Bearing state monitoring sampling optimization method and system based on reinforcement learning

Abstract

The invention relates to the technical field of industrial equipment monitoring, in particular to a bearing state monitoring sampling optimization method and system based on reinforcement learning, and the method and system are characterized in that multi-dimensional parameters such as equipment operation conditions, real-time vibration signals and bearing key degrees are integrated, and dynamic baseline prediction, health index correction and sampling strategy optimization are adopted. And finally, a self-adaptive monitoring sampling strategy which is highly matched with the real-time health state of the bearing is generated, so that the conversion from uniform monitoring to on-demand monitoring is realized, the overall efficiency and the intelligent level of the monitoring system are remarkably improved on the premise of guaranteeing the equipment safety, and scientific analysis and accurate optimization of the bearing state monitoring sampling strategy are realized.

Description

Technical Field

[0001] This invention relates to the field of industrial equipment monitoring technology, specifically to a bearing condition monitoring sampling optimization method and system based on reinforcement learning. Background Technology

[0002] As a core component of rotating machinery, the operating condition of bearings directly affects the safety, stability, and efficiency of the entire equipment and even the production line. Therefore, condition monitoring and health management of bearings, and the implementation of predictive maintenance, are key technologies for ensuring equipment reliability and reducing operation and maintenance costs in the industrial field.

[0003] Existing technologies fail to consider the dynamic impact of changes in operating conditions on the vibration baseline when assessing bearing health status, leading to misjudgments of the health status of the same bearing under different operating conditions. Furthermore, the assessment process does not incorporate the differences in the criticality of bearings within the equipment, making the assessment results unable to reflect actual maintenance priorities. This isolated and static assessment method prevents monitoring strategies from adaptively adjusting to accurate and quantifiable health status results, resulting in wasted resources and insufficient risk management. Summary of the Invention

[0004] To address the problems in related technologies, this invention provides a reinforcement learning-based method and system for optimizing bearing condition monitoring sampling, thereby overcoming the aforementioned technical issues in existing related technologies.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a bearing condition monitoring sampling optimization method based on reinforcement learning, comprising the following steps:

[0006] Acquire the equipment operating condition parameters, real-time vibration signals, and critical bearing parameters of the target bearing;

[0007] A health prediction baseline characterizing the health status of the target bearing is generated based on the equipment operating condition parameters and a preset baseline model, and the initial health index of the target bearing is calculated in combination with real-time vibration signals.

[0008] Based on the equipment operating condition parameters, an operating mode feature vector representing the dynamic characteristics of the bearing operating state is generated, and the initial health index is corrected and compensated in combination with the bearing criticality parameters to obtain a weighted health index.

[0009] Determine whether the weighted health index is less than a preset health threshold. If so, use the preset safety sampling strategy as the bearing monitoring sampling strategy and end the operation.

[0010] If not, the equipment operating condition parameters, operating mode feature vectors, and weighted health indicators are combined to obtain composite state characterization data.

[0011] The composite state representation data is matched with the historical sampling strategy dataset for similarity, and the historical sampling strategy data with similarity greater than a preset similarity threshold are filtered out and summarized to obtain the initial sampling strategy set.

[0012] The composite state characterization data and the initial sampling strategy set are input into a pre-built bearing sampling strategy optimization model for analysis to obtain the bearing monitoring sampling strategy.

[0013] Preferably, the specific steps for obtaining the equipment operating condition parameters, real-time vibration signals, and critical bearing parameters of the target bearing are as follows:

[0014] Set an initial sampling strategy; the initial sampling strategy includes at least a sampling time interval and a sampling duration.

[0015] The operating parameters and vibration signals of the target bearing are collected by multi-source sensors according to the initial sampling strategy to obtain the equipment operating parameters and real-time vibration signals of the target bearing.

[0016] Obtain the equipment number corresponding to the target bearing, and match the bearing criticality parameter corresponding to the target bearing in the bearing criticality parameter database based on the equipment number;

[0017] The bearing criticality parameter database contains bearing criticality parameters corresponding to different equipment numbers.

[0018] Preferably, the specific steps for generating a health prediction baseline characterizing the health status of the target bearing based on the equipment operating condition parameters and a preset baseline model, and calculating the initial health index of the target bearing in conjunction with real-time vibration signals, are as follows:

[0019] S21. Input the operating condition parameters of the equipment into the preset final baseline model to perform health baseline prediction and generate a health prediction baseline characterizing the health status of the target bearing.

