A method and system for signal analysis for aircraft hydraulic power system failure prediction

Abstract

The application provides a signal analysis aircraft hydraulic power system fault prediction method and system, comprising: obtaining actuating instructions and sensing signals under working conditions, generating a time alignment state matrix through time axis alignment and time domain superposition averaging; inputting a fluid pipe wall coupling physical model containing adjustable physical parameters to solve reversely, reconstructing three-dimensional flow field state parameters of pipe microelements; calculating total entropy production rate of the microelements and performing physical cause orthogonal decomposition to extract entropy production component time sequence characteristics; comparing with baseline template quantization of a benchmark state characteristic database to generate a characteristic residual vector; inputting a parameter optimization model based on a gradient descent method to update adjustable parameters, obtaining identified pipe wall equivalent roughness and local convective heat transfer coefficient; and obtaining a prediction result through cross comparison with a historical measured atlas experience library. The application solves the problems that existing aircraft hydraulic pipeline monitoring is difficult to capture early weak degradation characteristics online, and is difficult to effectively distinguish underlying physical damage mechanisms.

Description

Technical Field

[0001] This invention belongs to the technical field of aircraft condition monitoring and fault prediction, and relates to a fault prediction method and system for aircraft hydraulic power systems based on signal analysis. Background Technology

[0002] The hydraulic power system in aircraft operation is the power source for critical functions such as flight control, and the structural integrity and fluid transport performance of its piping system profoundly affect flight safety. With the accumulation of service time, hydraulic pipelines are subjected to the combined effects of physical factors such as high-pressure fluid erosion, temperature cycling, and vibration loads, gradually leading to progressive degradation phenomena such as increased internal wall roughness and material fatigue. This microscopic performance degradation is not easily visually apparent through macroscopic parameters such as system average pressure or flow rate in the early stages, but its continuous evolution affects system efficiency and, under certain operating conditions, poses the risk of pipeline leakage or rupture.

[0003] For health monitoring of hydraulic pipeline systems, the commonly used technical solutions in this field mainly include two forms: periodic offline non-destructive testing and online alarm based on operating parameter thresholds. Periodic offline testing technology often utilizes physical testing methods such as ultrasonic or eddy current testing to obtain macroscopic defect information of the pipe wall. The online alarm method based on operating parameter thresholds, on the other hand, collects macroscopic operating data such as system return oil pressure and pump outlet flow in real time, compares these sampled data with preset safety benchmark values, and determines the macroscopic operating status of the hydraulic system accordingly.

[0004] The aforementioned existing technologies have room for improvement in the early prediction of the gradual and covert degradation process of hydraulic pipelines. Specifically, the execution cycle of periodic offline non-destructive testing is relatively long and often requires equipment downtime, making it difficult to continuously track the dynamic characteristics of pipeline degradation during the intervals of such static testing. Furthermore, online alarm logic based on macroscopic parameter thresholds is highly dependent on the salience of fault characteristics; early, weak degradation signals are often masked by normal system fluctuations and high-frequency noise, limiting the sensitivity of early warnings. In addition, the aforementioned techniques relying on macroscopic numerical characterization or discontinuous physical flaw detection face difficulties in distinguishing pipeline damage caused by different underlying physical factors such as mechanical wear or thermal fatigue, making it difficult to provide accurate root cause analysis for subsequent maintenance decisions. Summary of the Invention

[0005] In a first aspect, the present invention provides a method for predicting faults in an aircraft hydraulic power system using signal analysis, comprising the following steps:

[0006] The actuation command signal and sensor signal under the actuation condition are acquired. The sensor signal is dynamically aligned on the time axis to eliminate the time deviation of each cycle. The processed signal is then superimposed and averaged in the time domain to generate a time-aligned state matrix.

[0007] The time-aligned state matrix is ​​input into the preset fluid pipe wall coupled physical model for inverse calculation to reconstruct the three-dimensional flow field state parameters of the target pipe micro-element. The fluid pipe wall coupled physical model includes adjustable physical parameters.

[0008] The total entropy production rate of the flow field micro-element is calculated based on the three-dimensional flow field state parameters, and the physical genesis orthogonal decomposition operation is performed to extract the temporal characteristics of the entropy production component.

[0009] Based on the current operating conditions, retrieve the baseline feature template matched in the preset baseline state feature database, and perform multi-dimensional quantization comparison between the baseline feature template and the temporal features of the entropy production component to generate a feature residual vector.

[0010] Input the feature residual vector into the preset parameter optimization model based on gradient descent, update the adjustable physical parameters until the preset convergence tolerance threshold is met, and obtain the convergent physical parameters.

[0011] The convergent physical parameters are cross-referenced with a pre-set historical measured component map experience database to obtain the pipeline condition prediction results.

[0012] A further aspect of the present invention involves dynamically aligning the sensing signal along its time axis to eliminate duration deviations in each cycle, and then performing time-domain superposition and averaging of the processed signals to generate a time-aligned state matrix, comprising the following steps:

[0013] Receive actuation command signals as timing trigger references;

[0014] Within a preset time window, the fluctuation signals output by the high-frequency pressure sensor array and temperature sensor array arranged on the target pipeline are captured as sensing signals.

[0015] Based on the timing triggering benchmark, the multi-cycle fluctuation signals within the preset time window are dynamically time-warped and matched, and the time axis is aligned with the fluctuation signal of a reference cycle to eliminate the phase shift caused by the inconsistent duration of each cycle; then, all the fluctuation signals after time axis alignment are time-domain superimposed and averaged to generate a time-aligned state matrix.

[0016] A further aspect of this invention involves inputting a time-aligned state matrix into a preset fluid-pipe wall coupled physical model for inverse calculation to reconstruct the three-dimensional flow field state parameters of the target pipe element, including the following steps:

[0017] Initialize the fluid-pipeline coupling physical model by obtaining the inherent geometric configuration dimensions of the target pipeline and the initial fluid properties under the current operating conditions;

[0018] The equivalent roughness of the pipe wall and the local convective heat transfer coefficient are defined as adjustable physical parameters.

[0019] The time-aligned state matrix is ​​used as the measured boundary condition and input into the fluid-tube wall coupled physical model to perform inverse calculation of transient force and thermal response.

[0020] Extract the solution distribution results of the fluid pipe wall coupled physical model, and output three-dimensional flow field state parameters including local velocity vector distribution, shear stress distribution and temperature gradient field of micro-element.

[0021] A further aspect of this invention involves calculating the total entropy yield of flow field micro-elements based on three-dimensional flow field state parameters and performing physical causal orthogonal decomposition to extract temporal features of the entropy yield components, including the following steps:

[0022] By integrating a pre-defined numerical model of an irreversible fluid dynamics process and utilizing the extracted three-dimensional flow field state parameters, the total entropy production rate of the corresponding micro-element during its continuous evolution within a pre-defined time window is calculated.

[0023] Orthogonal decomposition and dimensionality reduction extraction were performed on the total entropy production rate to separate the viscous dissipative entropy production component sequence and the thermal conduction entropy production component sequence;

[0024] Temporal envelope curve fitting feature extraction is performed on the viscous dissipation entropy production component sequence and the thermal conduction entropy production component sequence to obtain the peak evolution time sequence node and the relative phase shift angle. The peak evolution time sequence node, the relative phase shift angle and the viscous dissipation entropy production component sequence and the thermal conduction entropy production component sequence are combined to form the entropy production component time sequence feature.

[0025] A further aspect of this invention involves retrieving a baseline feature template from a pre-defined baseline state feature database based on the current operating condition, performing multi-dimensional quantization comparison between the baseline feature template and the entropy production component time-series features, and generating a feature residual vector, including the following steps:

[0026] Based on the current operating condition and the corresponding motor category number, retrieve the corresponding baseline feature template from the baseline state feature database;

[0027] The viscous dissipative entropy production component sequence and the thermal conduction entropy production component sequence contained in the temporal features of the entropy production component are matched with the corresponding reference sequence in the baseline feature template at the same time measure length by dynamic time warping distortion distance matching calculation, and their independent nonlinear deviation time series are measured respectively.

[0028] Each nonlinear deviation time series is normalized and mapped based on a preset scaling factor. Principal component analysis is then performed on each normalized time series to reduce the dimensionality of features and extract the principal component feature vectors. Vector concatenation is then performed on the extracted principal component feature vectors to calculate and generate the feature residual vector.

