Metalearning-driven crossed roller bearing state evaluation method and device and medium

Abstract

The invention discloses a meta-learning-driven crossed roller bearing state evaluation method and device and a medium. The method aims at improving the service state evaluation performance of the crossed roller bearing under the condition of fault data volume scarcity. According to the method, the strong feature mining capability of the deep neural network and the thought of'learning 'of meta learning are utilized, and the fault discrimination capability learned on the rolling bearing is migrated to the evaluation of the service state of the crossed roller bearing. In the meta-training stage, multiple sets of state evaluation tasks are randomly sampled on a rolling bearing data set, a powerful service state evaluation meta-model is obtained through training, and the constructed model comprises a feature extractor and a service state recognizer. In the meta test stage, the service state recognition model device is finely adjusted by using the model-independent meta learning algorithm, the model parameters are updated, the training of the service state evaluation model suitable for the crossed roller bearing is quickly completed, and the accuracy of state evaluation is improved. And a method is provided for evaluating the service state of the crossed roller bearing in a small sample industrial scene.

Description

Technical Field

[0001] The present invention relates to the field of fault prediction and health management, and in particular to a meta-learning driven crossed roller bearing state assessment method, device and medium. Background Art

[0002] Patent document CN 115600134 A proposes a bearing transfer learning fault diagnosis method based on domain dynamic adversarial adaptation, designs dynamic adversarial factors, and constructs a domain adversarial fault diagnosis model by adversarially matching source domain data and target domain data from different working conditions, so as to solve the fault diagnosis problem of bearings when target domain data is insufficient under different working conditions. The model mainly includes four modules: feature extraction module: using a multi-layer convolutional neural network to extract deep features of source domain data and target domain data; class discriminator module: taking source domain data as samples, taking implicit features extracted by multi-layer convolutional neural networks as deep features, calculating the probability that source domain data belongs to a certain fault, and judging which fault category the data belongs to; global domain discriminator: taking source domain data and target domain data as samples, taking fault implicit features as features, judging whether the input data comes from the source domain or the target domain; subdomain discriminator module: inputting sample features into the class discriminator, calculating the probability that the corresponding sample belongs to a certain fault category, and performing weighted calculation with samples not input into the class discriminator to obtain subdomain features, and then building several neural networks to realize category judgment of subdomain features.

[0003] The above method mainly has the following shortcomings:

[0004] 1. The above-mentioned fault diagnosis model based on transfer learning is suitable for the transfer diagnosis of rolling bearings under different working conditions, using a large amount of source domain data from different working conditions to improve the diagnosis results of a small amount of target domain data. However, for crossed roller bearings, due to their different internal structures, the data distribution is quite different from that of rolling bearings. The ability to adapt to the data distribution across devices using transfer learning alone is insufficient, which limits the diagnosis results.

[0005] 2. The above fault diagnosis model has a complex structure, including feature extractor, class discriminator, global pre-discriminator and subdomain discriminator modules. The algorithm complexity is high, and it is necessary to further develop a simple and fast fault diagnosis model suitable for industrial scenarios. Summary of the invention

[0006] In order to solve at least one of the problems existing in the prior art, the present invention proposes a crossed roller bearing condition assessment method, device and medium driven by meta-learning, and utilizes advanced artificial intelligence technologies such as meta-learning to transfer the condition assessment capability learned on ordinary rolling bearings to crossed roller bearings, thereby realizing the service condition assessment of crossed roller bearings.

[0007] To achieve the purpose of the present invention, the present invention provides a meta-learning driven crossed roller bearing state assessment method, device and medium, the method comprising:

[0008] The following steps are involved:

[0009] Obtaining service status data of ordinary rolling bearings and crossed roller bearings, including acceleration vibration signals and corresponding status labels, and using the service status data of ordinary rolling bearings as meta-training data and the service status data of crossed roller bearings as meta-testing data;

[0010] Based on the service status data of the ordinary rolling bearing, the frequency domain data of the ordinary rolling bearing is obtained, and based on the frequency domain data, a meta-training task is obtained, and the meta-training task is input into a service status evaluation meta-model for meta-training, wherein the service status evaluation meta-model includes a feature extraction module and a service status identifier, the feature extractor is used to extract deep features, and the service status identifier is used to determine the fault type;

[0011] The service status data of the crossed roller bearings are subjected to Fourier transformation and input into the constructed service status assessment model. The model parameters are transferred using a model-independent meta-learning algorithm to obtain a service status assessment model suitable for crossed roller bearings, wherein the network structures in the service status assessment model and the service status assessment meta-model are the same.

