Turbine blade intelligent design system capable of continuously learning from experience

Abstract

The invention discloses a turbine blade intelligent design system capable of continuously learning from experience, and the system comprises a design experience structured extraction module which is responsible for analyzing task demands to generate retrieval vectors, supporting vectors by a multi-modal database, carrying out atlas retrieval, and storing new knowledge to form a closed loop; the whole-process data capture module monitors and collects design data in the whole process; the experience-driven LLM module integrates historical experience and task requirements, and generates an intelligent reasoning prompt to drive a large language model to output an optimization scheme. According to the method, by constructing a dynamic closed-loop intelligent design system, the efficiency and quality of turbine blade design are remarkably improved, and a self-evolution knowledge system continuously accumulates design experience, so that the system capability is continuously enhanced along with use; semantic association and causal reasoning are achieved through the vectorization retrieval and knowledge graph technology, and the design problem is accurately positioned; professional and reliable intelligent suggestions are provided by combining large language model application of real-time context enhancement and periodic model fine tuning.

Description

Technical Field

[0001] This invention relates to the field of intelligent turbine blade design technology, and more particularly to an intelligent turbine blade design system that continuously learns from experience. Background Technology

[0002] Existing intelligent turbine blade design technologies suffer from a fundamental, fatal flaw: static closure and lack of memory accumulation. While current technologies have evolved into two main paradigms—automated processes based on surrogate models (Kriging / neural networks) and numerical optimization algorithms, and knowledge engineering systems based on If-Then rules or decision trees—both are essentially "one-off tools." Their knowledge bases are fixed after deployment, relying entirely on historical data or expert experience injected during the development phase. New design schemes, CFD simulation results, optimization paths, and engineer feedback generated during operation are merely scattered piles of raw logs or unstructured documents, lacking automatic extraction, structured organization, and proactive reuse mechanisms. More seriously, even with the introduction of Large Language Models (LLMs), their application remains at the level of "passive reasoning" by calling pre-stored information, failing to deeply couple with knowledge update mechanisms. This results in key parameter combinations and performance trade-off logic from successful designs not being fed back to the model, and geometric defects and CFD divergence lessons from failed cases not being distilled into avoidance rules. Consequently, the system inevitably repeats the same errors in subsequent tasks, severely limiting its robustness and intelligence.

[0003] The aforementioned technical deficiencies manifest in three distinct levels: First, at the knowledge representation level, experiential assets exist in fragmented and non-semantic forms. Traditional logs or case libraries only support keyword retrieval and simple matching, completely lacking the ability to understand and reason about deep causal semantics such as "leading edge overload - flow separation - efficiency decay," thus preventing the accurate recall of valuable engineering wisdom. Second, at the learning mechanism level, the system lacks continuous learning capabilities. It has neither an automatic closed-loop process for capturing, refining, and storing new knowledge nor a fine-tuning strategy for internalizing incremental experience into model parameters. When faced with new working conditions, new materials, or interdisciplinary constraints, it can only rely on manual rule reconstruction or retraining of proxy models, resulting in extremely poor adaptability. Third, at the intelligent evolution level, although LLM possesses powerful language understanding capabilities, it is treated as a "stateless executor" used only once. Its reasoning context cannot be dynamically integrated with the implicit knowledge of historical successful cases or the negative list of failed cases, making "getting smarter with use" a mere empty promise. The system remains at the passive response level and cannot achieve intelligent leaps through autonomous learning and proactive evolution from experience.

[0004] These fundamental limitations collectively lead to disastrous consequences: the inability to transform successful experiences (such as effective parameter combinations and performance trade-off strategies) or lessons learned from each design iteration into reusable knowledge assets; the inability to dynamically update the system's knowledge base, resulting in poor adaptability to new operating conditions, new materials, or interdisciplinary constraints, still requiring manual intervention to reconstruct rules or retrain proxy models; and, more importantly, the existence of design experience in a fragmented and non-semantic form, making it difficult to efficiently retrieve and reason about, thus hindering the effective transfer of valuable engineering wisdom between teams or projects. This "static knowledge management paradigm" has become a core bottleneck restricting the improvement of turbine blade design efficiency and breakthroughs in innovation capabilities. Summary of the Invention

[0005] This invention provides a turbine blade intelligent design system that learns continuously from experience, in order to solve the technical problems of existing turbine blade design, which cannot systematically accumulate and continuously learn design experience, making it difficult to accumulate, reuse and evolve engineering knowledge, and resulting in scattered storage of relevant information and waste of implicit experience.