[0020] The initial health index of the target bearing is calculated based on the health prediction baseline and real-time vibration signal; the calculation formula is as follows:

[0021] ,

[0022] in, Indicates initial health indicators, and These represent the initial sampling termination time and the initial sampling start time, respectively. and These represent the first and second vibration signals, respectively. The weights and adjustment parameters corresponding to each feature , and Representing time respectively No. The measured value, baseline predicted value, and preset standard deviation corresponding to each feature. This represents the total number of vibration signal features.

[0023] Preferably, the following steps are taken: First, an operating mode feature vector representing the dynamic characteristics of the bearing's operating state is generated based on the equipment operating condition parameters. This vector is then used to correct and compensate the initial health index by combining it with the bearing criticality parameters, resulting in a weighted health index. Second, it is determined whether the weighted health index is less than a preset health threshold. If so, a preset safety sampling strategy is used as the bearing monitoring sampling strategy, and the operation ends. If not, the specific steps for combining the equipment operating condition parameters, the operating mode feature vector, and the weighted health index to obtain composite state characterization data are as follows:

[0024] Construct a working condition-operation mode mapping relationship; the working condition-operation mode mapping relationship is used to map continuous equipment operating condition parameters into operating mode feature vectors that characterize the dynamic characteristics of bearing operating state;

[0025] Based on the mapping relationship between the equipment operating condition parameters and the operating condition-operating mode, a dynamic characteristic vector representing the bearing operating state is generated.

[0026] The operating mode feature vector includes the strength of the bearing operating state belonging to various preset operating modes;

[0027] The real-time risk coefficient of the target bearing is calculated based on the characteristic vector of the operating mode.

[0028] The initial health index is corrected and compensated based on the real-time risk coefficient and bearing criticality parameters to obtain a weighted health index.

[0029] Determine whether the weighted health index is less than a preset health threshold;

[0030] If so, the preset safety sampling strategy will be used as the bearing monitoring sampling strategy, and the operation will end.

[0031] If not, the equipment operating condition parameters, operating mode feature vectors, and weighted health indicators are combined to obtain composite state characterization data.

[0032] By correcting and compensating the calculated initial health indicators using the operating mode feature vector generated from the operating condition parameters and the criticality parameters of the target bearing, the possibility of misjudgment caused by ignoring the differences in operating modes and bearing importance is reduced, ensuring that the health status information on which subsequent sampling strategy decisions are based is closer to the actual situation.

[0033] Preferably, the specific steps for performing similarity matching between the composite state representation data and the historical sampling strategy dataset, filtering out historical sampling strategy data with similarity greater than a preset similarity threshold, and summarizing them to obtain the initial sampling strategy set are as follows:

[0034] By acquiring historical sampling strategy data of several bearings in different operating states through the Internet of Things, a historical sampling strategy dataset is obtained; the historical sampling strategy data includes historical composite state characterization data of different bearings in different operating states and the historical bearing monitoring sampling strategies adopted.

[0035] Calculate the similarity between the composite state representation data and the historical sampling strategy data in the historical sampling strategy dataset;

[0036] If the similarity between the composite state characterization data and each historical sampling strategy data in the historical sampling strategy dataset is greater than a preset similarity threshold, then the historical sampling strategy data is retained, and the historical bearing monitoring sampling strategies contained in all the retained historical sampling strategy data are summarized to obtain an initial sampling strategy set.

[0037] By quickly filtering out an initial sampling strategy set that is similar to the current state from historical data through similarity matching, and combining composite state representation data and preset model to generate bearing monitoring sampling strategies, the model's response speed is improved. This avoids the inefficiency of the model calculating from scratch and overcomes the limitation that relying solely on historical experience may not be able to cope with new situations, thus achieving a balance between strategy generation efficiency and adaptability.

[0038] Preferably, the specific steps for inputting the composite state characterization data and the initial sampling strategy set into a pre-constructed bearing sampling strategy optimization model for analysis to obtain the bearing monitoring sampling strategy are as follows:

[0039] The similarity between each historical sampling strategy data and other historical sampling strategy data in the historical sampling strategy dataset is calculated sequentially. The historical bearing monitoring sampling strategies contained in the historical sampling strategy data with a similarity greater than the preset similarity threshold are assigned to each historical sampling strategy data as the historical initial sampling strategy set corresponding to the historical sampling strategy data.

