Intelligent power distribution and fault diagnosis method for aircraft power supply system

Abstract

The application relates to the technical field of aircraft power supply systems and discloses an intelligent power distribution and fault diagnosis method for an aircraft power supply system, which comprises the following steps: acquiring real-time current signals and load working condition information of an aircraft power bus, extracting a time domain entropy jump index representing energy domain stability of the current signals and a phase synchronism index representing time domain stability, and constructing a double-domain composite entropy feature; and performing a cooperative diagnosis decision based on the composite entropy feature and the load working condition information, wherein the application realizes sharp capture of real fault precursors through double review of energy domains and time domains of the current signals, and simultaneously shows tolerance to normal load dynamics in aircraft operation through cooperative judgment in combination with the load working condition, so that the application effectively avoids the restriction between safety and availability in the prior art and provides a possibility for realizing predictive health management.

Description

Technical Field

[0001] This invention relates to an intelligent power distribution and fault diagnosis method for an aircraft power system, belonging to the technical field of aircraft power systems. Background Technology

[0002] As a core component ensuring flight safety, the current aircraft power system is designed and operated in accordance with a mature technical approach based on static redundancy and passive isolation. This approach constructs a basic safety barrier by setting fixed current or voltage thresholds for each power supply system and supplementing it with physical redundancy backups for critical equipment. However, as modern aircraft continue to increase their requirements for power quality and operational reliability, the inherent logical dilemma of this technical approach, which relies on static thresholds for rigid judgment, is becoming increasingly prominent, especially in the irreconcilable conflict between system safety and availability.

[0003] On the one hand, to avoid any potential overload risks, protection thresholds are usually set conservatively. This makes the system prone to unnecessary protective false triggers when multiple subsystem loads change drastically and synchronously within the normal range during aircraft takeoff and landing or high-maneuver flight. This leads to frequent unloading of non-critical loads, which not only reduces the system's mission availability but also imposes additional operational burdens on the crew. On the other hand, if the protection threshold is relaxed to reduce false triggers, the system's sensitivity to slowly evolving latent faults will be significantly reduced. For example, intermittent weak arcs caused by long-term insulation fatigue or transient resistance changes caused by connector fretting. The early electrical characteristics of these faults are low in energy and short in duration. Their signals are completely drowned out by the system's normal background noise and cannot reach any adjusted protection threshold. This allows the system to accumulate risks that could lead to electrical accidents under a seemingly stable operating state.

[0004] Specifically, existing technologies suffer from the following shortcomings: 1. The diagnostic mechanism lacks the ability to effectively identify the essential differences between normal load dynamic changes and the electrical characteristics of actual faults, leading to a trade-off between false alarm rates and missed alarm rates under complex operating conditions; 2. The monitoring method relies on passive responses to current or voltage amplitudes, resulting in inherent blind spots for early fault precursors where the amplitude is not significant but the signal morphology has changed; 3. The existing technological framework forces system design to compromise between two undesirable states: excessive sensitivity leading to reduced availability and excessive sluggishness leading to compromised safety, lacking a solution that can simultaneously improve both. Therefore, how to establish a new diagnostic method that transcends simple reliance on current signal amplitude, analyzes the inherent stability and structural changes of the signal, and thus accurately identifies early latent fault precursors while exhibiting high tolerance to normal load dynamics under complex aircraft operating environments, is the technical problem this invention aims to solve. Summary of the Invention

[0005] This invention provides an intelligent power distribution and fault diagnosis method for aircraft power systems. Its main purpose is to solve the problem that existing technologies rely on static amplitude thresholds, which cannot simultaneously ensure the keen perception of early latent faults and the high tolerance to dynamic changes in normal loads, resulting in a trade-off between system safety and availability.

[0006] To achieve the above objectives, the present invention provides an intelligent power distribution and fault diagnosis method for an aircraft power system, the method comprising the following steps:

[0007] Step a: Obtain the real-time current signal on the aircraft power bus and the operating condition information of the current effective load associated with the real-time current signal;

[0008] Step b: Based on the real-time current signal, perform a dual-domain composite entropy feature extraction. The dual-domain composite entropy feature extraction includes: calculating a time-domain entropy transition index that characterizes the energy domain stability of the real-time current signal. The time-domain entropy transition index is used to quantify the rate of change of disorder of the real-time current signal within adjacent time windows.

[0009] Step c: When the time domain entropy transition exponent exceeds the arbitration threshold, a phase synchronization index characterizing the stability of the time structure domain of the real-time current signal is triggered and calculated. The phase synchronization index is used to quantify the periodic regularity of the oscillation waveform of the real-time current signal.

[0010] Step d: Based on the extracted dual-domain composite entropy features and the obtained current effective load operating condition information, a collaborative diagnostic decision is executed. The collaborative diagnostic decision compares the dual-domain composite entropy features with a pre-stored knowledge base that associates load operating conditions with normal entropy feature patterns to distinguish between a normal load dynamic event and a real electrical fault event, and generates power distribution control instructions accordingly.

[0011] Preferably, the step of calculating a time-domain entropy transition index characterizing the energy domain stability of a real-time current signal specifically includes: continuously calculating the standard deviation of the real-time current signal in units of a first predetermined time window; obtaining the current window calculated value of the standard deviation and the previous window calculated value; and using the ratio of the current window calculated value of the standard deviation to the previous window calculated value as the time-domain entropy transition index.

