Generator fault diagnosis method and related device

Abstract

The invention belongs to a fault diagnosis method, and provides a generator fault diagnosis method and a related device for solving the technical problems that an existing automatic diagnosis scheme is difficult to cover complex fault scenes, or the professional degree requirement is extremely high, and the decision-making process in the diagnosis process is opaque. And inputting into a pre-trained fault diagnosis agent to obtain an interpretable reasoning path and a diagnosed fault occurrence reason. Wherein when the fault diagnosis agent is trained, a historical fault case file is analyzed through a large language model, a knowledge base composed of fault reasoning paths is obtained, the knowledge base composed of the obtained fault reasoning paths serves as a learning target of the fault diagnosis agent, and a more intelligent guarantee is provided for safe and stable operation of power generation equipment.

Description

Technical Field

[0001] This application pertains to a fault diagnosis method, specifically relating to a generator fault diagnosis method and related devices. Background Technology

[0002] As core equipment in the power system, the operational reliability of generator sets is of paramount importance. Currently, fault diagnosis in power plants mainly relies on maintenance engineers to comprehensively analyze diverse and heterogeneous data, including vibration characteristics, fluid analysis, electrical parameters, temperature monitoring, and historical maintenance records. This model, which depends on the experience of senior experts, has significant bottlenecks: on the one hand, expert knowledge is difficult to standardize and efficiently transfer; on the other hand, manual analysis is inefficient and may delay the best handling time in emergency situations.

[0003] Existing automated diagnostic solutions, such as rule-based expert systems, suffer from high rule base maintenance costs and struggle to cover complex fault scenarios like generator excitation system anomalies and bearing Babbitt alloy wear. Supervised learning models require large amounts of labeled, high-quality data, but actual production fault reports are mostly unstructured text. Manually labeling precise causes like "rotor turn-to-turn short circuit" or "cooling water system blockage" requires extensive expertise, is extremely costly, and the model's decision-making process is opaque, often considered a "black box," leading to low trust from field engineers. Therefore, there is an urgent need for an intelligent diagnostic solution that can automatically learn from historical data, has a clear and transparent reasoning process, and adapts to new fault patterns. Summary of the Invention

[0004] This application addresses the technical problem that existing automated diagnostic solutions are unable to cover complex fault scenarios or require extremely high levels of expertise, and that the decision-making process during diagnosis is opaque. It provides a generator fault diagnosis method and related devices.

[0005] To achieve the above objectives, this application adopts the following technical solution: Firstly, this application proposes a generator fault diagnosis method, including: Obtain a fault description; The fault description is input into a pre-trained fault diagnosis agent to obtain an interpretable reasoning path and the diagnosed cause of the fault. The training method for the fault diagnosis agent includes: By parsing historical failure case files using a large language model, a knowledge base composed of failure reasoning paths is obtained; The knowledge base composed of the obtained fault reasoning paths is used as the learning target of the fault diagnosis agent. The fault diagnosis agent is trained to obtain a pre-trained fault diagnosis agent.

[0006] Furthermore, the historical fault case files are in the form of unstructured files.

[0007] Furthermore, the method for parsing historical fault case files using a large language model includes: We employ a large language model fine-tuned with domain knowledge and use thought chain technology to analyze historical failure case files.

[0008] Furthermore, the method for training the fault diagnosis agent includes: Construct a simulated diagnostic environment and set the agent's state to a new fault phenomenon description. The agent's actions are to generate the next reasoning direction or select the next fault feature to focus on.

[0009] Furthermore, the reward function used in training the fault diagnosis agent is: The closer the logical similarity between the reasoning path generated by the agent and the standard reasoning path in the knowledge base, which consists of the fault reasoning path, and the more accurate the fault diagnosis, the higher the reward.

[0010] Furthermore, the training of the fault diagnosis agent also includes: If the fault diagnosis agent skips any reasoning path and directly and incorrectly derives the cause of the fault, a penalty will be imposed.

[0011] Furthermore, after obtaining the interpretable reasoning path and the diagnosed cause of the fault, the method further includes: determining the maintenance sequence based on the obtained interpretable reasoning path.

[0012] Secondly, this application proposes a generator fault diagnosis system, comprising: The data module is used to obtain fault descriptions; The diagnostic module is used to input the fault description into a pre-trained fault diagnosis agent to obtain an interpretable reasoning path and the diagnosed cause of the fault. The training method for the fault diagnosis agent includes: By parsing historical failure case files using a large language model, a knowledge base composed of failure reasoning paths is obtained; The knowledge base composed of the obtained fault reasoning paths is used as the learning target of the fault diagnosis agent. The fault diagnosis agent is trained to obtain a pre-trained fault diagnosis agent.