[0040] The final bearing sampling strategy optimization model is constructed based on the historical sampling strategy dataset and the historical initial sampling strategy set corresponding to each historical sampling strategy data in the historical sampling strategy dataset.

[0041] The composite state characterization data and the initial sampling strategy set are input into the pre-constructed final bearing sampling strategy optimization model for analysis to obtain the bearing monitoring sampling strategy.

[0042] The present invention also includes a bearing condition monitoring sampling optimization system based on reinforcement learning, comprising a data acquisition module, an initial health index measurement module, a weighted health index analysis module, an initial sampling strategy matching module, and a bearing monitoring sampling strategy optimization module;

[0043] The data acquisition module is used to acquire the equipment operating condition parameters, real-time vibration signals, and critical parameters of the target bearing.

[0044] The initial health index measurement module is used to generate a health prediction baseline characterizing the health status of the target bearing based on the equipment operating condition parameters and a preset baseline model, and to calculate the initial health index of the target bearing in combination with real-time vibration signals.

[0045] The weighted health index analysis module is used to generate a dynamic characteristic vector representing the bearing's operating state based on the equipment's operating condition parameters, and to correct and compensate the initial health index by combining it with the bearing's criticality parameters to obtain a weighted health index; it determines whether the weighted health index is less than a preset health threshold. If so, it uses a preset safety sampling strategy as the bearing monitoring sampling strategy and ends the current operation; otherwise, it combines the equipment's operating condition parameters, the operating mode characteristic vector, and the weighted health index to obtain composite state characterization data.

[0046] The initial sampling strategy matching module is used to perform similarity matching between the composite state representation data and the historical sampling strategy dataset, filter out historical sampling strategy data with similarity greater than a preset similarity threshold, and summarize them to obtain the initial sampling strategy set.

[0047] The bearing monitoring sampling strategy optimization module is used to input the composite state characterization data and the initial sampling strategy set into a pre-built bearing sampling strategy optimization model for analysis, so as to obtain the bearing monitoring sampling strategy.

[0048] By employing the above technical solution, the present invention provides a bearing condition monitoring sampling optimization method and system based on reinforcement learning, which has at least the following beneficial effects:

[0049] 1. This invention integrates multiple parameters such as equipment operating conditions, real-time vibration signals, and the criticality of the bearing, and employs dynamic baseline prediction, health index correction, and sampling strategy optimization to ultimately generate an adaptive monitoring sampling strategy that highly matches the real-time health status of the bearing. This achieves a shift from uniform monitoring to on-demand monitoring, significantly improving the overall efficiency and intelligence level of the monitoring system while ensuring equipment safety. It also realizes the scientific analysis and precise optimization of the bearing condition monitoring sampling strategy.

[0050] 2. This invention corrects and compensates the calculated initial health index by using the operating mode feature vector generated from the operating condition parameters and the criticality parameter of the target bearing. This reduces the possibility of misjudgment caused by ignoring the differences in operating modes and bearing importance, and ensures that the health status information on which subsequent sampling strategy decisions are based is closer to the actual situation.

[0051] 3. This invention quickly selects an initial sampling strategy set that is similar to the current state from historical data through similarity matching. Combined with composite state representation data and preset model, it generates bearing monitoring sampling strategies, which improves the model's response speed. This avoids the inefficiency of the model calculating from scratch and overcomes the limitation that relying solely on historical experience may not be able to cope with new situations, thus achieving a balance between strategy generation efficiency and adaptability. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, the drawings can be obtained from these drawings without creative effort.

[0053] Figure 1 A flowchart of the bearing condition monitoring sampling optimization method provided by the present invention;

[0054] Figure 2 This is a schematic diagram of the modules of the bearing condition monitoring sampling optimization system provided by the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Example 1 is as follows:

[0057] Addressing the limitations of existing technologies, this embodiment proposes a reinforcement learning-based sampling optimization method for bearing condition monitoring. For example... Figure 1 As shown, the method includes the following steps:

[0058] Acquire the equipment operating condition parameters, real-time vibration signals, and critical bearing parameters of the target bearing;

[0059] The process of acquiring the equipment operating condition parameters, real-time vibration signals, and critical bearing parameters of the target bearing includes the following steps:

[0060] Set an initial sampling strategy; the initial sampling strategy includes at least a sampling time interval and a sampling duration.

[0061] The operating parameters and vibration signals of the target bearing are collected by multi-source sensors according to the initial sampling strategy to obtain the equipment operating parameters and real-time vibration signals of the target bearing.

