Online simulation method and system for mechanical and electrical product wear test

Abstract

The invention relates to the field of mechanical and electrical product abrasion testing, and provides an online simulation method and system for mechanical and electrical product abrasion testing, and the method comprises the following steps: collecting historical multi-dimensional monitoring data in the operation process of a mechanical and electrical product from a historical database; establishing an initial mechanical and electrical product wear prediction model based on a Bayesian network and a convolutional neural network; training and testing an initial mechanical and electrical product wear prediction model through the historical multi-dimensional monitoring data to obtain a mechanical and electrical product wear online prediction model; and collecting electromechanical product operation multi-dimensional monitoring data in the electromechanical product operation process in real time, inputting the electromechanical product operation multi-dimensional monitoring data into the electromechanical product wear online prediction model, and outputting a wear prediction result of the electromechanical product. According to the method, historical multi-dimensional monitoring data is collected, the Bayesian network and the convolutional neural network are combined to establish an initial mechanical and electrical product wear prediction model, and online wear prediction is realized after the model is trained and tested, so that the wear condition of the mechanical and electrical product is monitored and predicted in real time, and the wear test accuracy and efficiency are improved.

Description

Technical Field

[0001] The invention relates to the field of wear testing of electromechanical products, and in particular to an online simulation method and system for wear testing of electromechanical products. Background Art

[0002] In the process of predictive maintenance, it is important to know the condition of electromechanical switching elements, including such as electromechanical relays or contactors or electromechanical valves, and in particular hydraulic and / or pneumatic actuators, and to replace worn actuators or switching elements before a possible failure in order to avoid stops or downtime.

[0003] At present, wear prediction of electromechanical products mainly uses the network structure of deep learning models or classical machine learning models, such as convolutional neural networks, recurrent neural networks or Bayesian networks, etc. However, the convergence speed of convolutional neural networks and recurrent neural networks is limited, and Bayesian networks are highly dependent on data. Using the above models alone to predict wear of electromechanical products has low prediction efficiency, low prediction accuracy, and lack of model interpretability. Summary of the invention

[0004] In view of this, the present invention proposes an online simulation method and system for wear testing of electromechanical products. By collecting historical multi-dimensional monitoring data, combining Bayesian networks and convolutional neural networks, an initial electromechanical product wear prediction model is established. After training and testing the model, online wear prediction is realized, thereby realizing real-time monitoring and prediction of the wear of electromechanical products, improving the accuracy and efficiency of wear testing, and at the same time increasing the interpretability of the model through the Bayesian network, solving the problems of low efficiency and low accuracy of current wear prediction methods for electromechanical products.

[0005] The technical solution of the present invention is implemented as follows: On the one hand, the present invention provides an online simulation method for wear test of electromechanical products, comprising the following steps:

[0006] S1, collects historical multi-dimensional monitoring data of electromechanical products during operation from the historical database;

[0007] S2, establish the initial wear prediction model of electromechanical products based on Bayesian network and convolutional neural network;

[0008] S3, training and testing the initial electromechanical product wear prediction model through the historical multi-dimensional monitoring data to obtain an online electromechanical product wear prediction model;

[0009] S4, collecting multi-dimensional monitoring data of the electromechanical product during its operation in real time, inputting the multi-dimensional monitoring data into an online wear prediction model for the electromechanical product, and outputting a wear prediction result for the electromechanical product.

[0010] On the basis of the above technical solution, preferably, step S1 includes:

[0011] The historical multi-dimensional monitoring data includes operating status monitoring characteristics, usage history and environmental characteristics, and fault history characteristics;

[0012] The operating status monitoring characteristics include vibration data characteristics, temperature data characteristics, current data characteristics, and power data characteristics;

[0013] The usage history and environmental characteristics include usage duration characteristics, workload characteristics, environmental temperature characteristics, and maintenance record characteristics;

[0014] The fault history characteristics include the time of occurrence of each fault, the type of fault, and the cause of the fault.

