Hypersphere and refined parameter-based health state evaluation method for deriving bearing system

Abstract

The invention provides a hyper-sphere and refined parameter-based health state evaluation method for deriving a bearing system. The method comprises the following steps of: obtaining historical sample data; calculating and deducing the health degree of the bearing system by using a hypersphere model; calculating the health degree of each key parameter; the health degree of the hypersphere model and the health degree of the key parameters are integrated, and the health degree of the bearing system at the # imgabs0 # moment is obtained; and correcting and deducing the health degree of the bearing system by using an independence weight method and a TOPSIS method. According to the method, the problems that subjective participation is needed, the evaluation mode is single and the health state of a complex mechanical system is difficult to accurately evaluate in a traditional health monitoring system are solved, and the health state of the bearing system can be more accurately predicted, evaluated and deduced by comprehensively applying a machine learning technology and a statistical method. Therefore, potential faults can be found in advance, the non-planned downtime is shortened, the reliability of equipment is improved, the service life of the equipment is prolonged, and the method is of great significance to safe and stable operation of industrial production.

Description

Technical Field

[0001] The invention relates to the field of equipment health evaluation in the hydropower industry, and in particular to a method for evaluating the health status of a bearing system based on a hypersphere and refined parameters. Background Art

[0002] The derivation bearing is also called the thrust guide joint bearing in the umbrella unit. It is often called the heart of the hydro-turbine generator set. Its function is to bear the axial water thrust of the hydro-turbine generator set and the weight of the rotating parts, as well as the radial mechanical and electromagnetic unbalanced forces of some rotating parts. The working performance of the derivation bearing system directly affects whether the hydro-turbine generator set can operate long-term, safely and reliably. With the continuous improvement of the reliability requirements of the hydropower industry, how to effectively evaluate and monitor the health status of the derivation bearing system has become a key issue that needs to be solved urgently.

[0003] At present, some progress has been made in the health status assessment technology for hydropower units and their key components. In Chinese patent CN114841533A, a comprehensive weighted health assessment method for the top cover drainage system of a hydropower unit is proposed. This method decomposes the top cover drainage system at multiple levels and combines the subjective weight method (hierarchy analysis method, AHP) and the objective weight method (anti-entropy weight method) to achieve quantitative evaluation of the health status of each level. In addition, Chinese patent CN115577808A provides a method for evaluating the health status of a hydropower unit based on multi-source information fusion. This method comprehensively considers the distance between the online monitoring data and the limit value and the subjective scoring of the offline data by manual experience, and uses the hierarchical analysis method to determine the weight of each indicator, thereby achieving a comprehensive evaluation of the health status of the hydropower unit.

[0004] Although the above technical solutions can evaluate the health status of hydropower units to a certain extent, the data requires subjective participation in the evaluation and the evaluation method is single. It only reflects the overall health status from the health status of each component, and does not consider the health status of the equipment from the overall level. Therefore, it is necessary to develop a new method to solve the above problems, so as to more accurately monitor and evaluate the health status of the bearing system, thereby ensuring the efficient operation of the hydropower generator set. Summary of the invention

[0005] The main purpose of the present invention is to provide a method for evaluating the health status of a bearing system based on a hypersphere and refined parameters, so as to solve the problems that the health status evaluation of equipment in the current hydropower field requires subjective participation and the evaluation method is single.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for evaluating the health status of a bearing system based on a hypersphere and refined parameters, comprising the following steps: Step 1: Obtain historical sample data; Step 2: Calculate and derive the health of the bearing system; Step 2.1: Filter healthy historical sample data, preprocess the data, and obtain processed healthy sample data ; Step 2.2: Use healthy samples to train the hypersphere model; Step 2.3: Use the hypersphere model to calculate the bearing system health ; Step 3: Calculate the health of each key parameter; Step 3.1: Use the LSTM long short-term memory network model to calculate the predicted values ​​of tile temperature, oil level, and swing, and The residuals between the actual and predicted values ​​of the bearing temperature, lower guide swing and lower guide oil level at the moment; Step 3.2: Exploitation The rule determines whether the cooling water flow is abnormal and calculates the abnormal situation Cooling water flow rate at all times The minimum distance between the upper and lower limits of the interval; Step 3.3: Calculate using quadratic parabolic membership function The health of each parameter at all times ; Step 4: Combine the health of the hypersphere model and the health of key parameters to obtain the bearing system Health at all times; Integrate the hypersphere model health and key parameter health to derive the bearing system Health at all times: ; Step 5: Modify the derived bearing system health, use the independence weight method to determine the initial weight of each key parameter health, use the TOPSIS method to determine the final weight of each key parameter health, and determine the derived bearing system health.

[0007] In the preferred solution, in step 2.1, filter out healthy historical sample data Specifically expressed as: ; in, Representation Matrix No. Line The elements of the column, , is the type of input data, , The number of input data for each category; In the preferred solution, in step 2.1, the historical sample data of health The preprocessing includes the following steps: S111. Historical sample data on health Perform value compensation and frequency reduction processing, and use active power as the judgment condition to select data when the unit is in stable operation, and remove null values ​​and abnormal points in the data; S112, normalizing the input parameters, the normalization method is: ; S113. Obtain final healthy sample data through the above processing , specifically expressed as: ; in, Representation Matrix No. Line The elements of the column, , is the type of input data, , The number of input data of each category after data preprocessing.

