Digital twin motor electric pump full life cycle health monitoring and life prediction platform

Abstract

The invention relates to the technical field of machinery, in particular to a digital twin motor electric pump full life cycle health monitoring and life prediction platform, which introduces mechanical and automatic technologies, an automatic control system is matched with a self-adaptive clamp, and the position can be automatically adjusted to adapt to measured objects of different sizes. The compatibility of the experiment table and the workpiece clamping efficiency are improved, vibration generated during operation is absorbed, and the test accuracy is improved; a test bench servo power system and a control system are used to simulate normal working states of an electric pump and a motor in a vehicle, and a real working condition scene is simulated. The platform is based on a multi-sensor integration thought to reduce the size and cost of the test bench and improve the precision of life prediction. A multi-physical-quantity mathematical model is built, the residual life of a measured object is accurately predicted, the efficiency is improved, and the prediction period is shortened.

Description

Technical Field

[0001] This invention relates to the field of mechanical technology, specifically to a digital twin electric motor pump full life cycle health monitoring and life prediction platform. Background Technology

[0002] In the field of mechanical equipment testing, motors and electric pumps, as core power components in industrial production and transportation, directly affect the overall reliability and maintenance costs of equipment due to the stability of their operating status and the accuracy of their remaining lifespan. With the development of intelligent manufacturing technology, traditional periodic maintenance models can no longer meet the needs of efficient and accurate maintenance. There is an urgent need to build a full life cycle health inspection and lifespan prediction system to realize the transformation from passive maintenance to predictive maintenance. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the purpose of this invention is to provide a digital twin electric motor pump full life cycle health monitoring and lifespan prediction platform.

[0004] The technical solution adopted by this invention to solve its technical problem is:

[0005] A digital twin electric motor pump full life cycle health monitoring and life prediction platform includes: a test bench, a dual-line ball screw module, an electromagnetic servo hydraulic fixture, a sensor integration module, a control module, and a test bench servo power system;

[0006] A double-wire ball screw module is embedded in the central groove of the test bench. The double-wire ball screw module includes a guide rail, on which two slides are slidably mounted. A grid-shaped sliding support platform is mounted on the opposite side of the two slides. The top surface of the grid-shaped sliding support platform is flush with the top surface of the slide. An electromagnetic servo hydraulic clamp is mounted on the top surface of the slide. The slide is driven by a slide servo motor and a double-wire ball screw structure.

[0007] The electromagnetic servo hydraulic clamp includes a square housing, within which multiple independent liquid chambers are arranged in an array. Each liquid chamber contains a hexagonal ejector pin, which extends through the liquid chamber from one side of the square housing to form an array arrangement. A piezoelectric ceramic sensor is integrated at the front end of the ejector pin. A modular blocking piston is arranged inside the square housing opposite to the liquid chamber. An electromagnetic telescopic rod connected to the modular blocking piston is mounted on the square housing. The square housing is connected to an oil storage tank, in which a spring-loaded reset piston is mounted.

[0008] The sensor integration module is used to collect the operating parameters of electric pumps and motors. It includes an electric pump sensor module and a motor sensor module. The electric pump sensor module includes a pressure sensor, a flow sensor, and an electric pump temperature sensor. The motor sensor module includes a current sensor, a voltage sensor, a speed sensor, and a motor temperature sensor.

[0009] The test bench servo power system is set on one side of the central groove on the test bench frame, and is used to connect with the motor and electric pump under test, so as to simulate the driving state of the motor and electric pump under test under real working conditions.

[0010] The control module includes an industrial computer and a controller that are interconnected. The controller is electrically connected to the test bench servo power system, slide table servo motor, electromagnetic telescopic rod, and sensor integrated module. The industrial computer integrates a life prediction model based on a random forest regression model pre-trained. After obtaining feature inputs based on the operating parameters of the electric pump and motor, the industrial computer predicts the remaining life of the electric pump and motor through the life prediction model.

[0011] As a preferred embodiment, a further technical solution of the present invention is:

[0012] Preferably, the electromagnetic servo hydraulic fixture is connected to the top surface of the slide table via a dovetail groove guide rail structure; the grid-shaped sliding support platform is connected to the slide table via a locking structure; the ejector pin adopts a three-section sleeve telescopic structure, and each section is equipped with a scraper ring at the bottom end. The scraper ring is used to scrape the impurities attached to the inner wall into the square shell when the ejector pin retracts; a protective baffle is provided on the test bench on the side of the test bench servo power system.

[0013] Preferably, the liquid chamber is divided into multiple areas according to the design dimensions, and each area is equipped with a modular blocking piston. The modular blocking piston includes a rectangular body, a plunger adapted to the liquid chamber is provided on the rectangular body, a rubber sealing ring is provided on the circumference of the rectangular body, and a connecting rod is provided on the side of the rectangular body opposite to the plunger. The connecting rod passes through the square shell and connects to the electromagnetic telescopic rod.

[0014] Preferably, the platform clamping process includes:

[0015] Initial state and condition: The modular blocking piston is at the lowest end of its stroke, the spring of the oil tank's reset piston is in a compressed state, and the oil flows into the square housing through the oil tank and enters each liquid chamber, thereby pushing the ejector pin to fully extend and with its top end flush.

[0016] Clamping state: The electric pump or motor under test is placed on the grid-shaped sliding support platform. The slide table drives the grid-shaped sliding support platform and the electromagnetic servo hydraulic clamp to move. After the grid-shaped sliding support platforms on both sides are inserted into each other, the ejector pin gradually contacts the surface of the electric pump or motor, and different axial displacements are generated due to the surface geometry.

[0017] Locking phase: When the pressure at the tip of the ejector pin reaches the preset threshold, the electromagnetic telescopic rod drives the modular blocking piston to move, pumping redundant oil back to the storage tank and cutting off the hydraulic circuit, sealing the liquid chamber to achieve hydraulic locking, thereby forming an envelope-type adaptive fit.