[0012] Preferably, the step of calculating a phase synchronization index characterizing the stability of the real-time current signal in the time structure domain is implemented by calculating the variance of the zero-crossing time difference of the real-time current signal oscillation waveform. Specifically, the calculation includes: extracting the oscillation waveform segment of the real-time current signal corresponding to the time-domain entropy transition exponent exceeding the arbitration threshold; determining the timestamp sequence of multiple consecutive zero-crossing points of the oscillation waveform segment; and calculating the variance σ of the time interval between adjacent timestamps in the timestamp sequence. Δt ,in Where, Δt i Let N be the number of zero-crossing points in the i-th time interval. The variance is the average across all time intervals; collaborative diagnostic decision-making further includes, if and only if the variance σ Δt Only when the time-domain entropy transition index is less than the phase synchronization threshold will the event represented by the time-domain entropy transition index be treated as a valid event that may indicate a real electrical fault and processed accordingly.

[0013] Preferably, the collaborative diagnostic decision further includes an iterative signature stripping step for decoupling concurrent normal dynamic events. This step is triggered when a mismatch is detected between the dual-domain composite entropy feature and the knowledge base, and two or more loads are simultaneously in an active state. The step includes: selecting one load from the multiple concurrent loads and retrieving the normal entropy feature pattern associated with the selected load from the knowledge base; stripping the feature component corresponding to the normal entropy feature pattern of the selected load from the currently extracted dual-domain composite entropy feature to obtain a residual entropy feature; and comparing the residual entropy feature with the normal entropy feature pattern of another concurrent load. If they match, the current event is determined to be an interpretable concurrent normal dynamic event.

[0014] Preferably, the method further includes a cross-flight trend analysis step for predicting progressive failures, which includes: converting a low-risk level diagnostic conclusion generated in a single flight and associated with the same load into an event symbol, and archiving it together with the flight's unique identification information to form a historical event symbol sequence; periodically calculating a pattern overlap between the event symbol sequence of the latest flight and the historical event symbol sequences of previous flights; and automatically increasing the risk level of the diagnostic conclusion for the load and the event symbol when the pattern overlap exceeds a trend confirmation threshold multiple times consecutively.

[0015] Preferably, the method further includes a diagnostic threshold adaptive calibration step for compensating for changes in the system environment. This step includes: calculating a system entropy baseline characterizing the current background noise level of the power bus during a flight phase quiet period determined by the aircraft flight phase and load operating status; and dynamically adjusting the arbitration threshold based on the ratio of the system entropy baseline to an initial entropy baseline.

[0016] Preferably, the method further includes an active health inspection step for detecting latent faults in a dormant load circuit. This step includes: for a dormant load circuit in a non-operating state, injecting a standardized sub-activation probe pulse with insufficient energy to activate the dormant load circuit by controlling the power controller of the dormant load circuit; simultaneously acquiring and analyzing the current response signal triggered by the standardized sub-activation probe pulse on the power bus to generate a response feature signature characterizing the integrity of the dormant load circuit; and comparing the response feature signature with a pre-stored reference signature corresponding to the health state of the dormant load circuit to determine whether a latent fault exists in the dormant load circuit.

[0017] Preferably, the method further includes an adjunct analysis step for monitoring component performance degradation, the step comprising: performing time smoothing on the standard deviations of multiple consecutive real-time current signals calculated within a first predetermined time window to obtain a real-time performance index characterizing the current continuous entropy level of the load device; comparing the real-time performance index with a pre-stored performance entropy benchmark associated with the load device and its current operating condition; and generating a warning signal indicating performance degradation of the load device when the real-time performance index is continuously higher than a predetermined range above the performance entropy benchmark.

[0018] Preferably, the generated power distribution control commands include three response levels, which are divided according to the probability and severity of electrical fault events determined by the collaborative diagnostic decision. The three response levels include the observation level, which only records the event; the intervention level, which limits the power of non-critical loads; and the emergency level, which executes isolation protection actions.

[0019] Preferably, the current effective load status information is obtained by monitoring load control commands on the aircraft data bus or by monitoring the status signals of the solid-state power controllers that control each load.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] 1. This invention effectively avoids the inherent constraints between safety and availability in existing aircraft power systems by constructing a dual-domain composite entropy feature extraction method and collaborative diagnostic decision-making. Specifically, the calculation of the time-domain entropy transition index enables the system to capture sudden changes in the disorder of the current signal energy domain caused by early physical faults. When this sudden change in energy is significant, the system does not directly identify it as a fault, but further triggers a verification of the phase synchronization index of the signal's time structure domain stability. This dual verification mechanism of the energy domain and time structure domain allows the system to accurately distinguish between disordered oscillations originating from real physical faults and ordered oscillations originating from high-frequency harmonics under extreme operating conditions. Combined with the current effective load operating information, the final collaborative diagnostic decision-making allows the system to exhibit high tolerance to normal high-power load dynamics while maintaining necessary sensitivity to weak fault precursors. Thus, without sacrificing either aspect, it simultaneously improves the system's safety redundancy and mission availability.

