Life prediction method for hydraulic systems of flight simulators
By denoising and synchronizing the multi-source operating parameters of the hydraulic motion system of the flight simulator, the degradation characteristics of the hydraulic pump, control valve, cylinder seals and hydraulic oil are formed respectively. By performing unidirectional normalization and component contribution fusion, the problem of difficulty in uniformly constructing the degradation state of multiple components and multiple mechanisms is solved, and efficient prediction of remaining service life and model updating are achieved.
Patent Information
- Application Number
- CN202611287245.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-24
- Publication Date
- 2026-11-17
AI Technical Summary
Under long-term, high-frequency, and multi-action operating conditions, the degradation state of the hydraulic motion system of a flight simulator is difficult to construct in a unified manner due to the coupling of multiple components and mechanisms. The cross-component effects of hydraulic oil viscosity changes on cylinder seals and hydraulic pumps are difficult to reflect during the degradation state generation stage. Fixed degradation models and fixed component contribution relationships result in insufficient predictability and responsiveness of remaining service life.
Data on hydraulic pressure, temperature, flow rate, and actuator displacement of the flight simulator's hydraulic motion system are collected, denoised, and synchronized with time to form degradation characteristics of the hydraulic pump, control valve, cylinder seals, and hydraulic oil. These characteristics are then input into the corresponding degradation models, normalized, and fused to form a comprehensive degradation state sequence of the hydraulic system. The remaining service life is predicted using a time-series life prediction model, and the parameters of the multi-factor degradation model are updated through the operational archive database.
It enables the handling of continuous degradation trends while preserving the physical degradation sources of different components, improves the accuracy and responsiveness of remaining service life prediction, forms a data closed loop from state generation and system fusion to model update, enhances the responsiveness to the accelerated degradation stage of local components, and reduces the lack of cross-component coupling effects.
Smart Images

Figure CN123110185A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of aviation training equipment condition monitoring and hydraulic system health management technology, specifically to a method for predicting the lifespan of a flight simulator's hydraulic system. Background Technology
[0002] The hydraulic motion system of a flight simulator includes a hydraulic pump, control valves, hydraulic lines, cylinders, a hydraulic oil circulation unit, and a motion control unit. During training, the flight simulator's hydraulic motion system continuously undergoes various motion platform movements. The hydraulic pump continuously builds and releases pressure according to changes in the load of the motion platform, the control valves repeatedly perform switching actions, and the cylinders continuously reciprocate. Hydraulic oil circulates between the hydraulic pump, control valves, hydraulic lines, and cylinders, absorbing heat. Therefore, the flight simulator's hydraulic motion system will simultaneously undergo multiple degradation processes during long-term operation.
[0003] After prolonged operation, the clearances, internal leakage, and friction conditions of the internal moving parts of a hydraulic pump change, resulting in variations in actual oil supply flow, pressure build-up capacity, and volumetric efficiency. During long-term, high-frequency switching of the control valve, continuous relative motion occurs between the valve core and valve sleeve, altering the hydraulic force exerted on the valve core and the flow state at the valve orifice due to the pressure difference between the valve's inlet and outlet. The cylinder seals are affected by friction, pressure peaks, and oil temperature. Under circulation, heating, and shearing, the hydraulic oil undergoes physical property changes, with variations in viscosity further influencing the internal leakage of the hydraulic pump, the flow state of the control valve, and the resistance to cylinder movement.
[0004] Hydraulic pumps, control valves, cylinder seals, and hydraulic oil exhibit different physical degradation mechanisms. Evaluating the state of a single component alone cannot fully characterize the overall lifespan of a flight simulator's hydraulic motion system. While sensor signals such as pressure, temperature, flow rate, and displacement reflect operational changes, each signal is simultaneously influenced by the states of multiple components and training motion conditions. Directly converting multiple sensor signals into a single health parameter can easily lead to the loss of differences between various sources of physical degradation during the state construction phase. Furthermore, different degradation states differ in dimensions and directions of change. Hydraulic pump volumetric efficiency typically decreases with increasing degradation, while the wear of control valve spools, cylinder seals, and hydraulic oil typically increases with increasing degradation. Using these directly as a unified lifespan prediction input reduces the physical consistency of the system's degradation sequence.
[0005] The operating state of a flight simulator's hydraulic motion system changes with cumulative operating time, environmental conditions, hydraulic oil condition, and maintenance activities. A life prediction model that maintains fixed parameters over a long period will deviate from the current equipment condition. When the rate of change of a certain type of degradation state is significantly higher than that of other degradation states, the contribution relationship of fixed components will weaken the influence of the corresponding accelerated degradation state on the overall system degradation state. Changes in hydraulic oil viscosity simultaneously affect the cylinder seal contact area and the internal clearance leakage state of the hydraulic pump. If the cylinder seal wear degradation model, the hydraulic pump volumetric efficiency degradation model, and the hydraulic oil viscosity aging degradation model remain completely independent, the cross-component effects cannot be directly reflected in the degradation state generation stage.
[0006] Therefore, it is necessary to generate the degradation states of hydraulic pumps, control valves, cylinder seals, and hydraulic oil from synchronous multi-source operating parameters, convert degradation states with different dimensions and directions of change into a continuous system-level degradation state sequence, adjust the component contribution relationship according to the local degradation rate changes of different components, establish a coupling compensation relationship between degradation models for the propagation of hydraulic oil viscosity changes to cylinder seals and hydraulic pumps, and use the system-level degradation state sequence to predict the remaining service life. At the same time, the previous multi-factor degradation model is updated based on the new state information generated by subsequent training subjects, so that the life prediction continuously corresponds to the actual degradation state of the hydraulic motion system of the flight simulator. Summary of the Invention
[0007] The purpose of this invention is to provide a life prediction method for hydraulic systems of flight simulators, which solves the technical problems of the difficulty in uniformly constructing the degradation state of multi-component and multi-mechanism coupling under long-term high-frequency and multi-action operation of hydraulic motion systems of flight simulators, the difficulty in reflecting the cross-component influence of hydraulic oil viscosity changes on cylinder seals and hydraulic pumps during the degradation state generation stage, and the insufficient persistence and responsiveness of remaining service life prediction due to fixed degradation models and fixed component contribution relationships.
[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:
[0009] Methods for predicting the lifespan of hydraulic systems in flight simulators include:
[0010] The hydraulic motion system of the flight simulator collects oil pressure, oil temperature, flow rate, and actuator displacement during the action sequence of training subjects. Noise reduction and time synchronization are performed to obtain synchronized preprocessed operating parameter data.
[0011] The degradation features corresponding to the hydraulic pump, control valve, cylinder seals and hydraulic oil are formed from the synchronous preprocessing operation parameter data, and then input into the corresponding degradation model to obtain four types of degradation state quantities;
[0012] The four types of degradation state quantities are normalized in the same direction and fused according to the component contribution relationship to obtain the comprehensive degradation state quantity sequence of the hydraulic system;
[0013] The sequence of comprehensive degradation state quantities of the hydraulic system is input into the time-series life prediction model to obtain the predicted value of the remaining service life, and the predicted value of the remaining service life and the corresponding degradation state quantities are written into the operation archive database.
[0014] The parameters of the multi-factor degradation model are updated based on the statistical characteristics of the operating parameters and the degradation status assessment results generated by subsequent training subjects. The updated four types of degradation status quantities are then used for subsequent prediction of remaining useful life.
[0015] As a preferred technical solution of the present invention, the degradation characteristics of the hydraulic pump include cumulative working time, number of pressure cycles and oil temperature integral. The hydraulic pump volumetric efficiency degradation model uses a nonlinear autoregressive neural network to establish the mapping relationship between the hydraulic pump volumetric efficiency degradation state quantity and the cumulative working time, number of pressure cycles and oil temperature integral.
[0016] As a preferred technical solution of the present invention, the degradation characteristics of the control valve include the valve core switching frequency and the pressure difference between the inlet and outlet of the control valve. The wear degradation model of the control valve core adopts a support vector regression model to establish the mapping relationship between the wear degradation state quantity of the control valve core and the valve core switching frequency and the pressure difference between the inlet and outlet of the control valve.
