An adaptive health management method and system for heavy machine tool hydraulic systems

By optimizing the sensor layout using directed graphical models and principal component analysis, and combining local state observers and slow time-varying updates, the problems of unscientific sensor layout and the inability of diagnostic models to adapt in the hydraulic systems of heavy CNC machine tools were solved, achieving efficient and accurate fault diagnosis and location.

CN123113368APending Publication Date: 2026-11-17WUHAN HEAVY MACHINE TOOL GRP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611441458.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-16
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

The sensor layout of the hydraulic system of heavy CNC machine tools lacks scientific basis, making fault tracing difficult and the diagnostic model unable to be updated adaptively, resulting in low diagnostic efficiency and high false alarm rate.

Method used

A directed graph model is used to divide functional sub-regions, and principal component analysis is combined to optimize the sensor layout. A local state observer is established to adjust the residual alarm threshold in real time. Fault tracing is performed through the directed graph topology structure to achieve slow time-varying adaptive updating of the model.

Benefits of technology

It achieves scientific sensor layout, precise fault location, and adaptive model, significantly improving diagnostic efficiency and accuracy, reducing false alarm rate, and shortening fault troubleshooting time.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123113368A_ABST
    Figure CN123113368A_ABST
Patent Text Reader

Abstract

This invention relates to an adaptive health management method and system for a heavy-duty machine tool hydraulic system. The method includes: establishing a directed graph model of the heavy-duty machine tool hydraulic system and dividing it into several functional sub-regions; performing principal component analysis (PCA) within each functional sub-region, determining the optimal combination of pressure measurement points based on the comprehensive contribution rate of each measuring point to the principal components and residual subspace, combined with cost constraints; establishing a local state observer corresponding to each functional sub-region to calculate the pressure residual in real time; using PCA to identify operating conditions in real time and dynamically adjusting the residual alarm threshold; when the residual exceeds the limit, tracing and locating the fault source based on the directed graph topology and assessing the severity, triggering a graded response. This invention achieves scientific sensor layout, deep fusion of mechanisms and data, adaptive anti-interference of operating conditions, and precise fault location, ensuring the safe and reliable operation of the heavy-duty machine tool hydraulic system throughout its entire lifecycle.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of condition monitoring and fault diagnosis technology for heavy-duty CNC machine tools, and in particular to an adaptive health management method and system for hydraulic systems of heavy-duty machine tools. Background Technology

[0002] Heavy-duty CNC machine tools are core foundational equipment in the high-end equipment manufacturing industry. Their hydraulic systems perform crucial functions such as tool cutting, hydrostatic guideway operation, lubrication, and clamping, directly determining the machine tool's machining accuracy, operational reliability, and service life. A failure in the hydraulic system will lead to equipment downtime and significant economic losses. Statistics show that approximately 70% of heavy-duty CNC machine tool failures originate from the hydraulic system, and over 80% of downtime is spent on fault diagnosis and location. Therefore, achieving efficient and healthy management of the hydraulic system is of paramount importance for ensuring safe equipment operation and reducing unplanned downtime.

[0003] Currently, the health management of heavy machine tool hydraulic systems mainly suffers from the following technical deficiencies: First, the placement of sensors lacks scientific basis. In existing technologies, the installation locations of pressure sensors are mostly determined based on engineering experience (such as pump outlet, cylinder inlet, etc.), lacking quantitative assessment of the value of the information from the measurement points. This leads to both redundant measurement points resulting in wasted costs and missing fault information from critical components. Existing research has used Principal Component Analysis (PCA) for dimensionality reduction of hydraulic system data, but only as a preprocessing step for model training, without using the PCA contribution rate to guide sensor layout optimization.

[0004] Secondly, the diagnostic logic is disconnected from the physical mechanism. Existing diagnostic methods either rely solely on data-driven approaches (such as neural networks), lacking the utilization of causal relationships between hydraulic system components; or they rely solely on fixed threshold alarms, which cannot adapt to the characteristics of drastic load changes during heavy machine tool processing. Although some studies have proposed devices and methods for monitoring the condition of hydraulic systems, using deep neural networks combined with Dempster-Shafer evidence theory for fault diagnosis, this method does not establish a causal relationship model between system components, resulting in insufficient fault tracing capabilities.

[0005] Third, the diagnostic model cannot be adaptively updated as the equipment ages. Existing technologies typically complete model training offline and then put it into online operation. However, after long-term operation of the hydraulic system, factors such as oil aging and component wear can cause the original model to become mismatched, resulting in a continuous increase in false alarm and false negative rates.

[0006] Therefore, there is an urgent need for a health management method for heavy machine tool hydraulic systems that can integrate physical mechanisms and data analysis, achieve scientific sensor layout, accurate fault location, and adaptive modeling. Summary of the Invention

[0007] This invention provides an adaptive health management method and system for a heavy machine tool hydraulic system, which solves the technical problems of lack of basis for sensor layout, difficulty in fault tracing, and inability of model to be adaptively updated in the prior art.

[0008] In a first aspect of the present invention, an adaptive health management method for a heavy-duty machine tool hydraulic system is provided, comprising: A directed graph model of the hydraulic system of a heavy machine tool is established, and the hydraulic system is divided into several functional sub-regions based on the directed graph model; Principal component analysis is performed in each of the functional sub-regions. Based on the comprehensive contribution rate of each measurement point to the principal component and residual subspace, and combined with the preset cost constraints, the optimal combination of pressure measurement points is determined. Establish a local state observer corresponding to each of the functional sub-regions, input the pressure signal collected by the optimal pressure measurement point combination into the corresponding local state observer, and calculate the pressure residual output by each local state observer in real time. Principal component analysis is used to identify the current operating mode in real time, and the residual alarm threshold of each local state observer is dynamically adjusted according to the current operating mode. When the pressure residual of any functional sub-region exceeds the corresponding residual alarm threshold, the fault is traced and located based on the topology of the directed graph model, and the severity of the fault is assessed based on the fault tracing results, triggering the corresponding graded response.