[0012] Training the service status assessment model to obtain a trained service status assessment model;

[0013] The test data of the crossed roller bearing is input into the trained service status assessment model to obtain the service status assessment result.

[0014] Furthermore, the service status assessment model is trained using an Adam optimizer to obtain a trained service status assessment model.

[0015] Furthermore, the method further comprises:

[0016] The obtained service status assessment results are evaluated using accuracy indicators.

[0017] Furthermore, the data of common rolling bearings are Crossed roller bearing data x (i) (i=1,2,…,n r ) is a sample of rolling bearing with service status label, y (i) (i=1,2,…,n r ) is the label of the rolling bearing sample, x (i) (i=1,2,…,n c) is a sample of a cross roller bearing with a service status label, y (i) (i=1,2,…,n c ) is the label of the cross roller bearing sample; n r and n c are the sample sizes of rolling bearings and crossed roller bearings, respectively, where n c <<n r ; Service status category in represents the number of cross roller bearing sample label categories, Indicates the number of rolling bearing sample label types.

[0018] Furthermore, for ordinary rolling bearing data Perform Fourier transform to convert the service status data of ordinary rolling bearings from time domain to frequency domain data. The specific expression is:

[0019]

[0020] Where ω is frequency and t is time; f(t) is a function in the time domain, representing the original signal; F(ω) represents the component of the input function f(t) at frequency ω; e -iωt is a complex exponential function, representing a sine wave with a frequency of ω, i is an imaginary unit; dt represents the integral element, which is used to solve the integral of a continuous function in the time domain. ∞ is positive infinity, -∞ is negative infinity, is the integral symbol, which means integrating the function over the entire real number axis.

[0021] Furthermore, for the transformed samples and their labels, a small number of samples are randomly selected multiple times as multi-group meta-training tasks and input into the constructed service status assessment meta-model.

[0022] Furthermore, in the service status assessment meta-model, the feature extractor includes an input layer, a convolution layer, a pooling layer and an activation function layer. The convolution layer is used to extract local features of the input data and generate a feature map. The pooling layer is used to reduce the spatial size of the feature map. The activation function layer is used to enable the model to learn complex function mapping relationships. The service status identifier includes a fully connected layer and a classification layer. The fully connected layer is used to integrate and map the extracted features to category probabilities. The classification layer is used to normalize the category probabilities and output the probabilities of multiple service status labels.

[0023] Furthermore, the feature extractor includes an input layer, two convolutional layers, two pooling layers, and two RELU activation function layers; the service status identifier includes two fully connected layers and a Softmax layer. Among them, the Softmax layer is a multi-label classifier that outputs the probabilities of multiple service status labels, and the label with the highest probability is the sample label.

[0024] The loss function of the entire model is:

[0025]

[0026] Among them, y i,j and points Respectively represent the true label value of the jth category of the ith sample and the probability value of the model output. If the ith sample belongs to the jth category, then y i,j =1, otherwise y i,j =0; n represents the number of rolling bearing samples, r represents the number of rolling bearing categories; ∑ is the summation symbol, and log() represents the natural logarithm function. If the model output value is different from the true label value, the cross entropy loss will increase. Therefore, by minimizing the cross entropy loss, the model can be better fitted and trained to improve the classification accuracy.

[0027] Furthermore, a small amount of crossed roller bearing data is Fourier transformed and input into the service status assessment model. A model-independent meta-learning training algorithm is adopted, the feature extractor in the service status meta-model is fixed, and the parameters in the service status identifier are fine-tuned and updated using the crossed roller bearing data to obtain a service status assessment model suitable for crossed roller bearings.