[0008] In the preferred embodiment, step 2.2 specifically includes the following sub-steps: S221, learning a large number of historical health samples of the system to generate a minimum hypersphere containing all health states, that is, the optimal hypersphere; S222, generating a hypersphere model node; S223. Obtain a bearing system health hypersphere model.

[0009] In the preferred solution, in step S222, a hypersphere model node is generated, which specifically includes the following process: deleting 70% of the state points in the hypersphere space, retaining 30% of the evenly distributed state points as the initial nodes, so that the distance between the deleted state points and the nearest nodes is minimized, backfilling the deleted points, calculating the distance between the deleted points and the nodes, and then using the Gaussian membership function to calculate the health index HPI of the deleted points relative to the nodes. If the HPI is less than a preset value, the deleted points are added as nodes.

[0010] In the preferred solution, in step 3.1, the LSTM long short-term memory network model is used to calculate the predicted values ​​of the tile temperature, oil level, and swing, and The residuals of the actual values ​​and the predicted values ​​of the bearing temperature, lower guide swing and lower guide oil level at the moment include the following sub-steps: S311. Calculation of tile temperature health: Use the historical sample data in step 1 to obtain the input parameters related to the tile temperature, and perform value filling, frequency reduction, null value processing and normalization on the input parameters, and use the LSTM multi-input multi-output model to predict the predicted value of the tile temperature. ,calculate The residual between the actual value and the predicted value of the temperature at the moment ; S312, calculation of swing and oil level health: using the historical sample data in step 1, obtain the input parameters related to the swing and oil level, and perform value filling, frequency reduction, null value processing and normalization on the input parameters, and use the LSTM multi-input multi-output model to predict the X-direction swing and Y-direction swing of the lower guide to obtain the predicted value of the lower guide swing as well as The residual between the actual value and the predicted value of the guide swing at time ; Use the LSTM multi-input single-output model to predict the oil level and obtain the predicted value of the lower oil level as well as The residual between the actual value and the predicted value of the oil level at time .

[0011] In the preferred solution, in step 3.2, the average cooling water flow rate under the historical healthy state is calculated and standard deviation ,use Principle, determine whether the cooling water flow is abnormal, if it is abnormal, calculate the abnormal The minimum distance between the cooling water flow rate and the upper or lower limit of the interval at any given moment .

[0012] In the preferred solution, in step 3.3, the key parameter health is calculated. The specific process is: according to the obtained in steps 3.1 and 3.2 The residuals of each key parameter at the moment are constructed The residual matrix of each key parameter at the moment: ; use In principle, the allowable residual matrix of various key parameters is obtained: ; in, It represents the allowable residual of tile temperature, which is 3 times the standard deviation of the predicted value of tile temperature; It represents the allowable residual error of the swing, which is 3 times the standard deviation of the predicted value of the swing; It represents the allowable residual of oil level, which is 3 times the standard deviation of the predicted oil level value; It represents the allowable residual of cooling water flow, which is 3 times the standard deviation of cooling water flow under historical healthy status.

[0013] The difference between the alarm value and the predicted value of each key parameter is taken as the maximum residual of the parameter. The maximum residual matrix of each key parameter is: ; in, represents the maximum residual error of Wattage. represents the maximum residual error of the swing, Indicates the maximum residual error of the oil level, Indicates the maximum residual of cooling water flow; The health of each key parameter is determined by a quadratic parabolic membership function, namely: Health at all times: ; in, , , represents the number of key parameters of each category, Indicates The key parameters of the class The health of the parameters, Indicates The key parameters of the class The residuals of the parameters, Indicates The key parameters of the class The allowed residuals of the parameters, Indicates The key parameters of the class The maximum residual error of the parameters.

[0014] In the preferred solution, in step 4, the initial weights of the health of each key parameter are determined by the independence weight method, and the final weights of the health of the bearing system and the health of each key parameter are determined by the TOPSIS method, which specifically includes the following sub-steps: Step 4.1: Use the independence weight method to determine the initial weights of the health of each key parameter. Expand it and get: ; in, Indicates the total number of key parameters; Before All health levels at all times, calculate the correlation coefficient between the health levels of each key parameter , then the initial weights of the health of each key parameter are: ; in, , For the The standard deviation of the health of key parameters; Normalize the weights to get the weights of the key parameter health: ; in, For the The weight of the health of the key parameters; Step 4.2: Use the TOPSIS method (top-bottom distance method) to determine the final weights of the derived bearing system health and the health of each key parameter; Derive initial weights for the health of the bearing system and key parameters ,in, To derive the bearing system health The initial weight of The initial weights obtained in step S51 are normalized to obtain weights: ; in, , After normalization, The health weight, , Respectively represent the maximum and minimum values ​​of the initial weights; No. The weight of each health degree is: ; The final health of the bearing system is derived as: .