[0029] A further aspect of this invention involves inputting the feature residual vector into a preset parameter optimization model based on gradient descent, updating the adjustable physical parameters until a preset convergence tolerance threshold is met, and obtaining the convergent physical parameters, including the following steps:

[0030] The feature residual vector is transferred to the parameter optimization model based on gradient descent and transformed into a parameter correction vector;

[0031] The extracted parameter correction vector includes the function of correcting the step size, backtracking into the fluid pipe wall coupled physical model, and superimposing the initial parameter values ​​of the adjustable physical parameters.

[0032] The closed-loop derivation operation, including parameter update process, is triggered repeatedly based on the dynamic model parameters after the modification takes effect, until the L2 value of the feature residual vector is lower than the convergence tolerance threshold, and the locked parameters are derived into convergent physical parameters.

[0033] A further aspect of this invention involves cross-referencing the convergent physical parameters with a pre-defined historical measured component map experience database to obtain pipeline condition prediction results, including the following steps:

[0034] Extract the local roughness increment and local convective heat transfer coefficient drift attenuation of the corresponding values ​​of the convergent physical parameters, and convert them into the evolution increment of the current service state of the system after merging.

[0035] The evolution increment of the current service status of the system is anchored to the disaster type evolution path by performing clustering node pattern similarity calculation in the historical measured component map experience database.

[0036] The data feature clusters of the actual remaining safe operating capacity percentage are calculated by weighted summation based on the built-in service margin attenuation derivation function mapped by the disaster type evolution path, and then encapsulated as pipeline status prediction results.

[0037] A further aspect of the present invention includes, before performing time-domain superposition averaging and time-series alignment processing:

[0038] Acquire fundamental frequency vibration data of the aircraft's hydraulic pump source;

[0039] Configure the notch band parameters of the adaptive comb filter based on the fundamental frequency vibration data;

[0040] The target wave signal is obtained by inputting the wave signal into an adaptive comb filter for harmonic filtering.

[0041] The target fluctuation signal is used as the new fluctuation signal.

[0042] A further aspect of the present invention, after generating the temporal features of the entropy-generating components, includes:

[0043] Extract the reference phase offset angle range under historical normal operating conditions and compare it with the relative phase offset angle contained in the temporal characteristics of the entropy production component;

[0044] If the relative phase offset angle deviates from the reference phase offset angle range by more than the preset outlier tolerance, it is determined that the temporal characteristics of the current entropy production component are contaminated.

[0045] Trigger a resampling command, discard the current time-aligned state matrix, and return to the startup step of "acquiring the actuation command signal and sensor signal under the actuation condition, performing dynamic time axis alignment on the sensor signal to eliminate the duration deviation of each cycle, and performing time-domain superposition and averaging on the processed signal to generate a time-aligned state matrix".

[0046] Secondly, the present invention provides a fault prediction system for aircraft hydraulic power systems based on signal analysis, comprising the following modules:

[0047] The data synchronization acquisition module is used to acquire the actuation command signal and sensor signal under the actuation condition, perform dynamic time axis alignment on the sensor signal to eliminate the duration deviation of each cycle, and perform time domain superposition and averaging on the processed signal to generate a time alignment state matrix.

[0048] The flow field state reconstruction module is used to input the time-aligned state matrix into the preset fluid pipe wall coupled physical model for inverse calculation, and reconstruct the three-dimensional flow field state parameters of the target pipe micro-element. The fluid pipe wall coupled physical model includes adjustable physical parameters.

[0049] The entropy production feature extraction module calculates the total entropy production rate of the flow field micro-element based on the three-dimensional flow field state parameters and performs physical causal orthogonal decomposition operation to extract the temporal features of the entropy production component.

[0050] The feature residual generation module is used to retrieve the baseline feature template matched in the preset baseline state feature database according to the current operating conditions, and compare the baseline feature template with the temporal features of the entropy production component in a multidimensional quantization to generate a feature residual vector.

[0051] The parameter optimization identification module is used to input the feature residual vector into the preset parameter optimization model based on gradient descent, update the adjustable physical parameters until the preset convergence tolerance threshold is met, and obtain the convergent physical parameters.

[0052] The degradation state prediction module is used to cross-compare converged physical parameters with a preset historical measured component map experience database to obtain pipeline state prediction results.

[0053] In summary, the present invention has the following beneficial technical effects:

[0054] This invention suppresses non-condition-related random broadband noise by synchronously acquiring actuation command signals and sensor signals under specific actuation conditions and performing time alignment and time-domain superposition averaging. This technique generates a time-aligned state matrix that effectively preserves the transient fluctuations in fluid pressure and temperature related to specific actuation processes. This provides measured boundary conditions with a certain signal-to-noise ratio for subsequent inverse solving of the physical model, thus improving the accuracy of the physical model in reconstructing the real flow field state.

[0055] This invention employs a fluid-pipe wall coupled physical model incorporating adjustable physical parameters, using the measured time-aligned state matrix as dynamic boundary conditions for inverse solving. By setting the equivalent roughness of the pipe wall and the local convective heat transfer coefficient, which are conventionally difficult to measure directly, as adjustable variables, and by using closed-loop iterative optimization to guide the model output to fit the measured data, this scheme achieves indirect online identification of key physical parameters reflecting the micro-health status of pipelines, thereby increasing the physical interpretability of micro-state assessment.

[0056] This invention solves for the total entropy production rate of micro-elements based on the reconstructed three-dimensional flow field state parameters and performs orthogonal decomposition according to physical causes. This method transforms dynamic flow field data into characteristic quantities related to fundamental physical dissipation mechanisms, enabling the extraction of temporal characteristics of viscous dissipation entropy production components and thermal conduction entropy production components that characterize different pipe material damage mechanisms. This achieves decoupling and quantification of different damage paths, such as mechanical erosion wear and material thermal fatigue, increasing the evaluation dimensions and objectivity of pipeline micro-degradation state identification. This invention clusters and compares the evolution increment of convergent physical parameters obtained online with a built-in historical measured component map experience database, anchoring the evolution path of disaster-causing types and calling a matching service margin decay derivation function for quantitative calculation. This mechanism establishes a mapping relationship between the current physical state point and the degradation map of the entire service life, calculating the remaining safe operating capacity percentage based on the identification of major micro-failure causes, thus forming an analytical closed loop that combines state identification and life assessment, providing quantitative data support for predictive maintenance at the system level. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings are used to provide a further understanding of the present invention.

[0058] Figure 1A flowchart illustrating an embodiment of this application is disclosed.

[0059] Figure 2 Structural schematic diagrams of embodiments of this application are disclosed. Detailed Implementation

[0060] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0062] The following is in conjunction with the appendix Figure 1 - Figure 2 A preferred description of the present invention is provided below.

[0063] See attached document Figure 1 This invention proposes a fault prediction method for aircraft hydraulic power systems based on signal analysis, comprising the following steps:

[0064] S1. Obtain the actuation command signal and sensor signal under the actuation condition, perform dynamic time axis alignment on the sensor signal to eliminate the time deviation of each cycle, and perform time-domain superposition and averaging on the processed signal to generate a time-aligned state matrix.

[0065] S2. Input the time-aligned state matrix into the preset fluid pipe wall coupled physical model for inverse calculation to reconstruct the three-dimensional flow field state parameters of the target pipe micro-element. The fluid pipe wall coupled physical model includes adjustable physical parameters.

[0066] S3. Calculate the total entropy production rate of the flow field micro-element based on the three-dimensional flow field state parameters and perform physical causal orthogonal decomposition operation to extract the temporal characteristics of the entropy production component.

[0067] S4. Based on the current operating conditions, retrieve the baseline feature template matched in the preset baseline state feature database, perform multi-dimensional quantization comparison between the baseline feature template and the temporal features of the entropy production component, and generate a feature residual vector.

[0068] S5. Input the feature residual vector into the preset parameter optimization model based on gradient descent, update the adjustable physical parameters until the preset convergence tolerance threshold is met, and obtain the convergent physical parameters.

[0069] S6. Cross-compare the converged physical parameters with the preset historical measured component map experience database to obtain the pipeline status prediction results.

