Hydraulic pump energy efficiency optimization and predictive maintenance system and method based on digital twinning

Abstract

The present application relates to the technical field of fluid machinery operation and maintenance, in particular to a hydraulic pump energy efficiency optimization and predictive maintenance system and method based on digital twinning, comprising: constructing a hydraulic pump digital twin carrying a multi-physical field coupling simulation model to realize synchronous linkage of the physical pump entity and the twin; collecting real-time operation data of the equipment to drive simulation model operation, generating synchronous twin state data, and predicting hydraulic system load change trend; relying on an improved model predictive control algorithm to solve the optimal control sequence and adjust the hydraulic pump operation condition in real time; continuously monitoring the residual error between the measured data and the twin data, triggering fault diagnosis through a dynamic threshold, and accurately identifying potential faults; fusing fault information and twin state data to evaluate the equipment health degree and predict the remaining useful life, and implementing operation and maintenance and energy efficiency optimization strategies combined with the evaluation results. The present application realizes dynamic adaptation of working conditions, optimizes operation energy consumption, and strengthens fault prediction and whole-cycle operation and maintenance control ability.

Description

Technical Field

[0001] The present invention relates to the technical field of fluid machinery operation and maintenance, and particularly to a hydraulic pump energy efficiency optimization and predictive maintenance system and method based on digital twin. Background Art

[0002] The hydraulic pump is the core power component of the hydraulic transmission system and is widely used in various equipment such as industrial machinery and fluid transmission. At present, the daily operation and control of hydraulic pumps mostly adopt the operation mode of independent parameter collection and basic simulation analysis. The operation regulation of equipment is mostly passive adjustment according to the current real-time working conditions, lacking the anticipation and analysis of the subsequent load changes in the hydraulic system. The conventional simulation modeling method only conducts simulation operations for a single physical dimension, and it is difficult to achieve global state synchronous mapping between the virtual model and the physical entity of the equipment.

[0003] The existing technology control means are relatively single. The conventional control strategy is difficult to adapt to the dynamic fluctuating load requirements of the hydraulic system. The conventional operation algorithm has limited applicable scenarios and cannot output adaptable control instructions according to the comprehensive operation state of the equipment. The equipment fault monitoring generally uses fixed numerical thresholds to complete the abnormal determination. During the complex working condition switching process of the hydraulic pump, the operation parameters will have normal fluctuations, and the fixed determination standard is prone to abnormal recognition deviation, making it difficult to accurately identify hidden fault problems. The development of equipment operation and maintenance work lacks the support of data linkage. The state detection and life analysis are independent of each other and cannot comprehensively reflect the actual loss situation of the equipment.

[0004] The conventional technologies in the industry cannot achieve the virtual-real data linkage control under multi-physical field collaborative simulation and are difficult to complete dynamic energy efficiency adjustment relying on predictive load data. The fixed abnormal determination mechanism cannot adapt to the full working condition operation state of the hydraulic pump, and the fault diagnosis, health assessment, and life prediction cannot form a data fusion linkage, making it difficult to support the development of refined operation and maintenance operations. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies existing in the prior art and to propose a hydraulic pump energy efficiency optimization and predictive maintenance system and method based on digital twin.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: A hydraulic pump energy efficiency optimization and predictive maintenance method based on digital twin, including: Establish and synchronize the physical pump entity and its digital twin, and the digital twin includes a multi-physical field coupling simulation model; Collect the real-time operation data of the physical pump entity, drive the multi-physical field coupling simulation model to perform synchronous simulation, and generate twin state data; Use the load prediction model to predict the future demand of the hydraulic system and obtain the predicted load curve; Based on the predicted load curve and the twin state data, the optimal control sequence is solved using an improved model predictive control algorithm. The first control quantity in the optimal control sequence is sent to the actuator of the physical pump to adjust the pump's operating state; Simultaneously, the residual between the real-time operating data and the twin state data is monitored; When the residual exceeds a preset dynamic threshold, a fault diagnosis process is triggered to identify potential fault modes and locations. By integrating the fault diagnosis results with the twin status data, the health status of the pump is assessed and its remaining service life is predicted; Based on the remaining service life and operational energy efficiency assessment results, maintenance decisions and energy efficiency optimization strategies are formulated and implemented.

[0007] As a further aspect of the present invention, the establishment and synchronization of the physical pump entity and its digital twin includes: Obtain the three-dimensional geometric model, material properties, assembly relationships, and initial boundary conditions of the physical pump; Based on the aforementioned three-dimensional geometric model, material properties, and assembly relationships, a coupled numerical model reflecting fluid dynamics, structural mechanical properties, and thermodynamic properties is constructed in a simulation environment. The sensor network collects measured data of the physical pump under various typical operating conditions, including outlet pressure, flow rate, multi-point temperature and shell vibration signal. The key parameters in the coupled numerical model are identified and calibrated using the measured data to ensure that the model output and the measured data are consistent within a preset error range. A data channel based on an industrial communication protocol is established, and the real-time sensor data stream of the physical pump is used as the boundary condition for dynamic operation and input into the calibrated coupled numerical model to drive the model to perform real-time synchronous calculations and generate the twin state data.

[0008] As a further aspect of the present invention, the prediction of future demand for the hydraulic system using a load prediction model includes: Extract the action sequence, pressure demand, and flow demand of the hydraulic system actuators from historical databases and real-time data streams to form a load demand time series. The load demand time series is cleaned, outliers are removed, and standardization is performed. The processed time series data is input into the load forecasting model, which employs an improved time series forecasting algorithm. The improved time series prediction algorithm performs multi-scale decomposition on the input time series to obtain components that characterize the changing trends of different time periods. For each component, Long Short-Term Memory (LSTM) network units are used for feature learning and trend capture. The prediction results of each component are adaptively weighted and fused to generate a continuous predicted load curve for a future prediction time domain. The predicted load curve includes the pressure demand curve and the flow demand curve.

[0009] As a further aspect of the present invention, the working principle of the improved time series prediction algorithm is as follows: The multi-scale decomposition uses the empirical mode decomposition method to decompose the original load demand time series into a set of multiple intrinsic mode function components and a residual component; For each intrinsic modulus function component, an independent long short-term memory network sub-model is constructed. The input of the long short-term memory network sub-model is the historical window data of the current component, and the output is the future predicted value of the intrinsic modulus function component. For the residual components, an autoregressive integral moving average model is used for prediction; During the fusion phase, a dynamic weighting coefficient is assigned to the prediction result of each component. The dynamic weighting coefficient is adjusted in reverse according to the prediction error of the corresponding component in the historical prediction period. The component with the smaller error receives a higher weight. The weighted prediction results of all components are superimposed to obtain the final load demand prediction sequence, i.e., the predicted load curve.

[0010] As a further aspect of the present invention, the step of solving for the optimal control sequence using the improved model predictive control algorithm includes: Construct an objective function with the primary optimization objective of minimizing the total system energy consumption in the prediction time domain and the secondary optimization objective of controlling the gradual increase; Multiple constraints are set, including pump outlet pressure constraints, upper and lower limits of motor speed constraints, speed change rate constraints, and flow constraints to avoid cavitation. In each control cycle, the current pump operating state is used as the initial state, and the predicted load curve is used as the future disturbance input. Forward rolling simulation is performed on the multiphysics coupling simulation model of the digital twin. During the simulation, the improved model predictive control algorithm is used to solve the objective function that satisfies all the constraints online, and the optimal control quantity sequence for several control cycles from the current moment is obtained. The optimal control quantity sequence mainly refers to the motor speed setpoint sequence. The first control variable is extracted from the solved optimal control variable sequence and output as the setpoint for the current control cycle.

