Multi-dimensional similarity measurement and adaptive weight adjustment combined transfer learning method

Abstract

The invention discloses a multi-dimensional similarity measurement and adaptive weight adjustment combined transfer learning method, and relates to the technical field of model training, and the method comprises the steps: for any to-be-detected mechanical equipment, calling at least two historical maintenance records of the to-be-detected mechanical equipment in different maintenance mechanisms, carrying out integration processing and cleaning processing on all the historical maintenance records to generate maintenance data of the mechanical equipment to be detected; performing feature extraction on the maintenance data and the sample data of the different damage categories through a pre-training model to obtain a first feature vector and a second feature vector, and calculating the similarity between the maintenance data and the sample data of the different damage categories according to the first feature vector and the second feature vector to obtain the similarity between the maintenance data and the sample data of the different damage categories; and taking the service life of the damage category corresponding to the damage category sample data with the highest similarity as a detection result of the to-be-detected mechanical equipment. According to the invention, the potential problem of the to-be-detected mechanical equipment can be effectively identified, and the detection accuracy and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the field of model training technology, and in particular to a transfer learning method combining multi-dimensional similarity measurement with adaptive weight adjustment. Background Art

[0002] In the field of mechanical equipment maintenance, the amount and characteristics of data for different damage types vary significantly, and data acquisition is costly. Traditional single-damage model training methods struggle to fully utilize limited data resources, resulting in limited model generalization and diagnostic accuracy. Existing transfer learning methods often lack a detailed understanding of sample relevance and importance when processing source and target data samples, making it difficult to effectively utilize useful information in the source data. This impacts the model's adaptability and diagnostic accuracy on the target data, further hindering the ability to accurately determine the damage category or condition of the mechanical equipment based on the source data. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to address the deficiencies of the existing technology and specifically provide a transfer learning method combining multi-dimensional similarity measurement with adaptive weight adjustment, as follows:

[0004] 1) In the first aspect, the present invention provides a transfer learning method combining multi-dimensional similarity measurement with adaptive weight adjustment. The specific technical solution is as follows:

[0005] For any mechanical equipment to be inspected, retrieve at least two historical maintenance records of the mechanical equipment to be inspected from different maintenance institutions, integrate and clean all historical maintenance records, and generate maintenance data for the mechanical equipment to be inspected;

[0006] Performing feature extraction on the maintenance data and sample data of different damage categories using a pre-trained model to obtain a first feature vector corresponding to the maintenance data and a second feature vector corresponding to the sample data of different damage categories; calculating the similarity between the maintenance data and the sample data of different damage categories based on the first feature vector and the second feature vector; and using the service life of the damage category corresponding to the sample data of the damage category with the highest similarity as the detection result of the mechanical equipment to be detected;

[0007] The weight of the pre-trained model is determined by adaptive adjustment.

[0008] The beneficial effects of the transfer learning method of a multi-dimensional similarity measurement combined with adaptive weight adjustment provided by the present invention are as follows:

[0009] First, by integrating and cleaning historical maintenance records from different maintenance organizations, comprehensive and high-quality maintenance data can be generated, providing a reliable data basis for subsequent analysis and detection; second, by using pre-trained models for feature extraction, the rich feature representation capabilities learned by pre-trained models on large-scale data can be fully utilized, thereby more accurately extracting the features of maintenance data and damage category sample data; third, by calculating the similarity between feature vectors to determine the damage category, this method can effectively identify potential problems of the mechanical equipment to be inspected and improve the accuracy and efficiency of detection; finally, the adaptive adjustment method of the pre-trained model weights can further optimize the model performance, making it better adapted to specific maintenance data and damage categories, and further improving the reliability of the detection results.

[0010] Based on the above solution, the present invention can also be improved as follows.

[0011] Furthermore, the cleaning process is specifically as follows:

[0012] Abnormal data is eliminated, and data in different formats under the same project are processed uniformly.

[0013] Furthermore, the similarity includes: feature similarity, label similarity and time series similarity.

[0014] Furthermore, the calculation process of the feature similarity is specifically as follows:

[0015]

[0016] Among them, S feat (x i ,y j ) represents feature similarity, f(·) represents the feature vector extracted by the convolutional neural network, x i represents the first eigenvector, y j represents the second eigenvector.

