Production equipment intelligent dynamic maintenance method based on prediction model
By applying CNN and SVR prediction models in equipment maintenance, monitoring the operating status of the equipment in real time and predicting RUL, the problem of premature or too late maintenance in traditional fixed-cycle maintenance methods is solved, and more efficient equipment maintenance and reduced maintenance costs are achieved.
Patent Information
- Application Number
- CN202411862086.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-17
- Publication Date
- 2025-04-11
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional fixed-cycle maintenance methods cannot flexibly adjust the maintenance timing according to the actual health of the equipment, resulting in premature or too late maintenance, resulting in waste of resources, equipment failure and increased maintenance costs.
The prediction model based on CNN and SVR is adopted to monitor and analyze the operating status of the equipment in real time, predict the remaining service life of the equipment (RUL), and dynamically adjust the maintenance plan according to the actual status of the equipment.
By dynamically adjusting the maintenance plan, the waste of resources, equipment failures and increased maintenance costs caused by premature or late maintenance in traditional fixed maintenance are avoided, and the overall efficiency of equipment and maintenance accuracy are improved.
Smart Images
Figure CN119809601A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of equipment maintenance, and in particular to an intelligent dynamic maintenance method for production equipment based on a prediction model. Background Art
[0002] Current equipment maintenance technology mainly relies on fixed-cycle maintenance strategies, which usually set maintenance schedules based on equipment operating hours, production volume, or usage times. Although this type of maintenance method is easy to implement, it ignores the actual operating conditions of the equipment and cannot flexibly adjust the maintenance timing according to the health status of the equipment. This leads to the following problems:
[0003] 1. Premature maintenance: In the maintenance plan, the equipment is maintained after reaching a fixed maintenance cycle, rather than the current health status of the equipment. The equipment may still be in good operating condition and can continue to operate, but because the maintenance cycle has expired, the company must perform maintenance. This leads to unnecessary waste of time and resources, reduces the overall efficiency of equipment use, and increases maintenance costs.
[0004] 2. Late maintenance: In some cases, equipment may fail prematurely during a fixed maintenance cycle, resulting in unexpected equipment downtime and production interruption. Due to the lack of real-time equipment status monitoring and prediction, companies are unable to identify potential equipment failure risks in a timely manner, thereby increasing maintenance costs and losses caused by downtime.
[0005] 3. Lack of real-time monitoring and prediction capabilities: Fixed maintenance methods do not take into account the real-time operating status of the equipment and cannot predict and monitor the health of the equipment. This means that companies cannot understand potential problems with the equipment in a timely manner, nor can they foresee the remaining life (RUL) of the equipment. Equipment failures often occur suddenly, and maintenance time and resources cannot be planned in advance, further increasing maintenance and potential risks and costs.
[0006] 4. Waste of resources and increased operating costs: The traditional fixed maintenance model easily leads to waste of maintenance resources. Too early maintenance will lead to reduced equipment efficiency and increase unnecessary maintenance costs; while too late maintenance increases the risk of sudden failures, which may lead to more expensive emergency repairs and production stoppages. Whether it is too early or too late maintenance, it will have an adverse impact on the company's production plan and operating costs. Summary of the invention
[0007] The purpose of the present invention is to provide an intelligent dynamic maintenance method for production equipment based on a predictive model. The operating status of the equipment is monitored and analyzed in real time through a CNN model and an SVR model, and the RUL of the equipment is predicted. The maintenance plan is dynamically adjusted according to the actual status of the equipment to avoid problems such as resource waste, equipment failure, and increased maintenance costs caused by premature or late maintenance in traditional fixed maintenance.
[0008] To achieve the above object, the present invention provides a method for intelligent dynamic maintenance of production equipment based on a prediction model, comprising the following steps:
[0009] S1. Equipment data collection, processing and conversion;
[0010] S2, build a CNN-SVR prediction model to predict the remaining service life of the equipment;
[0011] S3. Build a production scheduling model. The production scheduling model and CNN-SVR prediction model are controlled by the system. The production scheduling model flexibly adjusts the production plan according to the health status of the equipment.
