Grease injection quantity prediction method based on multiple parameters

Abstract

The invention discloses a grease injection quantity prediction method based on multiple parameters, and the method comprises the following steps: 1, collecting multi-source data, and carrying out the preprocessing of the multi-source data; 2, constructing an initial feature matrix; mapping the high-dimensional features to a low-dimensional principal component space; the feature data with the highest correlation with the grease injection amount are screened out; 3, constructing a residual multi-head attention long-short-term memory network as a prediction model; 4, inputting the feature data screened in the step 2 into the trained prediction model to obtain the grease adding amount required by the bearing; 5, converting the predicted value into a specific control instruction; and step 6, continuously performing dynamic optimization and online learning. According to the invention, the bearing state, environment, structure and grease performance parameters are simultaneously incorporated into the grease injection quantity prediction model, so that the problem of large limitation of a single parameter model is solved; on-line learning and Bayesian optimization are introduced, so that the model can be adaptively adjusted along with the change of the equipment state.

Description

Technical Field

[0001] This invention relates to the field of equipment maintenance technology, and specifically to a method for predicting grease injection volume based on multiple parameters. Background Technology

[0002] In the field of mechanical equipment maintenance, bearing grease lubrication is a crucial step in ensuring the normal operation of equipment. Too much or too little grease can lead to equipment failure and shorten the service life of bearings.

[0003] Current methods for predicting grease volume mostly rely on single or limited parameters such as rotational speed and temperature, which cannot fully reflect the actual operating conditions of bearings. Traditional models are fixed once established and cannot adapt to dynamic factors such as equipment aging and environmental changes. Very few methods consider the impact of environmental parameters such as dust concentration and oxygen concentration in the air on the performance of lubricating grease. Methods based on simple regression or rules are difficult to capture the complex nonlinear relationships between parameters. Summary of the Invention

[0004] Therefore, the present invention provides a multi-parameter-based fat injection volume prediction method to solve the problems in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A multi-parameter-based method for predicting fat injection volume includes the following steps:

[0007] Step 1: Collect data from multiple sources and preprocess it;

[0008] Step 2: Align the preprocessed multi-source data according to timestamps to construct an initial feature matrix; use the kernel principal component analysis algorithm to perform nonlinear feature extraction on the initial feature matrix, mapping high-dimensional features to a low-dimensional principal component space; use the recursive feature elimination method combined with the support vector machine model to select the feature data with the highest correlation to the amount of fat injected.

[0009] Step 3: Construct a residual multi-head attention long short-term memory network as a prediction model. The prediction model includes an LSTM base layer, a multi-head attention mechanism, residual connections, and a fully connected output layer.

[0010] Step 4: Input the feature data selected in Step 2 into the trained prediction model to obtain the required amount of grease for the bearing; determine whether the predicted amount of grease is within the preset threshold range, and if it exceeds the range, adopt a rule-based safety strategy.

[0011] Step 5: Convert the predicted values ​​into specific control commands;

[0012] Step 6: The grease injection equipment performs grease injection upon receiving the control command. After the grease injection is executed, the operating status of the bearing is continuously monitored to evaluate the actual effect of the grease injection. These newly generated data pairs are labeled and added to the training dataset. Through regular incremental learning, the prediction model can absorb new knowledge, dynamically adjust its internal parameters, and achieve self-evolution and continuous improvement of its prediction capabilities.

[0013] Furthermore: Step 1 involves collecting multi-source data from four dimensions: bearing system parameters, environmental parameters, bearing specification parameters, and grease performance parameters. The collected raw data is then preprocessed, including outlier handling, smoothing filtering, and normalization.

[0014] Furthermore: the bearing condition parameters include bearing temperature, bearing speed, and vibration acceleration; environmental parameters include dust concentration and oxygen concentration in the air; bearing specification parameters include bearing inner diameter, outer diameter, and roller diameter; and grease performance parameters include grease dielectric constant, consistency, and dropping point.

[0015] Furthermore: The calculation method of the recursive feature elimination method is as follows: Let the initial feature set be S0, and iteratively delete the least important features until m features remain:

[0016] J(S k =CrossValAccuracy(SVM(S k ));

[0017] Among them, J(S) k ) represents the current feature subset S k Cross-validation accuracy; SVM(S k ) is based on S k The training support vector machine; termination condition: |Sk|=m, finally retaining the feature combination with the highest correlation to the amount of fat injected.