[0017] Furthermore, the tag similarity calculation process is specifically as follows:

[0018]

[0019] Among them, S label (x i ,y j ) represents the label similarity, g(·) represents the function that maps the label text to the semantic space, l i Indicates the label corresponding to the maintenance data, l j Indicates the labels corresponding to sample data of different damage categories.

[0020] Furthermore, the calculation process of the time series similarity is specifically as follows:

[0021]

[0022] Among them, S time (x i ,y j ) represents the time series similarity, DTW(T i ,T j ) represents the time series T i and T j The dynamic time warping distance between i Represents the time series corresponding to the maintenance data, T j Represents the time series corresponding to sample data of different damage categories.

[0023] 2) In a second aspect, the present invention also provides a transfer learning system that combines multi-dimensional similarity measurement with adaptive weight adjustment. The specific technical solution is as follows:

[0024] The processing module is used to: for any mechanical equipment to be detected, retrieve at least two historical maintenance records of the mechanical equipment to be detected from different maintenance institutions, and integrate and clean all historical maintenance records to generate maintenance data of the mechanical equipment to be detected;

[0025] The detection module is configured to: perform feature extraction on the maintenance data and sample data of different damage categories using a pre-trained model to obtain a first feature vector corresponding to the maintenance data and a second feature vector corresponding to the sample data of different damage categories; calculate the similarity between the maintenance data and the sample data of different damage categories based on the first feature vector and the second feature vector; and use the service life of the damage category corresponding to the sample data of the damage category with the highest similarity as the detection result of the mechanical equipment to be detected;

[0026] The weight of the pre-trained model is determined by adaptive adjustment.

[0027] 3) In a third aspect, the present invention further provides an electronic device, comprising a processor, wherein the processor is coupled to a memory, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor so that the electronic device implements any of the above methods.

[0028] 4) In a fourth aspect, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores at least one computer program, and the at least one computer program is loaded and executed by a processor to enable a computer to implement any of the above methods.

[0029] It should be noted that the beneficial effects achieved by the technical solutions of the second to fourth aspects of the present invention and the corresponding possible implementation methods can be found in the above-mentioned technical effects of the first aspect and its corresponding possible implementation methods, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0031] Figure 1 A flow chart of a transfer learning method combining multi-dimensional similarity measurement with adaptive weight adjustment according to an embodiment of the present invention;

[0032] Figure 2 This is a structural framework diagram of an electronic device of the present invention. DETAILED DESCRIPTION

[0033] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0034] like Figure 1 As shown, a transfer learning method combining multi-dimensional similarity measurement with adaptive weight adjustment in an embodiment of the present invention includes the following steps:

[0035] S1, for any mechanical equipment to be inspected, retrieve at least two historical maintenance records of the mechanical equipment to be inspected from different maintenance institutions, and integrate and clean all the historical maintenance records to generate maintenance data of the mechanical equipment to be inspected;

[0036] S2, extracting features from the maintenance data and sample data of different damage categories using a pre-trained model to obtain a first feature vector corresponding to the maintenance data and a second feature vector corresponding to the sample data of different damage categories; calculating similarities between the maintenance data and the sample data of different damage categories based on the first feature vector and the second feature vector; and using the service life of the damage category corresponding to the sample data of the damage category with the highest similarity as the detection result of the mechanical equipment to be detected;

[0037] The weight of the pre-trained model is determined by adaptive adjustment.

[0038] The beneficial effects of the transfer learning method of a multi-dimensional similarity measurement combined with adaptive weight adjustment provided by the present invention are as follows:

[0039] First, by integrating and cleaning historical maintenance records from different maintenance organizations, comprehensive and high-quality maintenance data can be generated, providing a reliable data basis for subsequent analysis and detection; second, by using pre-trained models for feature extraction, the rich feature representation capabilities learned by pre-trained models on large-scale data can be fully utilized, thereby more accurately extracting the features of maintenance data and damage category sample data; third, by calculating the similarity between feature vectors to determine the damage category, this method can effectively identify potential problems of the mechanical equipment to be inspected and improve the accuracy and efficiency of detection; finally, the adaptive adjustment method of the pre-trained model weights can further optimize the model performance, making it better adapted to specific maintenance data and damage categories, and further improving the reliability of the detection results.

[0040] Different maintenance organizations refer to any organization that can inspect or repair mechanical equipment and generate maintenance records.

[0041] Mechanical equipment can be diagnostic equipment:

[0042] Imaging equipment: including X-ray machines, CT scanners, MRI (magnetic resonance imaging) equipment, ultrasonic diagnostic equipment, etc., used to obtain images of the internal structure and function of the human body.