[0009] In a second aspect of the present invention, an adaptive health management system for a heavy-duty machine tool hydraulic system is provided, comprising: The directed graph modeling module is used to build a directed graph model of the hydraulic system of heavy machine tools, and divide the hydraulic system into several functional sub-regions based on the directed graph model. The measurement point optimization module is used to perform principal component analysis in each of the functional sub-regions, and determine the optimal combination of pressure measurement points based on the comprehensive contribution rate of each measurement point to the principal component and the residual subspace, combined with the preset cost constraints. The state observation module is used to establish a local state observer corresponding to each of the functional sub-regions, input the pressure signal collected by the optimal pressure measurement point combination into the corresponding local state observer, and calculate the pressure residual output by each local state observer in real time. The threshold adaptive module is used to identify the current operating mode in real time using principal component analysis and dynamically adjust the residual alarm threshold of each local state observer according to the current operating mode. The fault diagnosis and response module is used to trace and locate the fault based on the topology of the directed graph model when the pressure residual of any functional sub-zone exceeds the corresponding residual alarm threshold, and to assess the severity of the fault based on the fault tracing results and trigger the corresponding graded response. The model update module is used to collect operational data with residuals not exceeding the limit and confirmed to be fault-free by manual review at a preset period as update data. The update data is used to update the mean vector and covariance matrix of the principal component analysis model, and the physical parameters of each local state observer are fine-tuned under the physical constraints of the directed graph model. The updated model and observer parameters are then put into online operation.

[0010] In a third aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor executes the computer program to implement the steps of the method as described in the first aspect of the present invention.

[0011] In a fourth aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the method as described in the first aspect of the present invention.

[0012] The adaptive health management method and system for heavy-duty machine tool hydraulic systems provided by this invention have the following advantages compared with the prior art: (1) Scientific sensor layout. The variance contribution rate and residual subspace contribution rate of principal component analysis are introduced into the sensor layout optimization. The optimal combination of pressure measurement points is selected with the goal of maximizing the comprehensive contribution rate and minimizing information redundancy. This overcomes the subjectivity and blindness of traditional experience-based point selection and reduces the number of sensors while ensuring the integrity of diagnostic information.

[0013] (2) Deep integration of mechanism and data. The directed graph model provides physical constraints and causal propagation paths, enabling the selection of principal component analysis measurement points to cover all functional sub-regions, the design of local state observers to reflect the physical characteristics of subsystems, and the fault tracing to proceed along the directed edges, forming a collaborative mechanism in which the physical model guides data analysis and the data analysis verifies the physical model.

[0014] (3) Adaptive anti-interference under working conditions. The current working condition mode is identified in real time through principal component analysis, and the residual alarm threshold is dynamically adjusted according to the current working condition mode, which effectively solves the problem of frequent false alarms of fixed threshold under the variable load condition of heavy machine tools.

[0015] (4) Self-maintaining model throughout its entire lifecycle. Through a slow time-varying adaptive update mechanism, the diagnostic model ages synchronously with the equipment, avoiding model mismatch problems caused by oil aging and component wear.

[0016] (5) Precise fault location. By combining the local state observer residual over-limit with the directed graph topological order for source reasoning, the fault location is accurate to the component level, which greatly shortens the fault troubleshooting time. Attached Figure Description

[0017] Figure 1 Flowchart of the adaptive health management method for heavy machine tool hydraulic systems provided by the present invention; Figure 2 An example of a directed graph for a hydraulic system provided by the present invention; Figure 3 This is a schematic diagram illustrating the principle of residual calculation for the local observer provided by the present invention. Figure 4 A flowchart for fault propagation and tracing based on directed graph topology provided by this invention; Figure 5 The flowchart of slow time-varying adaptive update provided by the present invention; Figure 6 A structural block diagram of the adaptive health management system for a heavy-duty machine tool hydraulic system provided by the present invention; Figure 7 This is a structural block diagram of the electronic device provided by the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. 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.

[0019] Figure 1 A flowchart illustrating the adaptive health management method for a heavy-duty machine tool hydraulic system provided by this invention. (Refer to...) Figure 1 The method includes the following steps: Step S1: Establish a directed graph model of the hydraulic system of the heavy machine tool, and divide the hydraulic system into several functional sub-regions based on the directed graph model.

[0020] Specifically, the directed graph model is a topological structure model established using the components of the hydraulic system as nodes, the hydraulic oil flow path and energy transfer direction as directed edges, and the pressure-flow coupling relationship between components as edge weights. It is used to characterize the system's physical structure and energy transfer path. Based on this directed graph model, the hydraulic system is divided into several functional sub-regions according to their functional types. The directed graph model provides candidate measurement point spatial constraints for subsequent sensor layout optimization, provides a basis for subsystem partitioning for observer design, and provides causal propagation paths for fault tracing.

[0021] Step S2: Perform principal component analysis in each of the functional sub-regions. Based on the comprehensive contribution rate of each measurement point to the principal component and residual subspace, and combined with the preset cost constraints, determine the optimal combination of pressure measurement points.

[0022] Principal component analysis (PCA) is a multivariate statistical analysis method that achieves dimensionality reduction and feature extraction by projecting the original high-dimensional data onto low-dimensional principal component subspaces and residual subspaces. The combined contribution rate of each measurement point to the principal components and residual subspaces is used to quantitatively evaluate the ability of that measurement point to represent the system's operational status information.

[0023] Specifically, within each functional sub-region, historical pressure data of candidate measuring points under various operating conditions are collected, principal component analysis is performed, the comprehensive contribution rate of each candidate measuring point is calculated, and under the condition of meeting cost constraints, the optimal combination of pressure measuring points that can fully reflect the operating status of the sub-region is selected as the pressure collection points.

[0024] Step S3: Establish a local state observer corresponding to each of the functional sub-regions, input the pressure signal collected by the optimal pressure measurement point combination into the corresponding local state observer, and calculate the pressure residual output by each local state observer in real time.

[0025] The local state observer is a state estimator built upon the pressure-flow dynamic mathematical model of each functional sub-region, used to estimate the internal state of the system based on system inputs and observed outputs. The pressure residual is the difference between the measured pressure value and the pressure estimate output by the observer.

[0026] In this embodiment, a local state observer is built for each functional sub-region. The real-time pressure signal collected by the optimal pressure measuring point is input into the corresponding observer. Through state estimation calculation, the difference between the measured pressure and the estimated pressure is obtained to obtain the real-time pressure residual, which is used for subsequent anomaly detection.

[0027] Step S4: Use principal component analysis to identify the current operating mode in real time, and dynamically adjust the residual alarm threshold of each local state observer according to the current operating mode.

