Failure physics and data driving fused plunger pump life prediction method and system
By integrating failure physics and data-driven methods, a plunger pump degradation mechanism and neural network model were established, solving the problems of difficult physical modeling and data scarcity in plunger pump life prediction, and achieving high-precision life prediction and improved interpretability.
Patent Information
- Application Number
- CN202511560881.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-29
- Publication Date
- 2026-01-20
AI Technical Summary
Traditional methods for predicting the lifespan of piston pumps suffer from difficulties in physical modeling, data scarcity, and low data quality, making it difficult to achieve high-precision lifespan predictions.
A method combining physical failure and data-driven approaches is adopted. This method establishes a RUL prediction method based on the degradation mechanism of plunger pumps using physical features and physical information loss functions, which is fused with neural networks. This includes constructing a performance degradation model that characterizes wear and leakage of the slipper pair, and using the sparrow optimization algorithm to optimize a multi-feature LSTM neural network, which is then trained using sensor data.
It improves the accuracy and interpretability of plunger pump life prediction, ensures that the prediction results conform to physical laws, alleviates the small sample problem, and improves data utilization efficiency.
Smart Images
Figure CN121365481A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application relates to the technical field of residual life prediction of a plunger pump, in particular to a plunger pump life prediction method and system fusing failure physics and data driving. BACKGROUND
[0002] The plunger pump is a core power component of a hydraulic system, and its life prediction is crucial to guarantee the safe operation of the hydraulic system. However, due to the complex structure, high cost and harsh working environment of the plunger pump, physical modeling is difficult, data is scarce and data quality is low, and traditional methods based on single failure physics or data driving cannot obtain high-precision life prediction. The failure physics method has the advantage of high prediction accuracy, but the complex factors such as load, pressure distribution and geometric motion relationship of the plunger pump friction pair make it difficult to build a failure physics model. The data-driven method has a high dependence on data, but the plunger pump has a long life and high cost, and cannot be subjected to long-term and multi-sample degradation tests, making it difficult to obtain degradation and failure data. Moreover, the degradation data collected on site often contains a large amount of invalid information due to the complex and harsh working conditions, which brings great challenges to the high-precision prediction of the residual service life of the plunger pump. SUMMARY
[0003] In order to solve the problems of physical modeling difficulty, data scarcity and low data quality in the traditional plunger pump life prediction method, the application provides a plunger pump life prediction method and system fusing failure physics and data driving. The application establishes a plunger pump degradation mechanism and neural network fusion RUL prediction method based on physical characteristics and physical information loss function, to solve the small sample problem in plunger pump life prediction and the problem of deviation of prediction results from physical laws and low accuracy, and to realize high-precision life prediction of the plunger pump.
[0004] The technical scheme adopted by the plunger pump life prediction method and system fusing failure physics and data driving provided by the application is as follows:
[0005] A plunger pump life prediction method fusing failure physics and data driving comprises the following steps,
[0006] S1: Establish a whole-pump degradation mechanism model, taking the plunger pump shoe pair as the research object, first calculate the shoe pair oil film thickness based on the three-point oil film thickness method, then calculate the total wear of the shoe pair based on the Archard wear formula, and combine the pressure flow characteristics to construct a performance degradation model representing the mutual representation of shoe pair wear and leakage;
[0007] S2: Construct a sparrow optimization algorithm optimized multi-feature LSTM neural network, build a multi-dimensional time sequence LSTM neural network framework, initially set the number of network layers, the number of neurons, the learning rate and the maximum iteration number, normalize the data by maximum and minimum value, and save the iteration optimized LSTM hyperparameters by sparrow optimization algorithm;
[0008] S3: Screening and inputting physical guiding features, inputting the structure parameters, material parameters and working condition parameters of the plunger pump into the performance degradation model of S1, iteratively outputting the time sequence data of wear and leakage, screening high-quality physical guiding features through correlation evaluation and time dependence evaluation, and inputting the high-quality physical guiding features and the running data collected by the sensor into the multi-feature LSTM neural network constructed in S2;
[0009] S4: Constructing a physical information loss function and training the model, simplifying the physical information equation containing unknown parameters based on the performance degradation model of S1, solving the optimal parameters of the equation by gradient descent method, constructing data loss function and physical information loss function respectively, and obtaining composite total loss function by weighted fusion, training the multi-feature LSTM neural network by back propagation, and outputting the plunger pump life prediction result.