[0062] The multi-source sensor includes at least a speed sensor, a torque sensor, and a vibration sensor;

[0063] The operating parameters include at least the spindle speed and load;

[0064] Obtain the equipment number corresponding to the target bearing, and match the bearing criticality parameter corresponding to the target bearing in the bearing criticality parameter database based on the equipment number;

[0065] The bearing criticality parameter database contains bearing criticality parameters corresponding to different equipment numbers.

[0066] A health prediction baseline characterizing the health status of the target bearing is generated based on the equipment operating condition parameters and a preset baseline model, and the initial health index of the target bearing is calculated in combination with real-time vibration signals.

[0067] The calculation of the initial health index of the target bearing includes the following steps:

[0068] The operating condition parameters of the equipment are input into a preset final baseline model to perform health baseline prediction, thereby generating a health prediction baseline that characterizes the health status of the target bearing.

[0069] The final baseline model adopts a multilayer perceptron model; the construction process of the baseline model includes the following steps:

[0070] Historical operating data of several bearings are acquired through the Internet of Things to obtain a historical operating dataset; the historical operating data includes historical equipment operating condition parameters of different bearings and corresponding standard health index curves.

[0071] Construct an initial baseline model and set a first training data ratio, such as 8:2 or 7.5:2.5. The specific ratio can be adjusted reasonably according to the actual situation. Divide the historical running dataset into data using the first training data ratio to obtain the first training dataset and the first test dataset.

[0072] Set a first training error threshold, such as 5%-10%, which can be adjusted reasonably according to the actual situation. Input the first training dataset into the initial baseline model for training. Continuously adjust the parameters of the initial baseline model according to the training results until the training error is less than the first training error threshold, and obtain the trained baseline model.

[0073] Set a first test precision, such as 90%-95%, which can be adjusted reasonably according to the actual situation. Input the first test dataset into the trained baseline model for testing, and calculate the accuracy of the test results. If the accuracy of the test results is greater than the first test precision, the final baseline model is obtained; otherwise, re-execute the training operation until the accuracy of the test results is greater than the first test precision.

[0074] The structure of the initial baseline model can be seen in Table 1 below:

[0075] Table 1

[0076] Model Name Model type Model Structure Initial baseline model Multilayer perceptron model Input Layer: Number of nodes: 3-5 neurons; Input Features: Working condition parameter vector; First Dense Layer: Number of nodes: 8-16 neurons; Activation Function: ReLU activation function; Weight Initialization: He normal initialization; Connection Type: Fully connected; Second Dense Layer: Number of nodes: 4-8 neurons; Activation Function: ReLU activation function; Weight Initialization: He normal initialization; Output Layer: Number of nodes: 1-3 neurons; Activation Function: Linear activation function; Parameter Settings: Optimization Algorithm: Adam optimizer; Initial Learning Rate: 0.001; Learning Rate Decrease: Multiplied by 0.9 every 100 iterations; Gradient Clipping: Threshold 1.0 to prevent gradient explosion; Early Stopping Strategy: Training stops if the loss does not decrease for 20 consecutive iterations; Main Loss Function: Mean Squared Error Loss; Number of Training Rounds: Dynamically controlled, maximum 500 rounds.

[0077] The initial health index of the target bearing is calculated based on the health prediction baseline and real-time vibration signal; the calculation formula is as follows:

[0078] ,

[0079] in, Indicates initial health indicators, and These represent the initial sampling termination time and the initial sampling start time, respectively. and These represent the first and second vibration signals, respectively. The weights and adjustment parameters corresponding to each feature , and Representing time respectively No. The measured value, baseline predicted value, and preset standard deviation corresponding to each feature. This represents the total number of vibration signal features;

[0080] The vibration signal characteristics include at least the effective value of vibration, the peak value of vibration, and the kurtosis of vibration.

[0081] Based on the equipment operating condition parameters, a dynamic characteristic of the bearing operating state is generated to form an operating mode feature vector. The initial health index is then corrected and compensated by combining the bearing criticality parameters to obtain a weighted health index. It is determined whether the weighted health index is less than a preset health threshold. If so, the preset safety sampling strategy is used as the bearing monitoring sampling strategy, and the operation ends. If not, the equipment operating condition parameters, the operating mode feature vector, and the weighted health index are combined to obtain composite state characterization data.