[0028] The service status assessment model optimizer is an Adam optimizer. In the meta-training stage, the learning rate of the Adam optimizer is 0.001 and the number of iterations is 2000; in the meta-testing stage, the learning rate of the Adam optimizer is 0.001 and the number of iterations is 300.

[0029] In a second aspect, the present invention provides a meta-learning driven cross roller bearing service status assessment system, comprising the following modules:

[0030] A data acquisition module is used to acquire service status data of ordinary rolling bearings and crossed roller bearings, including acceleration vibration signals and corresponding status labels, and use the service status data of ordinary rolling bearings as meta-training data and the service status data of crossed roller bearings as meta-test data;

[0031] A meta-training module, used to obtain frequency domain data of ordinary rolling bearings based on the service status data of ordinary rolling bearings, and to obtain meta-training tasks based on the frequency domain data, and to input the meta-training tasks into a service status evaluation meta-model for meta-training, wherein the service status evaluation meta-model includes a feature extraction module and a service status identifier, the feature extractor is used to extract deep features, and the service status identifier is used to determine the fault type;

[0032] A migration module is used to perform Fourier transformation on the service status data of the crossed roller bearing, input the data into the constructed service status assessment model, and use the model-independent meta-learning algorithm to migrate the model parameters to obtain a service status assessment model suitable for the crossed roller bearing, wherein the network structure in the service status assessment model and the service status assessment meta-model is the same;

[0033] An evaluation model training module is used to train the service status evaluation model to obtain a trained service status evaluation model;

[0034] The evaluation module is used to input the test data of the cross roller bearing into the trained service status evaluation model to obtain the service status evaluation result.

[0035] In a third aspect, the present invention provides a meta-learning driven cross roller bearing service status assessment device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method when executing the computer program.

[0036] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements the steps of the method when executed by a processor.

[0037] Compared with the prior art, the present invention has at least the following beneficial effects:

[0038] The method of the present invention proposes a method for evaluating the service status of a crossed roller bearing in a small sample industrial scenario, which has high accuracy. With the help of meta-learning technology, the ability to diagnose the service status of rolling bearings is transferred to the service status evaluation problem of crossed roller bearings through a model-independent meta-learning strategy, which effectively improves the service status evaluation effect when the amount of data is insufficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is an algorithm flow chart of a meta-learning driven crossed roller bearing state assessment method provided in an embodiment of the present invention;

[0040] Figure 2 is a schematic diagram of vibration data in the time domain and the frequency domain after Fourier transformation according to an embodiment of the present invention;

[0041] Figure 3 It is a schematic diagram of the structure of the service status assessment metamodel provided by an embodiment of the present invention;

[0042] Figure 4 is a confusion matrix diagram of prediction results of the meta-learning driven crossed roller bearing service status assessment method provided by an embodiment of the present invention;

[0043] Figure 5This is a comparison chart of the accuracy of the meta-learning driven crossed roller bearing service status assessment method provided by an embodiment of the present invention and other methods.

[0044] Figure 6 A schematic diagram of the composition of a meta-learning driven crossed roller bearing service status assessment device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0046] Example 1

[0047] The meta-learning driven crossed roller bearing state assessment method provided in this embodiment uses meta-learning technology to first divide ordinary rolling bearing data and crossed roller bearing data into a meta-training data set and a meta-testing data set; secondly, the meta-training data set is used to train a service state assessment meta-model; then a model-independent meta-learning algorithm is built, and a service state identifier is updated using the meta-testing data set through a fixed feature extractor to obtain a service state assessment model suitable for crossed roller bearings. The service state assessment model is used to implement service state assessment of crossed roller bearings in small sample industrial scenarios.

[0048] Specifically, Figure 1 As shown in FIG. 1 , the meta-learning driven cross roller bearing condition assessment method includes the following steps:

[0049] Step 1: Obtain the service status data of ordinary rolling bearings and crossed roller bearings, including acceleration vibration signals and corresponding service degree labels; use the service status data of ordinary rolling bearings as meta-training data, a small amount of labeled service status data of crossed roller bearings as meta-test data, and unlabeled crossed roller bearing data as test data.

