Bearing fault diagnosis system, method and program product based on large language model

Through a bearing fault diagnosis system based on a large language model, a large language model is used to process bearing fault data, extract time domain and frequency domain characteristics and combine label information for diagnosis, the problem of insufficient flexibility and adaptability speed in the existing technology is solved, which is significantly Improves diagnostic efficiency and accuracy and enhances the stability of rotating machinery.

CN118817313BActive Publication Date: 2024-12-06HANGZHOU INTERNATIONAL INNOVATION INSTITUTE OF BEIHANG UNIVERSITY +2
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411287669.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-14
Publication Date
2024-12-06
Estimated Expiration
2044-09-14

AI Technical Summary

Technical Problem

The lack of methods and systems for bearing fault diagnosis using large language models in the prior art, resulting in limitations in flexibility and adaptability, making it difficult to effectively process complex bearing fault data.

Method used

The bearing fault diagnosis system based on the large language model is adopted to obtain the vibration electrical signal through the vibration sensor, and the feature extraction module is used to extract the time domain and frequency domain feature vectors, and embed it in combination with the feature name tag and bearing tag, and input it to the large part of the frozen layer. Language model for troubleshooting.

Benefits of technology

It significantly improves the efficiency and accuracy of the bearing fault diagnosis system, improves the learning and memory ability of different fault modes, quickly adapts to new working conditions and fault types, and enhances the overall operating stability of the rotating machinery.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118817313B_ABST
    Figure CN118817313B_ABST
Patent Text Reader

Abstract

A bearing fault diagnosis system, method and program product based on a large language model, belonging to the field of artificial intelligence technology. In the system. A vibration sensor is used to obtain a vibration electrical signal of a bearing; a feature extraction module extracts a time domain feature vector and a frequency domain feature vector according to a sampling value of the vibration electrical signal; a splicing module splices the time domain feature vector and the frequency domain feature vector to form a splicing vector; a first embedding module embeds a feature name label vector in front of or behind each feature in the splicing vector to form a fine-tuning vector; a second embedding module embeds a bearing label vector in front of or behind the fine-tuning vector to form an input vector and then provides it to a fault diagnosis model; the fault diagnosis model performs fault diagnosis on the bearing according to the input vector, and the fault diagnosis model is trained by a large language model with some layers frozen. The present invention significantly improves the efficiency and accuracy of the bearing fault diagnosis system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a bearing fault diagnosis system, method and program product based on a large language model, and belongs to the technical field of artificial intelligence. Background Art

[0002] The Chinese invention patent application with publication number CN115525866A discloses a deep learning rolling bearing fault diagnosis method and system, which includes obtaining rolling bearing fault data; performing topological data analysis on each section of rolling bearing fault data to extract a topological feature matrix; performing fast Fourier transform on each section of rolling bearing fault data and calculating its modulus to extract a frequency domain feature matrix; using a convolutional neural network to fuse the topological feature matrix and the frequency domain feature matrix to obtain a fault feature matrix; based on the fault feature matrix, using a fault classification network to obtain a fault classification result. The invention can extract complementary fault features and improve the model diagnosis accuracy, and the model has stronger robustness under a strong noise background.

[0003] However, there is no method or system for fault diagnosis using a large language model in the prior art. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the present invention provides a bearing fault diagnosis system, method and program product based on a large language model, which utilizes the advantages of large language models in processing complex data patterns to improve the learning and memory capabilities of different fault modes; by fine-tuning a small number of parameters, it can quickly adapt to new operating conditions and fault types, thereby overcoming the limitations of traditional models in flexibility and adaptation speed; and can significantly improve the efficiency and accuracy of the bearing fault diagnosis system.

[0005] To achieve the above-mentioned purpose, the present invention also provides a bearing fault diagnosis system based on a large language model, which includes a vibration sensor, which is used to obtain the vibration electrical signal of the bearing. The system also includes a feature extraction module, a splicing module, a first embedding module, a second embedding module and a fault diagnosis model, wherein the feature extraction module extracts a time domain feature vector and a frequency domain feature vector according to the sampling value of the vibration electrical signal; the splicing module splices the time domain feature vector and the frequency domain feature vector to form a splicing vector; the first embedding module embeds a feature name label vector in front of or behind each feature in the splicing vector to form a fine-tuning vector; the second embedding module embeds a bearing label vector in front of or behind the fine-tuning vector to form an input vector and then provides it to the fault diagnosis model; the fault diagnosis model performs fault diagnosis on the bearing according to the input vector, and the fault diagnosis model is trained by a large language model with some layers frozen.

[0006] To achieve the above-mentioned object of the invention, the present invention also provides a bearing fault diagnosis method based on a large language model, which comprises the following steps:

[0007] Step 1: Obtain the vibration electrical signal of the bearing through a vibration sensor;

[0008] Step 2: Extracting time domain feature vectors and frequency domain feature vectors according to the sampling values ​​of the vibration electrical signal through the feature extraction module;

[0009] Step 3: Splice the time domain feature vector and the frequency domain feature vector to form a spliced ​​vector through a splicing module;

[0010] Step 4: embed the feature name label vector in front of or behind each feature in the concatenated vector through the first embedding module to form a fine-tuning vector; embed the bearing label vector in front of or behind the fine-tuning vector through the second embedding module to form an input vector and then provide it to the fault diagnosis model;

[0011] Step 5: The fault diagnosis model performs fault diagnosis on the bearing according to the input vector, and the fault diagnosis model is trained by a large language model with some layers frozen.

