Fault diagnosis and analysis method and system for turn-to-turn short circuit of permanent magnet synchronous motor

Abstract

The invention discloses a permanent magnet synchronous motor turn-to-turn short circuit fault diagnosis analysis method and system, and belongs to the technical field of motor fault diagnosis, and the method comprises the following steps: S1, building a fine tuning data set in the field of permanent magnet synchronous motor turn-to-turn short circuit faults, the data set comprising a fault question and answer data set and a fault current diagnosis data set; and S2, loading a pre-trained large model weight, and performing fine adjustment on the large model weight by adopting the data set in the S1. According to the invention, a permanent magnet synchronous motor turn-to-turn short circuit fault diagnosis experiment can be carried out in a laboratory environment by using the device, and signals having the same information features as those in an actual industrial condition are obtained, so that a basic sample set is provided for training a deep learning network for turn-to-turn short circuit fault diagnosis; and a basic experiment platform and means are provided for design and optimization of the permanent magnet synchronous motor, and a fine-tuning large model can better carry out real-time turn-to-turn short circuit fault diagnosis on the PMSM on a related real-time prediction system.

Description

Technical Field

[0001] The present invention belongs to the technical field of motor fault diagnosis, and in particular relates to a fault diagnosis and analysis method and system for a permanent magnet synchronous motor turn-to-turn short circuit. Background Art

[0002] Permanent magnet synchronous motors (PMSMs) have been widely used in many industrial scenarios due to their high efficiency, easy maintenance, and low rotor inertia, such as manufacturing systems, electric vehicles, wind turbines, and ship propulsion engines. However, inter-turn short circuit (ITSC) fault, as one of the most common faults of PMSMs, may further develop into catastrophic consequences, paralyzing the motor system and even threatening the safety of personnel. In order to ensure the high stability and reliability of PMSMs, it is very important to use a reasonable fault diagnosis system to monitor their conditions in a timely and accurate manner.

[0003] In actual industrial application scenarios, the severity of the inter-turn short-circuit fault of the permanent magnet synchronous motor is different, and the corresponding symptoms are also different. Accurately diagnosing and analyzing the inter-turn short-circuit fault of the motor and providing corresponding treatment methods require professional knowledge in the field of motor faults.

[0004] The Chinese patent number CN111551849 applied by Tianjin University was filed on April 9, 2020 and granted on August 18, 2020. The name of the invention is "Online diagnosis method for inter-turn short-circuit fault of dual-Y phase-shifted 30° dual-redundant permanent magnet synchronous motor". The case discloses a method for judging the inter-turn short-circuit fault of the coil based on the relationship between the absolute value of the arithmetic mean of the square difference of the spatial voltage vector modulus and the square ratio of the angular frequency of the two sets of three-phase windings in the motor. The disadvantage is that this method depends on the specific motor structure and parameters, and changes in the conditions during the operation of the motor may affect the accuracy of the diagnosis results.

[0005] The Chinese patent number CN118731690 applied by North China Electric Power University (Baoding) was applied on September 2, 2024, and the authorization date was November 5, 2024. The name of the invention is generator stator turn-to-turn short-circuit fault diagnosis method, device, system and medium. The case discloses a generator stator turn-to-turn short-circuit fault diagnosis method, device, system and medium. On the basis of generating high-quality data expansion samples, the dual-path wavelet method is used to extract data from multiple types of sensors and then perform feature fusion and fault diagnosis. Its shortcomings are that it uses two data sources, current and vibration, which affects the operation of the motor, and it cannot have the text output function of a large model.

[0006] The Chinese patent number CN118918923 applied by Beijing University of Chemical Technology was filed on August 19, 2024, and published on November 8, 2024. The name of the invention is a method for early warning and diagnosis of rotating machinery faults based on fine-tuning of a large speech model. The case discloses a method for early warning and diagnosis of rotating machinery faults based on fine-tuning of a large speech model. The method extracts richer features by utilizing the generalization ability of a pre-trained large speech model, and only requires fine-tuning of a lightweight autoencoder, which significantly reduces training time and cost. The disadvantage is that the use of a large speech model has a greater cost and training complexity, and the ability of the large model to organize language output is not fully utilized to achieve full utilization of the performance of the large model.