[0070] In one embodiment of the present invention, step S1 includes the following steps:

[0071] The system receives an actuation command signal as a timing trigger reference. Within a preset time window, it captures the fluctuation signals output by the high-frequency pressure sensor array and temperature sensor array arranged on the target pipeline as sensing signals. Based on the timing trigger reference, it performs dynamic time warping and matching on the multi-cycle fluctuation signals within the preset time window, aligning them with the fluctuation signals of a reference cycle on the time axis to eliminate phase shifts caused by inconsistent durations of each cycle. Subsequently, it performs time-domain superposition and averaging on all the time-aligned fluctuation signals to generate a time-aligned state matrix.

[0072] Specifically, the system synchronously acquires actuation command signals and sensor signals under specific actuation conditions to generate a time-aligned state matrix for subsequent flow field reconstruction. In this embodiment, a specific actuation condition refers to a hydraulic system operation driven by flight control computer commands, with defined start and end action boundaries, such as landing gear retraction / extension or flap deflection. First, the data processing unit receives actuation command signals characterizing specific maneuvers of the aircraft in real time via the standard communication bus of the flight control computer. This command signal is a digital pulse or a specific message. The system uses an edge detection algorithm, such as capturing the rising edge of the digital signal, to identify the start time of the deterministic actuation command and defines this moment as the timing trigger reference, thereby filtering out random background electrical signal interference under unrelated conditions. It should be understood that the timing trigger reference is a precise system clock point, triggered by the rising edge of the actuation command signal or the parsing of the specific message.

[0073] The system activates a preset data acquisition window based on a timing trigger benchmark. Within this window, multiple high-frequency pressure and temperature sensor arrays deployed on the target hydraulic pipeline begin operating, acquiring data on transient fluctuations in the pipeline at a fixed, synchronized sampling frequency. To ensure the capture of transient details, the sampling frequency of the high-frequency pressure and temperature sensor arrays must be at least 10 times the main pulsation frequency of the tested hydraulic system, typically not less than 5 kHz. Furthermore, the length of the preset time window is set according to the physical duration of the specific actuation process, typically ranging from 1 to 2 seconds, to fully cover the main process of a transient excitation.

[0074] The acquired raw analog signals are converted into discrete digital signal sequences by an analog-to-digital converter and stored in a high-speed data buffer along with precise hardware timestamps. Subsequently, based on the zero point of the time-triggered reference, the digital sequences of transient fluctuation signals acquired by all sensor channels within the time window are timestamped to ensure that the data of each channel are strictly synchronized on the time axis.

[0075] It should be further explained that, due to the inherent flow and pressure pulsations generated by the main pump (such as a piston pump) in the aircraft's hydraulic system during high-speed rotation, these high-order harmonics can severely interfere with the extraction of flow field fluctuations. Therefore, after extracting the digital sequence of transient fluctuation signals from each channel, and before performing time-domain superposition and averaging, the data processing unit synchronously reads the speed sensor data on the hydraulic pump drive shaft via the onboard bus. The obtained pump speed value is multiplied by the number of pistons built into the main pump to calculate the actual fundamental frequency vibration data of the aircraft's hydraulic pump source. Subsequently, using this fundamental frequency and its integer harmonics as the center frequency, the notch band parameters and passband attenuation coefficient of the built-in adaptive comb filter are dynamically configured. The extracted transient fluctuation signals from each channel are input into this adaptive comb filter, and the deterministic periodic harmonic components generated by the pump source are precisely filtered out using multi-level notch characteristics, thereby obtaining the target fluctuation signal stripped of major background mechanical noise.

[0076] Subsequently, the target fluctuation signal is used as a new transient fluctuation signal, and time-domain superposition averaging is performed on the signal data from multiple repeated acquisition cycles triggered by the same actuation command, which have been time-aligned. Specifically, for the first... The sensor channel acquires the data at the first sensor channel. Time series collected during each operation cycle ,in For the trigger reference of this cycle Discrete-time index; to eliminate phase shifts caused by inconsistent durations in different actuation cycles, the time series of each actuation cycle is dynamically time-normalized and compared with the time series of a reference cycle (e.g., the first cycle). Perform timeline alignment; assuming the first timeline obtained through dynamic time warping is... The mapping function from each actuation cycle to the reference cycle is: The aligned time series is then represented as Subsequently, for all The time series of the aligned actuation cycles are averaged over time in the time domain, and the calculation formula is as follows: Finally, the time-domain summation results of all sensor channels are averaged. Together they form a time-aligned state matrix Its dimensions are This enhances the deterministic signal components related to the actuation process while suppressing random broadband noise. It should be noted that the number of time-domain superposition averaging processes, i.e., the number of actuation cycles involved in the averaging, is typically between 10 and 100. More cycles result in better suppression of random noise but prolong data accumulation time. Through the synergy of multi-cycle time-domain superposition averaging and time-series alignment, the masking effect of non-operating-condition random broadband frequency response noise on weak flow field fluctuations can be eliminated.

[0077] Finally, all sensor channel data, after time-series alignment and time-domain overlay averaging, are arranged by channel index and sampling time point to form a three-dimensional data matrix, namely the time-aligned state matrix. In this embodiment, the time-aligned state matrix is ​​a matrix with one dimension. Multiply A three-dimensional tensor multiplied by 1, where This represents the total number of sensor channels participating in the data acquisition. This represents the number of sampling points for a single channel within a time window; its value is the product of the sampling frequency and the length of the time window.

[0078] It should be noted that the above parameters are based on the assumptions of the working characteristics of a typical commercial aircraft hydraulic system. The preset time window is set to 1.5 seconds, which is based on the statistical analysis of the typical duration of landing gear retraction. The average number of time-domain superpositions is set to 20, which is based on the consideration of balancing the need to improve the data signal-to-noise ratio and the real-time requirements of online diagnosis.

[0079] For example, suppose the system recognizes the rising edge of the landing gear retraction actuation command and records the absolute system time at that moment. This serves as the timing trigger reference. The preset time window length is set to 1.5 seconds. Three high-frequency pressure and temperature sensors, positioned on the target pipeline, sample synchronously at a frequency of 10 kHz. Within the time window... to Within seconds, each sensor acquires 15,000 data points. The system reads historical sensor data from the cache for the most recent 20 landing gear retraction actions within this window. For the first pressure sensor channel, the signal sequence of the first cycle is used as the reference sequence. For the subsequent 20 extracted historical cycle sequences, due to load fine-tuning causing slight stretching or compression in the fluctuation response on the time axis, the system first uses a dynamic time warping algorithm to calculate the nonlinear mapping function between these 20 historical sequences and the first cycle sequence. Perform sequence continuation or compression alignment. After normalization, relative to... At the 500th reference sampling point, all corresponding data points after mapping and alignment are extracted. These 20 corrected values ​​are, for example, 101.5 MPa, 101.6 MPa, 101.4 MPa, ..., 101.5 MPa. The aligned values ​​are then subjected to an arithmetic mean of time-domain superposition averaging, yielding 101.5 MPa. This value is used as the final noise-reduced data at the 500th time index of the first pressure sensor channel in the time-aligned state matrix. This averaging operation is performed on each time index point of each sensor channel according to the above regularization mapping rules. Finally, for the three pressure sensors and one temperature sensor, a time-aligned state matrix with dimensions of 4×15000×1 is generated, where each row represents a transient signal sequence of one sensor, aligned within a 1.5-second time window after 20 superposition averaging operations.

[0080] In one embodiment of the present invention, step S2 includes the following steps:

[0081] The inherent geometric configuration dimensions of the target pipeline and the initial fluid properties under the current operating conditions are obtained to initialize the fluid-pipe-wall coupled physical model; the equivalent roughness of the pipe wall and the local convective heat transfer coefficient are defined as adjustable physical parameters; the time-aligned state matrix is ​​used as the measured boundary condition to input the fluid-pipe-wall coupled physical model for inverse calculation of transient mechanical and thermal response; the solution distribution results of the fluid-pipe-wall coupled physical model are extracted, and the three-dimensional flow field state parameters, including the local velocity vector distribution, shear stress distribution, and temperature gradient field of the micro-element, are output.