[0017] As a preferred embodiment of the present invention, the degradation characteristics of the cylinder seal include the number of cylinder reciprocating motions, the peak load pressure, and the oil temperature. The cylinder seal wear degradation model determines the wear degradation state of the cylinder seal based on the cumulative effect of the number of cylinder reciprocating motions, the load effect of the peak load pressure, and the temperature correction amount of the oil temperature. The degradation characteristics of the hydraulic oil include the cumulative working time and the oil temperature. The hydraulic oil viscosity aging degradation model determines the hydraulic oil viscosity aging degradation state based on the temperature-related aging rate and the cumulative working time.
[0018] As a preferred technical solution of the present invention, the denoising and time synchronization includes: using 5-layer db8 wavelet decomposition and soft thresholding to reduce high-frequency noise in multi-source operating parameter data; using adjacent normal data points to linearly interpolate data points that exceed the mean ± 3 times the standard deviation of the corresponding signal sliding statistical window and are determined to be isolated anomalies; and aligning multi-source operating parameter data with different sampling rates to a 500Hz reference time axis according to the timestamp.
[0019] As a preferred technical solution of the present invention, the same-direction normalization includes directional conversion according to the change direction between different degradation state quantities and degradation degrees, so that an increase in the normalized value corresponds to a deeper degradation degree; the component contribution relationship includes the hydraulic pump volumetric efficiency degradation state quantity, the control valve core wear degradation state quantity, the cylinder seal wear degradation state quantity, and the hydraulic oil viscosity aging degradation state quantity, corresponding to fusion weights of 30%, 25%, 30%, and 15%, respectively.
[0020] As a preferred technical solution of the present invention, the time-series lifetime prediction model adopts a long short-term memory recurrent neural network. The long short-term memory recurrent neural network receives a continuous sequence of comprehensive degradation state quantities of the hydraulic system, updates the hidden state and cell state through input gate, forget gate and output gate, and outputs the remaining lifetime prediction value by the linear output layer according to the updated hidden state.
[0021] As a preferred technical solution of the present invention, the operation archive database stores the statistical characteristics of multi-source operation parameter data, four types of degradation state quantities, comprehensive degradation state quantities of hydraulic system, predicted value of remaining service life and model version according to training subjects; after the maintenance cycle is reached or the conditions for updating new training samples are met, new samples with corresponding calibration information are extracted from the operation archive database, incremental model updates are performed on at least one degradation model, and it is determined whether to adopt the updated model parameters based on the error changes of independent verification data.
[0022] As a preferred technical solution of the present invention, maintenance warning information is generated based on the comparison result between the remaining service life prediction value and the maintenance warning threshold. The maintenance warning information includes four types of degradation state quantities, the comprehensive degradation state quantity of the hydraulic system, the remaining service life prediction value, and priority inspection objects determined based on the four types of degradation state quantities.
[0023] As a preferred technical solution of the present invention, the multi-source operating parameter data is collected by pressure sensors and temperature sensors arranged at the hydraulic pump outlet, oil cylinder inlet and outlet and front and rear ends of the control valve, flow sensors arranged in the hydraulic main pipeline and displacement sensors that detect the position of the oil cylinder piston, and the data collection duration covers at least one complete training subject action sequence.
[0024] As a preferred technical solution of the present invention, the hydraulic motion system of the flight simulator adopts a dual-pump redundant structure consisting of a main hydraulic pump and a backup hydraulic pump, which respectively form the volumetric efficiency degradation state quantity of the main hydraulic pump and the volumetric efficiency degradation state quantity of the backup hydraulic pump. The corresponding hydraulic pump volumetric efficiency degradation state quantity is selected to participate in the formation of the comprehensive degradation state quantity sequence of the hydraulic system according to the actual load state of the main hydraulic pump and the backup hydraulic pump.
[0025] As a preferred technical solution of the present invention, an edge computing node is set up on the flight simulator side. The edge computing node receives multi-source operating parameter data and completes denoising, time synchronization and extraction of operating parameter statistical features. It then sends the operating parameter statistical features and degradation status assessment results to the remote prediction platform. The remote prediction platform saves the operating file according to the flight simulator equipment identity and updates the corresponding degradation model parameters. It then sends the verified updated degradation model parameters to the corresponding edge computing node.
[0026] The present invention also discloses a life prediction system for a hydraulic system of a flight simulator, including a signal acquisition module, a synchronization preprocessing module, a multi-factor degradation modeling module, a degradation state fusion module, a time-series life prediction and archive update module, and a processor and memory that support the synchronization preprocessing module, the multi-factor degradation modeling module, the degradation state fusion module, and the time-series life prediction and archive update module.
[0027] The signal acquisition module is connected to the pressure sensor, temperature sensor, flow sensor and displacement sensor in the hydraulic motion system of the flight simulator, and sends the acquired multi-source operating parameter data to the synchronous preprocessing module;
[0028] The synchronous preprocessing module sends the synchronous preprocessing operation parameter data to the multi-factor degradation modeling module;
[0029] The multi-factor degradation modeling module sends the degradation state values of hydraulic pump volumetric efficiency, control valve core wear, cylinder seal wear, and hydraulic oil viscosity aging to the degradation state fusion module.
[0030] The degradation state fusion module sends the comprehensive degradation state sequence of the hydraulic system to the time-series life prediction and archive update module;
[0031] The time-series lifetime prediction and archive update module outputs the remaining lifetime prediction value and writes the remaining lifetime prediction value and the degradation status assessment results formed by subsequent training subjects into the running archive database to update the degradation model parameters in the multi-factor degradation modeling module.
[0032] As a preferred embodiment of the present invention, in the step of normalizing the four types of degradation state quantities in the same direction and fusing them according to the component contribution relationship to obtain the comprehensive degradation state quantity sequence of the hydraulic system, a dynamic adaptive weight allocation strategy is adopted to determine the component contribution relationship. The specific process includes:
[0033] Construct a state evolution matrix that includes the volumetric efficiency degradation state of the hydraulic pump, the wear degradation state of the control valve core, the wear degradation state of the cylinder seals, and the aging degradation state of the hydraulic oil viscosity.
[0034] Calculate the local degradation gradient of each type of degenerate state variable in the state evolution matrix within a preset time window;
[0035] A logistic nonlinear activation function with the local degradation gradient as the independent variable is established, and the degradation acceleration penalty factor corresponding to various degradation state variables is calculated.
[0036] The preset static benchmark weights are multiplied by the degradation acceleration penalty factor and normalized to obtain the dynamic fusion weights corresponding to various degradation state quantities at the current state time. The dynamic fusion weights are then used to obtain the comprehensive degradation state quantity sequence of the hydraulic system.
[0037] As a preferred embodiment of the present invention, in the step of inputting the corresponding degradation models to obtain four types of degradation state quantities, a hydraulic viscosity-temperature coupling evolution compensation mechanism is introduced to correct the outputs of the cylinder seal wear degradation model and the hydraulic pump volumetric efficiency degradation model. The specific process includes:
[0038] Based on the hydraulic oil viscosity aging and degradation state quantity output by the hydraulic oil viscosity aging and degradation model, and the calibration mapping relationship between the hydraulic oil viscosity aging and degradation state quantity and the current dynamic viscosity of the hydraulic oil, the physical viscosity decay coefficient at the current state moment is determined.
[0039] The physical viscosity decay coefficient is used as a boundary condition to input the wear and degradation model of the cylinder seal, and a viscosity-related hydrodynamic lubrication film thickness decay equation is established. The temperature correction amount in the wear and degradation model of the cylinder seal is then corrected based on the hydrodynamic lubrication film thickness decay equation.
[0040] An internal leakage amplification factor is constructed based on the physical viscosity decay coefficient. The internal leakage amplification factor is then expanded into an internal leakage compensation tensor with the same dimension as the hidden layer feature matrix of the nonlinear autoregressive neural network. The hidden layer feature matrix and the internal leakage compensation tensor are subjected to a Hadamard product operation. The compensated hidden layer feature matrix obtained from the operation is then input into the subsequent network layers of the nonlinear autoregressive neural network to obtain the hydraulic pump volumetric efficiency degradation state quantity after coupling compensation.