[0013] Thirdly, this application proposes an electronic device, including: a memory and one or more processors; the memory is coupled to the processors; wherein the memory stores computer program code, the computer program code including computer instructions, and when the computer instructions are executed by the processor, the electronic device performs the steps of the generator fault diagnosis method described above.

[0014] Fourthly, this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the generator fault diagnosis method described above.

[0015] Compared with the prior art, this application has the following beneficial effects: This application proposes a generator fault diagnosis method. After obtaining the fault description, it is input into a pre-trained fault diagnosis agent to obtain an interpretable reasoning path and the diagnosed cause of the fault. During the training of the fault diagnosis agent, a large language model is used to parse historical fault case files to obtain a knowledge base composed of fault reasoning paths. This knowledge base serves as the learning target for the fault diagnosis agent. This application automates domain knowledge acquisition, overturning the traditional method of relying on manual knowledge compilation. Secondly, the unique reasoning path generation capability transforms the diagnosis process from a black box to a white box, greatly enhancing the credibility of the results and user acceptance. Finally, since the reinforcement learning agent learns a general fault analysis strategy, even for complex faults not previously encountered in the training database, such as the simultaneous occurrence of "turbine efficiency decline" and "generator power oscillation," the agent can still derive reasonable conclusions by combining learned reasoning modules, demonstrating excellent generalization ability and thus providing a more intelligent guarantee for the safe and stable operation of power generation equipment.

[0016] This application also proposes a generator fault diagnosis system, an electronic device, and a computer-readable storage medium, which possess all the advantages of the aforementioned generator fault diagnosis methods. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic flowchart of the generator fault diagnosis method of this application; Figure 2This is an overall architecture diagram of the generator fault diagnosis method of this application; Figure 3 This is a flowchart of the knowledge extraction process for the large language model in this application embodiment; Figure 4 This is a schematic diagram of the reinforcement learning agent training in the embodiments of this application; Figure 5 This is a schematic diagram of the generator fault diagnosis system of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0020] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0022] In the description of the embodiments of this application, it should be noted that if terms such as "upper," "lower," "horizontal," or "inner" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use, they are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this application. In addition, terms such as "first" and "second" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0023] Furthermore, the use of the term "horizontal" does not imply that the component must be absolutely horizontal, but rather that it can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0024] In the description of the embodiments of this application, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0025] Generator sets are core power equipment in power systems, widely used in various power generation fields such as thermal power, hydropower, nuclear power, and wind power. As the power industry develops towards intelligence and large-scale operation, generator sets are becoming increasingly complex, integrating multiple subsystems such as excitation systems, cooling systems, lubrication systems, and electrical control systems. The various data generated during operation exhibit diverse and heterogeneous characteristics. Currently, in various power production scenarios, generator set fault diagnosis is a core aspect of equipment operation and maintenance management. Its core objective is to promptly identify potential hidden dangers or existing faults, clarify the causes of faults, and guide subsequent maintenance, thereby ensuring the safe and stable operation of the unit and avoiding significant losses such as large-scale power outages and production interruptions caused by shutdowns.

[0026] In current generator set operation and maintenance practices, fault diagnosis generally relies on the experience of senior maintenance engineers. Engineers need to integrate diverse and heterogeneous data from various sources, such as vibration characteristics, oil analysis results, electrical parameters, temperature monitoring data, and historical maintenance records, and make comprehensive judgments based on their accumulated professional knowledge to ultimately determine the cause of the fault and the solution. This diagnostic model, which depends on manual experience, has revealed several prominent problems in practical applications. On the one hand, the diagnostic experience of experts is a product of personal knowledge accumulation, making it difficult to form standardized judgment processes and knowledge systems. This results in extremely low efficiency in the transfer of experience, requiring young engineers to accumulate experience over a long period to reach the diagnostic level of senior experts, making it impossible to quickly fill the gap in professional diagnostic talent within the industry. On the other hand, the manual analysis process is time-consuming, especially when dealing with massive amounts of diverse and heterogeneous data. Engineers need to sift through and compare various types of information one by one, significantly reducing diagnostic efficiency. Furthermore, in the event of an emergency fault in a generator set, every second of delay can lead to the escalation of the fault, causing not only more serious equipment damage but also potentially triggering chain reactions such as grid fluctuations and power outages, missing the optimal fault handling opportunity, and ultimately resulting in huge economic losses and social impacts.