[0015] Based on the above technical solution, preferably, step S2 includes:

[0016] The initial electromechanical product wear prediction model includes an input layer, a feature extraction layer, a feature learning layer, a feature fusion layer, and an output layer. The input layer is connected to the feature extraction layer and the feature learning layer respectively, the feature learning layer and the feature extraction layer are connected to the feature fusion layer respectively, and the feature fusion layer is connected to the output layer;

[0017] The input layer is used to receive input features, the feature extraction layer is used to extract time-frequency features of the input data, the feature learning layer is used to learn the probabilistic relationship features between variables in the input data, the feature fusion layer is used to fuse the output of the feature extraction layer and the output of the feature learning layer, and the output layer is used to receive the output of the feature fusion layer and output the prediction result of the model.

[0018] On the basis of the above technical solution, preferably, the feature extraction layer adopts a convolutional neural network structure, including a convolutional layer and a ReLU activation layer, and the feature extraction function expression of the feature extraction layer is:

[0019] H t =ReLU(W t X+α);

[0020] Among them, X is the input feature of the input layer, H t is the time-frequency feature of X, W t is the weight of the time-frequency feature corresponding to the wear state, and α is the feature extraction bias term.

[0021] On the basis of the above technical solution, preferably, the feature learning layer adopts a Bayesian network structure, including a first fully connected layer, a second fully connected layer and a batch normalization layer, and the probability relationship feature function expression of the feature learning layer is:

[0022]

[0023] H x =Sigmoid(ω2ReLU(ω1X+τ1)+τ2);

[0024] Among them, X is the input feature of the input layer, H x is the initial probability feature, Z x To improve the stability and generalization performance of the probability relationship feature of X, β is the learnable scaling parameter of the batch normalization layer, is the offset parameter of the batch normalization layer, μ is the preset mean of the batch normalization layer, and λ 2 is the preset variance of the batch normalization layer, ε is the anti-zero constant, ω1 and ω2 are the weights of the first and second fully connected layers, τ1 and τ2 are the bias terms of the first and second fully connected layers, respectively.

[0025] On the basis of the above technical solution, preferably, the feature fusion function expression of the feature fusion layer is:

[0026] H r =Attention(γH t +(1-γ)Z x );

[0027] Among them, H r is the fusion feature, Attention(·) is the attention mechanism, and γ is the learning weight of the feature extraction layer.

[0028] On the basis of the above technical solution, preferably, the prediction result function expression of the output layer is:

[0029] Y = Sigmoid(η[H r ,X prior ]+δ);

[0030] Among them, Y is the prediction result, η is the weight matrix of the output layer, and H r is the fusion feature, X prior is the prior information, [H r ,X prior ] is the concatenation of fusion features and prior information, and δ is the bias vector of the output layer.

[0031] On the basis of the above technical solution, preferably, step S3 includes:

[0032] S31, divide the historical multi-dimensional monitoring data into a training set and a test set in a ratio of 7:3;

[0033] S32, iteratively training the initial electromechanical product wear prediction model through the training set, configuring the optimizer, loss function and hyperparameters until the iteration stop condition is reached, thereby obtaining the trained initial electromechanical product wear prediction model, wherein the iteration stop condition includes: the number of iterations reaches a preset value and the loss function converges;

[0034] S33, use the test set to test the initial electromechanical product wear prediction model after training, and evaluate the model through the comprehensive accuracy evaluation value. If the model evaluation meets the requirements, the training and testing are completed, and the electromechanical product wear online prediction model is obtained. If the model evaluation does not meet the requirements, return to step S32 to adjust the hyperparameters and re-train the model iteratively.

[0035] On the basis of the above technical solution, preferably, the calculation formula of the comprehensive accuracy evaluation value is:

[0036]

[0037] Among them, K is the comprehensive accuracy evaluation value, ω i is the weight of the i-th prediction item of the wear prediction result, p i is the accuracy of the i-th prediction item of the wear prediction result, and n is the number of prediction items of the wear prediction result;

[0038] The wear prediction result is a multi-dimensional feature vector, and each dimension corresponds to a prediction item.

[0039] On the other hand, the present invention also provides an online simulation system for wear test of electromechanical products, the system comprising:

[0040] The data acquisition module is used to collect historical multi-dimensional monitoring data during the operation of electromechanical products from the historical database;

[0041] Model building module, used to build initial mechatronic product wear prediction model based on Bayesian network and convolutional neural network;

[0042] A training and verification module, used to train and test the initial electromechanical product wear prediction model through the historical multi-dimensional monitoring data to obtain an online prediction model for electromechanical product wear;

[0043] The wear prediction module is used to collect multi-dimensional monitoring data of electromechanical products during their operation in real time, input the multi-dimensional monitoring data into an online wear prediction model for electromechanical products, and output wear prediction results for electromechanical products.

