Mechanical equipment life prediction method based on statistical alignment element gating recursive unit

Abstract

The invention relates to a mechanical equipment life prediction method based on a statistical alignment element gating recursive unit. The method comprises the following steps: acquiring full-life vibration data of the rotary mechanical equipment as training data; designing a statistical alignment measure to calibrate the difference between the training data and the target data distribution; constructing a mechanical equipment life prediction model, inputting the calibrated training data into the model, and carrying out subtask learning; and integrating knowledge learned by the sub-tasks, performing cross-sub-task learning, performing optimization based on a secondary gradient, updating a meta prediction factor, and predicting a future degradation trend of the mechanical equipment based on the meta prediction factor. Effective knowledge migration among different fields is realized, priori knowledge in historical big data is applied to life prediction of unknown fields under the condition of limited data, the performance of cross-domain prediction is improved, and effective life prediction can be performed under the condition of insufficient samples.

Description

Technical Field

[0001] This application relates to the technical fields of artificial intelligence and mechanical engineering, and particularly to a method for predicting the lifespan of mechanical equipment based on a statistically aligned meta gated recurrent unit. Background Art

[0002] The big data era has given a significant boost to modern industrial scenarios, including advanced prediction and health management technologies. At the same time, artificial intelligence technologies provide powerful decision-making capabilities, promoting machine prediction and health management. In recent years, mechanical degradation prediction and remaining useful life prediction based on artificial intelligence have attracted increasing attention.

[0003] Generally speaking, most mechanical degradation predictions rely on data-driven methods, and data-driven technologies based on artificial intelligence have inspired a large number of studies. In particular, the powerful feature extraction capabilities and natural extrapolation structures of two popular deep learning architectures, convolutional neural networks and recurrent neural networks, enable the established intelligent algorithms to handle mechanical degradation predictions even under non-stationary vibration signals. However, these methods are directly based on the collected signals and usually assume that these signals are in the same distribution. In actual industrial applications, due to domain drift phenomena caused by changes in working conditions, environmental noise, quality differences between workpieces, etc., this ideal condition cannot be met.

[0004] In the current literature on cross-domain degradation prediction, deep learning and transfer learning methods have been successfully integrated. The main drawback of this method is the assumption that there is sufficient target domain data available for extracting domain-invariant knowledge. In actual industrial scenarios, it may be expensive and impossible to collect sufficient and representative state data for prediction. Even under the same training set and test set distributions, limited data-driven prediction and health management algorithms may obtain poor fault diagnosis or lifespan prediction results. This remains a fundamental and unsolved problem in current mechanical prediction and health management. In addition, when domain migration phenomena occur, it is still necessary to deeply explore the establishment of a prediction algorithm that is sensitive and perfect for a small number of conditional samples.

[0005] Therefore, the existing data-driven prediction methods based on artificial intelligence face two major dilemmas: 1) In actual applications, the limited measured monitoring data weakens the generalization of the constructed prediction algorithm to a certain extent, and this situation will be further exacerbated considering the domain migration phenomenon; 2) Although big data technologies provide many possibilities for processing large amounts of data in mechanical prediction and health management; how to efficiently integrate big data, construct a sensitive prediction model, and easily adapt to unknown tasks driven by limited data remains an unsolved and undiscovered problem. Summary of the Invention

[0006] Based on this, it is necessary to provide a mechanical equipment life prediction method based on a statistically aligned gated recurrent unit for the above technical problems.

[0007] In a first aspect, the present application provides a mechanical equipment life prediction method based on a statistically aligned gated recurrent unit. The method includes:

[0008] Obtain the full-life vibration data of the rotating mechanical equipment as training data;

[0009] Design a statistical alignment metric to calibrate the difference between the training data and the target data distribution;

[0010] Construct a mechanical equipment life prediction model, input the calibrated training data into the model, and perform subtask learning;

[0011] Integrate the knowledge learned from the subtasks, then perform cross-subtask learning, and perform optimization based on the second-order gradient to update the meta-predictor, and predict the future degradation trend of the mechanical equipment based on the meta-predictor.