[0007] The Chinese patent number CN118729471 applied by Zhuhai Gree Electric Appliances Co., Ltd. was applied for on July 22, 2024, and the publication date was October 1, 2024. The name of the invention is air conditioning and its fault analysis method, device, storage medium and program product. The case discloses a large model-based air conditioning fault analysis method. This method calls the pre-trained large vertical model in the air conditioning field to output the corresponding fault detection or fault diagnosis results, and calls the large vertical model to output the corresponding processing suggestions based on the obtained fault prediction results. Its shortcomings are that a fault diagnosis model other than the large model is used to assist in fault diagnosis, and the large model itself is not used to diagnose faults, and the model structure is redundant.

[0008] The existing technologies have the following defects that need to be solved: 1. Model-based methods are affected by motor parameters and structure. 2. Most data-based fault diagnosis methods can only output fault categories, and ultimately experts in professional fields are still required to specify treatment plans. 3. They fail to fully utilize the ability of large models in language organization output and are lacking in providing detailed, understandable fault descriptions and solutions. 4. They fail to reasonably convert fault data into natural language descriptions and implement fault diagnosis functions through the large model itself.

[0009] Based on this, the present invention designs a fault diagnosis and analysis method and system for inter-turn short circuit of a permanent magnet synchronous motor to solve the above-mentioned problem. Summary of the invention

[0010] The purpose of the present invention is to solve the problems in the above-mentioned background technology and to propose a fault diagnosis and analysis method and system for turn-to-turn short circuit of a permanent magnet synchronous motor.

[0011] In order to achieve the above object, the present invention adopts the following technical solutions:

[0012] A fault diagnosis and analysis method for a permanent magnet synchronous motor turn-to-turn short circuit, characterized in that it comprises the following steps:

[0013] Step S1, forming a fine-tuning data set in the field of permanent magnet synchronous motor turn-to-turn short circuit fault, the data set including a fault question-and-answer data set and a fault current diagnosis data set;

[0014] Step S2, load the pre-trained large model weights and fine-tune them using the data set in S1;

[0015] Step S3, receiving user question input and determining the input type; when the input is a question related to motor failure, calling the fine-tuned large model in step S2 to output a corresponding answer; when the input includes motor current signal data and fault diagnosis requirements, executing step S4;

[0016] Step S4, performing data processing on the motor current signal included in the input in step S3, calling the corresponding fault level of the fine-tuned large model output current signal in S2 and the corresponding answer to the question included in the input.

[0017] Optionally, the step S1 specifically includes:

[0018] Step S1-1, collecting three-phase current data of the permanent magnet synchronous motor and performing data preprocessing;

[0019] Step S1-2, constructing a fault field question-answering dataset of knowledge related to the background, causes, symptoms, and solutions of turn-to-turn short circuit faults;

[0020] Step S1-3, combining the current data processed in step S1-1 with the corresponding text description of the fault severity level, cause, symptom, and recommended treatment method to construct a fault current diagnosis data set;

[0021] Step S1-4: Combine the fault field question-answering dataset and the fault current diagnosis dataset into a large model fine-tuning dataset.

[0022] Optionally, the step S2 specifically includes:

[0023] Step S2-1, obtain a question-answering dataset in the field of turn-to-turn short-circuit faults, call the preset large language model weights, and obtain a basic model in the field of turn-to-turn short-circuit faults;

[0024] Step S2-2, obtain a fault current diagnosis data set, process the fault current signal, convert it into data in natural language format, and use the corresponding fault description as a label, input it into the basic model for supervised training until the model performance reaches the preset standard, and obtain a large model for permanent magnet synchronous motor turn-to-turn short-circuit fault diagnosis.

[0025] Optionally, when the input is a question related to a motor failure, the fine-tuned large model is called to output a corresponding answer, and the step S3 specifically includes:

[0026] Step S3-1, semantically understand the user's question input to determine whether it is a question related to motor failure;

[0027] Step S3-2, input the question into the model fine-tuned in step S2, and obtain the answer output by the model;

[0028] Step S3-3: post-process the answer output by the model, integrate the information, generate a coherent text, and finally output it to the user.