[0050] In this step, the service status data of ordinary rolling bearings is The service status data of crossed roller bearings is x (i) (i=1,2,…,n r ) is a sample of ordinary rolling bearing with service status label, y (i) (i=1,2,…,n r ) is the label of the common rolling bearing sample, x (i) (i=1,2,…,n c ) is a sample of a cross roller bearing with a service status label, y (i) (i=1,2,…,n c ) is the label of the cross roller bearing sample; n r and n care the sample sizes of ordinary rolling bearings and crossed roller bearings, respectively, where n c < <n r ; Service status category in represents the number of cross roller bearing sample label categories, Indicates the number of rolling bearing sample label types.

[0051] Step 2: Perform Fourier transform on the original signal of the ordinary rolling bearing, i.e., the service status data, to obtain the frequency domain data of the ordinary rolling bearing, and randomly extract multiple groups of small amounts of data as meta-training tasks, and input them into the service status evaluation meta-model for meta-training, wherein the service status evaluation meta-model includes a feature extractor and a service status identifier.

[0052] In this step, the data of ordinary rolling bearings Perform Fourier transform, the specific expression is:

[0053]

[0054] Where F(ω) represents the component of the input function f(t) at frequency ω; ω is frequency, t is time; f(t) is a function in the time domain, representing the original signal; e -iωt is a complex exponential function, representing a sine wave with a frequency of ω, i is an imaginary unit; dt represents the integral element, which is used to solve the integral of a continuous function in the time domain. ∞ is positive infinity, -∞ is negative infinity, is the integral symbol, which means integrating the function over the entire real number axis.

[0055] For the transformed samples and their labels, a small number of samples are randomly selected multiple times as multi-group training tasks and input into the service status evaluation meta-model. The service status evaluation meta-model includes a feature extractor and a service status identifier. The feature extractor is used to extract deep features and input them into the service status identifier; the service status identifier is used to determine the fault type. The feature extractor includes an input layer, two convolutional layers, two pooling layers and two activation function layers. The convolutional layer extracts the features of the input data through convolution operations and generates feature maps. These feature maps capture the local features of the input data. The pooling layer reduces the spatial size of the feature map by downsampling, reduces the number of parameters and computational complexity, and the activation function layer is a nonlinear activation function, which enables the model to learn complex function mapping relationships. The number of convolutional layers, pooling layers and activation function layers is not limited to two layers, and can be adjusted according to specific applications and task requirements. The function of the activation function layer is the RELU activation function, which is highly efficient and can effectively alleviate the gradient disappearance problem. In other embodiments, other activation functions can also be used; the service status identifier includes two fully connected layers and a Softmax layer (classification layer). The fully connected layer is used to integrate and map the extracted features to category probabilities, and the softmax layer is used to normalize these category probabilities to make the final fault type judgment. Among them, the Softmax layer is a multi-label classifier used to output the probabilities of multiple service status labels, and the label with the highest probability is the sample label. The loss function of the entire service status evaluation meta-model is:

[0056]

[0057] Among them, y i,j and They represent the true label value of the jth category of the ith sample and the probability value of the model output. If the ith sample belongs to the jth category, then y i,j =1, otherwise y i,j =0; n represents the number of rolling bearing samples, r represents the number of rolling bearing categories; Σ is the summation symbol, and log() represents the natural logarithm function. If the model output value is different from the true label value, the cross entropy loss will increase. Therefore, by minimizing the cross entropy loss, the model can be better fitted and trained to improve the classification accuracy.

[0058] Step 3: The small amount of cross-roller bearing data obtained is first subjected to Fourier transform, and then input into the constructed service status assessment model. The model parameters are migrated using the model-independent meta-learning algorithm, the parameters of the feature extractor are fixed, and the parameters of the service status identifier are updated to obtain a service status assessment model suitable for cross-roller bearings.

[0059] In this step, the parameters of the service condition assessment metamodel are transferred to obtain a service condition assessment model suitable for the service condition assessment of the crossed roller bearing. The network structure of the service condition assessment metamodel and the service condition assessment model is the same, but the model parameters are different.

