Fault association rule mining method for fault data in flight management and control system

Abstract

The invention belongs to the technical field of aircraft prediction and health management, and particularly relates to a fault association rule mining method for fault data in a flight management and control system, which comprises the following steps: S1, extracting specified different batches of fault historical data from a flight management and control fault database; s2, counting the simultaneous occurrence frequency of a pair of associated faults in different batches, and calculating the occurrence frequency of the pair of associated faults; s3, according to a preset frequency threshold value, label pairs, namely fault pairs, with significant association are screened out; s4, extracting a time difference between the label pairs; s5, performing statistical analysis on the time difference distribution between each pair of labels to evaluate the association strength; s6, comparing the time difference between the label pairs with a time difference threshold to obtain the association strength between the fault pairs; and S7, comparing and analyzing time difference distribution of the fault pairs among different batches, observing distribution characteristics of time differences, and judging whether stable association exists or not.

Description

Technical Field

[0001] The present invention belongs to the technical field of aircraft prediction and health management, and in particular relates to a fault association rule mining method for fault data in a flight control system. Background Art

[0002] Aircraft management and flight control systems are key components of modern aircraft and are related to aircraft safety and reliability. Rapidly and accurately locating flight control system faults is of great significance for improving aircraft dispatch rates and reliable operation. However, with the development of aviation science and technology, the interconnection between the flight control system and other external systems has become increasingly complex, and there are many fault propagation paths, which brings great challenges to cascade fault diagnosis.

[0003] In the past, the fault diagnosis of flight control systems mainly relied on manual experience for judgment. These methods are often inefficient and full of uncertainty when processing complex fault data. Specifically, when faced with the coexistence of multiple faults, single fault analysis often fails to reveal potential correlations, resulting in serious limitations on the ability to predict and prevent faults. First, traditional static analysis models often ignore real-time changes in the dynamic flight environment. This means that during the flight, the response speed of fault diagnosis may not keep up with the development of the actual situation, which in turn affects flight safety. For example, if a fault changes during the flight or interacts with other faults, the static model may not be able to identify it in time, resulting in delayed response time. Second, many traditional diagnostic methods do not take into account changes in environmental factors and operating conditions. In actual flight operations, multiple factors such as meteorological conditions, flight altitude, and airborne system status will affect the manifestation and impact of faults. If these factors are not fully incorporated into fault analysis, misjudgment may result. For example, in bad weather, some sensors may behave abnormally, and traditional models may mistake them for faults, increasing the risk of aircraft operation.

[0004] In addition, analysis methods that rely too much on human experience are prone to subjective bias. When pilots or maintenance personnel make judgments based on experience, they may ignore data-driven methods, resulting in reduced decision-making accuracy. This subjectivity not only makes fault diagnosis results inconsistent, but may also lead to inappropriate responses in similar situations.

[0005] In summary, traditional methods mainly rely on manual experience for judgment, which often appear to be inefficient and highly uncertain when processing complex fault data. In particular, when faced with the coexistence of multiple faults, single fault analysis often fails to reveal potential correlations, resulting in limited fault prediction and prevention capabilities. Previous static analysis models have difficulty adapting to real-time changes in dynamic flight environments, resulting in insufficient fault identification and response speed, which in turn affects flight safety. In addition, many traditional diagnostic methods rarely consider changes in environmental factors and operating conditions, which makes it easy to make misjudgments in practical applications and increases the risk of aircraft operations. Summary of the invention

[0006] The problem to be solved by the present invention is that the efficiency and accuracy of fault data analysis in a flight control system are low, and potential fault relationships cannot be quickly identified.

[0007] The technical solution of the present invention:

[0008] A method for mining fault association rules for fault data in a flight control system includes the following steps:

[0009] S1. Extracting designated fault history data of different batches from the flight control failure database;

[0010] S2. Count the frequency of a pair of related faults occurring simultaneously in different batches, and calculate the frequency of occurrence of the pair of related faults;

[0011] S3. According to a preset frequency threshold, filter out label pairs with significant association, i.e., fault pairs;

[0012] S4, extracting the time difference between the tag pairs;

[0013] S5, performing statistical analysis on the distribution of the time difference between each pair of tags to evaluate their association strength;

[0014] S6. Compare the time difference between the tag pairs with the time difference threshold to obtain the correlation strength between the fault pairs;

[0015] S7. Compare and analyze the time difference distribution of fault pairs between different batches, observe the distribution characteristics of the time difference, and determine whether there is a stable correlation.