[0041] Compared with the prior art, the present invention has the following beneficial effects:
[0042] This invention employs a multi-mechanism coupled degradation life prediction method for hydraulic motion systems in flight simulators. It denoises and synchronizes the time of multi-source operating parameter data (oil pressure, oil temperature, flow rate, and actuator displacement) to generate degradation state quantities for hydraulic pump volumetric efficiency, control valve spool wear, cylinder seal wear, and hydraulic oil viscosity aging. These four types of degradation state quantities are then normalized and their component contributions are fused to form a comprehensive degradation state quantity sequence for the hydraulic system. This processing enables the time-series life prediction model to handle continuous degradation trends while preserving the physical degradation sources of different components, improving the correlation between remaining service life prediction and hydraulic pump performance degradation, control valve wear, sealing performance changes, and hydraulic oil aging. The system stores statistical characteristics of operating parameters, the four types of degradation state quantities, remaining service life prediction values, and model versions in an operational archive database. The multi-factor degradation model parameters are updated based on calibration information generated from subsequent training subjects, ensuring that life prediction continuously reflects the coupled degradation state of multiple components in the flight simulator's hydraulic motion system, forming a data closed loop from state generation, system fusion, life prediction to model updating.
[0043] Furthermore, the dynamic adaptive weight allocation strategy calculates the local degradation gradient of various degradation state quantities based on the state evolution matrix, forms a degradation acceleration penalty factor through the logistic nonlinear activation function, and adjusts the static benchmark weights to dynamic fusion weights. This enhances the response capability of the comprehensive degradation state quantities of the hydraulic system to the accelerated degradation stage of local components, avoiding the weakening effect of fixed weights on the overall lifespan of the system due to accelerated degradation. The hydraulic viscosity-temperature coupling evolution compensation mechanism corrects the hydrodynamic lubrication film thickness change in the cylinder seal wear degradation model based on the physical viscosity decay coefficient, and performs Hadamard product compensation on the hidden layer feature matrix of the nonlinear autoregressive neural network through the internal leakage compensation tensor. This allows the hydraulic oil viscosity change to propagate to the degradation state of cylinder seals and hydraulic pump volumetric efficiency, establishing a cross-model data connection between hydraulic medium degradation and mechanical component degradation, reducing the lack of cross-component coupling influence of independent degradation models. Attached Figure Description
[0044] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.
[0045] Figure 1 This is a flowchart of the method of the present invention.
[0046] Figure 2 This is a flowchart of the synchronous preprocessing subprocess of the present invention.
[0047] Figure 3 This is a flowchart of the hydraulic viscosity-temperature coupling evolution compensation subprocess of the present invention.
[0048] Figure 4 This is a flowchart of the dynamic adaptive weight allocation sub-process of the present invention. Detailed Implementation
[0049] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.
[0050] The following is in conjunction with the appendix Figures 1-4 The embodiments of the present invention will be described in detail below.
[0051] The technical solutions in the embodiments of the present invention will be clearly and completely described. The hydraulic motion system of the flight simulator includes a hydraulic power source, hydraulic control elements, hydraulic actuators, and a hydraulic oil circulation circuit. The hydraulic pump supplies pressurized oil to the hydraulic circuit, the control valve changes the oil flow and direction according to the motion control process, the oil cylinder drives the motion platform to produce displacement and attitude changes, and the hydraulic oil circulates between the hydraulic pump, control valve, hydraulic pipeline, and oil cylinder.
[0052] In practice, a complete training subject can be considered as one state observation cycle, or it can be divided into multiple state moments according to a fixed time window. Model training and online prediction use the same state time granularity, ensuring that the historical degradation states in the runtime archive have a consistent time definition with the current degradation state.
[0053] Example 1: This example uses a single flight simulator hydraulic motion system. Pressure sensors are installed at the hydraulic pump outlet, cylinder inlet and outlet, and the front and rear ends of the control valve; temperature sensors are installed at locations that reflect the thermal state of the oil supply and return; flow sensors are installed in the main hydraulic pipeline; and displacement sensors are used to detect the piston position of the main actuator cylinder. Pressure, temperature, flow, and displacement together constitute multi-source operating parameter data.
[0054] The pressure sensor has a measurement range of 0–35 MPa and an accuracy class of 0.2%; the temperature sensor has a measurement range of -40–150℃ and an accuracy of ±0.5℃; the flow sensor has a measurement accuracy of 0.5%; and the displacement sensor is a magnetostrictive displacement sensor with a measurement range of 0–500 mm and a linearity of 0.05%. The high-speed operating parameter channel uses a 500Hz sampling frequency, and the data acquisition duration covers at least one complete training subject action sequence.
[0055] In the specific implementation, step S1: Collect multi-source operating parameter data; the hydraulic pump outlet pressure is used to describe the pressure supply status and pressure cycle; the cylinder inlet and outlet pressures are used to describe the actuator load status; the pressures before and after the control valve are used to calculate the pressure difference between the control valve inlet and outlet; the oil temperature is used to determine the hydraulic pump's thermal exposure and hydraulic oil aging status; the flow rate is used to characterize the actual oil supply status of the hydraulic pump; and the displacement is used to identify the cylinder stroke, direction of movement, and number of reciprocating movements. Each sampling record includes equipment identity, training subject identifier, channel identifier, sampling time, and sampled value.
[0056] Step S2: Generate synchronous preprocessed operating parameter data; perform 5-level decomposition of pressure, flow, and displacement signals using db8 wavelet, and reconstruct the signals after soft thresholding of detail coefficients. The wavelet threshold is determined based on the noise statistical characteristics of the corresponding decomposition layer, and is used to reduce the impact of sensor electronic noise, mechanical vibration coupling, and electromagnetic interference on the construction of degradation features.
[0057] The pressure signal is calculated using a sliding statistical window to determine its mean and standard deviation. If a data point exceeds the mean of the current window by ±3 times the standard deviation, the corresponding data point is marked as an anomaly candidate. The anomaly candidate is then assessed based on its duration, flow rate changes, and displacement changes to determine whether it represents a genuine hydraulic action or an isolated anomaly. When pressure changes persist across multiple sampling points and are accompanied by flow rate or displacement changes, the corresponding pressure changes are retained. When an anomaly occurs in isolation and the preceding and following data are continuous, linear interpolation is performed using adjacent normal data points.
[0058] After noise reduction and anomaly processing, each sensor channel is mapped to a 500Hz reference time axis based on its timestamp. Low-sampling-rate temperature signals are interpolated to calculate the temperature value at a corresponding unified time point, ensuring that each unified time point contains corresponding pressure, temperature, flow rate, and displacement data, forming synchronized preprocessed operating parameter data.
[0059] Step S3: Generate four types of degradation state variables; the cumulative working time of the hydraulic pump is obtained by accumulating the time the hydraulic motion system is in an effective pressure supply state; the number of pressure cycles is obtained by statistically analyzing the complete loading and unloading process of the hydraulic pump outlet pressure; and the oil temperature integral is obtained by accumulating the synchronous temperature signal along the effective working time. The cumulative working time, the number of pressure cycles, and the oil temperature integral are jointly input into the hydraulic pump volumetric efficiency degradation model.
[0060] A nonlinear autoregressive neural network was used to establish the mapping between the three types of hydraulic pump degradation characteristics and the hydraulic pump volumetric efficiency degradation state. The network's hidden layer had 32 neurons, the activation function was a hyperbolic tangent function, the learning rate was set to 0.001, and the input window was set to 100 sampling periods. A time-correlation relationship was established between hydraulic pump performance calibration data and the corresponding degradation characteristics of the training subjects, which was used to adjust the parameters of the hydraulic pump volumetric efficiency degradation model.
[0061] The valve spool switching frequency of the control valve is statistically analyzed based on changes in the control valve signal. When the control valve signal cannot be directly obtained, valve spool switching events are identified by combining the pressure relationship between the front and rear ends of the control valve and the displacement direction of the actuator. To reduce repeated counting caused by small pressure oscillations, duration conditions and minimum time intervals between adjacent events are set for valve spool switching events.
[0062] The pressure difference between the inlet and outlet of the control valve is formed by the upstream and downstream pressures. Within a state window, the average pressure difference, maximum pressure difference, and pressure difference during the switching process are generated. The support vector regression model generates the wear and degradation state of the control valve spool based on the spool switching frequency and the pressure difference between the inlet and outlet of the control valve. The radial basis function parameter is set to 0.01, the penalty coefficient is set to 100, and the insensitive loss width is set to 0.1.