[0022] 2. This invention introduces an iterative signature stripping step and a cross-flight trend analysis step, giving the diagnostic system deep analytical capabilities in complex scenarios and predictive capabilities across time dimensions. The iterative signature stripping step enables the system to accurately identify pseudo-abnormal signals composed of multiple superimposed normal events when facing busy conditions with multiple normal loads concurrent, such as takeoffs and landings. This is achieved by logically stripping and residually verifying known normal entropy characteristic patterns, thus avoiding unnecessary false alarms. The cross-flight trend analysis step, by associating low-risk event symbols recorded in a single flight with the flight's unique identification information and calculating their pattern overlap in continuous flight missions, allows seemingly unrelated and isolated weak abnormal signals in a single flight to be chained into a meaningful behavioral chain pointing to the progressive degradation of a specific component. This elevates the diagnostic approach from a passive response to momentary faults to an active prediction of long-term health trends.

[0023] 3. This invention also achieves expanded coverage and precise diagnosis of the health status of the entire aircraft's electrical system throughout its entire lifecycle through an active health inspection step for detecting dormant circuits and an adaptive calibration step for diagnostic thresholds to compensate for changes in the system environment. The active health inspection step utilizes a solid-state power controller to inject standardized sub-activation probe pulses and analyze their response signature characteristics, historically expanding the diagnostic scope from operating load circuits to dormant and backup circuits in a non-operating state, avoiding diagnostic blind spots caused by zero current. Simultaneously, the adaptive calibration step for diagnostic thresholds autonomously calibrates the system's current background noise level during flight-phase quiet periods and dynamically adjusts the arbitration threshold used for diagnostic decisions, ensuring that the accuracy of the diagnostic scale does not drift due to changes in the external electromagnetic environment or the gradual aging of the aircraft. Ultimately, this allows the entire diagnostic system to expand its spatial coverage while maintaining diagnostic consistency and reliability over a long timeframe. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the overall process of an intelligent power distribution and fault diagnosis method for an aircraft power system according to the present invention.

[0025] Figure 2 This is a schematic diagram of the cross-flight trend analysis steps of the present invention;

[0026] Figure 3 This is a flowchart illustrating the iterative signature stripping step in the collaborative diagnostic decision-making process of this invention.

[0027] Figure 4 This is an interactive timing diagram of the dual-domain composite entropy feature extraction process of the present invention.

[0028] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0030] This invention provides an intelligent power distribution and fault diagnosis method for an aircraft power system. Its operation begins with the synchronous acquisition of data from the aircraft power bus, followed by a stability assessment of the signal through the extraction of a dual-domain composite entropy feature, and finally, a collaborative diagnostic decision based on this feature and the real-time operating information of the load. Specifically, the method first performs a data acquisition step to collect the real-time current signal on the power bus and simultaneously acquire the operating information of the current effective load associated with the real-time current signal, providing a basic data stream with physical scenario context for subsequent diagnostic analysis. Subsequently, the system enters a feature extraction stage, which constructs a dual-domain composite entropy feature by calculating a time-domain entropy transition index characterizing the energy domain stability of the real-time current signal and a phase synchronization index characterizing its time structure domain stability. Finally, in a collaborative diagnostic decision step, the system... The extracted dual-domain composite entropy features are compared with a pre-stored knowledge base that associates load conditions with normal entropy feature patterns to distinguish between a normal load dynamic event and a real electrical fault event, and corresponding power distribution control commands are generated accordingly. In the specific high-dynamic operating environment of an aircraft, one challenge faced by its internal power system is to maintain the ability to identify early latent faults such as intermittent weak arcs caused by long-term insulation fatigue or transient resistance changes caused by connector micro-movements while simultaneously dealing with drastic synchronous changes in the load power of multiple subsystems within the normal range during takeoff, landing, or high-maneuver flight. To address this challenge, the method of this invention, after acquiring the real-time current signal of the aircraft power bus, first performs a calculation of the time-domain entropy transition index. This calculation continuously calculates the standard deviation of the real-time current signal in units of a first predetermined time window and obtains the current window calculated value σ of the standard deviation. t The calculated value σ from the previous window t-1 And then through formula H t =σ t / σ t-1 A ratio is calculated as the time-domain entropy transition exponent. This method focuses on the rate of change of signal disorder within adjacent time windows, rather than the absolute magnitude of the signal amplitude. This enables the system to identify abnormal energy dissipation caused by early physical faults that leads to transitions in signal disorder, while also being able to distinguish smooth high-power load startup processes.