[0082] Specifically, the time-aligned state matrix is ​​reverse-engineered using a fluid-to-wall coupled physical model including adjustable variables to reconstruct the three-dimensional flow field state parameters of the target pipeline element. In this step, the fluid-to-wall coupled physical model is a computational fluid dynamics model used to simulate the fluid dynamics and thermodynamic processes inside a pipeline. Its key feature is its ability to solve unsteady-state problems and consider heat exchange between the fluid and the solid wall. First, the computing host retrieves the inherent geometric dimensions of the target pipeline from a pre-set digital twin model database. These dimensions define the pipeline's inner diameter, wall thickness, bending radius, and length. Simultaneously, based on the current operating temperature and pressure measured by onboard sensors, the initial fluid properties under the current operating conditions, including density and dynamic viscosity, are queried and loaded from the built-in fluid property parameter library. Specific heat capacity and thermal conductivity are used as geometric and physical property data to initialize a fluid-pipe-wall coupled physical model. This model is based on fundamental governing equations of fluid mechanics and heat transfer, such as the Navier-Stokes equations and the energy conservation equations, and is constructed by discretization using the finite volume method. Specifically, its spatial discretization employs a second-order upwind scheme, and the SIMPLE algorithm is configured for iterative solution of the pressure-velocity coupled flow field.

[0083] In this model, the equivalent wall roughness and local convective heat transfer coefficient, characterizing the physical state of the pipe wall, are set as adjustable physical parameters, with their initial values ​​set to the baseline healthy state values ​​at the time of equipment delivery. Specifically, the equivalent wall roughness is an equivalent physical quantity used to quantify the resistance effect of pipe wall surface roughness on fluid flow; its initial value is set based on pipe manufacturing standards, for example, for new hydraulic steel pipes, its typical range is 1.5 μm to 10 μm. The local convective heat transfer coefficient is a physical quantity characterizing the intensity of convective heat transfer between the fluid and the pipe wall, with units of W / (m²·K); its initial value is usually estimated based on the standard Nusselt number empirical formula.

[0084] Next, the time-aligned state matrix is ​​used as the measured dynamic boundary condition and loaded into the fluid-pipe-wall coupled physical model. Specifically, the time-series data of each sensor channel in the matrix is ​​applied to the nodes in the model mesh corresponding to the actual physical locations of the sensors, serving as a mandatory constraint for the inverse calculation of the transient force-thermal response. The numerical solver is then started, and the flow field distribution within the model is adjusted through iterative calculations until the residuals between the pressure and temperature responses predicted by the model at the sensor nodes and the measured values ​​in the time-aligned state matrix converge to within a preset accuracy range.

[0085] After the inverse solution converges, the system extracts the complete solution results of the fluid-pipe wall coupled physical model across all mesh elements in the entire computational domain, covering all time steps within a preset time window. These results are structured into three-dimensional flow field state parameters, providing complete physical field input for subsequent entropy production analysis. In this embodiment, the three-dimensional flow field state parameters are a high-dimensional dataset including spatial three-dimensional coordinates, time, and multiple physical quantities. It accurately describes the spatiotemporal evolution details of the flow field inside the target pipeline segment under a specific operating condition. Its data structure includes the local velocity vector distribution, shear stress distribution, and temperature gradient field of each micro-element at each moment. In particular, the shear stress value of the inner wall of the pipeline in the shear stress distribution... It is calculated based on the local velocity vector distribution obtained from the solution, and its calculation follows Newton's law of internal friction. Specifically, the formula for calculating the shear stress at the wall surface is: ;

[0086] in, This represents the shear stress on the inner wall of the pipeline. This represents the dynamic viscosity of the fluid, a value provided by the fluid's initial physical properties. This indicates the flow velocity parallel to the wall surface. This represents the coordinates of the normal direction perpendicular to the wall. Indicates the location on the wall. At the point where the velocity equals 0, the rate of change of the velocity along the normal direction is directly provided by the solution of the fluid-pipe-wall coupled physical model.

[0087] For example, the inherent geometric dimensions of the target pipeline are retrieved, assuming the pipeline is a stainless steel pipe with an inner diameter of 20 mm and a wall thickness of 2 mm. Simultaneously, the initial fluid properties are obtained, and based on the current operating conditions, the fluid density is set to 850 kg / m³, and the dynamic viscosity is... The initial value is 0.015 Pa·s. The system initializes the fluid-pipe wall coupled physical model, setting the initial value of the equivalent roughness of the pipe wall to 5 μm and the initial value of the local convective heat transfer coefficient to [value missing]. Next, the system uses the 4x15000x1 time-aligned state matrix generated by S1 as the boundary condition input. Three pressure data points and one temperature data point from the matrix are applied to the corresponding three pressure monitoring points and one temperature monitoring point in the model's computational domain, respectively. Then, the system initiates the inverse solver. After iterative calculations, the entire three-dimensional flow field within the model is reconstructed. Finally, the three-dimensional flow field state parameters are extracted from the converged model solution. For example, in the analysis... At a specific moment, for a key wall element on the outer side of the pipe bend, the system extracts the normal velocity gradient at that element. for Calculate the shear stress at that point using the formula. The Pa is 520. Simultaneously, the system also extracts the local velocity vector distribution and temperature gradient field of the micro-element. The system integrates this data for all micro-elements across all 15,000 time steps to form a complete three-dimensional flow field state parameter, which is then used for entropy production decomposition calculation in the next step, S3.

[0088] In one embodiment of the present invention, step S3 includes the following steps:

[0089] By integrating a pre-defined numerical model of an irreversible fluid dynamics process and utilizing extracted three-dimensional flow field state parameters, the total entropy production rate of the corresponding infinitesimal element continuously evolving within a pre-defined time window is calculated. Orthogonal decomposition and dimensionality reduction extraction are performed on the total entropy production rate to separate the viscous dissipative entropy production component sequence and the thermal conduction entropy production component sequence. Temporal envelope curve fitting feature extraction is performed on the viscous dissipative entropy production component sequence and the thermal conduction entropy production component sequence to obtain the peak evolution time sequence node and the relative phase shift angle. The peak evolution time sequence node, the relative phase shift angle, and the viscous dissipative entropy production component sequence and the thermal conduction entropy production component sequence are combined to form the temporal feature of the entropy production component.

[0090] Specifically, the total entropy production rate of the flow field micro-elements is calculated based on the three-dimensional flow field state parameters, and a physical causal orthogonal decomposition operation is performed to extract the temporal characteristics of the independent entropy production components that characterize the damage mechanisms of different pipe materials. First, the data processing unit receives the three-dimensional flow field state parameters output in step S2. The system then uses a numerical model based on irreversible fluid dynamics processes to traverse each spatial micro-element and time step defined in the three-dimensional flow field state parameters. During this traversal, for any micro-element at any time point, the system uses its included local velocity vector distribution, shear stress distribution, and micro-element temperature gradient field data to calculate the total entropy production rate at that spatiotemporal point. It should be understood that the numerical model of irreversible fluid dynamics processes is a calculation program based on the second law of thermodynamics, used to quantify energy dissipation in the flow field.

[0091] Next, based on the second law of thermodynamics, orthogonal decomposition and dimensionality reduction are performed on the calculated total entropy yield along different physical irreversible causes. In this embodiment, the orthogonal decomposition is based on the thermodynamic local equilibrium assumption, reducing the total entropy yield per unit volume. It is decomposed into a viscous dissipation term caused by the fluid viscosity effect and a heat conduction term caused by the heat conduction effect. The specific calculation formula is as follows:

[0092] ;

[0093] Among them, the viscous dissipative entropy production component The calculation formula is: ;

[0094] Entropy production component of heat conduction The calculation formula is: ;

[0095] In the above formula, It represents the total entropy yield per unit volume; This represents the entropy production component of viscous dissipation; This represents the entropy production component of heat conduction; For fluid dynamic viscosity, the initial fluid property parameters defined in S2 are usually used as constant values ​​when simplifying engineering calculations. Alternatively, physical field variables based on the dynamic updating of infinitesimal absolute temperature and pressure can be used according to actual accuracy requirements. The absolute temperature of the infinitesimal element is provided by the temperature gradient field of the infinitesimal element. It is a viscous dissipation function, the value of which depends on the square of the velocity gradient and is calculated from the local velocity vector distribution; The thermal conductivity of the fluid can also be approximated by a constant derived from the initial physical properties of the fluid, or by dynamically updated physical field variables as needed. This represents the temperature gradient vector, which is directly extracted from the temperature gradient field of the infinitesimal element.