[0015] The present invention provides a method for evaluating the health status of a bearing system based on a hypersphere and refined parameters, which combines the hypersphere model with dynamic monitoring of key parameters to quantify the health status of the system. First, the bearing health is derived using the optimized hypersphere model, and then the LSTM long short-term memory network model and The law obtains the changing trend of key parameters and evaluates the health of these parameters, and then uses the quadratic parabolic membership function to calculate the health of each parameter. Finally, by comprehensively considering the overall health of the hypersphere model and the specific health of each key parameter, and using the independence weight method and TOPSIS method to determine the relative importance of these parameters, a comprehensive health evaluation of the entire bearing system is obtained. The present invention can accurately identify early signs of failure in the bearing system, timely warn of potential problems, effectively extend the service life of the equipment and reduce maintenance costs, which is of great significance to improving the reliability and safety of industrial equipment.

[0016] The beneficial effects of the present invention are: 1) The present invention calculates and infers the health of the bearing from two dimensions, namely, the system and the components, and then corrects the health of the derived bearing system by the health of the components, which more accurately reflects the health status of the derived bearing.

[0017] 2) The present invention utilizes the hypersphere model to calculate and derive the health of the bearing system, thereby reflecting the health of the bearing system from a global perspective.

[0018] 3) The present invention uses a weighting method that combines the independence weight method and the TOPSIS method to calculate the health weight of each key parameter, fully considering the characteristics of the data itself and eliminating the differences caused by subjective evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 It is a flow chart of the bearing system health status assessment method based on the hypersphere and refined parameters of the present invention; Figure 2 It is the optimal hypersphere in the embodiment of the present invention; Figure 3 It is the principle of generating nodes of the hypersphere model in the embodiment of the present invention; Figure 4 It is a training hypersphere model in an embodiment of the present invention; Figure 5 This is a schematic diagram of the LSTM in an embodiment of the present invention; DETAILED DESCRIPTION Example 1 like Figures 1 to 5 As shown, a method for evaluating the health status of a bearing system based on a hypersphere and refined parameters includes the following steps: Step 1: Obtain historical sample data; Step 2: Calculate and derive the health of the bearing system; Step 2.1: Filter healthy historical sample data, preprocess the data, and obtain processed healthy sample data ; Step 2.2: Use healthy samples to train the hypersphere model; Step 2.3: Use the hypersphere model to calculate the bearing system health ; Step 3: Calculate the health of each key parameter; Step 3.1: Use the LSTM long short-term memory network model to calculate the predicted values ​​of tile temperature, oil level, and swing, and The residuals between the actual and predicted values ​​of the bearing temperature, lower guide swing and lower guide oil level at the moment; Step 3.2: Exploitation The rule determines whether the cooling water flow is abnormal and calculates the abnormal situation Cooling water flow rate at all times The minimum distance between the upper and lower limits of the interval; Step 3.3: Calculate using quadratic parabolic membership function The health of each parameter at all times ; Step 4: Combine the health of the hypersphere model and the health of key parameters to obtain the bearing system Health at all times; Integrate the hypersphere model health and key parameter health to derive the bearing system Health at all times: ; Step 5: Modify the derived bearing system health, use the independence weight method to determine the initial weight of each key parameter health, use the TOPSIS method to determine the final weight of each key parameter health, and determine the derived bearing system health.

[0020] The present invention proposes a health status assessment method for bearing systems based on hypersphere and refined parameters, which achieves accurate assessment of system health by integrating hypersphere model and key parameter monitoring. Historical sample data is collected and a data set representing the health status is selected from it. A hypersphere model is trained using this data, which can be used to calculate the health of the system at any time. At the same time, an LSTM long short-term memory network is used to predict the trend of key parameters, and the residual is calculated by comparing it with the actual value to evaluate the health status of each parameter. In addition, using The rule is used to check whether there is an abnormality in the cooling water flow and calculate the health status accordingly. Finally, the health status obtained by the comprehensive hypersphere model and the health status of each key parameter are determined by using independence and TOPSIS methods to determine the weights, thereby obtaining the comprehensive health status of the entire bearing system. This method can effectively monitor the operating status of the bearing system, detect potential faults in a timely manner, provide a scientific basis for maintenance decisions, and thus improve the reliability and economy of the system.

[0021] In the preferred solution, in step 2.1, filter out healthy historical sample data Specifically expressed as: ; in, Representation Matrix No. Line The elements of the column, , is the type of input data, , The number of input data for each category; By screening historical sample data, it is ensured that the data set used for subsequent modeling only contains valid samples representing the health status, effectively eliminating data interference under abnormal operating conditions, and improving the accuracy of subsequent hypersphere model training. By clearly defining the structure and composition of health data, it is ensured that the model can accurately learn the characteristics under normal operating mode, so that in practical applications, the health status of the bearing system can be more accurately identified and evaluated, the false alarm rate can be reduced, and the effect of predictive maintenance can be improved.

[0022] In the preferred solution, in step 2.1, the historical sample data of health The preprocessing includes the following steps: S111. Historical sample data on health Perform value compensation and frequency reduction processing, and use active power as the judgment condition to select data when the unit is in stable operation, and remove null values ​​and abnormal points in the data; S112, normalizing the input parameters, the normalization method is: ; S113. Obtain final healthy sample data through the above processing , specifically expressed as: ; in, Representation Matrix No. Line The elements of the column, , is the type of input data, , The number of input data of each category after data preprocessing.

