Complex aerospace system fault intelligent diagnosis method and device based on correlation modeling

Abstract

The invention discloses an intelligent fault diagnosis method and device for a complex aerospace system based on correlation modeling, and relates to the technical field of aerospace system fault diagnosis, and the method comprises the steps: constructing a testability model of the aerospace system; the distinguishing capability of different faults is simplified, and test points are optimized; automatically generating a fault diagnosis strategy based on the optimized testability model; performing simulation evaluation on the fault diagnosis strategy, and counting a fault detection rate and a fault isolation rate; and performing iterative optimization by judging whether the fault detection rate and the fault isolation rate meet preset requirements or not. According to the method, the technical problems of low diagnosis efficiency and insufficient fault detection and isolation accuracy caused by dependence on artificial experience and incomplete static fault tree coverage of complex spaceflight system fault diagnosis in the prior art are solved, intelligentization and precision of complex spaceflight system fault diagnosis are realized, and the fault diagnosis efficiency is improved. And the fault detection rate and the isolation rate are improved.

Description

Technical Field

[0001] This invention relates to the field of aerospace system fault diagnosis technology, specifically to a method and equipment for intelligent fault diagnosis of complex aerospace systems based on correlation modeling. Background Technology

[0002] Complex aerospace systems, such as reusable rockets, are highly integrated with multiple subsystems including propulsion, control, and electrical systems. They must withstand harsh operating conditions during launch and recovery, and rapid and accurate fault location is crucial to ensure mission safety and system reusability. Traditional aerospace system fault diagnosis relies heavily on manual analysis by engineers based on measurement parameters and experience. This is not only time-consuming but also susceptible to subjective judgment, leading to missed diagnoses and misjudgments. While some research has employed fault trees and expert systems to assist diagnosis, static fault trees struggle to cover all fault scenarios as system complexity increases. Expert knowledge compilation is time-consuming and difficult to update, failing to adapt to dynamically changing fault modes. In recent years, while correlation models and other technologies have been applied in test analysis, the design of diagnostic solutions for complex aerospace systems still lacks a systematic approach. Existing methods struggle to efficiently establish quantitative correlations between faults and sensor signals, and cannot improve diagnostic performance through quantitative evaluation and iterative optimization. This results in insufficiently targeted diagnostic strategies, low fault detection and isolation rates, and an inability to meet the demands of complex aerospace systems for intelligent, precise, and efficient fault diagnosis.

[0003] Existing technologies suffer from technical problems such as reliance on human experience for fault diagnosis of complex aerospace systems, incomplete coverage of static fault trees, resulting in low diagnostic efficiency and insufficient accuracy in fault detection and isolation. Summary of the Invention

[0004] This application provides a method and equipment for intelligent fault diagnosis of complex aerospace systems based on correlation modeling, which addresses the technical problems in the prior art where fault diagnosis of complex aerospace systems relies on human experience, has incomplete static fault tree coverage, resulting in low diagnostic efficiency, and insufficient accuracy in fault detection and isolation.

[0005] In view of the above problems, this application provides a method and equipment for intelligent fault diagnosis of complex aerospace systems based on correlation modeling.

[0006] The first aspect of this application provides an intelligent fault diagnosis method for complex aerospace systems based on correlation modeling, the method comprising: A testable model of the aerospace system is constructed, including a testable graphical model reflecting the relationship between components and signals and a mathematical correlation model. Correlation analysis is used to calculate the ability of each sensor to distinguish different faults, simplifying the testable model and optimizing test points. Based on the optimized testable model, a fault diagnosis strategy is automatically generated, including diagnostic tree design and fault dictionary compilation. The optimized mathematical correlation model is used to simulate and evaluate the fault diagnosis strategy, and the fault detection rate and fault isolation rate are statistically analyzed. Iterative optimization is performed by determining whether the fault detection rate and fault isolation rate meet preset requirements.