[0096] Through the above calculations, the system independently extracts the viscous dissipative entropy production component and the thermal conduction entropy production component, and organizes their values ​​for all time steps within a preset time window into sequences for the viscous dissipative entropy production component and the thermal conduction entropy production component. Finally, the system executes a feature extraction algorithm for time-domain envelope curve fitting on these two independent time series. In this process, the time-domain envelope curve fitting uses signal processing techniques such as Hilbert transform to extract the instantaneous amplitude of the signal to generate two independent envelope curves. The algorithm calculates and identifies the peak points of the two envelope curves, records the precise time of occurrence of these peak points, and forms peak evolution time sequence nodes. The system further calculates the time difference between the peak points of the two envelope curves and converts it into a relative phase offset angle. It should be noted that the relative phase offset angle is a physical quantity characterizing the time difference between the peak occurrences of two time series signals, and its calculation method is to divide the peak time difference by the characteristic period of the operating condition and then multiply by 360 degrees.

[0097] Finally, the system integrates these two complete time series (i.e., the viscous dissipation entropy production component series and the heat conduction entropy production component series), as well as the peak evolution time series nodes extracted from them and the relative phase offset angle between them, into a structured feature array, namely the entropy production component time series features, for subsequent state comparison analysis.

[0098] For example, following the example in S2, calculations are performed based on the acquired three-dimensional flow field state parameters. Taking a key wall element on the outer side of a pipe bend at a given time as an example, the physical quantities at this point are extracted from the three-dimensional flow field state parameters: absolute temperature T is 353.15 K, and viscous dissipation function... The calculated value is The square of the magnitude of the temperature gradient vector for To simplify calculations, the fluid properties in this embodiment are inherited from the initial constant value of S2: dynamic viscosity. for And obtain the thermal conductivity from the property library. for Substituting into the formula to calculate: viscous dissipative entropy production component. for Entropy production component of heat conduction Approximately This calculation is repeated for all 15,000 time steps within the preset time window to generate complete sequences of viscous dissipation entropy production and heat conduction entropy production. Subsequently, envelope fitting is performed on these two sequences, revealing that the peak of the viscous dissipation entropy production sequence occurs at... Seconds, while the peak of the heat conduction entropy production component sequence appears in The system records these two time points as peak evolution time series nodes. The time difference between the two is 0.06 seconds. Assuming the characteristic period of this operating condition is 1.0 second, the relative phase offset angle is calculated to be 21.6 degrees. Finally, the system packages this set of data—the viscous peak node at 0.75 seconds, the thermal conduction peak node at 0.81 seconds, and the relative phase at 21.6 degrees—to construct the entropy production component time series feature and outputs it to the next step.

[0099] Furthermore, before comparing this feature output, the system introduces a self-checking mechanism for extreme fluid physics phenomena. Because a sudden large opening of a hydraulic valve can cause local fluid pressure to drop below the saturated vapor pressure, triggering a violent cavitation collapse phenomenon, the high-frequency disordered thermal pulses generated by cavitation collapse can cause a phase abrupt distortion in the originally delayed heat conduction response. A baseline phase shift angle interval between heat conduction and viscous dissipation is statistically extracted from the pipeline's historical data without cavitation during its normal service life, and the upper limit of the interval tolerance is typically set to plus or minus three standard deviations of the baseline mean. Comparing the calculated relative phase shift angle with this baseline phase shift angle interval, if the relative phase shift angle is found to far exceed the outlier tolerance set by the aforementioned standard deviation, it is determined that the currently extracted entropy production component time-series feature has been contaminated by extreme cavitation collapse noise and no longer possesses the physical meaning of characterizing gradual degradation. At this point, the system immediately triggers a forced resampling command. On the one hand, it erases and discards the time-aligned state matrix generated this time in memory. On the other hand, it sends a reset signal to the data processing unit and directly returns to step S1 to wait for the next actuation command signal issued by the flight control computer. It relies on the subsequent new acquisition cycle to avoid single, occasional severe cavitation samples.

[0100] In one embodiment of the present invention, step S4 includes the following steps:

[0101] Based on the current operating condition and the corresponding maneuver category number, the corresponding baseline feature template is retrieved from the baseline state feature database. The viscous dissipative entropy production component sequence and the thermal conduction entropy production component sequence contained in the entropy production component time series feature are matched with the corresponding baseline sequence in the baseline feature template at the same time measure length to calculate the dynamic time warping distortion distance, and their respective independent nonlinear deviation time series are measured. Each nonlinear deviation time series is normalized and mapped based on a preset scaling factor. Principal component analysis is performed on each normalized time series to perform feature dimensionality reduction and extract the principal component feature vector. Vector concatenation operation is performed on each extracted principal component feature vector to calculate and generate the feature residual vector.

[0102] Specifically, based on the current operating condition, the baseline feature template matching the one in the system's built-in baseline state feature database is retrieved for multi-dimensional quantization comparison with the temporal features of the entropy production components, generating a feature residual vector for driving the adaptive iteration of closed-loop parameters. First, the data processing unit initiates a query request to the system's internal memory based on the motor category number associated with the identified actuation command signal. Then, in the baseline state feature database, the baseline feature template corresponding to the precisely matched motor category number under the factory-degraded, stable state is retrieved. It should be understood that the baseline state feature database is a pre-built database with a key-value pair structure, where the key is the motor category number and the value is the corresponding baseline feature template. The database is pre-built by performing hydraulic pipeline steady-state calibration tests or high-fidelity simulations on newly manufactured equipment, and its baseline feature templates are structured data storing the baseline entropy production component sequence and its temporal features under ideal healthy conditions.

[0103] Next, the sequence data contained in the temporal features of the entropy production components generated in step S3, namely the viscous dissipative entropy production component sequence and the thermal conduction entropy production component sequence, are compared in pairs with the benchmark sequence stored in the baseline feature template. This comparison is achieved by executing the Dynamic Time Warping (DTW) warp distance matching algorithm.

[0104] In this embodiment, the Dynamic Time Warping algorithm is an algorithm used to measure the similarity between two time series that may have different lengths. It finds the optimal matching path between the two by non-linearly warping the time axis. The core of the algorithm is to construct a cumulative distance matrix D and find the optimal path through dynamic programming. The recursive formula for the cumulative distance is:

[0105] ;

[0106] in, Indicates the first sequence The points and the first two sequences The cumulative normalized distance between points; Represents corresponding points in two sequences and The local distance between them is usually expressed as Euclidean distance, i.e. It should be noted that here... and Specifically, it represents the magnitude of the entropy yield of the sequence to be matched at the corresponding time node. The DTW algorithm essentially matches the evolution pattern of the physical dissipation amplitude when the flow field undergoes a specific action. and These represent the indexes of the time data points on the first and second feature time series, respectively. , The terms in the derivation represent the known cumulative normalized distances on the paths of the preceding adjacent nodes.

[0107] This calculation quantifies the similarity between the current viscous dissipation sequence and the reference viscous dissipation sequence, as well as between the current heat conduction sequence and the reference heat conduction sequence. It also quantifies the independent nonlinear deviation time series in both the viscous dissipation state dimension and the heat conduction response dimension. It should be noted that the nonlinear deviation time series is derived from the local distance along the optimal path. The sequence is composed of elements, each of which represents the local difference at the corresponding time point on the optimal nonlinear alignment path.

[0108] Finally, considering the significant difference in physical magnitude between viscous dissipation and heat conduction, to avoid masking the smaller thermal conduction deviation characteristics during subsequent parameter optimization gradient calculations, a normalization mapping process is first performed on the time series of these two nonlinear deviations representing different physical dimensions. Specifically, let the time series of the normalized viscous dissipation entropy production component be... The time series of the heat conduction entropy production component is ,in Let be the sequence length. To eliminate redundant information and achieve feature reduction, principal component analysis (PCA) is performed on each sequence for order reduction; calculation... The covariance matrix is ​​used to extract the principal component eigenvectors. (dimension is) ,in Similarly, we can obtain (dimension is) ,in Subsequently, the two dimensionality-reduced vectors are fully connected and concatenated in one dimension, and vector concatenation is performed to generate feature residual vectors. : The final result That is, the dimension is The feature residual vector. In this embodiment, the dimension of the feature residual vector is the sum of the lengths of the two nonlinear deviation time series, and the magnitude of its vector norm is directly related to the degree to which the current system state deviates from the healthy baseline.