[0011] As a further aspect of the present invention, monitoring the residual between the real-time operating data and the twin state data includes: During the synchronous simulation of the digital twin, the calculated twin state data is acquired in real time, including the simulated outlet pressure, flow rate, key point temperature, and vibration spectrum. Simultaneously, real-time operating data of the physical pump entity is acquired from sensors, including actual outlet pressure, flow rate, key point temperature, and vibration spectrum. Calculate the difference between the twin state data and the real-time operation data corresponding to the same physical quantity to obtain a set of real-time residual sequences; Sliding window statistics are performed on the real-time residual sequence to calculate the mean and standard deviation of each residual within the window; The mean and standard deviation are input into a preset threshold calculation model to obtain the dynamic threshold corresponding to each monitoring signal. The dynamic threshold varies with the device working time, average load and ambient temperature. When the absolute value of any residual continuously exceeds its corresponding dynamic threshold for a preset duration, it is determined to be an abnormal event, triggering the fault diagnosis process.

[0012] As a further aspect of the present invention, the fault diagnosis process includes: Once fault diagnosis is triggered, the signal type that generates the persistent abnormal residual and its physical meaning are identified. Multiple fault simulation models related to the signal type are called from the model library of the digital twin, and the fault simulation models are preset with model parameter changes under different fault modes; The boundary conditions of the current physical pump's dynamic operation are input into each of the aforementioned fault simulation models to perform simulations and obtain simulation data under various assumed fault modes. Calculate the similarity between the real-time operating data and the simulation data under various hypothetical failure modes; The top few fault modes with the highest similarity were selected as candidate fault modes; By combining the pump's maintenance history and current load information, confidence assessments are performed on candidate failure modes to determine the potential failure modes and their locations.

[0013] As a further aspect of the present invention, the assessment of the pump's health status and prediction of its remaining service life includes: Based on the fault modes identified in the fault diagnosis process, the performance degradation rate model under the fault modes is obtained from the historical database and physical model. Based on the performance degradation rate model and the current equipment status, the performance index decline curves of the pump at different time points in the future are deduced. Define a comprehensive health index, which is a weighted fusion of the normalized values ​​of three performance indicators of the hydraulic pump: efficiency, outlet pressure pulsation amplitude, and vibration energy. Substituting the derived future performance indicators into the calculation formula of the comprehensive health index, a predicted curve of the health index changing over time is obtained. A failure threshold is set for the health index. When the prediction curve drops to the failure threshold, the corresponding time point is the predicted failure time. Calculate the time length from the current moment to the predicted failure time, and use it as the remaining service life.

[0014] As a further aspect of the present invention, the formulation and execution of maintenance decisions and energy efficiency optimization strategies includes: The remaining service life is compared with a preset maintenance cycle threshold. If it is less than the threshold, a preventive maintenance work order is generated. The work order includes the fault location identified by the fault diagnosis process, the suggested maintenance operation, and the required spare parts. Based on the twin state data and real-time operating data, the overall energy efficiency of the pump unit under the current operating conditions is calculated; Simulation tests were conducted on various potential energy efficiency optimization strategies in a digital twin, including adjusting the speed-pressure matching curve, optimizing the valve setting parameters in the hydraulic circuit, and suggesting system standby strategies. Compare the predicted energy efficiency improvement and implementation costs of various strategies after simulation testing; Select the strategy that satisfies cost constraints and maximizes energy efficiency improvement, and generate a specific optimization parameter setting scheme; The maintenance work order and the optimized parameter setting scheme are integrated into an execution command and sent to the field control station and maintenance management system.

[0015] As a further aspect of the present invention, the present invention also includes a hydraulic pump energy efficiency optimization and predictive maintenance system based on digital twins. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the hydraulic pump energy efficiency optimization and predictive maintenance method based on digital twins as described above.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: A digital twin of a hydraulic pump, integrating a multiphysics-coupled simulation model, was constructed to achieve real-time synchronous operation between the physical equipment and the virtual model. Real-time operational data from the physical equipment was collected to drive simulation calculations and generate real-time twin state data. A load prediction model was introduced to capture the load variation patterns of the hydraulic system. Based on an improved model predictive control algorithm, the optimal control sequence was calculated by combining load change trend data and twin state data. The control parameters at the beginning of the sequence were selected to adjust the equipment's operating state. The multiphysics coupling mode can cover the multi-dimensional operating characteristics of the equipment, synchronously simulating the actual working state of the equipment. The dynamic control quantity solution mode closely matches the dynamic changes in load and adapts to the continuous operation and adjustment requirements under complex working conditions.

[0017] The system collects physical equipment operation data and twin simulation data, continuously monitors residuals, and uses dynamic thresholds to replace fixed judgment criteria for anomaly screening. When numerical deviations exceed the fluctuation range, the system automatically initiates a fault diagnosis process to locate the fault location and type. It integrates various information collected for fault diagnosis with equipment operation status data recorded throughout the twin's lifecycle to classify equipment health levels and predict remaining service life. The dynamic thresholds can autonomously adjust their fluctuation range according to changes in operating conditions, adapting to parameter change characteristics across different equipment operating cycles. This multi-source data fusion analysis mode broadens the scope of equipment status monitoring and improves the comprehensive assessment of equipment wear and tear. Attached Figure Description

[0018] Figure 1 This is a flowchart of the hydraulic pump energy efficiency optimization and predictive maintenance method based on digital twins as described in this invention; Figure 2 A flowchart for establishing and synchronizing the operation of a physical pump entity and its digital twin; Figure 3 A flowchart illustrating the process of using a load forecasting model to predict future demand for hydraulic systems. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] See Figure 1 This invention provides a method for optimizing the energy efficiency and predictive maintenance of hydraulic pumps based on digital twins. The core of this method lies in constructing a digital twin highly synchronized with the physical pump entity, and using it to achieve a closed loop of operation optimization and health management. The overall implementation scheme is as follows: A physical pump entity and its digital twin are established and synchronized. This digital twin includes a multiphysics coupled simulation model that reflects the complex physical processes inside the pump. Real-time operating data of the physical pump entity is continuously collected through a sensor network, and this data drives the multiphysics coupled simulation model in the digital twin to perform synchronous simulation, thereby generating twin state data reflecting the internal state of the pump. Simultaneously, a load prediction model is used to predict the future pressure and flow requirements of the hydraulic system, obtaining a predicted load curve. Based on this predicted load curve and the twin state data provided by the digital twin, an improved model predictive control algorithm is used to continuously solve for the optimal control sequence over a future period.

[0022] In each control cycle, the first control variable in the optimal control sequence is sent to the actuator of the physical pump, such as the motor driver, to adjust operating parameters such as pump speed and achieve energy efficiency optimization. During operation, the residual between the real-time operating data of the physical pump and the twin state data generated by the digital twin is continuously monitored. When the residual exceeds a threshold dynamically calculated based on the operating conditions, a fault diagnosis process is triggered to identify potential fault modes and locations. Furthermore, the fault diagnosis results and twin state data are fused to assess the pump's current health status and predict its remaining service life. Based on the predicted remaining service life and the operating energy efficiency assessed based on the digital twin, specific maintenance decisions and energy efficiency optimization strategies are formulated and executed as instructions.