[0012] Optionally, in an embodiment of the present application, the core formula of the statistical alignment metric is:

[0013]

[0014] where p is the intrinsic dimension, is the discrete estimate value, U(ε), U(ε 1 ), U(ε 2 ) are the stack numbers, and ε, ε 1 , ε 2 are the conditional variables.

[0015] Optionally, in an embodiment of the present application, the design of the statistical alignment metric to calibrate the difference between the training data and the target data distribution includes:

[0016] Construct a statistic to calculate the distance between the training data and the target data;

[0017] Construct a loss function including the statistical alignment metric, and minimize the loss function to solve the mapping matrix.

[0018] Optionally, in an embodiment of the present application, the construction of the statistic to calculate the distance between the training data and the target data includes:

[0019] Use the maximum mean discrepancy metric to quantify the data difference as:

[0020]

[0021] where n s and n tis the number of samples from the training data and the target data, is the weight of the i-th training sample, is the weight of the i-th target sample, is the i-th training sample, is the i-th target sample, denotes the reproducing kernel Hilbert space, which is used to measure the weighted distance between the training data and the kernel target data, and φ(·) denotes the kernel function.

[0022] Optionally, in an embodiment of the present application, when constructing the mechanical equipment life prediction model, inputting the calibrated training data into the model, the subtask learning includes:

[0023] Using an encoder to extract the time series features of the support dataset, and using a decoder to predict the degradation trend according to the time series features.

[0024] Optionally, in an embodiment of the present application, the optimization formula for the second-order gradient is:

[0025]

[0026] where, denotes the parameter Φ obtained by subtask learning i is the gradient of the test set loss, denotes the task parameter Φ i is the gradient of the global parameter θ, which reflects the indirect influence of the support set learning on the test set optimization, and α is the learning rate, is the test data set, P(T) is the task distribution in multi-task learning, and T i is the label related to this task.

[0027] In a second aspect, the present application also provides a mechanical equipment life prediction device based on a statistically aligned meta gated recurrent unit. The device includes:

[0028] A data acquisition module, configured to acquire the full-life vibration data of the rotating mechanical equipment as training data;

[0029] A difference calibration module, configured to design a statistical alignment metric to calibrate the difference between the distributions of the training data and the target data;

[0030] A learning module, configured to construct a mechanical equipment life prediction model, input the calibrated training data into the model, and perform subtask learning;

[0031] A mechanical equipment life prediction module, configured to integrate the knowledge learned by the subtasks, then perform cross-subtask learning, and perform optimization based on the second-order gradient to update the meta-predictor, and predict the future degradation trend of the mechanical equipment based on the meta-predictor.

[0032] In a third aspect, the present application also provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program, and the processor executes the steps of the methods described in the above respective embodiments.

[0033] In a fourth aspect, the present application also provides a computer-readable storage medium. On the computer-readable storage medium, there is stored a computer program, and when the computer program is executed by a processor, the steps of the methods described in the above respective embodiments are implemented.

[0034] For the above mechanical equipment life prediction method based on statistical alignment meta gated recurrent unit, first, obtain the full-life vibration data of the rotating mechanical equipment as training data; then, design a statistical alignment metric to calibrate the difference between the training data and the target data distribution; then, construct a mechanical equipment life prediction model, input the calibrated training data into the model for subtask learning; finally, integrate the knowledge learned from the subtasks, then perform cross-subtask learning, and perform optimization based on the second-order gradient to update the meta-predictor, and predict the future degradation trend of the mechanical equipment based on the meta-predictor. That is to say, by introducing meta-learning and domain adaptation technologies, as well as the statistical alignment meta gated recurrent unit algorithm, effective knowledge transfer between different domains is achieved. In the case of limited data, the prior knowledge in historical big data is applied to the life prediction of unknown domains, improving the performance of cross-domain prediction and enabling effective life prediction in the case of insufficient samples. Description of the Drawings

[0035] Figure 1 It is an application environment diagram of the mechanical equipment life prediction method based on statistical alignment meta gated recurrent unit in an embodiment;

[0036] Figure 2 It is a flowchart of the mechanical equipment life prediction method based on statistical alignment meta gated recurrent unit in an embodiment;

[0037] Figure 3 It is a schematic diagram of the original vibration signal in an embodiment;