[0023] By preprocessing the healthy historical sample data, the quality of the data set used to build the model is ensured. By filling and reducing the frequency of the data, and using active power as the judgment condition, the data of the unit during stable operation is selected, which can effectively remove the null values ​​and abnormal points in the data and ensure the consistency and reliability of the data. Next, the input parameters are normalized. This step helps to eliminate the differences between different dimensions and magnitudes, so that the model can treat each feature more fairly, thereby improving the accuracy and generalization ability of the model. Through the preprocessing steps, a high-quality healthy sample data set is finally obtained, which is conducive to the subsequent learning of the hypersphere model, and can ensure that the model can better learn the features in the healthy state, so as to more accurately evaluate the health status of the bearing system.

[0024] In the preferred embodiment, step 2.2 specifically includes the following sub-steps: S221, learning a large number of historical health samples of the system to generate a minimum hypersphere containing all health states, that is, the optimal hypersphere; S222, generating a hypersphere model node; S223. Obtain a bearing system health hypersphere model.

[0025] The health status of the bearing system is evaluated by constructing a hypersphere model. First, a large amount of historical health sample data is used to learn and generate an optimal hypersphere, which can contain the minimum range of all data points under normal operating conditions to characterize the health status of the system. Then, hypersphere model nodes are created in the optimal hypersphere as the basis for evaluating new data. Finally, a complete hypersphere model of the health of the bearing system is obtained through integration. This process can not only help identify abnormal conditions that deviate from the normal operating mode, but also quantify the health of the bearing system, which has important practical value for predictive maintenance and fault diagnosis.

[0026] In the preferred solution, in step S222, a hypersphere model node is generated, which specifically includes the following process: deleting 70% of the state points in the hypersphere space, retaining 30% of the evenly distributed state points as the initial nodes, so that the distance between the deleted state points and the nearest nodes is minimized, backfilling the deleted points, calculating the distance between the deleted points and the nodes, and then using the Gaussian membership function to calculate the health index HPI of the deleted points relative to the nodes. If the HPI is less than a preset value, the deleted points are added as nodes.

[0027] The efficiency and accuracy of the model are optimized by generating hypersphere model nodes. First, 70% of the state points are removed, and only 30% of the evenly distributed points are retained as initial nodes to ensure that these nodes can cover the entire health state space and maintain the minimum spacing. Then, the removed points are backfilled and the distance between them and each node is calculated, and the health index (HPI) of these points is evaluated using the Gaussian membership function. When the HPI of a point is lower than the preset threshold, it indicates that the point is of high importance and needs to be added as a new node. This method effectively reduces the complexity of the model while ensuring that the model can accurately capture abnormal conditions, thereby improving the accuracy of fault prediction.

[0028] In the preferred solution, in step 3.1, the LSTM long short-term memory network model is used to calculate the predicted values ​​of the tile temperature, oil level, and swing, and The residuals of the actual values ​​and the predicted values ​​of the bearing temperature, lower guide swing and lower guide oil level at the moment include the following sub-steps: S311. Calculation of tile temperature health: Use the historical sample data in step 1 to obtain the input parameters related to the tile temperature, and perform value filling, frequency reduction, null value processing and normalization on the input parameters, and use the LSTM multi-input multi-output model to predict the predicted value of the tile temperature. ,calculate The residual between the actual value and the predicted value of the temperature at the moment ; S312, calculation of swing and oil level health: using the historical sample data in step 1, obtain the input parameters related to the swing and oil level, and perform value filling, frequency reduction, null value processing and normalization on the input parameters, and use the LSTM multi-input multi-output model to predict the X-direction swing and Y-direction swing of the lower guide to obtain the predicted value of the lower guide swing as well as The residual between the actual value and the predicted value of the guide swing at time ; Use the LSTM multi-input single-output model to predict the oil level and obtain the predicted value of the lower oil level as well as The residual between the actual value and the predicted value of the oil level at time .

[0029] By using the LSTM long short-term memory network model to predict the trend of key operating parameters such as watt temperature, swing and oil level and their residuals, the health status of the equipment can be effectively monitored. By preprocessing the historical data related to watt temperature and using the LSTM model for prediction, an accurate watt temperature prediction value can be obtained, and the deviation between the actual value and the predicted value can be calculated based on this. The swing and oil level are predicted separately and the residuals are calculated to detect abnormal changes in a timely manner. This method not only helps to reduce the false alarm rate, but also improves the prediction accuracy, ensuring the timely and accurate identification of potential equipment problems, thereby achieving effective preventive maintenance.

[0030] In the preferred solution, in step 3.2, the average cooling water flow rate under the historical healthy state is calculated and standard deviation ,use Principle, determine whether the cooling water flow is abnormal, if it is abnormal, calculate the abnormal The minimum distance between the cooling water flow rate and the upper or lower limit of the interval at any given moment .

[0031] By statistically analyzing the historical data of cooling water flow, the mean and standard deviation are calculated, and the The principle is used to determine whether the cooling water flow exceeds the normal range. If an abnormal cooling water flow is detected, the minimum distance between the cooling water flow and the upper and lower limits of the normal range is further calculated to quantify the degree of abnormality. This process can detect abnormal changes in cooling water flow in a timely manner, provide key information for the health status assessment of the bearing system, and help take measures in advance to avoid equipment damage or performance degradation caused by cooling system failure.