[0012] To achieve the above object, the present invention further provides a computer program product, which includes computer program code, and the computer program code can be called by a processor to execute the above method.

[0013] Compared with the prior art, the present invention has the following beneficial effects:

[0014] 1. Adopting a bearing fault diagnosis framework based on a large language model (LLM), integrating the long-term memory characteristics and fine-tuning technology of LLM, it can take advantage of the advantages of large language models in processing complex data patterns and improve the learning and memory capabilities of different fault modes;

[0015] 2. By fine-tuning a small number of parameters, it can quickly adapt to new operating conditions and fault types, thus overcoming the limitations of traditional models in terms of flexibility and adaptation speed;

[0016] 3. It can significantly improve the efficiency and accuracy of the bearing fault diagnosis system and further enhance the overall operating stability of rotating machinery. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a block diagram of the composition of a bearing fault diagnosis system based on a large language model provided in the first embodiment of the present invention.

[0018] Figure 2 It is a block diagram of the composition of the time domain feature extraction module provided in the first embodiment of the present invention.

[0019] Figure 3 It is a block diagram of the frequency domain feature extraction module provided in the first embodiment of the present invention.

[0020] Figure 4 It is a large language model with some layers frozen provided by the first embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0022] First embodiment

[0023] Figure 1 is a block diagram of a bearing fault diagnosis system based on a large language model provided by a first embodiment of the present invention, such as Figure 1 As shown, the first embodiment of the present invention provides a bearing fault diagnosis system based on a large language model, including a vibration sensor, which is used to obtain the vibration electrical signal of the bearing. The bearing fault diagnosis system also includes a feature extraction module, a splicing module, a first embedding module, a second embedding module and a fault diagnosis model, wherein the feature extraction module extracts a time domain feature vector and a frequency domain feature vector according to the sampling value of the vibration electrical signal; the splicing module splices the time domain feature vector and the frequency domain feature vector to form a splicing vector; the first embedding module embeds a feature name label vector in front of or behind each feature in the splicing vector to form a fine-tuning vector; the second embedding module embeds a bearing label vector in front of or behind the fine-tuning vector to form an input vector and then provides it to the fault diagnosis model; the fault diagnosis model performs fault diagnosis on the bearing according to the input vector, and the fault diagnosis model is trained by a large language model with some layers frozen.

[0024] In the first embodiment, each feature is distinguished by embedding a feature name label vector before or after each feature in the concatenated vector; and the data of a bearing is distinguished by embedding a bearing label vector before or after the fine-tuning vector.

[0025] In the first embodiment, various small sample fault data of various bearings are first used and fine-tuned before being input into a large language model (LL model), and the large language model is trained to obtain a fault diagnosis model. After training, the small sample fault data are input into the input embedding layer of the fault diagnosis model, and the fault type is output by the linear output layer of the fault diagnosis model.

[0026] The feature extraction module provided by the present invention includes a time domain feature extraction module and a frequency domain feature extraction module.

[0027] Figure 2 is a block diagram of a time domain feature extraction module provided by the first embodiment of the present invention, such as Figure 2As shown, the time domain feature extraction module includes a time domain mean calculator, a time domain standard deviation calculator, an absolute mean calculator, a peak calculator, a waveform index calculator, a pulse index calculator, a peak index calculator, a skewness calculator, a kurtosis calculator, a kurtosis index calculator, a root mean square calculator, a deviation calculator and a time domain feature vector generator, wherein the mean calculator calculates the mean p according to the time domain sampling value of the bearing vibration electrical signal. 1 ; The time domain standard deviation calculator is based on the time domain sampling value and time domain mean value p of the bearing vibration electrical signal. 1 Calculate the time domain standard deviation p 2 ; The absolute mean value calculator calculates the absolute mean value p according to the time domain sampling value of the bearing vibration electrical signal 4 ; The peak value calculator calculates the peak value p according to the time domain sampling value of the bearing vibration electrical signal 5 ; The waveform index calculator is based on the time domain standard deviation p 2 and the absolute mean p 4 Calculate the waveform index p 11 ; Pulse index calculator based on the absolute mean p 4 and peak value p 5 Calculate the pulse index p 12 ; The peak index calculator is based on the time domain standard deviation p 2 and peak value p 5 Calculate the peak index p 10 The skewness calculator calculates the skewness p based on the time domain sampling value of the bearing vibration electrical signal. 6 ; The kurtosis calculator calculates the kurtosis p according to the time domain sampling value of the bearing vibration electrical signal 7 ; Kurtosis index calculator based on skewness p 6 and kurtosis p 7 Calculate the kurtosis index p 9 ; The RMS calculator calculates the RMS p based on the time domain sampling value of the bearing vibration electrical signal 3 ; The deviation calculator calculates the deviation p based on the time domain sampling value of the bearing vibration electrical signal 8 The time domain feature vector generator generates a time domain feature vector P according to the output values ​​of the time domain mean calculator, the time domain standard deviation calculator, the absolute mean calculator, the peak calculator, the waveform index calculator, the pulse index calculator, the peak index calculator, the skewness calculator, the kurtosis calculator, the kurtosis index calculator, the root mean square calculator and the deviation calculator. T =[p 1 , p 2 , p 3 , p 4 , p 5 , p 6 , p 7 , p 8 , p 9 , p10 , p 11 , p 12 In the first embodiment, the time domain sampling value x(n) of the bearing vibration electrical signal x(t) is sampled by a sampler at a sampling frequency f k Sampling is performed.