[0044] Compared with the prior art, the online simulation method and system for wear test of electromechanical products of the present invention have the following advantages:

[0045] Beneficial effects:

[0046] (1) By collecting historical multi-dimensional monitoring data, combining Bayesian network and convolutional neural network to establish an initial wear prediction model for electromechanical products, and realizing online wear prediction after training and testing the model, it can achieve real-time monitoring and prediction of the wear of electromechanical products, thereby improving the accuracy and efficiency of wear testing;

[0047] (2) By segmenting the characteristics of multi-dimensional monitoring data, we can fully understand the operating conditions and historical failure conditions of electromechanical products, provide a more comprehensive and accurate data basis for establishing the initial wear prediction model of electromechanical products, monitor the operating status, usage environment and historical failure conditions of electromechanical products, improve the accuracy and reliability of wear prediction, help to timely discover potential problems and take corresponding maintenance measures, and extend the service life of electromechanical products;

[0048] (3) An initial wear prediction model for electromechanical products is established. The input layer receives feature inputs. The feature extraction layer uses a convolutional neural network to extract time-frequency features. The feature learning layer uses a Bayesian network to learn the probabilistic relationship between variables. The feature fusion layer effectively fuses features at different levels. Finally, the output layer gives the prediction results of the wear status of the electromechanical products. Combining the advantages of Bayesian networks and convolutional neural networks, it can comprehensively capture the key features of the wear status of electromechanical products, improve the accuracy and reliability of the prediction, and thus achieve accurate online wear prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 It is a flow chart of the online simulation method for wear test of electromechanical products of the present invention;

[0051] Figure 2 A network structure diagram of the initial electromechanical product wear prediction model of the present invention;

[0052] Figure 3 It is a composition network structure diagram of the feature extraction layer of the present invention;

[0053] Figure 4 It is a composition network structure diagram of the feature learning layer of the present invention;

[0054] Figure 5 This is a structural diagram of the online simulation system for wear testing of electromechanical products of the present invention. DETAILED DESCRIPTION

[0055] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] See also Figure 1 , provides an online simulation method for wear test of electromechanical products, comprising the following steps:

[0057] S1, collects historical multi-dimensional monitoring data of electromechanical products during operation from the historical database;

[0058] S2, establish the initial wear prediction model of electromechanical products based on Bayesian network and convolutional neural network;

[0059] S3, training and testing the initial electromechanical product wear prediction model through the historical multi-dimensional monitoring data to obtain an online electromechanical product wear prediction model;

[0060] S4, collecting multi-dimensional monitoring data of the electromechanical product during its operation in real time, inputting the multi-dimensional monitoring data into an online wear prediction model for the electromechanical product, and outputting a wear prediction result for the electromechanical product.

[0061] Specifically, this embodiment uses a combination of Bayesian network and convolutional neural network to establish an initial prediction model, which fully utilizes the advantages of the two models and improves prediction efficiency and accuracy.

[0062] The initial model was trained and tested using historical multi-dimensional monitoring data, and an online prediction model was obtained that can predict the wear status of electromechanical products in real time.

[0063] By collecting the operating data of electromechanical products in real time, inputting the online prediction model, and outputting the wear prediction results of the electromechanical products, it can help to detect the wear of the electromechanical products in advance, provide important reference for maintenance and care, and improve the reliability and service life of the electromechanical products.

[0064] Step S1 includes:

[0065] The historical multi-dimensional monitoring data includes operating status monitoring characteristics, usage history and environmental characteristics, and fault history characteristics;

[0066] The operating status monitoring characteristics include vibration data characteristics, temperature data characteristics, current data characteristics, and power data characteristics;

[0067] The usage history and environmental characteristics include usage duration characteristics, workload characteristics, environmental temperature characteristics, and maintenance record characteristics;

[0068] The fault history characteristics include the time of occurrence of each fault, the type of fault, and the cause of the fault.

[0069] Specifically, step S1 comprehensively collects multi-dimensional monitoring data during the operation of electromechanical products, including operating status characteristics, usage history and environmental characteristics, fault history characteristics, etc., providing a rich data foundation for the subsequent establishment of a prediction model.

