A method, device, equipment and medium for expanding the capacity of rotating machinery fault data based on AI large model and frequency domain characteristics
Through the data expansion method based on AI large model and frequency domain characteristics, the problem of poor capacity expansion of rotary mechanical fault data is solved, the accuracy of fault diagnosis is improved, and the performance of fault diagnosis model is enhanced.
Patent Information
- Application Number
- CN202410970488.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-19
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-07-19
AI Technical Summary
In the prior art, the effectiveness of rotary machinery fault data expansion is poor, resulting in low accuracy in fault diagnosis.
Using data expansion method based on AI big model and frequency domain features, the fault data set is obtained, the absolute mean processing is calculated to obtain the impact data set, the frequency domain feature is extracted to build a prompt template, and input it into the pre-trained AI big model for expansion, and generate the expansion Dataset.
It improves the effectiveness of data expansion and the accuracy of rotary machinery fault diagnosis, and enhances the performance of the fault diagnosis model.
Smart Images

Figure CN118520302B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of mechanical fault diagnosis, and in particular to a method, device, equipment and medium for expanding the fault data of rotating machinery based on an AI large model and frequency domain characteristics. Background Art
[0002] Rotating machinery is an important piece of industrial equipment. Due to the complex working environment, it is very easy to fail. Currently, the commonly used rotating machinery includes rolling bearings and rotor friction. Rolling bearings are called "industrial joints" and are widely used in many industrial scenarios such as aerospace, wind power generation, and processing and manufacturing. The rotor friction system, as the core of rotating machinery, plays a vital role in the normal operation of mechanical equipment. The failure of rotating machinery and rotor friction faults will directly affect the normal operation of the machinery, resulting in a decrease in production quality and production efficiency. In the fault diagnosis of rotating machinery, data expansion can enhance the limited real fault data through a series of technical means to improve the performance and generalization ability of the fault diagnosis model.
[0003] A large model usually refers to a large language model (LLM), also commonly referred to as an AI large model. The emergence and development of large language models is a milestone in the field of natural language processing and deep learning research. Tracing back to its development trajectory, we can clearly see four significant development stages from statistical language models, neural language models, pre-trained language models to large language models. In this process, the model has gradually developed from simple statistical prediction of word sequence probabilities to understanding and generating language through deep learning methods, and to today's large language models. Currently, commonly used LLMs include GPT-4o, Llama-3-8B, iFlytek Spark Cognitive Large Model, etc.
[0004] At present, Generative Adversarial Network (GAN) is a widely used data augmentation method, but it also has some significant disadvantages. The training process is not only complex and difficult to stabilize, but also prone to mode collapse when expanding data. In addition, the fault sample data is unbalanced and there is a lack of a large language model for data expansion, resulting in poor effectiveness of data expansion. The diagnostic accuracy of the expanded data when used in rotating machinery fault diagnosis is also low. Summary of the invention
[0005] The embodiment of the present invention provides a method for expanding rotating machinery fault data based on an AI large model and frequency domain features to solve the problems of poor effectiveness of data expansion and low accuracy of rotating machinery fault diagnosis in the prior art.
[0006] In a first aspect, an embodiment of the present invention provides a method for expanding the capacity of rotating machinery fault data based on an AI large model and frequency domain features, comprising:
[0007] Get a fault dataset containing a specified number of data samples;
[0008] Calculating the absolute mean of all the data samples, and processing the fault data set according to the absolute mean to obtain an impact data set;
[0009] Extracting frequency domain features according to the fault data set, and constructing a prompt template according to the frequency domain features;
[0010] The fault data set, the impact data set and the prompt template are input into a pre-trained AI large model, and the fault data set is expanded to obtain an expanded data set.
[0011] Further, the fault data set includes a rolling bearing fault data set and a rotor rubbing fault data set, and the impact data set includes a rolling bearing impact data set and a rotor rubbing impact data set.
[0012] Further, the processing of the fault data set according to the absolute mean to obtain the impact data set includes:
[0013] Traverse all data samples in the fault data set;
[0014] When the data sample is greater than or equal to a preset threshold and the absolute value of the data sample is greater than or equal to the absolute mean, the absolute value of the difference between the data sample and the absolute mean is calculated and recorded as the impact data corresponding to the data sample;
[0015] When the data sample is smaller than the preset threshold and the absolute value of the data sample is smaller than the absolute mean, the sum of the data sample and the absolute mean is calculated and recorded as the impact data corresponding to the data sample;
[0016] All the impact data are collected to obtain an impact data set.
[0017] Further, when the fault data set is a rotor rubbing fault data set, extracting frequency domain features according to the fault data set and constructing a prompt template according to the frequency domain features includes:
[0018] Inputting the rotor rubbing fault data set into a pre-trained AI big model;
[0019] Calculating a fast Fourier transform of the rotor rub fault data set;
[0020] Normalizing the frequency spectrum according to the fast Fourier transform and the total number of the data samples to obtain a normalized result;
[0021] Extracting rotor rubbing frequency domain features according to the normalized results; wherein the rotor rubbing frequency domain features include main frequency components, amplitude and amplitude difference;
[0022] A rotor rubbing prompt template is constructed according to the rotor rubbing frequency domain characteristics.