[0028] In the first embodiment, the mean value calculator calculates the mean value p according to the time domain sampling value x(n) of the bearing vibration electrical signal x(t). 1 , the calculation formula is:

[0029] ,

[0030] Where n is the time domain index point and N is the sampling period.

[0031] In the first embodiment, the time domain standard deviation calculator calculates the time domain sample value x(n) and the time domain mean value p of the bearing vibration electrical signal x(t) according to the mean value. 1 Calculate the time domain standard deviation p 2, The calculation formula is:

[0032] .

[0033] In the first embodiment, the absolute mean value calculator calculates the absolute mean value p according to the time domain sampling value x(n) of the bearing vibration electrical signal x(t). 4 , the calculation formula is:

[0034] .

[0035] In the first embodiment, the peak value calculator calculates the peak value p according to the time domain sampling value x(n) of the bearing vibration electrical signal x(t). 5 , the calculation formula is:

[0036] .

[0037] In the first embodiment, the waveform index calculator is based on the time domain standard deviation p 2 and calculate the absolute mean p 4 Calculate the waveform index p 11 , the calculation formula is:

[0038] .

[0039] In the first embodiment, the pulse index calculator is based on the absolute mean value p 4 and peak value p 5 Calculate the pulse index p 12 , the calculation formula is:

[0040] .

[0041] In the first embodiment, the peak index calculator is based on the time domain standard deviation p 2 and peak value p 5 Calculate the peak index p 10, The calculation formula is:

[0042]

[0043] In the first embodiment, the skewness calculator calculates the skewness p according to the time domain sampling value x(n) of the bearing vibration electrical signal x(t). 6

[0044] .

[0045] In the first embodiment, the kurtosis calculator calculates the kurtosis p according to the time domain sampling value x(n) of the bearing vibration electrical signal x(t). 7 , the calculation formula is:

[0046] .

[0047] In the first embodiment, the kurtosis index calculator is based on the skewness p 6 and kurtosis p 7 Calculate the kurtosis index p 9 , the calculation formula is:

[0048] .

[0049] In the first embodiment, the RMS calculator calculates the RMS p according to the time domain sampling value x(n) of the bearing vibration electrical signal x(t) by the following formula: 3 , which is calculated by the following formula:

[0050] .

[0051] In the first embodiment, the deviation calculator calculates the deviation p according to the time domain sampling value x(n) of the bearing vibration electrical signal x(t). 8 , which is calculated by the following formula:

[0052] .

[0053] Figure 3 is a block diagram of a frequency domain feature extraction module provided by the first embodiment of the present invention, such as Figure 3The frequency domain feature extraction module shown includes a frequency domain feature extraction module, which includes a frequency mean calculator, a frequency deviation calculator, a frequency skewness calculator, a frequency kurtosis calculator, a center of gravity frequency calculator, a frequency standard deviation calculator, a variable parameter calculator, an eighth-order moment calculator, a sixteenth-order moment calculator, a regularity calculator, a frequency root mean square calculator, an average frequency calculator and a frequency domain feature vector generator, wherein the frequency mean calculator calculates the frequency mean according to the frequency domain signal of the bearing vibration electrical signal sampling value The frequency deviation calculator is based on the frequency domain signal and frequency mean value p of the bearing vibration electrical signal sampling value. 13 Calculate the frequency deviation p 14 The frequency skewness calculator is based on the frequency domain signal and frequency mean value p of the bearing vibration electrical signal sampling value. 13 and frequency deviation p 14 Calculate the frequency skewness p 15 The frequency kurtosis calculator is based on the frequency domain signal and frequency mean p of the bearing vibration electrical signal sampling value. 13 and frequency deviation p 14 Calculate the frequency kurtosis p 16 The center of gravity frequency calculator calculates the center of gravity frequency p according to the frequency domain signal and sampling frequency of the bearing vibration electrical signal sampling value. 17 The frequency standard deviation calculator is based on the frequency domain signal, sampling frequency and center of gravity frequency p of the bearing vibration electrical signal sampling value. 17 Calculate the frequency standard deviation p 18 ; The variable parameter calculator is based on the center of gravity frequency p 17 Frequency standard deviation p 18 Calculate the variable parameter p 22 ; The eighth-order moment calculator is based on the center of gravity frequency p 17 , sampling frequency, frequency standard deviation p 18 The eighth-order moment p is calculated by the frequency domain signal of the sampling value of the bearing vibration electrical signal 23 ; The sixteenth-order moment calculator is based on the center of gravity frequency p 17 , sampling frequency, frequency standard deviation p 18 The sixteenth order moment p is calculated by the frequency domain signal of the bearing vibration electrical signal sampling value 24 ; The regularity calculator calculates the regularity p according to the frequency domain signal and sampling frequency of the bearing vibration electrical signal sampling value. 21 The frequency RMS calculator calculates the frequency RMS p according to the frequency domain signal and sampling frequency of the bearing vibration electrical signal sampling value. 19 The average frequency calculator calculates the average frequency p according to the frequency domain signal and sampling frequency of the bearing vibration electrical signal sampling value. 20The frequency domain feature vector generator generates a frequency domain feature vector according to the output values ​​of the frequency mean calculator, frequency deviation calculator, frequency skewness calculator, frequency kurtosis calculator, center of gravity frequency calculator, frequency standard deviation calculator, variable parameter calculator, eighth-order moment calculator, sixteenth-order moment calculator, regularity calculator, frequency root mean square calculator and average frequency calculator: P F =[p 13 , p 14 ,p 15 , p 16 , p 17 , p 18 , p 19 , p 20 , p 21 , p 22 , p 23 , p 24 In the first embodiment, the frequency domain signal s(k) of the sampled value x(n) of the bearing vibration electrical signal x(t) is obtained by performing a sub-transformation on the sampled value x(n).