[0109] For example, following the example of S3, the data processing unit receives the temporal characteristics of the entropy production component output by S3. The system identifies the maneuver category number associated with this feature as "LG-RET-01", corresponding to the landing gear retraction action. Using this number as an index, the corresponding baseline feature template is retrieved from the baseline state feature database. This template shows that, under healthy conditions, the viscosity peak node should be 0.72 seconds, the heat conduction peak node should be 0.79 seconds, the relative phase should be 14.4 degrees, and it includes two baseline entropy production component sequences, namely the baseline viscosity sequence and the baseline heat conduction sequence. Next, the current viscous dissipation entropy production component sequence generated by S3 is dynamically time-warped and matched with the baseline viscosity sequence in the template. This calculation generates the first nonlinear deviation time series, which, after being normalized by the corresponding scaling factor, is assumed to have a length of 128 data points and a sequence value of [0.12, 0.15, ..., 0.88]. Simultaneously, the same calculations and normalization mappings are performed on the current heat conduction entropy production component sequence and the reference heat conduction sequence to generate a second nonlinear deviation time series, also with a length of 128 data points and sequence values ​​[0.03, 0.04, ..., 0.21]. Finally, to eliminate redundancy and achieve feature reduction, the system performs principal component analysis on both of these 128-point nonlinear deviation time series. It is assumed that after the reduction process, the first 32 principal components are extracted to form feature vectors. and Then, a fully connected vector concatenation process is performed to combine the two vectors to generate a feature residual vector with a total dimension of 64 (i.e., 32+32). The specific numerical assumption is as follows: The 64-dimensional feature residual vector, after being reduced in order and purified, is passed to step S5 as the input for optimizing model parameters.

[0110] In one embodiment of the present invention, step S5 includes the following steps:

[0111] The feature residual vector is transmitted to the parameter optimization model based on gradient descent and transformed into a parameter correction vector. The effect correction step size of the parameter correction vector is extracted, backtracked into the fluid pipe wall coupled physical model, and the initial parameter values ​​of the adjustable physical parameters are progressively modified. Based on the dynamic model parameters after the modification takes effect, a closed-loop derivation operation including the parameter update process is repeatedly triggered until the L2 value of the feature residual vector is lower than the convergence tolerance threshold, and the locked parameters are derived into convergent physical parameters.

[0112] Specifically, the feature residual vector is fed into a gradient descent-based parameter optimization model, and the adjustable physical parameters are adaptively penalized and updated until the model's internal properties approximate the actual detected physical response, thus obtaining accurate and unbiased convergent physical parameters. First, the feature residual vector generated in step S4 is used as the initial optimization target and transmitted to a pre-constructed gradient descent-based parameter optimization model employing the fastest descent solution strategy. It should be noted that the gradient descent-based parameter optimization model is a gradient descent-based optimization algorithm, such as gradient descent or its adaptive moment estimation Adam optimizer variant, used to adjust model parameters by predicting bias through backpropagation. In this embodiment, the fluid-pipe-wall coupling physical model can also achieve gradient tracing at the underlying computational architecture using any of the following methods: First, employing adjoint solver technology to accurately analyze the sensitivity derivative of the target residual with respect to adjustable physical parameters by solving the adjoint partial differential equation of the corresponding fluid control equation; Second, constructing a white-box reconstruction model of a fluid numerical simulator based on a differentiable physics engine, such as a differentiable automatic differential programming framework JAX, so that all tensor operations in the flow field discretization calculation support automatic differentiation, thus achieving native compatibility with the computational graph of the deep learning optimizer; Third, using a physical knowledge neural network PINN to pre-train a high-fidelity surrogate model to replace the original finite volume method computation domain, thereby achieving the unification of forward solution and backward differentiation. Within this model, the feature residual vector is used as the basis for calculating the initial gradient direction. Supported by the aforementioned differentiable architecture or adjoint equation, the feature residual vector is transformed into a parameter correction vector including the step size of the correction, through chain rule or spatial derivative analytical operations, to offset the current model prediction bias. Specifically, the parameter correction vector is a vector with the same dimension as the adjustable physical parameter, and each component, i.e. the action correction step, indicates the direction and magnitude of the corresponding physical parameter to be adjusted.

[0113] Next, the correction step size for each adjustable physical parameter, namely the equivalent roughness of the pipe wall and the local convective heat transfer coefficient, is extracted from the parameter correction vector. These steps are then backtracked and fed into the fluid-pipe wall coupled physical model defined in S2. By modifying the current values ​​of the equivalent roughness of the pipe wall and the local convective heat transfer coefficient in a cumulative manner using these correction steps, an update of the physical model parameters is completed. Subsequently, based on the updated dynamic model parameters, a complete full-link forward prediction loop calculation is immediately initiated. This loop covers the entire process from the fluid force reconstruction analysis in S2, to the entropy production feature extraction in S3, and finally to the regeneration of the regression residual vector in S4.

[0114] The system enters a closed-loop derivation process that executes repeatedly. After each iteration, the system calculates the global norm 2 of the newly generated feature residual vector and compares it with a preset convergence tolerance threshold set internally by the system. In this embodiment, the global norm 2 is a method for measuring the length of a vector in a vector space, used to quantify the overall deviation between the model's prediction results and the measured results. After the k-th iteration, its calculation formula is:

[0115] ;

[0116] in, Indicates the first The global L2 norm value of the feature residual vector generated by the next iteration; This is the sequence number of the current iteration in the closed-loop derivation calculation process; It is the residual vector The One component; It is the total dimension of the residual vector, such as 64 obtained after principal component reduction and combination in example S4.

[0117] The condition for stopping the iteration is: ;

[0118] in This represents the preset convergence tolerance threshold. It should be understood that... The setting is based on ensuring the accuracy of parameter identification while avoiding excessive calculation time and numerical oscillation. The typical value range is 1×10. -6 Up to 1×10 -4 .

[0119] If the norm value is not lower than the threshold, the cycle of parameter correction, model update, and forward prediction is repeated. Until the global L2 value of the feature residual vector calculated in a certain iteration first shrinks to below the threshold, the computing host will instruct the iterative feedback loop to stop, and export and store the latest version of adjustable physical parameters that are locked at this moment and enable the model to reach steady-state equilibrium as the final result as the convergent physical parameters.

[0120] For example, the initial feature residual vector generated by S4, i.e., the dimension is reduced and concatenated to 64. This is fed into a parameter optimization model based on gradient descent. After calculation, the model outputs a two-dimensional parameter correction vector, for example, [+0.2×10]. -6[m, -15 W / (m²·K)]. The effect of the step size is extracted, updating the equivalent pipe wall roughness (initial value 5 μm in S2) to 5.2 μm, and the local convective heat transfer coefficient (initial value 1200 W / (m²·K)) to 1185 W / (m²·K). Subsequently, the forward calculations from S2 to S4 are re-executed using these new parameters, yielding a new characteristic residual vector with reduced values ​​for each component. The global norm 2 of this new vector is calculated, assuming a result of 0.95. A preset convergence tolerance threshold is set. 1×10 -5 Since 0.95 is much larger than 1×10 -5 The next iteration continues. Assume that after 15 iterations, the equivalent roughness of the pipe wall is updated to 5.8 μm, and the local convective heat transfer coefficient is updated to 1172 W / (m²·K). Based on these parameters, the characteristic residual vector generated in the latest forward calculation has a global norm of 8.7 × 10⁻⁶. -6 Because 8.7×10 -6 Less than the threshold 1×10 -6 The convergence condition was met. The system then stopped the iteration loop and output and stored the current pair of parameter values, namely roughness 5.8 μm and thermal conductivity 1172 W / (m²·K), as the final convergence physical parameters.

[0121] In one embodiment of the present invention, step S6 includes the following steps:

[0122] Extract the corresponding values ​​of the convergent physical parameter records to form the local roughness increment and the local convective heat transfer coefficient drift attenuation, and convert them into the current service state evolution increment of the system after merging. Perform clustering node pattern similarity calculation on the current service state evolution increment of the system in the historical measured component map experience database to anchor the disaster type evolution path. Calculate the actual remaining safe operating capacity percentage data feature cluster by weighted summation based on the built-in service margin attenuation derivation function mapped by the disaster type evolution path, and encapsulate it as the pipeline status prediction result.

[0123] Specifically, the physical essence of the converged physical parameters is analyzed and deviated from the initial value bias, and cross-referenced with historical measured component maps to generate pipeline condition prediction results, including quantified remaining cycle information, to guide planned operation and maintenance. First, the data processing unit extracts the converged physical parameters that were stopped and fixed in step S5, namely the updated and settled equivalent pipe wall roughness and local convective heat transfer coefficient. Then, the initial factory health value of the pipeline component is retrieved from the equipment file database. By subtracting the factory value from the equivalent pipe wall roughness in the converged physical parameters, the local roughness increment is obtained. Similarly, subtracting the factory-set local convective heat transfer coefficient from the corresponding value in the convergence physical parameters yields the amount of drift attenuation of the local convective heat transfer coefficient. These two differences are combined into a two-dimensional vector, which together constitute the increment of the system's current service status evolution.