[0031] Given that when aircraft engine thrust reaches its maximum under specific operating conditions such as go-around, it may cause high-frequency harmonics with large energy but periodic regularity to be superimposed on the power bus of the generator and its frequency converter controller. The characteristics of these harmonics, as shown in the calculated time-domain entropy transition index, may be similar to random oscillations caused by real physical faults. Therefore, the method of this invention is configured such that the time-domain entropy transition index H... tA phase synchronization index calculation is triggered only when the threshold value is exceeded, serving as a secondary arbitration procedure. This procedure first extracts the oscillation waveform segment of the real-time current signal corresponding to the time-domain entropy transition index exceeding the arbitration threshold, and determines the timestamp sequence of multiple consecutive zero-crossing points of this waveform segment. Subsequently, it calculates the variance σ of the time interval between adjacent timestamps in this timestamp sequence. Δt To quantify the periodic regularity of the oscillation waveform of the real-time current signal, among which Where, Δt i Let N be the number of zero-crossing points in the i-th time interval. The average value across all time intervals is used. By introducing a check on the stability of the signal's time structure domain, the system can distinguish between disordered oscillations originating from real physical faults and ordered oscillations originating from high-frequency harmonics under specific operating conditions, thus improving diagnostic accuracy. The core of collaborative diagnostic decision-making lies in comparing the extracted dual-domain composite entropy features with a pre-stored knowledge base. This knowledge base pre-stores data that associates the dynamic behavior of different loads under various normal operating conditions with corresponding normal entropy feature patterns. During flight phases with intensive electrical events such as takeoff and landing, multiple normal loads often start or run concurrently. Their respective normal entropy transition patterns, when superimposed on the bus, may form a new form not found in the database. To address this concurrent scenario, collaborative diagnostic decision-making... The strategy further includes an iterative signature stripping step for decoupling concurrent normal dynamic events. This step is triggered when a mismatch is detected between the dual-domain composite entropy feature and the knowledge base, and two or more loads are simultaneously in an active state. It selects one load from the multiple concurrent loads and retrieves the normal entropy feature pattern associated with the selected load from the knowledge base. Then, it strips the feature component corresponding to the normal entropy feature pattern of the selected load from the currently extracted dual-domain composite entropy feature to obtain a residual entropy feature. Finally, it compares the residual entropy feature with the normal entropy feature pattern of another concurrent load. If they match, the current event is determined to be an interpretable concurrent normal dynamic event, thereby effectively avoiding misjudgment under busy aircraft operating conditions.

[0032] To achieve trend analysis of progressive failures, the method of this invention further includes a cross-flight trend analysis step, which aims to correlate seemingly random weak anomalous signals scattered across different flights to identify time-series features pointing to the continuous degradation of the same physical component. This step converts diagnostic conclusions indicating low-risk levels generated in a single flight and associated with the same load into an event symbol, and archives them together with the unique identification information of that flight to form a historical event symbol sequence. Subsequently, the system periodically calculates a pattern overlap degree between the event symbol sequence of the latest flight and the historical event symbol sequences of previous flights, for example, using the longest common subsequence algorithm. When the pattern overlap exceeds the trend confirmation threshold multiple times consecutively, the system will automatically upgrade the risk level of the diagnostic conclusion for that load and event symbol, thus providing a technical approach for predictive maintenance. Considering that the overall background electrical noise level of an aircraft may drift under different external electromagnetic environments or due to the aging of the entire aircraft's electrical system, thereby affecting the applicability of fixed diagnostic thresholds, to maintain the accuracy of the diagnostic scale, the method of this invention further includes a diagnostic threshold adaptive calibration step. This step calculates a system entropy baseline σ characterizing the current background noise level of the power bus within a flight phase quiet period jointly determined by the aircraft's flight phase and load operating status. baseline And based on the system entropy baseline and an initial entropy baseline σ initial The ratio, expressed by the formula Threshold adaptive =Threshold preset ×(σ baseline / σ initial By dynamically adjusting the arbitration threshold and other preset thresholds used by the decision-making logic, the diagnostic system can adapt to environmental changes and its own state evolution, thus maintaining diagnostic consistency.

[0033] For critical equipment on aircraft that is in a large number of dormant or backup states, traditional passive monitoring methods have diagnostic limitations due to their zero-current characteristics. To cover this aspect, the method of this invention further includes an active health inspection step. This step targets dormant load circuits in a non-operating state. By controlling the solid-state power controller of the circuit, a standardized sub-activation probe pulse with insufficient energy to activate the dormant load circuit is injected. Simultaneously, the current response signal triggered by the probe pulse on the power bus is collected and analyzed to generate a response feature signature characterizing the integrity of the dormant load circuit. Finally, by comparing the response feature signature with a pre-stored benchmark signature corresponding to the health status of the dormant load circuit, it is determined whether there is a latent fault in the dormant load circuit, thereby extending the diagnostic scope to the entire aircraft electrical system. In addition, to monitor the performance degradation of components rather than direct faults, the method of this invention also includes a companion analysis step, which aims to identify components that, although not faulty, show a downward trend in operating performance or health. This step performs time smoothing on the standard deviation of multiple consecutive real-time current signals calculated within a first predetermined time window to obtain a real-time performance index characterizing the current continuous entropy level of the load device. The absolute magnitude is related to the internal friction and energy dissipation level of the component during operation. Subsequently, the system compares this real-time performance index with a pre-stored performance entropy benchmark associated with the load device and its current operating conditions. When the real-time performance index continuously exceeds a predetermined range of the performance entropy benchmark, an early warning signal indicating performance degradation of the load device is generated, providing data support for condition-based maintenance. As the final output of the diagnostic decision, the generated power distribution control command includes three response levels, which are divided according to the probability and severity of electrical fault events determined by the collaborative diagnostic decision. This includes recording only the event for subsequent analysis. The observation level of trend analysis is an intervention level designed to ensure the operation of the core system by limiting the power of non-critical loads, and an emergency level designed to execute isolation and protection actions when high-risk events such as physical short circuits or severe arcs are detected. The current effective load status information is obtained through redundant paths. The preferred path is to directly monitor the load control commands on the aircraft data bus, such as ARINC429 or AFDX, while the backup path is to monitor the status feedback signals of the solid-state power controllers that control each load. This ensures that the core context information required for the diagnostic system can still be reliably obtained even if there is a communication anomaly in a single information path.