[0023] In one embodiment of the present invention, establishing and synchronizing the physical pump entity with its digital twin requires the construction of a digital model and data integration. See also... Figure 2The process involves obtaining a precise 3D geometric model of the physical pump, the material properties of each component, assembly relationships, and known initial boundary conditions. Based on this 3D geometric model, material properties, and assembly relationships, a coupled numerical model is constructed in a professional simulation software environment. This model must simultaneously reflect the interactions between the pump's internal fluid dynamics, structural mechanical properties, and thermodynamic properties. A sensor network deployed on the physical pump is used to collect measured data under various typical operating conditions, including outlet pressure, flow rate, temperature at multiple key points, and casing vibration signals. The collected measured data is then used to reverse-engineer and calibrate key parameters in the initially constructed coupled numerical model, such as the friction coefficient and heat transfer coefficient, ensuring that the error between the model's output and the measured data under the same input remains within a preset allowable range.

[0024] A data channel based on an industrial communication protocol is established, and real-time sensor data streams from the physical pump, such as inlet pressure and motor speed, are continuously input as boundary conditions to the calibrated coupled numerical model. This drives the model to perform real-time synchronous calculations, thereby generating twin state data corresponding to the physical entity.

[0025] In practice, the process of establishing and synchronizing the physical pump entity with its digital twin is a closed-loop process from acquiring physical information to calibrating the virtual model. The specific implementation is as follows: Before constructing the digital twin, a precise three-dimensional geometric model of the physical pump needs to be obtained from the product design and manufacturing archives. This model defines the spatial shape and dimensions of all flow channels, cavities, and moving parts of the pump. Simultaneously, the material properties of each component need to be acquired, including elastic modulus, density, specific heat capacity, and thermal conductivity, as well as clear component assembly relationships and tolerances. Furthermore, initial boundary conditions required for simulation need to be set, such as initial inlet pressure and initial ambient temperature. Based on the obtained three-dimensional geometric model, material properties, and assembly relationships, a coupled numerical model is constructed in a professional simulation software environment. This coupled numerical model is spatially discretized using the finite element method and computational fluid dynamics methods. Fluid dynamics equations, structural mechanics equations, and thermodynamic energy equations are set in the solver, and the bidirectional coupling relationship between these equations is defined to reflect the interaction between flow field pressure fluctuations, structural stress and strain, and temperature field during pump operation.

[0026] Flow control in the fluid domain employs incompressible Reynolds-averaged Navier-Stokes (NAS) data. Stokes' equations and the continuity equation are used to describe the turbulent motion of hydraulic oil in the complex flow channels within a pump. The continuity equation is as follows:

[0027] The momentum equation is

[0028] symbol Represents the velocity vector of the fluid. The density of the hydraulic oil, For fluid pressure, The effective viscosity is obtained by superimposing the molecular viscosity and the turbulent viscosity according to the selected turbulence model. This represents the volume force in the rotating reference frame. The turbulence model chosen is the shear stress transport model to balance computational accuracy in both the near-wall and mainstream regions. Its specific transport equations are consistent with the standard form and will not be elaborated further.

[0029] The structural mechanical response of a solid domain is described based on the linear elastic dynamic equations:

[0030] in For the displacement field of a solid, For the density of solid materials, For stress tensor, Force per unit volume. Constitutive relations obey the generalized Hooke's law, and the stress tensor is expressed through...

[0031] With strain tensor

[0032] Related, Shear modulus Let Lamé constant be . The trace of the strain tensor It is a unit tensor.

[0033] Thermodynamic properties are obtained through the transient heat conduction equation in the solid domain and the convection in the fluid domain. The diffusion energy equation reflects this together. The heat conduction equation in the solid domain is expressed as follows:

[0034] symbol For temperature, Specific heat capacity of solid, Thermal conductivity of solids This is the volumetric heat source in the solid domain. The energy equation for the fluid domain includes convection and viscous dissipation terms, expressed as follows:

[0035] and These are the specific heat capacity and thermal conductivity of the fluid, respectively. This is a viscous dissipation function, representing the heat generation effect of fluid friction.

[0036] Multiphysics coupling in flow The solid-state interface is achieved through bidirectional data transfer. Pressure is calculated in the fluid domain. With viscous stress vector As a mechanical load applied to the surface of the solid domain, it forms the boundary condition.

[0037] The unit outward normal vector of the interface; the displacement generated by the solid domain. The values ​​are then propagated back to the fluid domain to update the fluid mesh node coordinates, ensuring mesh continuity. Thermally coupled boundary conditions require temperature continuity on both sides of the interface.

[0038] Continuous with heat flux density:

[0039] A description of the unified interface heat exchange process.

[0040] Example of mesh parameters: The fluid domain employs a combination of polyhedral core mesh and near-wall prismatic layer meshes. The global maximum mesh size for key flow domains such as pump suction port, pressure port, and plunger cavity is controlled to the order of 1 to 2 mm. Local refinement is applied at gap leakage channels, with the mesh size reduced to the order of 0.1 mm to accurately capture leakage flow characteristics. Ten prismatic layer meshes are generated at the wall surface, with the height of the first layer set between 30 and 80 based on the target y+ value, and an interlayer growth factor of 1.2. The solid domain primarily uses hexahedral dominant meshes, with an average mesh size of approximately 2 mm for the shell and shaft components. Mesh refinement is applied to stress concentration areas. The total number of meshes in the overall coupled model is maintained at the order of millions to ensure numerical stability and solution efficiency.

[0041] Boundary condition examples: The fluid inlet boundary is set as a total pressure inlet, with the total pressure value given based on the characteristics of the upstream accumulator or oil supply system of the physical pump; the fluid outlet boundary is set as a mass flow outlet, with the flow rate value taken from real-time operating data; the rotor region encloses rotating components, and rotational motion set by the sensor's measured speed is applied through sliding mesh technology. Solid domain constraints: Full constraints are applied at the pump body anchor bolt mounting holes, and radial displacement constraints are applied at the journal bearing positions while retaining circumferential rotational freedom. Thermal boundary conditions: A natural convection heat transfer boundary is applied to the outer surface of the pump casing with the surrounding air, the convection heat transfer coefficient is set to a constant, and the far-field ambient temperature is taken from the field temperature sensor signal; internal flow... The solid-state interface automatically completes the heat transfer calculation through the aforementioned thermal coupling conditions.

[0042] After constructing a preliminary coupled numerical model, a sensor network deployed on the physical pump was used to collect measured data under various typical operating conditions. The sensor network included pressure sensors, flow meters, thermocouples, and accelerometers. The measured data covered temperature signals at multiple key points such as pump outlet pressure, system flow rate, bearing housing, and distribution plate, as well as vibration signals of the casing in three orthogonal directions. The collected measured data was used to perform reverse identification and calibration of key parameters in the coupled numerical model. Specifically, the measured sensor data under the same operating conditions was input into the model as boundary conditions. The simulation was run, and the model's output was compared with the corresponding sensor's measured data. By adjusting key parameters in the coupled numerical model, such as the friction coefficient, volumetric efficiency coefficient, convective heat transfer coefficient, and damping coefficient, the root mean square error between the model's output pressure, flow rate, temperature, and vibration spectrum and the measured data was brought within a preset allowable range, for example, with an upper limit of 3%. After calibration, the model output can accurately reproduce the actual behavior of the physical pump under multiple operating conditions. After calibration, a data channel based on an industrial communication protocol is established. The data channel typically uses OPCUA or industrial Ethernet protocol. The data stream collected by the sensors of the physical pump during real-time operation, including real-time speed, inlet pressure, and oil temperature, is continuously input into the calibrated coupled numerical model as dynamic boundary conditions. This drives the model to perform real-time synchronous calculations. The simulation calculation step size is strictly consistent with the data acquisition cycle. The digital twin's calculation engine outputs twin state data corresponding to the physical pump's state in real time. The twin state data includes simulated outlet pressure, instantaneous flow rate, key point temperature distribution cloud map, and vibration characteristic frequency.