[0029] Optionally, when the input includes motor current signal data and fault diagnosis requirements, calling the fine-tuned large model to perform fault diagnosis specifically includes:

[0030] Perform data processing on the input current signal, including denoising, standardization and dimensionality reduction transformation;

[0031] The processed data is combined with the user's question input and input into the model fine-tuned in step S2 to obtain the answer to the model output;

[0032] The answers output by the model are post-processed to integrate information, generate coherent text, and finally output to the user.

[0033] The diagnostic analysis method further includes: performing data processing on the input current signal, including:

[0034] The sliding window average denoising method is used to remove the noise from the input three-phase current signal of the permanent magnet synchronous motor;

[0035] From the three-phase current data after noise removal, the short-circuit fault phase current and another healthy phase current are selected for phase shift difference analysis. The expression is:

[0036]

[0037]

[0038]

[0039] In the formula, and Represent the phases of the fault phase A and the other healthy phase B respectively, and x represents the data obtained after phase shift difference analysis. The data is then normalized;

[0040] The normalized data is subjected to discrete Fourier transform and main frequency component extraction to achieve dimensionality reduction. The expression is:

[0041]

[0042]

[0043] X'[k]=X[k], k=0, 1,...,M-1

[0044] Where N is the normalized data length, Re(X[k]) and Im(X[k]) represent the length of the data.

[0045] X is the real and imaginary parts of the data after discrete Fourier transformation, and M is the number of points containing the main frequency information in the transformed data.

[0046] The fault diagnosis and analysis system of the inter-turn short circuit of the permanent magnet synchronous motor includes:

[0047] A data acquisition unit, used for acquiring three-phase current data of the permanent magnet synchronous motor;

[0048] A data processing unit, used for processing the collected data;

[0049] Model unit, which contains a pre-trained and fine-tuned large model for outputting fault levels and answering user questions;

[0050] A receiving unit, used for receiving questions;

[0051] A first output unit, for calling a pre-trained large model to output a corresponding answer result when the question is a question related to a motor failure;

[0052] The second output unit is used to call the fine-tuned large model to perform fault diagnosis when the question contains motor current signal data and fault diagnosis requirements, and output corresponding answers to corresponding fault levels and related questions.

[0053] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0054] 1. In the present invention, the permanent magnet synchronous motor fault experimental device used can simulate the permanent magnet synchronous motor turn-to-turn short-circuit fault in a laboratory environment. The device can be used to carry out the permanent magnet synchronous motor turn-to-turn short-circuit fault diagnosis experiment in a laboratory environment, and obtain signals with the same information characteristics as those under actual industrial conditions, thereby providing a basic sample set for training the deep learning network for turn-to-turn short-circuit fault diagnosis, and providing a basic experimental platform and means for the design and optimization of permanent magnet synchronous motors, which helps the fine-tuned large model to better perform real-time turn-to-turn short-circuit fault diagnosis on PMSM on the relevant real-time prediction system;

[0055] The construction of the fine-tuning data set in the fault diagnosis and analysis method of the permanent magnet synchronous motor turn-to-turn short circuit is based on the specific performance of the permanent magnet synchronous motor fault experimental device under fault conditions and the relevant descriptions of the turn-to-turn short circuit fault in relevant literature and books. It can comprehensively summarize the symptoms, causes and solutions corresponding to different fault degrees from multiple angles, and professionals without domain knowledge can quickly handle real-time sudden turn-to-turn short circuit faults.

[0056] The three-phase current data of the permanent magnet synchronous motor is used as the fault field data source, which reduces the impact of background noise and other mechanical interference. By integrating Hall sensors in the inverter to collect the stator current of the motor, there is no need to install additional sensors, which reduces the impact on the operation of the motor.