[0043] Endoscopic equipment: such as fiber endoscopes, electronic endoscopes, etc., used to examine internal organs of the human body.

[0044] In vitro diagnostic equipment: including laboratory testing equipment and point-of-care testing equipment, used to test blood, urine and other samples.

[0045] Treatment equipment:

[0046] Surgical instruments: including general surgical instruments, light-guided surgical instruments (such as laser therapy machines), and auxiliary surgical instruments (such as anesthesia machines, ventilators, and extracorporeal circulation equipment).

[0047] Radiotherapy equipment: such as deep X-ray therapy machine, cobalt-60 therapy machine, accelerator, gamma knife, etc.

[0048] Interventional treatment equipment: such as angiography machines, interventional catheters, etc.

[0049] Rehabilitation therapy equipment: such as rehabilitation robots, physical therapy equipment, functional training equipment, etc.

[0050] Patient monitoring equipment

[0051] Vital signs monitoring equipment: such as electrocardiogram (ECG) machine, electroencephalogram (EEG) machine, electromyogram (EMG) machine, multi-parameter monitor, etc.

[0052] Respiratory support equipment: such as ventilators, oxygen concentrators, oxygen cylinders, etc.

[0053] The specific process of integrating and cleaning all historical maintenance records to generate maintenance data for the mechanical equipment to be tested is as follows:

[0054] The integration process is:

[0055] According to the unique number of the mechanical equipment to be inspected, all data related to the unique number are integrated and processed.

[0056] The cleaning process is:

[0057] Remove duplicate data: Check whether there are duplicate records in the data set consisting of unique labels and corresponding data, and delete them.

[0058] The criteria for determining duplicate records are: maintenance records or inspection records of the same mechanical equipment at different maintenance agencies within the same time period.

[0059] The same time period refers to a preset period, such as 1 hour or 6 hours.

[0060] Handling missing data: Fill in missing data.

[0061] The specific filling process is as follows:

[0062] Determine the data category of the missing data. If it is time data or maintenance cost data, then, in combination with the time series of the data set, use the time corresponding to the previous column and the time corresponding to the next column of the column where the missing data is located as the start time and end time of the target time range of the missing data. In the historical database, retrieve the first n groups of relevant historical data with the highest similarity to the type and model of the mechanical equipment to be detected, and determine the maintenance cycle of the maintenance type corresponding to the missing data in these n groups of relevant historical data. Based on this maintenance cycle, determine the first maintenance record under this maintenance type, and determine the target time that overlaps with the maintenance cycle within the target time range. This target time is determined as the filling data for filling processing. Determine the maintenance fee based on the filling results of the time data.

[0063] When the data category of the missing data is damage area data or component replacement data, feature extraction is performed on the current image information of the mechanical equipment to be inspected to determine the repair area, and the damage area is determined by the repair area. The current image information is compared with the original image information of the mechanical equipment to be inspected to obtain the component replacement information.

[0064] The specific process of retrieving the top n groups of relevant historical data that are most similar to the type and model of the mechanical equipment to be detected is as follows:

[0065] Determine in the historical database whether there are historical mechanical devices of the same or similar type as the mechanical device to be detected, where similarity refers to the same function and similar size or refers to the top m historical mechanical devices with the largest number of identical parts as the mechanical device to be detected.

[0066] Search historical mechanical equipment for the first historical mechanical equipment having the same or similar equipment model as the mechanical equipment to be detected, where similarity refers to the existence of a predecessor-descendant relationship with the equipment model of the mechanical equipment to be detected (such as the first generation, second generation and third generation products of a certain machine).

[0067] According to the component corresponding to the missing data and the names of other first components adjacent to the component, the component and the first component are labeled, and the weights are adjusted to obtain the first weights corresponding to the component and the first component. According to the first weight and the second weights (preset weights) corresponding to different components, the scores corresponding to all first historical mechanical equipment are calculated, that is, the similarities, that is, the similarity calculation results are arranged in descending order, and the relevant historical data corresponding to the first historical mechanical equipment corresponding to the first n similarities are output.