[0028] Specifically, the operating mode characterizes the operating status of the heavy machine tool hydraulic system under different loads; the residual alarm threshold is the criterion for judging whether the pressure residual is abnormal. In this embodiment, principal component analysis is used to spatially project the real-time pressure data to identify the current operating mode of the hydraulic system, and the residual alarm threshold is dynamically updated according to the identified operating mode to adapt to the pressure fluctuation characteristics under different loads.

[0029] Step S5: When the pressure residual of any functional sub-region exceeds the corresponding residual alarm threshold, the fault is traced and located based on the topology of the directed graph model, and the severity of the fault is assessed based on the fault tracing results, triggering the corresponding graded response.

[0030] Specifically, when the pressure residual output of a certain functional sub-region exceeds the residual alarm threshold under the current operating conditions, an anomaly is determined; fault source tracing and location are carried out based on the causal relationship of the topology oil circuit in the pre-established directed graph model to determine the location of the fault; the severity of the fault is assessed based on the fault information obtained from the source tracing, and corresponding graded response measures are implemented according to the severity.

[0031] The adaptive health management method for heavy-duty machine tool hydraulic systems provided in this invention deeply integrates physical mechanisms with data analysis. It optimizes sensor layout through principal component analysis contribution rate, overcoming the subjectivity of traditional experience-based point selection and effectively reducing the number of sensors. It utilizes a directed graph model to provide physical constraints and causal propagation paths, enabling observer design to reflect the physical characteristics of the subsystem and fault tracing to proceed along directed edges, significantly improving diagnostic reliability. Through condition-adaptive threshold modulation, it solves the problem of frequent false alarms with fixed thresholds under varying load conditions. Finally, it achieves precise fault location from the system level to the component level through directed graph topology tracing, greatly shortening fault troubleshooting time.

[0032] Based on the above embodiments, step S1, which involves establishing a directed graph model of the heavy machine tool hydraulic system and dividing the hydraulic system into several functional sub-regions based on the directed graph model, specifically includes: The power components, control components, actuators, and key pipeline nodes of the hydraulic system are represented as nodes in a directed graph; the hydraulic oil flow path and energy transfer direction are represented as edges in the directed graph; and the pressure between components is... The flow coupling relationship model or hydraulic admittance is used as the weight of the corresponding directed edge; based on the directed graph model, the hydraulic system is divided into the power source area, the control valve area and the actuator area according to the functional type.

[0033] Among them, the power element refers to the power device that provides pressurized oil to the hydraulic system; the control element refers to the device used to control and regulate the oil pressure, flow rate, and flow direction; the actuator refers to the device that converts hydraulic energy into mechanical energy; and the critical pipeline node refers to the connection point in the hydraulic pipeline where the pressure or flow rate changes significantly. The pressure... The flow coupling relationship model is used to characterize the dynamic mapping relationship between input and output signals between adjacent components; the hydraulic admittance is used to characterize the impedance characteristics of the component to pressure fluctuations.

[0034] Figure 2 An example of a directed graph for a hydraulic system provided by this invention. For example... Figure 2 As shown, in this embodiment, for the hydraulic system of a heavy machine tool, its key components are treated as nodes in a directed graph, specifically including: Node 1: Variable displacement piston pump; Node 2: Pilot-operated relief valve; Node 3: Three-position four-way solenoid directional valve; Node 4: One-way throttle valve; Node 5: Double-rod hydraulic cylinder (worktable cutter mechanism); Node 6: Accumulator; Node 7: Return oil filter.

[0035] In the directed graph model, the direction of the directed edges represents the hydraulic oil flow direction, specifically: 1→2→3→4→5. Simultaneously, there are oil circuits connecting nodes 3 and 6, and nodes 5 and 7. The weight of the directed edges is determined by the pressure-flow transmission relationship between the components. For example, the edge weight from node 1 (variable displacement piston pump) to node 2 (pilot-operated relief valve) is jointly determined by the pump's flow-pressure characteristic curve and the relief valve's opening pressure.

[0036] Based on the above directed graph model, the hydraulic system is divided into three functional sub-regions according to functional type: Sub-region A (power source region): includes node 1 (variable piston pump) and node 2 (pilot-operated relief valve); Sub-area B (control valve area): includes node 3 (three-position four-way solenoid directional valve) and node 4 (one-way throttle valve). Sub-area C (actuator area): includes node 5 (double rod hydraulic cylinder), node 6 (accumulator) and node 7 (return oil filter).

[0037] At the same time, candidate stress measurement points (such as...) are set at key locations in each sub-region. Figure 2 The points P1, P3, P5, P8, P9, P11, and P12 in the table provide a constraint space for the subsequent selection of the optimal combination of pressure measurement points.

[0038] This embodiment achieves a digital topological representation of the physical structure by establishing a directed graph model and dividing it into functional sub-regions. This model not only provides spatial constraints for sensor layout to ensure full coverage and provides physical boundaries for the independent design of local state observers, but also provides accurate causal propagation paths for subsequent fault tracing, improving the accuracy and efficiency of health management and location.

[0039] Based on the above embodiments, in step S2, principal component analysis is performed in each of the functional sub-regions. The optimal combination of pressure measurement points is determined based on the comprehensive contribution rate of each measurement point to the principal components and residual subspace, combined with preset cost constraints. Specifically, this includes the following sub-steps S21-S24: S21, set up candidate pressure measurement points in each of the functional sub-regions and collect historical pressure data under various working conditions.

[0040] Specifically, 12 candidate pressure measurement points (P1~P12) are set up in each functional sub-region, and their specific distribution is as follows: Sub-area A: P1 (pump outlet), P2 (pump outlet after filtering), P3 (relief valve inlet); Sub-area B: P4 (reversing valve P port), P5 (reversing valve A port), P6 (reversing valve B port), P7 (before the throttle valve), P8 (after the throttle valve); Sub-area C: P9 (rodless chamber of hydraulic cylinder), P10 (rod chamber of hydraulic cylinder), P11 (accumulator interface), P12 (before the return oil filter).

[0041] Pressure sensors were installed at the 12 candidate measurement points, and pressure data were collected under three working conditions: no load, light load, and heavy load. 1,000 samples were collected for each working condition, resulting in a total of 3,000 sets of historical pressure data.

[0042] S22, standardize the historical pressure data of each functional sub-region, calculate the covariance matrix and solve for the corresponding eigenvalues ​​and eigenvectors, and determine the number of principal components based on the preset threshold of the cumulative variance contribution rate.