[0057] The fault data processing method adopted by the present invention can perform denoising, dimension reduction and feature extraction operations on the three-phase current data of the motor, and describe the motor fault data through natural language as an effective input for fine-tuning the large model;

[0058] According to the technical solution of the present invention, by building a large model for motor fault diagnosis and analysis, the problems of untimely and inaccurate analysis of turn-to-turn short-circuit faults in this field, which require professional motor knowledge, are solved, and an efficient fault analysis platform for the air-conditioning field is provided.

[0059] 2. In the present invention, by constructing a fine-tuning data set and loading a pre-trained large model, the present invention can realize intelligent diagnosis of turn-to-turn short-circuit faults and improve the accuracy of diagnosis. The present invention does not require professional knowledge in the field of motor faults and can quickly handle real-time sudden turn-to-turn short-circuit faults, thereby reducing dependence on professionals. Through automated and intelligent diagnostic processes, the present invention can quickly output diagnostic results and answer questions, thereby improving the efficiency of fault diagnosis. The method of the present invention can be extended to other types of motor fault diagnosis and has good scalability and application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A schematic diagram of a method and system for fault diagnosis and analysis of inter-turn short circuit of a permanent magnet synchronous motor proposed by the present invention;

[0061] Figure 2 A flow chart of the steps of training a turn-to-turn short circuit fault diagnosis and analysis model based on a preset language model for a permanent magnet synchronous motor turn-to-turn short circuit fault diagnosis and analysis method and system proposed by the present invention;

[0062] Figure 3 A flow chart of the steps of processing data of three-phase current of a permanent magnet synchronous motor in a method and system for fault diagnosis and analysis of inter-turn short circuit of a permanent magnet synchronous motor proposed by the present invention;

[0063] Figure 4The original three-phase current of the permanent magnet synchronous motor and the current characteristic sample diagram after data processing used in the fault diagnosis and analysis method and system for the inter-turn short circuit of the permanent magnet synchronous motor proposed in the present invention, wherein (a) is a healthy state diagram, (b) is a minor fault diagram, (c) is a moderate fault diagram, and (d) is a severe fault diagram;

[0064] Figure 5 A diagram showing the training process of the fault diagnosis model of the permanent magnet synchronous motor turn-to-turn short circuit fault diagnosis and analysis method and the system proposed by the present invention;

[0065] Figure 6 This is a confusion matrix diagram of the diagnosis results of four fault states for the permanent magnet synchronous motor turn-to-turn short-circuit fault diagnosis and analysis model of the permanent magnet synchronous motor turn-to-turn short-circuit fault diagnosis and analysis method and system proposed in the present invention. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] Please see attached Figure 1 -Attached Figure 6 The present invention provides a technical solution: a fault diagnosis and analysis method for a permanent magnet synchronous motor turn-to-turn short circuit, characterized in that it comprises the following steps:

[0068] Step S1, forming a fine-tuning data set in the field of permanent magnet synchronous motor turn-to-turn short circuit fault, the data set including a fault question-and-answer data set and a fault current diagnosis data set;

[0069] Step S2, load the pre-trained large model weights and fine-tune them using the data set in S1;

[0070] Step S3, receiving user question input and determining the input type; when the input is a question related to motor failure, calling the fine-tuned large model in step S2 to output a corresponding answer; when the input includes motor current signal data and fault diagnosis requirements, executing step S4;

[0071] Step S4, performing data processing on the motor current signal included in the input in step S3, calling the corresponding fault level of the fine-tuned large model output current signal in S2 and the corresponding answer to the question included in the input.

[0072] Specifically, the large model for turn-to-turn short-circuit fault diagnosis and analysis will output the fault diagnosis results through natural language, and provide more detailed symptom descriptions, treatment plans and other additional information based on the instructions in the inquiry, so that personnel who do not have professional knowledge can perform preliminary processing of the fault to avoid further expansion of the fault and cause more serious consequences.

[0073] The above query can achieve multiple rounds of question-answering output. When the input includes current data and fault diagnosis instructions, the fault category result in the first round of output is used as supplementary information for subsequent instruction input, so that the model can maintain the current fault context and output corresponding additional information for appropriate faults. An example is as follows:

[0074] First round output: The current motor inter-turn short-circuit fault level is [serious], and the short-circuit turns ratio is about 25%.

