Fuel injector solenoid valve reliability prediction method based on multi-physics field coupling

Abstract

The invention provides a fuel injector solenoid valve reliability prediction method based on multi-physics field coupling, and relates to the technical field of solenoid valve reliability prediction, and the method comprises the steps: collecting real-time operation data and historical operation data, constructing a multi-physics field coupling model of a solenoid valve, sequentially inputting the historical operation data into the model, and carrying out the prediction of the reliability of the solenoid valve. A finite element analysis method is used for solving the model, multi-physical field distribution characteristics of the electromagnetic valve are obtained, historical reliability parameters are extracted from the multi-physical field distribution characteristics, category labels are added to the historical reliability parameters, real-time operation data are input into the multi-physical field coupling model, and real-time reliability parameters are obtained; and inputting the real-time reliability parameter into the BI-LSTM model, and outputting a prediction result. According to the method, the multi-physical field model and the machine learning model are combined, and the mechanism advantages of the physical model and the data driving advantages of the machine learning model can be comprehensively utilized, so that the reliability of the electromagnetic valve is predicted more accurately.

Description

Technical Field

[0001] The present application relates to the technical field of predicting reliability of solenoid valves, and in particular to a reliability prediction method for fuel injector solenoid valves based on multi-physical field coupling. Background Art

[0002] In the field of new energy internal combustion engines, whether it is the high-pressure common rail diesel fuel injectors and gasoline direct injection (GDI) fuel injectors used in hybrid vehicles (including plug-in hybrid electric vehicles (PHEVs), extended-range hybrid electric vehicles (REEVs), etc.), or gas fuel injectors suitable for clean energy internal combustion engines such as natural gas and hydrogen, the reliability of their solenoid valves is crucial to the stable operation of the entire system.

[0003] The Chinese invention patent application with publication number CN110852014A uses theoretical knowledge to establish the electromagnetic differential equations and valve core dynamics equations of the solenoid valve dynamic process based on multiple physical fields involved in the solenoid valve; based on the established mathematical and physical model, finite element software is used to calculate some of the physical field input parameters required in the equation; and the solenoid valve mathematical model is then jointly simulated through mathematical calculation software to further study the performance laws of the solenoid valve.

[0004] The above patent only relies on specific software calculations to obtain limited physical field parameters, and does not take into account that in new energy internal combustion engines, the reliability of the fuel injector solenoid valve not only depends on its own physical structure and working principle, but is also closely related to the operating status of the entire internal combustion engine system.

[0005] Since this patent does not take into account the operating conditions of new energy internal combustion engines, it is difficult to accurately predict the reliability of the fuel injector solenoid valve in actual use, and cannot provide effective support for the maintenance and fault prevention of new energy internal combustion engines. Summary of the Invention

[0006] In order to improve the accuracy of reliability prediction of fuel injector solenoid valves in new energy internal combustion engines, the present application provides a fuel injector solenoid valve reliability prediction method based on multi-physics field coupling.

[0007] This application adopts the following technical solutions: The reliability prediction method of a fuel injector solenoid valve based on multi-physics field coupling includes the following steps: Data acquisition: Collect real-time operating data of the solenoid valve and multiple sets of historical operating data of similar solenoid valves to build a multi-physics field coupling model of the solenoid valve. The operating data includes electromagnetic force data, temperature data, fluid pressure data, vibration data, operating voltage, and operating current. Extracting reliability parameters: Each set of historical operating data is sequentially input into a multi-physics field coupling model, and the multi-physics field coupling model is solved using a finite element analysis method to obtain the multi-physics field distribution characteristics of the solenoid valve. Historical reliability parameters are extracted from the multi-physics field distribution characteristics and category labels are added to the historical reliability parameters. The historical reliability parameters include: the temperature gradient of the electromagnetic coil, the stress concentration factor of the valve core, and the pressure pulse amplitude of the fluid flow; Prediction: Input real-time operation data into the multi-physics field coupling model to obtain real-time reliability parameters. Use the machine learning algorithm to build a BI-LSTM model. Use historical reliability parameters and category labels to train the BI-LSTM model to obtain a trained BI-LSTM model. Input the real-time reliability parameters into the trained BI-LSTM model and output the prediction results.

[0008] This application first collects real-time solenoid valve operating data (electromagnetic force, temperature, fluid pressure, vibration, operating voltage, and operating current) as well as multiple sets of historical operating data for similar solenoid valves. This application then uses this collected data to construct a multi-physics coupling model. This model comprehensively considers the interactions between multiple physical fields, such as electromagnetic, thermal, fluid, and mechanical fields, to more realistically reflect the actual operating conditions of the solenoid valve. Subsequently, this application solves the multi-physics coupling model using finite element analysis to obtain the multi-physics distribution characteristics of the solenoid valve. Historical reliability parameters are extracted from these multi-physics distribution characteristics and assigned category labels to these historical reliability parameters. Subsequently, this application inputs the real-time operating data into the multi-physics coupling model to obtain real-time reliability parameters. A machine learning algorithm is then used to construct a Bidirectional Long Short-Term Memory (BI-LSTM) model. The BI-LSTM model is trained using the historical reliability parameters and category labels, enabling it to learn patterns and features in the historical data. This application then inputs the real-time reliability parameters into the trained BI-LSTM model and outputs a prediction indicating whether the solenoid valve is currently reliable. This application combines multi-physics field models and machine learning models, and can comprehensively utilize the mechanism advantages of physical models and the data-driven advantages of machine learning models to more accurately predict the reliability of solenoid valves.