[0032] In the preferred solution, in step 3.3, the key parameter health is calculated. The specific process is: according to the obtained in steps 3.1 and 3.2 The residuals of each key parameter at the moment are constructed The residual matrix of each key parameter at the moment: ; use In principle, the allowable residual matrix of various key parameters is obtained: ; in, It represents the allowable residual of tile temperature, which is 3 times the standard deviation of the predicted value of tile temperature; It represents the allowable residual error of the swing, which is 3 times the standard deviation of the predicted value of the swing; It represents the allowable residual of oil level, which is 3 times the standard deviation of the predicted oil level value; It represents the allowable residual of cooling water flow, which is 3 times the standard deviation of cooling water flow under historical healthy status.

[0033] The residual is allowed to be smaller than the maximum residual. If it exceeds this residual, no alarm will be triggered, but it will deviate from the data during normal operation.

[0034] The difference between the alarm value and the predicted value of each key parameter is taken as the maximum residual of the parameter. The maximum residual matrix of each key parameter is: ; in, represents the maximum residual error of Wattage. represents the maximum residual error of the swing, Indicates the maximum residual error of the oil level, Indicates the maximum residual of cooling water flow; The health of each key parameter is determined by a quadratic parabolic membership function, namely: Health at all times: ; in, , , represents the number of key parameters of each category, Indicates The key parameters of the class The health of the parameters, Indicates The key parameters of the class The residuals of the parameters, Indicates The key parameters of the class The allowed residuals of the parameters, Indicates The key parameters of the class The maximum residual error of the parameters.

[0035] The operating status of the bearing system is quantified by calculating the health of each key parameter. A residual matrix is ​​constructed based on the residuals obtained in steps 3.1 and 3.2, and the allowable residual matrix and the maximum residual matrix are defined to reflect the normal fluctuation range and abnormal threshold of the key parameters. A quadratic parabolic membership function is used to determine the health of each key parameter at a specific moment. This method can convert complex operating parameters into easy-to-understand and intuitive health indicators, which helps to quickly identify potential fault points and provides strong support for predictive maintenance and fault diagnosis. In addition, in this way, not only can the health status of each parameter be evaluated individually, but the overall performance of all key parameters can also be comprehensively considered, thereby achieving comprehensive monitoring and management of the overall health status of the bearing system.

[0036] In the preferred solution, in step 4, the initial weights of the health of each key parameter are determined by the independence weight method, and the final weights of the health of the bearing system and the health of each key parameter are determined by the TOPSIS method, which specifically includes the following sub-steps: Step 4.1: Use the independence weight method to determine the initial weights of the health of each key parameter. Expand it and get: ; in, Indicates the total number of key parameters; Before All health levels at all times, calculate the correlation coefficient between the health levels of each key parameter , then the initial weights of the health of each key parameter are: ; in, , For the The standard deviation of the health of key parameters; Normalize the weights to get the weights of the key parameter health: ; in, For the The weight of the health of the key parameters; Step 4.2: Use the TOPSIS method (top-bottom distance method) to determine the final weights of the derived bearing system health and the health of each key parameter; Derive initial weights for the health of the bearing system and key parameters ,in, To derive the bearing system health The initial weight of The initial weights obtained in step S51 are normalized to obtain weights: ; in, , After normalization, The health weight, , Respectively represent the maximum and minimum values ​​of the initial weights; No. The weight of each health degree is: ; The final health of the bearing system is derived as: .

[0037] The weights of the health of each key parameter are determined by the independence weight method and the TOPSIS method, and then the overall health status of the bearing system is comprehensively evaluated. The initial weights of the health of each key parameter are calculated using the independence weight method, which helps to quantify the impact of different parameters on the health status of the system. The final weights are determined using the TOPSIS method. This method can more accurately reflect the relative importance of each parameter and comprehensively calculate the final health of the bearing system based on these weights. This method can not only improve the accuracy of the evaluation, but also provide a more scientific basis for maintenance decisions, effectively guide maintenance work, and ensure the reliable operation of the system.

[0038] Example 2 Further illustrate with reference to Example 1, Figure 1-5 As shown in the figure, the bearing system health status assessment method based on the hypersphere and refined parameters is derived. The specific process is as follows: Step 1: Obtain historical sample data; Step 2: Calculate and derive the health of the bearing system; This method uses the hypersphere model to calculate and derive the overall health of the bearing system.

[0039] The calculation method of bearing system health is derived as follows: Data preprocessing. The input active power, speed, water head, bearing temperature, oil temperature, oil level, and swing for calculating the health of the bearing system in this embodiment are recorded as ; in, Representation Matrix No. Line The elements of the column, , is the type of input data, , The number of input data for each category; Select historical samples to infer the health status of the bearing system , for the sample Perform value compensation and frequency reduction processing; then, based on active power as the judgment condition, select the data when the unit is in stable operation, and remove the null values ​​and abnormal points in the data; finally, obtain the healthy sample after data preprocessing ; in, Representation Matrix No. Line The elements of the column, , is the type of input data, , The number of input data of each category after data preprocessing.