[0007] A second aspect of this application provides an electronic device, comprising: a memory for storing executable instructions; and a processor for implementing a method for intelligent fault diagnosis of complex aerospace systems based on correlation modeling when executing the executable instructions stored in the memory.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: A testable model of the aerospace system is constructed, including a testable graphical model reflecting the relationship between components and signals and a mathematical correlation model. Correlation analysis is used to calculate the ability of each sensor to distinguish different faults, simplifying the testable model and optimizing test points. Based on the optimized testable model, a fault diagnosis strategy is automatically generated, including diagnostic tree design and fault dictionary compilation. The optimized mathematical correlation model is used to simulate and evaluate the fault diagnosis strategy, statistically analyzing the fault detection rate and fault isolation rate. Iterative optimization is performed by determining whether the fault detection rate and fault isolation rate meet preset requirements. This achieves intelligent and precise fault diagnosis of complex aerospace systems, improving the technical effectiveness of fault detection and isolation rates. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A flowchart illustrating a method for intelligent fault diagnosis of complex aerospace systems based on correlation modeling, provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0011] Explanation of reference numerals in the attached drawings: Input device 201, processor 202, memory 203, output device 204. Detailed Implementation

[0012] This application provides a method and equipment for intelligent fault diagnosis of complex aerospace systems based on correlation modeling. It addresses the technical problems in the prior art where fault diagnosis of complex aerospace systems relies on human experience, has incomplete static fault tree coverage, resulting in low diagnostic efficiency, and insufficient accuracy in fault detection and isolation.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] Example 1, as Figure 1 As shown, this application provides an intelligent fault diagnosis method for complex aerospace systems based on correlation modeling, the method comprising: Step S100: Construct a testable model of the aerospace system, including a testable graphical model and a mathematical correlation model that reflect the relationship between components and signals.

[0015] Specifically, based on the subsystem division of aerospace systems, such as propulsion, control, electrical, and structural subsystems, and the connection logic of key components in each subsystem, a testable graphical model reflecting the relationship between components and signals is drawn. Taking the propulsion system as an example, the functional relationships of core components such as fuel pumps, engines, and control valves are first clarified, and the sensor installation locations corresponding to each measurable parameter, such as pressure, temperature, speed, and valve opening, are marked, such as pump outlet pressure sensor P1, combustion chamber pressure sensor P2, turbine tachometer N, and valve opening sensor θ. The impact path of component failure on sensor signals is clearly defined by block diagram lines. For example, a fuel pump failure will directly affect the P1 and N signals, forming an intuitive and visual relationship map. On this basis, a matrix form is further constructed. The mathematical correlation model uses rows of a matrix to correspond to all sensor signals, such as P1, P2, N, θ, etc., and columns to correspond to potential fault modes of the system, such as fuel pump jamming, unstable engine combustion, valve sticking, sensor failure, etc. Matrix elements are represented by 1 / 0 to indicate whether there is an impact or the qualitative level. For example, "strong impact / weak impact" quantitatively characterizes the degree of impact of a fault on the corresponding signal. For example, "fuel pump jamming" corresponds to elements of rows P1 and N with 1, and elements of row P2 with 0. At the same time, it also clarifies the set of fault modes that each signal anomaly may be associated with. Finally, it forms a testable graphic model that combines the component-signal visualization path with the mathematical correlation model of fault-signal quantitative correlation, which fully covers the basic correlation data required for diagnosis.

[0016] Step S200: Calculate the ability of each sensor to distinguish different faults through correlation analysis, simplify the test model and optimize the test points.