[0038] Figure 4 It is a schematic diagram of the prediction results of the degradation trends of four bearings with different prediction times and data sizes in an embodiment;

[0039] Figure 5 It is a schematic diagram of the root mean square error results of the degradation trend prediction with different prediction times and data samples in an embodiment;

[0040] Figure 6Schematic diagram of the comparison results of the root mean square error of life prediction for the prediction data set using the present method and the existing life prediction method in an embodiment;

[0041] Figure 7 Schematic diagram of the life prediction results in an industrial environment in an embodiment;

[0042] Figure 8 Schematic diagram of the life prediction error in an industrial environment in an embodiment;

[0043] Figure 9 Schematic diagram of the comparison results of the root mean square error of life prediction for the actual industrial data set using the present method and the existing life prediction method in an embodiment;

[0044] Figure 10 Structure block diagram of a mechanical equipment life prediction device based on a statistically aligned meta gated recurrent unit in an embodiment;

[0045] Figure 11 Internal structure diagram of a computer device in an embodiment. Detailed implementation manners

[0046] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0047] The mechanical equipment life prediction method based on a statistically aligned meta gated recurrent unit provided by an embodiment of the present application can be applied to an application environment as shown in Figure 1 . Among them, the terminal communicates with the server through the network. The data storage system can store the data that the server needs to process. The data storage system can be integrated on the server, or can be placed in the cloud or other network servers. Among them, the terminal can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. The portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server can be implemented by an independent server or a server cluster composed of multiple servers.

[0048] In one embodiment, as shown in Figure 2 , a mechanical equipment life prediction method based on a statistically aligned meta gated recurrent unit is provided. Taking the method applied to the server in Figure 1 as an example for description, the method includes the following steps:

[0049] S201: Obtain the full-life vibration data of the rotating mechanical equipment as training data.

[0050] In the embodiment of the present application, first, the full-life vibration data of rotating machinery is collected. Specifically, the full-life data of the bearing is collected through accelerometers installed horizontally and vertically. The sampling frequency is 25.6 kHz, and time-domain and frequency-domain features are extracted from the vibration signals from healthy operation to failure, so as to form a degradation feature matrix. And the historical degradation signals are quantitatively defined as different proportions of the total data, that is, one-tenth (level 1), one-twentieth (level 2), one-thirtieth (level 3), one-fortieth (level 4), and one-fiftieth (level 5) of the total data samples, and stored offline as training data. By defining the limited data as different proportions, it means that during the training process, domain adaptation can be performed by adjusting the proportion of training data and target data, so as to minimize the domain drift caused by data scarcity or imbalance, enabling the model to better integrate cross-domain knowledge, thereby improving the generalization ability of the model.

[0051] S203: Design a statistical alignment metric to calibrate the difference between the distributions of the training data and the target data.

[0052] In the embodiment of the present application, a statistical alignment metric is designed to calibrate the difference between the source data and the target data distributions, thereby solving the domain drift problem existing in cross-domain degradation trend prediction. The designed statistical alignment metric can estimate the appropriate dimension of the subspace and has a good effect in determining the intrinsic dimension of manifold learning.

[0053] Specifically, in an embodiment of the present application, the core formula of the statistical alignment metric is:

[0054]

[0055] where p is the intrinsic dimension, is the discrete estimate value, U(ε), U(ε 1 ), U(ε 2 ) are the number of stacks, and ε, ε 1 , ε 2 are conditional variables.

[0056] In an embodiment of the present application, the intrinsic dimension and its discrete estimate value are used to judge the complexity of the manifold and applied to manifold learning, so as to improve the robustness and generalization ability of the model in cross-domain prediction tasks.

[0057] In an embodiment of the present application, the design of the statistical alignment metric to calibrate the difference between the distributions of the training data and the target data includes:

[0058] Construct a statistic to calculate the distance between the training data and the target data;

[0059] The structure includes a loss function that statistically aligns metrics, and minimizes the loss function to solve for the mapping matrix.