[0010] Further improvement of the technical scheme of the application is that the formula for calculating the oil film thickness of the sliding shoe pair in step S1 is:
[0011]
[0012] Wherein, (r, θ, h) is the coordinates of any point C on the bottom surface of the sliding shoe in the cylindrical coordinate system, R2 is the distribution circle radius of the plunger, Indicates the oil film thickness on the outer circumference of the sliding shoe at intervals of 120°.
[0013] Further improvement of the technical scheme of the application is that the formula for calculating the total wear of the sliding shoe in step S1 is:
[0014]
[0015] Wherein, Indicates the total wear of the sliding shoe, Indicates the number of plungers, Indicates the relative motion stroke of the sliding shoe relative to the swash plate.
[0016] Further improvement of the technical scheme of the application is that the performance degradation model of the sliding shoe pair wear and leakage is established based on the pressure flow characteristics in step S1, and the system performance degradation state equation is:
[0017] ;
[0018] Simplify and build physical equation:
[0019] ;
[0020] The plunger pump leakage model established after further simplification is:
[0021]
[0022] Wherein, The outer radius of the slipper sealing belt is represented by r, The inner radius of the slipper sealing belt is represented by R, The oil dynamic viscosity is represented by mu, The simplified parameter is represented by a, The slipper pair oil chamber pressure is represented by p, The shell oil pressure is represented by P, The slipper wear leakage accounts for the total leakage proportion, The wear speed is represented by V, The initial oil film thickness is represented by h0, , , , , The to-be-fitted parameter is represented by a, The time sequence is represented by t.
[0023] Further improvement of the technical scheme of the application is that the step S2 comprises the following steps,
[0024] S2.1, the sparrow optimization algorithm is used to optimize the hyperparameters of the model, after the first layer training is completed, the layer nodes of the first layer are regarded as the input nodes of the second layer, and the second layer is trained, and the subsequent layers are trained in turn.
[0025] S2.2, the positions of the producers, scavengers and early warners are updated according to the fitness value, and the individuals with better fitness are screened out, and the global optimal fitness value is updated. If the stop condition is not reached, the iterative optimization is continued until the condition is met, and the calculation is stopped, and the optimized LSTM hyperparameters are saved.
[0026] Further improvement of the technical scheme of the application is that the correlation evaluation in the step S3 adopts Pearson correlation coefficient calculation, and the formula is:
[0027]
[0028] Wherein, The Pearson correlation coefficient is represented by r, The values of two variables are represented by x and y, The mean value of the variable is represented by mu;
[0029] The time dependence evaluation adopts autocorrelation function calculation, and the formula is:
[0030]
[0031] wherein, denotes the autocorrelation coefficient of the time series at lag denotes the value of the time series at time denotes the mean value of the time series denotes the total length of the data of the time series, denotes the time difference at lag
[0032] A fusion failure physics and data-driven plunger pump life prediction system is used to implement the above prediction method, comprising a sensor data acquisition unit, a cloud database, a computer server, a remaining life prediction module and a webpage operation panel; wherein the sensor data acquisition unit is in communication connection with the computer server, and is used to acquire dynamic operation data of the plunger pump and transmit the dynamic operation data to the computer server; the cloud database is in communication connection with the computer server, and is used to store the dynamic operation data, static parameters and prediction results; the remaining life prediction module is integrated in the computer server, and is used to execute the steps of degradation model establishment, SSA-LSTM construction, feature screening and loss function calculation; the webpage operation panel is in communication connection with the computer server, and is used to input static parameters, configure prediction tasks and display prediction results.
[0033] Further improvement of the above technical scheme of the present application is that the sensor data acquisition unit comprises a pressure sensor, a flow sensor, a rotating speed sensor and a data acquisition card; wherein the pressure sensor comprises a shoe pair oil chamber pressure sensor and a shell oil pressure sensor; the flow sensor is used to acquire the whole pump leakage amount, the rotating speed sensor is used to acquire the cylinder body rotating speed, and the data acquisition card is used to convert an analog signal into a digital signal.
[0034] Further improvement of the above technical scheme of the present application is that the remaining life prediction module comprises a data preprocessing submodule, a failure mechanism modeling submodule, an SSA-LSTM training submodule, a feature screening submodule and a loss function calculation submodule; wherein the data preprocessing submodule is used to execute data cleaning, missing value completion and normalization processing; the failure mechanism modeling submodule is used to construct a performance degradation model and a whole pump leakage model; the SSA-LSTM training submodule is used to optimize LSTM hyperparameters based on the sparrow optimization algorithm and train the network; the feature screening submodule is used to screen high-quality physical guide features based on correlation and time dependence evaluation; and the loss function calculation submodule is used to construct a data loss function, a physical information loss function and a composite total loss function.