[0009] Optionally, the method further includes: Simulation: Build a digital twin model, embed the multi-physics field coupling model into the digital twin model, and input historical operating data into the digital twin model in chronological order of collection time. The digital twin model simulates the degradation process of the solenoid valve and obtains simulated degradation data of the solenoid valve at different stages. Simulation judgment: obtain the real degradation data at the same time as the simulated degradation data, calculate the difference between the simulated degradation data and the real degradation data, record it as the first data, and determine whether the first data is greater than the preset difference threshold. If so, correct the digital twin model and re-acquire the simulated degradation data based on the corrected digital twin model; if not, execute the step of calculating the degradation degree; Calculating the degree of degradation: Obtaining the operating data of a preset time period before the current moment, recording it as the second data, inputting the second data and the real-time operating data into the digital twin model in the order of acquisition time to obtain predicted degradation data, and calculating the predicted degree of degradation based on the predicted degradation data; Degradation judgment: Determine whether the predicted degradation degree is greater than the preset degradation degree threshold. If so, a degradation alarm signal is issued; if not, this step is executed again after a preset time interval.

[0010] This application creates a virtual solenoid valve model by constructing a digital twin model and embedding a multi-physics field coupling model into it. This model can simulate the physical characteristics and behavior of the solenoid valve under various operating conditions. The digital twin model is highly flexible and scalable, and can easily integrate different physical models and data sources, providing a powerful platform for degradation simulation of solenoid valves. Subsequently, this application inputs the historical operating data into the digital twin model in the order of acquisition time, so that the digital twin model can simulate the degradation process of the solenoid valve during actual operation and obtain simulated degradation data of different periods. By adopting the above scheme, this application can fully consider the historical operating information of the solenoid valve, so that the simulation results are closer to the actual situation.

[0011] This application also obtains real degradation data at the same time as the simulated degradation data, calculates the difference between the simulated and real degradation data, and determines whether the difference exceeds a preset difference threshold. By adopting this approach, the application can promptly detect errors in the digital twin model and, by adjusting the parameters or structure of the digital twin model, more accurately reflect the actual degradation of the solenoid valve, thereby improving the reliability and accuracy of the digital twin model. When the difference exceeds the preset threshold, the digital twin model is corrected and simulated degradation data is reacquired based on the corrected digital twin model. This process continuously optimizes the model, gradually bringing the simulation results of the digital twin model closer to the actual situation. Subsequently, the application obtains operating data (i.e., second data) for a preset period of time before the current time, and inputs the second data and real-time operating data into the digital twin model in the order of acquisition time to obtain predicted degradation data. Based on the predicted degradation data, a predicted degradation degree is calculated. The predicted degradation degree is then determined to be greater than a preset degradation degree threshold. If so, a degradation alarm signal is issued, enabling personnel to promptly identify potential faults and degradation risks of the solenoid valve.

[0012] Optionally, the method further includes: Update the model: Increase the mesh density of the target area in the digital twin model to obtain a new digital twin model, extract the new multi-physics field coupling model in the new digital twin model, and perform the steps of extracting reliability parameters.

[0013] In the digital twin model, the grid density directly affects the accuracy of the simulation. By increasing the grid density of the target area, the present application can divide the calculation units of the target area more finely and capture more subtle physical phenomena and changes. The target area is often the area where problems are prone to occur or key performance indicators are concentrated during the operation of the solenoid valve. By increasing the grid density of this area, the simulation and analysis capabilities of these local areas can be enhanced, and the physical characteristics and behavior laws of the local areas can be deeply understood. The new digital twin model obtained after increasing the grid density is an optimization and improvement of the original model. While retaining the overall framework and functions of the original model, the new digital twin model performs a more detailed simulation of the target area, which can better reflect the actual operation of the solenoid valve. Since the grid density of the new digital twin model in the target area has changed, the distribution and interaction of its internal physical fields will also change accordingly. Therefore, the present application also extracts a new multi-physics field coupling model in the new digital twin model, so that the new multi-physics field coupling model can reflect these changes in a timely manner and improve the timeliness and accuracy of the analysis results. Subsequently, the present application uses the new multi-physics field coupling model to extract reliability parameters and obtain more accurate results. The new multi-physics coupling model simulates the operating state of the solenoid valve more precisely and can calculate reliability parameters more accurately, thereby improving the accuracy of the BI-LSTM model training samples and, in turn, the accuracy of the prediction results.

[0014] Optionally, the method further includes: Establishing a degradation model: Draw a degradation curve based on the simulated degradation data, and establish a degradation model for fitting the degradation curve. The calculation model of the degradation model is as follows: ; in, is the degree of degradation at time t; for the degree of degradation of the moment; is a parameter determined by nonlinear least squares method The value of Calculation reliability: based on parameters Calculate and output the reliability of the solenoid valve. The calculation model is as follows: ; ; Among them, n is the intermediate variable; is the service life of the solenoid valve.

[0015] This application draws a degradation curve and establishes a degradation model based on real degradation data, making full use of the large amount of data accumulated during historical operation. These data contain information on the performance changes of the solenoid valve under different operating conditions, and can truly reflect the degradation law of the solenoid valve. By establishing a model in a data-driven manner, this application reduces the deviation that may be caused by relying solely on theoretical assumptions and empirical formulas, and improves the accuracy and reliability of the degradation model. Subsequently, this application uses a nonlinear least squares method to determine the parameters in the degradation model, which can enable the degradation model to better fit the actual degradation curve. Subsequently, this application calculates the reliability based on the parameters in the degradation model, and can combine the degree of degradation with the length of use of the solenoid valve to reflect the reliability status of the solenoid valve at different stages of use. By calculating the reliability, the probability of the solenoid valve operating normally within a specific time can be quantified, providing a scientific basis for equipment maintenance decisions.