[0060] In one embodiment of the present application, the domain drift problem is solved by aligning the latent statistics between the training data and the target data, including two basic processes: 1) constructing appropriate statistics and calculating the distance between the training data and the target data; 2) transforming unsupervised domain adaptation into an optimization problem, constructing a loss function that includes a statistical alignment metric, and minimizing the loss function to solve for the mapping matrix W.

[0061] In one embodiment of the present application, the constructing of statistics to calculate the distance between the training data and the target data includes:

[0062] Using the maximum mean discrepancy metric to quantify the data discrepancy as:

[0063]

[0064] where n s and n t are the number of samples from the training data and the target data, is the weight of the i-th training sample, is the weight of the i-th target sample, is the i-th training sample, is the i-th target sample, represents the reproducing kernel Hilbert space, which is used to measure the weighted distance between the training data and the target data, and φ(·) represents the kernel function.

[0065] By introducing the second-order statistical covariance matrix to assist in aligning the latent features, while constructing a loss function that includes a statistical alignment metric, which is specifically expressed as follows:

[0066]

[0067] where n is the total number of samples, that is, the sum of the training data samples and the target data samples, W s is the training data sample weight vector, W t is the target data sample weight vector, and respectively represent the covariance matrices of the training data and the target data, and are symmetric positive definite manifolds, and respectively represent the transposes of W s and W t δ s (·) represents a regularization function, which is used to balance the proportion of the loss terms or control the weight of the loss, and is expressed as follows:

[0068]

[0069] Among them, P is the covariance matrix of the data, representing the distribution of the training sample features, Q is the covariance matrix of the target data, representing the distribution of the training sample features, and det is the determinant operation, representing the determinant value of the matrix.

[0070] The optimization problem of minimizing the loss function is actually a non-convex constraint problem, that is

[0071]

[0072] Among them, belongs to the Riemannian manifold, and the formula is solved by the Riemannian gradient descent method, and its update rule is expressed as follows:

[0073]

[0074] Among them, γ is the update step size, τ(·) is the retraction operation and moves along the descent direction to solve, while ensuring that ω (t+1) remains on the Riemannian manifold .

[0075] S205: Build a mechanical equipment life prediction model, input the calibrated training data into the model, and perform subtask learning.

[0076] In the embodiment of the present application, a mechanical equipment life prediction model is built, the calibrated training data is input into the model, and subtask learning is performed. The input of this process is the support set data of subtask Γi i in the training set including time series signals and corresponding degradation labels; the content of learning is to learn subtask-specific parameters Φ i through the support data set,

[0077] Specifically, in an embodiment of the present application, the building of the mechanical equipment life prediction model, inputting the calibrated training data into the model, and performing subtask learning includes:

[0078] Using an encoder to extract the time series features of the support data set, and using a decoder to predict the degradation trend according to the time series features.

[0079] In an embodiment of the present application, in the built mechanical equipment life prediction (encoder-decoder) model, the encoder extracts the time series features of the support data set, the decoder predicts the degradation trend according to the features extracted by the encoder, and uses the mean square error as the loss function to optimize the subtask parameter Φ i , specifically expressed as follows:

[0080]

[0081] Among them, Φ i is the parameter updated by the time backpropagation method, θ is the initial model parameter, and β represents the learning rate. is the support set data of the subtask Γ i in the training set.

[0082] Meanwhile, the mean square error is adopted as the loss function in this step, which is expressed as follows:

[0083]

[0084] Among them, and y *,(t) are the predicted value and the actual value at the time step t, represents the training data set of the subtask Γ i in the training set.

[0085] S207: Integrate the knowledge learned by the subtasks, then perform cross-subtask learning, and perform optimization based on the second-order gradient to update the meta-predictor, and predict the future degradation trend of the mechanical equipment based on the meta-predictor.

[0086] In the embodiment of the present application, based on subtask-level learning, the parameters Φ i learned by the input subtasks and the test data set are used for cross-subtask learning and knowledge integration. The learning content includes: using the test data sets of multiple tasks, combining the knowledge learned from the support data set, and optimizing the global parameter θ. And using second-order gradient optimization, the influence of the support data set on the task-specific parameters is indirectly incorporated into the update of the global parameter, so as to further adjust and update the meta-predictor.