[0006] A turbine blade intelligent design system that continuously learns from experience includes:

[0007] The design experience structure extraction module is used to parse turbine blade design task requirements and generate query vectors, supporting multimodal database retrieval; it also performs structured processing on newly generated full lifecycle data during the design process, transforming it into structured knowledge entries and establishing knowledge graph associations.

[0008] A multimodal design experience self-accumulation database is used to receive query vectors and automatically match and return a collection of historical design experience covering the performance parameters of successful cases and the reasons for parameters of failed cases through vector retrieval and graph query. Newly generated structured knowledge entries are stored to complete the design-learning closed loop.

[0009] The full-process data capture module is used to monitor the execution process and results of the design scheme. It automatically captures full lifecycle data, including input parameters, process iteration data, simulation results and user feedback, providing raw data sources for the structured extraction of experience.

[0010] The experience-driven LLM reasoning enhancement module integrates historical design experience with new task requirements to construct context-aware prompts that include task objectives, constraints, and lessons learned. It uses retrieval-enhanced generation technology to generate reasoning prompts with experience references. The reasoning prompts are then input into the large language model LLM to drive the generation of initial design schemes, optimization strategies, or risk warnings for the new task.

[0011] A smart design method for turbine blades that learns continuously from experience includes:

[0012] S1. Receive the turbine blade design task submitted by the user. The design experience structured extraction module parses the task requirements and generates a query vector. The vector is retrieved and the graph is queried through the multimodal design experience self-accumulation database. The system automatically matches and returns a set of historical design experience covering the parameter performance of successful cases and the parameter reasons of failed cases.

[0013] S2, the experience-driven LLM reasoning enhancement module integrates historical design experience sets with new task requirements to construct context-aware prompts that include task objectives, constraints, and historical experience; and uses retrieval enhancement generation technology to generate reasoning prompts with experience reference.

[0014] S3. Input the inference prompts into the Large Language Model (LLM) and generate an initial design scheme, optimization strategy, or risk warning for the new task; input the generated scheme into the automated design process to perform parametric modeling and CFD simulation verification.

[0015] S4. Monitor the execution process and results of the initial design scheme through the full-process data capture module, automatically capture the full life cycle data of the task, and use the design experience structured extraction module to process the newly generated data and transform it into structured knowledge items, and establish knowledge graph associations.

[0016] The full lifecycle data includes input parameters, process iteration data, simulation results, and user feedback.

[0017] S5. Store the generated structured knowledge entries into the multimodal design experience self-accumulation database to complete the design-learning closed loop; based on the preset strategy, periodically start the periodic model fine-tuning process to internalize the accumulated design knowledge into the LLM through incremental fine-tuning, so as to realize the iterative upgrade of the system reasoning ability.

[0018] This invention achieves three fundamental beneficial effects, all stemming from an innovative system architecture rather than a simple aggregation of functions:

[0019] First, it achieves an intelligent leap from "memorylessness" to "continuous evolution." Traditional systems are mostly static models, with no capability growth after task completion. This invention constructs a dynamic closed loop of "design-capture-refinement-storage-application." Its multimodal experience database, through structured, vectorized, and graph-based collaborative storage, transforms tacit knowledge (such as design ideas and root causes of failure) from each practice into explicit assets that machines can understand. This means that the system's knowledge boundary dynamically expands over time, and every engineer interaction and every simulation success or failure becomes evolutionary nourishment. This self-growing knowledge system ensures that the system can learn patterns from repetitive work, and the quality and efficiency of decision-making when facing similar tasks improve non-linearly with experience accumulation, truly achieving "getting smarter with use."

[0020] Secondly, this invention achieves a dual breakthrough in retrieval efficiency and reasoning depth. Traditional solution libraries rely on keyword matching and cannot understand the semantic relationship between "leading edge thickening" and "improvement of stall at high angle of attack." This invention maps text to a high-dimensional semantic space through vectorized storage, enabling accurate retrieval of relevant experience even with vastly different expressions. More importantly, the knowledge graph explicitly constructs causal logic chains such as "parameter A → impact on performance B → risk avoidance D," allowing the system to perform multi-step reasoning, tracing the root cause of problems or predicting design consequences. For example, it can answer "which geometric parameters have caused flow separation in the blades," a deep mechanistic reasoning capability that is completely lacking in existing technologies.