[0018] Furthermore, the prediction model is trained using the Huber loss function and the AdamW optimizer, and the learning rate and batch size hyperparameters are automatically optimized using a Bayesian optimization algorithm.

[0019] Furthermore: the Huber loss function is:

[0020] ;

[0021] Where y is the actual amount of fat injected; This is the predicted amount of grease to be injected; δ is the threshold value. This is the historical average.

[0022] The Bayesian optimization objective is: ;

[0023] Where θ is the hyperparameter vector; MAPE val To verify the mean absolute percentage error of the validation set, the optimal hyperparameters are efficiently searched using a Gaussian process surrogate model.

[0024] Furthermore, by utilizing newly collected operational data and actual grease injection effect data, incremental learning is performed on the prediction model to achieve dynamic optimization of the model.

[0025] Furthermore: In step 5, the theoretical grease injection time is obtained by dividing the predicted grease injection volume by the grease pump flow rate, and the theoretical number of grease injection cycles is determined by multiplying the bearing speed by the grease injection time; in the actual grease injection process, a two-stage strategy is adopted, namely, high flow rate for rapid filling in the early stage and standard flow rate for precise control in the later stage.

[0026] Furthermore: the incremental learning update rule is as follows:

[0027] ;

[0028] Where Dnew represents the newly added sample pool; (x t ,y t () represents the newly collected working condition-grease injection volume pair;

[0029] The formula for calculating the grease injection control parameters is:

[0030] ;

[0031] Among them, V pred To predict the amount of grease injected; Q is the grease pump flow rate; n rpm T represents the bearing speed; theory Theoretical liposuction time; N turns The theoretical number of fat injection rings.

[0032] This invention has the following advantages: It incorporates bearing condition, environment, structure, and grease performance parameters into the grease injection volume prediction model simultaneously, solving the problem of the limitations of single-parameter models; it combines residual connections, multi-head attention mechanisms, and LSTM networks, which not only captures time-series dependencies but also improves the model's expressive power and training stability; and it introduces online learning and Bayesian optimization, enabling the model to adaptively adjust to changes in equipment condition, solving the problem of poor adaptability of traditional static models.

[0033] This invention can reduce grease consumption, reduce bearing failures, and extend equipment life, thus providing certain economic benefits.

[0034] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. Attached Figure Description

[0035] To more intuitively illustrate the prior art and this application, exemplary drawings are provided below. It should be understood that the specific shapes and structures shown in the drawings should not generally be regarded as limiting conditions for implementing this application; for example, based on the technical concept disclosed in this application and the exemplary drawings, those skilled in the art are able to easily make conventional adjustments or further optimizations to the addition / reduction / classification, specific shapes, positional relationships, connection methods, size ratios, etc. of certain units (components).

[0036] Figure 1 This is a flowchart illustrating an implementation of a multi-parameter-based fat injection volume prediction method according to one embodiment of this application. Detailed Implementation

[0037] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these embodiments are merely for further explanation of the present invention and should not be construed as limiting the scope of protection of the present invention. Technical engineers in the field can make some non-essential improvements and adjustments to the present invention based on the above-described content. 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.

[0038] Please see Figure 1 A multi-parameter-based fat injection volume prediction method includes the following steps:

[0039] Step 1: Multi-source data acquisition and preprocessing;

[0040] Bearing system parameters were collected from four dimensions: bearing condition parameters, environmental parameters, bearing specification parameters, and grease performance parameters. The collected raw data were preprocessed, including outlier handling, smoothing filtering, and normalization.

[0041] The bearing condition parameters include bearing temperature, bearing speed, and vibration acceleration; the environmental parameters include dust concentration and oxygen concentration in the air; the bearing specification parameters include the bearing inner diameter, outer diameter, and roller diameter; and the grease performance parameters include the grease dielectric constant, consistency, and dropping point.

[0042] Step 2: Feature fusion and dimensionality reduction;

[0043] The preprocessed multi-source data were aligned according to timestamps to construct an initial feature matrix. The kernel principal component analysis algorithm was used to extract nonlinear features from the initial feature matrix, mapping high-dimensional features to a low-dimensional principal component space. The recursive feature elimination method combined with the support vector machine model was used to select the feature combination with the highest correlation to the amount of fat injected.