[0043] Specifically, taking sub-region A as an example (containing 3 candidate measurement points P1~P3, and a data matrix of 3000×3), principal component analysis is performed: First, the data is standardized using the following formula:

[0044] In the formula, X std For standardized data, μ The mean, σ Standard deviation, X This is the original historical pressure data.

[0045] Next, the covariance matrix is ​​calculated using the following formula:

[0046] In the formula, is the transpose of the standardized data, and n is the total number of samples (n=3000 in this example). Let be the covariance matrix.

[0047] Then, we solve for the eigenvalues ​​and corresponding eigenvectors of the covariance matrix. Let the eigenvalues ​​be arranged in descending order as follows: The number of principal components is determined based on a preset threshold for the cumulative variance contribution rate (set to 85% in this embodiment). k The calculation formula is as follows:

[0048] In the formula, k The number of principal components obtained through screening. Let be the i-th feature value after sorting. Select the top eigenvalues ​​that meet the cumulative variance contribution rate requirement. k Each component is used as an effective principal component.

[0049] S23, calculate the variance contribution rate of each candidate measurement point to the principal component and the contribution rate to the residual subspace respectively; based on the variance contribution rate and residual subspace contribution rate of each candidate measurement point, obtain the comprehensive contribution rate of each candidate measurement point.

[0050] Specifically, the contribution rate of a measurement point to the variance of a single principal component is calculated by multiplying the square of the eigenvector component corresponding to that measurement point by the corresponding eigenvalue. The contribution rate of a measurement point to the residual subspace (i.e., the Q statistic) is the sum of the squares of the projected components of the measurement point's data in the residual subspace. The Q statistic is used to characterize the magnitude of residual fluctuations not explained by the principal component model.

[0051] The overall contribution rate of each candidate measurement point is calculated using a weighted fusion method. The formula for calculating the overall contribution rate is as follows: Overall contribution rate = Principal component variance contribution rate × + Q statistic contribution rate ×

[0052] In the formula, The first preset weight corresponding to the variance contribution rate. The second preset weight corresponding to the contribution rate of the residual subspace is, in this embodiment... The value is 0.6. The value is 0.4.

[0053] After calculation and sorting, in sub-region A, P1 has the highest contribution rate (0.42), followed by P3 (0.31), and P2 has the lowest (0.27); in sub-region B, P5 has the highest contribution rate, followed by P8; in sub-region C, P9 has the highest contribution rate, followed by P11.

[0054] S24. With the optimization objective of maximizing the total contribution rate and minimizing the information redundancy between measurement points, and taking into account the cost constraint of the number of sensors, an optimization algorithm is used to solve the optimal combination of pressure measurement points in each functional sub-region.

[0055] In this embodiment, the cost constraint is that each sub-region must retain at least 2 measuring points, and the total number of measuring points must not exceed 7. A greedy algorithm is used to solve the problem with the goal of maximizing the total comprehensive contribution rate. The optimized result is: retaining measuring points P1, P3 (sub-region A), P5, P8 (sub-region B), and P9, P11 (sub-region C), for a total of 6 measuring points. This combination of measuring points reduces the number of candidate measuring points by 50% compared to the original 12, and covers all functional sub-regions. The optimized combination of measuring points serves as the sensor layout scheme for subsequent online monitoring.

[0056] This embodiment scientifically solves the optimal combination of pressure measurement points based on PCA contribution rate and cost constraints, overcoming the subjectivity of traditional manual experience in point selection and effectively reducing the number of sensors and deployment costs. At the same time, the measurement points cover all functional sub-regions and minimize information redundancy, ensuring a comprehensive and high-fidelity characterization of the operating status of each region, laying a data foundation for subsequent high-precision status observation.

[0057] Based on the above embodiments, in step S3, a local state observer corresponding to each of the functional sub-regions is established. The pressure signal collected by the optimal pressure measurement point combination is input into the corresponding local state observer, and the pressure residual output by each local state observer is calculated in real time. Specifically, this includes the following sub-steps S31 to S33: S31. Based on the pressure-flow dynamic mathematical model of each functional sub-region, establish the state space model corresponding to each functional sub-region.

[0058] Figure 3 This is a schematic diagram illustrating the principle of residual calculation for the local observer provided by the present invention. Figure 3 As shown, local state observers are established for each functional sub-region. For the divided power source region, control valve region, and actuator region, pressure is established by matching the dynamic characteristics of the corresponding components. A dynamic mathematical model of flow is used to construct an independent state-space model. The following detailed modeling explanation uses the actuator region (sub-region C) as an example. The core state equations are constructed based on the hydraulic cylinder flow continuity equation and force balance equation. The specific formulas are as follows:

[0059]

[0060] In the formula, For load traffic, For the effective area of ​​the piston, x For piston displacement, The internal leakage coefficient is... This is the total volume of the cavity. For effective bulk modulus, The load pressure of the hydraulic cylinder. m For load quality, The viscous damping coefficient is... K For the load spring stiffness, This is the external load force.

[0061] Based on the above dynamic equations, select the state vector. Input vector Establish a continuous state-space model:

[0062]

[0063] In the formula, y is the observable output, i.e., the pressure measured by the pressure sensor. p L , The first derivative of the state vector with respect to time. The speed of the hydraulic cylinder piston (the first derivative of displacement with respect to time); For state vectors, x For the displacement of the hydraulic cylinder piston, For the input vector, A The state matrix is ​​obtained by rearranging the hydraulic cylinder dynamics equations and represents the internal state coupling relationship of the system. B The input matrix represents the relationship between the input quantities and the system state; C The observation output matrix is ​​used to map the system state to the observable output. D The feedforward matrix represents the coefficients through which the input vector is directly fed into the output vector.

[0064] S32, Based on the state-space model, a Kalman filter is designed as the local state observer. The real-time pressure signal collected by the optimal pressure measurement point combination is used as the observation value and input into the Kalman filter for state estimation to obtain the pressure estimate value.

[0065] First, the continuous state-space model of each functional sub-region is discretized to obtain a discrete state-space model adapted for online operation, as shown in the following formula:

[0066]

[0067] In the formula, x k for k The state vector at any given time; y k for k Pressure observations at any given time; A d The discrete state transition matrix, B d For discrete control input matrix, C d It is a discrete observation matrix; w k-1 for k Process noise at time -1 v k for k Measurement noise at time, k This represents the current sampling time.