[0023] Further, when the fault data set is a rolling bearing fault data set, extracting frequency domain features according to the fault data set and constructing a prompt template according to the frequency domain features includes:
[0024] Inputting the rolling bearing fault data set into a pre-trained AI big model;
[0025] Performing envelope analysis on the rolling bearing fault data set to obtain an envelope spectrum;
[0026] Extracting the rolling bearing frequency domain characteristics from the envelope spectrum; wherein the rolling bearing frequency domain characteristics include main frequency components, amplitude and amplitude difference;
[0027] A rolling bearing prompt template is constructed according to the rolling bearing frequency domain characteristics.
[0028] Further, the envelope analysis is performed on the rolling bearing fault data set to obtain an envelope spectrum, including:
[0029] Performing bandpass filtering on the rolling bearing fault data set to obtain a filtered signal;
[0030] Performing Hilbert transform on the filtered signal to obtain an envelope signal;
[0031] Perform Fourier transform on the envelope signal to obtain an envelope spectrum.
[0032] Furthermore, the method further comprises:
[0033] The expanded data set is verified using a convolutional neural network and a support vector machine to obtain a verification result.
[0034] In a second aspect, an embodiment of the present invention provides a rotating machinery fault data expansion device based on an AI large model and frequency domain features, comprising:
[0035] A fault data acquisition module is used to acquire a fault data set containing a specified number of data samples;
[0036] An impact data generation module, used for calculating the absolute mean of all the data samples, and processing the fault data set according to the absolute mean to obtain an impact data set;
[0037] A frequency domain feature extraction module, used to extract frequency domain features according to the fault data set, and construct a prompt template according to the frequency domain features;
[0038] The fault data expansion module is used to input the fault data set, the impact data set and the prompt template into a pre-trained AI large model, expand the fault data set, and obtain an expanded data set.
[0039] In a third aspect, an embodiment of the present invention provides an electronic device, including:
[0040] Memory for storing computer programs;
[0041] A processor, configured to execute the computer program;
[0042] Wherein, when the processor executes the computer program, it implements the rotating machinery fault data expansion method based on AI large model and frequency domain characteristics as described in any one of the first aspects above.
[0043] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed, the method for expanding the capacity of rotating machinery fault data based on the AI large model and frequency domain features as described in any one of the first aspects above is implemented.
[0044] Compared with the prior art, an embodiment of the present invention provides a method for expanding rotating machinery fault data based on an AI big model and frequency domain features, which has the beneficial effect of: by obtaining a fault data set containing a specified number of data samples, calculating the absolute mean of all the data samples, processing the fault data set according to the absolute mean to obtain an impact data set, extracting frequency domain features based on the fault data set, and constructing a prompt template based on the frequency domain features, inputting the fault data set, the impact data set and the prompt template into a pre-trained AI big model, expanding the fault data set to obtain an expanded data set, which can improve the effectiveness of data expansion and the accuracy of rotating machinery fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] In order to more clearly illustrate the technical features of the embodiments of the present invention, the drawings required for use in 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 paying any creative work.
[0046] Figure 1It is a flow chart of an embodiment of a method for expanding the capacity of rotating machinery fault data based on an AI large model and frequency domain features provided by the present invention;
[0047] Figure 2 It is a rotor rubbing spectrum diagram of an embodiment of a method for expanding the capacity of rotating machinery fault data based on an AI large model and frequency domain features provided by the present invention;
[0048] Figure 3 It is a rolling bearing envelope spectrum of an embodiment of a method for expanding the capacity of rotating machinery fault data based on an AI large model and frequency domain features provided by the present invention;
[0049] Figure 4 It is a rotor friction expansion data spectrum diagram of an embodiment of a method for expanding the capacity of rotating machinery fault data based on an AI large model and frequency domain characteristics provided by the present invention;
[0050] Figure 5 It is a rolling bearing expansion data envelope spectrum of an embodiment of a method for expanding the fault data of rotating machinery based on an AI large model and frequency domain characteristics provided by the present invention;
[0051] Figure 6 It is a structural schematic diagram of an embodiment of a rotating machinery fault data expansion device based on an AI large model and frequency domain features provided by the present invention;
[0052] Figure 7 It is a structural schematic diagram of an embodiment of an electronic device provided by the present invention. DETAILED DESCRIPTION
[0053] 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.
[0054] In the first aspect, the embodiment of the present invention provides a method for expanding the capacity of rotating machinery fault data based on an AI large model and frequency domain features, see Figure 1 , which is a flow chart of an embodiment of a method for expanding the capacity of rotating machinery fault data based on an AI large model and frequency domain features provided by the present invention.
[0055] like Figure 1 As shown, the method comprises the following steps:
[0056] S1: Get a fault data set containing a specified number of data samples;
[0057] S2: Calculate the absolute mean of all the data samples, and process the fault data set according to the absolute mean to obtain an impact data set;
[0058] S3: extracting frequency domain features according to the fault data set, and constructing a prompt template according to the frequency domain features;
[0059] S4: Input the fault data set, the impact data set and the prompt template into a pre-trained AI large model, expand the fault data set, and obtain an expanded data set.