[0063] The number of reciprocating motions of the hydraulic cylinder seals is counted based on the change in the cylinder's displacement direction and the effective stroke conditions. One effective reciprocating motion is counted after the cylinder completes a reciprocating motion that meets the minimum stroke conditions; small oscillations near the position are not counted as complete reciprocating motions. The peak load pressure is determined based on the cylinder's inlet and outlet pressures and the piston's direction of movement, while the oil temperature is taken from the synchronous temperature data within the corresponding action window.
[0064] The hydraulic cylinder seal wear degradation model uses the number of reciprocating motions to form a cumulative wear baseline. The wear contribution corresponding to the current action is corrected based on the peak load pressure, and the wear contribution corresponding to the current action is corrected based on the oil temperature. The wear contribution from different action windows accumulates continuously along the training course and cumulative running time, resulting in the hydraulic cylinder seal wear degradation state quantity.
[0065] The hydraulic oil viscosity aging degradation model uses the cumulative effective working time and oil temperature of the hydraulic oil as inputs. The cumulative exposure time of the hydraulic oil at different temperature levels is used to form the thermal aging state, and the periodic viscosity test results of the hydraulic oil are used to correct temperature-related aging parameters. When the hydraulic oil is replaced, the cumulative working time of the hydraulic oil is re-established based on the replacement record, and the new hydraulic oil viscosity aging degradation state continues to form with the new service cycle.
[0066] Step S4: Form a comprehensive degradation state sequence of the hydraulic system; the degradation state of hydraulic pump volumetric efficiency usually manifests as a decrease in volumetric efficiency as the degree of degradation deepens, while the wear of control valve core, wear of cylinder seals and aging of hydraulic oil usually increase as the degree of degradation deepens. Therefore, before fusion, the four types of degradation states are unified in direction and scale.
[0067] For state variables whose original state values increase with increasing degradation, forward normalization is used; for state variables whose original state values decrease with increasing degradation, reverse normalization is used. After normalization, all four types of degenerate states satisfy the condition that an increase in value within the range of 0 to 1 corresponds to a deeper degree of degradation.
[0068] In this embodiment, the four types of degradation states are used to form the overall degradation state quantity of the hydraulic system according to the following relationship:
[0069] ;
[0070] in, For the first The comprehensive degradation state quantity of the hydraulic system corresponding to each state moment; The state numbers are arranged according to a uniform state time granularity; For the first Normalized value of the hydraulic pump volumetric efficiency degradation state quantity corresponding to each state moment; For the first The normalized value of the wear and degradation state of the control valve core at each state moment; For the first Normalized value of the wear and degradation state of the cylinder seals at each state moment; For the first The normalized value of the hydraulic oil viscosity aging and degradation state quantity corresponding to each state moment.
[0071] The , , and All are normalized dimensionless degenerate state quantities, and 0.30, 0.25, 0.30, and 0.15 are dimensionless weights. Since it is a dimensionless quantity, the dimensions on both sides of the formula are consistent.
[0072] For example, when the normalized value of the hydraulic pump volumetric efficiency degradation state quantity at the same time is 0.35, the normalized value of the control valve core wear degradation state quantity is 0.28, the normalized value of the cylinder seal wear degradation state quantity is 0.52, and the normalized value of the hydraulic oil viscosity aging degradation state quantity is 0.55, the four contributions are 0.105, 0.070, 0.156, and 0.0825, respectively, and the comprehensive degradation state quantity of the hydraulic system is 0.4135.
[0073] Multiple comprehensive degradation state quantities of the hydraulic system are continuously generated according to the same time granularity and arranged in chronological order to obtain a sequence of comprehensive degradation state quantities of the hydraulic system. The normalized boundary and the corresponding model version are stored together. When the model version is updated, the normalized parameters of the old version are retained, so that historical states can be traced according to the corresponding model version.
[0074] Step S5: Predict remaining useful life and update the model; The long short-term memory recurrent neural network receives the comprehensive degradation state sequence of the continuous hydraulic system, uses the input gate, forget gate and output gate to recursively encode the historical degradation state, and outputs the predicted remaining useful life value corresponding to the current state through the linear output layer.
[0075] The hidden layer of the Long Short-Term Memory (LSTM) recurrent neural network has 64 units, and the sequence input length is set to 200 time steps. The model training uses the Adam optimization method with a learning rate of 0.0005. The time steps are consistent with the granularity of the generation of the comprehensive degradation state variables of the hydraulic system. When the comprehensive degradation state variables of the hydraulic system are formed state by state along a 500Hz time axis, 200 time steps correspond to 0.4s; when the comprehensive degradation state variables of the hydraulic system are formed according to a longer state window, the time steps correspond to the duration of the state window.
[0076] After each training course is completed, the running archive database saves the training course identifier, start and end time, pressure statistics, temperature statistics, flow statistics, cylinder displacement statistics, four types of degradation state quantities, comprehensive degradation state quantities of the hydraulic system, predicted remaining service life, and model version.
[0077] The performance test results of the hydraulic pump, the maintenance and inspection results of the control valve, the test results of the cylinder sealing performance, and the test results of the hydraulic oil viscosity are established with time-related relationships to the degradation characteristics of the corresponding training subjects. When the maintenance cycle is reached or the new calibration data meets the update conditions, the corresponding data is extracted from the operation archive database to update the relevant degradation model.
[0078] During model updates, data with corresponding calibration information is divided into parameter update data and independent validation data. After parameter updates are completed, the updated model error is calculated on the independent validation data. When the validation error of the updated model is lower than that of the original model and meets the preset acceptance criteria, a new model version is generated; when the updated validation error does not meet the acceptance criteria, the original model parameters are used.
[0079] In some embodiments, the update data is composed of the 100 most recent training records with valid calibration information, of which 80 training records are used for parameter updates and 20 training records are used for independent validation. The current model update process ends when the improvement in validation error between two consecutive update rounds is less than 1%. The updated model parameters are directly used for subsequent degradation state calculations, allowing the model update results to re-enter the remaining lifetime prediction data chain.
[0080] When the predicted remaining service life is lower than the maintenance warning threshold, a maintenance warning is generated. The maintenance warning information includes the current hydraulic pump volumetric efficiency degradation status, control valve spool wear degradation status, cylinder seal wear degradation status, hydraulic oil viscosity aging degradation status, overall hydraulic system degradation status, and the predicted remaining service life. Priority inspection targets are determined based on the relative status of the four types of degradation status.
[0081] Example 2: This example uses a dual-pump redundant hydraulic motion system for a flight simulator, consisting of a main hydraulic pump and a backup hydraulic pump. The same data processing relationships as in Example 1 are used for multi-source operating parameter acquisition, synchronous preprocessing, formation of control valve core wear and degradation states, formation of cylinder seal wear and degradation states, formation of hydraulic oil viscosity aging and degradation states, and prediction of remaining service life.
[0082] The main hydraulic pump and the standby hydraulic pump respectively store the cumulative operating time, pressure cycle count, oil temperature integral, and hydraulic pump volumetric efficiency degradation status. When the main hydraulic pump is under actual load, the degradation characteristics of the main hydraulic pump are updated according to the actual operating time and pressure cycle of the main hydraulic pump; when the standby hydraulic pump is not under load, the operating time and pressure cycle of the main hydraulic pump are not copied to the standby hydraulic pump.
[0083] After the standby hydraulic pump is engaged to supply pressure or performs a takeover operation, the degradation characteristics of the standby hydraulic pump are updated based on the actual duration of its participation in pressure supply, pressure data, and oil temperature. The operation record database stores the current primary hydraulic pump's identity and the corresponding degradation status of the main and standby hydraulic pumps.
[0084] When generating the overall degradation state of the hydraulic system, the volumetric efficiency degradation state of the hydraulic pumps involved in the fusion is determined based on the actual load conditions of the two pumps. By distinguishing the actual operating loads of the two hydraulic pumps, the system avoids the backup hydraulic pump from exhibiting the same degradation trajectory as the main hydraulic pump when it is in a long-term non-load state.