[0087] In an embodiment of the present application, the optimization formula of the second-order gradient is:

[0088]

[0089] Among them, represents the gradient of the parameters Φ i learned by the subtasks with respect to the loss of the test set, represents the gradient of the task parameters Φ i with respect to the global parameter θ, reflecting the indirect influence of the support set learning on the test set optimization, α is the learning rate, is the test data set, P(T) is the task distribution in multi-task learning, and T i is the label related to this task.

[0090] After that, input the query data set And the optimized global parameter θ is used to further adjust the meta-predictor. The learning content includes: further optimizing the meta-predictor θ on the query dataset agent to improve the prediction ability. According to the adjusted θ agent predict the future degradation trend of mechanical equipment. Specifically, according to the query set from the training samples Update and adjust the learned meta-predictor θ after sub-task and cross-sub-task optimization agent , which can be expressed as:

[0091]

[0092] where represents the query set data of sub-task Γ in the training set i , and α is the learning rate. Output the optimized meta-predictor θ agent , as well as the prediction result of the degradation trend of the mechanical equipment, so as to predict the future degradation trend of the mechanical equipment by querying the dataset of the test samples.

[0093] In an embodiment of the present application, the full-life data of the bearing is collected by accelerometers installed horizontally and vertically, as Figure 3 shown. The data from the healthy operation to failure of bearings 1-1, 1-2, 1-3, and 1-4 are used to verify the prediction performance of this method. Time-domain and frequency-domain features are extracted from the vibration signals of bearings 1-1, 1-2, 1-3, and 1-4 from healthy operation to failure, so as to form a degradation feature matrix. Four kinds of prediction data are designed (prediction data 1: bearings 1-2, 1-3, and 1-4 are used as source data, and bearing 1-1 is used as target data; prediction data 2: bearings 1-1, 1-3, and 1-4 are used as source data, and bearing 1-2 is used as target data; prediction data 3: bearings 1-1, 1-2, and 1-4 are used as source data, and bearing 1-3 is used as target data; prediction data 4: bearings 1-1, 1-2, and 1-3 are used as source data, and bearing 1-4 is used as target data) to comprehensively evaluate the proposed mechanical equipment life prediction method based on statistical alignment meta-gated recurrent unit. As Figure 4 shown, the four bearing degradation trend prediction results with different prediction times and data sizes are presented. Each column represents the same prediction time. As Figure 5 shown, the root mean square error results of the degradation trend prediction for different prediction times and data samples are given. From Figure 4 and Figure 5 , it can be seen that at different prediction moments, even with limited five-level data, the final degradation trend can be reasonably predicted, and this method also has excellent robustness to changes in data size.

[0094] In an embodiment of the present application, through comparative analysis with existing life prediction methods, mainly including (AR, GRU, Encoder-decoder, MURU, GDAU, and DSASA), as Figure 6 shown, the comparison results of the root mean square error of life prediction for the prediction dataset using the present method and existing life prediction methods are presented. It can be seen that the error of the mechanical equipment life prediction method based on the statistically aligned meta gated recurrent unit is the lowest among the existing methods, verifying the effectiveness of the present method in mechanical equipment life prediction.

[0095] In an embodiment of the present application, the industrial petrochemical pump motor-rotor dataset in the industrial environment (the models of two petrochemical plants are: P2209C and P1021B) is used to further verify the effectiveness of the present method. The life prediction results are as Figure 7 shown. It can be seen that the present method is still effective in the industrial environment. At the same time, the life prediction error of industrial data is as Figure 8 shown. It can be seen that even in the extreme case where the sample size is reduced to 50% of the original samples, the present method can well align the potential distribution and complete cross-domain life prediction. Additionally, through comparative analysis with existing life prediction methods, mainly including (AR, GRU, Encoder-decoder, MURU, GDAU, and DSASA), as Figure 9 shown, the comparison results of the root mean square error of life prediction for the actual industrial dataset using the present method and existing life prediction methods are presented. It can be seen that the life prediction error of the present method is the lowest, further verifying the superiority and effectiveness of the present method in dealing with the problem of actual industrial data life prediction.