[0082] The process of obtaining composite state characterization data includes the following steps:

[0083] Construct a working condition-operation mode mapping relationship; the working condition-operation mode mapping relationship is used to map continuous equipment operating condition parameters into operating mode feature vectors that characterize the dynamic characteristics of bearing operating state;

[0084] The working condition-operation mode mapping relationship is established by registration and analysis based on a large amount of historical normal bearing data, thereby establishing the correspondence between operation mode and speed and load combination;

[0085] Based on the mapping relationship between the equipment operating condition parameters and the operating condition-operating mode, a dynamic characteristic vector representing the bearing operating state is generated.

[0086] The operating modes include, but are not limited to, stable operation mode, load fluctuation mode, impact overload mode, and start-stop transient mode;

[0087] The operating mode feature vector includes the strength of the bearing operating state belonging to various preset operating modes;

[0088] The real-time risk coefficient of the target bearing is calculated based on the feature vector of the operating mode; the calculation formula is as follows:

[0089] ,

[0090] in, Indicates the real-time risk coefficient. and They represent the first The intensity and basic risk weights corresponding to each operating mode This represents the total number of classes of operating modes contained in the operating mode feature vector;

[0091] The initial health index is corrected and compensated based on the real-time risk coefficient and bearing criticality parameters to obtain a weighted health index; the correction and compensation formula is as follows:

[0092] ,

[0093] in, Indicates a weighted health indicator. Indicates the initial weights. This represents the balance coefficient between the operating mode and the criticality of the bearing. It can be set to 0.6, but can be adjusted appropriately according to actual conditions. Indicates the criticality parameter of the bearing;

[0094] Determine whether the weighted health index is less than a preset health threshold;

[0095] If so, the preset safety sampling strategy will be used as the bearing monitoring sampling strategy, and the operation will end.

[0096] If not, the equipment operating condition parameters, operating mode feature vectors, and weighted health indicators are combined to obtain composite state characterization data.

[0097] The safety sampling strategy refers to the high-frequency sampling strategy adopted by the system when it determines that the bearing health condition is in a dangerous state.

[0098] By correcting and compensating the calculated initial health indicators using the operating mode feature vector generated from the operating condition parameters and the criticality parameters of the target bearing, the possibility of misjudgment caused by ignoring the differences in operating modes and bearing importance is reduced, ensuring that the health status information on which subsequent sampling strategy decisions are based is closer to the actual situation.

[0099] The composite state representation data is matched with the historical sampling strategy dataset for similarity, and the historical sampling strategy data with similarity greater than a preset similarity threshold are filtered out and summarized to obtain the initial sampling strategy set.

[0100] The process of obtaining the initial sampling strategy set includes the following steps:

[0101] By acquiring historical sampling strategy data of several bearings in different operating states through the Internet of Things, a historical sampling strategy dataset is obtained; the historical sampling strategy data includes historical composite state characterization data of different bearings in different operating states and the historical bearing monitoring sampling strategies adopted.

[0102] The historical bearing monitoring sampling strategy includes at least the selection of sampling time interval, the setting of sampling duration, and the adjustment of sampling resolution.

[0103] Calculate the similarity between the composite state representation data and each historical sampling strategy data in the historical sampling strategy dataset; the calculation formula is as follows:

[0104] ,

[0105] in, This indicates that the composite state characterization data is related to the first... Similarity of historical sampling strategy data This represents the feature vector corresponding to the composite state characterization data. Indicates the first The feature vector corresponding to the historical composite state representation data contained in each historical sampling strategy data;

[0106] If the similarity between the composite state characterization data and each historical sampling strategy data in the historical sampling strategy dataset is greater than a preset similarity threshold, then the historical sampling strategy data is retained, and the historical bearing monitoring sampling strategies contained in all the retained historical sampling strategy data are summarized to obtain an initial sampling strategy set.

[0107] By quickly filtering out an initial sampling strategy set that is similar to the current state from historical data through similarity matching, and combining composite state representation data and preset model to generate bearing monitoring sampling strategies, the model's response speed is improved. This avoids the inefficiency of the model calculating from scratch and overcomes the limitation that relying solely on historical experience may not be able to cope with new situations, thus achieving a balance between strategy generation efficiency and adaptability.

[0108] The composite state characterization data and the initial sampling strategy set are input into a pre-constructed bearing sampling strategy optimization model for analysis to obtain the bearing monitoring sampling strategy.