[0044] Kernel principal component analysis (KPCA) projects data onto a reproducing kernel Hilbert space using a nonlinear mapping to solve the generalized eigenvalue problem. Where K is the kernel matrix, K ij =k(x i ,x j ), λ is the Gaussian radial basis kernel function; α is the eigenvector; λ is the eigenvalue.

[0045] Recursive Feature Elimination-SVM (RFE-SVM) uses an initial feature set S0 to iteratively remove the least important features until m features remain.

[0046] J(S k =CrossValAccuracy(SVM(S k ));

[0047] Among them, J(S) k ) represents the current feature subset S k Cross-validation accuracy; SVM(S k ) is based on S k The training support vector machine; termination condition: |Sk|=m, finally retaining the feature combination with the highest correlation to the amount of fat injected.

[0048] Step 3: Build and train the prediction model;

[0049] A residual multi-head attention long short-term memory network is constructed as the core prediction model, which includes:

[0050] (1) The LSTM base layer consists of two LSTM networks with 128 units each, used to capture long-term dependencies in time series.

[0051] (2) Multi-head attention mechanism: the LSTM output is divided into 8 heads, each focusing on a different feature subspace;

[0052] (3) Residual connections: introducing cross-layer connections in the network to prevent gradient vanishing;

[0053] (4) Fully connected output layer, which maps the output of the attention mechanism to a single fat injection volume prediction value;

[0054] The prediction model training uses the Huber loss function and AdamW optimizer, and utilizes Bayesian optimization algorithm to automatically optimize hyperparameters such as learning rate and batch size.

[0055] The Huber loss function is:

[0056] ;

[0057] Where y is the actual amount of fat injected; This is the predicted amount of grease to be injected; δ is the threshold value. This is the historical average.

[0058] The Bayesian optimization objective is:

[0059] ;

[0060] Where θ is the hyperparameter vector; MAPE val To verify the mean absolute percentage error of the validation set, the optimal hyperparameters are efficiently searched using a Gaussian process surrogate model.

[0061] Step 4: Predict the amount of grease to be injected;

[0062] Input the feature data selected in step 2 into the trained prediction model to obtain the required amount of grease for the bearing; determine whether the predicted grease amount is within the preset threshold range, and if it exceeds the range, adopt a rule-based safety strategy; use the newly collected operating status data and actual grease injection effect data to perform incremental learning on the prediction model to achieve dynamic optimization of the model.

[0063] The kernel principal component analysis algorithm retains more than 90% of the original information; the evaluation model of the recursive feature elimination method adopts support vector machine.

[0064] Step 5: Convert the predicted values ​​into specific control commands;

[0065] The theoretical grease injection time is obtained by dividing the predicted grease volume by the grease pump flow rate, and the theoretical number of grease injection cycles is determined by multiplying the bearing speed by the grease injection time. In the actual grease injection process, a two-stage strategy is adopted, namely, high flow rate for rapid filling in the early stage and standard flow rate for precise control in the later stage, so as to achieve a balance between efficiency and accuracy.

[0066] Step 6: Continuously perform dynamic optimization and online learning; after grease injection, continuously monitor the operating status of the bearing (such as temperature and vibration changes) to evaluate the actual effect of the grease injection; these newly generated "grease injection effect-operating condition data" pairs will be labeled and added to the training dataset; a sliding window mechanism is adopted to always keep the latest 100,000 data points in use; through regular incremental learning, the prediction model can absorb new knowledge and dynamically adjust its internal parameters, thereby continuously adapting to slow changes such as equipment aging, seasonal changes, and grease performance degradation, achieving self-evolution and continuous improvement of prediction capabilities.

[0067] The incremental learning update rule is as follows:

[0068] ;

[0069] Where Dnew is the newly added sample pool (containing the latest 100,000 data entries); (x t ,y t () represents the newly collected working condition-grease injection volume pair;

[0070] The formula for calculating the grease injection control parameters is:

[0071] ;

[0072] Among them, V pred To predict the injection volume (mL); Q is the grease pump flow rate (mL / s); n rpm T is the bearing speed (RPM); theory Theoretical fat injection time (s); N turns The theoretical number of fat injection rings.