[0085] When performance test data of the main hydraulic pump is obtained during maintenance, the corresponding data is used to update the main hydraulic pump degradation model; if the corresponding calibration data of the standby hydraulic pump is not obtained, the parameters of the standby hydraulic pump degradation model are not changed. After the hydraulic oil is replaced, the cumulative working time of the hydraulic oil is recalculated, and the existing degradation state of the control valves and cylinder seals is continuously retained according to the actual operation process.
[0086] Example 3: In this example, an edge computing node is set up on the flight simulator side, and a remote prediction platform is set up to communicate with multiple flight simulators. The edge computing node of each flight simulator is connected to the data acquisition unit corresponding to the pressure sensor, temperature sensor, flow sensor and displacement sensor, and performs timestamp verification, signal denoising, abnormal data point identification, time synchronization and extraction of statistical features of operating parameters locally.
[0087] Edge computing nodes further complete the calculation of four types of degradation state quantities and online prediction of remaining useful life. When the network connection is normal, the edge computing node sends the device identity, training subject identifier, statistical characteristics of operating parameters, four types of degradation state quantities, comprehensive degradation state quantity of hydraulic system, predicted value of remaining useful life, and current model version to the remote prediction platform.
[0088] The remote prediction platform establishes operational files for each device, and the data for different flight simulators retains device identity, model configuration, and model version information. When the historical states of flight simulators of the same type are used for degradation model training, the current degradation state of each flight simulator is still continuously formed according to the corresponding device's operational file, avoiding the connection of different device states into a single degradation time series.
[0089] After acquiring new performance calibration data, the remote prediction platform updates the corresponding degradation model and evaluates the model update results using independent validation data. New model parameters that meet the model acceptance criteria are compiled into a complete parameter package, which includes the model version and corresponding configuration identifier, and then sent to the corresponding flight simulator edge computing node.
[0090] After the edge computing node completes parameter integrity verification, it activates the new model version. In the event of network interruption or incomplete parameter packages, it continues to use the most recently verified model version. The updated model recalculates the component degradation state and continues the process of predicting the overall degradation state formation and remaining service life of the hydraulic system.
[0091] Comparative Example 1: Comparative Example 1 uses the same pressure, temperature, flow rate, and displacement inputs as Example 1, as well as the same synchronous preprocessing process. Only the process of constructing the four types of degradation states of hydraulic pump, control valve, cylinder seal, and hydraulic oil separately is cancelled. The pressure statistics, temperature statistics, flow rate statistics, and displacement statistics are normalized and directly formed into a single comprehensive state. The single comprehensive state sequence is then input into the time-series lifetime prediction model.
[0092] When using a single integrated state, different physical degradation sources are not separately mapped before time-series lifetime prediction. For example, when pressure changes are simultaneously affected by the hydraulic pump output state and control valve action, the single integrated state cannot retain the state information of the two degradation sources separately. Example 1 generates four types of degradation states and then performs system-level fusion to ensure that the time-series lifetime prediction input maintains the correspondence between different component degradation sources.
[0093] Comparative Example 2: Comparative Example 2 adopts the same multi-source operating parameter acquisition, synchronous preprocessing, four types of degradation state generation, same-direction normalization, comprehensive degradation state formation of hydraulic system and time-series life prediction process as Example 1, only canceling the incremental update function of the operating archive database on the previous degradation model.
[0094] Once deployed, the model always uses the initial degraded model parameters. The statistical characteristics of new operating parameters and maintenance calibration information generated by subsequent training subjects are only saved and not used for model updates. When the hydraulic motion system of the flight simulator has been running for a long time, the condition of components changes, or the hydraulic oil is replaced, the deviation between the data distribution corresponding to the initial model and the current state cannot be corrected by subsequent calibration data.
[0095] Example 1 updates the previous degradation model by using new calibration data in the running archive. The updated model continues to generate four new degradation state quantities and re-enters the comprehensive degradation state sequence and remaining service life prediction process of the hydraulic system, thereby maintaining the continuous data relationship between component state generation, system degradation fusion, service life prediction and model update.
[0096] Furthermore, for example, the normalized values of the volumetric efficiency degradation state of the hydraulic pump corresponding to six consecutive state moments are 0.31, 0.32, 0.33, 0.34, 0.35, and 0.36, respectively; the normalized values of the wear degradation state of the control valve core are 0.24, 0.25, 0.26, 0.27, 0.28, and 0.29, respectively; the normalized values of the wear degradation state of the cylinder seals are 0.46, 0.47, 0.49, 0.50, 0.52, and 0.53, respectively; and the normalized values of the viscosity aging degradation state of the hydraulic oil are 0.49, 0.50, 0.51, 0.53, 0.55, and 0.56, respectively.
[0097] The comprehensive degradation state quantity of the hydraulic system corresponding to the first state time is 0.3645, the comprehensive degradation state quantity of the hydraulic system corresponding to the second state time is 0.3745, the comprehensive degradation state quantity of the hydraulic system corresponding to the third state time is 0.3875, the comprehensive degradation state quantity of the hydraulic system corresponding to the fourth state time is 0.3990, the comprehensive degradation state quantity of the hydraulic system corresponding to the fifth state time is 0.4135, and the comprehensive degradation state quantity of the hydraulic system corresponding to the sixth state time is 0.4235, thus forming a sequence of comprehensive degradation state quantities of the hydraulic system arranged in chronological order.
[0098] After obtaining new component performance calibration information in subsequent states, incremental updates are performed on the corresponding degradation model. When the normalized value of the hydraulic pump volumetric efficiency degradation state quantity at the next updated state is 0.37, the normalized value of the control valve spool wear degradation state quantity is 0.30, the normalized value of the cylinder seal wear degradation state quantity is 0.55, and the normalized value of the hydraulic oil viscosity aging degradation state quantity is 0.58, the comprehensive degradation state quantity of the hydraulic system is 0.4380. The updated comprehensive degradation state quantity of the hydraulic system, together with the previous state, is entered into the time-series life prediction model, and the corresponding prediction results and model version are continued to be written into the runtime database.
[0099] Through the above process, the multi-source operating parameter data are synchronously preprocessed to form four types of degradation states with clear physical meanings. The four types of degradation states form a unified hydraulic system comprehensive degradation state quantity sequence. The hydraulic system comprehensive degradation state quantity sequence is used to predict the remaining service life. The subsequent performance calibration data further updates the parameters of the previous degradation model, so that the updated degradation state re-enters the subsequent service life prediction process, forming a complete data flow closed loop.
[0100] Example 4: Based on the multi-mechanism degradation state generation and system-level degradation state fusion described in Example 1, this example introduces a hydraulic viscosity-temperature coupling evolution compensation mechanism and a dynamic adaptive weight allocation strategy in the generation and fusion stages of the hydraulic pump volumetric efficiency degradation state quantity, control valve core wear degradation state quantity, cylinder seal wear degradation state quantity and hydraulic oil viscosity aging degradation state quantity.
[0101] As the hydraulic motion system of a flight simulator enters the mid-to-late stages of service, the degradation rates of different components may vary. When a component enters an accelerated degradation phase, its impact on the overall lifespan of the hydraulic motion system increases. If fixed component contribution weights are continued, the contribution ratio of each component in the overall system degradation state remains unchanged, making it difficult to reflect changes in the degradation rate of local components in a timely manner. Simultaneously, changes in the physical properties of hydraulic oil can affect the contact area of cylinder seals and the internal clearance of the hydraulic pump. A decrease in hydraulic oil viscosity alters the lubrication state of the sealing contact area and changes the internal leakage state of the hydraulic pump. Therefore, this embodiment establishes a cross-model compensation relationship for hydraulic oil viscosity changes among four degradation models and dynamically adjusts the system-level fusion weights based on the local change rate of each degradation state.
[0102] The hardware platform implementing the hydraulic viscosity-temperature coupling evolution compensation mechanism maintains a multi-source operating parameter data acquisition channel consisting of pressure sensors, temperature sensors, flow sensors, and displacement sensors. After the multi-factor degradation modeling module receives the synchronously preprocessed operating parameter data, the hydraulic oil viscosity aging degradation model first calculates the hydraulic oil state at the current moment. Based on the real-time oil temperature and historical cumulative working time, the hydraulic oil viscosity aging degradation model outputs the hydraulic oil viscosity aging degradation state quantity corresponding to the current moment.