[0060] In a specific implementation, a fault data set containing a specified number of data samples is first obtained. These data samples may come from monitoring data of actual equipment or from a public data set. The fault data set may include a rolling bearing fault data set and a rotor rub fault data set. The absolute mean of all data samples is calculated for the rolling bearing fault data set and the rotor rub fault data set respectively. The absolute mean is used as a threshold for determining whether the data belongs to impact data. The fault data set is processed according to the absolute mean to obtain an impact data set. The impact data set includes a rolling bearing impact data set and a rotor rub impact data set. The fault data set is input into a pre-trained AI big model to extract frequency domain features, and a prompt template is constructed based on the frequency domain features. The fault data set, the impact data set and the constructed prompt template are input into the AI big model again, and the fault data set is expanded to obtain an expanded data set.
[0061] In summary, a fault data set containing a specified number of data samples is obtained, the absolute mean of all the data samples is calculated, the fault data set is processed according to the absolute mean to obtain an impact data set, the fault data set is input into a pre-trained AI big model, the frequency domain features are extracted, and a prompt template is constructed according to the frequency domain features, the fault data set, the impact data set and the prompt template are input into the AI big model again, the fault data set is expanded to obtain an expanded data set. The present invention can provide sufficient data diversity and coverage for the AI big model by extracting impact data, thereby improving the effectiveness of data expansion, thereby improving the accuracy of rotating machinery fault diagnosis and enhancing the performance of the fault diagnosis model.
[0062] In an optional implementation, the fault data set includes a rolling bearing fault data set and a rotor rubbing fault data set, and the impact data set includes a rolling bearing impact data set and a rotor rubbing impact data set.
[0063] Specifically, the health status of rolling bearings plays a very important role in the safe service performance of mechanical equipment. Among the failure cases of rotating machinery, 45%-55% are caused by the failure of rolling bearings. As the core of rotating machinery, the rotor system also plays a vital role in the normal operation of mechanical equipment. Rotor rubbing failure will directly affect the normal operation of the machinery, resulting in a decrease in production quality and production efficiency. Therefore, when it comes to fault diagnosis of rotating machinery, it is particularly important to use diversified data including rolling bearing fault data sets and rotor rubbing fault data sets. The rolling bearing fault data sets and rotor rubbing fault data sets are processed to obtain rolling bearing impact data sets and rotor rubbing impact data sets.
[0064] The rolling bearing fault data set can be obtained from a public data set. For example, the public data set can select the XJTU-SY rolling bearing accelerated life test data set, which contains the operating data of 15 rolling bearings, the sampling frequency is 25.6kHz, the sampling interval is 1min, and the sampling time is 1.28s each time. For each sampling, the acquired data is saved as a CSV file, in which the first column is the horizontal vibration acceleration signal and the second column is the vertical vibration acceleration signal. In the embodiment of the present application, the horizontal vibration acceleration signal is used to select the data when the rolling bearing fails from the public data set. By calculating the root mean square value of the csv data, it can be determined whether the data is fault data. When the data is greater than the preset fault threshold, it can be determined that the data is fault data. The fault data of all rolling bearings are selected from the public data set as the rolling bearing fault data set of the embodiment of the present application.
[0065] The rotor rubbing fault data set can select multiple rotor rubbing abnormal data samples on a specified date. Each data sample has 1024 data points, namely wave time domain data, and also includes timestamp, rotation speed, sampling frequency and other information. This information can be collected by different sensors, such as vibration sensors and displacement sensors.
[0066] It should be noted that, according to the mechanical system and failure mode in actual operation, the fault data set may also include other types of fault data sets, such as gear fault data sets, motor fault data sets or other rotating mechanical component fault data sets, which are not specifically limited here.
[0067] In an optional implementation manner, the processing the fault data set according to the absolute mean to obtain the impact data set includes:
[0068] Traverse all data samples in the fault data set;
[0069] When the data sample is greater than or equal to a preset threshold and the absolute value of the data sample is greater than or equal to the absolute mean, the absolute value of the difference between the data sample and the absolute mean is calculated and recorded as the impact data corresponding to the data sample;
[0070] When the data sample is smaller than the preset threshold and the absolute value of the data sample is smaller than the absolute mean, the sum of the data sample and the absolute mean is calculated and recorded as the impact data corresponding to the data sample;
[0071] All the impact data are collected to obtain an impact data set.
[0072] Specifically, in order to make the data generated by the AI large model closer to the real data and allow the data range to be freely adjusted during generation, impact data needs to be introduced. First, the absolute mean of all data samples in the fault data set is calculated. The calculation formula is as follows:
[0073] ;
[0074] in, represents the absolute mean, N represents the total number of data samples in the fault data set, Represents a data sample.
[0075] The data in the fault data set that exceeds the absolute mean is taken as impact data, and the fault data set is processed according to the absolute mean. The processing rules are as follows:
[0076] When the data sample is greater than or equal to the preset threshold and the absolute value of the data sample is greater than or equal to the absolute mean, the absolute mean is subtracted from the data sample, and the difference is recorded as the impact data corresponding to the data sample;
[0077] When a data sample is smaller than a preset threshold and the absolute value of the data sample is smaller than the absolute mean, the data sample is added with the absolute mean, and the sum is recorded as the impact data corresponding to the data sample;
[0078] In a specific implementation, the preset threshold may be set to 0, or may be set to other values according to specific requirements, which is not specifically limited here.