[0018] Preferably, the pre-training process for the electric pump includes:

[0019] Data collection phase:

[0020] The electric pump was driven by a servo power system on a test bench to simulate actual operating conditions inside a vehicle. The test pressure was controlled to be no more than 10 MPa and the oil temperature was maintained at 40-50℃, covering the entire lifespan of the electric pump from early to mid-stages to the decline stage. Data points were collected at fixed time intervals, including pump speed n, displacement q, and actual output flow rate. Actual operating torque Working pressure P and oil temperature data;

[0021] Data preprocessing stage: Calculate the volumetric efficiency of the electric pump based on the collected data points. , Mechanical efficiency , ;

[0022] The volumetric efficiency, mechanical efficiency and known remaining life were fitted with a quadratic polynomial. The residuals of each data point were calculated by residual analysis. Outliers outside the 95% confidence interval were removed, and the nearest mean replacement method was used to fill in the missing data.

[0023] After filling in the missing data, the data points are normalized to map the volumetric efficiency and mechanical efficiency to the [0,1] interval, thus eliminating the difference in dimensions.

[0024] Feature engineering phase:

[0025] Normalized volumetric efficiency and mechanical efficiency were used as the core input features of the model. The effectiveness of the features was verified by the feature importance scoring function of the random forest model, confirming that the contribution of mechanical efficiency to the prediction results was no less than 75% and the contribution of volumetric efficiency was no less than 24%.

[0026] Model building and training phase: A random forest regression model is adopted, and the data points are divided into training set and test set in a 7:3 ratio. Both training set and test set contain data of each life stage of electric pump. The model parameters are initialized.

[0027] The model parameters were adjusted using a phased, two-parameter independent optimization strategy:

[0028] First, optimize the number of trees n_estimators. Within the first preset range, perform a large-scale positioning with a first step length to determine the optimal range within the first preset range. Within the optimal range, perform a fine search with a second step length to determine the value of n_estimators. The second step length is less than the first step length.

[0029] Then, the tree depth max_depth is optimized and adjusted with a third step size within the second preset range. The value of max_depth is determined by combining the overfitting gap analysis method.

[0030] Model validation phase:

[0031] Calculate model performance metrics based on the test set to ensure R 2 ≥0.97, MSE≤1.9×10 7 RMSE ≤ 4400, and residual mean ≤ 350, standard deviation ≤ 4400, with residuals randomly distributed near the zero line;

[0032] Nine lifetime stage nodes (10%, 20%, ..., 90%) were extracted from the original data. The optimized model was used to predict the remaining lifetime, ensuring that the prediction deviation percentage of each node was ≤ ±2%.

[0033] Comparison of training set and test set R 2 Difference, ensuring ΔR 2 <0.05, the training of the electric pump remaining life prediction model is complete.

[0034] Preferably, the model pre-training process for motors is the same as that for electric pumps, the difference being the different data collection points. Specifically:

[0035] Adjust the motor load to below rated operating conditions to stabilize the motor winding temperature rise at 60-80K, simulating a typical automotive motor operating load, and covering data points collected throughout the motor's entire lifespan, including active power. Apparent power S, bus voltage U, current I, and motor shaft extension torque Rotation speed Temperature before operation and the temperature after operation ;

[0036] The core characteristic of the failure mechanism calculation based on the motor's power loss leading to a decrease in power factor and energy conversion efficiency, and heat loss leading to an increase in temperature, is: power factor. motor efficiency Temperature rise ;

[0037] Residual analysis was performed on the core features to identify outliers outside the 95% confidence interval, fill in the missing data, and perform normalization.

[0038] The power factor after normalization Motor efficiency and temperature rise The core input features of the model are used to verify the effectiveness of the features through the feature importance scoring function of the random forest model. Contribution rate = 53.76%, motor efficiency =37.92%, power factor =8.32%, confirming that temperature rise is the dominant factor in predicting motor lifespan.

[0039] Preferably, the oil storage tank is externally connected to a high-pressure oil filter and a bypass circulation pump to achieve continuous filtration after the oil flows back to the storage tank, thus maintaining the cleanliness of the hydraulic circuit oil.

[0040] The present invention, which adopts the above technical solution, has the following prominent features compared with the prior art:

[0041] This platform enables real-time monitoring, performance evaluation, and accurate prediction of remaining lifespan for motors and pumps. By integrating a high-precision mechanical structure with an intelligent control system, it boasts excellent versatility and automation, while also improving testing accuracy. Specifically, the slide table and electromagnetic servo hydraulic clamping system achieve adaptive clamping and rigid fixation of workpieces of different sizes, enhancing clamping efficiency and testing reliability. The test bench servo power system simulates the operating conditions of motors and pumps under real-world conditions, effectively reducing experimental errors. The integrated sensor module is rationally arranged according to physical causality and signal flow sequence, simultaneously collecting multi-dimensional data such as pressure, flow rate, temperature, current, voltage, and speed, providing a high-quality data foundation for health status assessment.

[0042] In terms of lifetime prediction modeling, this invention proposes key performance indicators for electric pumps and motors—volume efficiency and mechanical efficiency of electric pumps, and power factor, motor efficiency, and standardized temperature rise of motors—and constructs lifetime prediction models based on random forest regression models. Through data preprocessing, outlier removal, feature normalization, and backtracking validation, the prediction accuracy and generalization ability of the models are significantly improved. Experimental results show that the determination coefficients of the prediction models for electric pumps and motors reach 0.8682 and 0.9777, respectively, verifying the effectiveness and reliability of the models.

[0043] By introducing mechanical and automation technologies, the automated control system, coupled with adaptive fixtures, can automatically adjust its position to adapt to test objects of different sizes, improving the compatibility of the test bench and workpiece clamping efficiency. Simultaneously, it absorbs vibrations generated during operation, enhancing test accuracy. The test bench's servo power system and control system simulate the normal operating conditions of electric pumps and motors inside a vehicle, mimicking real-world working scenarios. The platform is based on a multi-sensor integration approach to reduce the size and cost of the test bench and improve the accuracy of lifespan prediction. A multi-physical quantity mathematical model is built to accurately predict the remaining lifespan of the test object, improving efficiency and reducing the prediction cycle.