[0043] In some embodiments, the calibration process for key parameters can be implemented using automated optimization algorithms, such as the friction coefficient. Taking the identification of [data] as an example, its optimization objective is to minimize the error function between the model output and the measured data. The expression for the error function E can be constructed as follows:

[0044] Where: symbol Represents the total error between the model output and the measured data, with the sign... Represents the coefficient of friction to be identified, symbol A vector representing other parameters to be identified, with symbols Representing the Data points, symbols Represents the total number of measured data points used for calibration, with the symbol... The representative of the coupled numerical model in the first... The output value at each data point, sign The representative sensor in the first Measured values ​​collected at each data point, sign Representative bestowed upon the first The weighting coefficients for each data point are used to adjust the importance of data from different operating conditions in the overall calibration. The optimization algorithm iteratively adjusts the parameters. The coupled numerical model is continuously run until the error function E falls below a preset threshold, at which point the model parameters are considered to have been identified and calibrated. It is understood that the measured data collected through the sensor network must cover the main operating range of the physical pump, including high, medium, and low pressure and flow conditions, as well as different temperature states such as cold start and thermal steady state, to ensure the accuracy of the coupled numerical model across the entire operating range. Optionally, vibration signal calibration can be performed in the frequency domain, comparing the shell vibration spectrum calculated by the model with the measured vibration spectrum, focusing on adjusting the structural damping parameters to match the amplitude of the main frequency components.

[0045] In practical implementation, the synchronization process relies on a stable and low-latency data link. Sensor data from the physical pump is aggregated and protocol-converted through an industrial gateway, and sent to the server running the digital twin with a fixed timestamp. Upon receiving the data, the server immediately updates it with the boundary conditions of the current calculation step of the coupled numerical model. After the coupled numerical model solver completes the calculation of one step, the output twin state data is stored and published for use by other modules, thereby achieving step-by-step synchronization and state mapping between the physical pump entity and the digital twin on a time scale. In some embodiments, to cope with sensor failures or communication delays, the digital twin can briefly operate in predictive mode, that is, using the state of the previous moment and the model itself for forward calculation, and then perform synchronization correction after the data link is restored. It can be understood that the multiphysics coupled simulation model of the digital twin is dynamic, not a static three-dimensional visualization model. Its internal states, such as the flow field pressure distribution, structural stress field, and temperature field, continuously evolve with the operation of the physical pump and are represented externally through the twin state data. Optionally, if the physical pump entity undergoes significant changes, such as the replacement of key components, it is necessary to re-execute part or all of the model parameter calibration process to ensure that the digital twin continues to maintain high-fidelity synchronization with the updated physical pump entity.

[0046] In one embodiment of the present invention, predicting the future demand of the hydraulic system using a load prediction model is a prerequisite for optimized control. See also... Figure 3The system extracts the action sequence, pressure demand, and flow demand of each actuator in the hydraulic system from historical databases and real-time data streams to form a load demand time series. This load demand time series undergoes data cleaning to remove outliers and is standardized. The processed time series is then input into a load prediction model employing an improved time series prediction algorithm. This improved algorithm performs multi-scale decomposition on the input time series, obtaining a series of components representing the changing trends at different time periods. For each component, a Long Short-Term Memory (LSTM) network unit is used for independent feature learning and trend capture. The prediction results corresponding to each component are adaptively weighted and fused to generate a continuous predicted load curve for a future prediction time domain. This curve includes the system's future pressure demand and flow demand curves. The specific working principle of the improved time series prediction algorithm involves decomposition and reconstruction. Multi-scale decomposition uses the Empirical Mode Decomposition (EMD) method to decompose the original load demand time series into a set of multiple intrinsic mode function components and a residual component. For each intrinsic mode function component, an independent LSM network sub-model is constructed. The input of this sub-model is the historical window data of the current component, and the output is the future predicted value of that component. For the residual components, an autoregressive integral moving average model is used for prediction. During the fusion phase, a dynamic weighting coefficient is assigned to the prediction result of each component. This dynamic weighting coefficient is adjusted inversely based on the prediction error of the corresponding component within the historical prediction period; components with smaller prediction errors receive higher weights. The weighted prediction results of all components are then superimposed to obtain the final load demand prediction sequence, i.e., the predicted load curve.

[0047] Based on the predicted load curve and twin state data, an improved model predictive control algorithm is used to solve for the optimal control sequence. A multi-objective optimization function is constructed, with minimizing the total system energy consumption in the predicted time domain as the primary optimization objective and smoothing the control increment as a secondary objective. Multiple constraints are also set, including the allowable range of pump outlet pressure, upper and lower limits of motor speed, speed change rate limit, and a minimum flow constraint to avoid cavitation. In each control cycle, the current operating state of the pump is used as the initial state, and the predicted load curve is used as the known future disturbance input. Forward rolling simulation is performed on the multiphysics coupled simulation model of the digital twin. During the simulation, the improved model predictive control algorithm is used to solve the optimization objective function that satisfies all constraints online, obtaining the optimal control sequence for several control cycles from the current moment. This sequence mainly refers to the motor speed setpoint sequence. The first control quantity is extracted from the solved optimal control sequence and output as the setpoint for the current control cycle to the actuator of the physical pump.

[0048] In practical implementation, a load forecasting model is used to predict the future demand of the hydraulic system. The input data comes from a historical database and real-time data streams. The historical database extracts the timestamps of the action commands of each actuator, target pressure values, and target flow rates during past operating cycles. The real-time data stream captures the current motion state of the actuators and valve opening signals. This information is arranged chronologically to form a load demand time series, which includes pressure demand subsequences and flow rate demand subsequences. The load demand time series is then cleaned. Cleaning rules include setting reasonable physical ranges for pressure and flow rates, removing outliers exceeding the sensor's range, and smoothing transient spike noise caused by signal interference using median filtering. The cleaned series is then standardized, converting pressure and flow rates of different dimensions to similar numerical ranges. Specifically, this is done by subtracting the mean from each subsequence and then dividing by its standard deviation.

[0049] The processed time series data is input into a load forecasting model, which employs an improved time series forecasting algorithm. The core step of this improved algorithm is decomposition followed by fusion. The input time series is decomposed into multiple components at different time scales. Each component represents the changing trend of different periods in the original sequence. For each component, an independent Long Short-Term Memory (LSTM) network unit is used for feature learning and trend capture. The LSM network unit can remember long-term dependencies in the time series. After obtaining the independent forecast results for each component, these results are adaptively weighted and fused. The adaptive weighted fusion dynamically allocates weights based on the historical performance of each component predictor, generating continuous predicted load curves for a future forecast time domain. These predicted load curves include the pressure demand curves and flow demand curves for the system over several future control periods.