[0096] In the above-mentioned mechanical equipment life prediction method based on the statistically aligned meta gated recurrent unit, first, the full-life vibration data of the rotating mechanical equipment is obtained as the training data; then, a statistical alignment metric is designed to calibrate the difference between the training data and the target data distribution; then, a mechanical equipment life prediction model is constructed, and the calibrated training data is input into the model for subtask learning; finally, the knowledge learned from the subtasks is integrated, followed by cross-subtask learning, and optimization based on the second-order gradient is performed to update the meta-prediction factor, and the future degradation trend of the mechanical equipment is predicted based on the meta-prediction factor. That is to say, by introducing meta-learning and domain adaptation technologies, as well as the statistically aligned meta gated recurrent unit algorithm, effective knowledge transfer between different domains is achieved. In the case of limited data, the prior knowledge in historical big data is applied to the life prediction of unknown domains, improving the performance of cross-domain prediction and enabling effective life prediction in the case of insufficient samples.

[0097] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are sequentially shown according to the indications of the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same moment, but can be executed at different moments. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0098] Based on the same inventive concept, an embodiment of the present application also provides a device for predicting the service life of a mechanical device based on a statistically aligned meta gated recurrent unit for implementing the method for predicting the service life of a mechanical device based on a statistically aligned meta gated recurrent unit described above. The implementation solutions provided by this device to solve problems are similar to the implementation solutions described in the above method. Therefore, the specific limitations in one or more embodiments of the device for predicting the service life of a mechanical device based on a statistically aligned meta gated recurrent unit provided below can refer to the limitations on the method for predicting the service life of a mechanical device based on a statistically aligned meta gated recurrent unit in the above text, and will not be repeated here.

[0099] In one embodiment, as Figure 10 shown, a device 1000 for predicting the service life of a mechanical device based on a statistically aligned meta gated recurrent unit is provided, including: a data acquisition module 1001, a difference calibration module 1003, a learning module 1005, and a mechanical device service life prediction module 1007, where:

[0100] The data acquisition module 1001 is configured to acquire the full-life vibration data of a rotating mechanical device as training data.

[0101] The difference calibration module 1003 is configured to design a statistical alignment metric to calibrate the difference between the training data and the target data distribution.

[0102] The learning module 1005 is configured to construct a mechanical device service life prediction model, input the calibrated training data into the model, and perform subtask learning.

[0103] The mechanical device service life prediction module 1007 is configured to integrate the knowledge learned in the subtasks, then perform cross-subtask learning, and perform optimization based on the second-order gradient to update the meta-prediction factor, and predict the future degradation trend of the mechanical device based on the meta-prediction factor.

[0104] In an embodiment of the present application, the core formula of the statistical alignment metric is:

[0105]

[0106] Among them, p is the intrinsic dimension, is the discrete estimated value, U(ε), U(ε 1 ), U(ε 2 ) are the number of stacks, ε, ε 1 , v 2 are conditional variables.

[0107] In an embodiment of the present application, the difference calibration module is further configured to:

[0108] Construct a statistic to calculate the distance between the training data and the target data;

[0109] Construct a loss function including a statistical alignment metric, and minimize the loss function to solve the mapping matrix.

[0110] In an embodiment of the present application, the difference calibration module is further configured to:

[0111] Quantify the data difference using the maximum mean difference metric as:

[0112]

[0113] where n s and n t are the number of samples from the training data and the target data, is the weight of the i-th training sample, is the weight of the i-th target sample, is the i-th training sample, is the i-th target sample, represents the reproducing kernel Hilbert space, which is used to measure the weighted distance between the training data and the kernel target data, and φ(·) represents the kernel function.

[0114] In an embodiment of the present application, the learning module is further configured to:

[0115] Extract the time series features of the support data set using an encoder, and perform a degradation trend prediction using a decoder based on the time series features.

[0116] In an embodiment of the present application, the optimization formula for the second-order gradient is:

[0117]

[0118] where represents the gradient of the test set loss with respect to the parameters Φ i learned by the subtask, represents the task parameters Φ iThe gradient of the global parameter θ reflects the indirect impact of support set learning on the optimization of the test set, and α is the learning rate. is the test data set, P(T) is the task distribution in multi-task learning, and T i is the label representing the task relevant to this task.