[0054] In the first embodiment, the frequency mean value calculator calculates the frequency mean value p according to the frequency domain signal s(k) of the sample value x(n) of the bearing vibration electrical signal x(t). 13 , the calculation formula is:

[0055] ,

[0056] In the formula, The sampling signal frequency domain signal; k is the frequency domain index; K is the frequency domain period.

[0057] In the first embodiment, the frequency deviation calculator calculates the frequency domain signal s(k) and the frequency mean value P of the sampling value x(n) of the bearing vibration electrical signal x(t). 13 Calculate the frequency deviation P 14 , the calculation formula is:

[0058] .

[0059] In the first embodiment, the frequency skewness calculator calculates the frequency domain signal s(k) and the frequency mean value p of the sampling value x(n) of the bearing vibration electrical signal x(t). 13 and frequency deviation p 14 Calculate the frequency skewness p 15 , the calculation formula is:

[0060] .

[0061] In the first embodiment, the frequency kurtosis calculator calculates the frequency domain signal s(k) and the frequency mean of the sampling value x(n) of the bearing vibration electrical signal x(t). and frequency deviation Calculate frequency kurtosis , specifically calculated by the following formula:

[0062] .

[0063] In the first embodiment, the center of gravity frequency calculator is based on the frequency domain signal s(k) and the sampling frequency f of the bearing vibration electrical signal x(t) sampling value x(n). k Calculate the center of gravity frequency p 17 , which is calculated by the following formula:

[0064] ,

[0065] Where, f k is the sampling frequency.

[0066] In the first embodiment, the frequency standard deviation calculator calculates the frequency domain signal s(k) of the sampled value x(n) of the bearing vibration electrical signal x(t) and the sampling frequency f by the following formula: k , center of gravity frequency p 17 Calculate the frequency standard deviation p 18 , which is calculated by the following formula:

[0067] .

[0068] In the first embodiment, the variable parameter calculator is based on the center of gravity frequency p 17 Frequency standard deviation p 18 Calculate the variable parameter p 22 , which is calculated by the following formula:

[0069] .

[0070] The eighth-order moment calculator is based on the center of gravity frequency p 17 , sampling frequency f k , frequency standard deviation p 18 The eighth-order moment p is calculated by the frequency domain signal s(k) of the sampling value x(n) of the bearing vibration electrical signal x(t) 23 , which is calculated by the following formula:

[0071] .

[0072] In the first embodiment, the sixteenth-order moment calculator is based on the centroid frequency p 17 , sampling frequency f k , sampling frequency frequency standard deviation p 18The sixteenth-order moment p is calculated by the frequency domain signal s(k) of the sampling value x(n) of the bearing vibration electrical signal x(t) 24 , which is calculated by the following formula:

[0073] .

[0074] In the first embodiment, the regularity calculator is based on the sampling frequency f k , the calculation regularity p of the frequency domain signal s(k) of the sampling value x(n) of the bearing vibration electrical signal x(t) 21 , which is calculated by the following formula:

[0075] .

[0076] In the first embodiment, the frequency RMS calculator is based on the sampling frequency f k The frequency domain signal s(k) of the bearing vibration electrical signal x(t) sampling value x(n) calculates the frequency root mean square p 19 , which is calculated by the following formula:

[0077] .

[0078] In the first embodiment, the average frequency calculator is based on the sampling frequency f k The average frequency p is calculated by the frequency domain signal s(k) of the bearing vibration electrical signal x(t) sampling value x(n) 20 , which is calculated by the following formula:

[0079] .