[0050] The improved time-series prediction algorithm involves specific signal decomposition and neural network prediction processes. Multi-scale decomposition is achieved through the Empirical Mode Decomposition (EMD) method, which adaptively decomposes the original load demand time series into a set of multiple intrinsic mode function (IMF) components and a residual component. Each IMF component contains an inherent oscillation mode ranging from high to low frequency. For each IMF component, an independent Long Short-Term Memory (LSTM) network sub-model is constructed. The input of the LSM sub-model is a fixed-length historical window of data for the current component, and the output is the predicted value sequence of that IMF component in the future prediction time domain. For the residual component obtained after EMF decomposition, which typically exhibits a monotonic trend, an autoregressive integral moving average (ARM) model is used for prediction. In the fusion stage, a dynamic weight coefficient is assigned to the prediction result of each component. The assignment of the dynamic weight coefficient is adjusted inversely based on the prediction error of the corresponding component within the historical prediction period. The prediction error is calculated by comparing the historical prediction value with the actual value. Components with smaller prediction errors receive higher weights during fusion. The weighted prediction results of all components are superimposed. The superposition operation is a simple linear summation, which yields the final load demand prediction sequence. This final sequence is the predicted load curve.

[0051] In practical implementation, based on the predicted load curve and the twin state data provided by the digital twin, an improved model predictive control algorithm is used to solve for the optimal control sequence. An objective function is constructed with the primary optimization objective of minimizing the total system energy consumption in the prediction time domain. The total system energy consumption is calculated using a multiphysics coupled simulation model in the digital twin. The model predictive control algorithm also considers the smoothness of the control increment, treating the control increment as a secondary optimization objective, thus forming a multi-objective optimization problem. Multiple constraints are set, including that the pump outlet pressure must be maintained below the system safety valve setpoint, the motor speed must be between the rated speed and the minimum stable speed specified on the motor nameplate, the rate of change of speed must be less than the maximum allowable acceleration / deceleration rate of the driver, and the flow constraint ensures that the pump's suction flow rate is greater than the minimum flow rate to prevent cavitation.

[0052] At the start of each control cycle, the current pump operating state reflected in the digital twin is used as the initial state, and the predicted load curve output by the load prediction model is used as the known future disturbance input. Forward rolling simulation is performed on the multiphysics coupled simulation model of the digital twin, covering the entire prediction time domain. During the simulation, the improved model predictive control algorithm solves the multi-objective optimization function that satisfies all constraints online. The solution process can employ numerical optimization methods such as gradient descent or sequential quadratic programming to obtain the optimal control quantity sequence for several control cycles starting from the current moment. The optimal control quantity sequence mainly refers to the motor speed setpoint sequence. The first control quantity is extracted from the solved optimal control quantity sequence. This control quantity is the motor speed setpoint that should be executed immediately in the current control cycle, and this setpoint is sent to the physical pump's motor driver via the fieldbus.

[0053] In some embodiments, the calculation of dynamic weighting coefficients in the improved time-series prediction algorithm can be based on an adaptive formula that is dynamically adjusted according to historical prediction errors. The component in the first Weight coefficients at each time point The calculation formula is as follows:

[0054] Where: symbol Representing the Each component in Timing fusion weights, symbols Represents a positive scaling factor, symbol Representing the Each component in The prediction error at time (i.e., the previous prediction period), sign It is the previous sampling period (t) (Time 1) The difference between the measured value and the estimated value of a physical quantity, denoted by the digital twin. The total number of components, symbol This is a summation index. This formula ensures that the component with the smaller the historical prediction error has, the larger its weight, and that the sum of all weights is 1. It can be understood that the prediction time domain length of the load forecasting model needs to match the prediction time domain length of the model predictive control algorithm to ensure sufficient future information for optimization.

[0055] Optionally, in scenarios where load demand exhibits a strong periodic pattern, in some embodiments, solving the optimal control sequence using the model predictive control algorithm is an online optimization problem with high real-time computation requirements. Methods such as simplifying the model or pre-compiling an offline optimization table can be used to accelerate the online solution process. It can be understood that extracting only the first control variable from the optimal control sequence for execution is a rolling optimization strategy. In the next control cycle, the system will re-optimize based on the latest state and the updated predicted load curve, thereby achieving feedback correction for uncertainties. Optionally, when the load demand predicted by the load prediction model exceeds the safe operating range of the physical pump, the solution to the optimal control sequence will trigger a constraint handling mechanism. The optimization result will prioritize meeting equipment safety constraints, possibly at the expense of some energy efficiency.

[0056] In one embodiment of the present invention, monitoring the residual between real-time operating data and digital twin state data serves as a trigger mechanism for fault diagnosis. During the synchronous simulation of the digital twin, the calculated digital twin state data is acquired in real time, including simulated outlet pressure, flow rate, key point temperature, and vibration spectrum. Simultaneously, real-time operating data of the physical pump entity is acquired from the physical sensor network, including actual outlet pressure, flow rate, key point temperature, and vibration spectrum. The difference between the digital twin state data and the real-time operating data corresponding to the same physical quantity is calculated to obtain a set of real-time residual sequences. A sliding window statistical analysis is performed on this real-time residual sequence to calculate the average and standard deviation of each residual within the window. The statistically obtained average and standard deviation are input into a preset threshold calculation model to obtain a dynamic threshold corresponding to each monitoring signal. This dynamic threshold is not a fixed value but changes dynamically with factors such as equipment operating time, average load, and ambient temperature. When the absolute value of any residual continuously exceeds its corresponding dynamic threshold for a preset duration, it is determined to be an abnormal event, triggering the subsequent fault diagnosis process.

[0057] The fault diagnosis process is initiated upon detection of abnormal residuals. After triggering fault diagnosis, the signal type generating the persistent abnormal residuals and its physical meaning are identified. Multiple fault simulation models related to the identified signal type are retrieved from the digital twin's model library; these models predefine the parameter variation patterns under different fault modes. The current physical pump's operating boundary conditions are input into each fault simulation model for simulation, obtaining simulation data under various hypothetical fault modes. The similarity between the physical pump's real-time operating data and the simulation data under various hypothetical fault modes is calculated. The fault modes with the highest similarity are selected as candidate fault modes. Combining the pump's historical maintenance records and current load information, the confidence level of the candidate fault modes is evaluated to determine the most likely potential fault mode and location.

[0058] In practice, monitoring the residuals between real-time operational data and digital twin data is a continuous online process. The digital twin data is generated synchronously from the multiphysics coupled simulation model of the digital twin, including simulated outlet pressure, simulated instantaneous flow rate, simulated key point temperature distribution, and simulated vibration spectrum. Real-time operational data of the physical pump entity is acquired through a sensor network, including actual outlet pressure, actual flow rate, actual key point temperature, and actual vibration spectrum. The physical pump entity and the digital twin are strictly aligned using timestamps. The difference between the digital twin data and the real-time operational data corresponding to the same physical quantity is calculated. This difference calculation is performed independently for each monitored variable at each sampling time, forming a set of real-time residual sequences. These sequences include pressure residuals, flow rate residuals, temperature residuals, and vibration residuals. A sliding window statistical analysis is performed on the real-time residual sequences, covering a fixed time period of data. Within each sliding window, the mean and standard deviation of the residuals for each monitored variable are calculated. The calculated residual mean and residual standard deviation are input into a preset threshold calculation model. This model dynamically outputs a dynamic threshold for each monitoring signal based on the device's current operating time, average load, and ambient temperature. If the absolute value of the residual for any monitoring signal continuously exceeds its corresponding dynamic threshold for a preset duration (e.g., more than three consecutive sampling periods), it is considered an abnormal event, triggering the subsequent fault diagnosis process. Table 1 shows an exemplary data comparison segment demonstrating an anomaly detected and diagnosed within a sliding window.