[0068] Based on the discrete state-space model, a local state observer is designed using the standard Kalman filter recursive algorithm. The iterative recursive formula is as follows: (1) State prediction:

[0069] (2) Error covariance prediction:

[0070] (3) Kalman gain:

[0071] (4) Status update:

[0072] (5) Error covariance update:

[0073] In the above formula, for k Prior state estimate at time step Let be the posterior state estimate at time k. Let be the prior error covariance matrix. Let be the posterior error covariance matrix. for k Kalman gain at time step; I It is an identity matrix.

[0074] During online operation, the pressure signals collected in real time from the optimal pressure measurement points corresponding to each functional sub-region are used as observation values ​​and input into the Kalman filter of the corresponding sub-region. Through the above recursive iterative calculation, the pressure estimate of each sub-region is output in real time. The local state observers of the power source region, control valve region, and actuator region operate synchronously and in parallel, independently completing the state estimation of each sub-region.

[0075] S33, Residual Calculation.

[0076] At any moment k The real-time pressure signal collected by the optimal combination of pressure measuring points is used as the observation value. y k The pressure estimate is obtained by inputting the Kalman filter for state estimation. The actual measured pressure value y k The difference between the pressure estimate and the pressure residual is used as the pressure residual. r k :

[0077] The observer design methods for sub-regions A and B are similar, establishing state-space models based on the pump's flow-pressure characteristic equation and the valve's throttling characteristic equation, respectively. During online operation, the three local state observers perform parallel calculations, outputting the pressure residual sequence for each sub-region.

[0078] This embodiment establishes a local state observer and introduces the Kalman filter algorithm to deeply integrate the physical mechanism model with the measured data, thereby achieving accurate calculation of the pressure residuals of each functional sub-region. By using parallel collaborative calculation of multiple sub-region observers, it effectively overcomes the problems of high dimensionality and computational complexity of a single global observer, and improves the real-time performance and reliability of state estimation.

[0079] Based on the above embodiments, step S4, which involves using principal component analysis to identify the current operating mode in real time and dynamically adjusting the residual alarm thresholds of each local state observer according to the current operating mode, specifically includes the following sub-steps S41~S43: S41, clustering is performed using the score vectors of historical pressure data under multiple operating conditions in the principal component analysis subspace to obtain the cluster centers corresponding to each operating condition mode and establish an operating condition mode library; at the same time, statistical analysis is performed on the residual data under each operating condition mode to obtain the basic alarm thresholds corresponding to each operating condition mode.

[0080] Specifically, using the historical pressure data under multiple operating conditions (including no-load, light-load, and heavy-load conditions) collected in step S2, the score vector of each sample in the principal component analysis subspace is calculated using the following formula:

[0081] In the formula, t The sample principal component score vector is the dimensionality-reduced feature vector obtained by mapping the original high-dimensional data onto the principal component loading matrix. X std For standardized historical stress data, W k For the front k The loading matrix corresponding to each effective principal component.

[0082] The score vectors of all samples are clustered using the K-means clustering algorithm. In this embodiment, the number of clusters is set to 3, corresponding one-to-one with three typical operating conditions. After clustering, three cluster centers are obtained, corresponding to the idle, light load, and heavy load operating conditions, respectively. A complete operating condition pattern library is constructed based on the three cluster centers to achieve standardized storage of operating condition features.

[0083] Statistical analysis was performed on the pressure residual data corresponding to each operating condition, and the mean value of the residual data for each operating condition was calculated. and standard deviation ,by This serves as the basic alarm threshold for the corresponding operating condition.

[0084] S42, during online operation, real-time pressure data is projected onto the principal component analysis subspace to obtain a real-time score vector, the distance between the real-time score vector and each cluster center is calculated, and the current working condition mode is identified.

[0085] Specifically, pressure data from each optimal combination of pressure measuring points is collected in real time. X new The sample is standardized according to the standardized parameters in step S2 to obtain the result. X new,std The vector is then projected onto the principal component analysis subspace to calculate the real-time score vector. .

[0086] Calculate the real-time score vector t new The Euclidean distance between the current working condition and the three cluster centers in the working condition pattern library is used to select the working condition pattern corresponding to the smallest Euclidean distance and identify it as the current working condition pattern.

[0087] S43, retrieve the corresponding basic alarm threshold according to the current operating mode, and dynamically compensate the basic alarm threshold in combination with the physical parameters of the components in the directed graph model to obtain the adjusted residual alarm threshold.

[0088] Specifically, based on the identified current operating condition mode, the corresponding basic alarm threshold is retrieved from the operating condition mode library. T baseBy combining the physical parameters in the directed graphical model with dynamic compensation, the final residual alarm threshold is obtained. T final The calculation formula is:

[0089] in, T comp This is a dynamic compensation amount. T comp The parameters are calculated based on the physical parameters of key components such as the energy storage capacity and pipeline cavity in the directed graph model, and are used to eliminate interference caused by normal load fluctuations.

[0090] In this embodiment, the specific alarm threshold after dynamic modulation is set as follows: the residual alarm threshold under no-load conditions is set to ±0.5MPa, under light-load conditions to ±0.8MPa, and under heavy-load conditions to ±1.2MPa, which effectively avoids false alarms under heavy-load conditions.

[0091] This embodiment achieves high-precision real-time identification of operating conditions by establishing a working condition pattern library and using a score vector distance metric. By combining directed graph physical parameters with dynamic modulation and compensation of the alarm threshold, the problem of frequent false alarms or missed alarms caused by fixed thresholds under variable load conditions is solved, significantly improving the adaptability and robustness of anomaly detection under different operating states.

[0092] Based on the above embodiments, in step S5, when the pressure residual of any functional sub-region exceeds the corresponding residual alarm threshold, the fault is traced and located based on the topology of the directed graph model, and the severity of the fault is assessed based on the fault tracing results, triggering a corresponding graded response. This specifically includes the following sub-steps S51 to S54: S51, when the pressure residual of any of the functional sub-regions exceeds the corresponding residual alarm threshold and continues to exceed the preset sampling period, the fault diagnosis process is triggered.

[0093] Figure 4 This invention provides a flowchart for fault propagation and tracing based on directed graph topology. (See attached flowchart.) Figure 4 As shown, during online monitoring, the pressure residuals output by the local state observers of each functional sub-region are acquired in real time. If the absolute value of the pressure residual in any functional sub-region exceeds the dynamically adjusted residual alarm threshold under that operating condition, and this over-limit state continues for more than the preset sampling period, the system is determined to be abnormal, and a fault diagnosis process is immediately triggered to eliminate misjudgments caused by transient interference. In this embodiment, the preset sampling period is set to 3 sampling periods.