[0109] The obtained bearing monitoring sampling strategy includes the following steps:

[0110] The similarity between each historical sampling strategy data and other historical sampling strategy data in the historical sampling strategy dataset is calculated sequentially. The historical bearing monitoring sampling strategies contained in the historical sampling strategy data with a similarity greater than the preset similarity threshold are assigned to each historical sampling strategy data as the historical initial sampling strategy set corresponding to the historical sampling strategy data.

[0111] The final bearing sampling strategy optimization model is constructed based on the historical sampling strategy dataset and the historical initial sampling strategy set corresponding to each historical sampling strategy data in the historical sampling strategy dataset.

[0112] The final bearing sampling strategy optimization model adopts a dual-depth Q-network model; the construction process of the final bearing sampling strategy optimization model includes the following steps:

[0113] Construct an initial bearing sampling strategy optimization model and set a second training data ratio, such as 8:2 or 7.5:2.5. The specific ratio can be adjusted reasonably according to the actual situation. Based on the second training data ratio, divide the historical sampling strategy dataset and the historical initial sampling strategy set corresponding to each historical sampling strategy data in the historical sampling strategy dataset into a second training dataset and a second test dataset.

[0114] Set a second training error threshold, such as 5%-10%, which can be adjusted reasonably according to the actual situation. Input the second training dataset into the initial bearing sampling strategy optimization model for training. Continuously adjust the parameters of the initial bearing sampling strategy optimization model according to the training results until the training error is less than the second training error threshold, and obtain the trained bearing sampling strategy optimization model.

[0115] Set a second test precision, such as 90%-95%, which can be adjusted reasonably according to the actual situation. Input the test data in the second test dataset into the trained bearing sampling strategy optimization model for testing, and calculate the accuracy of the test results. If the accuracy of the test results is greater than the second test precision, the final bearing sampling strategy optimization model is obtained; otherwise, re-execute the training operation until the accuracy of the test results is greater than the second test precision.

[0116] The structure of the initial bearing sampling strategy optimization model can be seen in Table 2 below:

[0117] Table 2

[0118] Model Name Model type Model Structure Initial bearing sampling strategy optimization model Dual-depth Q-network model Feature layer extraction module: Primary feature extraction layer: Fully connected layer 1: 128 neurons, input dimension 18→128, using ReLU activation function, with an additional batch normalization layer for stable training; Dropout layer: dropout rate 0.2, to prevent overfitting; Fully connected layer 2: 64 neurons, 128→64 dimension transformation, ReLU activation, forming a shared feature basis; State feature extraction module: LSTM layer: 64 units, bidirectional structure, extracting temporal features; 1D convolutional layer: 32 filters, kernel size 5, extracting frequency domain features; Global pooling layer: compressing feature dimension; Dual Q network core module: Dense layer 1:1 28 neurons, ReLU activation; Dense layer 2: 64 neurons, ELU activation; Output layer: N neurons, linear activation; Target network: same structure as the online network, parameters are periodically synchronized from the online network; Feature fusion module: concatenation of device status features and vibration features; Dense layer: 96 neurons, feature fusion; Parameter settings: Optimization algorithm: Adam optimizer; Initial learning rate: 0.0008; Learning rate decay: learning rate multiplied by 0.95 every 10 training epochs; Gradient clipping: gradient L2 norm threshold set to 1.5 to avoid instability in the early stages of training; Early stopping strategy: stop if there is no improvement in average reward for 25 consecutive epochs;

[0119] The composite state characterization data and the initial sampling strategy set are input into the pre-constructed final bearing sampling strategy optimization model for analysis to obtain the bearing monitoring sampling strategy.

[0120] Simultaneously, the total reward value after the system executes the bearing monitoring sampling strategy is calculated based on the reward function. Based on the total reward value and using the time difference error or strategy gradient method, the parameters of the final bearing sampling strategy optimization model are updated using gradient.

[0121] The reward function is constructed by comprehensively considering the minimization of fault missed detection risk, the control of sampling resource consumption, and the constraints of data storage cost.

[0122] The dual-deep Q-network effectively overcomes the systematic overestimation of the value of bearing sampling actions in traditional deep Q-networks by decoupling action selection and value assessment, thus obtaining a more accurate and unbiased policy evaluation. Simultaneously, the dual-network mechanism significantly improves training stability, reduces value estimation fluctuations, and ensures reliable convergence of the sampling strategy. This allows the model to more accurately balance the long-term relationship between sampling frequency, timing, energy consumption, and fault detection capabilities. Ultimately, while ensuring monitoring performance, it achieves optimal adaptive sampling of bearing conditions, extending equipment lifespan and reducing maintenance costs.