[0059] Table 1: Export Pressure Monitoring Residuals and Dynamic Thresholds

[0060] In practical implementation, the fault diagnosis process is initiated immediately after an abnormal event is identified. The process identifies the signal type and its physical meaning that generates the persistent abnormal residual; for example, in the table above, it identifies an abnormal outlet pressure signal. Multiple fault simulation models related to the abnormal signal type are retrieved from the digital twin's model library. These models pre-define model parameter changes under different fault modes. For example, for an abnormal outlet pressure, the retrieved simulation models might include a "distributor plate wear model," a "plunger slipper loosening model," and a "damping orifice blockage model." The current physical pump's operating boundary conditions, including speed, inlet pressure, and oil temperature, are input into each fault simulation model for simulation, obtaining simulation data under various hypothetical fault modes. The simulation data primarily consists of simulated outlet pressure values ​​corresponding to these fault modes. The similarity between the physical pump's real-time operating data and the simulation data under various hypothetical fault modes is calculated. Similarity can be measured using correlation coefficients or dynamic time warping distances. The fault modes with the highest similarity are selected as candidate fault modes, such as "distributor plate wear" and "damping orifice blockage." Based on the pump's maintenance history and current load information, the confidence level of candidate failure modes was assessed. The maintenance history showed that the pump had not recently replaced the damping orifice, while the load showed that the system was under high pressure and high flow, which would accelerate the wear of the distribution plate. After comprehensive evaluation, "distribution plate wear" was determined to be the most likely potential failure mode, and the failure location was determined to be the distribution plate area.

[0061] In some embodiments, the dynamic threshold can be calculated based on an adaptive formula that comprehensively considers the device's operating status and the dynamic threshold. The calculation formula is as follows:

[0062] Where: symbol Represents the calculated dynamic threshold, symbol Represents the basic threshold constant, symbol Represents the working time of the equipment coefficient, sign Represents average load coefficient, sign Represents the absolute value of the ambient temperature deviation The coefficient. It can be understood that the dynamic threshold is not a fixed value; it varies with the cumulative operating time of the device. Increase, average load The rise and ambient temperature Deviation from reference temperature The degree of adaptability increases, which aligns with the natural degradation of equipment performance with use and aging. Optionally, for complex signals such as vibration spectra, residual calculation can be performed in the frequency domain. For example, comparing the amplitude differences between the actual vibration spectrum and the twin vibration spectrum in characteristic frequency bands (such as bearing passage frequencies) and using the amplitude difference as the vibration residual. In some embodiments, the fault simulation model can construct a fault feature library by pre-injecting different types of fault parameters (such as increasing fit clearance or changing damping coefficients) into the digital twin and running simulations. It is understood that confidence assessment is a decision-making step that integrates multi-source information. In addition to maintenance history and operating conditions, rule-based expert knowledge can also be introduced, such as "in low-temperature startup, increased pressure pulsation is more likely to be related to high oil viscosity than wear." Optionally, when multiple monitoring signals (such as pressure and vibration) simultaneously exhibit abnormal residuals, the fault diagnosis process will call a multi-signal joint fault simulation model for comprehensive diagnosis to improve diagnostic accuracy.

[0063] In one embodiment of the present invention, assessing the health status of the pump and predicting its remaining service life requires combining fault information and a performance degradation model. Based on the specific fault mode identified through the fault diagnosis process, a performance degradation rate model for that fault mode is obtained from a historical failure database and a physical failure model. Based on the obtained performance degradation rate model and the current equipment status assessed through a digital twin, the decline curves of the pump's key performance indicators at different future time points are extrapolated. A comprehensive health index is defined, which is a weighted fusion of the normalized values ​​of three key performance indicators: hydraulic pump efficiency, outlet pressure pulsation amplitude, and vibration energy. The extrapolated performance indicator values ​​for each future time point are substituted into the calculation formula of the comprehensive health index to obtain a predicted curve of the health index changing over time. A failure threshold for the health index is set; when the predicted health index curve drops to this threshold, the corresponding time point is the predicted failure time. The time length from the current moment to the predicted failure time is calculated; this length is the predicted remaining service life.

[0064] In practical implementation, assessing the pump's health status and predicting its remaining service life begins with identifying specific failure modes in the fault diagnosis process. For example, the fault diagnosis process identifies "distributor plate wear" as a potential failure mode. Based on the identified "distributor plate wear" failure mode, historical performance degradation data of the same model pump under the same or similar failure modes is retrieved from a historical database. A physical model describing the degradation law of key performance parameters under the "distributor plate wear" failure mode is obtained from a physical model library. Combining historical data and the physical model, a performance degradation rate model applicable to the current failure mode is obtained. This performance degradation rate model describes the mathematical relationship, such as volumetric efficiency, as a function of operating time or cumulative load. Based on the performance degradation rate model and the current equipment state obtained through digital twin assessment, using the current state as the starting point, the performance degradation rate model is integrated forward to extrapolate the decline curves of the pump's key performance indicators at different future time points. Key performance indicators include volumetric efficiency, outlet pressure pulsation amplitude, and vibration energy.

[0065] A comprehensive health index is defined, which is a weighted fusion of three key performance indicators of the hydraulic pump after normalization. These three indicators are pump efficiency, outlet pressure pulsation amplitude, and vibration energy. Normalization maps indicator values ​​of different dimensions and ranges to a range of 0 to 1. Weighting assigns different weight coefficients to each indicator to reflect its impact on the health status. The predicted performance indicator values ​​at different future time points are substituted into the comprehensive health index calculation formula to obtain the comprehensive health index values ​​at a series of future time points. These values ​​are then connected to form a predicted curve of the health index changing over time. A failure threshold for the comprehensive health index is set. This failure threshold is an empirical value between 0 and 1, representing the critical health level at which equipment function fails. When the predicted comprehensive health index curve drops to the failure threshold, the corresponding time point is the predicted equipment failure time. The time length from the current moment to the predicted failure time is calculated; this time length is the predicted remaining service life. Table 2 shows a segment of the comprehensive health index prediction based on performance degradation projection.

[0066] Table 2: Prediction Trajectory of Hydraulic Pump Comprehensive Health Index

[0067] In some embodiments, the comprehensive health index can be calculated using a linear weighted sum, as shown in the following formula:

[0068] Where: the symbol HI represents the calculated comprehensive health index, and the symbol... Normalized value of representative efficiency index Weights, signs Normalized value of pressure pulsation amplitude Weights, signs Represents the normalized value of vibrational energy The weights. Normalized values ​​of efficiency indicators. It is the ratio of the current efficiency value to the efficiency value of a brand new pump, and the normalized value of the pressure pulsation amplitude. It is the ratio of the current pressure pulsation value to the maximum permissible pressure pulsation value, and the normalized value of vibration energy. It is the ratio of the current vibration energy value to the maximum allowable vibration energy value. The sum of all weighting coefficients is 1, i.e. It is understood that the performance degradation rate model can be an empirical model based on historical data fitting, or an analytical model derived from physical failure mechanisms, such as a physical relationship model between wear depth and leakage. Optionally, when extrapolating the performance index decline curve, different future load profiles can be considered, i.e., assuming the equipment will operate under different load intensities in the future, resulting in multiple performance degradation trajectories and remaining service life prediction ranges under different load assumptions. In some embodiments, the failure threshold can be dynamically adjusted according to the criticality of the equipment and maintenance strategy; for critical equipment, a more conservative threshold can be used for early warning. It is understood that the constituent indicators and weights of the comprehensive health index can be adjusted according to the specific pump type and main failure modes; for example, for failures characterized by vibration, vibration energy can be given a higher weight. Optionally, the performance degradation extrapolation can be re-executed periodically (e.g., weekly) or when the equipment operating status changes significantly, updating the initial point of the extrapolation with the latest equipment status data, thereby dynamically correcting the predicted value of the remaining service life.