[0094] S52, according to the upstream and downstream topology order of the directed graph model, check the residual contribution rate of the local state observer corresponding to each functional sub-region in turn, and determine the upstream node that first shows residual exceeding the limit and has the largest residual contribution rate as the root cause node of the fault.

[0095] like Figure 4 As shown, fault tracing follows the logical path of "observer residual exceeding limit trigger → assessing the residual status of each sub-region → determining the first sub-region to exceed the limit → locating the fault root cause sub-region → calculating the residual contribution rate of the measurement points within the sub-region → locating the specific component node". Its core tracing logic is that the fault propagates from top to bottom along the directed graph (i.e., in the direction of A→B→C), and the upstream node where the residual exceeds the limit first is the fault root cause.

[0096] Specifically, following the hydraulic oil flow path and energy transfer direction defined by the directed graph model (i.e., upstream and downstream topological order), the residual states of the local state observers corresponding to the power source region (sub-region A), control valve region (sub-region B), and actuator region (sub-region C) are checked sequentially. Global residual assessment and sub-region location: If only the residual of sub-region C exceeds the limit, while sub-regions A and B are normal, then the fault is determined to originate from inside the actuator area (such as leakage inside the cylinder or failure of the accumulator), and the root cause of the fault is locked in node 5 or node 6 within sub-region C.

[0097] If the residuals of both sub-regions B and C exceed the limit, and sub-region B exceeds the limit before sub-region C, based on the directed graph topology (A→B→C), it is determined that the fault propagates from upstream to downstream, and the root cause is located in sub-region B (control valve region). The root cause of the fault is locked at node 3 or node 4.

[0098] If the residuals of all three sub-regions exceed the limit, and sub-region A exceeds the limit first, then the root cause is determined to be in sub-region A (power source region), and the root cause of the fault is locked at node 1 (such as pump wear) or node 2 (such as relief valve spring failure).

[0099] Precise node localization within a sub-region: Once the root cause of the fault is pinpointed to a specific sub-region, the contribution rate of the residual components corresponding to each measuring point within that sub-region is further analyzed. Taking sub-region A as an example, if the contribution rate of the residual corresponding to P1 accounts for more than 65% of the total residual of sub-region A, the fault is further pinpointed to the pump outlet location. Combining the frequency characteristics of the residuals (such as plunger wear typically manifesting as periodic pressure pulsations related to pump speed), the specific faulty component node and fault type are ultimately determined.

[0100] S53, calculate the fault severity index based on the residual excess amplitude of the fault root cause node and the preset node weight coefficient.

[0101] Specifically, after identifying the root cause node of the fault, the fault severity index is calculated according to the following formula. S :

[0102] In the formula, |r| is the absolute value of the current residual. The dynamic alarm threshold under the current operating conditions. These are node weight coefficients, with key core components having higher weight coefficients, used to differentiate the impact of failures at different nodes.

[0103] S54, determine the severity level of the fault based on the fault severity index, and trigger the graded response measures corresponding to the severity level of the fault.

[0104] Specifically, based on the calculated fault severity index S The faults are classified into three levels and corresponding response measures are triggered: Level 1 Response (Early Warning): When 1.0 ≤ S A reading less than 1.5 indicates a slight degradation in system performance. In this case, the human-machine interface should prompt the operator to pay closer attention and conduct regular inspections, without shutting down the system temporarily.

[0105] Level 2 response (alarm): When 1.5 ≤ S A value less than 2.5 indicates a significant degradation in system performance. At this point, the system will issue an alarm, recommending maintenance after the current processing task is completed, and providing a recommended maintenance plan and spare parts list.

[0106] Level 3 Response (Emergency Shutdown): When S When a transient pressure drop logic is triggered at a value ≥2.5 or at a critical safety node (such as an accumulator or high-pressure pipeline), it indicates that the fault seriously endangers equipment safety. In this case, the system immediately triggers emergency shutdown protection and automatically pushes fault information to the remote maintenance center to avoid the risk of equipment damage.

[0107] The final output includes the fault node number, fault type, diagnostic confidence level, fault severity level, and personalized maintenance recommendations, completing the entire fault health management process.

[0108] This embodiment combines the observer residual out-of-limit nodes with the directed graph topological order for source tracing reasoning, which can pinpoint fault location from the system level to the component level, greatly shortening the fault investigation time. By introducing a fault severity index and a three-level graded response mechanism, differentiated handling from early warning to emergency shutdown is achieved, effectively balancing production continuity and equipment safety.

[0109] Based on the above embodiments, after triggering the corresponding graded response in step S5, the method further includes: Step S6, Slow Time-Varying Adaptive Update of the Model: Collect fault-free operation data according to a preset cycle, update the principal component analysis model and fine-tune the physical parameters of the local state observer, so as to realize the diagnostic model aging synchronously with the equipment.

[0110] Among them, slow time-varying adaptive update refers to the mechanism of gradually correcting the diagnostic model parameters by periodically collecting health data in response to the performance degradation (such as oil aging, increased clearance) caused by long-term operation of hydraulic systems; the physical constraint refers to the limitation of parameter variation range set based on the physical characteristics of components in the directed graph model, which is used to prevent the observer from becoming unstable due to sudden changes in model parameters.

[0111] Figure 5 This is a flowchart illustrating the slow time-varying adaptive update process provided by the present invention. Figure 5 As shown, the update process follows the logical path of "starting the timer → collecting recent data → confirming health status → updating PCA model parameters → fine-tuning observer physical parameters → putting the new parameters into operation". Step S6 specifically includes the following sub-steps S61~S64: S61 collects operating data according to a preset cycle, and uses the operating data that has been determined by the pressure residual diagnosis process to be within the limit and has been manually verified to be in a fault-free state as the updated data.

[0112] Specifically, a slow time-varying update cycle is set at 30 days. Within each update cycle, a data acquisition program is started to collect no fewer than 500 sets of recent operational data.

[0113] Next, the health status of the collected data is verified using the fault diagnosis process described in steps S3 to S5. Only when the pressure residuals output by the local state observers of all functional sub-regions do not exceed the corresponding residual alarm thresholds, and the diagnostic module does not trigger any alarm signals, is the data segment determined to be in a fault-free and healthy state. Simultaneously, the validity of the data is further confirmed by combining the results of manual review, and the operational data confirmed to be in a fault-free state is used as the baseline data (i.e., update data) for subsequent model updates.