[0027] To address the bottlenecks of manual diagnostic methods, the industry has gradually explored and applied various automated diagnostic solutions, attempting to improve diagnostic efficiency and accuracy through technological means. Among these, early applications include rule-based expert systems. These systems transform the diagnostic experience of senior engineers into a series of fixed judgment rules, building a rule base. When the system receives generator set operating data, it matches the pre-defined conditions in the rule base to arrive at a diagnostic result. In addition, with the development of artificial intelligence technology, machine learning-based diagnostic solutions have also emerged, especially supervised learning models. Supervised learning models require training with a large amount of high-quality data labeled with fault types and causes. After training, new generator set operating data is input into the model, which automatically determines the fault situation. The core idea of ​​these automated solutions is to replace subjective human judgment, achieving rapid fault diagnosis through standardized processes, thereby solving the problems of low efficiency and difficulty in passing on experience in manual diagnosis.

[0028] However, existing automated diagnostic solutions still have many insurmountable shortcomings, failing to fundamentally meet the diagnostic needs of actual production. Existing automated solutions either cannot adapt to complex fault scenarios or are limited by data conditions and lack transparency in decision-making, making it difficult to meet the efficiency, accuracy, and reliability requirements of actual operation and maintenance for fault diagnosis. Therefore, the industry urgently needs an intelligent diagnostic solution that can automatically learn fault characteristics from historical data, has a clear and transparent reasoning process, and can adapt to newly emerging fault patterns.

[0029] Based on the above, this application proposes a generator fault diagnosis method and related apparatus. The following is a detailed description of this application in conjunction with embodiments and accompanying drawings.

[0030] like Figure 1 The diagram shown is a flowchart of a generator fault diagnosis method according to this application, which may include: S101, obtain fault description.

[0031] It's important to note that fault descriptions are textual or structured records of various phenomena, characteristics, and abnormal operating parameters encountered when a generator set malfunctions. These descriptions serve as the foundational input data for subsequent diagnosis. Fault diagnosis presupposes a clear understanding of the abnormal behavior of the target. By collecting relevant descriptive information at the time of the fault, accurate diagnostic criteria are provided to the intelligent agent, ensuring that subsequent reasoning and diagnosis have clear objectives and supporting data. In practical applications, manual input is possible, where on-site maintenance engineers manually input the observed fault phenomena into the system. Alternatively, the system can automatically collect and convert data through equipment monitoring systems. For example, abnormal parameters detected by vibration and temperature sensors can be combined with equipment operating status information to automatically generate standardized fault description text. Voice input is also possible, where engineers verbally describe the fault phenomena, which are then converted into a textual fault description by a speech recognition system. Diverse acquisition methods can be adapted to different on-site scenarios, such as rapid voice input during emergency faults and manual input during routine inspections, enhancing the practicality and flexibility of the solution.

[0032] S102, the fault description is input into the pre-trained fault diagnosis agent to obtain an interpretable reasoning path and the diagnosed cause of the fault.

[0033] It's important to note that the fault diagnosis agent is an AI program with fault diagnosis reasoning capabilities. It can receive fault descriptions and output diagnostic results and reasoning processes. An interpretable reasoning path clearly displays the logical chain of reasoning from the fault description to the fault cause, including diagnostic criteria, rules, and historical case references used in the reasoning process, making the diagnostic process traceable and understandable. The pre-trained fault diagnosis agent has learned and mastered the diagnostic logic and knowledge from historical fault cases. Upon receiving a fault description, it calls upon its internally learned fault reasoning rules and knowledge base to match, analyze, and reason about the input fault description. While determining the cause of the fault, it records and outputs the entire reasoning process, forming an interpretable reasoning path, achieving a dual output of diagnostic results and reasoning basis. Therefore, it achieves automated fault diagnosis, significantly improving diagnostic efficiency, avoiding delays from manual analysis, and saving time for emergency fault handling. Simultaneously, the interpretable reasoning path solves the black-box decision-making problem of traditional supervised learning models, allowing on-site engineers to clearly understand the origin of the diagnostic results, increasing trust in the results, and reducing the workload of manual verification. Accurate fault diagnosis can directly guide subsequent maintenance work, reducing maintenance costs and downtime.