[0068] The specific process of determining the maintenance cycle of the maintenance type corresponding to the missing data in the n sets of relevant historical data is as follows:

[0069] Determine the first maintenance cycle of the maintenance type corresponding to the first historical mechanical equipment, and sort all the first maintenance cycles according to the descending order result above, and calculate the target maintenance cycle corresponding to the maintenance type according to the second weight corresponding to the descending order result and the first maintenance cycle, and determine the target maintenance cycle as the maintenance cycle of the maintenance type corresponding to the missing data determined in the n groups of related historical data.

[0070] Eliminate abnormal data: Identify and eliminate abnormal data based on the normal operating range of the equipment and maintenance experience. For example, data that exceeds the normal operating parameters of the equipment or data with obviously unreasonable repair costs will be eliminated.

[0071] Data consistency check: Checks whether the data is logically consistent. For example, the repair time should be later than the fault time, and the repair cost should match the repair content.

[0072] Data validation and quality assessment

[0073] Verify data accuracy: Verify data accuracy by comparing with actual maintenance records, equipment operating status, etc.

[0074] Assess data quality: Perform quality assessment on cleaned data to ensure data integrity, accuracy, and consistency.

[0075] The test result refers to the corresponding service life of the mechanical equipment to be tested.

[0076] Furthermore, the cleaning process is specifically as follows:

[0077] Abnormal data is eliminated, and data in different formats under the same project are processed uniformly.

[0078] Furthermore, the similarity includes: feature similarity, label similarity and time series similarity.

[0079] Furthermore, the calculation process of the feature similarity is specifically as follows:

[0080]

[0081] Among them, S feat (x i ,y j ) represents feature similarity, f(·) represents the feature vector extracted by the convolutional neural network, x i represents the first eigenvector, y j represents the second eigenvector.

[0082] Furthermore, the tag similarity calculation process is specifically as follows:

[0083]

[0084] Among them, S label (x i ,y j ) represents the label similarity, g(·) represents the function that maps the label text to the semantic space, l i Indicates the label corresponding to the maintenance data, l j Indicates the labels corresponding to sample data of different damage categories.

[0085] Furthermore, the calculation process of the time series similarity is specifically as follows:

[0086]

[0087] Among them, S time (x i ,y j ) represents the time series similarity, DTW(T i ,T j ) represents the time series T i and T j The dynamic time warping distance between i Represents the time series corresponding to the maintenance data, T j Represents the time series corresponding to sample data of different damage categories.

[0088] Example 1, 1. Data Cleaning

[0089] 1. Remove duplicate records: When integrating data from multiple repair shops, duplicate records may occur because different repair shops may record the same basic information about machinery, such as unique identification, repair details, inspection items, and costs. Data cleansing can identify and remove these duplicate records, ensuring data uniqueness and accuracy.

[0090] 2. Correcting erroneous data: Some data may contain incorrectly entered test indicator values. By setting reasonable value range thresholds, out-of-range data can be marked and corrected, thereby improving data quality.

[0091] 3. Standardize data formats: Different data may have different record formats. For example, dates may be in "YYYY-MM-DD" or "MM / DD / YYYY." During data migration, these date formats need to be standardized to facilitate subsequent data processing and analysis.

[0092] 4. Numerical normalization: For numerical data of different dimensions, in order to eliminate the influence of dimension, the Min-Max normalization method is used to uniformly map these values to the [0,1] interval, making different data numerically comparable.

[0093] 2. Multi-dimensional Similarity Calculation

[0094] Responsible for calculating the similarity between source data and target data samples in multiple dimensions such as features, labels, and time series.

[0095] The specific steps are as follows:

[0096] 1. For feature similarity calculation, use the pre-trained convolutional neural network to extract the feature vector of the sample, and then calculate the cosine similarity. For example, for the source data sample x i and target damaged category sample y j , its feature similarity S feat (x i ,y j ) can be expressed as:

[0097]

[0098] Where f(·) represents the feature vector extracted by the convolutional neural network.

[0099] 2. For label similarity calculation, convert the sample label text into a semantic vector and use the cosine similarity calculation method. Let the source data sample x i The label is l i , target damaged category sample yj The label is y j , its label similarity S label (x i ,y j ) can be expressed as:

[0100]

[0101] Here, g(·) represents the function that maps the label text to the semantic space, and the word vector is obtained by training the Word2Vec model.

[0102] 3. For time series similarity calculation, dynamic time warping algorithm is used. Let the source data sample x i The time series is T i ={t i1 ,t i2 ,..t in}, target damaged category sample y j The time series is T j ={t j1 ,t j2 ,..t jn}, its time series similarity S time (x i ,y j ) can be expressed as:

[0103]

[0104] Among them, DTW(T i ,T j ) represents the time series T i and T j The dynamic time warping distance between them.