[0034] Example 1: When a wide-body passenger aircraft operating a transoceanic route enters the stable cruise phase in the stratosphere, its power system simultaneously faces two parallel operating conditions. First, due to accumulated fatigue, microscopic cracks appear in the insulation material of the excitation winding of one generator, generating intermittent arcs with low energy lasting several microseconds during normal aircraft vibration. Second, due to high-altitude atmospheric disturbances, the aircraft performs a routine automatic fine-tuning of its attitude, causing multiple flight control surface actuators to synchronously request higher power. The bus current generates a peak within the design safety envelope within milliseconds, which briefly touches the traditional protection threshold set for the galley subsystem. In response to these parallel objective operating conditions, the intelligent power distribution and fault diagnosis method for the aircraft power system of this invention operates as follows: First, for the bus current peak generated by attitude fine-tuning, the system calculates its time-domain entropy transition index H. t At that time, because the power fluctuation, although large in amplitude, is smooth in process, the current window calculated value σ of its current signal standard deviation is... t The calculated value σ from the previous window t-1 The ratio remained around 1, not exceeding the preset arbitration threshold; when the system executes collaborative diagnostic decisions, it will use this H... t The index was compared with the effective load status information of the flight control servo motors obtained by monitoring the aircraft data bus at that time. It was confirmed that the entropy characteristic pattern was consistent with the normal working mode of the load stored in the knowledge base. Therefore, the event was judged as a normal load dynamic event, no power distribution control command was generated, and the power supply in the galley area remained normal.

[0035] Meanwhile, the sharp pulse characteristics of the intermittent weak electric arc generated by the generator excitation winding cause the current signal standard deviation to decrease in the current window calculation value σ. t The instantaneous increase causes the calculated time-domain entropy transition exponent H to... t The value exceeded the arbitration threshold; this value then triggered the calculation of the phase synchronization index. The system extracted the oscillation waveform segment corresponding to this arc and calculated the variance σ of the time interval between its consecutive zero crossings. ΔtSubsequently, a result less than the phase synchronization threshold was obtained, indicating that the oscillation waveform of the signal lacked high periodicity, thus distinguishing it from high-frequency harmonic interference. The time-domain entropy transition index's identification of signal disorder changes provided a prerequisite for verifying the phase synchronization index, while the phase synchronization index's identification of signal periodicity verified the former's identification result. The combination of these two factors led the system to confirm this event as a genuine electrical fault event. In the final stage of collaborative diagnostic decision-making, the system correlated this confirmed genuine electrical fault event with the then-current effective load condition information. Because the event could not be explained by the entropy characteristic pattern of any load in normal working condition, the system characterized it as an anonymous high-entropy event associated with the generator. Based on its energy level and intermittent characteristics, the system generated and executed an observation-level power distribution control command according to the preset response level rules, namely, the event's time, associated load, and H... t The value is recorded without sending an alarm to the cockpit. After the flight mission is completed, the recorded diagnostic conclusion indicating a low-risk level is converted into an event symbol and archived along with the flight's unique identification information into the generator's historical event symbol sequence, providing initial data points for subsequent cross-flight trend analysis. In this way, during a routine flight, the system's availability is not reduced due to normal power fluctuations, while an early physical sign pointing to a progressive failure is accurately captured and archived.

[0036] Example 2: To objectively verify the ability of the method of the present invention to distinguish between high-energy normal load dynamics and low-energy early fault signals, a ground test platform for an aircraft power system was built. The platform includes a programmable DC power supply to simulate the output of an aircraft generator, a standard 28V DC busbar, an aircraft galley oven connected via a solid-state power controller as a normal high-power load, and an arbitrary waveform generator coupled to the busbar via a high-frequency injector to simulate early physical faults. The first predetermined time window used for calculating the time-domain entropy transition index was set to 1 millisecond. This value was determined based on the typical duration of transient events in the monitored signal after balancing the real-time performance of data acquisition with the data processing load. The sampling frequency of the current signal was set to 10MHz to ensure that signal components with frequencies higher than the Nyquist frequency that may be included in the characteristics of early arc faults could be captured.

[0037] The experiment was conducted in two control phases. In the first phase, only the aircraft galley oven was activated, and its operating current steadily rose to the rated value within hundreds of milliseconds. During this process, the time-domain entropy transition index H calculated by the diagnostic system... tThe peak value was below 2.0, which did not trigger subsequent phase synchronization index calculations, and the system did not generate a fault diagnosis conclusion. In the second stage, under the condition of only a stable background load, the system injected a series of simulated arc pulses with a duration of 5 microseconds, energy below 1 millijoule, and random intervals into the busbar through an arbitrary waveform generator. During this process, the average current value of the busbar did not change observably, but the appearance of each pulse caused the time domain entropy to jump exponentially H. t When the instantaneous value reaches 8.0 or higher, it triggers the calculation of the phase synchronization index, and the calculated zero-crossing time difference variance σ Δt All values ​​were less than the preset phase synchronization threshold; Table 1 records the key data for the two stages.

[0038] Table 1: Comparison of diagnostic indicators for normal load dynamics and simulated fault events.