[0070] Step S1 collects operating status monitoring characteristics such as vibration, temperature, current, power, as well as usage history and environmental characteristics such as usage time, workload, ambient temperature, maintenance records, and fault history characteristics such as fault occurrence time, fault type and cause, comprehensively reflecting the operating status of the electromechanical product.

[0071] By collecting these multi-dimensional monitoring data, the wear status of electromechanical products can be analyzed and predicted more accurately. Step S1 makes full use of various monitoring data during the operation of electromechanical products, lays the foundation for accurate wear prediction, and improves the accuracy and reliability of the prediction model.

[0072] The feature segmentation of the multi-dimensional monitoring data included in step S1 helps to fully understand the operation status and historical failure conditions of electromechanical products, thereby providing a more comprehensive and accurate data basis for establishing a wear prediction model. Through the collection and analysis of these features, the operation status, use environment and historical failure conditions of electromechanical products can be better monitored, thereby improving the accuracy and reliability of wear prediction, helping to timely discover potential problems and take corresponding maintenance measures, and extending the service life of electromechanical products.

[0073] Step S2 includes:

[0074] See also Figure 2 The initial electromechanical product wear prediction model includes an input layer, a feature extraction layer, a feature learning layer, a feature fusion layer, and an output layer. The input layer is connected to the feature extraction layer and the feature learning layer respectively. The feature learning layer and the feature extraction layer are connected to the feature fusion layer respectively. The feature fusion layer is connected to the output layer.

[0075] The input layer is used to receive input features, the feature extraction layer is used to extract time-frequency features of the input data, the feature learning layer is used to learn the probabilistic relationship features between variables in the input data, the feature fusion layer is used to fuse the output of the feature extraction layer and the output of the feature learning layer, and the output layer is used to receive the output of the feature fusion layer and output the prediction result of the model.

[0076] Specifically, step S2 adopts a composite network structure including an input layer, a feature extraction layer, a feature learning layer, a feature fusion layer and an output layer, which makes full use of the feature extraction and learning capabilities of different levels and improves the prediction performance of the model.

[0077] The feature extraction layer adopts a convolutional neural network structure, which can effectively extract the time-frequency characteristics of the input data and capture the dynamic changes of the wear state. The feature learning layer adopts a Bayesian network structure, which can learn the probabilistic relationship characteristics between variables in the input data and enhance the interpretability of the model. The feature fusion layer combines the outputs of the feature extraction layer and the feature learning layer, making full use of the advantages of the two features. This composite network structure can effectively integrate different types of features and improve the accuracy and robustness of wear prediction of electromechanical products.

[0078] Step S2 establishes a composite neural network model with clear hierarchy and smooth information transmission. The model receives feature input through the input layer, the feature extraction layer uses convolutional neural network to extract time-frequency features, the feature learning layer uses Bayesian network to learn the probability relationship between variables, and the feature fusion layer effectively fuses features at different levels. Finally, the output layer gives the prediction results of the wear state of electromechanical products. The model design gives full play to the advantages of various neural networks, can fully capture the key features of the wear state of electromechanical products, improve the accuracy and reliability of prediction, and lay the foundation for accurate online wear prediction.

[0079] See also Figure 3 , the feature extraction layer adopts a convolutional neural network structure, including a convolutional layer and a ReLU activation layer, and the feature extraction function expression of the feature extraction layer is:

[0080] H t =ReLU(W t X+α);

[0081] Among them, X is the input feature of the input layer, H t is the time-frequency feature of X, W t is the weight of the time-frequency feature corresponding to the wear state, and α is the feature extraction bias term.

[0082] Specifically, the feature extraction layer adopts a convolutional neural network structure, which can effectively extract the time-frequency characteristics of the input data and capture the dynamic change characteristics of the wear state of electromechanical products.

[0083] Through the combination of convolutional layers and ReLU activation layers, the time-frequency features related to the wear state can be automatically learned, which improves the pertinence and robustness of feature extraction.

[0084] The weight W of the time-frequency feature in the feature extraction function tAnd the bias term α can adaptively adjust the extracted time-frequency features, select different variables in the input data based on different consideration scales, and further optimize the effect of feature extraction.

[0085] This feature extraction layer can efficiently extract key time-frequency features related to the wear status of electromechanical products from the original input data.