[0080] In the first embodiment, the feature name label vector can use a five-bit binary code instead of the feature name, such as Table 1:

[0081] Table 1

[0082]

[0083] In the first embodiment, since 24 features of the bearing vibration data need to be extracted, the 24 features can be distinguished by using five-bit binary labels. Compared with using natural language as labels, the calculation time of the large language model (LLM large model) can be saved.

[0084] In the first embodiment, the vibration characteristic data of different bearings can be distinguished by the bearing label vector.

[0085] The first embodiment integrates the long-term memory characteristics of the large language model and the fine-tuning technology, and can utilize the advantages of the large language model in processing complex data patterns to improve the learning and memory capabilities of different fault modes; by fine-tuning a small number of parameters, it can quickly adapt to new operating conditions and fault types, thereby overcoming the limitations of traditional models in flexibility and adaptation speed; it can significantly improve the efficiency and accuracy of the bearing fault diagnosis system, and further enhance the overall operating stability of the rotating machinery.

[0086] Figure 4 Schematic diagram of a structure based on a large language model with some layers frozen provided by the present invention. Figure 4 The large language model with some layers frozen as shown includes: the large language model with some layers frozen includes an input embedding layer, a position embedding layer, a multi-head attention mechanism layer, a first residual connection layer, a feedforward neural network and a second residual connection layer, a normalization layer and a linear output layer. In the present invention, the large language model with some layers frozen is a transformer model. The first embodiment of the present invention trains the large language model with some layers frozen to generate a fault diagnosis model through the following steps:

[0087] S4-1: Extract multiple data sets of faulty bearings from the database, each data set includes the current n pre-trained data vectors P n and the current n-th test data vector F n , the pre-training data vector P n The input embedding layer of the large language model with some layers frozen, and the output vector output from the linear output layer is:

[0088] ,

[0089] in, , is the time domain feature vector of the current time n, is the frequency feature vector of the current nth order, , , To freeze the current parameter vector of the large language model of some layers, freeze the parameters of the large language model of some layers, including: the key value P, query value Q, value V of the multi-head attention mechanism layer, and the matrix coefficient W and bias B of the feedforward neural network, K is a positive integer greater than or equal to 2; is the kth value of the output vector output from the second residual connection layer l at the current time n; F nk The k-th value of the current n-th test data vector, k=1, 2, ..., K;

[0090] S4-2: Assume the loss function L is:

[0091] ,

[0092] The gradients of the mean and variance are calculated as follows:

[0093] ,

[0094] S4-3: Update the current parameter vector of the large language model with some layers frozen according to the following formula:

[0095] ;

[0096] In the formula, is the learning rate, Express The gradient of

[0097] S4-4: Determine whether the loss function value is the minimum. If so, end the training. If not, make the new parameter vector Replace the current parameter vector , and returns to S4-1.

[0098] The first embodiment of the present invention can use the above training method to enable the large language model with some layers frozen to classify bearing faults and obtain faulty bearings.

[0099] Second embodiment

[0100] The second embodiment of the present invention only describes the contents that are different from the first embodiment, and the same contents are not repeated.

[0101] The second embodiment of the present invention provides a bearing fault diagnosis method based on a large language model, comprising the following steps:

[0102] Step 1: Obtain the vibration electrical signal of the bearing through a vibration sensor;

[0103] Step 2: Extracting time domain feature vectors and frequency domain feature vectors according to the sampling values ​​of the vibration electrical signal through the feature extraction module;

[0104] Step 3: Splice the time domain feature vector and the frequency domain feature vector to form a spliced ​​vector through a splicing module;

[0105] Step 4: embed the feature name label vector in front of or behind each feature in the concatenated vector through the first embedding module to form a fine-tuning vector; embed the bearing label vector in front of or behind the fine-tuning vector through the second embedding module to form an input vector and then provide it to the fault diagnosis model;

[0106] Step 5: The fault diagnosis model performs fault diagnosis on the bearing according to the input vector, and the fault diagnosis model is trained by a large language model with some layers frozen.

[0107] The beneficial effects of the second embodiment are the same as those of the first embodiment, and will not be described again here.

[0108] Third embodiment

[0109] A third embodiment of the present invention provides a computer program product, which includes a computer program code. The computer program code can be called by a processor to execute the method described in the second embodiment.

[0110] The beneficial effects of the third embodiment are the same as those of the second embodiment, and will not be described again here.

[0111] Fourth embodiment

[0112] A fourth embodiment of the present invention provides a computer storage medium for storing computer program codes, where the computer program codes can be called by a processor to execute the method described in the second embodiment.

[0113] The beneficial effects of the fourth embodiment are the same as those of the second embodiment, and will not be described again here.

[0114] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined. The meaning of "several" is one or more, unless otherwise clearly and specifically defined.