[0103] A calibration mapping relationship is established between the periodic viscosity test results of hydraulic oil and the viscosity aging degradation state of the hydraulic oil at the corresponding state time. Based on this calibration mapping relationship, the dynamic viscosity of the hydraulic oil at the current state time is determined, and the physical viscosity decay coefficient is further obtained. The physical viscosity decay coefficient satisfies the following relationship:
[0104] ;
[0105] in, Current state time The corresponding hydraulic oil dynamic viscosity, in Pa·s; The initial standard dynamic viscosity of the hydraulic oil is expressed in Pa·s. Current state time The corresponding physical viscosity decay coefficient is a dimensionless quantity, with a value ranging from 0 to 1. This is the time parameter corresponding to the current state, in seconds.
[0106] when An increase indicates an increase in the degree of degradation of the current hydraulic oil dynamic viscosity relative to its initial standard dynamic viscosity. The hydraulic oil viscosity aging degradation model thus outputs not only the hydraulic oil viscosity aging degradation state quantity for system-level fusion, but also, through calibration mapping relationships, outputs the physical viscosity degradation coefficient to the cylinder seal wear degradation model and the hydraulic pump volumetric efficiency degradation model.
[0107] The physical viscosity decay coefficient is used as a dynamic boundary condition input into the wear degradation model of the hydraulic cylinder seal. During the continuous reciprocating motion sequence of the hydraulic cylinder, the contact area between the piston rod and the seal is lubricated by hydraulic oil. This embodiment further introduces the hydrodynamic lubrication film thickness variation relationship into the original hydraulic cylinder seal wear calculation logic, determining the lubrication film thickness based on the instantaneous piston speed, hydraulic oil dynamic viscosity, and current load pressure.
[0108] The thickness of the hydrodynamic lubrication film is determined by the following relationship:
[0109] ;
[0110] in, Current state time The corresponding hydrodynamic lubrication film thickness, in meters (m); The comprehensive geometric material coefficient is obtained by calibration based on the surface roughness of the inner wall of the cylinder barrel and the elastic modulus of the sealing material. Its dimension is the 1 / 2 power of length, that is, the unit is m^0.5. The instantaneous velocity of the cylinder piston is calculated by differentially based on adjacent sampled values from the displacement sensor, and the unit is m / s; The definition is the same as the previous definition; The definition is the same as the previous definition; Current state time The corresponding hydraulic cylinder load pressure, in Pa; The definition is the same as the previous definition.
[0111] because The unit is Pa·s. The unit is m / s. The unit is Pa, therefore the dimension of the calculation result within the parentheses is length; the result within the parentheses is raised to the power of 0.5, resulting in a dimension of 1 / 2 power of length, which is then compared with the result whose dimension is also 1 / 2 power of length. Multiplying them together gives the result in units of m. The dimensions on both sides of the formula remain consistent.
[0112] Hydraulic cylinder seal wear degradation model based on The change range dynamically adjusts the original temperature correction amount. When the physical viscosity decay coefficient increases and leads to a decrease in the thickness of the hydrodynamic lubrication film, the wear contribution of the seals in the corresponding action window is increased; the local temperature rise generated by friction continues to be fed back to the cumulative wear base, so that the wear degradation state of the cylinder seals simultaneously reflects the coupling effect of the number of reciprocating motions, the peak load pressure, the oil temperature, and the change in hydraulic oil viscosity.
[0113] The hydraulic pump volumetric efficiency degradation model synchronously receives the physical viscosity decay coefficient and constructs an internal leakage amplification factor based on the physical viscosity decay coefficient. When the physical viscosity decay coefficient is 0, the internal leakage amplification factor adopts the baseline value corresponding to the initial viscosity state; as the physical viscosity decay coefficient increases, the internal leakage amplification factor increases according to the mapping relationship corresponding to the hydraulic pump performance calibration data, which is used to represent the impact of the decrease in hydraulic oil dynamic viscosity on the internal clearance leakage state of the hydraulic pump.
[0114] When processing the three input sequences—cumulative working time, number of pressure cycles, and oil temperature integral—the nonlinear autoregressive neural network forms a hidden layer feature matrix based on the network's forward propagation results. The internal leakage amplification factor is then expanded according to the number of rows and columns of the hidden layer feature matrix, forming an internal leakage compensation tensor of the same dimension as the hidden layer feature matrix. Corresponding elements in the internal leakage compensation tensor are filled with the internal leakage amplification factor at the current state, enabling element-wise multiplication between the internal leakage compensation tensor and the hidden layer feature matrix.
[0115] Perform the Hadamard product on the hidden layer feature matrix and the internal leakage compensation tensor. The operational relationship is as follows:
[0116] ;
[0117] in, For the nonlinear autoregressive neural network, the first The hidden layer feature matrix after Hadamard product compensation; For the nonlinear autoregressive neural network, the first The hidden layer feature matrix formed by the original forward propagation of the layer; This represents the hidden layer number in a nonlinear autoregressive neural network. This represents the Hadamard product operation between elements at the same position in a matrix. Current state time The corresponding internal leakage compensation tensor, matrix dimension and Completely identical; The definition is the same as the previous definition.
[0118] The input degradation features of a nonlinear autoregressive neural network have been normalized before entering the network, and the hidden layers use dimensionless neuron activation values. A dimensionless matrix; internal leakage compensation tensor Composed of dimensionless internal leakage amplification factors, it is therefore also a dimensionless matrix; obtained through the Hadamard product. It remains a dimensionless matrix, and the dimensions remain consistent before and after the operation.
[0119] After compensation Continuing into the nonlinear autoregressive neural network's subsequent activation function and output layer, the hydraulic pump volumetric efficiency degradation state quantity after hydraulic viscosity-temperature coupling evolution compensation is obtained. Thus, the output of the hydraulic oil viscosity aging degradation model simultaneously acts on the cylinder seal wear degradation model and the hydraulic pump volumetric efficiency degradation model through the physical viscosity decay coefficient, forming a cross-component coupled processing chain with data input-output relationships among the three degradation models.
[0120] After acquiring the four types of degradation state variables after hydraulic viscosity-temperature coupling evolution compensation, the degradation state fusion module initiates a dynamic adaptive weight allocation strategy. The processor sets a sliding time window in its local storage space, with a window length of 200 consecutive state moments. The degradation state variables of hydraulic pump volumetric efficiency, control valve spool wear, cylinder seal wear, and hydraulic oil viscosity aging are arranged in chronological order to construct a 4-row, 200-column state evolution matrix.
[0121] Each row of the state evolution matrix corresponds to a degenerate state variable, and each column corresponds to a state moment within a sliding time window. As a new state moment is generated, the oldest state moment is removed from the current sliding time window and the latest state moment is added, ensuring that the state evolution matrix maintains a 4-row, 200-column matrix structure.
[0122] When calculating the local degradation gradient of various degradation state variables, the rate of change of degradation state between adjacent state moments is calculated using first-order finite difference along the state-time axis. The calculation relationship is as follows:
[0123] ;
[0124] in, For the first Degenerate state variables at the current state time The corresponding local degradation gradient, in units of s⁻¹; For the first Degenerate state variables at the current state time The corresponding normalized degenerate state value is a dimensionless quantity; For the first The normalized degenerate state value corresponding to the previous adjacent state time of the degenerate state quantity is a dimensionless quantity; The time interval between two adjacent state moments, in seconds; The value is the serial number of the degradation state category, which can be 1, 2, 3 or 4, corresponding to the degradation state of hydraulic pump volumetric efficiency, degradation state of control valve core wear, degradation state of cylinder seal wear, and degradation state of hydraulic oil viscosity aging, respectively. The definition is the same as the previous definition.
[0125] because and All are dimensionless quantities. The unit is s, therefore The unit is s⁻¹. Within the sliding time window, the continuously obtained local degradation gradients can be averaged, or the current local degradation gradient at the end of the window can be used as the input to the logistic nonlinear activation function. In this embodiment, the local degradation gradient corresponding to the current state time is used.
[0126] The local degradation gradient is used to reflect the rate of change of the degradation state of different components at the current stage. When the local degradation gradient remains small, the corresponding component is in a relatively slow degradation stage; when the local degradation gradient increases, the normalized degradation state of the corresponding component changes faster per unit time, and the dynamic adaptive weight allocation strategy increases the influence of the corresponding degradation state quantity in the system-level fusion process.