[0079] According to the above processing rules, a rolling bearing impact data set and a rotor rubbing impact data set are obtained respectively. In order to facilitate subsequent operation processing, the rolling bearing fault data set and the rolling bearing impact data set can be merged and saved as a first data file, the first column of the first data file is the rolling bearing fault data set, and the second column is the rolling bearing impact data set. The rotor rubbing fault data set and the rotor rubbing impact data set are merged and saved as a second data file, the first column of the second data file is the rotor rubbing fault data set, and the second column is the rotor rubbing impact data set. It can also be directly input into the AI large model for expansion without saving, and no specific limitation is made here.
[0080] In an optional implementation, when the fault data set is a rotor rubbing fault data set, extracting frequency domain features according to the fault data set and constructing a prompt template according to the frequency domain features includes:
[0081] Inputting the rotor rubbing fault data set into a pre-trained AI big model;
[0082] Calculating a fast Fourier transform of the rotor rub fault data set;
[0083] Normalizing the frequency spectrum according to the fast Fourier transform and the total number of the data samples to obtain a normalized result;
[0084] Extracting rotor rubbing frequency domain features according to the normalized results; wherein the rotor rubbing frequency domain features include main frequency components, amplitude and amplitude difference;
[0085] A rotor rubbing prompt template is constructed according to the rotor rubbing frequency domain characteristics.
[0086] Specifically, when the fault data set is a rotor rubbing fault data set, the rotor rubbing fault data set is input into a pre-trained AI big model. The AI big model extracts frequency domain features by using GPT-4o. First, the fast Fourier transform of the rotor rubbing fault data set is calculated, and the spectrum is normalized by dividing by the total number of data points. The main frequency components within a preset range are determined, the amplitudes are extracted, and the amplitude difference between each amplitude and the highest amplitude is calculated.
[0087] It should be noted that when the rotor rubbing fault data set is input into the AI big model, a rotor dialogue template can be added. For example, the rotor dialogue template can be set as:
[0088] The data in the above table are rotor rubbing fault data. The speed of the equipment is 9517RPM and the sampling frequency is 5100Hz. The spectrum of the rotor rubbing fault data is plotted (the total number of data points should be divided in the plotted spectrum), and the first few main frequency components, amplitudes, and the difference between each amplitude and the highest amplitude are obtained (the main frequency components should remove the ones within 10hz difference, and only the main frequency components of 80-1000Hz are needed).
[0089] It should be noted that the values in the above rotor dialogue templates are not fixed values and can be adjusted according to different tasks in the specific implementation. Most of the training corpus of GPT-4o is composed of English, so the above rotor dialogue templates are translated into English.
[0090] The rotor rubbing fault dataset and the rotor dialogue template translated into English are input into GPT-4o. The answer of GPT-4o is as follows:
[0091] The data file contains: "Serial Number", "Rotor Rub Fault Data". For this analysis, we will focus on the "Rotor Rub Fault Data" column. The main steps are as follows:
[0092] 1. Calculate the Fast Fourier Transform (FFT) of the “rotor rubbing fault data”;
[0093] 2. Normalize the spectrum by dividing by the total number of data points;
[0094] 3. Determine the main frequency components in the range of 80-1000 Hz;
[0095] 4. Eliminate frequency components within 10 Hz of each other;
[0096] 5. Extract the amplitudes and calculate the difference between each amplitude and the highest amplitude.
[0097] Table 1 below is the main frequency components in the range of 80-1000 Hz, excluding components within 10 Hz of each other:
[0098] Table 1 Rotor friction frequency domain characteristics
[0099]
[0100] Dominant Frequencies: The most dominant frequency is 159.375 Hz with an amplitude of approximately 5.971. Significant Differences: All other significant frequencies have much lower amplitudes, followed by 0.499 Hz with an amplitude of 318.75, which differs from the maximum amplitude by 5.472. The rotor spectrum plot above shows the frequency amplitudes in the 80-1000 Hz range, highlighting the dominant frequency components.
[0101] For example, see Figure 2 , which is the rotor spectrum diagram in the embodiment of the present application.
[0102] In order to enable the AI big model to better understand and perform the task of expanding the data capacity of rotating machinery faults, it is necessary to construct a new rotor rubbing prompt template based on the above frequency domain features and data. The prompt template is usually called prompt engineering, also known as context prompt, which is a method to guide the behavior of the large language model LLM towards a specific result without updating the weights / parameters of the model. For example, the rotor rubbing prompt template can be constructed as follows:
[0103] Tip: Data Generation Tasks
[0104] Task: First, learn the time domain and frequency domain features of the rotor rubbing fault data in Table 1 above, including:
[0105] 1. The overall trend of rotor rubbing fault data;
[0106] 2. Maximum, minimum and standard deviation of rotor rubbing fault data;
[0107] 3. Several main frequency components, amplitudes and the difference between the amplitude of each main frequency component and the highest amplitude extracted from the rotor dialogue template.
[0108] Based on the time domain and frequency domain features learned above, five time domain data sets are generated. The requirements of the data sets are as follows:
[0109] 1. There are 2048 data points in each data set;
[0110] 2. Adding impact data to time domain data. Adding impact data to time domain data can perform data point replacement operations;
[0111] 3. The overall trend of the generated data in the time domain should be somewhat similar to the rotor rubbing fault data;
[0112] 4. The envelope spectrum of the generated data set shall include the specified frequency components, amplitudes and amplitude differences of the specified frequency components in Table 1 above;
[0113] Finally, draw the time domain graph and spectrum of the generated data set. The generated data is saved in a csv file. Each column represents a data set. Do not use scientific notation.