[0034] The training method for the fault diagnosis agent includes: parsing historical fault case files using a large language model to obtain a knowledge base composed of fault reasoning paths; using the obtained knowledge base composed of fault reasoning paths as the learning target of the fault diagnosis agent to train the fault diagnosis agent and obtain a pre-trained fault diagnosis agent.

[0035] The large language model is a large-scale language processing model trained on massive amounts of text data, possessing powerful text parsing, understanding, and generation capabilities. Historical fault case files are text files recording information such as the occurrence time, fault phenomena, diagnostic process, fault causes, and maintenance plans of past generator set faults. The resulting knowledge base stores structured data of diagnostic knowledge, including fault reasoning paths, providing data support for agent training and reasoning. Fault reasoning paths are the logical deduction process from fault phenomena to fault causes, representing a structured representation of expert diagnostic experience. Historical fault case files contain the diagnostic experience and reasoning logic of senior engineers, but this information often exists in unstructured text form, making it difficult for agents to learn directly. The large language model possesses powerful text parsing and semantic understanding capabilities, enabling it to extract key information such as fault phenomena, diagnostic basis, reasoning process, and fault causes from these unstructured texts, organize them into standardized fault reasoning paths, and then integrate these reasoning paths into a structured knowledge base, providing clear and learnable target knowledge for subsequent agent training. Therefore, this method achieves structured and standardized extraction of expert experience from historical failure cases, solving the problem of the difficulty in passing on expert knowledge. Automatic parsing via a large language model significantly reduces the workload and cost of manually compiling historical cases, improving the efficiency of knowledge base construction. The constructed structured knowledge base provides high-quality learning data for subsequent agent training, ensuring that the agent can learn accurate and effective diagnostic knowledge. The core of agent training is to enable it to learn and master the diagnostic logic inherent in the failure reasoning paths in the knowledge base. By using the failure reasoning paths in the knowledge base as learning targets, training tasks are constructed, and appropriate training algorithms are used to adjust the agent's model parameters. This continuously reduces the deviation between the agent's output and the standard reasoning paths and failure causes in the knowledge base, ultimately enabling the agent to possess stable and accurate diagnostic reasoning capabilities, completing the pre-training process.

[0036] The present application will now be described in further detail through a more detailed embodiment.

[0037] The generator fault diagnosis method proposed in this invention comprises a complete process of two stages: offline training and online application. In the offline training stage, the primary task is to deeply mine a large number of accumulated unstructured historical fault case files. These cases may describe complex phenomena such as "generator outlet voltage oscillation accompanied by abnormal rise in stator core temperature." In practical application, a large language model fine-tuned with domain knowledge is used, employing thought chain technology to analyze each case. The large language model not only extracts key fault labels such as "voltage oscillation" and "high core temperature," but more importantly, it constructs a complete reasoning path from phenomenon to root cause. For example, the reasoning path might be: "Voltage oscillation indicates instability in the excitation system or asymmetry in the rotor winding -> combined with high core temperature, it is suspected that a short circuit between rotor turns is causing local overheating, thus affecting the magnetic field symmetry -> the final cause is a short circuit between rotor turns due to insulation aging." Through this process, the messy text report can be transformed into a structured, logically rich knowledge base, with each case containing fault phenomena, abstract labels, reasoning chains, and root causes.

[0038] Subsequently, the core reinforcement learning training phase begins. This embodiment constructs a simulated diagnostic environment where the agent's state is a new description of the fault phenomenon, and its actions are generating the next reasoning direction or selecting the next fault feature to focus on. The core innovation of the training lies in using the standard reasoning path generated by the large language model in the previous stage as the agent's learning target. The reward function is designed such that the closer the agent's generated reasoning steps are logically to the standard path, and the more accurate its final diagnosis, the higher the reward. Conversely, if the agent makes an action such as skipping "checking the passing frequency in the bearing vibration spectrum" and directly drawing an incorrect conclusion, it will be penalized. Through extensive trial-and-error learning, the agent gradually masters a robust fault diagnosis strategy, going beyond simple pattern matching.

[0039] During the online diagnostic phase, when a new fault occurs on-site, such as "a sudden increase in temperature and a drop in lubricating oil pressure in the non-drive end bearing of the generator," a well-trained agent is activated. It receives the fault description and progressively generates a reasoning path similar to that of a human expert, such as, "First, the drop in oil pressure may stem from a clogged filter or a faulty oil pump; however, considering the sudden increase in bearing temperature, a failure in the lubrication circuit is the primary suspicion, leading to insufficient oil supply and overheating friction in the bearing; the focus should be on checking for leaks or blockages in the oil supply lines." Finally, it synchronously outputs this clear, explainable path and the diagnosed root cause, such as "a ruptured oil supply line," providing maintenance personnel with clear decision support. This method significantly improves the automation level of diagnostics and the transparency of decision-making.