[0060] In this step, a small amount of crossed roller bearing data is Fourier transformed and input into the constructed service status assessment model. A model-independent meta-learning training algorithm is used to fix the feature extractor in the service status meta-model and use the crossed roller bearing data to fine-tune and update the parameters of the service status identifier to obtain a service status assessment model suitable for crossed roller bearings.

[0061] Step 4: Use the optimizer to train the service condition assessment model for the crossed roller bearing in step 3 to obtain a trained service condition assessment model.

[0062] In this embodiment, the optimizer is an Adam optimizer. In the meta-training stage, the learning rate of the Adam optimizer is 0.001 and the number of iterations is 2000; in the meta-testing stage, the learning rate of the Adam optimizer is 0.001 and the number of iterations is 300.

[0063] Step 5: Input the unlabeled crossed roller bearing data into the trained service status assessment model to obtain the service status assessment results.

[0064] In this step, the unlabeled crossed roller bearing data is input into the trained service status assessment model to obtain the service status assessment results.

[0065] Finally, the accuracy is used to quantify the evaluation results. The accuracy calculation formula is: Where N is the number of all test samples, N corr Evaluate the correct number of samples for the state.

[0066] The above method of the present invention takes advantage of meta-learning, utilizes the state assessment ability learned in the ordinary rolling bearing service state assessment task, and combines the powerful feature extraction ability of convolutional neural network to realize the assessment of the service state of crossed roller bearings in small sample industrial scenarios. With the help of model-independent meta-learning algorithm, the training speed of the model can be quickly improved, and the accuracy of the service state assessment results can be improved.

[0067] In order to verify the effectiveness of the method proposed in the present invention, the embodiment of the present invention adopts Miller BBM-PAK data acquisition and analysis system, installs vibration sensors on a test bench close to the cross roller bearing, and collects data under four service conditions, where the sampling frequency fs is 51200 Hz and the sampling time T is 200 s.

[0068] 1. This embodiment is based on the definition of bearing fatigue spalling in GB / T 24607-2009: "Spalling depth ≥ 0.05mm; spalling area: roller bearing parts ≥ 1.0mm 2 ", rectangular pits with different fault depths are artificially implanted on the rolling surface of the roller of the cross roller bearing to simulate different stages of service status, as shown in Table 1.

[0069] Table 1 Crossed roller bearings in different service stages

[0070]

[0071] 2. The embodiment of the present invention designs 5 groups of small sample experimental scenarios, and the number of crossed roller bearings with labels is 10, 20, 30, 40 and 50 respectively. The experimental settings used in the present invention are shown in Table 2.

[0072] Table 2 Small sample scenario experimental settings

[0073]

[0074] 2. Perform Fourier transform on the data. Compared with time domain data, frequency domain data can provide detailed spectrum information, which is convenient for understanding signals and can improve the accuracy of service status identification. Taking the cross roller bearing as an example, the changes of signals in time domain and frequency domain under different service conditions are as follows: Figure 2 shown.

[0075] 3. Randomly extract multiple groups of small amounts of data from the rolling bearing data as meta-training tasks, and input them into the service status evaluation meta-model for meta-training, wherein the service status evaluation meta-model includes a feature extractor and a service status identifier; in this stage, the service status evaluation meta-model is trained by the Adam optimizer, wherein the learning rate of the Adam optimizer is 0.001 and the number of iterations is 2000. The network structure of the constructed service status evaluation meta-model is as follows: Figure 3 shown.

[0076] 4. Perform Fourier transform on a small amount of crossed roller bearing data in the training data and input it into the constructed service status assessment model. Use a model-independent meta-learning training algorithm, fix the feature extractor in the service status meta-model, and use the crossed roller bearing data to fine-tune and update the parameters in the service status identifier to obtain a service status assessment model suitable for crossed roller bearings.

[0077] 5. Input the test data into the above-trained cross roller bearing service status assessment model. The prediction results are shown in Table 3, and the confusion matrix is ​​as follows: Figure 4As shown, it can be seen that with the increase in the number of crossed roller bearings, the accuracy of the service status assessment results gradually improves, and when only 10 crossed roller bearing samples are involved in the training, the accuracy can reach more than 85%, which verifies the effectiveness of the method proposed in the present invention.