[0073] During liposuction, high flow rate is used for rapid filling in the early stage, and standard flow rate is used for fine-tuning in the later stage.

[0074] In step 1, outlier handling employs outlier group preprocessing based on seasonal factor decomposition; smoothing filtering uses a moving average filter; and normalization processing uses a minimum-maximum normalization method.

[0075] In step 3, the input sequence length of the residual multi-head attention long short-term memory network is 60 time steps; the optimization objective of the Bayesian optimization algorithm is the mean absolute percentage error on the validation set.

[0076] In step 4, the preset threshold range is 70%-130% of the historical fat injection volume; the incremental learning adopts a sliding window mechanism to retain the most recent 100,000 data points.

[0077] Implementation Case 1: Grease Injection of Main Shaft Bearings for Wind Turbine Units

[0078] Application scenario: The main shaft bearing of a 1.5MW wind turbine generator set in a wind farm operates in a harsh environment with high maintenance costs.

[0079] Implementation steps:

[0080] A PT100 temperature sensor and a triaxial acceleration vibration sensor are installed on the spindle bearing; a PM2.5 dust sensor and an electrochemical oxygen sensor are installed in the engine compartment; an online dielectric constant sensor is installed in the lubrication line; all sensors are connected to the central processing unit via a CAN bus.

[0081] Data collection: Data will be collected continuously for 3 months, covering different seasons and operating conditions;

[0082] Sampling frequencies: temperature 1Hz, vibration 20kHz, rotation speed 1Hz, environmental parameters 0.1Hz;

[0083] Simultaneously record bearing specifications: inner diameter 200mm, outer diameter 300mm, roller diameter 25mm.

[0084] Model training: Data from the first two months was used as the training set, and data from the last month was used as the test set; after feature dimensionality reduction, the original 12-dimensional parameters were compressed into 7 principal features, with a cumulative contribution rate of 92%; the optimal hyperparameters were obtained through Bayesian optimization: learning rate 0.002, batch size 64, and LSTM unit number 128.

[0085] After 500 training rounds, the model achieved R²=0.94 on the test set with a mean absolute percentage error of 6.3%.

[0086] Grease injection control: The predicted grease injection volume is converted into a grease pump control signal; a two-stage grease injection strategy is adopted: high flow rate (1.2 times the standard flow rate) is used for the first 30% of the time, and standard flow rate is used for the next 70% of the time; the bearing temperature change during the grease injection process is monitored in real time, and grease injection is stopped immediately if any abnormality occurs.

[0087] Performance evaluation: After 6 months of actual operation, compared with the traditional method: over-lubrication was reduced; grease consumption was reduced; failure rate due to insufficient lubrication was reduced; and bearing life was extended.

[0088] Implementation Case 2: Grease Injection into Rolling Mill Bearings in a Steel Plant

[0089] Application scenario: Support roll bearings of a hot strip mill in a steel plant, which operate at high temperatures, under heavy loads, and generate significant pollution.

[0090] Implement high-temperature adaptability improvements: Select high-temperature resistant sensors with a maximum operating temperature of 150℃; add a temperature compensation algorithm in the preprocessing stage to eliminate the impact of high temperature on sensor accuracy.

[0091] Heavy load condition adaptation: Add load current parameters as additional input features; assign higher weights to heavy load condition data during model training.

[0092] Environmental pollution response: Employ more frequent sensor-based automatic cleaning mechanisms; increase sensitivity analysis of dust concentration parameters during the feature selection phase.

[0093] Implementation results: After 4 months of operation, compared with the traditional method, the bearing replacement cycle was extended from 3 months to 5 months, grease consumption was reduced, and unplanned downtime due to bearing damage was reduced.

[0094] The present invention has the following beneficial effects: by simultaneously incorporating bearing condition, environment, structure and grease performance parameters into the grease injection volume prediction model, it solves the problem of the large limitations of single parameter models;

[0095] Combining residual connections, multi-head attention mechanisms, and LSTM networks not only captures time-series dependencies but also improves the model's expressive power and training stability.

[0096] By introducing online learning and Bayesian optimization, the model can adaptively adjust to changes in device status, solving the problem of poor adaptability of traditional static models;

[0097] This invention can reduce grease consumption, reduce bearing failures, and extend equipment life, thus providing certain economic benefits.