[0035] Due to the adoption of the above technical solutions, the technical progress achieved by the present application is:
[0036] The present application first establishes a sliding shoe wear mechanism model to explain the wear degradation process inside the sliding shoe, and further derives a degradation model of the entire pump, which simplifies the construction of the model and solves the difficulty of physical modeling of the piston pump. In long-term prediction tasks, multi-feature input is an effective solution to the small sample problem. Combining physical modeling with LSTM training, a physically guided high correlation feature is constructed. Not only does it make up for the lack of field data, but it also improves the accuracy and interpretability of the prediction results. The physically guided loss function ensures that the prediction results conform to the physical laws by adding physical constraints during neural network training. After introducing the physical loss, the neural network can adjust the prediction results during training to conform to the constraints of the physical model. For example, in the wear process of the piston pump, the leakage flow usually increases with time, while the traditional neural network may predict a non-physical downward trend due to insufficient data or noise. The physically guided neural network effectively avoids such problems by constraining the physical loss, ensuring that the prediction results conform to the real degradation process of the piston pump. In addition, the introduction of physical loss can also alleviate the small sample problem. Even with limited training data, the model can supplement prior information with the help of physical mechanisms to improve data utilization efficiency. BRIEF DESCRIPTION OF DRAWINGS
[0037] Figure 1 The flowchart of the present application is a fusion of failure physics and data-driven piston pump life prediction method;
[0038] Figure 2 The SSA algorithm optimization principle and flowchart in the embodiment of the present application;
[0039] Figure 3 The physical guided feature construction flowchart in the embodiment of the present application;
[0040] Figure 4 The physically guided loss function flowchart in the embodiment of the present application;
[0041] Figure 5 The real data and prediction data effect diagram of the piston pump from 800 to 1000 hours in the embodiment of the present application;
[0042] Figure 6 The RUL prediction result diagram of the piston pump over time in the embodiment of the present application. DETAILED DESCRIPTION
[0043] To make the purpose, technical solutions and advantages of the present application clearer and more explicit, the present application will be further described in detail below in combination with specific embodiments and with reference to the drawings. In the following description, the description of well-known structures and technologies is omitted to avoid unnecessary confusion of the concept of the present application.
[0044] Embodiment one
[0045] As Figure 1 shown, the embodiment provides a fusion failure physical and data-driven plunger pump life prediction method, comprising the following steps:
[0046] S1, on the basis of failure physical analysis of plunger pump, taking plunger pump slipper pair as the main research object, a slipper wear-leakage degradation model is constructed, and a whole pump leakage model is further established.
[0047] S1.1, first of all, according to the three-point oil film thickness method, the oil film thickness of the slipper pair is calculated, and the nominal oil film thickness corresponding to the bottom surface C point of the slipper can be expressed as:
[0048] ;
[0049] Wherein, (r, θ, h) is the coordinate of any point C on the bottom surface of the slipper in the cylindrical coordinate system, R2 is the distribution circle radius of the plunger, The oil film thickness on the outer circumference of the slipper is represented by 120°.
[0050] S1.2, based on Archard wear formula, the total wear volume of all slipper can be obtained when the cylinder rotates one revolution:
[0051] ;
[0052] Wherein, The total wear volume of the slipper is represented by 120°. The number of plungers is represented by 120°. The relative motion stroke of the slipper relative to the swash plate is represented by 120°.
[0053] S1.3, the leakage is generated through the micro gap between the slipper and the swash plate, and the pressure flow characteristic formula is:
[0054] ;
[0055] Wherein, The leakage flow of the slipper is represented by 120°. The oil room pressure of the slipper pair (MPa) is represented by 120°. The shell oil pressure (MPa) is represented by 120°. The outer radius of the slipper sealing band (m) is represented by 120°. The inner radius of the slipper sealing band (m) is represented by 120°. The oil dynamic viscosity (N·s / m2) is represented by 120°. The oil film thickness (m) is represented by 120°.