[0017] Specifically, based on the constructed mathematical correlation model, the system employs information entropy to measure the sensor's ability to distinguish fault information or diagnostic matrix rank analysis to evaluate the sensor combination's ability to isolate faults. Pre-defined calculation strategies are used to calculate the distinguishing contribution of each sensor to different fault modes. For example, analyzing two functionally overlapping temperature sensors T1 and T2, if their signal responses are completely consistent across all fault scenarios and the corresponding row elements in the matrix are identical, their diagnostic contribution is considered highly redundant. Then, a preset contribution threshold is set for each sensor. Sensors below this threshold offer no significant gain for fault isolation, thus identifying the sensor combination most critical for fault detection and isolation. Redundant test points with contribution below the threshold are then removed, such as redundant temperature sensor T2 and duplicate valve position sensors. Simultaneously, by backtracking the mathematical correlation model, it is ensured that all critical fault modes, such as "fuel pump jamming" and "unstable engine combustion," can be detected by at least one set of remaining test points. This means that each fault corresponds to at least one signal anomaly from a remaining sensor. Finally, the test model's structure is simplified and test points are optimized, reducing system monitoring complexity and cost while ensuring the effectiveness of diagnostic data.

[0018] Step S300: Based on the optimized test model, automatically generate a fault diagnosis strategy, including diagnostic tree design and fault dictionary compilation.

[0019] Specifically, in the diagnostic tree design phase, the quantitative correlation between signals and faults is first extracted from the optimized mathematical correlation model. Combined with the prior probability of fault occurrence, such as "unstable engine combustion" which is highly dangerous and has a high probability of occurrence, it is prioritized for detection. Key sensor signals are selected as initial test points, such as combustion chamber pressure sensor P2. Based on its test results, normal / abnormal oscillations divide the fault space into two branches: "eliminating unstable combustion faults" and "locking in unstable combustion faults". Then, for the "eliminating unstable combustion" branch, the next test point is recursively selected, such as pump outlet pressure P1 and turbine speed N. Based on the test results, such as low P1 and low N / normal P1 but abnormal N, the fault set is further subdivided until each branch can uniquely lock a fault mode, such as fuel pump jamming, turbine failure, or sensor failure. Finally, a diagnostic tree is formed with test steps as nodes and fault modes as leaf nodes, clarifying the optimal detection order. In the fault dictionary compilation stage, unique signal anomaly combinations corresponding to each fault mode are extracted from the optimized mathematical correlation model, i.e., fault fingerprints. For example, fuel pump jamming corresponds to "P1 low, N low, other signals normal", valve jamming corresponds to "θ does not match the command, P2 abnormal, the rest normal", and sensor fault P1 failure corresponds to "P1 reading extreme value, sensor self-test BIT fails, other signals normal". These fault-signal anomaly correspondences are organized into a structured table to form a fault dictionary that can be directly used for rapid comparison and diagnosis. It complements the diagnostic tree. The diagnostic tree guides the order of testing, and the fault dictionary supports how to judge after testing. Together, they constitute a complete fault diagnosis strategy.

[0020] Step S400: Use the optimized mathematical correlation model to simulate and evaluate the fault diagnosis strategy, and statistically analyze the fault detection rate and fault isolation rate.

[0021] Specifically, based on the optimized mathematical correlation model, simulation scenarios covering aerospace systems, such as reusable rockets with all potential failure modes, are preset. These scenarios include simulating fuel pump jamming, valve sticking, and sensor failure multiple times, totaling over a hundred failure scenarios. In each scenario, the model sets the output data of the simulated sensors. For example, in the fuel pump jamming scenario, signals of low P1 and low N are output; in the sensor failure scenario, signals corresponding to extreme sensor readings and BIT self-test anomalies are output. The generated fault diagnosis strategy is then applied to these simulation scenarios. Faults are inferred through a diagnostic tree according to a preset test order, and the diagnostic results are verified by comparing signal anomaly combinations with the fault dictionary. Finally, core performance indicators are statistically analyzed based on the simulation results. The Fault Detection Rate (FDR) is the proportion of successfully identified faults out of the total number of fault simulations. If all faults trigger at least one sensor signal anomaly or BIT alarm, the FDR can approach 100%. The Fault Isolation Rate (FIR) is the proportion of accurately identified specific faults out of the total number of detected faults. This is achieved by analyzing the fault set covered by the diagnostic tree and the differentiation of the fault dictionary, and statistically analyzing the proportion of unconfused faults. This quantifies the actual effectiveness of the diagnostic strategy and provides data support for subsequent optimization.