[0086] See also Figure 4 , the feature learning layer adopts a Bayesian network structure, including a first fully connected layer, a second fully connected layer and a batch normalization layer, and the probability relationship feature function expression of the feature learning layer is:

[0087]

[0088] H x =Sigmoid(ω2ReLU(ω1X+τ1)+τ2);

[0089] Among them, X is the input feature of the input layer, H x is the initial probability feature, Z x To improve the stability and generalization performance of the probability relationship feature of X, β is the learnable scaling parameter of the batch normalization layer, is the offset parameter of the batch normalization layer, μ is the preset mean of the batch normalization layer, and λ 2 is the preset variance of the batch normalization layer, ε is the anti-zero constant, ω1 and ω2 are the weights of the first and second fully connected layers, τ1 and τ2 are the bias terms of the first and second fully connected layers, respectively.

[0090] Specifically, the feature learning layer adopts a Bayesian network structure, which can effectively learn the probabilistic relationship characteristics between variables in the input data and enhance the interpretability of the model.

[0091] The feature learning layer, through the combination of the first fully connected layer and the second fully connected layer, can perform deep nonlinear transformation on the initial probability features, thereby improving the expressive power of feature learning.

[0092] By introducing the batch normalization layer, the feature learning layer improves the stability and generalization performance of the model and avoids overfitting. By introducing the learnable scaling parameter β and offset parameter of the batch normalization layer And the preset mean μ and variance λ 2 , can adaptively adjust the distribution of features, adjust the above parameters according to the selection of different variables, further optimize the effect of feature learning, and the setting of anti-zero constants helps to avoid the problem of gradient disappearance or explosion, thereby improving the training stability of the model.

[0093] This feature learning layer can effectively learn the probabilistic relationship features related to the wear state of electromechanical products from the original input data, and is used to predict the probabilistic relationship features of different wear states of electromechanical products.

[0094] The feature learning layer adopts a Bayesian network structure and a multi-layer fully connected structure, combined with batch normalization and parameter setting, which can effectively learn and extract the probabilistic relationship characteristics of the input data, improving the stability, generalization ability and performance of the model.

[0095] The feature fusion function expression of the feature fusion layer is:

[0096] H r =Attention(γH t +(1-γ)Z x );

[0097] Among them, H r is the fusion feature, Attention(·) is the attention mechanism, and γ is the learning weight of the feature extraction layer.

[0098] Specifically, the feature fusion layer uses the attention mechanism to fuse the outputs of the feature extraction layer and the feature learning layer, which can adaptively adjust the importance weights between different features and improve the pertinence of feature fusion.

[0099] By fusing the features H r The calculation of the time-frequency features extracted by the feature extraction layer and the probability relationship features learned by the feature learning layer are effectively integrated, making full use of the complementary information of features at different levels and enhancing the model's ability to understand the wear status of electromechanical products.

[0100] Through the learning weight γ of the feature fusion layer, the importance of the output of the feature extraction layer and the feature learning layer can be dynamically adjusted according to the characteristics of different input data.

[0101] Through the attention mechanism, the learning weights of the feature extraction layer can be effectively learned, so that the model can pay more attention to important features, improve the model's ability to capture key features, and enhance the model's representation ability.

[0102] This feature fusion layer can organically fuse multi-source heterogeneous features together, providing a richer and more effective feature representation for the subsequent output layer, which helps to improve the overall performance of the wear prediction model for electromechanical products.

[0103] The feature fusion layer uses the attention mechanism to adaptively fuse features at different levels, making full use of the supplementary information of multi-source features and enhancing the model's ability to understand and express the wear status of electromechanical products.

[0104] The prediction result function expression of the output layer is:

[0105] Y = Sigmoid(η[H r ,X prior ]+δ);

[0106] Among them, Y is the prediction result, η is the weight matrix of the output layer, and H r is the fusion feature, X prior is the prior information, [H r ,X prior ] is the concatenation of fusion features and prior information, and δ is the bias vector of the output layer.

[0107] Specifically, the output layer is composed of the fusion features H output by the feature fusion layer r With external prior information X prior The splicing makes full use of the feature information learned inside the model and the expert experience knowledge introduced by the Bayesian network as external domain knowledge, enhancing the reliability and pertinence of the prediction results.