[0056] S1.4. Then, the performance degradation model of the mutual representation of the wear and leakage of the slipper pair was established based on the pressure-flow characteristics, and the system performance degradation state equation was:
[0057] ;
[0058] Simplification and construction of physical equations:
[0059] ;
[0060] The leakage model of the plunger pump established after further simplification is:
[0061]
[0062] wherein, represents the outer radius of the slipper sealing band (m), represents the inner radius of the slipper sealing band (m), represents the dynamic viscosity of the oil (N·s / m2), represents a simplified parameter, represents the oil chamber pressure of the slipper pair (MPa), represents the oil pressure of the shell (MPa), is the proportion of the slipper wear leakage in the total leakage, is the wear speed, is the initial oil film thickness, , , , , is a parameter to be fitted, is a time series.
[0063] S2. Based on the sparrow optimization algorithm, a multi-feature LSTM neural network was constructed.
[0064] S2.1. The model framework of the multi-dimensional time series neural network was constructed, wherein the LSTM includes forward propagation LSTM and back propagation LSTM.
[0065] The state update of the LSTM layer is as follows:
[0066]
[0067]
[0068]
[0069]
[0070]
[0071] wherein, are the activation vectors of the forget gate, input gate and output gate respectively, are the weights and biases in the model respectively, is the state at time step t-1, is the input at time step t, are the sigmoid function and hyperbolic tangent function respectively with values between 0 and 1.
[0072] S2.2, normalize and denormalize the original data:
[0073] The maximum size normalization method is used to normalize the data:
[0074]
[0075] Denormalization:
[0076] wherein, represents the maximum value in the data set, represents the minimum value in the data set, represents the normalized value, represents the original value.
[0077] S2.3, initially set the number of neurons in each layer to 50 and 100 respectively, the learning rate to 0.001, the maximum number of iterations to 100, the sliding window size to 10, the dropout layer size to 0.1, and the number of network layers to 2.
[0078] S2.4, the sparrow optimization algorithm is used to optimize the hyperparameters of the model. After the training of the first layer is completed, the layer nodes of the first layer are regarded as the input nodes of the second layer, and the second layer is trained. Similarly, the subsequent layers are trained in turn.
[0079] S2.5, then, update the positions of the producers, scavengers and early warners according to the fitness value, and select individuals with better fitness, and update the global optimal fitness value. If the stopping condition is not met, continue the iterative optimization until the condition is met, and stop the calculation, and save the optimized LSTM hyperparameters. The optimization principle and process of the SSA optimization algorithm are shown in Figure 2 .
[0080] S3, construct high-quality physical guidance features.
[0081] S3.1, based on the model of the sliding shoe pair wear degradation constructed in S1, simulate the degradation process of the whole pump.
[0082] S3.2, Then input the collected structural parameters, material parameters and working condition parameters of the plunger pump into the mechanism model and iterate the operation.
[0083] S3.3, Then output the data of wear and leakage changing with time under corresponding pressure and rotating speed;
[0084] Introduce correlation evaluation and time-dependent evaluation analysis screening;
[0085] Pears correlation coefficient: ;
[0086] Wherein, represents the value of two variables, represents the mean value of the variable;
[0087] Autocorrelation function: ;
[0088] Wherein, represents the autocorrelation coefficient of time series at lag , represents the value of the time series at the time, represents the mean value of the time series, represents the total length of the time series data, represents the time difference of lag period. S3.4, Evaluate the quality of physical features by calculating the time, slipper wear, slipper leakage and three correlation coefficients of leakage feature.
[0089] S3.5, Then these high-quality physical guide features are part of the LSTM training set, so that the model can learn physical laws and adapt to real sensor data at the same time. The construction process is shown in .
[0090] S3.5, Then these high-quality physical guide features are part of the LSTM training set, so that the model can learn physical laws and adapt to real sensor data at the same time. The construction process is shown in Figure 3 .
[0091] S4, Construct physical information loss function.
[0092] S4.1, Based on the leakage model of the whole pump constructed in S1, the parameters of the mathematical degeneration equation are simplified, and the physical information equation containing unknown parameters is established.
[0093] S4.2, Select gradient descent method as parameter optimization algorithm, estimate physical equation parameter set.
[0094] S4.3, Based on the error between physical model and real label data, construct loss function with penalty term for regularization, see formula: ;
[0095] wherein, is a regularization coefficient, is a regularization loss.
[0096] S4.4, the parameter update direction is obtained by taking the partial derivative of the regularized loss function, and a gradient calculation formula is constructed.