[0069] In one embodiment of the invention, the final output of the method is to formulate and execute maintenance decisions and energy efficiency optimization strategies. The predicted remaining service life is compared with a preset maintenance cycle threshold. If the remaining service life is less than the threshold, the system automatically generates a preventative maintenance work order. The work order includes the fault location identified by the fault diagnosis process, suggested maintenance operations, and required spare parts information. Simultaneously, based on the twin state data provided by the digital twin and the real-time operating data of the physical pump, the overall energy efficiency of the pump unit under the current operating conditions is calculated. Various potential energy efficiency optimization strategies are simulated and tested in the digital twin. These strategies include adjusting the pump's speed-pressure matching curve, optimizing the setting parameters of the relief valve or proportional valve in the hydraulic circuit, and suggesting that the system enter a low-power standby strategy during idle periods. The predicted energy efficiency improvement and required implementation costs of various strategies after simulation testing are compared. The strategy that meets the cost constraints and has the largest energy efficiency improvement is selected, and a specific optimization parameter setting scheme is generated. Finally, the generated maintenance work order and optimization parameter setting scheme are integrated into executable instructions and sent to the field control station and maintenance management system for execution.

[0070] In practical implementation, maintenance decisions and energy efficiency optimization strategies are formulated and executed. Based on the predicted remaining service life and operational energy efficiency assessment results, the predicted remaining service life is compared with a preset maintenance cycle threshold. The maintenance cycle threshold is a time margin set according to the equipment maintenance strategy. If the remaining service life is less than the maintenance cycle threshold, the system automatically generates a preventive maintenance work order. The preventive maintenance work order includes the fault location identified by the fault diagnosis process, the recommended maintenance operation for the fault location, and a list of spare parts required to complete the maintenance operation. Based on the twin status data provided by the digital twin and the real-time operating data of the physical pump, the comprehensive energy efficiency of the pump set under the current operating conditions is calculated through an energy flow model. The comprehensive energy efficiency is the ratio of hydraulic output power to motor input electrical power.

[0071] Multiple potential energy efficiency optimization strategies were simulated and tested in a digital twin. These strategies included adjusting the pump's speed-pressure matching curve, optimizing the pressure setpoint of the relief valve or the opening control parameters of the proportional valve in the hydraulic circuit, and suggesting that the system enter a low-power standby strategy during idle periods. The predicted energy efficiency improvement and implementation cost of each optimization strategy were compared after simulation testing. The predicted energy efficiency improvement was calculated by comparing the comprehensive energy efficiency before and after the optimization strategy implementation. The implementation cost included the man-hours required for equipment adjustment, potential increased energy consumption, and potential risk costs. The energy efficiency optimization strategy that met the cost constraints and achieved the largest energy efficiency improvement was selected, and a specific optimization parameter setting scheme was generated. This scheme existed in the form of an instruction list or parameter file. The generated preventative maintenance work orders and the optimization parameter setting scheme were integrated into execution instructions. These instructions contained clear action commands, target parameters, and execution sequences, and were distributed to the field control station and maintenance management system via the factory network. The field control station executed the optimization parameter settings, and the maintenance management system received and scheduled the maintenance work orders.

[0072] In practical implementation, maintenance decisions rely on a logical judgment between remaining useful life and a preset threshold. The preset maintenance cycle threshold can be set based on factors such as equipment downtime costs, spare parts procurement cycles, and maintenance team arrangements. For example, the maintenance cycle threshold can be set to 30 days. When the predicted remaining useful life is 25 days, since 25 days is less than 30 days, the system determines that a maintenance work order needs to be generated. Simulation tests of energy efficiency optimization strategies are conducted in a virtual environment provided by the digital twin. The simulation tests set the same future load profile for each strategy to be evaluated. This load profile comes from the predicted load curve output by the load prediction model. After simulating the application of each optimization strategy, the digital twin outputs the system energy consumption and output power within the simulation cycle, and then calculates the predicted comprehensive energy efficiency. The comparison process requires quantitative evaluation. One way to do this is to calculate the energy efficiency improvement benefit index for each strategy. The formula for calculating the energy efficiency improvement benefit index is:

[0073] Where: symbol The energy efficiency improvement index is represented by the symbol. The symbol represents the predicted overall energy efficiency improvement (expressed as a percentage or absolute value) after implementing this energy efficiency optimization strategy. This represents the total cost required to implement the strategy. It can be monetized or normalized into a dimensionless cost index. The strategy selection logic is to choose the energy efficiency improvement benefit index. The best strategy is one whose total cost is... Within permissible budget constraints. In some embodiments, if the energy efficiency improvement index of all strategies is below a certain minimum requirement, or the implementation cost exceeds the budget, optimization may not be executed immediately, and only a maintenance work order may be generated. It is understood that the integration of maintenance work orders with optimization parameter setting schemes may involve the arrangement of execution sequences, such as applying a new speed-pressure matching curve after completing preventative maintenance and replacing worn parts. Optionally, before applying the optimization parameter setting scheme, a safety verification simulation can be performed in a digital twin to ensure that the new parameters will not cause system pressure to exceed limits or generate instability. In some embodiments, for multiple optimization strategies with similar costs, the strategy with more sustained energy efficiency improvement rather than the highest instantaneous peak can be selected based on the equipment's long-term operating plan. It is understood that the issuance of execution instructions requires a reliable communication mechanism and confirmation feedback; after the field control station executes parameter modifications, it needs to send back the actual parameters to confirm successful updates. Optionally, after receiving a maintenance work order, the maintenance management system can automatically check spare parts availability in the inventory system and trigger a spare parts requisition or procurement process.

[0074] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for hydraulic pump energy efficiency optimization and predictive maintenance based on digital twin, characterized in that, include: Establish and synchronize the physical pump entity with its digital twin, the digital twin containing a multiphysics coupled simulation model; The real-time operating data of the physical pump entity is collected to drive the multi-physics coupled simulation model to perform synchronous simulation and generate twin state data. The future demand of the hydraulic system is predicted using a load forecasting model, resulting in a predicted load curve. Based on the predicted load curve and the twin state data, the optimal control sequence is solved using an improved model predictive control algorithm. The first control quantity in the optimal control sequence is sent to the actuator of the physical pump to adjust the pump's operating state; Simultaneously, the residual between the real-time operating data and the twin state data is monitored; When the residual exceeds a preset dynamic threshold, a fault diagnosis process is triggered to identify potential fault modes and locations. By integrating the fault diagnosis results with the twin status data, the health status of the pump is assessed and its remaining service life is predicted; Based on the remaining service life and operational energy efficiency assessment results, maintenance decisions and energy efficiency optimization strategies are formulated and implemented.