[0123] Example 2 is as follows:

[0124] Please see Figure 2 A bearing condition monitoring sampling optimization system based on reinforcement learning includes a data acquisition module, an initial health index measurement module, a weighted health index analysis module, an initial sampling strategy matching module, and a bearing monitoring sampling strategy optimization module.

[0125] The data acquisition module is used to acquire the equipment operating condition parameters, real-time vibration signals, and critical parameters of the target bearing.

[0126] The initial health index measurement module is used to generate a health prediction baseline characterizing the health status of the target bearing based on the equipment operating condition parameters and a preset baseline model, and to calculate the initial health index of the target bearing in combination with real-time vibration signals.

[0127] The weighted health index analysis module is used to generate a dynamic characteristic vector representing the bearing's operating state based on the equipment's operating condition parameters, and to correct and compensate the initial health index by combining it with the bearing's criticality parameters to obtain a weighted health index; it determines whether the weighted health index is less than a preset health threshold. If so, it uses a preset safety sampling strategy as the bearing monitoring sampling strategy and ends the current operation; otherwise, it combines the equipment's operating condition parameters, the operating mode characteristic vector, and the weighted health index to obtain composite state characterization data.

[0128] The initial sampling strategy matching module is used to perform similarity matching between the composite state representation data and the historical sampling strategy dataset, filter out historical sampling strategy data with similarity greater than a preset similarity threshold, and summarize them to obtain the initial sampling strategy set.

[0129] The bearing monitoring sampling strategy optimization module is used to input the composite state characterization data and the initial sampling strategy set into a pre-built bearing sampling strategy optimization model for analysis, so as to obtain the bearing monitoring sampling strategy.

[0130] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0131] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0132] The preferred embodiments of the invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A bearing condition monitoring sampling optimization method, characterized in that, Includes the following steps: Acquire the equipment operating condition parameters, real-time vibration signals, and critical bearing parameters of the target bearing; A health prediction baseline characterizing the health status of the target bearing is generated based on the equipment operating condition parameters and a preset baseline model, and the initial health index of the target bearing is calculated in combination with real-time vibration signals. Based on the equipment operating condition parameters, an operating mode feature vector representing the dynamic characteristics of the bearing operating state is generated, and the initial health index is corrected and compensated in combination with the bearing criticality parameters to obtain a weighted health index. Determine whether the weighted health index is less than a preset health threshold. If so, use the preset safety sampling strategy as the bearing monitoring sampling strategy and end the operation. If not, the equipment operating condition parameters, operating mode feature vectors, and weighted health indicators are combined to obtain composite state characterization data. The composite state representation data is matched with the historical sampling strategy dataset for similarity, and the historical sampling strategy data with similarity greater than a preset similarity threshold are filtered out and summarized to obtain the initial sampling strategy set. The composite state characterization data and the initial sampling strategy set are input into a pre-built bearing sampling strategy optimization model for analysis to obtain the bearing monitoring sampling strategy.

2. The bearing condition monitoring sampling optimization method according to claim 1, characterized in that, The process of acquiring the equipment operating condition parameters, real-time vibration signals, and critical bearing parameters of the target bearing includes the following steps: Set an initial sampling strategy; the initial sampling strategy includes at least a sampling time interval and a sampling duration. The operating parameters and vibration signals of the target bearing are collected by multi-source sensors according to the initial sampling strategy to obtain the equipment operating parameters and real-time vibration signals of the target bearing. Obtain the equipment number corresponding to the target bearing, and match the bearing criticality parameter corresponding to the target bearing in the bearing criticality parameter database based on the equipment number; The bearing criticality parameter database contains bearing criticality parameters corresponding to different equipment numbers.

3. The bearing condition monitoring sampling optimization method according to claim 1, characterized in that, The calculation of the initial health index of the target bearing includes the following steps: The operating condition parameters of the equipment are input into a preset final baseline model to perform health baseline prediction, thereby generating a health prediction baseline that characterizes the health status of the target bearing. The initial health index of the target bearing is calculated based on the health prediction baseline and real-time vibration signal; the calculation formula is as follows: ， in, Indicates initial health indicators, and These represent the initial sampling termination time and the initial sampling start time, respectively. and These represent the first and second vibration signals, respectively. The weights and adjustment parameters corresponding to each feature , and Representing time respectively No. The measured value, baseline predicted value, and preset standard deviation corresponding to each feature. This represents the total number of vibration signal features.