[0021] Third, this invention deeply optimizes the professional efficiency of LLM through "real-time enhancement + periodic fine-tuning." Existing technologies treat LLM as a one-time general-purpose engine, lacking domain specificity. This invention equips it with dynamic "short-term memory": real-time retrieval of the most relevant historical successes and failures, constructing context-aware prompts, ensuring that every decision is based on evidence and effectively avoiding known pitfalls. Simultaneously, through "long-term on-the-job training": periodically using high-quality task-experience data to incrementally fine-tune the model, solidifying universal knowledge into intrinsic parameters. This mechanism of parallel short-term reference and long-term internalization makes LLM both accurate and efficient in highly complex turbine design scenarios, completely avoiding the "illusion" problem, transforming it from a general-purpose tool into a continuously growing professional collaborator. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of a turbine blade intelligent design system that continuously learns from experience in one embodiment of the present invention.

[0024] Figure 2 This is a flowchart of a turbine blade intelligent design method that continuously learns from experience in one embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] In one embodiment, such as Figure 1 As shown, a turbine blade intelligent design system that continuously learns from experience is provided, comprising:

[0027] The design experience structure extraction module is used to parse turbine blade design task requirements and generate query vectors, supporting multimodal database retrieval; it also performs structured processing on newly generated full lifecycle data during the design process, transforming it into structured knowledge entries and establishing knowledge graph associations.

[0028] In one embodiment, the design experience structuring extraction module serves as a core bridge connecting raw data and a self-accumulating database. Its core function is to utilize the semantic understanding and information abstraction capabilities of a Large Language Model (LLM) to transform unstructured and semi-structured data into structured entries that can be stored in the database. This module achieves the explicit expression of design wisdom through a standardized processing flow.

[0029] The information extraction module is used to automatically extract key performance summaries and flow phenomenon descriptions (such as "enhanced vortex in the blade tip channel" and "uneven Mach number distribution") from CFD report text.

[0030] The causal induction module combines process and result data to analyze and infer the causes of failure and generate structured entries. For example, if a CFD calculation diverges, the causal induction module will analyze its final geometric parameters and flow field state, and combine LLM domain knowledge to generate a structured entry such as "Cause: The blade trailing edge is too thin, resulting in poor mesh quality".

[0031] The semantic annotation module is used to automatically assign semantic tags to each design case, such as "high efficiency", "wide operating range", "transonic", and "leading edge reshaping". These semantic tags are vectorized for retrieval.

[0032] The knowledge graph construction module is used to automatically update the extracted entities (such as parameter entities, performance indicator entities, and cause entities) and their relationships to the knowledge graph storage unit.

[0033] A multimodal design experience self-accumulation database is used to receive query vectors and automatically match and return a collection of historical design experience covering the performance parameters of successful cases and the reasons for the parameters of failed cases through vector retrieval and graph query. Newly generated structured knowledge entries are stored to complete the design-learning closed loop.

[0034] In one embodiment, the multimodal design experience self-accumulation database adopts a hybrid architecture. This database is not a single tabular database, but achieves systematic management of design knowledge through three core storage units:

[0035] The structured data storage unit adopts a relational database architecture to systematically store scalar data with a clear format. The scalar data includes the initial constraints of the design task (such as operating parameters such as flow rate, pressure ratio, and speed), the finally determined geometric parameters, and key performance indicators (efficiency, stall margin, total pressure loss coefficient, etc.) to ensure the standardized management of basic design data.

[0036] The vectorized knowledge storage unit is used to construct a vector database for storing high-dimensional semantic information from unstructured and semi-structured data. All unstructured and semi-structured data (design concept descriptions, CFD analysis reports, user feedback, failure causes) are encoded into high-dimensional vectors (Embeddings) through a pre-trained language model (such as an embedded LLM or a dedicated model). Understandably, this mechanism enables the system to perform rapid retrieval based on semantic similarity, rather than simple keyword matching, effectively solving the problem of utilizing implicit experiential knowledge. For example, although the wording of "flow separation caused by blade leading edge overload" and "suction peak on the blade basin side caused by inlet airflow angle mismatch" is different, their vectors are spatially adjacent.

[0037] The knowledge graph storage unit constructs a graph-structured database to store the complex relationships between entities representing design experience. By abstracting each design case, design parameter, performance curve, and failure mode in the database as a node, and defining the causal, correlation, and constraint relationships between them as "edges," a reasonable relational network is formed. Understandably, for example, (Design Case A) → [adopted] → (parameter combination P1), (parameter combination P1) → [caused] → performance index M1), while (parameter combination P1) → [avoided] → (failure mode F1). This causal chain graph structure enables the system to possess deep relational reasoning capabilities, achieving intelligent reuse of design experience.