[0012] Preferably, in S1, the number of printings, the number of maintenances, the solder paste thickness, and the remaining service life RUL corresponding to the solder paste thickness are collected from the historical data of the equipment operation; after the data collection is completed, each data is normalized so that it is in the same numerical range, and the normalization formula is as follows:
[0013]
[0014] Among them, X normalized is the normalized value, X is the original data, which represents the eigenvalue to be normalized, X max is the maximum value in the data set, X min is the minimum value in the data set;
[0015] The number of printings and maintenance times are used as input features and converted into a two-dimensional matrix using the multiplication method. The formula of the multiplication method is as follows:
[0016] f = [f1, f2];
[0017]
[0018] Among them, f is a feature vector, which represents the set of printing times and maintenance times, f1 is the printing times, f2 is the maintenance times, and F m is a two-dimensional matrix generated by the multiplication method.
[0019] Preferably, S2 specifically includes the following steps:
[0020] S2.1. Construct a convolutional neural network (CNN) model, input the two-dimensional matrix obtained in S1, and store the two-dimensional matrix in a four-dimensional array in the format of [2,2,1,N]. In the four-dimensional array, 2 represents the spatial dimension of the matrix; 1 represents the number of channels, indicating a single-channel image; and N represents the number of samples. The stored four-dimensional array can be directly input into the CNN model, and key features are extracted through convolution and pooling operations to predict the solder paste thickness.
[0021] S2.2, construct a support vector regression SVR model, and use the solder paste thickness prediction results obtained in S2.1 to further predict the RUL of the device;
[0022] S2.3. Train the model and optimize the hyperparameters of the CNN model and SVR model through Bayesian optimization.
[0023] Preferably, S2.2 specifically includes the following steps:
[0024] S2.2.1, through the polynomial kernel function, the SVR model maps the input solder paste thickness to the high-order space;
[0025] S2.2.2, the SVR model searches for the optimal hyperplane in the high-level space, which can minimize the error between the solder paste thickness and the actual RUL of the device;
[0026] S2.2.3. Through regression fitting, the SVR model generates the RUL prediction value of the equipment.
[0027] Preferably, the Bayesian optimization of hyperparameters in S2.3 specifically includes the following steps:
[0028] S2.3.1. Update the proxy model and update the Gaussian process model according to the currently evaluated hyperparameter combination;
[0029] S2.3.2. Select a new evaluation point and rotate the next hyperparameter combination to be evaluated by maximizing the acquisition function.
[0030] S2.3.3. Evaluate the objective function, train the model, and calculate the objective function value;
[0031] S2.3.4. Update the optimal solution. If the result of the new evaluation point is better than the current optimal solution, update the optimal solution.
[0032] Preferably, S3 specifically includes the following steps:
[0033] S3.1, the production scheduling model dynamically adjusts the production load of the equipment according to the equipment RUL predicted by the CNN-SVR prediction model;
[0034] S3.2. The system dynamically makes maintenance and replacement decisions based on the predicted equipment RUL.
[0035] Therefore, the present invention adopts the above-mentioned intelligent dynamic maintenance method for production equipment based on the prediction model, monitors and analyzes the operating status of the equipment in real time through the CNN model and the SVR model, predicts the RUL of the equipment, and dynamically adjusts the maintenance plan according to the actual status of the equipment, avoiding the problems of resource waste, equipment failure and increased maintenance costs caused by premature or late maintenance in traditional fixed maintenance.
[0036] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 is a flow chart of a method for intelligent dynamic maintenance of production equipment based on a prediction model of the present invention;
[0038] Figure 2 It is a cycle-RUL diagram of machine 1 of the intelligent dynamic maintenance method for production equipment based on the prediction model of the present invention;
[0039] Figure 3 It is a cycle-RUL diagram of machine 2 of the intelligent dynamic maintenance method for production equipment based on the prediction model of the present invention. DETAILED DESCRIPTION
[0040] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.