[0075] Second round of input: What are the symptoms of this fault? + (Fault severity is [serious], short-circuit turns ratio 25%)

[0076] Second round of output: Severe faults are manifested in significantly increased vibration and noise of the motor body, severe heating of the coil, etc.

[0077] Third round input: How to deal with the current fault? + (fault level is [serious], short-circuit turns ratio 25%)

[0078] The third round of output: For severe failures, it is necessary to...

[0079] The step S1 specifically includes:

[0080] Step S1-1, collecting three-phase current data of the permanent magnet synchronous motor and performing data preprocessing;

[0081] Step S1-2, constructing a fault field question-answering dataset of knowledge related to the background, causes, symptoms, and solutions of turn-to-turn short circuit faults;

[0082] Specifically, corpus data in the field of turn-to-turn short-circuit faults are collected, for example, it may include at least one of professional literature, industrial cases, and question-and-answer forum content. The preset language model may be a general large language model, such as GPT (Generative Pre-Trained) released by Openai and BERT (Bidirectional Encoder Representations from Transformers) released by Google. Based on the corpus data in the field of turn-to-turn short-circuit fault diagnosis, a fault field question-and-answer data set of turn-to-turn short-circuit fault-related knowledge is constructed, and the general large model is called for fine-tuning training to obtain a basic model in the field of turn-to-turn short-circuit faults.

[0083] Specifically, the fine-tuning training of the large model includes using the fine-tuning data set, inputting it into the large model to output the result, and calculating the difference between the predicted result and the actual target through the loss function, that is, the loss value. According to the size of the loss value, the parameter weights of the model are adjusted in reverse to achieve the purpose of optimizing the model. This process will continue until the loss value gradually stabilizes and converges.

[0084] The fine-tuning dataset is specifically: multiple groups of professional dialogue data in the field of turn-to-turn short-circuit faults that contain inquiry instructions and corresponding answers. These inquiry instructions cover various knowledge introductions, functional consultations, troubleshooting and other scenarios for motor turn-to-turn short-circuit faults, and the corresponding answers are condensed from professional literature, industrial cases, domain experts and other sources to obtain accurate answers. The corpus data composed of inquiry instructions and corresponding answers can be regarded as labeled data, and the large model can be supervised for training, so that it can learn the corresponding knowledge in the field of turn-to-turn short-circuit faults and have the ability to make accurate corresponding answers to user inquiries.

[0085] Step S1-3, combining the current data processed in step S1-1 with the corresponding text description of the fault severity level, cause, symptom, and recommended treatment method to construct a fault current diagnosis data set;

[0086] Step S1-4: Combine the fault field question-answering dataset and the fault current diagnosis dataset into a large model fine-tuning dataset.

[0087] The step S2 specifically includes:

[0088] Step S2-1, obtain a question-answering dataset in the field of turn-to-turn short-circuit faults, call the preset large language model weights, and obtain a basic model in the field of turn-to-turn short-circuit faults;

[0089] Step S2-2, obtain a fault current diagnosis data set, process the fault current signal, convert it into data in natural language format, and use the corresponding fault description as a label, input it into the basic model for supervised training until the model performance reaches the preset standard, and obtain a large model for permanent magnet synchronous motor turn-to-turn short-circuit fault diagnosis.

[0090] When the input is a question related to a motor failure, the fine-tuned large model is called to output a corresponding answer. The step S3 specifically includes:

[0091] Step S3-1, semantically understand the user's question input to determine whether it is a question related to motor failure;

[0092] Step S3-2, input the question into the model fine-tuned in step S2, and obtain the answer output by the model;

[0093] Step S3-3: post-process the answer output by the model, integrate the information, generate a coherent text, and finally output it to the user.

[0094] When the input includes motor current signal data and fault diagnosis requirements, the fine-tuned large model is called to perform fault diagnosis, specifically including:

[0095] Perform data processing on the input current signal, including denoising, standardization and dimensionality reduction transformation;

[0096] The processed data is combined with the user's question input and input into the model fine-tuned in step S2 to obtain the answer to the model output;

[0097] The answers output by the model are post-processed to integrate information, generate coherent text, and finally output to the user.