[0119] Each module in the above mechanical equipment life prediction device based on the statistical alignment meta gated recurrent unit can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in the processor of the computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each of the above modules.

[0120] In one embodiment, a computer device is provided. The computer device can be a terminal, and its internal structure diagram can be as Figure 11 shown. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be achieved through WIFI, a mobile cellular network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a mechanical equipment life prediction method based on the statistical alignment meta gated recurrent unit. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, a touchpad, or a mouse, etc.

[0121] Those skilled in the art can understand that Figure 11 the structure shown in is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0122] In one embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, it implements the steps in the above method embodiments.

[0123] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0124] In one embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0125] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0126] Those of ordinary skill in the art can understand that all or part of the processes of implementing the above method embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above method embodiments. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., and are not limited thereto. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., and are not limited thereto.

[0127] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0128] The above-described embodiments merely represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method for predicting the life of mechanical equipment based on statistically aligned gated recursive units, characterized in that: The method comprises: Obtain the full life vibration data of rotating machinery equipment as training data; Designing a statistical alignment metric to calibrate the difference between the training data and the target data distribution; Build a mechanical equipment life prediction model, input the calibrated training data into the model, and perform subtask learning; The knowledge learned from the subtasks is integrated, and then cross-subtask learning is performed, and quadratic gradient-based optimization is performed to update the meta-prediction factor, and the future degradation trend of the mechanical equipment is predicted based on the meta-prediction factor.

2. The method for predicting the life of mechanical equipment based on statistically aligned gated recursive units according to claim 1, characterized in that: The core formula of the statistical alignment metric is: Where p is the intrinsic dimension, is a discrete estimate, U(ε), U(ε1), U(ε2) are the number of stacks, and ε, ε1, ε2 are conditional variables.

3. The method for predicting the life of mechanical equipment based on statistically aligned gated recursive units according to claim 1, characterized in that: The design of a statistical alignment metric to calibrate the difference between the training data and the target data distribution includes: Construct statistics to calculate the distance between training data and target data; A loss function including a statistical alignment metric is constructed, and the mapping matrix is ​​solved by minimizing the loss function.

4. The method for predicting the life of mechanical equipment based on statistically aligned gated recursive units according to claim 3 is characterized in that: The constructing of statistics to calculate the distance between training data and target data includes: The maximum mean difference metric is used to quantify the data differences as: Among them, n s and n t is the number of samples from the training data and the target data, is the weight of the i-th training sample, is the weight of the i-th target sample, is the i-th training sample, is the i-th target sample, represents the reproducing kernel Hilbert space, which is used to measure the weighted distance between the training data and the target data, and φ(·) represents the kernel function.

5. The method for predicting the life of mechanical equipment based on statistically aligned gated recursive units according to claim 1, characterized in that: The construction of the mechanical equipment life prediction model, inputting the calibrated training data into the model, and performing subtask learning includes: An encoder is used to extract time series features of a supporting data set, and a decoder is used to predict degradation trends based on the time series features.

6. The method for predicting the life of mechanical equipment based on statistically aligned gated recursive units according to claim 1, characterized in that: The optimization formula of the quadratic gradient is: in, represents the parameter Φ learned by the subtask i The gradient of the loss with respect to the test set, represents the task parameter Φ i The gradient of the global parameter θ reflects the indirect effect of support set learning on test set optimization. α is the learning rate. is the test data set, P(T) is the task distribution in multi-task learning, T i Indicates the labels related to this task.

7. A mechanical equipment life prediction device based on statistically aligned gated recursive units, characterized in that: The device comprises: A data acquisition module, used to acquire the full life vibration data of the rotating mechanical equipment as training data; A difference calibration module, used to design a statistical alignment metric to calibrate the difference between the training data and the target data distribution; The learning module is used to build a mechanical equipment life prediction model, input the calibrated training data into the model, and perform subtask learning; The mechanical equipment life prediction module is used to integrate the knowledge learned from the subtasks, perform cross-subtask learning, and perform quadratic gradient-based optimization to update the meta-prediction factor, and predict the future degradation trend of the mechanical equipment based on the meta-prediction factor.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, 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 6 are implemented.

Images