[0041] Unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0042] Embodiment 1
[0043] like Figure 1 As shown, the present invention provides a method for intelligent dynamic maintenance of production equipment based on a prediction model, comprising the following steps:
[0044] S1. Equipment data collection, processing and conversion;
[0045] Collect key parameters from the historical data of equipment operation, including:
[0046] Printing times: the number of times the device is printed;
[0047] Maintenance times: equipment maintenance records;
[0048] Solder paste thickness: a key physical property during equipment operation;
[0049] Remaining service life RUL corresponding to solder paste thickness;
[0050] After the data collection is completed, normalization is performed to make each data in the same numerical range. The normalization formula is as follows:
[0051]
[0052] Among them, X normalized is the normalized value, X is the original data, which represents the eigenvalue to be normalized, and X max is the maximum value in the data set, X min is the minimum value in the data set;
[0053] The number of printings and maintenance times are used as input features and converted into a two-dimensional matrix using the multiplication method. The formula of the multiplication method is as follows:
[0054] f = [f1, f2];
[0055]
[0056] Among them, f is a feature vector, which represents the set of printing times and maintenance times, f1 is the printing times, f2 is the maintenance times, and F m is a two-dimensional matrix generated by the multiplication method.
[0057] S2. Construct a CNN-SVR prediction model to predict the remaining service life of the equipment, which specifically includes the following steps:
[0058] S2.1. Construct a convolutional neural network (CNN) model, input the two-dimensional matrix obtained in S1, and store the two-dimensional matrix in a four-dimensional array in the format of [2,2,1,N]. In the four-dimensional array, 2 represents the spatial dimension of the matrix; 1 represents the number of channels, indicating a single-channel image; and N represents the number of samples. The stored four-dimensional array can be directly input into the CNN model, and key features can be extracted through convolution and pooling operations to predict the solder paste thickness.
[0059] CNN is a deep learning model that is particularly suitable for processing data with spatial structure. The core idea of the present invention is to construct the number of printing and maintenance times in the operation of the equipment into a two-dimensional matrix, and then convert the two-dimensional matrix into a four-dimensional array and input it into the CNN model. Through operations such as convolution and pooling, key features are extracted to predict the thickness of solder paste. The following is the structure of the CNN model.
[0060] Input layer (imageInputLayer): The input data is the normalized device parameters, including the number of printings and maintenance times. A two-dimensional matrix is constructed by multiplication method, and the two-dimensional matrix is converted into a four-dimensional array as input.
[0061] Convolutional layer (convolution2dLayer): The convolutional layer is the core component of CNN and is used to extract local features from the input data. Each convolutional layer scans the input image through convolution kernels (also called filters) of different sizes and performs convolution operations on local areas of the data to extract local features. The convolutional layer can identify the correlation between the number of prints and the number of maintenance times, and deeply explore the complex patterns in the operation of the equipment through multiple layers of convolution. For example, the changing trend of the number of prints under a certain number of maintenance times can reflect the workload status of the equipment, and these statuses will directly affect the change in solder paste thickness.
[0062] Batch Normalization Layer: Its important function is to reduce the variation of input distribution between layers, normalize the input data of each layer of the neural network, make the data distribution more stable, thereby speeding up the training process and making the network less sensitive to the choice of initial weights.
[0063] ReLU layer (reluLayer): performs element-by-element nonlinear activation on the input data, sets all negative values to 0, keeps positive values unchanged, and increases the nonlinear ability of the model.
[0064] Pooling layer (maxpooling2dLayer): In order to reduce data dimension and computational complexity, a pooling layer is usually added after the convolution layer in the CNN model. The role of the pooling layer is to downsample the convolution result to retain the most important features and reduce redundant information. The max pooling layer is used to reduce the dimension of the local area of the feature map and retain the most significant feature values. The principle of max pooling is to select the largest value in a region to represent the main feature of the region. This method can effectively reduce the amount of calculation of the model while retaining the key change points in the operating status of the equipment. For example, the max pooling layer can highlight the significant changes in the number of printing and maintenance times of the equipment in a specific time period, thereby helping the model to better predict changes in solder paste thickness.
[0065] Dropout layer: used to set the output of the first part of neurons to 0 during training to reduce model overfitting.