[0039]

[0040] Referring to Table 1, the recorded diagnostic indicators showed differences in the two phases of the experiment; although the kitchen oven start-up event caused a large change in current, it did not lead to a time-domain entropy transition index H due to its smooth process. t A significant increase; in contrast, lower-energy simulated arc pulses, due to the abrupt change they introduce in the energy domain, produce high H. t Peak value; the high H t The phase synchronization index is calculated based on the value trigger, and the result σ Δt The data correspondence, which is less than the threshold, indicates that the method can identify low-energy signals that represent the precursors of real faults from high-energy backgrounds that belong to the normal operating range, while not generating erroneous responses to normal load dynamics.

[0041] Example 3: This example combines Figures 1 to 3 The document describes intelligent power distribution and fault diagnosis methods for aircraft power systems, such as... Figure 1 As shown, it synchronously acquires real-time current and load condition information and transmits this data stream to the dual-domain composite entropy feature extraction step. This step is responsible for calculating the time-domain entropy transition index and phase synchronization index, and its internal arbitration threshold can be dynamically adjusted by a parallel diagnostic threshold adaptive calibration step. The feature extraction result is then sent to the collaborative diagnostic decision step, which compares and queries a knowledge base of pre-stored normal load entropy feature patterns to distinguish between normal dynamics and real faults, and finally leads to the power distribution control command generation step. Based on the diagnostic conclusion, it generates graded response commands. Among them, the archived low-risk events are used as inputs to the cross-flight trend analysis step to predict progressive faults.

[0042] like Figure 2As shown in the figure, the horizontal axis represents the flight sequence arranged in chronological order, the left vertical axis represents the pattern overlap, and the right vertical axis represents the risk level. The pattern overlap curve marked by solid squares in the figure shows that as the flight sequence progresses, its value gradually increases from 0.1 for flight 1 to 0.8 for flight 10. The risk level curve marked by dashed circles corresponds to this. When the pattern overlap first and continuously exceeds the trend confirmation threshold value of 0.5 indicated by the horizontal dashed line at flight 8, the risk level is also raised from medium to high, thus realizing the risk warning of potential failure trends.

[0043] like Figure 3 As shown, upon receiving the dual-domain composite entropy feature and load condition, the system first determines whether the composite feature matches a single pattern in the knowledge base. If it matches, it is directly determined to be a normal load dynamic; if it does not match, it proceeds to the next judgment node: whether multiple loads are operating concurrently. If not, it is directly determined to be a real electrical fault; if so, it triggers the iterative signature stripping step within the dashed box. This step iteratively selects a concurrent load, retrieves its normal entropy feature pattern from the knowledge base, and strips the feature component of the load from the current composite feature to obtain a residual entropy feature. Then, it determines whether the residual entropy feature matches the pattern of another concurrent load. If it matches, it is determined to be a concurrent normal dynamic; if it does not match, it is ultimately determined to be a real electrical fault.

[0044] like Figure 4 As shown, the process begins with a current signal providing a real-time data stream to the time-domain analyzer. The time-domain analyzer calculates the standard deviation between the current and previous window values ​​in a loop and sends this transferred standard deviation data to the entropy calculator, which calculates the time-domain entropy transition index and compares it with an arbitration threshold. If the index does not exceed the threshold, only the time-domain entropy feature is passed to the feature fusion unit and marked as a low-risk feature. If the index exceeds the arbitration threshold, phase synchronization analysis is triggered. At this point, the phase analyzer is activated, extracting oscillation waveform segments, determining zero-crossing timestamps, and calculating the time interval variance to obtain a phase synchronization index. Finally, this index, along with the time-domain entropy transition index, is sent to the feature fusion unit to construct a dual-domain composite entropy feature.

[0045] Example 4: Before deploying the diagnostic method of this invention on a new aircraft model for the first time, an offline system calibration and knowledge base generation process needs to be executed to adapt its diagnostic logic to the electrical characteristics of the aircraft model. This process is executed on a ground engineering workstation, and its input is a dataset containing the power bus current signal and synchronous load condition information recorded from several typical flight missions of the aircraft model. This process aims to initialize and configure the internal key thresholds and knowledge base before the diagnostic system is put into operation. The process first calibrates the key diagnostic thresholds to determine the time-domain entropy transition index H. tThe arbitration threshold is determined by analyzing the current signals during all flight phase silence periods in the dataset and calculating all H values. t The statistical distribution of the values ​​is used to set the initial value of the arbitration threshold as the mean of the distribution plus six times the standard deviation; this is to determine the phase synchronization index σ. Δt The threshold is determined by the system analysis dataset, which contains known signal segments that contain high-frequency harmonics from generators and frequency converters. The corresponding σ is then calculated. Δt The values ​​form the first set of data. At the same time, the signal segment injected with simulated random arc noise is analyzed, and its corresponding σ is calculated. Δt The values ​​form the second set of data, and the phase synchronization threshold is set to a value that minimizes the classification error rate of these two sets of data.