[0016] Further, step 1 specifically includes the following steps:

[0017] S11, reading the fault type code, timestamp and fault description information from the flight control fault record database;

[0018] S12. Remove invalid or duplicate data and fill in missing values ​​to ensure data integrity.

[0019] Furthermore, in step 2, it also includes:

[0020] According to the statistical results of the fault history data, an initial frequency threshold is set; based on the real-time monitoring and analysis results, the flight control system can adjust the frequency threshold; if the frequency of the current associated fault is significantly higher than the historical average level, the system will automatically increase the frequency threshold to reduce the false alarm rate; otherwise, the frequency threshold will be lowered to increase sensitivity;

[0021] When new fault data is recorded, the flight control system will automatically update the historical database and recalculate the frequency of faults, allowing the system to continuously learn and adapt to new data patterns;

[0022] The flight control system allows the user to manually set or adjust the frequency threshold.

[0023] Furthermore, step 3 also includes the following steps:

[0024] S31, calculating the relevance of the label pairs: using statistical methods to calculate the frequency and relevance of each pair of labels in the fault data;

[0025] S32, recording the number of simultaneous appearances of each pair of tags and the number of separate appearances of each pair of tags;

[0026] S33, comparing the calculated correlation degree with a preset frequency threshold, and retaining only those tag pairs whose correlation degree is higher than the frequency threshold;

[0027] S34. Arrange the filtered label pairs into a set.

[0028] Furthermore, step 3 also includes step S35, dynamically adjusting the frequency threshold: regularly evaluating and adjusting the frequency threshold according to actual application effects and data changes to optimize the selection of tag pairs and improve the sensitivity of fault detection.

[0029] Further, step 4 specifically includes:

[0030] First, specific labels are identified from the event sequence;

[0031] Then, the time difference between the tag pairs is calculated based on the sequence of their identifiers.

[0032] Furthermore, S5 performs statistical analysis on the time difference distribution between each pair of tags, including the following analysis methods:

[0033] S51. Evaluate the linear or nonlinear relationship between the time difference and the frequency of the fault event by calculating the Pearson correlation coefficient or the Spearman rank correlation coefficient;

[0034] S52, quantify the association strength and evaluate the association strength: calculate the average value of the time difference between each pair of tags to understand the typical time interval of the fault event; calculate the standard deviation of the time difference to evaluate the volatility of the time difference;

[0035] S53, after performing the association strength evaluation, analyzing the obtained average time difference and standard deviation, and determining a similarity threshold based on the average time difference and the standard deviation;

[0036] S54. Set a time difference threshold. Any time difference below this time difference threshold is considered a significant association, and statistical tests are used to verify whether there is significance; any time difference above this threshold is considered a weak association.

[0037] Further, in step S52, a smaller standard deviation indicates that the time difference between events is relatively consistent, while a larger standard deviation may indicate that there are larger differences between different events; the median and quantile are used to fully reflect the distribution characteristics of the time difference.

[0038] Further, step S7 also includes the following steps:

[0039] S71. Observe the distribution characteristics of the time difference, extract fault pairs between different batches from the fault data set, record the occurrence time of each fault in the fault pair, and calculate the time difference between the fault pairs. This step is achieved by the difference in timestamps to form a time difference data set.

[0040] S72. Periodicity analysis: Use spectrum analysis technology to process the time difference data and identify periodic patterns: By analyzing the main frequency in the frequency domain, it can be determined whether the fault pair presents periodic characteristics. The autocorrelation function and partial autocorrelation function are used to evaluate the periodicity of the time difference data to determine whether there are periodic fault pairs.

[0041] S73. Symmetry measurement: By plotting the distribution of time differences, observing their morphological characteristics, and analyzing the symmetry between the left and right sides; using statistical methods to quantify the symmetry and evaluate whether the distribution of time differences conforms to the normal distribution or other known distributions;

[0042] S74. Clustering phenomenon analysis: Use clustering analysis technology to group time difference data to find similarities and differences between different batches, and identify high-incidence and low-incidence time periods of fault pairs through clustering; use visualization tools to display clustering results;

[0043] S75. Stability evaluation: Combined with the results of time difference analysis, potential fault association rules are mined through Apriori algorithm or FP-Growth algorithm to evaluate the association strength between fault pairs in different batches and the stability of their time distribution.