[0016] Optionally, the method further includes: Construct an evaluation matrix: Obtain the vectors output by the BI-LSTM model after each round of training, construct an evaluation matrix based on the vectors, and normalize the rows or columns of the evaluation matrix to obtain a processed evaluation matrix, where the element in the i-th row and j-th column of the evaluation matrix represents the probability of misclassifying the i-th class as the j-th class; Obtaining a data set: Integrate the off-diagonal elements in the rows or columns of the evaluation matrix into a data set; Calculate the difference between any two off-diagonal elements in the data set, record it as the third data, and execute the sample generation step for the third data that is less than a preset threshold until all data sets are traversed; Generate samples: Use the GANs model to generate training samples corresponding to the third data, and use the training samples to retrain the trained BI-LSTM model to obtain a retrained BI-LSTM model, where the training samples include new reliability parameters and new category labels; In the prediction step, the real-time reliability parameters are input into the retrained BI-LSTM model.

[0017] This application obtains the vectors output by the BI-LSTM model after each round of training and constructs an evaluation matrix based on these vectors. The element in the i-th row and j-th column of the evaluation matrix represents the probability of misclassifying the i-th category as the j-th category. This construction method can intuitively reflect the classification performance of the BI-LSTM model between different categories. Subsequently, this application standardizes the rows or columns of the evaluation matrix to make the misclassification probabilities between different categories comparable. Subsequently, this application integrates the off-diagonal elements in the rows or columns of the evaluation matrix into a data set. The off-diagonal elements represent the misclassification of the model between different categories. By integrating these elements, the classification performance of the model between easily confused categories can be analyzed in a focused manner.

[0018] Subsequently, the present application calculates the difference between any two off-diagonal elements in the dataset, recording this as the third data. The sample generation step is then performed for any third data that is less than a preset threshold, identifying instances where the BI-LSTM model has similar classification performance between easily confused categories, i.e., category pairs whose difference is less than the preset threshold. When the difference is less than the preset threshold, a GANs model is used to generate training samples for the two categories corresponding to the third data. The GANs model has powerful generative capabilities and can generate samples with a distribution similar to that of real data. Generating these samples increases the amount of training data for the BI-LSTM model between these easily confused categories, helping the model better learn the characteristics of these categories and improving classification performance. Subsequently, the present application uses the generated training samples to retrain the trained BI-LSTM model, obtaining a retrained BI-LSTM model. The retraining process enables the BI-LSTM model to adapt to the newly generated samples, further optimizing model parameters and improving the BI-LSTM model's ability to discriminate between easily confused categories. The training samples include new reliability parameters and new category labels, which helps the BI-LSTM model learn richer feature information and improve the model's generalization ability. During the prediction step, this application inputs the real-time reliability parameters into the retrained BI-LSTM model. Because the retrained BI-LSTM model has been optimized for easily confused categories, it can more accurately predict the category to which the real-time reliability parameters belong, improving the accuracy and reliability of the prediction results.

[0019] Optionally, the values of the diagonal elements of the evaluation matrix are 0.

[0020] Optionally, the method further includes: Obtaining a vector: splicing the vectors into a characteristic matrix, calculating the maximum eigenvalue and the corresponding eigenvector of the characteristic matrix, obtaining the vector corresponding to the maximum value of the elements in the eigenvector, and recording it as the first vector; Correct judgment: determine whether the historical reliability parameter and category label corresponding to the first vector are correct. If so, divide the evaluation matrix according to the position of the maximum value in the eigenvector to obtain two sub-matrices; if not, reset the category label of the historical reliability parameter corresponding to the first vector, retrain the BI-LSTM model, and execute the step of constructing the evaluation matrix; Vector screening: Based on the vectors in the two sub-matrices, the loss values of the trained BI-LSTM model are obtained. The vectors in the sub-matrix with the smaller loss value are retained, and the vectors in the sub-matrix with the larger loss value are deleted.

[0021] This application concatenates the vectors output by the BI-LSTM model into a feature matrix and calculates its maximum eigenvalue and corresponding eigenvector, thereby obtaining the vector corresponding to the maximum value of the element in the eigenvector (the first vector). Subsequently, this application determines whether the historical reliability parameter and category label corresponding to the first vector are correct. If the judgment is correct, the evaluation matrix is partitioned according to the position of the maximum value in the eigenvector, resulting in two sub-matrices. If the judgment is incorrect, the category labels need to be reset and the model retrained. Subsequently, this application partitions the evaluation matrix according to the position of the maximum value in the eigenvector, resulting in two sub-matrices. Based on the vectors in each sub-matrix, the loss value of the trained BI-LSTM model is calculated. Vectors in the sub-matrix with the lowest loss value are retained, while vectors in the sub-matrix with the highest loss value are deleted. The loss value is an important indicator of how well the BI-LSTM model fits the vector. A lower loss value indicates a better fit for the vector, and the vector is more helpful in improving the performance of the BI-LSTM model. This screening method can remove poor-quality vectors and retain higher-quality vectors, thereby improving the quality of the evaluation matrix.

[0022] Optionally, the GANs model is trained using the historical reliability parameters and the category labels to obtain a trained GANs model; In the step of generating samples, the trained GANs model is used to generate training samples of two categories corresponding to the third data.

[0023] This application uses historical reliability parameters and category labels to train the GANs model, fully leveraging the rich historical data resources available. Through this approach, the GANs model can learn the distribution characteristics and patterns of samples of different categories in the historical data, thereby generating high-quality training samples.

[0024] Optionally, a PCA algorithm is used to reduce the dimensions of the historical reliability parameters and the real-time reliability parameters, and the historical reliability parameters and the real-time reliability parameters after the dimension reduction are used as new historical reliability parameters and new real-time reliability parameters.