[0022] Step S500: Iterative optimization is performed by determining whether the fault detection rate and the fault isolation rate meet the preset requirements.

[0023] Specifically, the statistically obtained Fault Detection Rate (FDR) and Fault Isolation Rate (FIR) are compared with the preset performance requirements of the aerospace system, such as FDR ≥ 99% and FIR ≥ 95%. If both indicators meet the requirements, the current diagnostic strategy is deemed qualified and no further optimization is needed. If the requirements are not met, such as an FIR of only 90%, indicating confusion between "valve jamming" and "fuel pump jamming," an iterative optimization process is initiated. Based on the reasons for the performance failure, an appropriate solution is selected from the preset optimization mechanism. For example, to address the problem of fault confusion, a new sensor is added, such as a valve drive current sensor, to distinguish between valve jamming and pump failure through current anomalies. Alternatively, the test order of the diagnostic tree is adjusted to prioritize the detection of signals that can distinguish between confused faults. Or, the component design is modified to enhance testability, such as optimizing the sensor installation position to improve signal sensitivity. After optimization and adjustment, the process returns to step S200, which involves selecting test points, and then proceeding to step S400, which involves performance evaluation. The FDR and FIR of the optimized diagnostic strategy are then quantitatively evaluated again until both indicators meet the preset requirements, forming a stable and reliable final diagnostic solution.

[0024] In one possible implementation, step S100 further includes: The mathematical correlation model represents the relationship between sensor readings and the status of key components in matrix form, where rows represent sensor signals, columns represent fault modes, and matrix elements represent the degree of impact of faults on signals.

[0025] Specifically, the matrix comprehensively lists all filtered sensor signals within the system in the row dimension, such as pump outlet pressure signal P1, combustion chamber pressure signal P2, turbine speed signal N, valve opening signal θ, temperature signal T1, and sensor self-test BIT signals in the reusable rocket propulsion system, ensuring that each key measurable parameter has a corresponding row item; and clearly marks all potential failure modes of the system in the column dimension, such as "fuel pump jamming", "unstable engine combustion", "valve jamming", "P1 sensor failure", "turbine failure", etc., covering all types of failure scenarios from component functional failure to sensor abnormality. Each element in the matrix is ​​used to precisely describe the degree of impact of a certain fault mode on the corresponding sensor signal. It is usually defined qualitatively or quantitatively. Qualitatively, "1", "0", and "0.5" can be used to represent "significant impact". For example, fuel pump jamming will cause the P1 signal to drop, and the corresponding element is 1, "no impact". If the fuel pump jamming has no direct effect on the P2 signal, the corresponding element is 0, "weak impact". If combustion instability has a slight disturbance to the temperature signal T1, the corresponding element is 0.5. Quantitatively, specific values ​​can be assigned according to the intensity threshold of the fault impact, such as the percentage of signal deviation. For example, if the fault causes the signal deviation to exceed 20%, the element is 0.8, and if the deviation is 5% to 20%, it is 0.4. Finally, a matrix structure with clear row, column, and element logic is formed, which not only clarifies "which signals a certain fault will affect", but also defines "which faults may cause a certain signal abnormality", providing a quantitative correlation basis for subsequent test point optimization and diagnostic strategy generation.

[0026] In one possible implementation, step S200 further includes: Step S210: Calculate the contribution of each sensor to the ability to distinguish different faults through a preset calculation strategy, wherein the preset calculation strategy is either information entropy or the rank of the diagnostic matrix.

[0027] Step S220: Based on the contribution of each sensor to the ability to distinguish different faults, remove test points whose contribution is lower than a preset threshold to complete the test point selection.