[0078] Table 3 Crossed roller bearing service status evaluation results

[0079]

[0080] The comparison results of the method proposed in the present invention and directly training and predicting the data using only a convolutional neural network are shown in Figure 5. It can be seen from the figure that the prediction result of the method of the present invention is better than the prediction result obtained when only a convolutional neural network is used.

[0081] It can be seen that the meta-learning driven crossed roller bearing service status assessment method provided by the present invention combines the powerful learning ability of meta-learning and the powerful feature extraction ability of deep convolutional neural networks, which can improve the accuracy of service status assessment in small sample industrial scenarios.

[0082] Example 2

[0083] The meta-learning driven cross roller bearing service status assessment system provided in this embodiment is used to implement the method described in Example 1. The system includes the following modules:

[0084] A data acquisition module is used to acquire service status data of ordinary rolling bearings and crossed roller bearings, including acceleration vibration signals and corresponding status labels, and use the service status data of ordinary rolling bearings as meta-training data and the service status data of crossed roller bearings as meta-test data;

[0085] A meta-training module, used to obtain frequency domain data of ordinary rolling bearings based on the service status data of ordinary rolling bearings, and to obtain meta-training tasks based on the frequency domain data, and to input the meta-training tasks into a service status evaluation meta-model for meta-training, wherein the service status evaluation meta-model includes a feature extraction module and a service status identifier, the feature extractor is used to extract deep features, and the service status identifier is used to determine the fault type;

[0086] A migration module is used to perform Fourier transformation on the service status data of the crossed roller bearing, input the data into the constructed service status assessment model, and use the model-independent meta-learning algorithm to migrate the model parameters to obtain a service status assessment model suitable for the crossed roller bearing, wherein the network structure in the service status assessment model and the service status assessment meta-model is the same;

[0087] An evaluation model training module is used to train the service status evaluation model to obtain a trained service status evaluation model;

[0088] The evaluation module is used to input the test data of the cross roller bearing into the trained service status evaluation model to obtain the service status evaluation result.

[0089] Furthermore, it also includes an accuracy evaluation module for evaluating the obtained service status evaluation results using accuracy indicators to evaluate the recognition results.

[0090] Example 3

[0091] like Figure 6 As shown, the meta-learning driven cross roller bearing service status assessment device provided in this embodiment includes a processor 60, a memory 61, and a computer program 62 stored in the memory 61 and executable on the processor 60, such as a meta-learning driven cross roller bearing service status assessment program. The processor 60 executes the computer program to implement the steps of the above embodiment 1.

[0092] Example 4

[0093] This embodiment provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the method described in Embodiment 1 are implemented.

[0094] For the meta-learning driven crossed roller bearing service status assessment system, device and medium disclosed in the embodiment, since they correspond to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0095] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable ordinary technicians in the field to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made based on the essence of the content of the present invention should be included in the protection scope of the present invention.

Claims

1. Meta-learning driven crossed roller bearing condition assessment method, characterized in that: The following steps are involved: Obtaining service status data of ordinary rolling bearings and crossed roller bearings, including acceleration vibration signals and corresponding status labels, and using the service status data of ordinary rolling bearings as meta-training data and the service status data of crossed roller bearings as meta-testing data; Based on the service status data of the ordinary rolling bearing, the frequency domain data of the ordinary rolling bearing is obtained, and based on the frequency domain data, a meta-training task is obtained, and the meta-training task is input into a service status evaluation meta-model for meta-training, wherein the service status evaluation meta-model includes a feature extraction module and a service status identifier, the feature extractor is used to extract deep features, and the service status identifier is used to determine the fault type; The service status data of the crossed roller bearings are subjected to Fourier transformation and input into the constructed service status assessment model. The model parameters are transferred using a model-independent meta-learning algorithm to obtain a service status assessment model suitable for crossed roller bearings, wherein the network structures in the service status assessment model and the service status assessment meta-model are the same. Training the service status assessment model to obtain a trained service status assessment model; The test data of the crossed roller bearing is input into the trained service status assessment model to obtain the service status assessment result.