[0114] After translating the above rotor prompt template into English and inputting it into GPT-4o, the AI model may not fully understand the requirements of the generated data set. In this case, the English rotor prompt template can be repeatedly input until a generated data set that meets the requirements is output.
[0115] In an optional implementation, when the fault data set is a rolling bearing fault data set, extracting frequency domain features according to the fault data set and constructing a prompt template according to the frequency domain features includes:
[0116] Inputting the rolling bearing fault data set into a pre-trained AI big model;
[0117] Performing envelope analysis on the rolling bearing fault data set to obtain an envelope spectrum;
[0118] Extracting the rolling bearing frequency domain characteristics from the envelope spectrum; wherein the rolling bearing frequency domain characteristics include main frequency components, amplitude and amplitude difference;
[0119] A rolling bearing prompt template is constructed according to the rolling bearing frequency domain characteristics.
[0120] Specifically, when the fault data set is a rolling bearing fault data set, the rolling bearing fault data set is input into a pre-trained AI big model. The AI big model extracts frequency domain features by using GPT-4o. First, envelope analysis is performed on the rolling bearing fault data set to obtain an envelope spectrum, and then the rolling bearing frequency domain features are extracted from the envelope spectrum.
[0121] It should be noted that when the rolling bearing fault data set is input into the AI big model, a rolling bearing dialogue template can be added. For example, the rolling bearing dialogue template can be set as:
[0122] The data in the above table are rolling bearing failure data. The speed of the equipment is 2100RPM and the sampling frequency is 25600Hz. The envelope spectrum of the rolling bearing outer ring failure data is drawn (the total number of data points must be divided when drawing the envelope spectrum), and the first few main frequency components, amplitudes, and the difference between each amplitude and the highest amplitude are obtained (the main frequency components with a difference of 10hz must be removed, and only the main frequency components of 30-1000Hz are needed).
[0123] It should be noted that the values in the above rolling bearing dialogue template are not fixed values and can be adjusted according to different tasks in the specific implementation. Most of the training corpus of GPT-4o is composed of English, so the above rolling bearing dialogue template is translated into English.
[0124] The rolling bearing fault dataset and the rolling bearing dialogue template translated into English are input into GPT-4o. The answer of GPT-4o is as follows:
[0125] To complete this task, you need to first read the uploaded data file, then calculate the envelope spectrum of the rolling bearing outer ring fault data, and then extract the main frequency components and their amplitudes in the envelope spectrum, as well as the difference between each amplitude and the highest amplitude. The specific steps are as follows:
[0126] 1. Read the data file;
[0127] 2. Perform envelope analysis, which includes: Bandpass filtering: perform bandpass filtering on the signal to remove unwanted frequency components; Hilbert transform: perform Hilbert transform on the filtered signal to obtain the envelope signal; Fourier transform: perform Fourier transform on the envelope signal to obtain the envelope spectrum;
[0128] 3. Draw the envelope spectrum;
[0129] 4. Extract the main frequency components, including the amplitude and the gap between each amplitude and the highest amplitude;
[0130] 5. Filter the frequency components, remove the frequency components with a phase difference of 10Hz, and retain the main frequency components in the range of 30-1000Hz.
[0131] Next, the main frequency components and their amplitudes are extracted, and the difference between each amplitude and the highest amplitude is calculated. The main frequency components, amplitudes, and the amplitude difference between each amplitude and the highest amplitude after cleaning and processing are shown in Table 2 below:
[0132] Table 2 Frequency domain characteristics of rolling bearings
[0133]
[0134] These frequency components are all in the range of 30-1000 Hz, and the difference between each frequency component is greater than 10 Hz.
[0135] For example, see Figure 3 , is the rolling bearing envelope spectrum in the embodiment of the present application.
[0136] In order to enable the AI big model to better understand and perform the task of expanding the rotating machinery fault data, it is necessary to construct a new rolling bearing prompt template based on the above frequency domain features and data. For example, the rolling bearing prompt template can be constructed as follows:
[0137] Tip: Data Generation Tasks
[0138] Task: First, learn the time domain and frequency domain characteristics of the rolling bearing outer ring fault data in the above table, including:
[0139] 1. The overall trend of rolling bearing outer ring failure data;
[0140] 2. Maximum, minimum and standard deviation of rolling bearing outer ring fault data;
[0141] 3. Several main frequency components, amplitudes, and the difference between the amplitude of each main frequency component and the highest amplitude extracted from the rolling bearing dialogue template.
[0142] Based on the time domain and frequency domain features learned above, five time domain data sets are generated. The requirements of the data sets are as follows:
[0143] 1. There are 15,000 data points in each data set;
[0144] 2. Adding impact data to time domain data. Adding impact data to time domain data can perform data point replacement operations;
[0145] 3. The overall trend of the generated data in the time domain should be somewhat similar to the rolling bearing outer ring failure data;
[0146] 4. The envelope spectrum of the generated data set shall include the specified frequency components, amplitudes and amplitude differences of the specified frequency components in Table 2 above;
[0147] Finally, draw the time domain diagram and envelope spectrum of the generated data set. The generated data is saved in a csv file. Each column represents a data set. Do not use scientific notation.
[0148] After translating the above rolling bearing prompt template into English and inputting it into GPT-4o, the AI large model may not fully understand the requirements of the generated data set. In this case, the English rolling bearing prompt template can be repeatedly input until a generated data set that meets the requirements is output.