[0105] 3. Adaptive Weight Adjustment

[0106] Based on the multi-dimensional similarity calculation results and model training feedback information, the weights of the source data and target damaged category samples are dynamically adjusted. The specific algorithm is as follows:

[0107] 1. Initialize sample weights: Based on the prior knowledge of the source data and target damage categories, for each source data sample x i and target damaged category sample y j Set initial weights and For example, a relatively high initial weight is given to source data samples that have a high similarity with the target damage category features.

[0108] 2. During the model training process, the contribution of each sample to the model training is evaluated by monitoring the model's loss function value L and accuracy A. For the kth iteration, the update amount of the sample weight is calculated and

[0109]

[0110] Where η is the learning rate, and are the gradients of the loss function with respect to the weights of the source data samples and the target damaged class samples, respectively.

[0111] 3. Combined with the reinforcement learning mechanism, the sample weights are adjusted according to the changes in the model's diagnostic accuracy on the target damage category. The reward function R is defined as:

[0112] R=A current -A previous

[0113] Among them, A current and A previous Represent the diagnostic accuracy of the current iteration and the previous iteration on the target damage category, respectively.

[0114] According to the reward signal R, update the sample weight:

[0115]

[0116] Here, α is the learning rate of reinforcement learning.

[0117] The above scheme can quickly lock the final target damage category sample, and output the service life corresponding to the damage category sample as the detection result.

[0118] In the above embodiments, although the steps are numbered S1, S2, etc., these are only specific embodiments given by the present invention. Those skilled in the art may adjust the execution order of S1, S2, etc. according to actual conditions, which is also within the scope of protection of the present invention. It can be understood that in some embodiments, some or all of the above embodiments may be included.

[0119] The present invention also provides a transfer learning system that combines multi-dimensional similarity measurement with adaptive weight adjustment. The specific technical solution is as follows:

[0120] The processing module is used to: for any mechanical equipment to be detected, retrieve at least two historical maintenance records of the mechanical equipment to be detected from different maintenance institutions, and integrate and clean all historical maintenance records to generate maintenance data of the mechanical equipment to be detected;

[0121] The detection module is configured to: perform feature extraction on the maintenance data and sample data of different damage categories using a pre-trained model to obtain a first feature vector corresponding to the maintenance data and a second feature vector corresponding to the sample data of different damage categories; calculate the similarity between the maintenance data and the sample data of different damage categories based on the first feature vector and the second feature vector; and use the service life of the damage category corresponding to the sample data of the damage category with the highest similarity as the detection result of the mechanical equipment to be detected;

[0122] The weight of the pre-trained model is determined by adaptive adjustment.

[0123] It should be noted that the beneficial effects of the transfer learning system of a multi-dimensional similarity metric combined with adaptive weight adjustment provided by the above embodiment are the same as the beneficial effects of the transfer learning method of a multi-dimensional similarity metric combined with adaptive weight adjustment, which will not be repeated here. In addition, when the system provided by the above embodiment realizes its functions, it only uses the division of the above-mentioned functional modules as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the system can be divided into different functional modules according to actual conditions to complete all or part of the functions described above. In addition, the system and method embodiments provided by the above embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0124] like Figure 2 As shown, an electronic device 300 according to an embodiment of the present invention includes a processor 320, which is coupled to a memory 310. The memory 310 stores at least one computer program 330. The at least one computer program 330 is loaded and executed by the processor 320 to enable the electronic device 300 to implement any of the above methods. Specifically:

[0125] The electronic device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors 320 (Central Processing Units, CPU) and one or more memories 310, wherein the one or more memories 310 store at least one computer program 330, and the at least one computer program 330 is loaded and executed by the one or more processors 320, so that the electronic device 300 implements a multi-dimensional similarity measurement combined with adaptive weight adjustment transfer learning method provided in the above embodiment. Of course, the electronic device 300 may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output. The electronic device 300 may also include other components for realizing device functions, which will not be described in detail here.

[0126] A computer-readable storage medium according to an embodiment of the present invention stores at least one computer program, and the at least one computer program is loaded and executed by a processor to enable a computer to implement any of the above methods.