[0066] Fully connected layer: The function of the fully connected layer is to convert the output of the pooling layer into a predicted value of solder paste thickness. The fully connected layer simulates the complex nonlinear relationship between device status and solder paste thickness through a series of weighted sum operations. This layer fuses the information from different convolutional layers and outputs the final prediction result based on the weights of the input features.
[0067] Regression Layer: Calculates the error between the output value and the target value, and the error result is used for back propagation to update the model weights.
[0068] S2.2, construct a support vector regression SVR model, and use the solder paste thickness prediction results obtained in S2.1 to further predict the RUL of the device;
[0069] The CNN model predicts the solder paste thickness of the equipment by processing the equipment's operating data (printing times and maintenance times), while the SVR model uses this solder paste thickness prediction result to further predict the RUL of the equipment. This combination fully utilizes the advantages of CNN in extracting data features and SVR's ability to handle nonlinear relationships, thereby providing a better reference for equipment maintenance and production scheduling.
[0070] The specific steps include:
[0071] S2.2.1. Through the polynomial kernel function, the SVR model maps the input solder paste thickness to a high-dimensional space; in this high-dimensional space, the nonlinear relationship between solder paste thickness and device RUL can be effectively captured and fitted.
[0072] S2.2.2. The SVR model searches for the optimal hyperplane in the high-order space. The optimal hyperplane can minimize the error between the solder paste thickness and the actual RUL of the device. Since SVR introduces error tolerance (ε), the model will ignore errors within a certain range during training, ensuring that the model is not affected by noise and has good generalization ability.
[0073] S2.2.3. Through regression fitting, the SVR model generates the RUL prediction value of the equipment. The RUL prediction value represents the time the equipment can continue to operate in the current state, providing a direct basis for equipment maintenance and replacement.
[0074] S2.3, training model, optimizing the hyperparameters of CNN model and SVR model through Bayesian optimization;
[0075] The hyperparameters for optimizing the CNN model mainly include the learning rate, the number of convolution kernels (filters) in the convolution layer, the size of the convolution kernel, the Dropout ratio, and the number of neurons in the fully connected layer. The hyperparameters for optimizing the SVR model mainly include the polynomial kernel function order, Box constraint, and Epsilon. The core idea of Bayesian optimization of hyperparameters is to build a proxy model (Gaussian process) to approximate the objective function, and use the model to select the point that is most likely to improve the hyperparameter combination, thereby effectively searching the hyperparameter space.
[0076] Bayesian optimization of hyperparameters specifically includes the following steps:
[0077] S2.3.1. Update the proxy model and update the Gaussian process model according to the currently evaluated hyperparameter combination;
[0078] S2.3.2. Select a new evaluation point and rotate the next hyperparameter combination to be evaluated by maximizing the acquisition function.
[0079] S2.3.3. Evaluate the objective function, train the model, and calculate the objective function value;
[0080] S2.3.4. Update the optimal solution. If the result of the new evaluation point is better than the current optimal solution, update the optimal solution.
[0081] After the Bayesian optimization of the CNN-SVR prediction model hyperparameters is completed, the CNN-SVR prediction model is trained, and the required equipment parameters (printing times, maintenance times, solder paste thickness, and the remaining service life corresponding to the solder paste thickness) are extracted from the historical data. The data is processed and converted through the normalization and multiplication methods described above, and finally used for CNN model training and SVR model fitting. After multiple trainings, the CNN model will finally obtain the best weights (trainable parameters), and the SVR model will find a function in high-dimensional space that can fit the data well.
[0082] When the model training is completed, the CNN-SVR prediction model is saved. The corresponding cumulative number of printings is obtained through the number of products produced in each cycle by the production scheduling model and fed back to the CNN-SVR prediction model. The CNN-SVR prediction model will then predict its corresponding remaining useful life (RUL) and take corresponding actions for the next cycle based on the predicted RUL.