[0114] S62, update the mean vector and covariance matrix of the principal component analysis model using the updated data.

[0115] Specifically, using the selected fault-free update data, the statistical parameters of the principal component analysis (PCA) model are recalculated and updated to match the system's current operating baseline. The update formula is as follows:

[0116]

[0117] In the formula, u new This is the updated mean vector of the PCA model. The updated covariance matrix, n The total number of samples, X i For the firsti Group update data sample.

[0118] By updating the mean vector u new Covariance Matrix This enables the PCA model to reflect the current normal state distribution of the equipment.

[0119] S63, under the physical constraints of the directed graph model, the physical parameters of each local state observer are fine-tuned using the least squares estimation method.

[0120] Specifically, in response to changes in the physical characteristics of the hydraulic system due to wear, the updated data is used to fine-tune key physical parameters in the local state observers of each functional sub-region. These physical parameters include the internal leakage coefficient. C ip Viscous damping coefficient B c wait.

[0121] The least squares estimation method is used to solve for the optimal parameter values, minimizing the error between the observer output and the actual measurement values. Simultaneously, a physical constraint is imposed on the directed graph model: the amount of parameter update in a single step cannot exceed 5% of the original parameter value. This constraint ensures that parameter adjustments are gradual, preventing problems such as observation instability and residual aberration caused by large fluctuations in observer parameters, and guaranteeing the continuity and stability of the observer's operation.

[0122] S64, put the updated principal component analysis model and local state observer parameters into the next cycle of online operation.

[0123] Specifically, after updating the PCA model parameters and observer physical parameters as described above, the new model parameters immediately replace the old parameters, and the system is put into operation for the next online monitoring cycle. At this point, it returns to the timed waiting state until the next 30-day update cycle arrives.

[0124] This embodiment uses a slow time-varying adaptive update mechanism to enable the diagnostic model to age synchronously with the gradual degradation of equipment performance, always maintaining the best match between the principal component analysis model and the local state observer and the actual system. This effectively avoids model mismatch problems caused by oil aging and component wear, allowing the system to maintain stable diagnostic performance throughout the entire equipment lifecycle and significantly reducing false alarm and false negative rates.

[0125] Figure 6 The structural block diagram of the adaptive health management system for the heavy machine tool hydraulic system provided by the present invention is shown below. Figure 6 The system includes: The directed graph modeling module is used to build a directed graph model of the hydraulic system of heavy machine tools, and divide the hydraulic system into several functional sub-regions based on the directed graph model. The measurement point optimization module is used to perform principal component analysis in each of the functional sub-regions, and determine the optimal combination of pressure measurement points based on the comprehensive contribution rate of each measurement point to the principal component and the residual subspace, combined with the preset cost constraints. The state observation module is used to establish a local state observer corresponding to each of the functional sub-regions, input the pressure signal collected by the optimal pressure measurement point combination into the corresponding local state observer, and calculate the pressure residual output by each local state observer in real time. The threshold adaptive module is used to identify the current operating mode in real time using principal component analysis and dynamically adjust the residual alarm threshold of each local state observer according to the current operating mode. The fault diagnosis and response module is used to trace and locate the fault based on the topology of the directed graph model when the pressure residual of any functional sub-zone exceeds the corresponding residual alarm threshold, and to assess the severity of the fault based on the fault tracing results and trigger the corresponding graded response. The model update module is used to collect operational data with residuals not exceeding the limit and confirmed to be fault-free by manual review at a preset period as update data. The update data is used to update the mean vector and covariance matrix of the principal component analysis model, and the physical parameters of each local state observer are fine-tuned under the physical constraints of the directed graph model. The updated model and observer parameters are then put into online operation.

[0126] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the above system and modules can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0127] Figure 7 A structural block diagram of the electronic device provided by the present invention, such as Figure 7 As shown, the present invention also provides an electronic device 700, which can be a mobile terminal, desktop computer, laptop, handheld computer, server, or other computing device. The electronic device 700 includes a processor 701 and a memory 702, wherein the memory 702 stores an adaptive health management program 703 for a heavy-duty machine tool hydraulic system. In one embodiment, when the processor 701 executes the adaptive health management program 703 for the heavy-duty machine tool hydraulic system, it implements the steps of an adaptive health management method for the heavy-duty machine tool hydraulic system.

[0128] This embodiment also provides a computer-readable storage medium storing an adaptive health management program for a heavy machine tool hydraulic system. When the adaptive health management program for the heavy machine tool hydraulic system is executed by a processor, it implements the steps of an adaptive health management method for the heavy machine tool hydraulic system.

[0129] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; 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; and these 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.

Claims

1. An adaptive health management method for a heavy-duty machine tool hydraulic system, characterized in that, include: A directed graph model of the hydraulic system of a heavy machine tool is established, and the hydraulic system is divided into several functional sub-regions based on the directed graph model; Principal component analysis is performed in each of the functional sub-regions. Based on the comprehensive contribution rate of each measurement point to the principal component and residual subspace, and combined with the preset cost constraints, the optimal combination of pressure measurement points is determined. Establish a local state observer corresponding to each of the functional sub-regions, input the pressure signal collected by the optimal pressure measurement point combination into the corresponding local state observer, and calculate the pressure residual output by each local state observer in real time. Principal component analysis is used to identify the current operating mode in real time, and the residual alarm threshold of each local state observer is dynamically adjusted according to the current operating mode. When the pressure residual of any functional sub-region exceeds the corresponding residual alarm threshold, the fault is traced and located based on the topology of the directed graph model, and the severity of the fault is assessed based on the fault tracing results, triggering the corresponding graded response.

2. The adaptive health management method for heavy machine tool hydraulic systems according to claim 1, characterized in that, The aforementioned establishment of a directed graph model for the heavy machine tool hydraulic system involves dividing the hydraulic system into several functional sub-regions based on the directed graph model, specifically including: The power components, control components, actuators, and key pipeline nodes of the hydraulic system are represented as nodes in a directed graph; the hydraulic oil flow path and energy transfer direction are represented as edges in the directed graph; and the pressure between components is... The flow coupling relationship model or hydraulic admittance is used as the weight of the corresponding directed edge; Based on the directed graph model, the hydraulic system is divided into a power source area, a control valve area, and an actuator area according to functional type.