[0028] Specifically, based on the constructed mathematical correlation model, rows correspond to sensor signals, columns correspond to fault modes, and elements represent the degree of influence of faults on signals. According to the diagnostic requirements of the aerospace system, one of the preset calculation strategies is selected for calculation. If the information entropy strategy is selected, the prior probability of each fault mode must first be statistically calculated. Then, based on the abnormal state of the sensor signal, such as "normal / abnormal" or "high / low," the information gain that the sensor signal can reduce the uncertainty of the fault set is calculated. The greater the information gain, the stronger the sensor's ability to distinguish different faults. For example, a pressure sensor can clearly distinguish between "fuel pump jamming" and "turbine fault" by signal strength, indicating high information gain and high contribution. If the diagnostic matrix rank strategy is selected, a sub-correlation matrix containing the row corresponding to the sensor signal needs to be extracted. The rank of the sub-matrix is ​​calculated to evaluate the sensor's ability to distinguish faults in the fault space. The higher the rank value, the more fault categories the sensor can help isolate. For example, the rank of the sub-matrix corresponding to a speed sensor is 5, indicating that it can participate in distinguishing 5 different types of faults, with a significant contribution. Finally, by using either strategy, a quantitative contribution value for each sensor's ability to distinguish different faults can be obtained, providing data for subsequent test point selection.

[0029] In conjunction with aerospace systems, such as reusable rockets, considering diagnostic accuracy requirements, hardware cost constraints, and operational complexity, a preset threshold for sensor contribution is established. This threshold must balance "diagnostic effectiveness" and "system simplicity." For example, based on historical diagnostic data or industry standards, only sensors with the top 75% contribution ranking or contribution values ​​higher than 0.6 are retained, assuming a contribution quantification range of 0-1. Subsequently, the calculated contribution quantification values ​​of each sensor are compared one by one with the preset threshold. Test points with contribution values ​​below the threshold are filtered out. For instance, two temperature sensors with overlapping functions and identical signal responses have highly redundant contributions; if one of them has a contribution value below the threshold, it is marked as an object to be removed. During the elimination process, the mathematical correlation model needs to be backtracked simultaneously to verify whether the test point set composed of the remaining sensors can cover all key fault modes of the system. That is, it is necessary to ensure that each potential fault can be detected by at least one signal anomaly of the remaining sensor, so as to avoid the risk of missed fault detection due to the elimination of test points. If it is found that a certain type of key fault has lost its detection basis, the threshold needs to be adjusted appropriately or some low contribution but indispensable test points need to be retained. Finally, the test point optimization is completed, which simplifies the system monitoring architecture, reduces hardware and operation and maintenance costs, and ensures the effectiveness of subsequent diagnostic strategies.

[0030] In one possible implementation, step S300 further includes: Step S310: Extract the mathematical correlation model from the optimized test model, select the initial test point according to the correlation between the signal and the fault, and divide the fault space into two parts according to the test results.

[0031] Step S320: Based on the fault space bisection result, recursively select the next test point, continue to analyze the correlation between the signal and the fault to perform fault space bisection until the fault mode can be uniquely isolated, and generate a fault decision tree as the diagnostic tree.

[0032] Specifically, the simplified mathematical correlation model is first extracted from the optimized test model, that is, test points with low contribution have been eliminated, and the quantitative correlation between core sensor signals and fault modes is retained. Then, combined with the prior probability of fault occurrence, such as the high harm and frequency of "unstable engine combustion" and the high cost of detection, and the convenient acquisition and high reliability of key pressure sensor signals, the signal most critical to fault differentiation is selected as the initial test point. Taking a reusable rocket propulsion system as an example, if unstable engine combustion will cause significant oscillation of the combustion chamber pressure signal P2, and this fault has the greatest impact on flight safety, then the P2 sensor signal is set as the initial test point. Subsequently, based on the results of the initial test point, such as abnormal oscillation of P2 or normal P2, the fault space composed of all potential fault modes of the system is divided into two parts: the fault subset corresponding to abnormal oscillation of P2 and the fault subset corresponding to normal P2, completing the first fault space bisection.