[0083] S3. Build a production scheduling model. The production scheduling model and CNN-SVR prediction model are controlled by the system. The production scheduling model flexibly adjusts the production plan according to the health status of the equipment. Specifically, it includes the following steps:
[0084] S3.1, the production scheduling model dynamically adjusts the production load of the equipment according to the equipment RUL predicted by the CNN-SVR prediction model;
[0085] The production scheduling model dynamically adjusts the production load of the equipment according to the RUL of the equipment. The system will set different production limits based on the remaining life of the equipment:
[0086] When the RUL of the equipment is higher than 8, the equipment is in the best condition and the maximum production capacity is set to 500.
[0087] When RUL drops to between 6 and 8, the production capacity will be reduced and the maximum production capacity is set to 350.
[0088] If the RUL drops further to between 3.5 and 6, the equipment will continue to operate, but the production capacity will be further reduced, with the maximum production capacity set at 200.
[0089] When the RUL falls below 3.5, the system triggers maintenance actions to prevent equipment failure.
[0090] S3.2. The system dynamically makes maintenance and replacement decisions based on the predicted equipment RUL.
[0091] Dynamic maintenance: When the RUL approaches the preset critical value (below 3.5), the system will automatically schedule maintenance operations. After the maintenance is completed, the CNN-SVR prediction model will re-predict the RUL of the equipment, and the system will update the production capacity setting of the production scheduling model based on the predicted equipment RUL. During the maintenance period, the accumulated production load of the equipment will be reset to extend the service life of the equipment.
[0092] Equipment replacement: If the predicted RUL after the maintenance operation is still below the threshold (4) or the equipment is about to expire, the system will replace the equipment. After the equipment is replaced, the production scheduling model resets the equipment's RUL to the maximum value and restores the equipment's production capacity to the optimal state.
[0093] like Figure 2 and Figure 3 As shown, the simulation results of the intelligent dynamic maintenance method for production equipment based on the prediction model of the present invention on machine 1 and machine 2 are shown. Taking 40 production cycles as an example, machine 1 and machine 2 respectively produce two different products. Since the number of prints required for the production of each product is different for these two products, and the demand for the two products in each cycle is also different, the RUL predictions of machine 1 and machine 2 in each cycle are also different. Therefore, the maintenance cycles of machine 1 and machine 2 are also different. Although this embodiment only takes 40 cycles as an example and does not involve the machine replacement cycle, according to this trend, the replacement cycles of machine 1 and machine 2 will also be different.
[0094] In order to more directly illustrate the accuracy of the CNN-SVR prediction model of the present invention, the mean square error (MSE), mean absolute error (MAE) and root mean square error (RMSE) of the prediction results of the test set are calculated as evaluation indicators for comparing the CNN-SVR prediction model of the present invention with the Random Forest prediction model, the MLP prediction model, the SVR prediction model and the LSBoost prediction model. The calculation formula is as follows:
[0095]
[0096]
[0097] in, is the predicted value of the i-th sample; yi is the true value of the ith sample; N is the total number of samples. The results are shown in Table 1:
[0098] Table 1 Comparison of performance indicators of various prediction models
[0099]
[0100] As can be seen from Table 1, the three indicators of RMSE (root mean square error), MAE (mean absolute error), and MSE (mean square error) calculated by the prediction model of the present invention are the smallest among all models, indicating that the prediction model of the present invention is more accurate. The RMSE calculated by the prediction model of the present invention is 0.265053, which is 7.59% less than the RMSE obtained by the SVR prediction model. And this gap is more obvious in the MSE index. The MSE value calculated by the LSBoost prediction model is 0.19996645, and the MSE value of the CNN+SVR prediction model is only 0.0727154, which is 12.73% lower than the MSE obtained by the LSBoost prediction model. In MAE, the value calculated by the prediction model of the present invention is also the smallest compared to other models. The above results fully illustrate that the prediction model of the present invention is superior to other prediction models in all aspects. The specific role of this step: conduct a comparative experiment to show the accuracy of the proposed prediction model. Because the maintenance of production equipment is based on the remaining useful life (RUL) predicted by the prediction model, if the remaining useful life (RUL) predicted by the prediction model is more accurate, the timing of production equipment maintenance will be more accurate.