[0038] The end-to-end data capture module is used to monitor the execution process and results of the design scheme. It automatically captures data throughout the entire lifecycle, including input parameters, process iteration data, simulation results, and user feedback, providing a raw data source for the structured extraction of experience.

[0039] In one embodiment, the full-process data capture module, acting as the system's "sensory organ," operates in a non-intrusive background mode to automate and comprehensively collect data throughout the entire lifecycle of the design task. This includes:

[0040] The initial design element capture module is used to automatically record the initial design elements input by the user and build a basic information framework for the task. The initial design elements include initial design requirements, geometric constraints, and performance objectives.

[0041] The process data tracking module is used to capture in real time the various sets of design parameters tried by the user during the optimization process, the generated intermediate 3D model files, the convergence history of CFD calculations, and residual curves.

[0042] The results data archiving module is used to capture in real time the various sets of design parameters tried by the user during the optimization process, the generated intermediate 3D model files, the convergence history of CFD calculations, and residual curves.

[0043] The data feedback module captures user evaluations of the design results (e.g., "This solution meets the requirements but is difficult to manufacture"), modification records of engineers manually adjusting parameters, and "successful cases" / "failed cases" tags, providing practical verification data for experience accumulation. This module, through a full-element data capture mechanism, builds a complete data foundation for subsequent knowledge mining and model optimization.

[0044] The experience-driven LLM reasoning enhancement module integrates historical design experience with new task requirements to construct context-aware prompts that include task objectives, constraints, and lessons learned. It uses retrieval-enhanced generation technology to generate reasoning prompts with experience references. The reasoning prompts are then input into the large language model LLM to drive the generation of initial design schemes, optimization strategies, or risk warnings for the new task.

[0045] In one embodiment, the experience-driven LLM inference enhancement module further includes:

[0046] The real-time context enhancement module is used when the system receives a new design task. This module first converts the task requirements into a query vector, retrieves the K most relevant historical experiences from the self-accumulated database, and formats the experience into a contextual prompt, which is then input into the LLM along with the new task description.

[0047] The periodic model fine-tuning module is used to periodically organize high-quality data pairs from the self-accumulated database into a fine-tuning dataset. This fine-tuning dataset is then used to incrementally fine-tune the basic LLM, thereby solidifying recurring and universally applicable design knowledge into the model's internal parameters.

[0048] Current turbine blade design systems suffer from significant knowledge management deficiencies: knowledge cannot be systematically accumulated and continuously learned. Key information generated during the design process, such as scheme parameters, performance data, failure reasons, and user feedback, can only be stored in scattered forms as raw logs or unstructured documents. This information cannot be automatically extracted into reasonable knowledge units, nor can it be actively invoked in subsequent tasks, resulting in a significant waste of implicit engineering experience.

[0049] This has triggered a series of chain reactions: due to a lack of knowledge accumulation, the system can only start from scratch and repeat trial and error when faced with similar tasks, significantly reducing design efficiency; in complex scenarios involving multiple operating conditions and interdisciplinary coupling, the lack of historical experience makes it difficult for the system to quickly identify effective design areas or avoid known risks, resulting in slow optimization convergence and poor stability. More seriously, existing systems cannot achieve intelligent evolution that "gets smarter with use"—engineers' long-term accumulated preferences, judgments, and innovative ideas cannot be captured and internalized by the system, and intelligent design tools remain at the level of "one-off tools," unable to grow into collaborators with domain knowledge. This static architecture severely hinders the digital inheritance and cross-team reuse of knowledge assets.

[0050] Therefore, this invention constructs an intelligent turbine blade design system with continuous learning capabilities: by establishing a structured, self-accumulating database to explicitly store design experience, and combining it with the knowledge internalization mechanism of a large language model (LLM), the system can automatically extract, associate, and reuse cross-task experience, optimizing reasoning strategies and design quality through continuous iteration, ultimately achieving autonomous evolution of design intelligence and sustainable value-added of knowledge assets. This system constructs a dynamic, self-accumulating design experience database and deeply couples it with a large language model (LLM), thereby achieving automatic capture, structured storage, intelligent retrieval, and reasoning application of design knowledge, enabling the system to continuously evolve in design practice.