[0025] Historical and real-time reliability parameters typically contain multiple characteristic dimensions. The PCA algorithm projects the original data into a new coordinate system through a linear transformation and selects the first few principal components with the largest variance as new characteristic dimensions, effectively reducing the data's dimensionality. This dimensionality reduction reduces redundant information while retaining the data's key features, making the data more concise and easier to process.

[0026] In summary, this application includes at least one of the following beneficial technical effects: 1. This application combines multi-physics field models and machine learning models, and can comprehensively utilize the mechanism advantages of physical models and the data-driven advantages of machine learning models to more accurately predict the reliability of solenoid valves.

[0027] 2. This application creates a virtual solenoid valve model by constructing a digital twin model and embedding a multi-physics field coupling model into it, which can simulate the physical characteristics and behaviors of the solenoid valve under various operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a flow chart of Example 1 of the present application; Figure 2 This is a flowchart of the steps from S21 simulation to S25 degradation determination in Example 2 of the present application; Figure 3 This is a flowchart of Example 3 of the present application. DETAILED DESCRIPTION

[0029] The following combination Figures 1 to 3 This application is described in further detail.

[0030] Example 1: This example discloses a fuel injector solenoid valve reliability prediction method based on multi-physics field coupling, referring to Figure 1 The method includes: S11 data collection, S12 reliability parameter extraction, and S13 prediction. First, the real-time operation data of the solenoid valve and multiple sets of historical operation data of the same type (including electromagnetic force, temperature, fluid pressure, vibration, operating voltage and current, etc.) are collected to build a multi-physics field coupling model; then, each set of historical operation data is input into the multi-physics field coupling model, and finite element analysis is used to solve the multi-physics field distribution characteristics, from which historical reliability parameters such as the electromagnetic coil temperature gradient, valve core stress concentration factor, and fluid flow pressure pulse amplitude are extracted and category labels are added; finally, the real-time operation data is input into the model to obtain real-time reliability parameters, and a BI-LSTM model is constructed using a machine learning algorithm. The model is trained with historical reliability parameters and labels, and the real-time reliability parameters are then input into the trained model to output prediction results. The process of this embodiment is as follows: S11 data acquisition collects real-time operating data of the solenoid valve and multiple sets of historical operating data of the same type of solenoid valves to build a multi-physics field coupling model of the solenoid valve. The operating data includes electromagnetic force data, temperature data, fluid pressure data, vibration data, operating voltage and operating current.

[0031] Real-time operating data reflects the current working status of the solenoid valve, including electromagnetic force data (reflecting the size and changes of the electromagnetic force when the solenoid valve is working, affecting the movement of the valve core), temperature data (reflecting the thermal status of the solenoid valve as a whole and key parts (such as the solenoid coil). Excessive temperature may affect the performance and life of the solenoid valve), fluid pressure data (related to the fluid system controlled by the solenoid valve. Abnormal pressure may cause seal failure or obstruction of valve core movement), vibration data (may be caused by the movement of internal parts of the solenoid valve, fluid impact, etc. Excessive vibration may cause parts to loosen or fatigue damage), working voltage and working current data (provide energy for the solenoid valve. Its fluctuation or abnormality will affect the electromagnetic performance and heating of the solenoid valve).

[0032] This step also collects multiple sets of historical operating data of the same type of solenoid valves. These data can provide long-term operating experience and trends, and help to explore the operating rules of the solenoid valves under different working conditions.

[0033] In actual operation, solenoid valves experience complex interactions among multiple physical fields, including electromagnetic, temperature, flow, and structural fields. For example, electromagnetic force acts on the valve core, causing it to move. This movement in turn causes changes in the flow field, which in turn alters the temperature distribution. This temperature change, in turn, affects electromagnetic properties such as the resistance of the electromagnetic coil. A multi-physics coupling model comprehensively considers and models these multiple physical fields to more accurately describe the actual operation of the solenoid valve. This embodiment utilizes professional simulation software (such as ANSYS and COMSOL) based on relevant theories of electromagnetics, thermodynamics, fluid mechanics, and solid mechanics to establish governing equations for each physical field and couple them together. For example, in electromagnetic field calculations, the electromagnetic force is calculated based on Maxwell's equations; in temperature field calculations, factors such as heating in the electromagnetic coil and convective heat transfer in the fluid are considered; in flow field calculations, the fluid flow is simulated based on the Navier-Stokes equations; and in structural field calculations, the stress distribution of components such as the valve core is analyzed based on stress-strain relationships.

[0034] S12 extracts reliability parameters, inputs each set of historical operating data into the multi-physics field coupling model in turn, and uses the finite element analysis method to solve the multi-physics field coupling model to obtain the multi-physics field distribution characteristics of the solenoid valve, extracts historical reliability parameters from the multi-physics field distribution characteristics, and adds category labels to the historical reliability parameters. The historical reliability parameters include: the temperature gradient of the electromagnetic coil, the stress concentration coefficient of the valve core, and the pressure pulse amplitude of the fluid flow.

[0035] Each set of historical operating data is input into the constructed multi-physics field coupling model in turn. The finite element analysis method discretizes the continuous physical field into a finite number of units, calculates and analyzes each unit, and then combines them to obtain the distribution of the entire physical field. For example, in the temperature field analysis, the solenoid valve is divided into many small temperature units, and the temperature value of each unit is calculated to obtain the temperature distribution cloud map of the entire solenoid valve. By solving the multi-physics field coupling model through finite element analysis, this embodiment can intuitively understand the multi-physics field distribution characteristics of the solenoid valve under different historical operating conditions, such as the distribution of the electromagnetic field, the gradient of the temperature field, the pressure distribution and velocity distribution of the flow field, the stress distribution of the structural field, etc.

