated here.
[0091] The computer-readable storage medium provided in the above embodiments of this disclosure and the distillation method provided in the embodiments of this disclosure are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the applications stored therein.
[0092] This disclosure also provides a computer program product; please refer to [reference needed]. Figure 6 The computer program product 600 carries program code, namely computer program 601. The instructions included in the computer program 601 can be used to execute the steps of the distillation method described in the above method embodiments. For details, please refer to the above method embodiments, which will not be repeated here.
[0093] The aforementioned computer program product can be implemented through hardware, software, or a combination thereof. In one optional embodiment, the computer program product is specifically embodied in a computer storage medium; in another optional embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.
[0094] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.
[0095] The block diagrams of devices, apparatuses, devices, and systems disclosed herein are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.
[0096] Additionally, as used herein, the "or" used in a list of items beginning with "at least one" indicates a separate list, such that a list of, for example, "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not imply that the described example is preferred or better than other examples.
[0097] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.
[0098] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.
[0099] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.
[0100] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.
Claims
1. A distillation method for aircraft engines, characterized in that, include: Collect multi-source heterogeneous data in the field of aero-engine maintenance; Based on the actual business processes in the field of aero-engine maintenance, the multi-source heterogeneous data is reconstructed in a task-oriented manner to obtain a distillation dataset in the form of command response. The distillation supervision signal is obtained by reasoning on the distillation dataset using a teacher model; wherein the distillation supervision signal includes at least one of the teacher model's reasoning result, reasoning process, and candidate result ranking information; The student model is trained on distillation based on the distillation supervision signal, and the trained student model is then applied to the field of aero-engine maintenance.
2. The distillation method as described in claim 1, characterized in that, Based on the teacher model, inference is performed on the distillation dataset to obtain distillation supervision signals, including: The distillation dataset is input into the teacher model, and black-box distillation is used to obtain the inference results output by the teacher model; wherein, black-box distillation means accessing only the input and output of the teacher model; And / or, white-box distillation technology is used to obtain the reasoning process output by the teacher model; wherein, the white-box distillation technology means supporting access to the teacher model's input, output, candidate result ranking information, and intermediate reasoning states; the reasoning process includes fault analysis, causal inference, and maintenance decision-making processes; Based on the reasoning results and / or the reasoning process and the candidate result ranking information, a distillation supervision signal is obtained.
3. The distillation method as described in claim 1, characterized in that, The distillation training is used to guide the student model to generate outputs similar to those of the teacher model under the same input; During the distillation training process, the decision-making tendency of the student model is guided by minimizing the difference between the output results of the teacher model and the student model. Furthermore, through a preference optimization mechanism, the student model selects the answer that aligns with the field of aero-engine maintenance from among multiple candidate results.
4. The distillation method as described in claim 1, characterized in that, The instruction response format includes at least one of the following: fault diagnosis question and answer task, maintenance decision reasoning task, and maintenance procedure process explanation and specification description task.
5. The distillation method as described in claim 1, characterized in that, The multi-source heterogeneous data is reconstructed in a task-oriented manner to obtain a distillation dataset in the form of an instruction response, including: The multi-source heterogeneous data is reconstructed in a task-oriented manner to obtain task samples in the form of instruction responses; The task samples are preprocessed to generate a distillation dataset.
6. The distillation method as described in claim 1, characterized in that, Applying the student model, which has completed distillation training, to the field of aircraft engine maintenance includes: The student model, having completed distillation training, is deployed on a computing platform and integrated with the aircraft engine maintenance business system.
7. A distillation apparatus for an aircraft engine, characterized in that, include: The data acquisition module is used to collect multi-source heterogeneous data in the field of aero-engine maintenance; The reconstruction module is used to perform task-oriented reconstruction of the multi-source heterogeneous data based on the actual business processes in the field of aero-engine maintenance, and obtain a distillation dataset in the form of instruction response. The teacher model inference module is used to perform inference on the distillation dataset through the teacher model to obtain distillation supervision signals; wherein, the distillation supervision signals include at least one of the teacher model's inference results, inference process, and candidate result ranking information; The distillation module is used to perform distillation training on the student model based on the distillation supervision signal, and then apply the student model that has completed the distillation training to the field of aero-engine maintenance.
8. A computer embedded device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the method according to any one of claims 1 to 6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 6.