[0051] like Figure 2 As shown, the present invention also provides a turbine blade intelligent design method that continuously learns from experience, comprising the following steps:

[0052] S1. Receive turbine blade design tasks submitted by users. The design experience structure extraction module parses the task requirements and generates query vectors. The vectors are then retrieved and graphs are searched through the multimodal design experience self-accumulation database. The system automatically matches and returns a collection of historical design experience that includes parameter performance of successful cases and parameter reasons for failure cases.

[0053] In one embodiment, step S1 further includes the following sub-steps:

[0054] S101. Receive the turbine blade design task submitted by the user, and have the design experience structured extraction module parse the task requirements and generate a query vector, which is then submitted to the multimodal design experience self-accumulation database.

[0055] S102. The multimodal design experience self-accumulation database returns the set of historical design experiences most relevant to the current task through vector retrieval and graph query. The set of historical design experience includes at least the parameters and performance data of successful cases and the parameters and cause analysis of failed cases.

[0056] S2, the experience-driven LLM reasoning enhancement module integrates historical design experience sets with new task requirements to construct context-aware prompts that include task objectives, constraints, and historical experience; and uses retrieval enhancement generation technology to generate reasoning prompts with experience reference.

[0057] S3. Input the inference prompts into the Large Language Model (LLM) and generate an initial design scheme, optimization strategy, or risk warning for the new task; input the generated scheme into the automated design process to perform parametric modeling and CFD simulation verification.

[0058] S4. The execution process and results of the initial design scheme are monitored through the full-process data capture module, automatically capturing the entire lifecycle data of the task. The newly generated data is processed and transformed into structured knowledge items using the design experience structuring module, and a knowledge graph is established. The full lifecycle data includes input parameters, process iteration data, simulation results, and user feedback.

[0059] S5. Store the generated structured knowledge entries into the multimodal design experience self-accumulation database to complete the design-learning closed loop; based on the preset strategy, periodically start the periodic model fine-tuning process to internalize the accumulated design knowledge into the LLM through incremental fine-tuning, so as to realize the iterative upgrade of the system reasoning ability.

[0060] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0061] In one embodiment, a turbine blade intelligent design system that continuously learns from experience is provided, which corresponds one-to-one with the turbine blade intelligent design method that continuously learns from experience in the above embodiments.

[0062] Specific limitations regarding the intelligent turbine blade design system based on continuous learning from experience can be found in the limitations of the intelligent turbine blade design method based on continuous learning from experience described above, and will not be repeated here. Each module in the aforementioned intelligent turbine blade design system based on continuous learning from experience can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0063] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When executed by a processor, the computer program implements the intelligent turbine blade design method based on continuous learning from experience as described in the above embodiments, for example... Figure 2 As shown, to avoid repetition, it will not be described again here. Alternatively, when the computer program is executed by the processor, it implements the functions of each module / unit in this embodiment of the intelligent turbine blade design system that continuously learns from experience; to avoid repetition, this will not be described again here.

[0064] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0065] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0066] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A turbo blade intelligent design system that continuously learns from experience, characterized by, The application relates to a design experience structured refining module, a multi-modal design experience self-accumulation database, a full-process data capturing module and an experience-driven LLM reasoning enhancement module. The design experience structured refining module is used for parsing a turbine blade design task requirement and generating a query vector, and supports multi-modal database retrieval. The multi-modal design experience self-accumulation database is used for receiving the query vector and performing vector retrieval and graph query, automatically matching and returning a historical design experience set covering successful case parameter performance and failure case parameter reason, and storing newly generated structured knowledge entries to complete a design-learning closed loop. The full-process data capturing module is used for monitoring a design scheme execution process and result, and automatically capturing full-life cycle data including input parameters, process iteration data, simulation results and user feedback, and providing an original data source for experience structured refining. The experience-driven LLM reasoning enhancement module is used for integrating the historical design experience set and the new task requirement, constructing a context-aware prompt containing a task target, a constraint condition and historical experience lessons, generating a reasoning prompt with experience reference by using retrieval enhancement generation technology, inputting the reasoning prompt into a large language model (LLM), and driving generation of an initial design scheme, an optimization strategy or a risk warning for the new task. The design experience structured refining module comprises an information extraction module, a cause-effect induction module and a semantic labeling module.

2. The system for intelligent design of turbine blades learning from experience continuously as claimed in claim 1, wherein, The information extraction module is used for automatically extracting a key performance summary and a flow phenomenon description from a CFD report text. The cause-effect induction module is used for analyzing and inferring failure reasons in combination with process data and result data, and generating a structured entry. The semantic labeling module is used for automatically labeling each design case with a semantic label, and the semantic label is vectorized for retrieval. The knowledge graph construction module is used for automatically updating the extracted entities and their relationships to a knowledge graph storage unit. The multi-modal design experience self-accumulation database comprises a structured data storage unit, a vectorized knowledge storage unit and a knowledge graph storage unit.