2. The meta-learning driven crossed roller bearing state assessment method according to claim 1, characterized in that: The method further comprises: The obtained service status assessment results are evaluated using accuracy indicators.

3. The meta-learning driven crossed roller bearing state assessment method according to claim 1, characterized in that: The Fourier transform tool is used to convert the collected service status data of ordinary rolling bearings from the time domain to the frequency domain data, and the expression of the Fourier transform is: Where ω is frequency and t is time; f(t) is a function in the time domain, representing the original signal; F(ω) represents the component of the input function f(t) at frequency ω; e -iωt is a complex exponential function, representing a sine wave with a frequency of ω, i is an imaginary unit; dt represents the integral element, which is used to solve the integral of a continuous function in the time domain, ∞ is positive infinity, -∞ is negative infinity, is the integral symbol, which means integrating the function over the entire real number axis.

4. The meta-learning driven crossed roller bearing state assessment method according to claim 1, characterized in that: In the service status assessment meta-model, the feature extractor includes an input layer, a convolution layer, a pooling layer and an activation function layer. The convolution layer is used to extract local features of the input data and generate a feature map. The pooling layer is used to reduce the spatial size of the feature map. The activation function layer is used to enable the model to learn complex function mapping relationships. The service status identifier includes a fully connected layer and a classification layer. The fully connected layer is used to integrate and map the extracted features to category probabilities. The classification layer is used to normalize the category probabilities and output the probabilities of multiple service status labels.

5. The meta-learning driven crossed roller bearing state assessment method according to claim 1, characterized in that: The loss function of the service status assessment meta-model is the cross entropy loss, and its specific expression is: Among them, y i,j and points Respectively represent the true label value of the jth category of the ith sample and the probability value of the model output. If the ith sample belongs to the jth category, then y i,j =1, otherwise y i,j =0; n represents the number of samples of rolling bearings, r represents the number of categories of rolling bearings; ∑ is the summation symbol, and log() represents the natural logarithm function.

6. The meta-learning driven crossed roller bearing state assessment method according to claim 1, characterized in that: The model-independent meta-learning algorithm fixes the feature extractor in the service status meta-model, uses the crossed roller bearing data to fine-tune and update the parameters in the service status identifier, and obtains a service status evaluation model suitable for crossed roller bearings.

7. Meta-learning driven cross roller bearing service status assessment system, characterized by: For implementing the method described in any one of claims 1 to 6, the system comprises the following modules: A data acquisition module is used to acquire service status data of ordinary rolling bearings and crossed roller bearings, including acceleration vibration signals and corresponding status labels, and use the service status data of ordinary rolling bearings as meta-training data and the service status data of crossed roller bearings as meta-test data; A meta-training module, used to obtain frequency domain data of ordinary rolling bearings based on the service status data of ordinary rolling bearings, and to obtain meta-training tasks based on the frequency domain data, and to input the meta-training tasks into a service status evaluation meta-model for meta-training, wherein the service status evaluation meta-model includes a feature extraction module and a service status identifier, the feature extractor is used to extract deep features, and the service status identifier is used to determine the fault type; A migration module is used to perform Fourier transformation on the service status data of the crossed roller bearing, input the data into the constructed service status assessment model, and use the model-independent meta-learning algorithm to migrate the model parameters to obtain a service status assessment model suitable for the crossed roller bearing, wherein the network structure in the service status assessment model and the service status assessment meta-model is the same; An evaluation model training module is used to train the service status evaluation model to obtain a trained service status evaluation model; The evaluation module is used to input the test data of the cross roller bearing into the trained service status evaluation model to obtain the service status evaluation result.

8. The meta-learning driven cross roller bearing service status assessment system according to claim 7, characterized in that: It also includes an accuracy evaluation module, which is used to evaluate the obtained service status evaluation results using accuracy indicators to identify the results.

9. A device for evaluating the service status of a crossed roller bearing based on meta-learning, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method according to claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images