[0046] Next, the process enters the automatic knowledge base generation stage. The system traverses every clearly recorded single load start-up, shutdown, or mode switching event in the dataset. When the system recognizes a load control command for landing gear lowering, it automatically extracts the current signal within a certain period before and after the command takes effect and calculates its corresponding time-domain entropy transition exponent H. t Sequence, and this contains multiple H t The sequence of peak values ​​and their relative time positions, as an entropy feature pattern representing the normal load dynamics of landing gear deployment, is stored in the knowledge base and indexed and associated with the corresponding load and operating condition information. The system repeats this process until all predefined normal entropy feature patterns of important loads in the dataset have been learned and archived. After calibration and learning, the system verifies the operation of the iterative signature stripping step through a replay verification process. The system replays a segment of approach and landing phase data containing two normal dynamic events, landing gear deployment and flap deployment, that partially overlap within a second. It detects a composite entropy feature pattern that cannot be directly matched in the knowledge base. At this point, the iterative signature stripping step is triggered. It first strips the known normal entropy feature pattern component of landing gear deployment from the composite entropy feature, generating a residual entropy feature. Then, it compares the residual entropy feature with the normal entropy feature pattern of flap deployment in the knowledge base. Since the two match in feature morphology, the system finally determines this composite event as an interpretable concurrent normal dynamic event. After this calibration and verification process is completed, a diagnostic system containing statistically calibrated diagnostic thresholds and a knowledge base filled with the aircraft's specific load behavior patterns is initialized, providing a configuration basis for subsequent deployment in actual flight.

[0047] Example 5: After an aircraft completes a Class A or Class C scheduled maintenance, or after replacing core components affecting the overall aircraft electrical baseline, such as generators or main busbars, a recalibration and baseline data update procedure for the diagnostic system must be performed on the ground. Upon initiation of this procedure, the aircraft power system is first run under preset ground silent test conditions to re-acquire and calculate an updated initial entropy baseline σ. initial The system will replace the original stored initial entropy baseline with this value as a reference for subsequent in-flight diagnostic threshold adaptive calibration steps. At the same time, maintenance personnel will sequentially trigger active health inspection steps for all dormant load circuits that are not in operation on the aircraft. Since the components of these circuits are confirmed to be in good condition after scheduled maintenance, the response feature signature of each dormant load circuit collected and generated by the system at this time will be stored in the system's non-volatile storage unit as its initial health reference signature.

[0048] To ensure the comparability of the response signature, it is defined as a vector containing multiple feature components. When a normalized sub-activation probe pulse is injected into the dormant circuit of a newly installed emergency fuel pump, the system synchronously analyzes the weak current response signal induced on the power bus and extracts its time-domain entropy transition exponential peak, response current peak, response duration, and zero-crossing time difference variance σ of the response signal. Δt This constitutes a multi-dimensional response feature signature; based on this, the system calculates and generates a health baseline signature for the fuel pump, which is a signature containing 8.7, 0.5A, 150μs, and 25.4ns. 2 The feature vector consists of four components, which is then bound to the component serial number of the fuel pump and stored. Through this procedure, the baseline data within the diagnostic system is synchronized with the current physical state of the aircraft to maintain the consistency of its diagnostics throughout its service life.

[0049] Example 6: During the final ground testing phase before the aircraft's first operational use, a baseline setting and parameter quantification procedure needs to be executed to initialize predictive functional modules such as cross-flight trend analysis and component performance degradation analysis in the diagnostic system. This procedure aims to establish a safe and conservative working state for these modules that rely on historical data accumulation. First, to establish the performance entropy baseline of key rotating components, the testers sequentially run the main hydraulic pump, environmental control system fan, and other equipment under rated operating conditions for a predetermined time. During this period, the system performs time smoothing on the standard deviations of multiple continuous current signals and obtains a stable value characterizing the continuous entropy level of the component in its new state, which is then stored in the system as its performance entropy baseline.

[0050] Regarding the judgment logic for component performance degradation, the generation condition for its warning signal is quantified as follows: when the real-time performance index is continuously higher than its corresponding performance entropy benchmark by 30% for five consecutive minutes, a warning indicating performance degradation of the load equipment is triggered. For the cross-flight trend analysis module, the procedure defines the pattern overlap as the ratio of the longest common subsequence length between the event symbol sequence of the latest flight and the historical event symbol sequence of previous flights to the length of the event symbol sequence of the latest flight. The trend confirmation threshold is determined through offline simulation. That is, in a synthetic historical event symbol sequence injected with a known progressive failure mode, the minimum pattern overlap value that can stably identify the mode is set as the initial value of the threshold. Through this procedure, the system obtains an initial predictive capability with clear quantitative logic even in the absence of real flight history data.