3. The system of claim 2, wherein, The structured data storage unit adopts a relational database architecture and is used for systematically storing scalar data with a clear format, and the scalar data includes initial constraint conditions of a design task, finally determined geometric parameters and key performance indicators. The vectorized knowledge storage unit is used for constructing a vector database for storing high-dimensional semantic information for unstructured and semi-structured data, and all unstructured and semi-structured data are encoded into high-dimensional vectors by using a pre-trained language model. The knowledge graph storage unit constructs a graph structure database to store complex relationships between design experience entities, defines causal relationships, correlations, constraints and other correlation relationships between each design case, each design parameter, each performance curve and each failure mode in the database as "edges" to form an inferable relationship network. The full-process data capturing module comprises an initial design element capturing module and a process data tracking module.

4. The system of claim 3, wherein, The initial design element capturing module is used for automatically recording user input initial design elements, and constructing a task basic information framework, wherein the initial design elements include initial design requirements, geometric constraint conditions and performance targets. The process data tracking module is used for capturing each set of design parameters tried by a user in an optimization process, generated intermediate three-dimensional model files, convergence history and residual curve of CFD calculation in real time. ​ A result data archiving module for capturing in real time various sets of design parameters tried by the user in the optimization process, the generated intermediate three-dimensional model files, the convergence history and residual curve of the CFD calculation; A data feedback module for capturing the user's evaluation of the design results, the modification records of the engineer's manual adjustment of parameters, and the labels of "success case" and "failure case" to provide practical verification data for experience accumulation.

5. The system of claim 1, wherein, The experience-driven LLM inference enhancement module includes: A real-time context enhancement module for, when the system receives a new design task, first converting the task requirements into a query vector, retrieving the most relevant K historical experiences in the self-accumulation database, and formatting the experiences into a context prompt to be input into the LLM together with the description of the new task; A periodic model fine-tuning module for periodically arranging high-quality data pairs in the self-accumulation database into a fine-tuning dataset, and using the fine-tuning dataset to incrementally fine-tune the basic LLM, so as to solidify the design knowledge that repeatedly occurs and has universal rules into the internal parameters of the model.

6. A method of intelligent design of turbine blades that learns from experience continuously, characterized in that, The turbine blade intelligent design system for continuous learning from experience according to any one of claims 1-5 comprises: S1, receiving a turbine blade design task submitted by a user, analyzing the task requirements by the design experience structured refining module and generating a query vector, performing vector retrieval and graph query in the multi-modal design experience self-accumulation database, automatically matching and returning a historical design experience set covering the parameters and performance of successful cases and the reasons for the parameters of failed cases; S2, the experience-driven LLM inference enhancement module integrates the historical design experience set and the new task requirements to build a context-aware prompt containing task objectives, constraint conditions and historical experiences, and generates an inference prompt with experience reference using retrieval enhancement generation technology; S3, inputting the inference prompt into the large language model LLM to generate an initial design scheme, optimization strategy or risk warning for the new task; inputting the generated scheme into the automated design process to perform parameterized modeling and CFD simulation verification; S4, monitoring the execution process and results of the initial design scheme through the full-process data capture module, automatically capturing the full life cycle data of the task, processing the newly generated data into structured knowledge items using the design experience structured refining module, and establishing a knowledge graph association; Wherein, the full life cycle data includes input parameters, process iteration data, simulation results and user feedback; S5, storing the generated structured knowledge items into the multi-modal design experience self-accumulation database to complete the design-learning closed loop; periodically starting the periodic model fine-tuning process based on the preset strategy, internalizing the accumulated design knowledge into the LLM through incremental fine-tuning, and realizing the iterative upgrade of the system inference ability.

7. The method of intelligent design of turbine blades learning from experience continuously according to claim 6, characterized in that, The step S1 includes the following steps: S101, receiving a turbine blade design task submitted by a user, analyzing the task requirements by the design experience structured refining module and generating a query vector, and submitting to the multi-modal design experience self-accumulation database; S102, the multi-modal design experience self-accumulation database returns a set of historical design experiences most relevant to the current task through vector retrieval and atlas query, and the set of historical design experiences at least includes parameter and performance data of successful cases and parameter and cause analysis of failed cases.

Images