[0036] Representative historical reliability parameters are extracted from the multi-physics field distribution characteristics. The solenoid coil's temperature gradient reflects temperature variations within the coil. Large temperature gradients can cause local overheating of the coil, affecting its insulation performance and lifespan. The valve core's stress concentration factor indicates the degree of stress concentration on the valve core, making areas of stress concentration prone to fatigue failure. The fluid flow pressure pulse amplitude reflects fluid pressure fluctuations. Excessive pressure pulses can lead to seal failure or increased valve core vibration.

[0037] The extracted historical reliability parameters are labeled by category, for example, based on the solenoid valve's operating conditions (e.g., normal operating conditions, light-load operating conditions, heavy-load operating conditions) or fault types (e.g., seal failure, valve core jamming). These labels provide supervision information for subsequent machine learning model training, helping the model learn the characteristics of reliability parameters under different operating conditions and faults.

[0038] S13 prediction inputs real-time operation data into the multi-physics field coupling model to obtain real-time reliability parameters, uses a machine learning algorithm to build a BI-LSTM model, uses historical reliability parameters and category labels to train the BI-LSTM model, obtains a trained BI-LSTM model, inputs real-time reliability parameters into the trained BI-LSTM model, and outputs prediction results.

[0039] This embodiment inputs the collected real-time operating data into the multi-physics field coupling model, and also uses the finite element analysis method to solve the multi-physics field coupling model, obtains the multi-physics field distribution characteristics of the solenoid valve in the real-time operating state, and extracts real-time reliability parameters from it. The real-time reliability parameters reflect the current working reliability and potential risks of the solenoid valve.

[0040] The bidirectional long short-term memory (BI-LSTM) network is an improved recurrent neural network (RNN) that can simultaneously consider past and future information. In solenoid valve reliability prediction, the BI-LSTM model can learn the temporal evolution of historical reliability parameters and the correlations between different parameters. For example, the historical temperature gradient trend may be correlated with the future stress concentration of the valve core. The BI-LSTM model is able to capture this complex temporal relationship.

[0041] This example uses the extracted historical reliability parameters and corresponding category labels to train a BI-LSTM model. During training, the BI-LSTM model continuously adjusts its internal parameters to minimize the error between the predicted historical data and the actual category labels. This training allows the BI-LSTM model to learn the characteristics and variation patterns of the solenoid valve reliability parameters.

[0042] In this embodiment, the acquired real-time reliability parameters are input into a trained BI-LSTM model. The BI-LSTM model predicts the reliability of the solenoid valve based on the learned knowledge, and obtains the probability of the solenoid valve failing in the future, the possible types of failures, etc.

[0043] This embodiment first collects real-time operating data of the solenoid valve (including electromagnetic force, temperature, fluid pressure, vibration, operating voltage and current, etc.) and multiple sets of historical operating data of similar solenoid valves to construct a coupling model that comprehensively considers the interaction of multiple physical fields, such as electromagnetic field, temperature field, flow field, and structural field. Next, using finite element analysis, the historical operating data is input into the coupling model to solve the multi-physical field distribution characteristics. Historical reliability parameters such as the solenoid coil temperature gradient, valve core stress concentration factor, and fluid pressure pulse amplitude are extracted from the coupled model, and category labels are added according to operating conditions or fault types. Finally, the real-time operating data is input into the coupling model to obtain real-time reliability parameters. The historical reliability parameters and labels are constructed and trained using the BI-LSTM model. The real-time reliability parameters are then input into the trained model to output prediction results, thereby realizing reliability prediction of the fuel injector solenoid valve.

[0044] Example 2: Reference Figure 2 The difference between this embodiment and embodiment 1 is that the method further includes: S21 simulation, build a digital twin model, embed the multi-physics field coupling model into the digital twin model, input the historical operation data into the digital twin model in the order of collection time, enable the digital twin model to simulate the degradation process of the solenoid valve, and obtain the simulated degradation data of the solenoid valve at different periods.

[0045] Digital twins make full use of physical models, sensor updates, operation history and other data, and integrate multi-disciplinary, multi-physical quantity, multi-scale, and multi-probability simulation models. Digital twin models can complete mapping in virtual space, thereby reflecting the entire life cycle of the solenoid valve.

[0046] The actual operation of a solenoid valve involves the interaction of multiple physical fields, such as electromagnetic, thermal, and mechanical fields. The electromagnetic field drives the valve core, and this movement generates frictional heat, which in turn affects the temperature distribution of the solenoid valve. This temperature change can affect the mechanical and electromagnetic properties of the solenoid valve material. A multiphysics coupling model is used to describe the interactions and mutual influences between these different physical fields. Embedding this model into a digital twin model enables a more realistic simulation of the solenoid valve's behavior under complex operating conditions. For example, when simulating the high-frequency switching of a solenoid valve, changes in the electromagnetic field cause the valve core to move rapidly, generating significant heat. The thermal field model can calculate the resulting temperature increase, which in turn affects the elastic modulus of the valve core material and, in turn, its kinematic characteristics. The multiphysics coupling model accurately reflects these complex interactions and can then be mapped into the digital twin model, enabling the integration of the multiphysics coupling model into the digital twin.

[0047] The solenoid valve's historical operating data is input into the digital twin model in chronological order. This historical operating data includes various parameters of the solenoid valve under different operating conditions, such as voltage, current, pressure, temperature, and valve core displacement. By inputting this data, the digital twin model can simulate the actual operation of the solenoid valve over a period of time, thereby simulating the solenoid valve's degradation process. For example, based on the time of each solenoid valve opening and closing, current level, and other information recorded in the historical data, the digital twin model can simulate the wear of the solenoid valve's internal components, such as the wear between the valve core and the valve seat. This can then generate simulated degradation data for the solenoid valve at different time periods, reflecting the gradual decline in solenoid valve performance over time.