[0115] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A bearing fault diagnosis system based on a large language model, comprising a vibration sensor, the vibration sensor is used to obtain a vibration electrical signal of the bearing, characterized in that: The system also includes an extraction module, a splicing module, a first embedding module, a second embedding module and a fault diagnosis model, wherein the feature extraction module extracts a time domain feature vector and a frequency domain feature vector according to the sampling value of the vibration electrical signal; the splicing module splices the time domain feature vector and the frequency domain feature vector to form a splicing vector; the first embedding module embeds a feature name label vector in front of or behind each feature in the splicing vector to form a fine-tuning vector; the second embedding module embeds a bearing label vector in front of or behind the fine-tuning vector to form an input vector and then provides it to the fault diagnosis model; the fault diagnosis model performs fault diagnosis on the bearing according to the input vector, and the fault diagnosis model is trained by a large language model with some layers frozen, and the large language model with some layers frozen includes an input embedding layer, a position embedding layer, a multi-head attention mechanism layer, a first residual connection layer, a feedforward neural network and a second residual connection layer, a normalization layer and a linear output layer.

2. The bearing fault diagnosis system based on large language model according to claim 1 is characterized in that: The feature extraction module includes a time domain feature extraction module, which includes a time domain mean calculator, a time domain standard deviation calculator, an absolute mean calculator, a peak calculator, a waveform index calculator, a pulse index calculator, a peak index calculator, a skewness calculator, a kurtosis calculator, a kurtosis index calculator, a root mean square calculator, a deviation calculator and a time domain feature vector generator, wherein the mean calculator calculates a mean p1 according to the time domain sampling value of the bearing vibration electrical signal; the time domain standard deviation calculator calculates a time domain standard deviation p2 according to the time domain sampling value of the bearing vibration electrical signal and the time domain mean p1; the absolute mean calculator calculates an absolute mean p4 according to the time domain sampling value of the bearing vibration electrical signal; the peak calculator calculates a peak value p5 according to the time domain sampling value of the bearing vibration electrical signal; the waveform index calculator calculates a waveform index p according to the time domain standard deviation p2 and the absolute mean p4. 11 ; Pulse Index Calculator calculates the pulse index p based on the absolute mean p4 and peak p5 12 ; The peak index calculator calculates the peak index p based on the time domain standard deviation p2 and peak value p5 10 ; The skewness calculator calculates the skewness p6 according to the time domain sampling value of the bearing vibration electrical signal; the kurtosis calculator calculates the kurtosis p7 according to the time domain sampling value of the bearing vibration electrical signal; the kurtosis index calculator calculates the kurtosis index p9 according to the skewness p6 and the kurtosis p7; the root mean square calculator calculates the root mean square p3 according to the time domain sampling value of the bearing vibration electrical signal; the deviation calculator calculates the deviation p8 according to the time domain sampling value of the bearing vibration electrical signal; the time domain feature vector generator generates the time domain feature vector P according to the output values ​​of the time domain mean calculator, the time domain standard deviation calculator, the absolute mean calculator, the peak calculator, the waveform index calculator, the pulse index calculator, the peak index calculator, the skewness calculator, the kurtosis calculator, the kurtosis index calculator, the root mean square calculator and the deviation calculator T =[p1, p2, p3, p4, p5, p6, p7, p8, p9, p 10 , p 11 ,p 12 ].

3. The bearing fault diagnosis system based on large language model according to claim 2 is characterized in that: The feature extraction module includes a frequency domain feature extraction module, which includes a frequency mean calculator, a frequency deviation calculator, a frequency skewness calculator, a frequency kurtosis calculator, a center of gravity frequency calculator, a frequency standard deviation calculator, a variable parameter calculator, an eighth-order moment calculator, a sixteenth-order moment calculator, a regularity calculator, a frequency root mean square calculator, an average frequency calculator and a frequency domain feature vector generator, wherein the frequency mean calculator calculates the frequency mean according to the frequency domain signal of the bearing vibration electrical signal sampling value The frequency deviation calculator is based on the frequency domain signal and frequency mean value p of the bearing vibration electrical signal sampling value. 13 Calculate the frequency deviation p 14 The frequency skewness calculator is based on the frequency domain signal and frequency mean value p of the bearing vibration electrical signal sampling value. 13 and frequency deviation p 14 Calculate the frequency skewness p 15 The frequency kurtosis calculator is based on the frequency domain signal and frequency mean p of the bearing vibration electrical signal sampling value. 13 and frequency deviation p 14 Calculate the frequency kurtosis p 16 The center of gravity frequency calculator calculates the center of gravity frequency p according to the frequency domain signal and sampling frequency of the bearing vibration electrical signal sampling value. 17 The frequency standard deviation calculator is based on the frequency domain signal, sampling frequency and center of gravity frequency p of the bearing vibration electrical signal sampling value. 17 Calculate the frequency standard deviation p 18 ; The variable parameter calculator is based on the center of gravity frequency p 17 Frequency standard deviation p 18 Calculate the variable parameter p 22 ; The eighth-order moment calculator is based on the center of gravity frequency p 17 , sampling frequency, frequency standard deviation p 18 The eighth-order moment p is calculated by the frequency domain signal of the sampling value of the bearing vibration electrical signal 23 ; The sixteenth-order moment calculator is based on the center of gravity frequency p 17 , sampling frequency, frequency standard deviation p 18 The sixteenth order moment p is calculated by the frequency domain signal of the bearing vibration electrical signal sampling value 24 ; The regularity calculator calculates the regularity p according to the frequency domain signal and sampling frequency of the bearing vibration electrical signal sampling value. 21 The frequency RMS calculator calculates the frequency RMS p according to the frequency domain signal and sampling frequency of the bearing vibration electrical signal sampling value. 19 The average frequency calculator calculates the average frequency p according to the frequency domain signal and sampling frequency of the bearing vibration electrical signal sampling value. 20 The frequency domain feature vector is generated according to the output values ​​of the frequency mean calculator, frequency deviation calculator, frequency skewness calculator, frequency kurtosis calculator, center of gravity frequency calculator, frequency standard deviation calculator, variable parameter calculator, eighth-order moment calculator, sixteenth-order moment calculator, regularity calculator, frequency root mean square calculator and average frequency calculator generator: P F =[p 13 , p 14 , p 15 , p 16 ,p 17 , p 18 , p 19 , p 20 , p 21 , p 22 , p 23 , p 24 ].