[0051] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for intelligent power distribution and fault diagnosis of an aircraft power system, characterized in that, The method includes the following steps: Step a: Obtain the real-time current signal on the aircraft power bus and the operating condition information of the current effective load associated with the real-time current signal; Step b: Based on the real-time current signal, perform a dual-domain composite entropy feature extraction. The dual-domain composite entropy feature extraction includes: calculating a time-domain entropy transition index that characterizes the energy domain stability of the real-time current signal. The time-domain entropy transition index is used to quantify the rate of change of disorder of the real-time current signal within adjacent time windows. Step c: When the time domain entropy transition exponent exceeds the arbitration threshold, a phase synchronization index characterizing the stability of the time structure domain of the real-time current signal is triggered and calculated. The phase synchronization index is used to quantify the periodic regularity of the oscillation waveform of the real-time current signal. Step d: Based on the extracted dual-domain composite entropy features and the obtained current effective load operating condition information, a collaborative diagnostic decision is executed. The collaborative diagnostic decision compares the dual-domain composite entropy features with a pre-stored knowledge base that associates load operating conditions with normal entropy feature patterns to distinguish between a normal load dynamic event and a real electrical fault event, and generates power distribution control instructions accordingly. The step of calculating a time-domain entropy transition index that characterizes the energy domain stability of a real-time current signal specifically includes: continuously calculating the standard deviation of the real-time current signal in units of a first predetermined time window; obtaining the current window calculated value of the standard deviation and the previous window calculated value; and using the ratio of the current window calculated value of the standard deviation to the previous window calculated value as the time-domain entropy transition index. The steps for calculating a phase synchronization index characterizing the stability of a real-time current signal in the time structure domain are as follows: The calculation involves calculating the variance of the zero-crossing time difference of the real-time current signal oscillation waveform. Specifically, this calculation includes: extracting the oscillation waveform segment of the real-time current signal corresponding to the time-domain entropy transition exponent exceeding the arbitration threshold; determining the timestamp sequence of multiple consecutive zero-crossing points of the oscillation waveform segment; and calculating the variance of the time interval between adjacent timestamps in the timestamp sequence. ,in ,in, For the first A time interval, For the total number of points that cross zero, The variance is the average across all time intervals; collaborative diagnostic decision-making further includes, if and only if the variance Only when the time-domain entropy transition index is less than the phase synchronization threshold will the event represented by the time-domain entropy transition index be treated as a valid event that may indicate a real electrical fault and processed accordingly.

2. The intelligent power distribution and fault diagnosis method for an aircraft power system according to claim 1, characterized in that, The collaborative diagnostic decision further includes an iterative signature stripping step for decoupling concurrent normal dynamic events. This step is triggered when a mismatch is detected between the dual-domain composite entropy feature and the knowledge base, and two or more loads are simultaneously in an active state. The step includes: selecting one load from multiple concurrent loads and retrieving the normal entropy feature pattern associated with the selected load from the knowledge base; stripping the feature component corresponding to the normal entropy feature pattern of the selected load from the currently extracted dual-domain composite entropy feature to obtain a residual entropy feature; and comparing the residual entropy feature with the normal entropy feature pattern of another concurrent load. If they match, the current event is determined to be an interpretable concurrent normal dynamic event.

3. The intelligent power distribution and fault diagnosis method for an aircraft power system according to claim 1, characterized in that, The method further includes a cross-flight trend analysis step for predicting progressive failures, which includes: converting a diagnostic conclusion indicating a low-risk level generated in a single flight and associated with the same load into an event symbol, and archiving it together with the flight's unique identification information to form a historical event symbol sequence; periodically calculating a pattern overlap between the event symbol sequence of the latest flight and the historical event symbol sequences of previous flights; and automatically increasing the risk level of the diagnostic conclusion for the load and the event symbol when the pattern overlap exceeds a trend confirmation threshold multiple times consecutively.

4. The intelligent power distribution and fault diagnosis method for an aircraft power system according to claim 1, characterized in that, The method further includes a diagnostic threshold adaptive calibration step for compensating for changes in the system environment. This step includes: calculating a system entropy baseline characterizing the current background noise level of the power bus during a flight phase quiet period determined by the aircraft flight phase and load operating status; and dynamically adjusting the arbitration threshold based on the ratio of the system entropy baseline to an initial entropy baseline.

5. The intelligent power distribution and fault diagnosis method for an aircraft power system according to claim 1, characterized in that, The method further includes an active health inspection step for detecting latent faults in a dormant load circuit. This step includes: for a dormant load circuit in a non-operating state, injecting a standardized sub-activation probe pulse with insufficient energy to activate the dormant load circuit by controlling the power controller of the dormant load circuit; simultaneously acquiring and analyzing the current response signal triggered by the standardized sub-activation probe pulse on the power bus to generate a response feature signature characterizing the integrity of the dormant load circuit; and comparing the response feature signature with a pre-stored reference signature corresponding to the health state of the dormant load circuit to determine whether there is a latent fault in the dormant load circuit.

6. The intelligent power distribution and fault diagnosis method for an aircraft power system according to claim 1, characterized in that, The method further includes an adjunct analysis step for monitoring component performance degradation, which includes: performing time smoothing on the standard deviations of multiple consecutive real-time current signals calculated within a first predetermined time window to obtain a real-time performance index characterizing the current continuous entropy level of the load device; comparing the real-time performance index with a pre-stored performance entropy benchmark associated with the load device and its current operating condition; and generating a warning signal indicating performance degradation of the load device when the real-time performance index is continuously higher than a predetermined range above the performance entropy benchmark.

7. The intelligent power distribution and fault diagnosis method for an aircraft power system according to claim 1, characterized in that, The generated power distribution control commands include three response levels, which are determined based on the probability and severity of electrical fault events as determined by collaborative diagnostic decisions. The three response levels are: observation level (which only records the event), intervention level (which limits the power of non-critical loads), and emergency level (which executes isolation protection actions).

8. The intelligent power distribution and fault diagnosis method for an aircraft power system according to claim 1, characterized in that, The current effective load status information is obtained by monitoring load control commands on the aircraft data bus or by monitoring the status signals of the solid-state power controllers that control each load.

Images