[0033] Starting with the binary fault subset, the sensor signal with the strongest ability to distinguish faults within the current subset is selected from the optimized mathematical correlation model as the next test point. For example, for this subset, the pump outlet pressure P1 and turbine speed N can be initially distinguished from fuel pump jamming by the difference in signal strength, i.e., low P1 and low N are different from other faults. Therefore, these two are selected as the next set of test points. Based on the test results, such as "P1 is low and N is low", "P1 is normal but N is abnormal", and "P1 and N are both normal", the current fault subset is binaryed again to obtain a more subdivided subset. The process then proceeds recursively according to the same logic: for each newly generated fault subset, the optimal test point is repeatedly selected and the process of dividing the subsets according to the relationship between the signal and the fault is repeated. For example, for the "turbine fault / sensor failure" subset, the sensor self-test BIT signal is selected as the test point, and it is further subdivided into two single fault subsets, "sensor failure" and "turbine fault", by "BIT abnormal / normal", until the fault subsets at the end of all branches contain only one fault mode. Finally, a fault decision tree, i.e. a diagnostic tree, is formed with test points as nodes, test results as branches, and single fault modes as leaf nodes.

[0034] In one possible implementation, step S300 further includes: Step S330: Extract the signal anomaly combination corresponding to each fault mode from the mathematical correlation model.

[0035] Step S340: Form a fault-symptom correspondence table based on the signal anomaly combinations corresponding to each fault mode, and establish the fault dictionary.

[0036] Specifically, using the optimized mathematical correlation model, rows correspond to core sensor signals, columns correspond to fault modes, and elements represent the degree of influence of faults on signals as data sources. For each fault mode corresponding to a column, such as "fuel pump jamming," "valve jamming," and "P1 sensor failure," the matrix element values ​​of the corresponding rows are analyzed one by one. Sensor signals with element values ​​of "significant influence," such as 1 or high quantization values, are selected to determine the type of signal anomaly triggered by the fault mode. For example, "fuel pump jamming" corresponds to an element of 1 in row P1 and an element of 1 in row N of the matrix, which triggers an anomaly of a decrease in both the P1 and N signals. At the same time, signals with element values ​​of no influence or weak influence are excluded. Finally, the unique signal anomaly combination that can be used as a diagnostic basis for each fault mode is extracted, namely the fault fingerprint.

[0037] The extracted fault mode-signal anomaly combination correspondences are organized into a structured table called a fault-symptom correspondence table. The table columns are clearly divided into dimensions such as fault mode name, signal anomaly combination, and key distinguishing features. For example, in the fuel pump jamming row, the signal anomaly combination column is filled with "P1 low, N low, P2 normal, θ normal, T1 normal," and the key distinguishing feature column indicates that P1 and N decrease simultaneously. In the P1 sensor failure row, the signal anomaly combination column is filled with "Extreme P1 reading," "BIT detection failed," and "Other signals normal," and the key distinguishing feature column indicates "Only P1 is abnormal and BIT alarm." This structured organization transforms scattered signal anomaly combinations into a clear and referable fault dictionary. During subsequent diagnosis, the suspected fault mode can be quickly identified by directly comparing the actual sensor signal anomalies with the fault fingerprints in the dictionary, providing supplementary verification for diagnostic tree reasoning.

[0038] In one possible implementation, step S400 further includes: Step S410: Preset various fault occurrence scenarios, simulate sensor output, and identify faults through a diagnostic tree.

[0039] Step S420: Based on the fault identification results, analyze the fault set covered by the diagnostic tree and the fault dictionary to statistically calculate the fault detection rate and fault isolation rate.

[0040] Specifically, based on the full range of potential failure modes of aerospace systems, such as reusable rockets, simulation scenarios are preset to cover different failure types, severity levels, and concurrent situations. For example, each single failure such as fuel pump jamming, valve sticking, and sensor failure is simulated 20 times, while a composite failure scenario of valve sticking and minor turbine anomalies is set up 5 times. For each scenario, according to the "fault-signal" association rules in the optimized mathematical correlation model, the real-time data of each sensor is simulated and output. For example, in the fuel pump jamming scenario, signals of low P1 and low N are output, and in the sensor failure scenario, signals of the corresponding extreme values ​​of sensor readings and BIT self-test anomalies are output. Then, the simulated sensor data is input into the generated diagnostic tree. According to the test order of the diagnostic tree, such as first detecting P2, then detecting P1 and N, and finally detecting BIT, the fault identification result for each scenario is output step by step, such as determining that it is fuel pump jamming or valve sticking.