[0127] Establish a logistic nonlinear activation function with local degenerate gradients as input, and compute the th... The current degradation acceleration penalty factor corresponding to the class degradation state variable:
[0128] ;
[0129] in, For the first Degenerate state variables at the current state time The corresponding degradation acceleration penalty factor is a dimensionless quantity; The definition is the same as the previous definition; The definition is the same as the previous definition; The upper limit coefficient for the penalty is a dimensionless quantity, and in this embodiment it is set to 1.5; Let be the shape adjustment parameter of the logistic nonlinear activation function, when and When represented by s⁻¹, The unit is s, so that the exponential term remains dimensionless; The definition is the same as the previous definition; The gradient warning threshold is calculated based on the rate of change of various degradation states in the historical operation archive database, with the unit being s⁻¹; is the base of the natural constant.
[0130] because and Having the same unit, the difference between the two is in units of s⁻¹, and the unit is s. Multiplying them together yields a dimensionless number; therefore, the independent variable of the exponential function is a dimensionless quantity. Since it is a dimensionless quantity, therefore It is a dimensionless quantity.
[0131] In this embodiment Setting it to 1.5 gives the degradation acceleration penalty factor a defined upper bound for regulation. The steepness of the curve of the control logic stith function as it transitions from the low-penalty region to the high-penalty region. Used to determine the location where the local degradation gradient enters the accelerated degradation regulation region.
[0132] After obtaining the degradation acceleration penalty factors corresponding to the four types of degradation state variables, the degradation state fusion module fuses the degradation acceleration penalty factors with the static benchmark weights. The static benchmark weights are maintained at 0.30 for hydraulic pumps, 0.25 for control valves, 0.30 for cylinder seals, and 0.15 for hydraulic oil. The dynamic fusion weights are calculated by multiplying each item and then normalizing the result.
[0133] ;
[0134] in, For the first Degenerate state variables at the current state time The corresponding dynamic fusion weights are dimensionless. For the first The static baseline weights corresponding to the class degenerate state variables are dimensionless. The definition is the same as the previous definition; This is the summation index used in the dynamic fusion weight normalization calculation, with a value ranging from 1 to 4; For the first The static baseline weights corresponding to the class degenerate state variables are dimensionless. For the first Degenerate state variables at the current state time The corresponding degradation acceleration penalty factor is a dimensionless quantity; and The definition is the same as the previous definition.
[0135] because , , and All of them are dimensionless quantities, therefore It is a dimensionless quantity. After the four types of dynamic fusion weights are normalized, their sum is 1, so that the comprehensive degradation state quantity of the hydraulic system after the dynamic weights replace the static weights still maintains a unified state scale of 0 to 1.
[0136] The degradation state fusion module multiplies the four types of normalized degradation state quantities by their corresponding dynamic fusion weights and then sums them to obtain the comprehensive degradation state quantity of the hydraulic system at the current state moment. The comprehensive degradation state quantities of the hydraulic system obtained at consecutive state moments are arranged in chronological order to form a sequence of comprehensive degradation state quantities of the hydraulic system, which is then input into the time-series life prediction model.
[0137] When the local degradation gradients corresponding to the four types of degradation state variables are all within a near-stable range, the differences between the degradation acceleration penalty factors are small, and the dynamic fusion weights maintain a proportional relationship close to the static baseline weights. When the local degradation gradient of a certain type of degradation state variable increases and exceeds the gradient warning threshold, the corresponding degradation acceleration penalty factor increases, causing the corresponding dynamic fusion weights to increase synchronously, thereby enhancing the system's overall degradation state response to local accelerated degradation.
[0138] The hydraulic viscosity-temperature coupled evolution compensation mechanism and the dynamic adaptive weight allocation strategy operate continuously on the data stream. The hydraulic oil viscosity aging degradation model first outputs the hydraulic oil viscosity aging degradation state quantity and physical viscosity decay coefficient; the physical viscosity decay coefficient enters the cylinder seal wear degradation model and the hydraulic pump volumetric efficiency degradation model to form the component degradation state after coupled compensation; the four types of degradation states enter the state evolution matrix to calculate the local degradation gradient; the local degradation gradient is converted into a degradation acceleration penalty factor through the logistic nonlinear activation function; the degradation acceleration penalty factor and the static benchmark weights together form the dynamic fusion weights; the dynamic fusion weights are used to form the comprehensive degradation state quantity sequence of the hydraulic system; the comprehensive degradation state quantity sequence of the hydraulic system is input into the long short-term memory recurrent neural network to obtain the remaining service life prediction value, thus forming a complete data dependency relationship.
[0139] In this embodiment, a six-degree-of-freedom flight simulator with a dual-pump redundant architecture was used to conduct accelerated degradation tests. The test training included sustained high-G turns, high-frequency pitch oscillations, and motion platform actions corresponding to emergency decompression landings, accumulating a total runtime of 3500 hours and generating 52,000 action sample cycles. During the test, the remaining service life prediction values output by the long short-term memory recurrent neural network and the service life baseline values formed from laboratory test results were recorded simultaneously. The mean absolute percentage error (MASE) and root mean square error (RMSE) were used to evaluate the remaining service life prediction deviations under different model configurations.
[0140] Table 1 records the prediction errors of remaining service life at different operating time points for the processing methods using static weights and independent degradation models and the processing methods using dynamic weights and hydraulic viscosity-temperature coupling compensation models.
[0141] Table 1 Performance Comparison Table;
[0142] Testing Phase and Method Configuration Runtime Node Mean Absolute Percentage Error (MAPE) Root Mean Square Error (RMSE) / h Original static weights and independent degradation model 1000h 4.12% 45.3 Original static weights and independent degradation model 2500h 8.76% 112.5 Original static weights and independent degradation model 3200h 15.43% 205.8 This embodiment uses a dynamic weighting and coupling compensation model. 1000h 3.85% 41.2 This embodiment uses a dynamic weighting and coupling compensation model. 2500h 4.22% 52.6 This embodiment uses a dynamic weighting and coupling compensation model. 3200h 5.17% 68.4
[0143] At the 1000-hour operating time point, all components are in a relatively stable degradation stage, and the physical state of the hydraulic oil changes little. Both processing methods maintain low prediction errors. When using the static weight and independent degradation model, the mean absolute percentage error is 4.12%, and the root mean square error is 45.3h; when using the dynamic weight and coupled compensation model, the mean absolute percentage error is 3.85%, and the root mean square error is 41.2h.
[0144] After 2500 hours of operation, the cumulative number of reciprocating motions of the cylinder increased, and the hydraulic oil exhibited more significant viscosity changes under high-pressure shearing and temperature effects. When using the static weighting and independent degradation model, the hydraulic oil viscosity change was only treated as an independent oil degradation state and entered into a fixed-proportion fusion. There was no direct compensation model for the wear degradation of the cylinder seals and the volumetric efficiency degradation model of the hydraulic pump. The overall degradation state of the system responded less to the state of local accelerated degradation, corresponding to an increase in the mean absolute percentage error to 8.76% and a root mean square error to 112.5h.
[0145] After adopting the hydraulic viscosity-temperature coupling evolution compensation mechanism of this embodiment, the physical viscosity decay coefficient is incorporated into the cylinder seal wear degradation model to calculate the change in hydrodynamic lubrication film thickness and correct the seal wear contribution based on the change in lubrication film thickness. Simultaneously, the physical viscosity decay coefficient is applied to the hidden layer feature matrix of the nonlinear autoregressive neural network through the internal leakage compensation tensor, ensuring that the hydraulic pump volumetric efficiency degradation state includes the internal leakage effect corresponding to the hydraulic oil viscosity change. At the 2500-hour operating time point, the mean absolute percentage error corresponding to this embodiment is 4.22%, and the root mean square error is 52.6 hours.
[0146] After 3200 hours of operation, the local degradation gradient corresponding to the cylinder seal increased. A dynamic adaptive weight allocation strategy calculates the local degradation gradient based on the state evolution matrix and increases the degradation acceleration penalty factor corresponding to the cylinder seal using a logistic nonlinear activation function. After dynamic fusion weight normalization, the dynamic fusion weight corresponding to the cylinder seal is adjusted from the static baseline weight of 0.30 to 0.48, thereby increasing the contribution ratio of the accelerated degradation state of the cylinder seal to the overall degradation state of the hydraulic system.