[0044] In summary, this lifespan prediction platform not only enables health status monitoring and lifespan prediction of motors and pumps throughout their entire lifecycle, but also provides a feasible technical path and experimental foundation for intelligent operation and maintenance and predictive maintenance of mechanical equipment, demonstrating promising engineering application prospects and widespread value. In the future, it can be further integrated with edge computing and cloud platforms to expand its practical applications in the Industrial Internet of Things (IIoT). Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the overall structure of the lifetime prediction platform in an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of the structure of the dual-line ball screw module in an embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram of the connection structure between the slide table and the electromagnetic servo hydraulic clamp in an embodiment of the present invention;

[0048] Figure 4 This is a cross-sectional view of the electromagnetic servo hydraulic clamp in an embodiment of the present invention;

[0049] Figure 5 This is a schematic diagram of the ejector pin structure in an embodiment of the present invention;

[0050] Figure 6 This is a schematic diagram showing the detailed structure of the scraper ring in an embodiment of the present invention;

[0051] Figure 7 This is a graph showing the relationship between the mechanical efficiency and remaining life of the electric pump in an embodiment of the present invention;

[0052] Figure 8 This is a graph showing the relationship between the volumetric efficiency and remaining life of the electric pump in an embodiment of the present invention;

[0053] Figure 9 These are mechanical efficiency and remaining life residual analysis diagrams in an embodiment of the present invention, where diagram a is the mechanical efficiency fitting curve and diagram b is the mechanical efficiency residual analysis.

[0054] Figure 10 Figure 1 shows the volumetric efficiency and residual analysis of remaining lifetime in an embodiment of the present invention. Figure 2a is the volumetric efficiency fitting curve, and Figure 3b is the volumetric efficiency residual analysis.

[0055] Figure 11 These are comparison diagrams of the distribution of electric pump features before and after normalization in embodiments of the present invention, where diagram a is the original mechanical efficiency distribution, diagram b is the normalized mechanical efficiency distribution, diagram c is the original volumetric efficiency distribution, and diagram d is the normalized solvent efficiency distribution.

[0056] Figure 12 This is a comparison chart of the actual remaining lifespan and the predicted remaining lifespan of the electric pump in an embodiment of the present invention;

[0057] Figure 13 This is a schematic diagram of the predictive residual analysis of the electric pump in an embodiment of the present invention;

[0058] Figure 14 This is a schematic diagram showing the order of importance of electric pump features in an embodiment of the present invention;

[0059] Figure 15 This is a schematic diagram of the optimization analysis of the electric pump prediction model parameter n_estimators in an embodiment of the present invention, where figure a is R 2 Learning curve, Figure b shows the MSE learning curve;

[0060] Figure 16 This is a schematic diagram of the optimization analysis of the max_depth parameter of the electric pump prediction model in an embodiment of the present invention, where figure a is R 2 Learning curve, Figure b shows the MSE learning curve;

[0061] Figure 17 This is a schematic diagram illustrating the maximum depth of overfit analysis of the electric pump prediction model in this embodiment of the invention;

[0062] Figure 18 This is a schematic diagram of the backtracking verification of the electric pump prediction model in an embodiment of the present invention, wherein Figure a is a comparison diagram of the actual remaining life and the predicted remaining life, and Figure b is the percentage of prediction deviation.

[0063] Figure 19 This is a schematic diagram of the overfitting test results of the electric pump prediction model in an embodiment of the present invention. Figure a shows the prediction results on the training set, Figure b shows the prediction results on the test set, and Figure c shows the results of the predictions on the training and test sets. 2 Comparison chart;

[0064] Figure 20 This is a schematic diagram of motor power factor residual analysis in an embodiment of the present invention;

[0065] Figure 21 This is a schematic diagram of motor efficiency residual analysis in an embodiment of the present invention;

[0066] Figure 22 This is a schematic diagram of motor temperature rise residual analysis in an embodiment of the present invention;

[0067] Figure 23 These are comparison diagrams of the distribution of motor features before and after normalization in embodiments of the present invention. In these diagrams, a is the original efficiency distribution diagram, b is the normalized efficiency distribution diagram, c is the original power factor distribution diagram, d is the normalized power factor distribution diagram, e is the original temperature rise distribution diagram, and f is the normalized temperature rise distribution diagram.

[0068] Figure 24 This is a distribution diagram of the actual remaining life and predicted remaining life of the motor in an embodiment of the present invention;

[0069] Figure 25 This is a schematic diagram showing the order of importance of motor features in an embodiment of the present invention;

[0070] Figure 26 This is a comparison chart of the actual remaining lifespan and the predicted remaining lifespan of the motor in an embodiment of the present invention;

[0071] Figure 27 This is a schematic diagram of backtracking verification of the motor prediction model in an embodiment of the present invention, wherein Figure a is a comparison diagram of the actual remaining life and the predicted remaining life, and Figure b is the percentage of prediction deviation.

[0072] Figure 28 This is a schematic diagram of the overfitting test results of the motor prediction model in an embodiment of the present invention. Figure a shows the prediction results on the training set, Figure b shows the prediction results on the test set, and Figure c shows the results of the predictions on the training and test sets. 2 Comparison chart.

[0073] Explanation of reference numerals in the attached drawings: 1. Servo power system for the test bench; 2. Protective baffle; 3. Double-line ball screw module; 4. Electromagnetic servo hydraulic fixture; 5. Industrial computer; 6. Test bench frame; 7. Guide rail; 8. Slide table; 9. Grid-shaped sliding support platform; 10. Slide table servo motor; 11. Double-line ball screw; 12. Square housing; 13. Liquid chamber; 14. Ejector pin; 15. Modular blocking piston; 16. Electromagnetic telescopic rod; 17. Oil reservoir; 18. Reset piston; 19. Lock; 20. Scraper ring; 21. Rectangular body; 22. Plunger; 23. Connecting rod; 24. First-stage telescopic ejector pin; 25. Second-stage telescopic ejector pin; 26. Third-stage telescopic ejector pin. Detailed Implementation

[0074] The present invention will be further illustrated below with reference to specific embodiments. The purpose of this illustration is solely to provide a better understanding of the invention. Therefore, the examples given do not limit the scope of protection of the present invention.