[0048] S22 simulates and judges, obtains the real degradation data at the same time as the simulated degradation data, calculates the difference between the simulated degradation data and the real degradation data, records it as the first data, and judges whether the first data is greater than the preset difference threshold. If so, the digital twin model is corrected, and the simulated degradation data is re-obtained based on the corrected digital twin model; if not, executes S23 to calculate the degree of degradation.

[0049] Acquire real degradation data at the same time as the simulated degradation data. Real degradation data can be obtained through actual testing methods. For example, regularly test the performance of the solenoid valve, measuring its key performance indicators such as flow rate, pressure loss, opening and closing time, and comparing them with the performance indicators in the solenoid valve's initial state to obtain the actual degradation of the solenoid valve. Real degradation data reflects the actual degree of degradation of the solenoid valve in the actual operating environment.

[0050] This embodiment also calculates the difference between the simulated degradation data and the actual degradation data, which is recorded as the first data. Then, it is determined whether the first data is greater than the preset difference threshold. The preset difference threshold is determined by using a statistical analysis algorithm and based on factors such as the performance requirements and operating accuracy of the solenoid valve. If the first data is greater than the preset difference threshold, it means that there is a large deviation between the simulation result of the digital twin model and the actual situation, and the digital twin model needs to be corrected. The correction method includes adjusting the parameters of the model, optimizing the algorithm of the multi-physics field coupling model, improving the boundary conditions of the digital twin model, etc. After the correction is completed, the simulation degradation data is re-acquired based on the corrected digital twin model, and this step is re-executed to improve the accuracy of the simulation. If the first data is not greater than the preset difference threshold, it means that the simulation result of the digital twin model is relatively accurate, and S23 can be executed to calculate the degree of degradation.

[0051] S24 calculates the degree of degradation, obtains the operating data of a preset time period before the current moment, records it as the second data, inputs the second data and the real-time operating data into the digital twin model in the order of collection time, obtains predicted degradation data, and calculates the predicted degree of degradation based on the predicted degradation data.

[0052] Operation data for a preset period of time before the current moment is obtained and recorded as the second data. The preset period of time can be determined based on the degradation characteristics of the solenoid valve and actual needs. For example, if the solenoid valve degrades slowly, the preset period of time can be set longer; if the degradation process is rapid, the preset period of time can be appropriately shortened. The operation data includes information about the operating status of the solenoid valve over a period of time.

[0053] In this embodiment, the second data and real-time operating data are input into the digital twin model in the order of their collection time. Based on this data, the digital twin model simulates the future degradation trend of the solenoid valve under its current operating state, thereby obtaining predicted degradation data. Subsequently, the predicted degradation degree is calculated based on this predicted degradation data. The predicted degradation degree can be quantified in various ways. For example, a key performance indicator (such as flow rate) of the solenoid valve can be compared with the initial performance indicator and the percentage of decrease calculated as the predicted degradation degree.

[0054] S25 degradation judgment, judging whether the predicted degradation degree is greater than a preset degradation degree threshold, if so, issuing a degradation alarm signal; if not, re-execute this step after a preset time interval.

[0055] Determine whether the predicted degradation level is greater than a preset degradation threshold. If so, the solenoid valve's performance has deteriorated to the point where it affects normal operation. A degradation alarm signal should be issued to allow for prompt maintenance measures, such as valve replacement or repair. If the predicted degradation level is not greater than the preset degradation threshold, the solenoid valve's performance remains within an acceptable range. This step can be repeated after a preset interval to continue monitoring the solenoid valve's degradation.

[0056] In other embodiments, the method further comprises: S26 updates the model, increases the grid density of the target area in the digital twin model, obtains a new digital twin model, extracts a new multi-physics field coupling model in the new digital twin model, and executes S12 to extract reliability parameters.

[0057] Increasing the mesh density in target areas of the digital twin model is crucial. These areas are typically critical areas of the solenoid valve that are prone to degradation or have a significant impact on performance, such as the contact area between the valve core and valve seat and the windings of the solenoid coil. Increasing the mesh density improves the digital twin model's computational accuracy in these areas, enabling more accurate simulation of physical phenomena. For example, when simulating friction and wear in a solenoid valve, increasing the mesh density in the contact area between the valve core and valve seat allows for more precise calculation of parameters such as contact stress and wear.

[0058] After obtaining the new digital twin model, a new multiphysics coupling model is extracted from it. Because changes in mesh density can affect the interactions between multiple physical fields, the multiphysics coupling model needs to be re-extracted to ensure it accurately reflects the physical characteristics of the solenoid valve. Then, step S12 is executed to extract reliability parameters.

[0059] In other embodiments, the method further comprises: S27 establishes a degradation model, draws a degradation curve based on the simulated degradation data, and establishes a degradation model for fitting the degradation curve. The calculation model of the degradation model is as follows: ; in, is the degree of degradation at time t; for the degree of degradation of the moment; is a parameter determined by nonlinear least squares method The value of .

[0060] Based on the simulated degradation data, a degradation curve is plotted. This curve can intuitively demonstrate the degradation trend of the solenoid valve. For example, with time as the horizontal axis and the degree of solenoid valve degradation as the vertical axis, the degradation curve can clearly show how the solenoid valve performance changes over time.

[0061] The nonlinear least squares method is a commonly used parameter estimation method. It determines the optimal value of parameter a by minimizing the sum of squares of the errors between the simulated degradation data and the predicted values of the degradation model, so that the degradation model can better fit the actual degradation curve.