[0098] Specifically, the three-phase current data of the permanent magnet synchronous motor is subjected to data processing such as denoising, dimension reduction, and feature extraction to reduce the complexity of the current data, so that the processed current data can be described by natural language, and then the fault diagnosis instruction data with a coherent text description is formed in combination with the fault level corresponding to the current data to form a fault current diagnosis data set;

[0099] Then the basic model of turn-to-turn short-circuit fault is called, and further fine-tuned and trained on it to obtain a large model for diagnosis and analysis of turn-to-turn short-circuit fault of permanent magnet synchronous motor;

[0100] Specifically, the fault diagnosis instruction data sample includes a diagnostic instruction query with current data description and an instruction output corresponding to the fault level, and multiple rounds of dialogue can be conducted to provide corresponding fault symptoms, solutions and other information. For example, the dialogue is:

[0101] Question: The current current is [X1, X2, X3, ..., Xn]. What is the current degree of the corresponding inter-turn short-circuit fault?

[0102] Answer: The current motor inter-turn short-circuit fault is serious, and the short-circuit turns ratio is about 25%.

[0103] Q: What are the symptoms of this fault?

[0104] Answer: The specific manifestations of this fault are significantly increased vibration and noise of the motor body, severe heating of the coil, etc.

[0105] In the above dialogue case, the current description is the current characteristic data after data processing. Specifically, Figure 3 As shown;

[0106] The diagnostic analysis method further includes: performing data processing on the input current signal, including:

[0107] The sliding window average denoising method is used to remove the noise from the input three-phase current signal of the permanent magnet synchronous motor;

[0108] From the three-phase current data after noise removal, the short-circuit fault phase current and another healthy phase current are selected for phase shift difference analysis. The expression is:

[0109]

[0110]

[0111]

[0112] In the formula, and Represent the phases of the fault phase A and the other healthy phase B respectively, and x represents the data obtained after phase shift difference analysis. The data is then normalized;

[0113] The normalized data is subjected to discrete Fourier transform and main frequency component extraction to achieve dimensionality reduction. The expression is:

[0114]

[0115]

[0116] X'[k]=X[k], k=0, 1,...,M-1

[0117] Where N is the normalized data length, Re(X[k]) and Im(X[k]) represent the length of the data.

[0118] X is the real and imaginary parts of the data after discrete Fourier transformation, and M is the number of points containing the main frequency information in the transformed data.

[0119] The fault diagnosis and analysis system of the inter-turn short circuit of the permanent magnet synchronous motor includes:

[0120] A data acquisition unit, used for acquiring three-phase current data of the permanent magnet synchronous motor;

[0121] A data processing unit, used for processing the collected data;

[0122] Model unit, which contains a pre-trained and fine-tuned large model for outputting fault levels and answering user questions;

[0123] A receiving unit, used for receiving questions;

[0124] A first output unit, for calling a pre-trained large model to output a corresponding answer result when the question is a question related to a motor failure;

[0125] The second output unit is used to call the fine-tuned large model to perform fault diagnosis when the question contains motor current signal data and fault diagnosis requirements, and output corresponding answers to corresponding fault levels and related questions.

[0126] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A fault diagnosis and analysis method for a permanent magnet synchronous motor turn-to-turn short circuit, characterized in that: The following steps are involved: Step S1, forming a fine-tuning data set in the field of permanent magnet synchronous motor turn-to-turn short circuit fault, the data set including a fault question-and-answer data set and a fault current diagnosis data set; Step S2, load the pre-trained large model weights and fine-tune them using the data set in S1; Step S3, receiving user question input and determining the input type; when the input is a question related to motor failure, calling the fine-tuned large model in step S2 to output a corresponding answer; when the input includes motor current signal data and fault diagnosis requirements, executing step S4; Step S4, performing data processing on the motor current signal included in the input in step S3, calling the corresponding fault level of the fine-tuned large model output current signal in S2 and the corresponding answer to the question included in the input.