3. The adaptive health management method for heavy machine tool hydraulic systems according to claim 1, characterized in that, Principal component analysis is performed in each of the aforementioned functional sub-regions. Based on the comprehensive contribution rate of each measurement point to the principal components and residual subspace, and in conjunction with preset cost constraints, the optimal combination of stress measurement points is determined, specifically including: Candidate pressure measurement points are set in each of the functional sub-regions to collect historical pressure data under various working conditions; The historical pressure data of each functional sub-region are standardized, the covariance matrix is ​​calculated and the corresponding eigenvalues ​​and eigenvectors are solved, and the number of principal components is determined based on the preset threshold of the cumulative variance contribution rate. Calculate the variance contribution rate of each candidate measurement point to the principal component and its contribution rate to the residual subspace. Based on the variance contribution rate and residual subspace contribution rate of each candidate measurement point, the comprehensive contribution rate of each candidate measurement point is obtained. With the optimization objective of maximizing the total contribution rate and minimizing the information redundancy between measurement points, and taking into account the cost constraint of the number of sensors, an optimization algorithm is used to solve for the optimal combination of pressure measurement points in each functional sub-region.

4. The adaptive health management method for heavy machine tool hydraulic systems according to claim 1, characterized in that, Establish a local state observer corresponding to each of the aforementioned functional sub-regions. Input the pressure signal acquired by the optimal pressure measurement point combination into the corresponding local state observer, and calculate the pressure residual output by each local state observer in real time. Specifically, this includes: Based on the pressure-flow dynamic mathematical model of each functional sub-region, a state-space model corresponding to each functional sub-region is established; wherein, the state equations of the state-space model include: In the formula, For load traffic, For the effective area of ​​the piston, x For piston displacement, The internal leakage coefficient is... This is the total volume of the cavity. For effective bulk modulus, For load pressure, m For load quality, The viscous damping coefficient is... K For the load spring stiffness, External load force; Based on the state-space model, a Kalman filter is designed as the local state observer. The real-time pressure signal collected by the optimal pressure measurement point combination is used as the observation value and input into the Kalman filter for state estimation to obtain the pressure estimate. The difference between the actual measured pressure value and the estimated pressure value is taken as the pressure residual.

5. The adaptive health management method for heavy machine tool hydraulic systems according to claim 1, characterized in that, The method of using principal component analysis to identify the current operating mode in real time and dynamically adjusting the residual alarm thresholds of each local state observer based on the current operating mode specifically includes: Clustering is performed using the score vectors of historical pressure data under multiple operating conditions in the principal component analysis subspace to obtain the cluster centers corresponding to each operating condition mode and establish an operating condition mode library; at the same time, statistical analysis is performed on the residual data under each operating condition mode to obtain the basic alarm thresholds corresponding to each operating condition mode. During online operation, real-time pressure data is projected onto the principal component analysis subspace to obtain a real-time score vector. The distance between the real-time score vector and each cluster center is calculated to identify the current working condition mode. The corresponding basic alarm threshold is retrieved based on the current operating mode, and the basic alarm threshold is dynamically compensated by combining the physical parameters of the components in the directed graph model to obtain the adjusted residual alarm threshold.

6. The adaptive health management method for a heavy-duty machine tool hydraulic system according to claim 1, characterized in that, When the pressure residual of any functional sub-region exceeds the corresponding residual alarm threshold, the fault is traced and located based on the topology of the directed graph model, and the severity of the fault is assessed based on the fault tracing results, triggering a corresponding graded response, specifically including: When the pressure residual of any of the functional sub-regions exceeds the corresponding residual alarm threshold and continues to exceed the preset sampling period, the fault diagnosis process is triggered. According to the upstream and downstream topological order of the directed graph model, the residual contribution rate of the local state observer corresponding to each functional sub-region is checked in turn, and the upstream node that first exceeds the residual limit and has the largest residual contribution rate is determined as the root cause node of the fault. The fault severity index is calculated based on the residual excess magnitude of the fault root cause node and a preset node weighting coefficient. S : In the formula, | r | represents the current absolute value of the residual. The dynamic alarm threshold under the current operating conditions. These are the node weight coefficients; The severity level of the fault is determined based on the fault severity index, and the graded response measures corresponding to the fault severity level are triggered.

7. The adaptive health management method for heavy machine tool hydraulic systems according to claim 1, characterized in that, After triggering the corresponding graded response, the method further includes: Operational data is collected according to a preset cycle. Operational data that has been determined by the pressure residual diagnosis process to be within the limit and has been manually verified to be in a fault-free state are used as updated data. The mean vector and covariance matrix of the principal component analysis model are updated using the updated data; Under the physical constraints of the directed graph model, the physical parameters of each local state observer are fine-tuned using the least squares estimation method. The updated principal component analysis model and local state observer parameters will be put into online operation for the next cycle.

8. An adaptive health management system for a heavy-duty machine tool hydraulic system, the system being used to execute the adaptive health management method for the heavy-duty machine tool hydraulic system according to any one of claims 1-7, characterized in that, The system includes: The directed graph modeling module is used to build a directed graph model of the hydraulic system of heavy machine tools, and divide the hydraulic system into several functional sub-regions based on the directed graph model. The measurement point optimization module is used to perform principal component analysis in each of the functional sub-regions, and determine the optimal combination of pressure measurement points based on the comprehensive contribution rate of each measurement point to the principal component and the residual subspace, combined with the preset cost constraints. The state observation module is used to establish a local state observer corresponding to each of the functional sub-regions, input the pressure signal collected by the optimal pressure measurement point combination into the corresponding local state observer, and calculate the pressure residual output by each local state observer in real time. The threshold adaptive module is used to identify the current operating mode in real time using principal component analysis and dynamically adjust the residual alarm threshold of each local state observer according to the current operating mode. The fault diagnosis and response module is used to trace and locate the fault based on the topology of the directed graph model when the pressure residual of any functional sub-zone exceeds the corresponding residual alarm threshold, and to assess the severity of the fault based on the fault tracing results and trigger the corresponding graded response. The model update module is used to collect operational data with residuals not exceeding the limit and confirmed to be fault-free by manual review at a preset period as update data. The update data is used to update the mean vector and covariance matrix of the principal component analysis model, and the physical parameters of each local state observer are fine-tuned under the physical constraints of the directed graph model. The updated model and observer parameters are then put into online operation.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the adaptive health management method for the heavy machine tool hydraulic system as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the adaptive health management method for the heavy machine tool hydraulic system as described in any one of claims 1 to 7.