[0062] S28 calculates reliability based on parameters Calculate and output the reliability of the solenoid valve. The calculation model is as follows: ; ; Among them, n is the intermediate variable; is the service life of the solenoid valve.

[0063] Reliability reflects a solenoid valve's ability to perform its specified function within a specified timeframe. By calculating the reliability of a solenoid valve over varying periods of use, its reliability level can be assessed, providing valuable insights for predicting its service life and developing maintenance plans. For example, when reliability drops below a certain level, the solenoid valve can be replaced or repaired in advance to minimize production interruptions or safety incidents caused by solenoid valve failure.

[0064] In the solenoid valve performance analysis process of this embodiment, a digital twin model is first constructed, a multi-physics coupling model is embedded, and historical operating data is input to simulate the degradation process and obtain simulated degradation data at different periods. Then, the actual degradation data at the same time is obtained, and the difference (first data) with the simulated data is calculated. If the difference is greater than a preset difference threshold, the model is corrected and the simulation data is re-acquired. Otherwise, the predicted degradation degree is calculated. That is, the operating data (second data) for a preset time period before the current time is input into the model with real-time data, and the degradation degree is calculated based on the predicted data. Then, it is determined whether the predicted degradation degree exceeds the preset threshold. If it exceeds, an alarm is issued. If it does not exceed, the determination is repeated after a preset interval. In addition, the grid density of the target area of the digital twin model can be increased to update the model, extract a new multi-physics coupling model, and extract reliability parameters. It is also possible to draw a degradation curve based on the simulated degradation data, and use the nonlinear least squares method to determine the parameters to establish the degradation model. Finally, the reliability of the solenoid valve under the service life is calculated based on the parameters, providing a basis for solenoid valve maintenance and life prediction, and ensuring its safe and stable operation.

[0065] Example 3: Reference Figure 3The difference between this embodiment and embodiment 2 is that the method further includes: S31 constructs an evaluation matrix, obtains the vector output by the BI-LSTM model after each round of training (that is, a vector composed of the probability distribution output by the BI-LSTM model after each round of training), constructs an evaluation matrix based on the vector, and standardizes the rows or columns of the evaluation matrix to obtain a processed evaluation matrix, where the element in the i-th row and j-th column of the evaluation matrix represents the probability of misclassifying the i-th class as the j-th class, and the value of the diagonal elements of the evaluation matrix is 0.

[0066] S32 obtains a data set, integrates the off-diagonal elements in the rows or columns of the evaluation matrix into one data set (that is, the probability distribution output after the same round of training, excluding the probability values of the correct classification results, is integrated into a data set), calculates the difference between any two off-diagonal elements in the data set, and records it as the third data (the third data represents the difference in the probability of misclassification. When this difference is very small, it means that the BI-LSTM model cannot specifically distinguish which category the current input data belongs to). For the third data that is less than the preset threshold, execute S33 to generate samples until all data sets are traversed.

[0067] S33 generates samples, trains the GANs model using historical reliability parameters and category labels to obtain a trained GANs model, uses the trained GANs model to generate training samples corresponding to the third data, and uses the training samples to train the trained BI-LSTM model again to obtain a retrained BI-LSTM model (the retrained BI-LSTM model strengthens training on unrecognizable input data and increases resolution capability), and the training samples include new reliability parameters and new category labels.

[0068] In the S13 prediction, the real-time reliability parameters are input into the retrained BI-LSTM model.

[0069] In other embodiments, the method further comprises: S34 obtains vectors, concatenates the vectors into a characteristic matrix, calculates the maximum eigenvalue of the characteristic matrix and the corresponding eigenvector, obtains the vector corresponding to the maximum value of the elements in the eigenvector, and records it as the first vector.

[0070] S35 correctly judges whether the historical reliability parameters and category labels corresponding to the first vector are correct. If so, the position of the maximum value in the eigenvector is used as the index, and the evaluation matrix is divided according to the index to obtain two sub-matrices (for example, the maximum value is in the 5th position in the eigenvector, then it is divided into two sub-matrices in the 5th row or 5th column of the evaluation matrix, and the 5th row or 5th column can be divided into any sub-matrix), and S36 vector screening is executed; if not, the category label of the historical reliability parameter corresponding to the first vector is reset, the BI-LSTM model is retrained, and S31 is executed to construct the evaluation matrix.

[0071] S36 vector screening, based on the vectors in the two sub-matrices, respectively obtains the loss value of the trained BI-LSTM model (that is, calculates the loss value corresponding to the elements in the two sub-matrices), retains the vectors in the sub-matrix with a smaller loss value (a smaller loss value indicates better training effect of the sub-matrix and better classification ability of the BI-LSTM model), deletes the vectors in the sub-matrix with a larger loss value, and executes S32 to obtain the dataset.

[0072] Example 4: The difference between this example and Example 1 is that the method further includes: using the PCA algorithm to reduce the dimension of the historical reliability parameters and the real-time reliability parameters, and using the reduced dimension historical reliability parameters and real-time reliability parameters as new historical reliability parameters and new real-time reliability parameters.