[0075] See Figures 1-6 This embodiment provides a digital twin electric motor pump full life cycle health monitoring and life prediction platform, including: test bench 6, double-line ball screw module 3, electromagnetic servo hydraulic fixture 4, sensor integrated module, control module and test bench servo power system 1;

[0076] A double-wire ball screw module 3 is embedded in the central groove of the test bench 6. The double-wire ball screw module 3 includes a guide rail 7. Two slides 8 are slidably arranged on the guide rail 7. A grid-shaped sliding support platform 9 is arranged on the opposite side of the two slides 8. The top surface of the grid-shaped sliding support platform 9 is flush with the top surface of the slides 8. An electromagnetic servo hydraulic clamp 4 is arranged on the top surface of the slides 8. The slides 8 are driven by a slide servo motor 10 and a double-wire ball screw 11.

[0077] The electromagnetic servo hydraulic clamp 4 includes a square housing 12, in which multiple independent liquid chambers 13 are arranged in an array. Each liquid chamber 13 is provided with a hexagonal ejector pin 14. The ejector pin 14 passes through the liquid chamber 13 and extends out from one side of the square housing 12 to form an array arrangement. The front end of the ejector pin 14 integrates a piezoelectric ceramic sensor. A modular blocking piston 15 is arranged in the square housing 12 opposite to the liquid chamber 13. An electromagnetic telescopic rod 16 is provided on the square housing 12 and is driven and connected to the modular blocking piston 15. The square housing 12 is connected to an oil storage tank 17. A spring-loaded reset piston 18 is provided in the oil storage tank 17.

[0078] The sensor integration module is used to collect the operating parameters of electric pumps and motors. It includes an electric pump sensor module and a motor sensor module. The electric pump sensor module includes a pressure sensor, a flow sensor, and an electric pump temperature sensor. The motor sensor module includes a current sensor, a voltage sensor, a speed sensor, and a motor temperature sensor.

[0079] The test bench servo power system 1 is set on one side of the central groove on the test bench frame 6, and is used to connect with the motor and electric pump under test, so as to simulate the driving state of the motor and electric pump under test under real working conditions.

[0080] The control module includes an industrial computer 5 and a controller that are interconnected. The controller is electrically connected to the test bench servo power system 1, the slide table servo motor 10, the electromagnetic telescopic rod 16, and the sensor integration module. The industrial computer 5 integrates a life prediction model based on a random forest regression model pre-trained. After obtaining feature inputs based on the working parameters of the electric pump and motor, the industrial computer 5 predicts the remaining life of the electric pump and motor through the life prediction model.

[0081] In practice, the electromagnetic servo hydraulic clamp 4 is connected to the top surface of the slide table 8 through the dovetail groove guide rail structure; the grid-shaped sliding support table 9 is connected to the slide table 8 through the locking buckle 19 structure; the ejector pin 14 adopts a three-section sleeve telescopic structure (including the first-stage telescopic ejector pin 24, the second-stage telescopic ejector pin 25, and the third-stage telescopic ejector pin 26), and each section is provided with a scraper ring 20 at the bottom end. The scraper ring 20 is used to scrape the impurities attached to the inner wall into the square shell 12 when the ejector pin 14 retracts; a protective baffle 2 is provided on the test bench 6 on one side of the test bench servo power system 1.

[0082] The liquid chamber 13 is divided into multiple areas according to the design dimensions. Each area is equipped with a modular blocking piston 15. The modular blocking piston 15 includes a rectangular body 21. A plunger 22 adapted to the liquid chamber 13 is provided on the rectangular body 21. A rubber sealing ring is provided around the rectangular body 21. A connecting rod 23 is provided on the side of the rectangular body 21 opposite to the plunger 22. The connecting rod 23 passes through the square housing 12 and connects to the electromagnetic telescopic rod 16.

[0083] Modular design allows for flexible configuration of the number of electromagnetic servo hydraulic fixtures 4 and the dimensions of the grid-shaped sliding support platform 9 according to the size and structure of the object under test. During the development of a digital twin-driven platform for the full lifecycle health monitoring and life prediction of motors and pumps, the efficiency, versatility, and reliability of workpiece clamping are key factors restricting the overall performance of the platform. Existing fixtures often use mechanical locking blocks (such as set screws) to laterally lock the ejector pins, resulting in uneven force distribution. This can easily cause localized loosening or excessive pressure on the ejector pin array, affecting clamping accuracy. Furthermore, mechanical locking typically relies on manual or external drive, resulting in slow response times that are difficult to meet the demands of high-speed automated testing. Long-term use also leads to screw wear and frequent maintenance. Simultaneously, existing ejector pins primarily rely on spring sleeves and guide post structures for elastic support, resulting in limited and unevenly distributed clamping force, making them unsuitable for clamping heavy or high-precision workpieces.

[0084] To address the problems of existing clamps, the electromagnetic servo hydraulic clamp 4 of this invention adopts a hexagonal ejector pin structure, which significantly improves the torsional resistance of the ejector pins 14. This effectively suppresses deflection and loosening caused by torque during clamping, thereby enhancing clamping stability. Each ejector pin 14 employs a three-stage sleeve-type telescopic mechanism. Under a large telescopic stroke, the force is transmitted sequentially from the first-stage to the third-stage telescopic ejector pins, with the circumscribed circle area increasing progressively, thus increasing the contact area with the workpiece and providing stronger clamping force. To achieve precise control and safe locking of the clamping force, each ejector pin 14 integrates a piezoelectric ceramic sensor at its front end to monitor the pressure in real time. When the monitored pressure reaches a preset threshold, the platform immediately triggers the hydraulic locking mechanism, transmitting pressure through the oil circuit to lock the current telescopic position, preventing over-clamping or loosening.