[0098] 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, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting fat injection volume based on multiple parameters, characterized in that, Includes the following steps: Step 1: Collect data from multiple sources and preprocess it; Step 2: Align the preprocessed multi-source data according to timestamps to construct an initial feature matrix; Nonlinear feature extraction is performed on the initial feature matrix to map high-dimensional features to a low-dimensional principal component space; feature data with the highest correlation to the amount of fat injected are selected. Step 3: Construct a residual multi-head attention long short-term memory network as a prediction model. The prediction model includes an LSTM base layer, a multi-head attention mechanism, residual connections, and a fully connected output layer. Step 4: Input the feature data selected in Step 2 into the trained prediction model to obtain the required amount of grease for the bearing; determine whether the predicted amount of grease is within the preset threshold range, and if it exceeds the range, adopt a rule-based safety strategy. Step 5: Convert the predicted values ​​into specific control commands; Step 6: The grease injection equipment performs grease injection upon receiving the control command. After the grease injection is executed, the operating status of the bearing is continuously monitored to evaluate the actual effect of the grease injection. These newly generated data pairs are labeled and added to the training dataset. Through regular incremental learning, the prediction model can absorb new knowledge, dynamically adjust its internal parameters, and achieve self-evolution and continuous improvement of its prediction capabilities.

2. The method for predicting fat injection volume based on multiple parameters according to claim 1, characterized in that, Step 1 involves collecting multi-source data from four dimensions: bearing status parameters, environmental parameters, bearing specification parameters, and grease performance parameters. The collected raw data is then preprocessed, including outlier handling, smoothing filtering, and normalization.

3. The method for predicting fat injection volume based on multiple parameters according to claim 2, characterized in that, The bearing condition parameters include bearing temperature, bearing speed, and vibration acceleration; environmental parameters include dust concentration and oxygen concentration in the air; bearing specification parameters include bearing inner diameter, outer diameter, and roller diameter; and grease performance parameters include grease dielectric constant, consistency, and dropping point.

4. The method for predicting fat injection volume based on multiple parameters according to claim 1, characterized in that, The calculation method of the recursive feature elimination method is as follows: Let the initial feature set be S0, and iteratively delete the least important features until m features remain: J(S k )=CrossValAccuracy(SVM(S k )); Among them, J(S) k ) represents the current feature subset S k Cross-validation accuracy; SVM(S k ) is based on S k The training support vector machine; termination condition: |Sk|=m, finally retaining the feature combination with the highest correlation to the amount of fat injected.

5. The method for predicting fat injection volume based on multiple parameters according to claim 1, characterized in that, The prediction model is trained using the Huber loss function and the AdamW optimizer, and the learning rate and batch size hyperparameters are automatically optimized using the Bayesian optimization algorithm.

6. The method for predicting fat injection volume based on multiple parameters according to claim 5, characterized in that, The Huber loss function is: ； Where y is the actual amount of fat injected; This is the predicted amount of grease to be injected; δ is the threshold value. This is the historical average. The Bayesian optimization objective is: ; Where θ is the hyperparameter vector; MAPE val To verify the mean absolute percentage error of the validation set, the optimal hyperparameters are efficiently searched using a Gaussian process surrogate model.

7. The method for predicting fat injection volume based on multiple parameters according to claim 1, characterized in that, By using newly collected operational data and actual grease injection effect data, the prediction model is incrementally learned to achieve dynamic optimization of the model.

8. The method for predicting fat injection volume based on multiple parameters according to claim 1, characterized in that, In step 5, the theoretical grease injection time is obtained by dividing the predicted grease volume by the grease pump flow rate, and the theoretical number of grease injection cycles is determined by multiplying the bearing speed by the grease injection time. In the actual grease injection process, a two-stage strategy is adopted, namely, high flow rate for rapid filling in the early stage and standard flow rate for precise control in the later stage.

9. The method for predicting fat injection volume based on multiple parameters according to claim 1, characterized in that, The incremental learning update rule is as follows: ； Where Dnew represents the newly added sample pool; (x t ,y t () represents the newly collected working condition-grease injection volume pair; The formula for calculating the grease injection control parameters is: ； Among them, V pred To predict the amount of grease injected; Q is the grease pump flow rate; n rpm T represents the bearing speed; theory Theoretical liposuction time; N turns The theoretical number of fat injection rings.

Images