4. The bearing fault diagnosis system based on large language model according to claim 3 is characterized in that: The large language model with some layers frozen is trained through the following steps to generate a fault diagnosis model: S4-1: Extract multiple data sets of faulty bearings from the database, each data set includes the current n pre-trained data vectors P n and the current n-th test data vector F n , the pre-training data vector P n The input embedding layer of the large language model with some layers frozen, and the output vector output from the linear output layer is: , in, , P Tn is the time domain feature vector of the current time n, P Fn is the frequency feature vector of the current nth order, , , To freeze the current parameter vector of the large language model of some layers, freeze the parameters of the large language model of some layers, including: the key value P, query value Q, value V of the multi-head attention mechanism layer, and the matrix coefficient W and bias B of the feedforward neural network, K is a positive integer greater than or equal to 2; is the kth value of the output vector output from the second residual connection layer l at the current time n; F nk The k-th value of the current n-th test data vector, k=1, 2, ..., K; S4-2: Assume the loss function L is: , The gradients of the mean and variance are calculated as follows: , S4-3: Update the current parameter vector of the large language model with some layers frozen according to the following formula: ; In the formula, is the learning rate, Express The gradient of S4-4: Determine whether the loss function value is the minimum. If so, end the training. If not, make the new parameter vector Replace the current parameter vector , and returns to S4-1.

5. A bearing fault diagnosis method based on a large language model, characterized in that: The steps include: Step 1: Obtain the vibration electrical signal of the bearing through a vibration sensor; Step 2: Extracting time domain feature vectors and frequency domain feature vectors according to the sampling values ​​of the vibration electrical signal through the feature extraction module; Step 3: Splice the time domain feature vector and the frequency domain feature vector to form a spliced ​​vector through a splicing module; Step 4: embed the feature name label vector in front of or behind each feature in the concatenated vector through the first embedding module to form a fine-tuning vector; embed the bearing label vector in front of or behind the fine-tuning vector through the second embedding module to form an input vector and then provide it to the fault diagnosis model; Step 5: The fault diagnosis model performs bearing fault diagnosis according to the input vector. The fault diagnosis model is trained by a large language model with some layers frozen. The large language model with some layers frozen includes an input embedding layer, a position embedding layer, a multi-head attention mechanism layer, a first residual connection layer, a feedforward neural network and a second residual connection layer, a normalization layer and a linear output layer.

6. The bearing fault diagnosis method based on large language model according to claim 5 is characterized in that: The feature extraction module includes a time domain feature extraction module, which includes a time domain mean calculator, a time domain standard deviation calculator, an absolute mean calculator, a peak calculator, a waveform index calculator, a pulse index calculator, a peak index calculator, a skewness calculator, a kurtosis calculator, a kurtosis index calculator, a root mean square calculator, a deviation calculator and a time domain feature vector generator, wherein the mean calculator calculates a mean p1 according to the time domain sampling value of the bearing vibration electrical signal; the time domain standard deviation calculator calculates a time domain standard deviation p2 according to the time domain sampling value of the bearing vibration electrical signal and the time domain mean p1; the absolute mean calculator calculates an absolute mean p4 according to the time domain sampling value of the bearing vibration electrical signal; the peak calculator calculates a peak value p5 according to the time domain sampling value of the bearing vibration electrical signal; the waveform index calculator calculates a waveform index p according to the time domain standard deviation p2 and the absolute mean p4. 11 ; Pulse Index Calculator calculates the pulse index p based on the absolute mean p4 and peak p5 12 ; The peak index calculator calculates the peak index p based on the time domain standard deviation p2 and peak value p5 10 ; The skewness calculator calculates the skewness p6 according to the time domain sampling value of the bearing vibration electrical signal; the kurtosis calculator calculates the kurtosis p7 according to the time domain sampling value of the bearing vibration electrical signal; the kurtosis index calculator calculates the kurtosis index p9 according to the skewness p6 and the kurtosis p7; the root mean square calculator calculates the root mean square p3 according to the time domain sampling value of the bearing vibration electrical signal; the deviation calculator calculates the deviation p8 according to the time domain sampling value of the bearing vibration electrical signal; the time domain feature vector generator generates the time domain feature vector P according to the output values ​​of the time domain mean calculator, the time domain standard deviation calculator, the absolute mean calculator, the peak calculator, the waveform index calculator, the pulse index calculator, the peak index calculator, the skewness calculator, the kurtosis calculator, the kurtosis index calculator, the root mean square calculator and the deviation calculator T =[p1, p2, p3, p4, p5, p6, p7, p8, p9, p 10 , p 11 ,p 12 ].