[0124] Next, the built-in historical measured component atlas experience database is read. It should be noted that this experience database is constructed using a large amount of bench accelerated life testing and historical maintenance data. It internally stores incremental sample vectors of service condition evolution at different degradation stages and their corresponding failure mode labels. Using the service condition evolution increment just calculated as the query feature, nearest neighbor clustering node pattern similarity calculation is performed within the feature vector space of this experience database to find the degradation mode node that is closest to the current increment vector in Euclidean distance. Based on this, the current degradation state is accurately labeled, thereby anchoring its corresponding disaster-causing evolution path.

[0125] Finally, based on the identified disaster type evolution path, the associated built-in service margin decay derivation function is retrieved from the experience base. In this embodiment, the service margin decay derivation function is used to calculate the percentage of a component's consumed lifespan based on the incremental changes in service status. For example, a linear function model can be used:

[0126] ;

[0127] in, This indicates the percentage of lifespan already consumed. and These are weighting coefficients associated with the roughness increment and the thermal conductivity decay, respectively. Their values ​​are obtained by fitting statistical data of the corresponding degradation paths in the empirical database, representing the contribution of different physical parameter changes to lifetime loss. Local roughness increment This refers to the increase in the current equivalent roughness of the pipe wall compared to the factory reference value. It also includes the decrease in the drift of the local convective heat transfer coefficient. This refers to the decrease in the current local convective heat transfer coefficient compared to the factory baseline value. The state evolution increment is a two-dimensional vector, whose components are... and .

[0128] This function calculates the percentage of performance loss of a component by combining the service status evolution increment, and then derives a data feature cluster of the component's actual remaining safe operating capacity percentage. Among these features is the component's remaining safe operating capacity percentage. for: ;

[0129] Further integrate the description of the causes of lesions associated with this path, and structure and package information such as the percentage of remaining lifespan, specific disaster type, and microscopic physical and chemical causes into a final pipeline status prediction result, and send it to the aircraft cockpit integrated display and control system for alarm or status display.

[0130] For example, the system extracts the following convergence physical parameters: equivalent pipe wall roughness of 5.8 μm and local convective heat transfer coefficient of 1172 W / (m²·K). The system retrieves the factory health values ​​for this pipeline from the equipment file: 5.0 μm and 1200 W / (m²·K). Based on this, the local roughness increment is calculated. The local convective heat transfer coefficient drift attenuation is 0.8 μm. The value is 28 W / (m²·K). These two values ​​constitute the state evolution increment vector [0.8×10]. -6

[28] . Next, this vector is input into the historical measured component atlas experience database for nearest neighbor matching. The calculation results show that the vector is closest to the cluster center labeled as Type II erosion wear, therefore the system labels the current degradation mode as Type II erosion wear. Then, the service margin decay derivation function associated with this mode is retrieved, with its weighting coefficient being... equal , It equals 0.005 m²·K / W. Substituting this into the formula, we can calculate the consumed lifespan: The value is 0.46. Therefore, the remaining safe operating capacity percentage is... The result is 54%. Finally, the pipeline condition prediction result is generated, which reads: "Current condition of the target pipeline: Class II erosion wear. Main cause: Significant increase in pipe wall roughness. Estimated remaining safe life: 54%." This structured information is then sent to the cockpit integrated display and control system.

[0131] In other embodiments of the present invention, after generating the pipeline condition prediction result, the method further includes:

[0132] Extract the data feature cluster of the actual remaining safe operating capacity percentage within the pipeline status prediction results; determine whether the data feature cluster of the actual remaining safe operating capacity percentage is lower than the preset critical safety threshold; if the determination is yes, generate a hydraulic system downgrade operation command and send it to the flight control computer to limit the preset actuation command signal trigger frequency and actuation rate.

[0133] Specifically, in addition to the screen display used for health management status, the generated pipeline status prediction results are also fed forward parameters into the aircraft's closed-loop execution network. Specifically, the health management computer extracts the data feature clusters representing the percentage of actual remaining safe operating capacity from the prediction results and determines in real time whether it is below the system's calibrated critical safety threshold, for example, a safety margin set at 15%. When it is determined to be below this threshold, it indicates that the microcracks or thinning of the target pipeline are highly susceptible to physical rupture under full-speed pressure. The system then generates a hydraulic system downgrade operation command with the target pipeline identifier. The system sends this downgrade operation command to the flight control computer via the avionics bus. After parsing the command, the flight control computer, during subsequent maneuvers related to the dangerous pipeline, such as landing gear retraction and extension, physically limits the actuation rate of the corresponding components by reducing the upper limit of the electro-hydraulic servo valve's electric drive current gain, while simultaneously extending the trigger time interval. This physical constraint mechanism significantly reduces the high-intensity water hammer peak overshoot formed by hydraulic oil in the pipeline, ultimately sacrificing some maneuver response efficiency to buy time for the pipeline to operate safely and ensure that the system can be degraded to the return-to-ship maintenance phase.

[0134] Example 2

[0135] See appendix Figure 2 Embodiment 2 of the present invention also proposes a fault prediction system for aircraft hydraulic power systems based on signal analysis, comprising the following modules:

[0136] The data synchronization acquisition module is used to acquire the actuation command signal and sensor signal under the actuation condition, perform dynamic time axis alignment on the sensor signal to eliminate the duration deviation of each cycle, and perform time domain superposition and averaging on the processed signal to generate a time alignment state matrix.

[0137] The flow field state reconstruction module is used to input the time-aligned state matrix into the preset fluid pipe wall coupled physical model for inverse calculation, and reconstruct the three-dimensional flow field state parameters of the target pipe micro-element. The fluid pipe wall coupled physical model includes adjustable physical parameters.

[0138] The entropy production feature extraction module calculates the total entropy production rate of the flow field micro-element based on the three-dimensional flow field state parameters and performs physical causal orthogonal decomposition operation to extract the temporal features of the entropy production component.

[0139] The feature residual generation module is used to retrieve the baseline feature template matched in the preset baseline state feature database according to the current operating conditions, and compare the baseline feature template with the temporal features of the entropy production component in a multidimensional quantization to generate a feature residual vector.

[0140] The parameter optimization identification module is used to input the feature residual vector into the preset parameter optimization model based on gradient descent, update the adjustable physical parameters until the preset convergence tolerance threshold is met, and obtain the convergent physical parameters.

[0141] The degradation state prediction module is used to cross-compare converged physical parameters with a preset historical measured component map experience database to obtain pipeline state prediction results.

[0142] Each of the modules can be implemented in whole or in part through software, hardware, or a combination thereof. It supports hardware embedded in or independent of the processor in the computer device, and also supports software stored in the memory of the computer device, so that the processor can call and execute the operations corresponding to each of the above modules.

[0143] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0144] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0145] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0146] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for predicting faults in an aircraft hydraulic power system using signal analysis, characterized in that, Includes the following steps: The actuation command signal and sensor signal under the actuation condition are acquired. The sensor signal is dynamically aligned on the time axis to eliminate the time deviation of each cycle. The processed signal is then superimposed and averaged in the time domain to generate a time-aligned state matrix. The time-aligned state matrix is ​​input into the preset fluid pipe wall coupled physical model for inverse calculation to reconstruct the three-dimensional flow field state parameters of the target pipe micro-element. The fluid pipe wall coupled physical model includes adjustable physical parameters. The total entropy production rate of flow field micro-elements is calculated based on three-dimensional flow field state parameters, and physical causal orthogonal decomposition is performed to extract temporal features of entropy production components. Specifically, this includes: fusing a preset numerical model of irreversible fluid dynamics processes, using the extracted three-dimensional flow field state parameters to calculate the total entropy production rate of the corresponding micro-elements continuously evolving within a preset time window; performing orthogonal decomposition and dimensionality reduction extraction on the total entropy production rate to separate the viscous dissipative entropy production component sequence and the thermal conduction entropy production component sequence; performing temporal envelope curve fitting feature extraction on the viscous dissipative entropy production component sequence and the thermal conduction entropy production component sequence to obtain the peak evolution time sequence nodes and relative phase shift angles, and combining the peak evolution time sequence nodes, relative phase shift angles, and viscous dissipative entropy production component sequences and thermal conduction entropy production component sequences to form temporal features of entropy production components. Based on the current operating conditions, retrieve the baseline feature template matched in the preset baseline state feature database, and perform multi-dimensional quantization comparison between the baseline feature template and the temporal features of the entropy production component to generate a feature residual vector. Input the feature residual vector into the preset parameter optimization model based on gradient descent, update the adjustable physical parameters until the preset convergence tolerance threshold is met, and obtain the convergent physical parameters. The convergent physical parameters are cross-referenced with a pre-set historical measured component map experience database to obtain the pipeline condition prediction results.