[0108] By learning the weight matrix η and the bias vector δ, end-to-end mapping from fusion features to the final prediction result Y is achieved, which simplifies the model structure and improves the prediction efficiency.

[0109] The output layer uses the Sigmoid activation function, which can map the predicted value to a probability output between 0 and 1, providing an intuitive probabilistic basis for subsequent decision-making and enhancing the interpretability of the model.

[0110] The output layer makes full use of internal features and external prior knowledge, which can better summarize and generalize to new electromechanical product wear scenarios and improve the generalization performance of the model.

[0111] The output layer takes into account the advantages of the feature fusion layer and prior knowledge, realizes end-to-end mapping from features to predicted outputs, generates interpretable probabilistic outputs, and improves the overall performance and generalization ability of the model.

[0112] Step S3 includes:

[0113] S31, divide the historical multi-dimensional monitoring data into a training set and a test set in a ratio of 7:3;

[0114] S32, iteratively training the initial electromechanical product wear prediction model through the training set, configuring the optimizer, loss function and hyperparameters until the iteration stop condition is reached, thereby obtaining the trained initial electromechanical product wear prediction model, wherein the iteration stop condition includes: the number of iterations reaches a preset value and the loss function converges;

[0115] S33, use the test set to test the initial electromechanical product wear prediction model after training, and evaluate the model through the comprehensive accuracy evaluation value. If the model evaluation meets the requirements, the training and testing are completed, and the electromechanical product wear online prediction model is obtained. If the model evaluation does not meet the requirements, return to step S32 to adjust the hyperparameters and re-train the model iteratively.

[0116] Specifically, step S3 divides the historical multi-dimensional monitoring data into a training set and a test set in a ratio of 7:3, which not only ensures the data volume of the training set, but also leaves enough test sets for model evaluation, ensuring the scientific nature of training and testing.

[0117] The initial electromechanical product wear prediction model is iteratively trained through the training set, and the optimizer, loss function and hyperparameters are reasonably configured until the preset iteration stop condition is reached, ensuring the convergence and optimization of the model on the training set and improving the model's fitting ability.

[0118] The test set is used to test the initial electromechanical product wear prediction model after training, and the model is comprehensively evaluated through the comprehensive accuracy evaluation value, which ensures the generalization performance of the model on new data and avoids overfitting.

[0119] In this embodiment, the loss function may adopt a cross entropy loss function, which may measure the difference between the predicted probability distribution and the true probability distribution and is suitable for classification-type prediction tasks.

[0120] The hyperparameters include the optimizer, the number of samples trained in each iteration, the number of iterations, etc.

[0121] Step S3 ensures the performance of the wear prediction model for electromechanical products on the training set and test set through reasonable data set division, effective model training, comprehensive model evaluation, and iterative optimization, improves the model's fitting ability and generalization performance, and provides a reliable foundation for subsequent online prediction.

[0122] The calculation formula of the comprehensive accuracy evaluation value is:

[0123]

[0124] Among them, K is the comprehensive accuracy evaluation value, ω i is the weight of the i-th prediction item of the wear prediction result, p i is the accuracy of the i-th prediction item of the wear prediction result, and n is the number of prediction items of the wear prediction result;

[0125] The wear prediction result is a multi-dimensional feature vector, and each dimension corresponds to a prediction item.

[0126] Specifically, in this embodiment, the comprehensive accuracy evaluation value K is calculated by weighted average, which reflects the comprehensive consideration of each prediction item, rather than evaluating a single prediction indicator. This multi-dimensional comprehensive evaluation method is closer to the needs of actual application scenarios.

[0127] By introducing the prediction item weight coefficient ω i , which can reflect the importance of different prediction items in the overall prediction results. According to actual application requirements, the weights of key prediction items can be increased in a targeted manner to ensure the prediction accuracy of key indicators.

[0128] The comprehensive accuracy evaluation value K not only takes into account the accuracy of each prediction item p i , and also incorporates the importance weight of the prediction item ω i This comprehensive evaluation method can more objectively reflect the comprehensive performance of the model on the training set and test set, and provide a basis for model evaluation.

[0129] Compared with a single accuracy rate or loss function, the comprehensive accuracy evaluation value K has stronger interpretability. It intuitively reflects the comprehensive performance of the model in various prediction dimensions and helps to analyze the advantages and disadvantages of the model.