[0149] In specific implementations, other dialogue templates may also be required. For example, when the generated frequency domain does not have a peak value of a specified frequency component, enter the following first dialogue template: In the envelope spectrum of the generated data, there is no peak value with a specified frequency component. When the amplitude of the generated time domain or frequency domain is obviously several times different from the given one, enter the following second dialogue template: Regenerate the data set and double the amplitude of both the time domain and the frequency domain. When no curve appears when the frequency domain image of the generated data set is drawn, there are blank data points in the generated data set. Enter the following third dialogue template: Regenerate the blank data points in the data set and do not simply replace them with 0.
[0150] It should be noted that, in the specific implementation operation, various types of dialogue templates can be constructed according to task requirements, and no specific limitation is made here.
[0151] In an optional implementation, performing envelope analysis on the rolling bearing fault data set to obtain an envelope spectrum includes:
[0152] Performing bandpass filtering on the rolling bearing fault data set to obtain a filtered signal;
[0153] Performing Hilbert transform on the filtered signal to obtain an envelope signal;
[0154] Perform Fourier transform on the envelope signal to obtain an envelope spectrum.
[0155] Specifically, when performing envelope analysis, GPT-4o first performs bandpass filtering on the rolling bearing fault data set to obtain a filtered signal, performs Hilbert transform on the filtered signal to obtain an envelope signal, and then performs Fourier transform on the envelope signal to obtain an envelope spectrum.
[0156] In an optional embodiment, the method further includes:
[0157] The expanded data set is verified using a convolutional neural network and a support vector machine to obtain a verification result.
[0158] Specifically, in order to test the quality of the generated expanded data set, the frequency domain analysis and correlation coefficient calculation are first performed on the expanded data set, and then the statistical properties such as the mean, variance, skewness and kurtosis of the expanded data set are compared with the original fault data set. The correlation coefficient is the basis for analyzing the correlation between the two sets of data. If the correlation coefficient between the two signals is larger, it means that the correlation of the signals is stronger. If the correlation coefficient between the two signals is smaller, it means that the correlation of the signals is smaller. The calculation formula of the correlation coefficient is as follows:
[0159] ;
[0160] in, represents the correlation coefficient, Indicates signal and The covariance of Indicates signal The variance of Indicates signal The variance of .
[0161] For example, see Figure 4 , is the frequency spectrum of the five rotor rubbing expansion data sets. For the rotor rubbing expansion data sets, the correlation coefficients are shown in Table 3 below:
[0162] Table 3 Rotor friction correlation coefficient table
[0163]
[0164] The statistical indicators of the rotor friction expansion data set are shown in Table 4 below:
[0165] Table 4 Rotor friction statistical index table
[0166]
[0167] according to Figure 4 It can be seen from the frequency spectrum of the rotor rubbing expansion data set that the main frequency components of the frequency spectrum of the rotor rubbing fault data set are not much different. According to Tables 3 and 4, it can be concluded that the correlation coefficients between each expansion data set are above 0.95, indicating that these expansion data sets are highly similar, but not exactly the same. In terms of statistical characteristics, the expansion data set and the rotor rubbing fault data set are similar in terms of mean, variance, skewness, and kurtosis.
[0168] For example, see Figure 5 , is the envelope spectrum of 5 rolling bearing expansion data sets. For the rolling bearing expansion data set, the correlation coefficient is shown in Table 5 below:
[0169] Table 5 Rolling bearing correlation coefficient table
[0170]
[0171] The statistical indicators of the rolling bearing expansion data set are shown in Table 6 below:
[0172] Table 6 Statistical index table of rolling bearings
[0173]
[0174] according to Figure 5 It can be seen from the envelope spectrum of the rolling bearing expansion data set that the main frequency components of the envelope spectrum of the rolling bearing fault data set are not much different. According to Tables 5 and 6, it can be concluded that the correlation coefficients between each expansion data set are around 0.90, indicating that these expansion data sets are highly similar, but not exactly the same. In terms of statistical characteristics, the expansion data set and the rolling bearing fault data set are similar in terms of mean, variance, skewness, and kurtosis.
[0175] It can be concluded that the above-mentioned expanded dataset of rotating machinery faults exhibits realistic time domain characteristics, such as periodic patterns and impact-like peaks, which mimic the real-world phenomena observed in the original data. The authenticity of the data trends and noise simulation methods is crucial, including the extent to which the synthetic data simulates the natural variability and randomness of the original data.
[0176] In order to verify the effectiveness of the expanded dataset, two models, convolutional neural network (CNN) and support vector machine (SVM), are used to perform binary classification on the original rotating machinery fault dataset and the dataset after adding the expanded dataset. The ratio of the training set to the test set can be 7:3.
[0177] For example, the number of training times of the convolutional neural network can be selected to be 20 times, and its network structure is shown in Table 7 below:
[0178] Table 7 Convolutional neural network structure
[0179]
[0180] The original rotating machinery fault data set is classified into two categories to obtain the original classification results. The data after adding the expanded data set is classified into two categories to obtain the expanded classification results. According to the original classification results and the expanded classification results, the validity verification results of the expanded data set can be obtained.