[0041] Next, the fault identification results based on all scenarios are analyzed statistically from two dimensions: fault detection and fault isolation. For the fault detection rate (FDR), the proportion of scenarios successfully identified as faulty by the diagnostic tree is counted out of the total number of fault scenarios. For example, if 98 out of 100 fault scenarios are successfully detected, the FDR is 98%, indicating whether there are any undetected faults, i.e., missed detections. For the fault isolation rate (FIR), the proportion of scenarios accurately identified as specific fault modes is counted out of the total number of detected fault scenarios. For example, if 95 out of 98 detected faults are accurately located to specific fault types, the FIR is 96.9%, indicating whether there is any confusion in the fault set covered by the diagnostic tree, such as misclassifying valve jamming as fuel pump jamming. At the same time, the accuracy of the diagnostic results is verified by combining the signal anomaly combination discrimination of the fault dictionary. For example, by comparing the θ of valve jamming with the command and the P2 anomaly feature in the fault dictionary, the signal matching deviation of misjudged scenarios is confirmed. Finally, the quantitative statistics of fault detection rate and fault isolation rate are completed, providing a performance basis for subsequent iterative optimization.

[0042] In one possible implementation, step S500 further includes: Step S510: Determine whether the fault detection rate and the fault isolation rate meet the preset requirements.

[0043] Step S520: If not, perform iterative optimization according to the preset optimization mechanism, wherein the preset optimization mechanism includes at least adding new sensors, adjusting diagnostic tree logic, and modifying component design.

[0044] Specifically, the statistically obtained Fault Detection Rate (FDR) and Fault Isolation Rate (FIR) are compared one by one with the pre-set diagnostic performance requirements of aerospace systems, such as reusable rockets. These requirements are typically set based on system safety, reliability standards, and operational needs, such as FDR ≥ 99% and FIR ≥ 95%. If both indicators reach or exceed the preset thresholds, it indicates that the current diagnostic solution can effectively meet the fault detection and isolation requirements, and no further adjustments are needed. If either indicator fails to meet the standard, such as an FIR of only 90%, indicating a potential misjudgment of valve jamming or fuel pump sticking, then the diagnostic solution needs to be optimized.

[0045] Based on the specific reasons for performance failure, an appropriate improvement plan is selected from the preset optimization mechanism: If the fault confusion stems from the existing sensors' inability to distinguish similar fault characteristics, such as valve jamming and fuel pump sticking both causing abnormal P2 signals and lacking a dedicated distinguishing signal, a new sensor is added, such as a valve drive current sensor, utilizing the characteristic of abnormal current when the valve is jammed and normal current when the pump malfunctions to achieve differentiation; if the fault is missed or misjudged due to an unreasonable diagnostic tree test order, such as prioritizing the detection of signals with low distinguishability, the diagnostic tree logic is adjusted, such as moving sensor signals that can distinguish core faults to earlier test nodes; if the fault is difficult to detect due to poor measurability of the component itself, such as the lack of suitable signal output for key components, the component design is modified, such as adding signal monitoring interfaces at key locations of the component to enhance measurability. After each optimization adjustment, the process from S200 test point selection to S400 performance evaluation needs to be re-executed to verify whether the fault detection rate and isolation rate meet the standards, until both indicators meet the preset requirements, forming a final stable diagnostic solution.

[0046] Example 2, Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, showing a block diagram of an exemplary electronic device suitable for implementing the embodiments of the present invention. Figure 2 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention. This electronic device is presented as a general-purpose computing device, and its components may include, but are not limited to, an input device 201, a processor 202, a memory 203, and an output device 204. The processor 202 may be one or more; by running software programs, instructions, and modules stored in the memory 203, the processor 202 executes various functional applications and data processing of the computer device, thereby realizing the aforementioned intelligent fault diagnosis method for complex aerospace systems based on correlation modeling.