[0085] Each modular blocking piston corresponds to a pin 14, forming an independent control unit. Each control unit can independently drive the electromagnetic telescopic rod 16 based on the pressure feedback signal to adaptively adjust the corresponding pin group, giving full play to the rapid response characteristics of electromagnetic drive, so that each pin group can complete position adjustment and optimization within milliseconds.

[0086] During implementation, a high-pressure oil filter and a bypass circulation pump can be connected to the outside of the oil storage tank to achieve continuous filtration after the oil flows back to the storage tank, thus maintaining the cleanliness of the hydraulic circuit oil.

[0087] During implementation, the platform's mounting process includes:

[0088] Initial state and condition: The modular blocking piston is at the lowest end of its stroke, the spring of the oil tank's reset piston is in a compressed state, and the oil flows into the square housing through the oil tank and enters each liquid chamber, thereby pushing the ejector pin to fully extend and with its top end flush.

[0089] Clamping state: The electric pump or motor under test is placed on the grid-shaped sliding support platform. The slide table drives the grid-shaped sliding support platform and the electromagnetic servo hydraulic clamp to move. After the grid-shaped sliding support platforms on both sides are inserted into each other, the ejector pin gradually contacts the surface of the electric pump or motor, and different axial displacements are generated due to the surface geometry.

[0090] Locking phase: When the pressure at the tip of the ejector pin reaches the preset threshold, the electromagnetic telescopic rod drives the modular blocking piston to move, pumping redundant oil back to the storage tank and cutting off the hydraulic circuit, sealing the liquid chamber to achieve hydraulic locking, thereby forming an envelope-type adaptive fit.

[0091] In implementation, the electric pump sensor module is connected to the output hydraulic circuit of the pump under test to synchronously monitor key parameters such as system pressure, output flow rate, and oil temperature. The electric pump sensor module adopts an integrated oil passage block design, centrally arranging various sensor interfaces on a specially designed oil pipe. This aims to simplify the hydraulic circuit structure, reduce space occupation, and achieve centralized data acquisition and neat cable layout. The design follows the physical causal sequence of the hydraulic measurement chain for sensor placement: pressure sensor, flow sensor, and temperature sensor in that order. This arrangement strategy aims to eliminate mutual interference between sensors. The primary consideration is that the impeller of the turbine flow sensor can cause throttling of the oil, resulting in pressure loss. Therefore, the pressure sensor is placed at the front to accurately capture the true pressure at the pump outlet; the flow and temperature sensors are placed sequentially afterwards, ultimately ensuring that the collected pressure, flow, and temperature data best match the actual operating conditions of the system.

[0092] The motor sensor module connects to the output of the motor under test and the electrical control system to monitor key parameters such as current, voltage, temperature, and speed during motor operation in real time. Temperature sensors are located at key temperature measurement points on the motor housing and windings to capture real-time temperature changes. The speed sensor, via an encoder mounted on the motor shaft, outputs pulse signals to calculate the real-time speed. To reduce signal crosstalk and noise, the sensor layout follows the signal flow sequence principle: current and voltage sensors are placed at the forefront to ensure the accuracy of electrical parameter acquisition; the speed sensor follows, used to synchronously record the mechanical output state; the temperature sensor is located within the motor body structure and is insulated and shielded to prevent electromagnetic interference from affecting temperature measurement accuracy.

[0093] During implementation, the pre-training process for the electric pump includes:

[0094] Data Acquisition Phase: The electric pump is driven by the test bench servo power system to simulate the actual operating conditions inside the vehicle. The test pressure is controlled to be no more than 10MPa and the oil temperature is maintained at 40-50℃, covering the entire life stage of the electric pump from early to mid-term to the decline stage. Data points are collected at fixed time intervals, including electric pump speed n, displacement q, and actual output flow rate. Actual operating torque Data on working pressure P and oil temperature can be collected; a set of data points can be collected every 10 hours, for a total of 300 data points, to ensure that the data evenly covers the life curve and avoid local data density that may cause model deviation.

[0095] Data preprocessing stage: Calculate the volumetric efficiency of the electric pump based on the collected data points. , Mechanical efficiency , ;

[0096] A quadratic polynomial fit was performed on the volumetric efficiency, mechanical efficiency, and known remaining life. Residuals for each data point were calculated using residual analysis, and outliers outside the 95% confidence interval were removed. Data with residuals exceeding the interval were considered outliers. Figure 9 , 10 As shown, outliers mostly originate from momentary sensor malfunctions or fluctuations in operating conditions. The nearest mean replacement method is used to fill in missing data, avoiding data loss caused by direct deletion and ensuring the continuity of the lifespan curve.

[0097] The system was fitted using quadratic polynomial regression, with a coefficient of determination R0. 2 The values ​​of 0.939 and 0.889, respectively, indicate that the model can effectively characterize the changes in mechanical efficiency and volumetric efficiency with remaining lifetime. The fitted curves show that volumetric efficiency gradually decreases with decreasing remaining lifetime, and the rate of decline tends to be moderate. In the residual analysis plot, a few residual points (marked by purple circles in the figure) exceed the confidence interval. These outliers may originate from measurement errors, operating condition fluctuations, or other uncontrolled factors. Most residuals are randomly distributed near the zero line and fall within the 95% confidence interval, indicating that the model fits well.

[0098] After filling in the missing data, the data points are normalized to map the volumetric efficiency and mechanical efficiency to the [0,1] interval, eliminating dimensional differences and ensuring that the original data distribution is as shown. Figure 11 As shown, normalization preserves the distribution characteristics of the original data, effectively eliminating dimensional differences and allowing mechanical efficiency and volumetric efficiency to be compared and analyzed on the same numerical scale. After normalization, both types of efficiency indicators exhibit approximately normal distribution characteristics, and their distribution intervals tend to be consistent. This provides a data foundation for the subsequent construction of a remaining life prediction model based on multi-feature fusion, and helps to improve the stability and convergence efficiency of model training.