[0130] The comprehensive accuracy evaluation described fully considers the multi-dimensional characteristics of the wear prediction task of electromechanical products, reflects the differences in the importance of different prediction items, achieves a comprehensive evaluation of the model performance, and enhances the interpretability of the model optimization.

[0131] In a feasible embodiment, step S4 includes the following steps:

[0132] At the beginning of step S4, the system will collect multi-dimensional monitoring data in real time during the operation of electromechanical products. These data may include different types of monitoring information such as vibration data, temperature data, current data, power data, as well as related information such as usage history, environmental characteristics and maintenance records.

[0133] (1) Data preprocessing:

[0134] The collected raw data needs to be preprocessed, including data cleaning, feature extraction, feature conversion and other steps. The raw data is converted into a format acceptable to the model for subsequent predictive analysis.

[0135] (2) Input to the wear prediction model:

[0136] The preprocessed data is input into the online prediction model for wear of electromechanical products. The model is the online prediction model for wear of electromechanical products trained in step S3, which combines the Bayesian network and the convolutional neural network.

[0137] (3) Model prediction:

[0138] After receiving the input data, the model performs wear prediction calculations. The model predicts the wear status of the electromechanical product based on the input multi-dimensional monitoring data of the electromechanical product. The prediction result is a multi-dimensional feature vector, where each dimension corresponds to a prediction item of the wear status of the electromechanical product.

[0139] (4) Output prediction results:

[0140] Finally, the model will output the wear prediction results of electromechanical products. This result can be recorded, displayed on the monitoring interface, or trigger further maintenance measures. The accuracy and real-time nature of the prediction results will have an important impact on subsequent maintenance and decision-making.

[0141] See also Figure 5 Based on the above method embodiments, the present invention also provides an online simulation system for wear test of electromechanical products, the system comprising:

[0142] The data acquisition module is used to collect historical multi-dimensional monitoring data during the operation of electromechanical products from the historical database;

[0143] Model building module, used to build initial mechatronic product wear prediction model based on Bayesian network and convolutional neural network;

[0144] A training and verification module, used to train and test the initial electromechanical product wear prediction model through the historical multi-dimensional monitoring data to obtain an online prediction model for electromechanical product wear;

[0145] The wear prediction module is used to collect multi-dimensional monitoring data of electromechanical products during their operation in real time, input the multi-dimensional monitoring data into an online wear prediction model for electromechanical products, and output wear prediction results for electromechanical products.

[0146] The above system embodiments and method embodiments correspond one to one, and for a brief description of the system embodiments, please refer to the method embodiments.

[0147] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An online simulation method for wear test of electromechanical products, characterized in that: The following steps are involved: S1, collects historical multi-dimensional monitoring data of electromechanical products during operation from the historical database; S2, establish the initial wear prediction model of electromechanical products based on Bayesian network and convolutional neural network; S3, training and testing the initial electromechanical product wear prediction model through the historical multi-dimensional monitoring data to obtain an online electromechanical product wear prediction model; S4, collecting multi-dimensional monitoring data of the electromechanical product during its operation in real time, inputting the multi-dimensional monitoring data into an online wear prediction model for the electromechanical product, and outputting a wear prediction result for the electromechanical product.

2. The online simulation method for wear test of electromechanical products according to claim 1, characterized in that: Step S1 includes: The historical multi-dimensional monitoring data includes operating status monitoring characteristics, usage history and environmental characteristics, and fault history characteristics; The operating status monitoring characteristics include vibration data characteristics, temperature data characteristics, current data characteristics, and power data characteristics; The usage history and environmental characteristics include usage duration characteristics, workload characteristics, environmental temperature characteristics, and maintenance record characteristics; The fault history characteristics include the time of occurrence of each fault, the type of fault, and the cause of the fault.

3. The online simulation method for wear test of electromechanical products according to claim 2, characterized in that: Step S2 includes: The initial electromechanical product wear prediction model includes an input layer, a feature extraction layer, a feature learning layer, a feature fusion layer, and an output layer. The input layer is connected to the feature extraction layer and the feature learning layer respectively, the feature learning layer and the feature extraction layer are connected to the feature fusion layer respectively, and the feature fusion layer is connected to the output layer; The input layer is used to receive input features, the feature extraction layer is used to extract time-frequency features of the input data, the feature learning layer is used to learn the probabilistic relationship features between variables in the input data, the feature fusion layer is used to fuse the output of the feature extraction layer and the output of the feature learning layer, and the output layer is used to receive the output of the feature fusion layer and output the prediction result of the model.