[0101] Therefore, the present invention adopts the above-mentioned intelligent dynamic maintenance method for production equipment based on the prediction model, monitors and analyzes the operating status of the equipment in real time through the CNN model and the SVR model, predicts the RUL of the equipment, and dynamically adjusts the maintenance plan according to the actual status of the equipment, avoiding the problems of resource waste, equipment failure and increased maintenance costs caused by premature or late maintenance in traditional fixed maintenance.
[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.
Claims
1. An intelligent dynamic maintenance method for production equipment based on a prediction model, characterized in that: The following steps are involved: S1. Equipment data collection, processing and conversion; S2, build a CNN-SVR prediction model to predict the remaining service life of the equipment; S3. Build a production scheduling model. The production scheduling model and CNN-SVR prediction model are controlled by the system. The production scheduling model flexibly adjusts the production plan according to the health status of the equipment.
2. The method for intelligent dynamic maintenance of production equipment based on a prediction model according to claim 1, characterized in that: In S1, the number of printings, the number of maintenances, the solder paste thickness, and the remaining service life RUL corresponding to the solder paste thickness are collected from the historical data of the equipment operation; after the data collection is completed, the data are normalized so that each data is in the same numerical range. The normalization formula is as follows: Among them, X normalized is the normalized value, X is the original data, and represents the eigenvalue to be normalized. max is the maximum value in the data set, X min is the minimum value in the data set; The number of printings and maintenance times are used as input features and converted into a two-dimensional matrix using the multiplication method. The formula of the multiplication method is as follows: f = [f1, f2]; Among them, f is a feature vector, which represents the set of printing times and maintenance times, f1 is the printing times, f2 is the maintenance times, and F m is a two-dimensional matrix generated by the multiplication method.
3. The intelligent dynamic maintenance method for production equipment based on the prediction model according to claim 2 is characterized in that: S2 specifically includes the following steps: S2.
1. Construct a convolutional neural network (CNN) model, input the two-dimensional matrix obtained in S1, and store the two-dimensional matrix in a four-dimensional array in the format of [2,2,1,N]. In the four-dimensional array, 2 represents the spatial dimension of the matrix; 1 represents the number of channels, indicating a single-channel image; and N represents the number of samples. The stored four-dimensional array can be directly input into the CNN model, and key features are extracted through convolution and pooling operations to predict the solder paste thickness. S2.2, construct a support vector regression SVR model, and use the solder paste thickness prediction results obtained in S2.1 to further predict the RUL of the device; S2.
3. Train the model and optimize the hyperparameters of the CNN model and SVR model through Bayesian optimization.
4. The method for intelligent dynamic maintenance of production equipment based on a prediction model according to claim 3 is characterized in that: S2.2 specifically includes the following steps: S2.2.1, through the polynomial kernel function, the SVR model maps the input solder paste thickness to the high-order space; S2.2.2, the SVR model searches for the optimal hyperplane in the high-level space, which can minimize the error between the solder paste thickness and the actual RUL of the device; S2.2.
3. Through regression fitting, the SVR model generates the RUL prediction value of the equipment.
5. The method for intelligent dynamic maintenance of production equipment based on a prediction model according to claim 3, characterized in that: The Bayesian optimization of hyperparameters in S2.3 specifically includes the following steps: S2.3.
1. Update the proxy model and update the Gaussian process model according to the currently evaluated hyperparameter combination; S2.3.
2. Select a new evaluation point and rotate the next hyperparameter combination to be evaluated by maximizing the acquisition function. S2.3.
3. Evaluate the objective function, train the model, and calculate the objective function value; S2.3.
4. Update the optimal solution. If the result of the new evaluation point is better than the current optimal solution, update the optimal solution.
6. The method for intelligent dynamic maintenance of production equipment based on a prediction model according to claim 1, characterized in that: S3 specifically includes the following steps: S3.1, the production scheduling model dynamically adjusts the production load of the equipment according to the equipment RUL predicted by the CNN-SVR prediction model; S3.
2. The system dynamically makes maintenance and replacement decisions based on the predicted equipment RUL.