[0099] Feature Engineering Phase: Based on the failure mechanism of electric pumps: The remaining lifespan of electric pumps is mainly constrained by sealing performance (affecting volumetric efficiency) and wear of moving parts (affecting mechanical efficiency). Therefore, normalized volumetric efficiency and mechanical efficiency are directly used as the core input features of the model. The effectiveness of the features is verified by the feature importance scoring function of the random forest model, confirming that the contribution of mechanical efficiency to the prediction results is not less than 75%, and the contribution of volumetric efficiency is not less than 24%. Figure 14 As shown, both types of features are valid inputs, and mechanical efficiency is the dominant factor.

[0100] Model building and training phase: A random forest regression model was adopted, with data points divided into training and test sets in a 7:3 ratio. Both sets included data from each stage of the electric pump's lifespan. Model parameters were initialized. The random forest regression model possesses advantages such as strong resistance to overfitting, good adaptability to nonlinear data, and the ability to handle high-dimensional features, adapting to the nonlinear relationship between volumetric efficiency, mechanical efficiency, and remaining lifespan. Figure 7 , 8 As shown, volumetric efficiency and mechanical efficiency decline nonlinearly with lifetime. Compared with a single decision tree, random forests can reduce random errors and improve prediction stability by integrating multiple trees for prediction.

[0101] The model parameters were adjusted using a phased, two-parameter independent optimization strategy:

[0102] First, optimize the number of trees n_estimators. Within the first preset range, perform a large-scale positioning with a first step length to determine the optimal range within the first preset range. Within the optimal range, perform a fine search with a second step length to determine the value of n_estimators. The second step length is less than the first step length.

[0103] Next, the tree depth max_depth is optimized and adjusted with a third step size within a second preset range. The value of max_depth is determined by combining the overfitting gap analysis method.

[0104] Specifically, the number of initial trees n_estimators is 100 by default, the tree depth max_depth is unlimited by default, and other parameters include minimum number of samples per leaf node min_samples_leaf=2, minimum number of samples per node split min_samples_split=5, and random seed random_state=42 (to ensure reproducible results).

[0105] In the optimization of the number of trees n_estimators, the range of n_estimators is first set to [10, 200], with a step size of 10. The model R under different parameters is calculated by 5-fold cross-validation. 2(Coefficient of determination) and MSE (mean squared error); when n_estimators≥80, R 2 It tends to stabilize (approximately 0.96), and the MSE tends to be minimum (approximately 2 × 10⁻⁶). 7 The initial optimal range is determined to be [70, 90]. Within the range [70, 90], a step size of 1 is set, and cross-validation is performed again. When n_estimators = 76, cross-validation R... 2 The value reached 0.9717, and the MSE reached 1.89×10. 7 ,like Figure 15 The figure shows the optimal number of trees.

[0106] During the optimization of the tree depth (max_depth), the range of max_depth is set to [2, 10], the step size is 1, and the overfitting gap quantization analysis method is used to calculate the training set and validation set R. 2 Difference ΔR 2 When max_depth=5, the training set R 2 =0.9897, validation set R 2 =0.9717, ΔR 2 =0.018 is the smallest, such as Figure 16 , 17 As shown, at this point, the model achieves the optimal balance between accuracy and generalization ability; that is, if the depth is too large, ΔR... 2 Increasing the depth increases the risk of overfitting; insufficient depth reduces R0. 2 Reduce underfitting. By organically combining independent parameter optimization with overfit gap analysis, the parameter tuning efficiency is improved by about 3 times while ensuring model performance, providing a new solution for high-dimensional parameter optimization problems.

[0107] Model validation phase: Calculate model performance metrics based on the test set, such as... Figure 12 As shown, ensure R 2 ≥0.97, MSE≤1.9×10 7 RMSE ≤ 4400, and residual analysis: such as Figure 13 The residuals shown have a mean of ≤350 and a standard deviation of ≤4400, and the residuals are randomly distributed near the zero line.

[0108] Nine lifetime stages (10%, 20%, ..., 90%) were extracted from the original data. The optimized model was used to predict the remaining lifetime, ensuring that the prediction deviation percentage for each stage was ≤ ±2%. Figure 18As shown, the bias converges with increasing data volume, indicating that the model possesses full-cycle reliability from early warning to accurate prediction at the end of the lifespan, validating the practical value and engineering feasibility of the proposed method in predicting the remaining life of electric pumps. This multi-timescale validation method not only confirms the effectiveness of the random forest model in predicting remaining life, but also provides a reliable basis for maintenance decisions at different stages in engineering practice.

[0109] Comparison of training set and test set R 2 Difference, ensuring ΔR 2 If the value is less than 0.05, then there is no significant overfitting. The test results are: Training set R... 2 =0.9935, test set R 2 =0.9716, ΔR 2 =0.018 (<0.05) As shown in Figure 19, it is confirmed that the model has no overfitting and good generalization ability, and the training of the electric pump remaining life prediction model is completed.

[0110] In practice, the model pre-training process for the motor is the same as that for the electric pump, the difference being the different data collection points. Specifically:

[0111] Adjust the motor load to below rated operating conditions to stabilize the motor winding temperature rise at 60-80K, simulating a typical automotive motor operating load, and covering data points collected throughout the motor's entire lifespan, including active power. Apparent power S, bus voltage U, current I, and motor shaft extension torque Rotation speed Temperature before operation and the temperature after operation ;

[0112] The core characteristic of the failure mechanism calculation based on the motor's power loss leading to a decrease in power factor and energy conversion efficiency, and heat loss leading to an increase in temperature, is: power factor. motor efficiency Temperature rise .