[0097] S4.5, the gradient is regularized, the freedom of the model is limited by introducing a penalty term in the loss function, which prevents the parameter optimization model from overfitting and improves the generalization ability, and the parameters are constrained and iterated through the gradient, which avoids gradient explosion or instability of the model. Based on the initial parameters, gradient descent method, learning rate, etc. to parameter iteration:
[0098] ;
[0099] wherein, is an initial parameter, is a learning rate.
[0100] S4.6, the leakage data collected by the sensor and the time are normalized, the units of input and output are adjusted, and the dimensional difference is eliminated.
[0101] S4.7, input the initial parameter estimation and hyperparameter, and estimate the parameter by gradient descent method, use regularization to correct the gradient, and use physical experience to limit the range of parameter update, based on the constructed mathematical equation of the degradation of the sliding shoe pair wear under ideal physical state, solve the parameter set with minimum loss, and apply it to the mathematical equation;
[0102] wherein, based on the leakage mechanism of plunger pump sliding shoe wear, the equation of unknown parameters is constructed, and the pseudo code of the parameter set solving process is shown in table 1:
[0103] Table 1 pseudo code for solving mechanism equation
[0104] Input: a leakage flow dataset , initial parameter values , an initial learning rate , a regularization coefficient , and a maximum number of iterations . 1: Step 1: Data pre-processing and initialization 2: Normalization of time and sensor tag data 3: Initialization of initial parameters, initialization of iteration counter, initialization of last round loss value 4: Definition of basic model of physical mathematical polynomial 5: Step 2: Gradient descent main loop 6: For n = 1 to N do: 7: Calculate loss function of current iteration round, mean square error + regularization 8: Compare optimal loss and optimal parameters under numerous iteration rounds 9: Calculate gradient and add regularization 10: Update parameters 11: Empirical limit on parameter range, 12: Judge whether the convergence condition is met, If the loss function meets the standard, Break 13: End for 14: Step 3: Check the iteration termination condition, if it does not converge within the maximum number of iterations, return the current optimal parameters. 15: Output optimal loss and optimal parameters, optimal parameter set is substituted into the equation and output, time is denormalized
[0105] S4.8, secondly, the data loss function and the physical information loss function are constructed by label data, physical data and prediction data respectively. Based on multi-dimensional evaluation index, the form is as follows:
[0106] ;
[0107] .
[0108] S4.9, finally, the composite total loss function losstotal is constructed. The data item and the physical item are weighted and fused, and the network parameters are trained and corrected in back propagation, which guides the optimization direction of the model;
[0109] ;
[0110] wherein, , , are the loss function of the neural network model, the weight of the data loss function, the weight of the physical information loss function, and the loss function flow of the physical guidance is shown in Figure 4 .
[0111] As shown in Figure 5 , the real data and the predicted data of the plunger pump for 800 to 1000 hours, the multi-feature time series neural network is constructed, after the optimization of the hyperparameters through the experiment and the SSA algorithm, the RUL prediction is carried out. It can be seen from the shape of the predicted curve of the pump that the multi-feature time series LSTM network can better predict the degradation trend of the plunger pump leakage data; at the same time, the predicted data is smoother, and the numerical value can accurately represent the real data. It is proved that the multi-feature neural network based on physics has an advantage in long-term prediction.
[0112] As shown in Figure 6 , taking the current hydraulic pump running time of 800 hours as the starting time point, using the constructed physical-data fusion model, the RUL of the plunger pump is iteratively predicted by updating the data, and it is consistent with the real remaining life.
[0113] Example Two
[0114] The embodiment provides a plunger pump remaining life prediction system fusing failure physics and data driving, which is used for realizing the above method, and comprises a sensor data acquisition unit, a cloud database, a computer server, a remaining life prediction module and a webpage operation panel; wherein the sensor data acquisition unit is in communication connection with the computer server, and is used for acquiring dynamic operation data of the plunger pump and transmitting the dynamic operation data to the computer server; the cloud database is in communication connection with the computer server, and is used for storing the dynamic operation data, static parameters and prediction results; the remaining life prediction module is integrated in the computer server, and is used for executing the steps of establishing a degradation model, constructing an SSA-LSTM, screening features and calculating a loss function; and the webpage operation panel is in communication connection with the computer server, and is used for inputting static parameters, configuring a prediction task and displaying a prediction result.