4. The online simulation method for wear test of electromechanical products according to claim 3, characterized in that: The feature extraction layer adopts a convolutional neural network structure, including a convolutional layer and a ReLU activation layer. The feature extraction function expression of the feature extraction layer is: H t =ReLU(W t X+α); Among them, X is the input feature of the input layer, H t is the time-frequency feature of X, W t is the weight of the time-frequency feature corresponding to the wear state, and α is the feature extraction bias term.

5. The online simulation method for wear test of electromechanical products according to claim 4, characterized in that: The feature learning layer adopts a Bayesian network structure, including a first fully connected layer, a second fully connected layer and a batch normalization layer. The probability relationship feature function expression of the feature learning layer is: H x =Sigmoid(ω2ReLU(ω1X+τ1)+τ2); Among them, X is the input feature of the input layer, H x is the initial probability feature, Z x To improve the stability and generalization performance of the probability relationship feature of X, β is the learnable scaling parameter of the batch normalization layer, is the offset parameter of the batch normalization layer, μ is the preset mean of the batch normalization layer, and λ 2 is the preset variance of the batch normalization layer, ε is the anti-zero constant, ω1 and ω2 are the weights of the first and second fully connected layers, τ1 and τ2 are the bias terms of the first and second fully connected layers, respectively.

6. The online simulation method for wear test of electromechanical products according to claim 5, characterized in that: The feature fusion function expression of the feature fusion layer is: H r =Attention(γH t +(1-γ)Z x ); Among them, H r is the fusion feature, Attention(·) is the attention mechanism, and γ is the learning weight of the feature extraction layer.

7. The online simulation method for wear test of electromechanical products according to claim 6, characterized in that: The prediction result function expression of the output layer is: Y=Sigmoid(η[H r ,X prior ]+d); Among them, Y is the prediction result, η is the weight matrix of the output layer, and H r is the fusion feature, X prior is the prior information, [H r ,X prior ] is the concatenation of fusion features and prior information, and δ is the bias vector of the output layer.

8. The online simulation method for wear test of electromechanical products according to claim 4, characterized in that: Step S3 includes: S31, divide the historical multi-dimensional monitoring data into a training set and a test set in a ratio of 7:3; S32, iteratively training the initial electromechanical product wear prediction model through the training set, configuring the optimizer, loss function and hyperparameters until the iteration stop condition is reached, thereby obtaining the trained initial electromechanical product wear prediction model, wherein the iteration stop condition includes: the number of iterations reaches a preset value and the loss function converges; S33, use the test set to test the initial electromechanical product wear prediction model after training, and evaluate the model through the comprehensive accuracy evaluation value. If the model evaluation meets the requirements, the training and testing are completed, and the electromechanical product wear online prediction model is obtained. If the model evaluation does not meet the requirements, return to step S32 to adjust the hyperparameters and re-train the model iteratively.

9. The online simulation method for wear test of electromechanical products according to claim 8, characterized in that: The calculation formula of the comprehensive accuracy evaluation value is: Among them, K is the comprehensive accuracy evaluation value, ω i is the weight of the i-th prediction item of the wear prediction result, p i is the accuracy of the i-th prediction item of the wear prediction result, and n is the number of prediction items of the wear prediction result; The wear prediction result is a multi-dimensional feature vector, and each dimension corresponds to a prediction item.

10. The online simulation system for wear test of electromechanical products is characterized by: The system comprises: The data acquisition module is used to collect historical multi-dimensional monitoring data during the operation of electromechanical products from the historical database; Model building module, used to build initial mechatronic product wear prediction model based on Bayesian network and convolutional neural network; A training and verification module, used to train and test the initial electromechanical product wear prediction model through the historical multi-dimensional monitoring data to obtain an online prediction model for electromechanical product wear; The wear prediction module is used to collect multi-dimensional monitoring data of electromechanical products during their operation in real time, input the multi-dimensional monitoring data into an online wear prediction model for electromechanical products, and output wear prediction results for electromechanical products.

Images