[0113] Residual analysis was performed on the core features to identify outliers outside the 95% confidence interval, such as... Figure 20 , 21 As shown in Figure 22, R 2 All values ​​are ≥0.90, indicating a good fit. Missing data were filled and normalized. Min-Max normalization was used to map the three main features to the [0,1] interval, eliminating dimensional differences. Figure 23 As shown.

[0114] The power factor after normalization Motor efficiency and temperature rise The core input features of the model are used to verify the effectiveness of the features through the feature importance scoring function of the random forest model. Contribution rate = 53.76%, motor efficiency =37.92%, power factor =8.32%, such as Figure 25 As shown, temperature rise is the dominant factor in predicting motor lifespan, and all three characteristics are effective inputs.

[0115] The same staged dual-parameter independent optimization strategy as that used for electric pumps was adopted to determine the optimal parameters: n_estimators=80, max_depth=6 (because the feature dimension of the motor is increased by 1, the tree depth is slightly greater than that of the electric pump to ensure full utilization of feature information).

[0116] Model validation results, such as Figure 24 , 26 The test set R shown 2 =0.9777, MSE=1.57×10 7 RMSE = 125.57; For example Figure 27 As shown, the prediction deviation at each stage is ≤ ±0.9%, indicating high reliability throughout the entire cycle; overfitting test: as shown Figure 28 As shown, the training set R 2 =0.9942, test set R 2 =0.9777, ΔR 2 =0.0165, no overfitting.

[0117] The prediction models for the remaining life of electric pumps and motors both achieve high accuracy, high reliability, and strong generalization ability, which can meet the needs of the digital twin platform for predicting the life of electric pumps and motors throughout their entire life cycle.

[0118] This invention enables real-time monitoring, performance evaluation, and accurate prediction of remaining lifespan for motors and pumps. The platform, integrating a high-precision mechanical structure and an intelligent control system, boasts excellent versatility and automation while improving testing accuracy. Specifically, the slide table and electromagnetic servo hydraulic clamping system achieve adaptive clamping and rigid fixation of workpieces of different sizes, enhancing clamping efficiency and testing reliability. The test bench servo power system simulates the operating conditions of motors and pumps under real-world operating conditions, effectively reducing experimental errors. The integrated sensor module is rationally arranged according to physical causality and signal flow sequence, simultaneously collecting multi-dimensional data such as pressure, flow rate, temperature, current, voltage, and speed, providing a high-quality data foundation for health status assessment.

[0119] In terms of lifetime prediction modeling, this invention proposes key performance indicators for electric pumps and motors—volume efficiency and mechanical efficiency of electric pumps, and power factor, motor efficiency, and standardized temperature rise of motors—and constructs lifetime prediction models based on random forest regression models. Through data preprocessing, outlier removal, feature normalization, and backtracking validation, the prediction accuracy and generalization ability of the models are significantly improved. Experimental results show that the determination coefficients of the prediction models for electric pumps and motors reach 0.8682 and 0.9777, respectively, verifying the effectiveness and reliability of the models.

[0120] In summary, this lifespan prediction platform not only enables health status monitoring and lifespan prediction of motors and pumps throughout their entire lifecycle, but also provides a feasible technical path and experimental foundation for intelligent operation and maintenance and predictive maintenance of mechanical equipment, demonstrating promising engineering application prospects and widespread value. In the future, it can be further integrated with edge computing and cloud platforms to expand its practical applications in the Industrial Internet of Things (IIoT).

[0121] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of the invention. All equivalent changes made based on the description and drawings of the present invention are included within the scope of the present invention.

Claims

1. A digital twin electro-pump full life cycle health monitoring and lifespan prediction platform, characterized in that, include: Test bench, dual-line ball screw module, electromagnetic servo hydraulic fixture, sensor integrated module, control module, and test bench servo power system; A double-wire ball screw module is embedded in the central groove of the test bench. The double-wire ball screw module includes a guide rail, on which two slides are slidably mounted. A grid-shaped sliding support platform is mounted on the opposite side of the two slides. The top surface of the grid-shaped sliding support platform is flush with the top surface of the slide. An electromagnetic servo hydraulic clamp is mounted on the top surface of the slide. The slide is driven by a slide servo motor and a double-wire ball screw structure. The electromagnetic servo hydraulic clamp includes a square housing, within which multiple independent liquid chambers are arranged in an array. Each liquid chamber contains a hexagonal ejector pin, which extends through the liquid chamber from one side of the square housing to form an array arrangement. A piezoelectric ceramic sensor is integrated at the front end of the ejector pin. A modular blocking piston is arranged inside the square housing opposite to the liquid chamber. An electromagnetic telescopic rod connected to the modular blocking piston is mounted on the square housing. The square housing is connected to an oil storage tank, in which a spring-loaded reset piston is mounted. The sensor integration module is used to collect the operating parameters of electric pumps and motors. It includes an electric pump sensor module and a motor sensor module. The electric pump sensor module includes a pressure sensor, a flow sensor, and an electric pump temperature sensor. The motor sensor module includes a current sensor, a voltage sensor, a speed sensor, and a motor temperature sensor. The test bench servo power system is set on one side of the central groove on the test bench frame, and is used to connect with the motor and electric pump under test, so as to simulate the driving state of the motor and electric pump under test under real working conditions. The control module includes an industrial computer and a controller that are interconnected. The controller is electrically connected to the test bench servo power system, slide table servo motor, electromagnetic telescopic rod, and sensor integrated module. The industrial computer integrates a life prediction model based on a random forest regression model pre-trained. After obtaining feature inputs based on the operating parameters of the electric pump and motor, the industrial computer predicts the remaining life of the electric pump and motor through the life prediction model.

2. The digital twin electro-pump full life cycle health monitoring and life prediction platform according to claim 1, characterized in that, The electromagnetic servo hydraulic fixture is connected to the top surface of the slide table via a dovetail groove guide rail structure; the grid-shaped sliding support platform is connected to the slide table via a locking structure; the ejector pin adopts a three-section sleeve telescopic structure, and each section is equipped with a scraper ring at the bottom. The scraper ring is used to scrape the impurities attached to the inner wall into the square shell when the ejector pin retracts; a protective baffle is set on the test bench on the side of the test bench servo power system.