[0181] In the specific implementation, the kernel function of SVM is selected as poly function, and a binary classification task is performed on it. The following Table 8 shows the accuracy of selecting different data points as training and test samples:
[0182] Table 8 Sample accuracy
[0183]
[0184] According to Table 8 above, the accuracy of the model has a significant effect after adding the expanded data set. Except for the 100% accuracy, the lowest accuracy increase is 2%, and the highest is 19%, which proves that the expanded data set is reliable and effective.
[0185] In the second aspect, the embodiment of the present invention provides a rotating machinery fault data expansion device based on AI large model and frequency domain characteristics, see Figure 6 , which is a structural schematic diagram of an embodiment of a rotating machinery fault data expansion device based on an AI large model and frequency domain features provided by the present invention.
[0186] like Figure 6 As shown, the device comprises:
[0187] A fault data acquisition module 21 is used to acquire a fault data set containing a specified number of data samples;
[0188] An impact data generating module 22 is used to calculate the absolute mean of all the data samples, and process the fault data set according to the absolute mean to obtain an impact data set;
[0189] A frequency domain feature extraction module 23, used to extract frequency domain features according to the fault data set, and construct a prompt template according to the frequency domain features;
[0190] The fault data expansion module 24 is used to input the fault data set, the impact data set and the prompt template into a pre-trained AI large model, expand the fault data set, and obtain an expanded data set.
[0191] In an optional implementation, the fault data set includes a rolling bearing fault data set and a rotor rubbing fault data set, and the impact data set includes a rolling bearing impact data set and a rotor rubbing impact data set.
[0192] In an optional implementation, the impact data generating module 22 is further configured to:
[0193] Traverse all data samples in the fault data set;
[0194] When the data sample is greater than or equal to a preset threshold and the absolute value of the data sample is greater than or equal to the absolute mean, the absolute value of the difference between the data sample and the absolute mean is calculated and recorded as the impact data corresponding to the data sample;
[0195] When the data sample is smaller than the preset threshold and the absolute value of the data sample is smaller than the absolute mean, the sum of the data sample and the absolute mean is calculated and recorded as the impact data corresponding to the data sample;
[0196] All the impact data are collected to obtain an impact data set.
[0197] In an optional implementation, when the fault data set is a rotor rubbing fault data set, the frequency domain feature extraction module 23 is further used to:
[0198] Inputting the rotor rubbing fault data set into a pre-trained AI big model;
[0199] Calculating a fast Fourier transform of the rotor rub fault data set;
[0200] Normalizing the frequency spectrum according to the fast Fourier transform and the total number of the data samples to obtain a normalized result;
[0201] Extracting rotor rubbing frequency domain features according to the normalized results; wherein the rotor rubbing frequency domain features include main frequency components, amplitude and amplitude difference;
[0202] A rotor rubbing prompt template is constructed according to the rotor rubbing frequency domain characteristics.
[0203] In an optional implementation, when the fault data set is a rolling bearing fault data set, the frequency domain feature extraction module 23 is further used to:
[0204] Inputting the rolling bearing fault data set into a pre-trained AI big model;
[0205] Performing envelope analysis on the rolling bearing fault data set to obtain an envelope spectrum;
[0206] Extracting the rolling bearing frequency domain characteristics from the envelope spectrum; wherein the rolling bearing frequency domain characteristics include main frequency components, amplitude and amplitude difference;
[0207] A rolling bearing prompt template is constructed according to the rolling bearing frequency domain characteristics.
[0208] In an optional implementation, the frequency domain feature extraction module 23 is further used for:
[0209] Performing bandpass filtering on the rolling bearing fault data set to obtain a filtered signal;
[0210] Performing Hilbert transform on the filtered signal to obtain an envelope signal;
[0211] Perform Fourier transform on the envelope signal to obtain an envelope spectrum.
[0212] In an optional embodiment, the device further includes a verification module;
[0213] The verification module is used to:
[0214] The expanded data set is verified using a convolutional neural network and a support vector machine to obtain a verification result.
[0215] In a third aspect, an embodiment of the present invention provides an electronic device, see Figure 7 , which is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention.
[0216] like Figure 7 As shown, the device includes:
[0217] A memory 31, used for storing computer programs;
[0218] A processor 32, configured to execute the computer program;
[0219] Wherein, when the processor 32 executes the computer program, it implements the rotating machinery fault data expansion method based on the AI large model and frequency domain characteristics as described in any of the above embodiments.
[0220] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory 31 and executed by the processor 32 to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the electronic device.
[0221] The processor 32 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.
[0222] The memory 31 can be used to store the computer program and / or module, and the processor 32 realizes various functions of the electronic device by running or executing the computer program and / or module stored in the memory 31, and calling the data stored in the memory 31. The memory 31 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory 31 can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.
[0223] It should be noted that the above electronic device includes, but is not limited to, a processor and a memory. Those skilled in the art can understand that Figure 7 The structural diagram is merely an example of the electronic device described above and does not constitute a limitation on the electronic device, and may include more components than shown in the figure, or a combination of certain components, or different components.
[0224] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed, the method for expanding the capacity of rotating machinery fault data based on the AI large model and frequency domain features as described in any of the above embodiments is implemented.
[0225] It should be understood that the present invention implements all or part of the processes in the above-mentioned method for expanding the capacity of rotating machinery fault data based on the AI large model and frequency domain features, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned method for expanding the capacity of rotating machinery fault data based on the AI large model and frequency domain features can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electric carrier signals and telecommunication signals.