[0073] The above are all preferred embodiments of the present application, and are not intended to limit the scope of protection of the present application. Therefore, any equivalent changes made based on the structure, shape, and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A fuel injector solenoid valve reliability prediction method based on multi-physics field coupling, characterized in that: include: Data acquisition: Collect real-time operating data of the solenoid valve and multiple sets of historical operating data of similar solenoid valves to build a multi-physics field coupling model of the solenoid valve. The operating data includes electromagnetic force data, temperature data, fluid pressure data, vibration data, operating voltage, and operating current. Extracting reliability parameters: Each set of historical operating data is sequentially input into a multi-physics field coupling model, and the multi-physics field coupling model is solved using a finite element analysis method to obtain the multi-physics field distribution characteristics of the solenoid valve. Historical reliability parameters are extracted from the multi-physics field distribution characteristics and category labels are added to the historical reliability parameters. The historical reliability parameters include: the temperature gradient of the electromagnetic coil, the stress concentration factor of the valve core, and the pressure pulse amplitude of the fluid flow; Prediction: Input real-time operation data into the multi-physics field coupling model to obtain real-time reliability parameters. Use the machine learning algorithm to build a BI-LSTM model. Use historical reliability parameters and category labels to train the BI-LSTM model to obtain a trained BI-LSTM model. Input the real-time reliability parameters into the trained BI-LSTM model and output the prediction results.

2. The fuel injector solenoid valve reliability prediction method based on multi-physics field coupling according to claim 1, characterized in that: The method further comprises: Simulation: Build a digital twin model, embed the multi-physics field coupling model into the digital twin model, and input historical operating data into the digital twin model in chronological order of collection time. The digital twin model simulates the degradation process of the solenoid valve and obtains simulated degradation data of the solenoid valve at different stages. Simulation judgment: obtain the real degradation data at the same time as the simulated degradation data, calculate the difference between the simulated degradation data and the real degradation data, record it as the first data, and determine whether the first data is greater than the preset difference threshold. If so, correct the digital twin model and re-acquire the simulated degradation data based on the corrected digital twin model; if not, execute the step of calculating the degradation degree; Calculating the degree of degradation: Obtaining the operating data of a preset time period before the current moment, recording it as the second data, inputting the second data and the real-time operating data into the digital twin model in the order of acquisition time to obtain predicted degradation data, and calculating the predicted degree of degradation based on the predicted degradation data; Degradation judgment: Determine whether the predicted degradation degree is greater than the preset degradation degree threshold. If so, a degradation alarm signal is issued; if not, this step is executed again after a preset time interval.

3. The fuel injector solenoid valve reliability prediction method based on multi-physics field coupling according to claim 2, characterized in that: The method further comprises: Update the model: Increase the mesh density of the target area in the digital twin model to obtain a new digital twin model, extract the new multi-physics field coupling model in the new digital twin model, and perform the steps of extracting reliability parameters.

4. The fuel injector solenoid valve reliability prediction method based on multi-physics field coupling according to claim 3, characterized in that: The method further comprises: Establishing a degradation model: Draw a degradation curve based on the simulated degradation data, and establish a degradation model for fitting the degradation curve. The calculation model of the degradation model is as follows: ； in, is the degree of degradation at time t; for the degree of degradation of the moment; is a parameter determined by nonlinear least squares method The value of Calculation reliability: based on parameters Calculate and output the reliability of the solenoid valve. The calculation model is as follows: ； ； Among them, n is the intermediate variable; is the service life of the solenoid valve.

5. The fuel injector solenoid valve reliability prediction method based on multi-physics field coupling according to any one of claims 1 to 4, characterized in that: The method further comprises: Construct an evaluation matrix: Obtain the vectors output by the BI-LSTM model after each round of training, construct an evaluation matrix based on the vectors, and normalize the rows or columns of the evaluation matrix to obtain a processed evaluation matrix, where the element in the i-th row and j-th column of the evaluation matrix represents the probability of misclassifying the i-th class as the j-th class; Obtaining a data set: Integrate the off-diagonal elements in the rows or columns of the evaluation matrix into a data set; Calculate the difference between any two off-diagonal elements in the data set, record it as the third data, and execute the sample generation step for the third data that is less than a preset threshold until all data sets are traversed; Generate samples: Use the GANs model to generate training samples corresponding to the third data, and use the training samples to retrain the trained BI-LSTM model to obtain a retrained BI-LSTM model, where the training samples include new reliability parameters and new category labels; In the prediction step, the real-time reliability parameters are input into the retrained BI-LSTM model.

6. The fuel injector solenoid valve reliability prediction method based on multi-physics field coupling according to claim 5, characterized in that: The values of the diagonal elements of the evaluation matrix are 0.

7. The fuel injector solenoid valve reliability prediction method based on multi-physics field coupling according to claim 6, characterized in that: The method further comprises: Obtaining a vector: splicing the vectors into a characteristic matrix, calculating the maximum eigenvalue and the corresponding eigenvector of the characteristic matrix, obtaining the vector corresponding to the maximum value of the elements in the eigenvector, and recording it as the first vector; Correct judgment: determine whether the historical reliability parameter and category label corresponding to the first vector are correct. If so, divide the evaluation matrix according to the position of the maximum value in the eigenvector to obtain two sub-matrices; if not, reset the category label of the historical reliability parameter corresponding to the first vector, retrain the BI-LSTM model, and execute the step of constructing the evaluation matrix; Vector screening: Based on the vectors in the two sub-matrices, the loss values of the trained BI-LSTM model are obtained. The vectors in the sub-matrix with the smaller loss value are retained, and the vectors in the sub-matrix with the larger loss value are deleted.

8. The fuel injector solenoid valve reliability prediction method based on multi-physics field coupling according to claim 5, characterized in that: The method further includes: training the GANs model using historical reliability parameters and category labels to obtain a trained GANs model; In the step of generating samples, the trained GANs model is used to generate training samples corresponding to the third data.

9. The fuel injector solenoid valve reliability prediction method based on multi-physics field coupling according to claim 1, characterized in that: The method further includes: using a PCA algorithm to reduce the dimensions of the historical reliability parameters and the real-time reliability parameters, and using the reduced dimensions of the historical reliability parameters and the real-time reliability parameters as new historical reliability parameters and new real-time reliability parameters.

Images