2. The digital twin based hydraulic pump energy efficiency optimization and predictive maintenance method of claim 1, wherein, The establishment and synchronization of the physical pump entity and its digital twin includes: Obtain the three-dimensional geometric model, material properties, assembly relationships, and initial boundary conditions of the physical pump; Based on the aforementioned three-dimensional geometric model, material properties, and assembly relationships, a coupled numerical model reflecting fluid dynamics, structural mechanical properties, and thermodynamic properties is constructed in a simulation environment. The sensor network collects measured data of the physical pump under various typical operating conditions, including outlet pressure, flow rate, multi-point temperature and shell vibration signal. The key parameters in the coupled numerical model are identified and calibrated using the measured data to ensure that the model output and the measured data are consistent within a preset error range. A data channel based on an industrial communication protocol is established, and the real-time sensor data stream of the physical pump is used as the boundary condition for dynamic operation and input into the calibrated coupled numerical model to drive the model to perform real-time synchronous calculations and generate the twin state data.

3. The method for optimizing hydraulic pump energy efficiency and predictive maintenance based on digital twins according to claim 1, characterized in that, The method of predicting future demand for hydraulic systems using a load forecasting model includes: Extract the action sequence, pressure demand, and flow demand of the hydraulic system actuators from historical databases and real-time data streams to form a load demand time series. The load demand time series is cleaned, outliers are removed, and standardization is performed. The processed time series data is input into the load forecasting model, which employs an improved time series forecasting algorithm. The improved time series prediction algorithm performs multi-scale decomposition on the input time series to obtain components that characterize the changing trends of different time periods. For each component, Long Short-Term Memory (LSTM) network units are used for feature learning and trend capture. The prediction results of each component are adaptively weighted and fused to generate a continuous predicted load curve for a future prediction time domain. The predicted load curve includes the pressure demand curve and the flow demand curve.

4. The method for optimizing hydraulic pump energy efficiency and predictive maintenance based on digital twins according to claim 3, characterized in that, The working principle of the improved time series prediction algorithm is as follows: The multi-scale decomposition uses the empirical mode decomposition method to decompose the original load demand time series into a set of multiple intrinsic mode function components and a residual component; For each intrinsic modulus function component, an independent long short-term memory network sub-model is constructed. The input of the long short-term memory network sub-model is the historical window data of the current component, and the output is the future predicted value of the intrinsic modulus function component. For the residual components, an autoregressive integral moving average model is used for prediction; During the fusion phase, a dynamic weighting coefficient is assigned to the prediction result of each component. The dynamic weighting coefficient is adjusted in reverse according to the prediction error of the corresponding component in the historical prediction period. The component with the smaller error receives a higher weight. The weighted prediction results of all components are superimposed to obtain the final load demand prediction sequence, i.e., the predicted load curve.

5. The method for optimizing hydraulic pump energy efficiency and predictive maintenance based on digital twins according to claim 1, characterized in that, Solving for the optimal control sequence using the improved model predictive control algorithm includes: Construct an objective function with the primary optimization objective of minimizing the total system energy consumption in the prediction time domain and the secondary optimization objective of controlling the gradual increase; Multiple constraints are set, including pump outlet pressure constraints, upper and lower limits of motor speed constraints, speed change rate constraints, and flow constraints to avoid cavitation. In each control cycle, the current pump operating state is used as the initial state, and the predicted load curve is used as the future disturbance input. Forward rolling simulation is performed on the multiphysics coupling simulation model of the digital twin. During the simulation, the improved model predictive control algorithm is used to solve the objective function that satisfies all the constraints online, and the optimal control quantity sequence for several control cycles from the current moment is obtained. The optimal control quantity sequence mainly refers to the motor speed setpoint sequence. The first control variable is extracted from the solved optimal control variable sequence and output as the setpoint for the current control cycle.

6. The method for optimizing hydraulic pump energy efficiency and predictive maintenance based on digital twins according to claim 1, characterized in that, Monitoring the residual between the real-time operating data and the twin state data includes: During the synchronous simulation of the digital twin, the calculated twin state data is acquired in real time, including the simulated outlet pressure, flow rate, key point temperature, and vibration spectrum. Simultaneously, real-time operating data of the physical pump entity is acquired from sensors, including actual outlet pressure, flow rate, key point temperature, and vibration spectrum. Calculate the difference between the twin state data and the real-time operation data corresponding to the same physical quantity to obtain a set of real-time residual sequences; Sliding window statistics are performed on the real-time residual sequence to calculate the mean and standard deviation of each residual within the window; The mean and standard deviation are input into a preset threshold calculation model to obtain the dynamic threshold corresponding to each monitoring signal. The dynamic threshold varies with the device working time, average load and ambient temperature. When the absolute value of any residual continuously exceeds its corresponding dynamic threshold for a preset duration, it is determined to be an abnormal event, triggering the fault diagnosis process.

7. The method for optimizing hydraulic pump energy efficiency and predictive maintenance based on digital twins according to claim 6, characterized in that, The fault diagnosis process includes: Once fault diagnosis is triggered, the signal type that generates the persistent abnormal residual and its physical meaning are identified. Multiple fault simulation models related to the signal type are called from the model library of the digital twin, and the fault simulation models are preset with model parameter changes under different fault modes; The boundary conditions of the current physical pump's dynamic operation are input into each of the aforementioned fault simulation models to perform simulations and obtain simulation data under various assumed fault modes. Calculate the similarity between the real-time operating data and the simulation data under various hypothetical failure modes; The top few fault modes with the highest similarity were selected as candidate fault modes; By combining the pump's maintenance history and current load information, confidence assessments are performed on candidate failure modes to determine the potential failure modes and their locations.

8. The method for optimizing hydraulic pump energy efficiency and predictive maintenance based on digital twins according to claim 1, characterized in that, The assessment of the pump's health status and prediction of its remaining service life includes: Based on the fault modes identified in the fault diagnosis process, the performance degradation rate model under the fault modes is obtained from the historical database and physical model. Based on the performance degradation rate model and the current equipment status, the performance index decline curves of the pump at different time points in the future are deduced. Define a comprehensive health index, which is a weighted fusion of the normalized values ​​of three performance indicators of the hydraulic pump: efficiency, outlet pressure pulsation amplitude, and vibration energy. Substituting the derived future performance indicators into the calculation formula of the comprehensive health index, a predicted curve of the health index changing over time is obtained. A failure threshold is set for the health index. When the prediction curve drops to the failure threshold, the corresponding time point is the predicted failure time. Calculate the time length from the current moment to the predicted failure time, and use it as the remaining service life.

9. The method for optimizing hydraulic pump energy efficiency and predictive maintenance based on digital twins according to claim 8, characterized in that, The formulation and implementation of maintenance decisions and energy efficiency optimization strategies include: The remaining service life is compared with a preset maintenance cycle threshold. If it is less than the threshold, a preventive maintenance work order is generated. The work order includes the fault location identified by the fault diagnosis process, the suggested maintenance operation, and the required spare parts. Based on the twin state data and real-time operating data, the overall energy efficiency of the pump unit under the current operating conditions is calculated; Simulation tests were conducted on various potential energy efficiency optimization strategies in a digital twin, including adjusting the speed-pressure matching curve, optimizing the valve setting parameters in the hydraulic circuit, and suggesting system standby strategies. Compare the predicted energy efficiency improvement and implementation costs of various strategies after simulation testing; Select the strategy that satisfies cost constraints and maximizes energy efficiency improvement, and generate a specific optimization parameter setting scheme; The maintenance work order and the optimized parameter setting scheme are integrated into an execution command and sent to the field control station and maintenance management system.

10. A hydraulic pump energy efficiency optimization and predictive maintenance system based on digital twins, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for optimizing the energy efficiency and predictive maintenance of hydraulic pumps based on digital twins as described in any one of claims 1 to 9.

Images