[0115] The above sensor data acquisition unit comprises a pressure sensor, a flow sensor, a rotating speed sensor and a data acquisition card; wherein the pressure sensor comprises a shoe pair oil chamber pressure sensor and a shell oil pressure sensor; the flow sensor is used for acquiring the leakage amount of the whole pump, the rotating speed sensor is used for acquiring the rotating speed of the cylinder body, and the data acquisition card is used for converting an analog signal into a digital signal.
[0116] The remaining life prediction module comprises a data preprocessing submodule, a failure mechanism modeling submodule, an SSA-LSTM training submodule, a feature screening submodule and a loss function calculation submodule; the data preprocessing submodule is used for performing data cleaning, missing value completion and normalization processing; the failure mechanism modeling submodule is used for constructing a performance degradation model and a whole pump leakage model; the SSA-LSTM training submodule is used for optimizing LSTM hyperparameters and training the network based on the sparrow optimization algorithm; the feature screening submodule is used for screening high-quality physical guidance features based on correlation and time dependence evaluation; and the loss function calculation submodule is used for constructing a data loss function, a physical information loss function and a composite total loss function.
[0117] In the above embodiment, the application provides a fusion failure physical and data-driven plunger pump life prediction method and system. The application first establishes a sliding shoe wear mechanism model to explain the wear degradation process inside the sliding shoe, and further derives the degradation model of the whole pump. From the whole to the local, the construction of the model is simplified, and the difficulty of physical modeling of the plunger pump is solved. In the long-term prediction task, multi-feature input is an effective solution to the small sample problem. The combination of physical modeling and LSTM training constructs a physical guidance high-correlation feature. Not only does it make up for the lack of field data, but also improves the accuracy and interpretability of the prediction results. The physical guidance loss function adds physical constraints in neural network training, ensuring that the prediction results conform to the physical law. After introducing the physical loss, the neural network can adjust the prediction results during the training process to make them conform to the constraints of the physical model. For example, in the wear process of the plunger pump, the leakage flow usually increases with time, while the traditional neural network may predict a non-physical downward trend due to insufficient data or noise. The physical guidance neural network effectively avoids such problems by the constraint of the physical loss, ensuring that the prediction results conform to the real degradation process of the plunger pump. In addition, the introduction of physical loss can also alleviate the small sample problem. Even with limited training data, the model can supplement prior information with the help of physical mechanisms to improve data utilization efficiency.
[0118] The above-described embodiments are merely preferred embodiments of the application and do not limit the concept and scope of the application. Various modifications and improvements to the technical solutions of the application made by those of ordinary skill in the art without departing from the design concept of the application shall fall within the protection scope of the application. The technical content claimed by the application has been fully recorded in the claims.
Claims
1. A fused failure physics and data-driven plunger pump life prediction method, characterized by: The method comprises the following steps, S1: establishing a whole-pump degradation mechanism model, taking a plunger pump sliding shoe pair as the research object, first calculating the oil film thickness of the sliding shoe pair based on the three-point oil film thickness method, then calculating the total wear of the sliding shoe based on the Archard wear formula, and combining the pressure flow characteristics to build a performance degradation model representing the wear and leakage of the sliding shoe pair; S2: constructing a multi-feature LSTM neural network optimized by the sparrow optimization algorithm, building a multi-dimensional time series LSTM neural network architecture, initially setting the number of network layers, the number of neurons, the learning rate and the maximum number of iterations, normalizing the data using the maximum and minimum values, and saving the LSTM hyperparameters optimized by the sparrow optimization algorithm; S3: screening and inputting physical guiding features, inputting the structural parameters, material parameters and working condition parameters of the plunger pump into the performance degradation model of S1, iteratively outputting time series data of wear and leakage, screening high-quality physical guiding features through correlation evaluation and time dependence evaluation, and inputting them together with the operating data collected by the sensor into the multi-feature LSTM neural network constructed in S2; S4: constructing a physical information loss function and training the model, simplifying the performance degradation model of S1 to obtain a physical information equation containing unknown parameters, solving the optimal parameters of the equation using the gradient descent method, constructing a data loss function and a physical information loss function respectively, and obtaining a composite total loss function by weighted fusion, training the multi-feature LSTM neural network through back propagation, and outputting the plunger pump life prediction result.
2. The fused failure physics and data-driven plunger pump life prediction method of claim 1, wherein, The formula for calculating the oil film thickness of the sliding shoe pair in step S1 is: Wherein, (r, θ, h) is the coordinate of any point C on the bottom surface of the sliding shoe in the cylindrical coordinate system, R2 is the distribution circle radius of the plunger, The oil film thickness on the outer circumference of the sliding shoe is represented by 120°.