2. The fault diagnosis and analysis method for inter-turn short circuit of a permanent magnet synchronous motor according to claim 1 is characterized in that: The step S1 specifically includes: Step S1-1, collecting three-phase current data of the permanent magnet synchronous motor and performing data preprocessing; Step S1-2, constructing a fault field question-answering dataset of knowledge related to the background, causes, symptoms, and solutions of turn-to-turn short circuit faults; Step S1-3, combining the current data processed in step S1-1 with the corresponding text description of the fault severity level, cause, symptom, and recommended treatment method to construct a fault current diagnosis data set; Step S1-4: Combine the fault field question-answering dataset and the fault current diagnosis dataset into a large model fine-tuning dataset.

3. The fault diagnosis and analysis method for inter-turn short circuit of a permanent magnet synchronous motor according to claim 1 is characterized in that: The step S2 specifically includes: Step S2-1, obtain a question-answering dataset in the field of turn-to-turn short-circuit faults, call the preset large language model weights, and obtain a basic model in the field of turn-to-turn short-circuit faults; Step S2-2, obtain a fault current diagnosis data set, process the fault current signal, convert it into data in natural language format, and use the corresponding fault description as a label, input it into the basic model for supervised training until the model performance reaches the preset standard, and obtain a large model for permanent magnet synchronous motor turn-to-turn short-circuit fault diagnosis.

4. The fault diagnosis and analysis method for inter-turn short circuit of a permanent magnet synchronous motor according to claim 1 is characterized in that: When the input is a question related to a motor failure, the fine-tuned large model is called to output a corresponding answer. The step S3 specifically includes: Step S3-1, semantically understand the user's question input to determine whether it is a question related to motor failure; Step S3-2, input the question into the model fine-tuned in step S2, and obtain the answer output by the model; Step S3-3: post-process the answer output by the model, integrate the information, generate a coherent text, and finally output it to the user.

5. The fault diagnosis and analysis method for inter-turn short circuit of a permanent magnet synchronous motor according to claim 1 is characterized in that: When the input includes motor current signal data and fault diagnosis requirements, the fine-tuned large model is called to perform fault diagnosis, specifically including: Perform data processing on the input current signal, including denoising, standardization and dimensionality reduction transformation; The processed data is combined with the user's question input and input into the model fine-tuned in step S2 to obtain the answer to the model output; The answers output by the model are post-processed to integrate information, generate coherent text, and finally output to the user.

6. The fault diagnosis and analysis method for inter-turn short circuit of a permanent magnet synchronous motor according to claim 1 is characterized in that: The diagnostic analysis method further includes: performing data processing on the input current signal, including: The sliding window average denoising method is used to remove the noise from the input three-phase current signal of the permanent magnet synchronous motor; From the three-phase current data after noise removal, the short-circuit fault phase current and another healthy phase current are selected for phase shift difference analysis. The expression is: In the formula, and They represent the phases of the faulty phase A and the other healthy phase B, respectively, and x represents the data obtained after the phase shift difference analysis, which is then normalized; The normalized data is subjected to discrete Fourier transform and main frequency component extraction to achieve dimensionality reduction. The expression is: X′[k]=X[k],k=0,1,...,M-1 Where N is the normalized data length, Re (X[k]) and Im (X[k]) They represent the real and imaginary parts of the data after discrete Fourier transform of the data X, and M is the number of points containing the main frequency information in the transformed data.

7. A fault diagnosis and analysis system for inter-turn short circuit of a permanent magnet synchronous motor, characterized in that: include: A data acquisition unit, used for acquiring three-phase current data of the permanent magnet synchronous motor; A data processing unit, used for processing the collected data; Model unit, which contains a pre-trained and fine-tuned large model for outputting fault levels and answering user questions; A receiving unit, used for receiving questions; A first output unit, for calling a pre-trained large model to output a corresponding answer result when the question is a question related to a motor failure; The second output unit is used to call the fine-tuned large model to perform fault diagnosis when the question contains motor current signal data and fault diagnosis requirements, and output corresponding answers to corresponding fault levels and related questions.

Images