[0040] Using healthy samples to train the hypersphere model, as shown in 4, includes the following steps: 1) Learn a large number of historical health samples of the system to generate a minimum hypersphere containing all health states, that is, the optimal hypersphere, such as Figure 2 As shown; 2) Generate hypersphere model nodes. Delete 70% of the state points in the hypersphere space, and keep about 30% of the evenly distributed state points as the initial nodes, so that the distance between the deleted state points and the nearest nodes is the smallest. Backfill the deleted points, calculate the distance between the deleted points and the nodes, and then use the Gaussian membership function to calculate the health index HPI of the deleted points relative to the nodes. If the HPI is less than the preset value, which is 0.95 in this embodiment, the deleted point will be added as a node. Figure 3 As shown; 3) Obtain the hypersphere model of the bearing system health. Input the new state point into the established hypersphere model to obtain the health of the system in this state.

[0041] The bearing system health degree is derived by using the hypersphere model to calculate the time .

[0042] Step 3: Calculate the health of each key parameter; The key parameters that affect the health of the derived bearing system are bearing temperature and swing. Since changes in oil level and cooling water flow can reflect changes in bearing temperature in advance, the health of the derived bearing system is corrected by changes in four parameters: bearing temperature, swing, oil level and cooling water flow.

[0043] The LSTM long short-term memory network can be used to obtain predicted values ​​for the tile temperature, oil level, and swing, and then the residual between the predicted value and the actual value is obtained; since the pressure behind the pressure reducing valve generally remains stable, the change in cooling water flow is generally not large, and the normal situation of cooling water flow can be judged by comparing the current cooling water flow with the average of the cooling water flow under the historical healthy state.

[0044] LSTM long short-term memory network is a time recurrent neural network, which is specially designed to solve the long-term dependency problem of general RNN (recurrent neural network). Figure 5 As shown: In the figure, Indicates the current input. represents the output of the previous hidden layer, To forget the gate, To select the gate, For input gating, For output gating, Indicates the previous state. Indicates the current state. represents the current hidden layer output, Indicates output.

[0045] LSTM is mainly composed of four stages: 1) Forget stage. This stage is mainly to selectively forget the input from the previous node and forget the gating for ; in, is the weight of the forget gate, is the bias of the forget gate. The forget gate is obtained by calculation To control the previous state Which ones should be discarded and which ones should be kept.

[0046] 2) Selective memory stage. In this stage, the input is selectively memorized. Select memory, the current input content is obtained by the previous calculation The selected gating signal is represented by to control.

[0047] ; ; in, is the weight of the selected gate, To select the gate bias, is the input weight, is the input bias. Add the results of the above two steps to get the value transmitted to the next state.

[0048] ; 3) Output stage. This stage will determine what will be considered as the output of the current state, mainly through To control.

[0049] ; in, is the weight of the output gate, is the bias of the output gate. Represents the current hidden layer output and the final output as follows: ; ; in, is the final output weight, is the bias of the final output.

[0050] The calculation steps for deriving the health of key parameters of the bearing system are as follows: Watt health calculation: Since there are multiple tile temperatures, the LSTM multi-input multi-output model is used here to predict the temperature of each tile.

[0051] First, the input parameters related to the tile temperature (active power, head, speed, tile temperature, oil temperature, cooling water temperature, cooling water flow, etc.) are recorded as .

[0052] The input parameters are padded, reduced and treated as null values, and the input parameters are normalized. The normalization method is: ; Get the final input .

[0053] Finally, the LSTM model is used to obtain the predicted value of the tile temperature ,calculate The residual between the actual value and the predicted value at the moment .

[0054] Calculation of swing and oil level health: The input parameters related to the swing and oil level are obtained, and the input parameters are supplemented, reduced, processed with null values, and normalized. The swing is divided into the lower guide X-direction swing and the lower guide Y-direction swing. The multi-input multi-output model can be used to predict the lower guide swing, and the multi-input single-output model can be used to predict the derived oil level. The predicted values ​​of the lower guide swing and the lower guide oil level are obtained respectively. and , The residuals between the actual value and the predicted value at each moment are and .

[0055] Cooling water flow health calculation: First, calculate the average cooling water flow rate under historical health conditions and standard deviation , using 3 Principle, compare whether the cooling water flow at the current moment is If it is within the range, it is considered that the cooling water flow is normal; otherwise, calculate The minimum distance between the cooling water flow rate and the upper or lower limit of the interval at any given moment .

[0056] Key parameter health calculation: Calculated using quadratic parabolic membership function The health of each parameter at all times .

[0057] According to the above steps, The residuals of each key parameter at the moment are constructed The residual matrix of each key parameter at the moment: ; use In principle, the allowable residual matrix of various key parameters is obtained: ; in, It represents the allowable residual of tile temperature, which is 3 times the standard deviation of the predicted value of tile temperature; It represents the allowable residual error of the swing, which is 3 times the standard deviation of the predicted value of the swing; It represents the allowable residual of oil level, which is 3 times the standard deviation of the predicted oil level value; It represents the allowable residual of cooling water flow, which is 3 times the standard deviation of cooling water flow under historical healthy status.

[0058] The residual is allowed to be smaller than the maximum residual. If it exceeds this residual, no alarm will be triggered, but it will deviate from the data during normal operation.