[0147] At the 3200-hour operating time point, the mean absolute percentage error was 15.43% and the root mean square error was 205.8h when using the static weight and independent degradation model; the mean absolute percentage error was 5.17% and the root mean square error was 68.4h when using the dynamic weight and coupled compensation model. Therefore, the dynamic adaptive weight allocation strategy can change the fusion ratio of different degradation states according to the local degradation gradient, and the hydraulic viscosity-temperature coupled evolution compensation mechanism can propagate the changes in hydraulic oil viscosity to the degradation states of cylinder seals and hydraulic pumps. These two mechanisms work together to form the comprehensive degradation state sequence of the hydraulic system.
[0148] The physical viscosity decay coefficient, hydrodynamic lubrication film thickness, internal leakage amplification factor, local degradation gradient, degradation acceleration penalty factor, dynamic fusion weight, comprehensive degradation state quantity of the hydraulic system, and remaining service life prediction value obtained at each state moment are all written into the operation archive database according to the corresponding equipment identity, training subject identifier, state moment, and model version. After receiving the operation archive data, the remote prediction platform saves each intermediate state according to the corresponding model version, so that the dynamic weight and hydraulic viscosity-temperature coupling compensation process can be traced back with historical states, and serves as the data basis for subsequent multi-factor degradation model parameter updates.
[0149] Through the above implementation process, the output of the hydraulic oil viscosity aging degradation model no longer participates in the final fusion as an independent degradation state, but further influences the cylinder seal wear degradation model and the hydraulic pump volumetric efficiency degradation model. The four types of degradation states are no longer fused at the system level according to fixed baseline weights, but the dynamic fusion weights corresponding to the current state are adjusted according to the local degradation gradient of each type of degradation state. The hydraulic viscosity-temperature coupling evolution compensation mechanism and the dynamic adaptive weight allocation strategy thus act on the component degradation state generation stage and the system-level degradation state fusion stage, respectively, and jointly affect the subsequent remaining service life prediction.
[0150] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.
[0151] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for predicting the lifespan of a hydraulic system in a flight simulator, characterized in that, include: The hydraulic motion system of the flight simulator collects oil pressure, oil temperature, flow rate, and actuator displacement during the action sequence of training subjects. Noise reduction and time synchronization are performed to obtain synchronized preprocessed operating parameter data. The degradation features corresponding to the hydraulic pump, control valve, cylinder seals and hydraulic oil are formed from the synchronous preprocessing operation parameter data, and then input into the corresponding degradation model to obtain four types of degradation state quantities; The four types of degradation state quantities are normalized in the same direction and fused according to the component contribution relationship to obtain the comprehensive degradation state quantity sequence of the hydraulic system; The sequence of comprehensive degradation state quantities of the hydraulic system is input into the time-series life prediction model to obtain the predicted value of the remaining service life, and the predicted value of the remaining service life and the corresponding degradation state quantities are written into the operation archive database. The parameters of the multi-factor degradation model are updated based on the statistical characteristics of the operating parameters and the degradation status assessment results generated by subsequent training subjects. The updated four types of degradation status quantities are then used for subsequent prediction of remaining useful life.
2. The method for predicting the lifespan of a flight simulator hydraulic system according to claim 1, characterized in that, The degradation characteristics of the hydraulic pump include cumulative working time, number of pressure cycles, and oil temperature integral. The hydraulic pump volumetric efficiency degradation model uses a nonlinear autoregressive neural network to establish the mapping relationship between the hydraulic pump volumetric efficiency degradation state quantity and the cumulative working time, number of pressure cycles, and oil temperature integral.
3. The method for predicting the lifespan of a flight simulator hydraulic system according to claim 1, characterized in that, The degradation characteristics of the control valve include the valve core switching frequency and the pressure difference between the inlet and outlet of the control valve. The valve core wear degradation model of the control valve adopts the support vector regression model to establish the mapping relationship between the valve core wear degradation state quantity and the valve core switching frequency and the pressure difference between the inlet and outlet of the control valve.
4. The method for predicting the lifespan of a flight simulator hydraulic system according to claim 1, characterized in that, The degradation characteristics of the cylinder seals include the number of cylinder reciprocating motions, the peak load pressure, and the oil temperature. The cylinder seal wear degradation model determines the wear degradation state of the cylinder seals based on the cumulative effect of the number of cylinder reciprocating motions, the load effect of the peak load pressure, and the temperature correction amount of the oil temperature. The degradation characteristics of the hydraulic oil include the cumulative working time and the oil temperature. The hydraulic oil viscosity aging degradation model determines the hydraulic oil viscosity aging degradation state based on the temperature-related aging rate and the cumulative working time.
5. The method for predicting the lifespan of a flight simulator hydraulic system according to claim 1, characterized in that, The denoising and time synchronization include: using 5-layer db8 wavelet decomposition and soft thresholding to reduce high-frequency noise in multi-source operating parameter data; using adjacent normal data points to linearly interpolate data points that exceed the mean ± 3 times the standard deviation of the corresponding signal sliding statistical window and are determined to be isolated anomalies; and aligning multi-source operating parameter data with different sampling rates to a 500Hz reference time axis according to the timestamp.
6. The method for predicting the lifespan of a flight simulator hydraulic system according to claim 1, characterized in that, The normalization in the same direction includes converting the direction according to the change direction between different degradation state quantities and degradation degrees, so that an increase in the normalized value corresponds to a deeper degradation degree; the component contribution relationship includes the hydraulic pump volumetric efficiency degradation state quantity, the control valve core wear degradation state quantity, the cylinder seal wear degradation state quantity, and the hydraulic oil viscosity aging degradation state quantity, which correspond to fusion weights of 30%, 25%, 30%, and 15%, respectively.
7. The method for predicting the lifespan of a flight simulator hydraulic system according to claim 1, characterized in that, The time-series lifetime prediction model employs a long short-term memory recurrent neural network. The long short-term memory recurrent neural network receives a continuous sequence of comprehensive degradation state variables of the hydraulic system, updates the hidden state and cell state through input gates, forget gates, and output gates, and outputs the remaining lifetime prediction value by a linear output layer based on the updated hidden state.
8. The method for predicting the lifespan of a flight simulator hydraulic system according to claim 1, characterized in that, The operation archive database stores statistical characteristics of multi-source operation parameter data, four types of degradation state quantities, comprehensive degradation state quantities of hydraulic system, predicted value of remaining service life, and model version according to training subjects. After the maintenance cycle is reached or the conditions for updating new training samples are met, new samples with corresponding calibration information are extracted from the operation archive database, incremental model updates are performed on at least one degradation model, and it is determined whether to adopt the updated model parameters based on the error changes of independent verification data.
9. The method for predicting the lifespan of a flight simulator hydraulic system according to claim 1, characterized in that, Maintenance warning information is generated based on the comparison between the predicted remaining service life and the maintenance warning threshold. The maintenance warning information includes four types of degradation status, the comprehensive degradation status of the hydraulic system, the predicted remaining service life, and priority inspection objects determined based on the four types of degradation status.
10. The method for predicting the lifespan of a flight simulator hydraulic system according to claim 1, characterized in that, In the step of normalizing the four types of degradation state variables in the same direction and fusing them according to the component contribution relationship to obtain the comprehensive degradation state variable sequence of the hydraulic system, a dynamic adaptive weight allocation strategy is adopted to determine the component contribution relationship. The specific process includes: Construct a state evolution matrix that includes the volumetric efficiency degradation state of the hydraulic pump, the wear degradation state of the control valve core, the wear degradation state of the cylinder seals, and the aging degradation state of the hydraulic oil viscosity. Calculate the local degradation gradient of each type of degenerate state variable in the state evolution matrix within a preset time window; A logistic nonlinear activation function with the local degradation gradient as the independent variable is established, and the degradation acceleration penalty factor corresponding to various degradation state variables is calculated. The preset static benchmark weights are multiplied by the degradation acceleration penalty factor and normalized to obtain the dynamic fusion weights corresponding to various degradation state quantities at the current state time. The dynamic fusion weights are then used to obtain the comprehensive degradation state quantity sequence of the hydraulic system.