7. The bearing fault diagnosis method based on large language model according to claim 6 is characterized in that: The feature extraction module includes a frequency domain feature extraction module, which includes a frequency mean calculator, a frequency deviation calculator, a frequency skewness calculator, a frequency kurtosis calculator, a center of gravity frequency calculator, a frequency standard deviation calculator, a variable parameter calculator, an eighth-order moment calculator, a sixteenth-order moment calculator, a regularity calculator, a frequency root mean square calculator, an average frequency calculator and a frequency domain feature vector generator, wherein the frequency mean calculator calculates the frequency mean according to the frequency domain signal of the bearing vibration electrical signal sampling value The frequency deviation calculator is based on the frequency domain signal and frequency mean value p of the bearing vibration electrical signal sampling value. 13 Calculate the frequency deviation p 14 The frequency skewness calculator is based on the frequency domain signal and frequency mean value p of the bearing vibration electrical signal sampling value. 13 and frequency deviation p 14 Calculate the frequency skewness p 15 The frequency kurtosis calculator is based on the frequency domain signal and frequency mean p of the bearing vibration electrical signal sampling value. 13 and frequency deviation p 14 Calculate the frequency kurtosis p 16 The center of gravity frequency calculator calculates the center of gravity frequency p according to the frequency domain signal and sampling frequency of the bearing vibration electrical signal sampling value. 17 The frequency standard deviation calculator is based on the frequency domain signal, sampling frequency and center of gravity frequency p of the bearing vibration electrical signal sampling value. 17 Calculate the frequency standard deviation p 18 ; The variable parameter calculator is based on the center of gravity frequency p 17 Frequency standard deviation p 18 Calculate the variable parameter p 22 ; The eighth-order moment calculator is based on the center of gravity frequency p 17 , sampling frequency, frequency standard deviation p 18 The eighth-order moment p is calculated by the frequency domain signal of the sampling value of the bearing vibration electrical signal 23 ; The sixteenth-order moment calculator is based on the center of gravity frequency p 17 , sampling frequency, frequency standard deviation p 18 The sixteenth order moment p is calculated by the frequency domain signal of the bearing vibration electrical signal sampling value 24 ; The regularity calculator calculates the regularity p according to the frequency domain signal and sampling frequency of the bearing vibration electrical signal sampling value. 21 The frequency RMS calculator calculates the frequency RMS p according to the frequency domain signal and sampling frequency of the bearing vibration electrical signal sampling value. 19 The average frequency calculator calculates the average frequency p according to the frequency domain signal and sampling frequency of the bearing vibration electrical signal sampling value. 20 The frequency domain feature vector generator generates a frequency domain feature vector according to the output values ​​of the frequency mean calculator, frequency deviation calculator, frequency skewness calculator, frequency kurtosis calculator, center of gravity frequency calculator, frequency standard deviation calculator, variable parameter calculator, eighth-order moment calculator, sixteenth-order moment calculator, regularity calculator, frequency root mean square calculator and average frequency calculator: P F =[p 13 , p 14 , p 15 , p 16 ,p 17 , p 18 , p 19 , p 20 , p 21 , p 22 , p 23 , p 24 ].

8. The bearing fault diagnosis method based on large language model according to claim 7 is characterized in that: The large language model with some layers frozen is trained through the following steps to generate a fault diagnosis model: S4-1: Extract multiple data sets of faulty bearings from the database, each data set includes the current n pre-trained data vectors and the current n test data vector , the pre-training data vector The input embedding layer of the large language model with some layers frozen, and the output vector output from the linear output layer is: , in, , is the time domain feature vector of the current time n, is the frequency feature vector of the current nth order, , , To freeze the current parameter vector of the large language model of some layers, freeze the parameters of the large language model of some layers, including: the key value P, query value Q, value V of the multi-head attention mechanism layer, and the matrix coefficient W and bias B of the feedforward neural network, K is a positive integer greater than or equal to 2; is the kth value of the output vector output from the second residual connection layer l at the current time n; F nk The k-th value of the current n-th test data vector, k=1, 2, ..., K; S4-2: Assume the loss function L is: , The gradients of the mean and variance are calculated as follows: , S4-3: Update the current parameter vector of the large language model with some layers frozen according to the following formula: ;, In the formula, is the learning rate, Express The gradient of S4-4: Determine whether the loss function value is the minimum. If so, end the training. If not, make the new parameter vector Replace the current parameter vector , and returns to S4-1.

9. A computer program product, characterized in that The method comprises computer program codes, wherein the computer program codes can be called by a processor to execute the method according to any one of claims 5 to 8.

Citation Information

Patent Citations

  • Deep learning rolling bearing fault diagnosis method and system

    CN115525866A

  • Fault diagnosis method for rolling bearing

    CN115221930A

  • Bearing fault diagnosis model training method and bearing fault diagnosis method and device

    CN117890111A