[0127] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc (CD-ROM), a magnetic tape, a floppy disk, an optical data storage device, or the like.

[0128] In an exemplary embodiment, a computer program product or computer program is also provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform any of the above methods.

[0129] It should be noted that the terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects and to define a specific order or precedence. Where appropriate, the order used for similar objects may be interchanged, such that the embodiments of the present application described herein can be implemented in an order other than the order shown or described.

[0130] Those skilled in the art will appreciate that the present invention may be implemented as a system, method, or computer program product. Therefore, the present disclosure may be implemented in the following forms: entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, the present invention may be implemented in the form of a computer program product embodied in one or more computer-readable media containing computer-readable program code.

[0131] Any combination of one or more computer-readable media can be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or device.

[0132] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A transfer learning method combining multi-dimensional similarity measurement with adaptive weight adjustment, characterized in that: include: For any mechanical equipment to be inspected, retrieve at least two historical maintenance records of the mechanical equipment to be inspected from different maintenance institutions, integrate and clean all historical maintenance records, and generate maintenance data for the mechanical equipment to be inspected; Performing feature extraction on the maintenance data and sample data of different damage categories using a pre-trained model to obtain a first feature vector corresponding to the maintenance data and a second feature vector corresponding to the sample data of different damage categories; calculating the similarity between the maintenance data and the sample data of different damage categories based on the first feature vector and the second feature vector; and using the service life of the damage category corresponding to the sample data of the damage category with the highest similarity as the detection result of the mechanical equipment to be detected; The weight of the pre-trained model is determined by adaptive adjustment.

2. The transfer learning method combining multi-dimensional similarity measurement with adaptive weight adjustment according to claim 1, characterized in that: The cleaning process is specifically as follows: Abnormal data is eliminated, and data in different formats under the same project are processed uniformly.

3. The transfer learning method combining multi-dimensional similarity measurement with adaptive weight adjustment according to claim 1, characterized in that: The similarity includes: feature similarity, label similarity and time series similarity.

4. The transfer learning method combining multi-dimensional similarity measurement with adaptive weight adjustment according to claim 3, characterized in that: The calculation process of the feature similarity is specifically as follows: Among them, S feat (x i ,y j ) represents feature similarity, f(·) represents the feature vector extracted by the convolutional neural network, x i represents the first eigenvector, y j represents the second eigenvector.

5. The transfer learning method combining multi-dimensional similarity measurement with adaptive weight adjustment according to claim 3, characterized in that: The specific calculation process of the label similarity is as follows: Among them, S label (x i ,y j ) represents the label similarity, g(·) represents the function that maps the label text to the semantic space, l i Indicates the label corresponding to the maintenance data, l j Indicates the labels corresponding to sample data of different damage categories.

6. The transfer learning method combining multi-dimensional similarity measurement with adaptive weight adjustment according to claim 3, characterized in that: The specific calculation process of the time series similarity is as follows: Among them, S time (x i ,y j ) represents the time series similarity, DTW(T i ,T j ) represents the time series T i and T j The dynamic time warping distance between i Represents the time series corresponding to the maintenance data, T j Represents the time series corresponding to sample data of different damage categories.

7. A transfer learning system with multi-dimensional similarity measurement combined with adaptive weight adjustment, characterized in that: include: The processing module is used to: for any mechanical equipment to be detected, retrieve at least two historical maintenance records of the mechanical equipment to be detected from different maintenance institutions, and integrate and clean all historical maintenance records to generate maintenance data of the mechanical equipment to be detected; The detection module is configured to: perform feature extraction on the maintenance data and sample data of different damage categories using a pre-trained model to obtain a first feature vector corresponding to the maintenance data and a second feature vector corresponding to the sample data of different damage categories; calculate the similarity between the maintenance data and the sample data of different damage categories based on the first feature vector and the second feature vector; and use the service life of the damage category corresponding to the sample data of the damage category with the highest similarity as the detection result of the mechanical equipment to be detected; The weight of the pre-trained model is determined by adaptive adjustment.

8. The transfer learning system combining multi-dimensional similarity measurement with adaptive weight adjustment according to claim 7, characterized in that: The cleaning process is specifically as follows: Abnormal data is eliminated, and data in different formats under the same project are processed uniformly.

9. An electronic device, characterized in that: The electronic device includes a processor coupled to a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor so that the electronic device implements the method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to enable a computer to implement the method according to any one of claims 1 to 6.

Images