3. The digital twin electro-pump full life cycle health monitoring and life prediction platform according to claim 1, characterized in that, The liquid chamber is divided into multiple areas according to the design dimensions. Each area is equipped with a modular blocking piston. The modular blocking piston includes a rectangular body with a plunger adapted to the liquid chamber. A rubber sealing ring is provided around the rectangular body. A connecting rod is provided on the side of the rectangular body opposite to the plunger. The connecting rod passes through the square shell and connects to the electromagnetic telescopic rod.

4. The digital twin electro-pump full life cycle health monitoring and life prediction platform according to claim 3, characterized in that, The platform mounting process includes: Initial state and condition: The modular blocking piston is at the lowest end of its stroke, the spring of the oil tank's reset piston is in a compressed state, and the oil flows into the square housing through the oil tank and enters each liquid chamber, thereby pushing the ejector pin to fully extend and with its top end flush. Clamping state: The electric pump or motor under test is placed on the grid-shaped sliding support platform. The slide table drives the grid-shaped sliding support platform and the electromagnetic servo hydraulic clamp to move. After the grid-shaped sliding support platforms on both sides are inserted into each other, the ejector pin gradually contacts the surface of the electric pump or motor, and different axial displacements are generated due to the surface geometry. Locking phase: When the pressure at the tip of the ejector pin reaches the preset threshold, the electromagnetic telescopic rod drives the modular blocking piston to move, pumping redundant oil back to the storage tank and cutting off the hydraulic circuit, sealing the liquid chamber to achieve hydraulic locking, thereby forming an envelope-type adaptive fit.

5. The digital twin electro-pump full life cycle health monitoring and life prediction platform according to claim 1, characterized in that, The pre-training process for electric pumps includes: Data collection phase: The electric pump was driven by a servo power system on a test bench to simulate actual operating conditions inside a vehicle. The test pressure was controlled to be no more than 10 MPa and the oil temperature was maintained at 40-50℃, covering the entire lifespan of the electric pump from early to mid-stages to the decline stage. Data points were collected at fixed time intervals, including pump speed n, displacement q, and actual output flow rate. Actual operating torque Working pressure P and oil temperature data; Data preprocessing stage: Calculate the volumetric efficiency of the electric pump based on the collected data points. , Mechanical efficiency , ; The volumetric efficiency, mechanical efficiency and known remaining life were fitted with a quadratic polynomial. The residuals of each data point were calculated by residual analysis. Outliers outside the 95% confidence interval were removed, and the nearest mean replacement method was used to fill in the missing data. After filling in the missing data, the data points are normalized to map the volumetric efficiency and mechanical efficiency to the [0,1] interval, thus eliminating the difference in dimensions. Feature engineering phase: Normalized volumetric efficiency and mechanical efficiency were used as the core input features of the model. The effectiveness of the features was verified by the feature importance scoring function of the random forest model, confirming that the contribution of mechanical efficiency to the prediction results was no less than 75% and the contribution of volumetric efficiency was no less than 24%. Model building and training phase: A random forest regression model is adopted, and the data points are divided into training set and test set in a 7:3 ratio. Both training set and test set contain data of each life stage of electric pump. The model parameters are initialized. The model parameters were adjusted using a phased, two-parameter independent optimization strategy: First, optimize the number of trees n_estimators. Within the first preset range, perform a large-scale positioning with a first step length to determine the optimal range within the first preset range. Within the optimal range, perform a fine search with a second step length to determine the value of n_estimators. The second step length is less than the first step length. Then, the tree depth max_depth is optimized and adjusted with a third step size within the second preset range. The value of max_depth is determined by combining the overfitting gap analysis method. Model validation phase: Calculate model performance metrics based on the test set to ensure R 2 ≥0.97, MSE≤1.9×10 7 RMSE ≤ 4400, and residual mean ≤ 350, standard deviation ≤ 4400, with residuals randomly distributed near the zero line; Nine lifetime stage nodes (10%, 20%, ..., 90%) were extracted from the original data. The optimized model was used to predict the remaining lifetime, ensuring that the prediction deviation percentage of each node was ≤ ±2%. Comparison of training set and test set R 2 Difference, ensuring ΔR 2 <0.05, the training of the electric pump remaining life prediction model is complete.

6. The digital twin electro-pump full life cycle health monitoring and life prediction platform according to claim 5, characterized in that, The model pre-training process for electric motors is the same as that for electric pumps, the difference being the data collection points. Specifically: Adjust the motor load to below rated operating conditions to stabilize the motor winding temperature rise at 60-80K, simulating a typical automotive motor operating load, and covering data points collected throughout the motor's entire lifespan, including active power. Apparent power S, bus voltage U, current I, and motor shaft extension torque Rotation speed Temperature before operation and the temperature after operation ; The core characteristic of the failure mechanism calculation based on the motor's power loss leading to a decrease in power factor and energy conversion efficiency, and heat loss leading to an increase in temperature, is: power factor. motor efficiency Temperature rise ; Residual analysis was performed on the core features to identify outliers outside the 95% confidence interval, fill in the missing data, and perform normalization. The power factor after normalization Motor efficiency and temperature rise The core input features of the model are used to verify the effectiveness of the features through the feature importance scoring function of the random forest model. Contribution rate = 53.76%, motor efficiency =37.92%, power factor =8.32%, confirming that temperature rise is the dominant factor in predicting motor lifespan.

7. The digital twin electro-pump full life cycle health monitoring and life prediction platform according to claim 1, characterized in that, The oil storage tank is equipped with a high-pressure oil filter and a bypass circulation pump to achieve continuous filtration after the oil flows back to the storage tank, thus maintaining the cleanliness of the hydraulic circuit oil.

Images