[0040] The following is an example of this application; First, we enter the offline training phase.

[0041] Knowledge is extracted from historical fault cases using a large language model, and this knowledge is then used to train a reinforcement learning agent. Implementation requires historical data, such as a maintenance report describing "a generator experiencing continuous hydrogen purity alarms accompanied by a slight increase in vibration." This unstructured text report is input into a large language model fine-tuned using extensive power industry literature and fault cases. Guided by carefully designed prompts, the model initiates a thought chain analysis. It first identifies key fault features in the report, such as "low hydrogen purity" and "increased bearing vibration," as core labels. The model then constructs a logical reasoning path: The first step is to confirm that the decrease in hydrogen purity is the main problem, which will directly affect cooling efficiency and safety. The second step is to correlate the observed increase in vibration, which may be due to changes in gas density caused by a decrease in purity or an imbalance caused by local overheating inside. The third step is to pinpoint the problem to the hydrogen system and related auxiliary systems, and analyze the common causes of decreased purity, such as leaks or dryer failure, while vibration may be a consequence. The fourth step, based on comprehensive assessment, is that the most likely common cause is a leak in the hydrogen sealing system, which allows humid air to intrude.

[0042] Ultimately, the root cause of the model output being consistent with the report was: "Aging and cracking of the hydrogen dryer outlet flange seal".

[0043] In this way, thousands of similar historical reports are automatically transformed into a structured, high-quality knowledge base containing a quadruple of {fault phenomenon, fault label, reasoning path, root cause}.

[0044] Next, we move on to the reinforcement learning training phase. The goal of this phase is to train an agent capable of simulating the aforementioned reasoning process. A virtual fault diagnosis environment is constructed. The agent's state space consists of textual descriptions of various fault phenomena, such as "generator hydrogen pressure is unstable, hydrogen replenishment rate is abnormally high." The agent's action space consists of generating reasoning steps or selecting diagnostic focuses, such as whether the next action is "check the sealing oil system" or "troubleshoot the hydrogen dryer." The key to training lies in the design of the reward function. When the agent faces a state "purity decrease accompanied by vibration," it begins to generate reasoning actions step by step. For each action generated, the system compares its current reasoning path fragment with "standard reasoning paths" for similar faults generated by a large language model in the knowledge base in real time. If the agent's reasoning direction, "prioritizing the elimination of large leaks," is consistent with the logic of the standard path, it receives a positive reward. If it makes unreasonable or leapfrog reasoning, ignoring purity and directly judging it as "rotor dynamic balance problem," it will be penalized. When the agent decides to end the reasoning and provides the final cause, it receives a reward if the cause matches the standard answer. Through a large number of such training rounds, and using algorithms such as proximal policy optimization, the policy network inside the intelligent body is gradually optimized, and eventually learns a robust fault diagnosis strategy.

[0045] Finally, the system enters the online diagnostic phase. When the power plant monitoring system detects new fault symptoms, such as "a sharp drop in generator hydrogen purity within a short period of time," the well-trained agent is activated. The received state is this new fault description. The agent then begins its reasoning process, progressively outputting its reasoning path: "First, the sharp drop in purity indicates an acute leak, not chronic degradation. Second, a leak detector should be used first to check the high-pressure side external pipelines and connecting flanges of the hydrogen system. Third, the sealing oil system should not be prioritized, as its faults are usually accompanied by abnormal oil levels." Ultimately, the agent gives the diagnostic conclusion: "High probability of leakage in the external hydrogen pipeline flange or valve seal." The entire process is clear and transparent, not only providing a conclusion but also demonstrating complete diagnostic logic, greatly enhancing the maintenance personnel's trust in the intelligent diagnostic results and providing clear guidance for their emergency repair work.

[0046] like Figure 5 The diagram shown is a schematic of a generator fault diagnosis system according to this application, which may include: The data module is used to obtain fault descriptions; The diagnostic module is used to input the fault description into a pre-trained fault diagnosis agent to obtain an interpretable reasoning path and the diagnosed cause of the fault. The training method for the fault diagnosis agent includes: By parsing historical failure case files using a large language model, a knowledge base composed of failure reasoning paths is obtained; The knowledge base composed of the obtained fault reasoning paths is used as the learning target of the fault diagnosis agent. The fault diagnosis agent is trained to obtain a pre-trained fault diagnosis agent.