[0059] The difference between the alarm value and the predicted value of each key parameter is taken as the maximum residual of the parameter. The maximum residual matrix of each key parameter is: ; in, represents the maximum residual error of Wattage. represents the maximum residual error of the swing, Indicates the maximum residual error of the oil level, Indicates the maximum residual of cooling water flow; The health of each key parameter is determined by a quadratic parabolic membership function, namely: Health at all times: ; in, , , represents the number of key parameters of each category, Indicates The key parameters of the class The health of the parameters, Indicates The key parameters of the class The residuals of the parameters, Indicates The key parameters of the class The allowed residuals of the parameters, Indicates The key parameters of the class The maximum residual error of the parameters.

[0060] Step 4: Combine the health of the hypersphere model and the health of key parameters to obtain the bearing system Health at all times; Integrate the hypersphere model health and key parameter health to derive the bearing system Health at all times: ; Step 5: Derive the correction of the health of the bearing system, use the independence weight method to determine the initial weight of the health of each key parameter, use the TOPSIS method to determine the final weight of the health of each key parameter, and determine the derived bearing system health.

[0061] The independence weight method is an objective weighting method. Its idea is to use the strength of collinearity between indicators to determine the weight. If a certain indicator is highly correlated with other indicators, it means that the information has a large overlap, which means that the weight of the indicator will be relatively low. On the contrary, if a certain indicator is weakly correlated with other indicators, it means that the indicator carries a large amount of information and should be given a higher weight. This method is suitable for determining the weight of indicators that have a certain correlation and belong to the same system. The specific calculation method is as follows: 1) Use the independence weight method to determine the initial weight of the health of each key parameter; First, the matrix Expand it and get: ; in, Indicates the total number of key parameters; Before All health levels at all times, calculate the correlation coefficient between the health levels of each key parameter , then the initial weights of the health of each key parameter are: ; in, , For the The standard deviation of the health of key parameters; Normalize the weights to get the weights of the key parameter health: ; in, For the The weight of the health of the key parameters; 2) Use the TOPSIS method to determine the final weights of the bearing system health and the health of each key parameter; TOPSIS method (top-to-bottom distance method) is a commonly used comprehensive evaluation method that can make full use of the information of the original data, and its results can accurately reflect the gap between the evaluation schemes. The basic process is to first unify the indicator type of the original data matrix (generally forward processing) to obtain a forward matrix, and then standardize the forward matrix to eliminate the influence of the dimension of each indicator, and find the best and worst solutions among the limited solutions, and then calculate the distance between each evaluation object and the best and worst solutions, and obtain the relative closeness of each evaluation object to the best solution, which is used as the basis for evaluating the pros and cons. The specific method is as follows: To derive the bearing system health The initial weight of the bearing system is assumed to be The initial weight , including the derivation of initial weights for the health of the bearing system and key parameters ,in, Indicates the total number of key parameters, The initial weights obtained in step S51 are normalized to obtain weights: ; in, , After normalization, The health weight, , Respectively represent the maximum and minimum values ​​of the initial weights; No. The weight of each health degree is: ; The final health of the bearing system is derived as: .

[0062] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for evaluating the health status of a bearing system based on a hypersphere and refined parameters, characterized in that: The following steps are involved: Step 1: Obtain historical sample data; Step 2: Calculate and derive the health of the bearing system; Step 2.1: Filter healthy historical sample data, preprocess the data, and obtain processed healthy sample data ; Step 2.2: Use healthy samples to train the hypersphere model; Step 2.3: Use the hypersphere model to calculate the bearing system health ; Step 3: Calculate the health of each key parameter; Step 3.1: Use the LSTM long short-term memory network model to calculate the predicted values ​​of tile temperature, oil level, and swing, and The residuals between the actual and predicted values ​​of the bearing temperature, lower guide swing and lower guide oil level at the moment; Step 3.2: Exploitation The rule determines whether the cooling water flow is abnormal and calculates the abnormal situation Cooling water flow rate at all times The minimum distance between the upper and lower limits of the interval; Step 3.3: Calculate using quadratic parabolic membership function The health of each parameter at all times ; Step 4: Combine the health of the hypersphere model and the health of key parameters to obtain the bearing system Health at all times; Integrate the hypersphere model health and key parameter health to derive the bearing system Health at all times: ； Step 5: Modify the derived bearing system health, use the independence weight method to determine the initial weight of each key parameter health, use the TOPSIS method to determine the final weight of each key parameter health, and determine the derived bearing system health.

2. A method for evaluating the health status of a bearing system based on a hypersphere and refined parameters according to claim 1, characterized in that: In step 2.1, filter out healthy historical sample data Specifically expressed as: ； in, Representation Matrix No. Line The elements of the column, , is the type of input data, , The number of input data for each category.

3. A method for evaluating the health status of a bearing system based on a hypersphere and refined parameters according to claim 2, characterized in that: In step 2.1, the historical sample data of health The preprocessing includes the following steps: S111. Historical sample data on health Perform value compensation and frequency reduction processing, and use active power as the judgment condition to select data when the unit is in stable operation, and remove null values ​​and abnormal points in the data; S112, normalizing the input parameters, the normalization method is: ； S113. Obtain final healthy sample data through the above processing , specifically expressed as: ； in, Representation Matrix No. Line The elements of the column, , is the type of input data, , The number of input data of each category after data preprocessing.