3. The fused failure physics and data-driven plunger pump life prediction method of claim 1, wherein, The formula for calculating the total wear of the sliding shoe in step S1 is: wherein, represents the total wear amount of the sliding shoe, represents the number of plungers, represents the relative movement stroke of the sliding shoe relative to the swash plate.
4. The fused failure physics and data-driven plunger pump life prediction method and system of claim 1, wherein, The performance degradation state equation of the system is: ； Simplify and build a physical equation: ； The plunger pump leakage model established after further simplification is: wherein, represents the outer radius of the shoe seal band, represents the inner radius of the shoe seal band, represents the oil dynamic viscosity, represents a simplified parameter, represents the shoe pair oil chamber pressure, represents the housing oil pressure, is the proportion of total leakage occupied by the shoe wear leakage, is the wear speed, is the initial oil film thickness, , , , , is the parameter to be fitted, is the time sequence.
5. The fused failure physics and data-driven plunger pump life prediction method of claim 1, wherein: The step S2 comprises the following steps, S2.1, the hyperparameters of the model are optimized by the sparrow optimization algorithm, after the first layer is trained, the layer nodes of the first layer are regarded as the input nodes of the second layer, and the second layer is trained; the subsequent layers are trained in turn. S2.2, update the positions of the producers, scavengers and early warners according to the fitness value, select the individuals with better fitness, and update the global optimal fitness value. If the stop condition is not met, continue to iterate and optimize until the condition is met, and then stop calculation and save the optimized LSTM hyperparameters.
6. The fused failure physics and data-driven plunger pump life prediction method of claim 1, wherein, The correlation evaluation in step S3 is calculated by using the Pearson correlation coefficient, and the formula is: wherein, denotes the Pearson correlation coefficient, denotes the value of two variables, denotes the mean value of a variable; The time dependence evaluation is calculated by using the autocorrelation function, and the formula is: wherein, denotes the autocorrelation coefficient of the time series at lag denotes the autocorrelation coefficient of the time series at lag denotes the value of the time series at time denotes the value of the time series at time denotes the value of the time series at time denotes the mean value of the time series denotes the mean value of the time series denotes the total length of the data of the time series denotes the time difference of the lag 7. A fused failure physics and data-driven plunger pump life prediction system, characterized by: The application discloses a prediction method for realizing any one of claims 1-6, comprising a sensor data acquisition unit, a cloud database, a computer server, a residual life prediction module and a webpage operation panel; wherein the sensor data acquisition unit is in communication connection with the computer server, is used for acquiring dynamic operation data of the plunger pump and transmitting the dynamic operation data to the computer server; the cloud database is in communication connection with the computer server, is used for storing the dynamic operation data, static parameters and prediction results; the residual life prediction module is integrated in the computer server, is used for executing the steps of degradation model establishment, SSA-LSTM construction, feature screening and loss function calculation; the webpage operation panel is in communication connection with the computer server, is used for inputting the static parameters, configuring the prediction task and displaying the prediction results.
8. The fused failure physics and data-driven plunger pump life prediction system of claim 7, wherein: The sensor data acquisition unit comprises a pressure sensor, a flow sensor, a rotating speed sensor and a data acquisition card; wherein the pressure sensor comprises a sliding shoe pair oil chamber pressure sensor and a shell oil pressure sensor; the flow sensor is used for acquiring the whole pump leakage amount, the rotating speed sensor is used for acquiring the cylinder body rotating speed, and the data acquisition card is used for converting an analog signal into a digital signal.
9. The fused failure physics and data-driven plunger pump life prediction system of claim 7, wherein: The residual life prediction module comprises a data preprocessing submodule, a failure mechanism modeling submodule, an SSA-LSTM training submodule, a feature screening submodule and a loss function calculation submodule; wherein the data preprocessing submodule is used for executing data cleaning, missing value completion and normalization processing; the failure mechanism modeling submodule is used for constructing a performance degradation model and a whole pump leakage model; the SSA-LSTM training submodule is used for optimizing LSTM hyperparameters and training a network based on a sparrow optimization algorithm; the feature screening submodule is used for screening high-quality physical guide features based on correlation and time dependence evaluation; and the loss function calculation submodule is used for constructing a data loss function, a physical information loss function and a composite total loss function.