[0047] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0048] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0049] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for intelligent fault diagnosis of complex aerospace systems based on correlation modeling, characterized in that, include: Construct testable models of aerospace systems, including testable graphical models and mathematical correlation models that reflect the relationship between components and signals; Correlation analysis was used to calculate the ability of each sensor to distinguish different faults, and the test model was simplified and test points were optimized. Based on the optimized testable model, a fault diagnosis strategy is automatically generated, including diagnostic tree design and fault dictionary compilation. The fault diagnosis strategy was simulated and evaluated using the optimized mathematical correlation model, and the fault detection rate and fault isolation rate were statistically analyzed. Iterative optimization is performed by determining whether the fault detection rate and the fault isolation rate meet preset requirements.

2. The intelligent fault diagnosis method for complex aerospace systems based on correlation modeling as described in claim 1, characterized in that, The mathematical correlation model represents the relationship between sensor readings and the status of key components in matrix form, where rows represent sensor signals, columns represent fault modes, and matrix elements represent the degree of impact of faults on signals.

3. The intelligent fault diagnosis method for complex aerospace systems based on correlation modeling as described in claim 1, characterized in that, Correlation analysis was used to calculate the ability of each sensor to distinguish different faults, and the test model was simplified and test points were optimized, including: The contribution of each sensor to the ability to distinguish different faults is calculated by a preset calculation strategy, wherein the preset calculation strategy is either information entropy or the rank of the diagnostic matrix. Based on the contribution of each sensor to different fault differentiation capabilities, test points with a contribution rate below a preset threshold are removed, thus completing the test point optimization.

4. The intelligent fault diagnosis method for complex aerospace systems based on correlation modeling as described in claim 1, characterized in that, Based on the optimized test model, a fault diagnosis strategy is automatically generated, including: Mathematical correlation models are extracted from the optimized test model. Initial test points are selected based on the correlation between signals and faults. The fault space is then divided into two parts based on the test results. Based on the fault space bisection results, the next test point is recursively selected, and the correlation between the signal and the fault is analyzed to perform fault space bisection until the fault mode can be uniquely isolated, generating a fault decision tree as the diagnostic tree.

5. The intelligent fault diagnosis method for complex aerospace systems based on correlation modeling as described in claim 4, characterized in that, Based on the optimized test model, a fault diagnosis strategy is automatically generated, including: Extract the signal anomaly combinations corresponding to each fault mode from the mathematical correlation model; A fault-symptom correspondence table is formed based on the signal anomaly combinations corresponding to each fault mode, and the fault dictionary is established.

6. The intelligent fault diagnosis method for complex aerospace systems based on correlation modeling as described in claim 5, characterized in that, The fault diagnosis strategy was simulated and evaluated using the optimized mathematical correlation model, and the fault detection rate and fault isolation rate were statistically analyzed, including: Pre-set various fault scenarios, simulate sensor outputs, and identify faults through a diagnostic tree; Based on the fault identification results, the fault detection rate and fault isolation rate are statistically calculated by analyzing the fault set covered by the diagnostic tree and the fault dictionary.

7. The intelligent fault diagnosis method for complex aerospace systems based on correlation modeling as described in claim 1, characterized in that, Iterative optimization is performed by determining whether the fault detection rate and the fault isolation rate meet preset requirements, including: Determine whether the fault detection rate and the fault isolation rate meet preset requirements; If not, iterative optimization is performed according to a preset optimization mechanism, which includes at least adding new sensors, adjusting diagnostic tree logic, and modifying component design.

8. An electronic device, characterized in that, The electronic device includes: Memory, used to store executable instructions; The processor, when executing executable instructions stored in the memory, implements the intelligent fault diagnosis method for complex aerospace systems based on correlation modeling as described in any one of claims 1 to 7.

Images