4. The method for evaluating the health status of a bearing system based on a hypersphere and refined parameters according to claim 1, characterized in that: Step 2.2 specifically includes the following sub-steps: S221, learning a large number of historical health samples of the system to generate a minimum hypersphere containing all health states, that is, the optimal hypersphere; S222, generating a hypersphere model node; S223. Obtain a bearing system health hypersphere model.

5. A method for evaluating the health status of a bearing system based on a hypersphere and refined parameters according to claim 4, characterized in that: In step S222, a hypersphere model node is generated, which specifically includes the following process: 70% of the state points in the hypersphere space are deleted, and 30% of the evenly distributed state points are retained as initial nodes, so that the distance between the deleted state points and the nearest nodes is minimized, the deleted points are backfilled, the distance between the deleted points and the nodes is calculated, and then the health index HPI of the deleted points relative to the nodes is calculated using the Gaussian membership function. If the HPI is less than the preset value, the deleted point is added as a node.

6. The method for evaluating the health status of a bearing system based on a hypersphere and refined parameters according to claim 1, characterized in that: In step 3.1, the LSTM long short-term memory network model is used to calculate the predicted values ​​of tile temperature, oil level, and swing, as well as The residuals of the actual values ​​and the predicted values ​​of the bearing temperature, lower guide swing and lower guide oil level at the moment include the following sub-steps: S311. Calculation of tile temperature health: Use the historical sample data in step 1 to obtain the input parameters related to tile temperature, and perform value filling, frequency reduction, null value processing and normalization on the input parameters, and use the LSTM multi-input multi-output model to predict the predicted value of tile temperature. ,calculate The residual between the actual value and the predicted value of the temperature at the moment ; S312, calculation of swing and oil level health: using the historical sample data in step 1, obtain the input parameters related to the swing and oil level, and perform value filling, frequency reduction, null value processing and normalization on the input parameters, and use the LSTM multi-input multi-output model to predict the X-direction swing and Y-direction swing of the lower guide to obtain the predicted value of the lower guide swing as well as The residual between the actual value and the predicted value of the guide swing at time ; Use the LSTM multi-input single-output model to predict the oil level and obtain the predicted value of the lower oil level as well as The residual between the actual value and the predicted value of the oil level at time .

7. The method for evaluating the health status of a bearing system based on a hypersphere and refined parameters according to claim 1, characterized in that: In step 3.2, calculate the average cooling water flow rate under historical healthy conditions and standard deviation ,use Principle, determine whether the cooling water flow is abnormal, if it is abnormal, calculate the abnormal The minimum distance between the cooling water flow rate and the upper or lower limit of the interval at any given moment .

8. The method for evaluating the health status of a bearing system based on a hypersphere and refined parameters according to claim 1, characterized in that: In step 3.3, calculate the health of key parameters. The specific process is as follows: Based on the data obtained in steps 3.1 and 3.2 The residuals of each key parameter at the moment are constructed The residual matrix of each key parameter at the moment: ； use In principle, the allowable residual matrix of various key parameters is obtained: ； in, It represents the allowable residual of tile temperature, which is 3 times the standard deviation of the predicted value of tile temperature; It represents the allowable residual error of the swing, which is 3 times the standard deviation of the predicted value of the swing; It represents the allowable residual of oil level, which is 3 times the standard deviation of the predicted oil level value; It represents the allowable residual of cooling water flow, which is 3 times the standard deviation of cooling water flow under the historical healthy state; The difference between the alarm value and the predicted value of each key parameter is taken as the maximum residual of the parameter. The maximum residual matrix of each key parameter is: ； in, represents the maximum residual error of Wattage. represents the maximum residual error of the swing, Indicates the maximum residual error of the oil level, Indicates the maximum residual of cooling water flow; The health of each key parameter is determined by a quadratic parabolic membership function, namely: Health at all times: ； in, , , represents the number of key parameters of each category, Indicates The key parameters of the class The health of the parameters, Indicates The key parameters of the class The residuals of the parameters, Indicates The key parameters of the class The allowed residuals of the parameters, Indicates The key parameters of the class The maximum residual error of the parameters.

9. The method for evaluating the health status of a bearing system based on a hypersphere and refined parameters according to claim 1, characterized in that: In step 4, the initial weights of the health of each key parameter are determined using the independence weight method, and the final weights of the health of the bearing system and the health of each key parameter are determined using the TOPSIS method, which specifically includes the following sub-steps: Step 4.1: Use the independence weight method to determine the initial weights of the health of each key parameter. Expand it and get: ； in, Indicates the total number of key parameters; Before All health levels at all times, calculate the correlation coefficient between the health levels of each key parameter , then the initial weights of the health of each key parameter are: ； in, , For the The standard deviation of the health of key parameters; Normalize the weights to get the weights of the key parameter health: ； in, For the The weight of the health of the key parameters; Step 4.2: Use the TOPSIS method to determine the final weights of the derived bearing system health and the health of each key parameter; Derive initial weights for the health of the bearing system and key parameters ,in, To derive the bearing system health The initial weight of The initial weights obtained in step S51 are normalized to obtain weights: ； in, , After normalization, The health weight, , Respectively represent the maximum and minimum values ​​of the initial weights; No. The weight of each health degree is: ； The final health of the bearing system is derived as: 。

Images