[0226] The above description is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited thereto. It should be pointed out that for those skilled in the art, several equivalent obvious variations and / or equivalent substitutions can be made without departing from the technical principles of the present invention. These obvious variations and / or equivalent substitutions should also be regarded as the protection scope of the present invention.
Claims
1. A method for expanding the capacity of rotating machinery fault data based on AI large model and frequency domain features, characterized in that: include: Get a fault dataset containing a specified number of data samples; Calculating the absolute mean of all the data samples, and processing the fault data set according to the absolute mean to obtain an impact data set; Extracting frequency domain features according to the fault data set, and constructing a prompt template according to the frequency domain features; Inputting the fault data set, the impact data set and the prompt template into a pre-trained AI large model, expanding the fault data set to obtain an expanded data set; When the fault data set is a rotor rubbing fault data set, extracting frequency domain features according to the fault data set and constructing a prompt template according to the frequency domain features include: Inputting the rotor rubbing fault data set into a pre-trained AI big model; Calculating a fast Fourier transform of the rotor rub fault data set; Normalizing the frequency spectrum according to the fast Fourier transform and the total number of the data samples to obtain a normalized result; Extracting rotor rubbing frequency domain features according to the normalized results; wherein the rotor rubbing frequency domain features include main frequency components, amplitude and amplitude difference; A rotor rubbing prompt template is constructed according to the rotor rubbing frequency domain characteristics.
2. The method for expanding the capacity of rotating machinery fault data based on the AI large model and frequency domain features according to claim 1, characterized in that: The fault data set includes a rolling bearing fault data set and a rotor rubbing fault data set, and the impact data set includes a rolling bearing impact data set and a rotor rubbing impact data set.
3. The method for expanding the capacity of rotating machinery fault data based on the AI large model and frequency domain features according to claim 1, characterized in that: The step of processing the fault data set according to the absolute mean to obtain an impact data set includes: Traversing all data samples in the fault data set; When the data sample is greater than or equal to a preset threshold and the absolute value of the data sample is greater than or equal to the absolute mean, the absolute value of the difference between the data sample and the absolute mean is calculated and recorded as the impact data corresponding to the data sample; When the data sample is smaller than the preset threshold and the absolute value of the data sample is smaller than the absolute mean, the sum of the data sample and the absolute mean is calculated and recorded as the impact data corresponding to the data sample; All the impact data are collected to obtain an impact data set.
4. The method for expanding the capacity of rotating machinery fault data based on the AI large model and frequency domain features according to claim 2, characterized in that: When the fault data set is a rolling bearing fault data set, extracting frequency domain features according to the fault data set and constructing a prompt template according to the frequency domain features includes: Inputting the rolling bearing fault data set into a pre-trained AI big model; Performing envelope analysis on the rolling bearing fault data set to obtain an envelope spectrum; Extracting the rolling bearing frequency domain characteristics from the envelope spectrum; wherein the rolling bearing frequency domain characteristics include main frequency components, amplitude and amplitude difference; A rolling bearing prompt template is constructed according to the rolling bearing frequency domain characteristics.
5. The method for expanding the capacity of rotating machinery fault data based on the AI large model and frequency domain features according to claim 4, characterized in that: The step of performing envelope analysis on the rolling bearing fault data set to obtain an envelope spectrum includes: Performing bandpass filtering on the rolling bearing fault data set to obtain a filtered signal; Performing Hilbert transform on the filtered signal to obtain an envelope signal; Perform Fourier transform on the envelope signal to obtain an envelope spectrum.
6. The method for expanding the capacity of rotating machinery fault data based on the AI large model and frequency domain features according to claim 1, characterized in that: The method further comprises: The expanded data set is verified using a convolutional neural network and a support vector machine to obtain a verification result.
7. A rotating machinery fault data expansion device based on AI large model and frequency domain characteristics, characterized in that: include: A fault data acquisition module is used to acquire a fault data set containing a specified number of data samples; An impact data generation module, used for calculating the absolute mean of all the data samples, and processing the fault data set according to the absolute mean to obtain an impact data set; A frequency domain feature extraction module, used to extract frequency domain features according to the fault data set, and construct a prompt template according to the frequency domain features; A fault data expansion module, used for inputting the fault data set, the impact data set and the prompt template into a pre-trained AI large model, expanding the fault data set to obtain an expanded data set; When the fault data set is a rotor rubbing fault data set, the frequency domain feature extraction module is further used to: Inputting the rotor rubbing fault data set into a pre-trained AI big model; Calculating a fast Fourier transform of the rotor rub fault data set; Normalizing the frequency spectrum according to the fast Fourier transform and the total number of the data samples to obtain a normalized result; Extracting rotor rubbing frequency domain features according to the normalized results; wherein the rotor rubbing frequency domain features include main frequency components, amplitude and amplitude difference; A rotor rubbing prompt template is constructed according to the rotor rubbing frequency domain characteristics.
8. An electronic device, characterized in that: include: Memory for storing computer programs; A processor, configured to execute the computer program; Wherein, when the processor executes the computer program, the method for expanding the fault data of rotating machinery based on the AI large model and frequency domain features as described in any one of claims 1 to 6 is implemented.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed, it implements the method for expanding the fault data of rotating machinery based on the AI large model and frequency domain features as described in any one of claims 1 to 6.