[0047] It should be noted that, in the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of each block is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple blocks may be combined or integrated into another device, or some features may be ignored or not executed. The modules described as separate components may or may not be physically separated. The components shown as modules may be one or more physical units, that is, they may be located in one place or distributed in multiple different places. Some or all of the modules can be selected to achieve the purpose of the solution in this embodiment according to actual needs.

[0048] Furthermore, in the various embodiments of the present invention, the modules can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The integrated unit described above can be implemented in hardware or as a software functional unit.

[0049] This application also provides an electronic device, which may include one or more processors, memory and communication interfaces.

[0050] The memory, communication interface, and processor are coupled together. For example, the memory, communication interface, and processor can be coupled together via a bus.

[0051] The communication interface is used for data transmission with other devices. The memory stores computer program code. This computer program code includes computer instructions, which, when executed by the processor, cause the electronic device to perform the steps of the generator fault diagnosis method described above.

[0052] The processor can be a processor or controller, such as a Central Processing Unit (CPU), a general-purpose processor, a Digital Signal Processor (DSP), an Application-Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with this disclosure. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc. The processor can be used to support an electronic device in performing the method steps provided in the above embodiments.

[0053] The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. These buses can be categorized as address buses, data buses, control buses, etc.

[0054] This application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the generator fault diagnosis method described above.

[0055] The computer-readable storage media involved in this application include random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage media known in the art.

[0056] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A generator fault diagnosis method, characterized in that, include: Obtain a fault description; The fault description is input into a pre-trained fault diagnosis agent to obtain an interpretable reasoning path and the diagnosed cause of the fault. The training method for the fault diagnosis agent includes: By parsing historical failure case files using a large language model, a knowledge base composed of failure reasoning paths is obtained; The knowledge base composed of the obtained fault reasoning paths is used as the learning target of the fault diagnosis agent. The fault diagnosis agent is trained to obtain a pre-trained fault diagnosis agent.

2. The generator fault diagnosis method according to claim 1, characterized in that, The historical failure case files are in unstructured format.

3. The generator fault diagnosis method according to claim 1, characterized in that, The method for parsing historical fault case files using a large language model includes: We employ a large language model fine-tuned with domain knowledge and use thought chain technology to analyze historical failure case files.

4. The generator fault diagnosis method according to claim 1, characterized in that, The method for training the fault diagnosis agent includes: Construct a simulated diagnostic environment and set the agent's state to a new fault phenomenon description. The agent's actions are to generate the next reasoning direction or select the next fault feature to focus on.

5. The generator fault diagnosis method according to claim 4, characterized in that, The reward function used in the training of the fault diagnosis agent is: The closer the logical similarity between the reasoning path generated by the agent and the standard reasoning path in the knowledge base, which consists of the fault reasoning path, and the more accurate the fault diagnosis, the higher the reward.

6. The generator fault diagnosis method according to claim 5, characterized in that, The training of the fault diagnosis agent also includes: If the fault diagnosis agent skips any reasoning path and directly and incorrectly derives the cause of the fault, a penalty will be imposed.

7. The generator fault diagnosis method according to claim 1, characterized in that, After obtaining the interpretable reasoning path and the diagnosed cause of the fault, the method further includes: determining the maintenance sequence based on the obtained interpretable reasoning path.

8. A generator fault diagnosis system, characterized in that, include: The data module is used to obtain fault descriptions; The diagnostic module is used to input the fault description into a pre-trained fault diagnosis agent to obtain an interpretable reasoning path and the diagnosed cause of the fault. The training method for the fault diagnosis agent includes: By parsing historical failure case files using a large language model, a knowledge base composed of failure reasoning paths is obtained; The knowledge base composed of the obtained fault reasoning paths is used as the learning target of the fault diagnosis agent. The fault diagnosis agent is trained to obtain a pre-trained fault diagnosis agent.

9. An electronic device, characterized in that, include: A memory, one or more processors; the memory is coupled to the processors; wherein the memory stores computer program code, the computer program code including computer instructions, and when the computer instructions are executed by the processor, the electronic device performs the steps of the generator fault diagnosis method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the generator fault diagnosis method as described in any one of claims 1-7.

Images