4. The bearing condition monitoring sampling optimization method according to claim 1, characterized in that, The process of obtaining composite state characterization data includes the following steps: Construct a working condition-operation mode mapping relationship; the working condition-operation mode mapping relationship is used to map continuous equipment operating condition parameters into operating mode feature vectors that characterize the dynamic characteristics of bearing operating state; Based on the mapping relationship between the equipment operating condition parameters and the operating condition-operating mode, a dynamic characteristic vector representing the bearing operating state is generated. The operating mode feature vector includes the strength of the bearing operating state belonging to various preset operating modes; The real-time risk coefficient of the target bearing is calculated based on the characteristic vector of the operating mode. The initial health index is corrected and compensated based on the real-time risk coefficient and bearing criticality parameters to obtain a weighted health index. Determine whether the weighted health index is less than a preset health threshold; If so, the preset safety sampling strategy will be used as the bearing monitoring sampling strategy, and the operation will end. If not, the equipment operating condition parameters, operating mode feature vectors, and weighted health indicators are combined to obtain composite state characterization data.

5. The bearing condition monitoring sampling optimization method according to claim 1, characterized in that, The process of obtaining the initial sampling strategy set includes the following steps: By acquiring historical sampling strategy data of several bearings in different operating states through the Internet of Things, a historical sampling strategy dataset is obtained; the historical sampling strategy data includes historical composite state characterization data of different bearings in different operating states and the historical bearing monitoring sampling strategies adopted. Calculate the similarity between the composite state representation data and the historical sampling strategy data in the historical sampling strategy dataset; If the similarity between the composite state characterization data and each historical sampling strategy data in the historical sampling strategy dataset is greater than a preset similarity threshold, then the historical sampling strategy data is retained, and the historical bearing monitoring sampling strategies contained in all the retained historical sampling strategy data are summarized to obtain the initial sampling strategy set.

6. The bearing condition monitoring sampling optimization method according to claim 1, characterized in that, The obtained bearing monitoring sampling strategy includes the following steps: Calculate the similarity between each historical sampling strategy data in the historical sampling strategy dataset and other historical sampling strategy data in turn. Assign historical bearing monitoring sampling strategies contained in historical sampling strategy data with similarity greater than the preset similarity threshold to each historical sampling strategy data as the historical initial sampling strategy set corresponding to that historical sampling strategy data. The final bearing sampling strategy optimization model is constructed based on the historical sampling strategy dataset and the historical initial sampling strategy set corresponding to each historical sampling strategy data in the historical sampling strategy dataset. The composite state characterization data and the initial sampling strategy set are input into the pre-constructed final bearing sampling strategy optimization model for analysis to obtain the bearing monitoring sampling strategy.

7. The bearing condition monitoring sampling optimization method according to claim 6, characterized in that, The final bearing sampling strategy optimization model adopts a dual-depth Q-network model.

8. A system for implementing the bearing condition monitoring sampling optimization method according to any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire the equipment operating condition parameters, real-time vibration signals, and critical parameters of the target bearing. The initial health index measurement module is used to generate a health prediction baseline characterizing the health status of the target bearing based on the equipment operating condition parameters and a preset baseline model, and to calculate the initial health index of the target bearing in combination with real-time vibration signals. The weighted health index analysis module is used to generate a dynamic characteristic vector representing the bearing's operating state based on the equipment's operating condition parameters, and to correct and compensate the initial health index by combining it with the bearing's criticality parameters to obtain a weighted health index; it determines whether the weighted health index is less than a preset health threshold. If so, it uses a preset safety sampling strategy as the bearing monitoring sampling strategy and ends the current operation; otherwise, it combines the equipment's operating condition parameters, the operating mode characteristic vector, and the weighted health index to obtain composite state characterization data. The initial sampling strategy matching module is used to perform similarity matching between the composite state representation data and the historical sampling strategy dataset, filter out historical sampling strategy data with similarity greater than a preset similarity threshold, and summarize them to obtain the initial sampling strategy set. The bearing monitoring sampling strategy optimization module is used to input the composite state characterization data and the initial sampling strategy set into a pre-built bearing sampling strategy optimization model for analysis, so as to obtain the bearing monitoring sampling strategy.

9. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method as described in any one of claims 1-7.

10. A computer storage medium storing computer-executable instructions thereon, characterized in that: When the computer-executable instructions are executed by a processor, they implement the steps of the method as described in any one of claims 1-7.

Images