2. The method for predicting faults in an aircraft hydraulic power system based on signal analysis according to claim 1, characterized in that, Dynamic time-axis alignment of the sensor signal is performed to eliminate duration deviations in each cycle, and the processed signal is then time-domain superimposed and averaged to generate a time-aligned state matrix. This includes the following steps: Receive actuation command signals as timing trigger references; Within a preset time window, the fluctuation signals output by the high-frequency pressure sensor array and temperature sensor array arranged on the target pipeline are captured as sensing signals. Based on the timing triggering benchmark, the multi-cycle fluctuation signals within the preset time window are dynamically time-warped and matched, and the time axis is aligned with the fluctuation signal of a reference cycle to eliminate the phase shift caused by the inconsistent duration of each cycle; then, all the fluctuation signals after time axis alignment are time-domain superimposed and averaged to generate a time-aligned state matrix.

3. The method for predicting faults in an aircraft hydraulic power system based on signal analysis according to claim 1, characterized in that, The time-aligned state matrix is ​​input into a preset fluid-pipe wall coupled physical model for inverse calculation to reconstruct the three-dimensional flow field state parameters of the target pipe element, including the following steps: Initialize the fluid-pipeline coupling physical model by obtaining the inherent geometric configuration dimensions of the target pipeline and the initial fluid properties under the current operating conditions; The equivalent roughness of the pipe wall and the local convective heat transfer coefficient are defined as adjustable physical parameters. The time-aligned state matrix is ​​used as the measured boundary condition and input into the fluid-tube wall coupled physical model to perform inverse calculation of transient force and thermal response. Extract the solution distribution results of the fluid pipe wall coupled physical model, and output three-dimensional flow field state parameters including local velocity vector distribution, shear stress distribution and temperature gradient field of micro-element.

4. The method for predicting faults in an aircraft hydraulic power system based on signal analysis according to claim 1, characterized in that, Based on the current operating conditions, retrieve the baseline feature template matched in the preset baseline state feature database, perform multi-dimensional quantization comparison between the baseline feature template and the entropy production component time-series features, and generate a feature residual vector, including the following steps: Based on the current operating condition and the corresponding motor category number, retrieve the corresponding baseline feature template from the baseline state feature database; The viscous dissipative entropy production component sequence and the thermal conduction entropy production component sequence contained in the temporal features of the entropy production component are matched with the corresponding reference sequence in the baseline feature template at the same time measure length by dynamic time warping distortion distance matching calculation, and their independent nonlinear deviation time series are measured respectively. Each nonlinear deviation time series is normalized and mapped based on a preset scaling factor. Principal component analysis is then performed on each normalized time series to reduce the dimensionality of features and extract the principal component feature vectors. Vector concatenation is then performed on the extracted principal component feature vectors to calculate and generate the feature residual vector.

5. The method for predicting faults in an aircraft hydraulic power system based on signal analysis according to claim 1, characterized in that, The feature residual vector is input into a preset parameter optimization model based on gradient descent, and the adjustable physical parameters are updated until a preset convergence tolerance threshold is met to obtain the convergent physical parameters. This includes the following steps: The feature residual vector is transferred to the parameter optimization model based on gradient descent and transformed into a parameter correction vector; The extracted parameter correction vector includes the function of correcting the step size, backtracking into the fluid pipe wall coupled physical model, and superimposing the initial parameter values ​​of the adjustable physical parameters. Based on the dynamic model parameters after the modification takes effect, the closed-loop derivation operation, including the parameter update process, is repeatedly triggered until the L2 value of the feature residual vector is lower than the convergence tolerance threshold, and the locked parameters are derived into convergent physical parameters.

6. The method for predicting faults in an aircraft hydraulic power system based on signal analysis according to claim 1, characterized in that, The convergence physical parameters are cross-referenced with a pre-set historical measured component map experience database to obtain pipeline condition prediction results, including the following steps: Extract the local roughness increment and local convective heat transfer coefficient drift attenuation of the corresponding values ​​of the convergent physical parameters, and convert them into the evolution increment of the current service state of the system after merging. The evolution increment of the current service status of the system is anchored to the disaster type evolution path by performing clustering node pattern similarity calculation in the historical measured component map experience database. The data feature clusters of the actual remaining safe operating capacity percentage are calculated by weighted summation based on the built-in service margin attenuation derivation function mapped by the disaster type evolution path, and then encapsulated as pipeline status prediction results.

7. The method for predicting faults in an aircraft hydraulic power system based on signal analysis according to claim 2, characterized in that, Before performing time-domain overlay averaging and time-series alignment processing, the following steps are also included: Acquire fundamental frequency vibration data of the aircraft's hydraulic pump source; Configure the notch band parameters of the adaptive comb filter based on the fundamental frequency vibration data; The target wave signal is obtained by inputting the wave signal into an adaptive comb filter for harmonic filtering. The target fluctuation signal is used as the new fluctuation signal.

8. The method for predicting faults in an aircraft hydraulic power system based on signal analysis according to claim 1, characterized in that, After generating the temporal features of the entropy-producing components, the following steps are also included: Extract the reference phase offset angle range under historical normal operating conditions and compare it with the relative phase offset angle contained in the temporal characteristics of the entropy production component; If the relative phase offset angle deviates from the reference phase offset angle range by more than the preset outlier tolerance, it is determined that the temporal characteristics of the current entropy production component are contaminated. Trigger a resampling command, discard the current time-aligned state matrix, and return to the startup step of "obtaining the actuation command signal and sensor signal under the actuation condition, performing dynamic time axis alignment on the sensor signal to eliminate the duration deviation of each cycle, and performing time-domain superposition and averaging on the processed signal to generate a time-aligned state matrix".

9. A fault prediction system for aircraft hydraulic power systems based on signal analysis, characterized in that, Includes the following modules: The data synchronization acquisition module is used to acquire the actuation command signal and sensor signal under the actuation condition, perform dynamic time axis alignment on the sensor signal to eliminate the duration deviation of each cycle, and perform time domain superposition and averaging on the processed signal to generate a time alignment state matrix. The flow field state reconstruction module is used to input the time-aligned state matrix into the preset fluid pipe wall coupled physical model for inverse calculation, and reconstruct the three-dimensional flow field state parameters of the target pipe micro-element. The fluid pipe wall coupled physical model includes adjustable physical parameters. The entropy production feature extraction module calculates the total entropy production rate of flow field micro-elements based on three-dimensional flow field state parameters and performs physical causal orthogonal decomposition to extract temporal features of entropy production components. Specifically, it includes: fusing a preset numerical model of irreversible fluid dynamics processes, using the extracted three-dimensional flow field state parameters to calculate the total entropy production rate of the corresponding micro-elements continuously evolving within a preset time window; performing orthogonal decomposition and dimensionality reduction extraction on the total entropy production rate to separate the viscous dissipative entropy production component sequence and the thermal conduction entropy production component sequence; performing time-domain envelope curve fitting feature extraction on the viscous dissipative entropy production component sequence and the thermal conduction entropy production component sequence to obtain the peak evolution time nodes and relative phase shift angles, and combining the peak evolution time nodes, relative phase shift angles, and the viscous dissipative entropy production component sequence and the thermal conduction entropy production component sequence together to form the temporal features of entropy production components. The feature residual generation module is used to retrieve the baseline feature template matched in the preset baseline state feature database according to the current operating conditions, and compare the baseline feature template with the temporal features of the entropy production component in a multidimensional quantization to generate a feature residual vector. The parameter optimization identification module is used to input the feature residual vector into the preset parameter optimization model based on gradient descent, update the adjustable physical parameters until the preset convergence tolerance threshold is met, and obtain the convergent physical parameters. The degradation state prediction module is used to cross-compare converged physical parameters with a preset historical measured component map experience database to obtain pipeline state prediction results.

Images