[0044] The advantages of the present invention compared with the prior art are:

[0045] High efficiency:

[0046] By counting and analyzing the frequency of faults, we can quickly filter out related fault tags and significantly improve the efficiency of fault data analysis.

[0047] accuracy:

[0048] Using time difference distribution analysis, it provides precise correlation strength assessment and can more accurately identify potential fault relationships than traditional methods.

[0049] Dynamic adaptability:

[0050] By comparing fault data from different batches, it has strong adaptability and can identify fault correlations under different production conditions.

[0051] Clear criteria for determining association:

[0052] Clear time difference distribution and threshold settings standardize the association judgment process and facilitate implementation and verification.

[0053] Multi-dimensional analysis:

[0054] It also considers periodicity, symmetry and clustering phenomena to provide a more comprehensive fault correlation analysis.

[0055] Reduce the risk of misjudgment:

[0056] The use of dense time differences and irregular judgment standards reduces the risk of misjudgment caused by data fluctuations.

[0057] Compared with previous technologies, the present invention has made significant breakthroughs in efficiency and accuracy. First, it makes potential fault identification more scientific and reliable by comprehensively extracting historical data and quantifying fault co-occurrence relationships. Secondly, the introduced significance screening mechanism focuses on the most risky fault combinations, which improves the priority of problem identification. At the same time, time difference analysis reveals the causal relationship between faults, providing deeper insights for fault diagnosis. In addition, stability analysis determines the reliability of fault relationships and provides support for long-term monitoring. These advantages not only optimize fault management and maintenance decisions, but also enhance the safety and reliability of aircraft, and contribute to safe operation in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is a flow chart of the method of embodiment 2 of the present invention;

[0059] Figure 2 This is a schematic diagram of fault pair frequency calculation in Embodiment 2 of the present invention;

[0060] Figure 3 This is a schematic diagram of time distribution morphology calculation of Embodiment 2 of the present invention; DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] One embodiment of the present invention is,

[0063] A method for mining fault association rules for fault data in a flight control system includes:

[0064] S1. Extract designated batches of fault history data from the flight control fault database to ensure the integrity and accuracy of the basic information for analysis.

[0065] In a more specific embodiment, the S1 step includes:

[0066] S11. Read the fault class code (fault identifier), timestamp and fault description information from the flight control system fault record database.

[0067] S12. Remove invalid or duplicate data and fill in missing values ​​to ensure data integrity. Standardize the data to eliminate the impact of noise data.

[0068] The present invention ensures the basicity and accuracy of the data by extracting information such as fault class codes and timestamps from the flight control system fault record database. By removing invalid or duplicate data, filling missing values ​​and performing standardization, this step not only improves the integrity and comparability of the data, but also eliminates interference between different dimensions. The significance of this process is that it lays a solid data foundation for subsequent association rule mining, enhances the reliability and scientificity of the analysis results, and effectively supports fault diagnosis and prediction, and improves the safety and operating efficiency of the flight control system.

[0069] S2. Count the co-occurrence frequencies of a pair of related faults in different batches and calculate the frequency of occurrence of the pair of related faults. This step is used to quantify the co-occurrence relationship between faults to identify potential fault associations.

[0070] In a more specific embodiment, the S2 step includes:

[0071] S21. According to the statistical results of the historical fault data, an initial frequency threshold is set. The frequency threshold can be used as a fault correlation benchmark. For example, the frequency threshold can be set by adding a certain standard deviation to the average value of the historical fault occurrence rate. In actual applications, the system will monitor the current operating environment and fault occurrence in real time, and compare and analyze with historical data.

[0072] S22. When new fault data is recorded, the system will automatically update the historical database and recalculate the frequency of faults, so that the system can continuously learn and adapt to new data patterns.

[0073] S23. In addition, the flight control system also allows users to manually set or adjust the frequency threshold so as to optimize it according to specific business needs or operating experience. This flexibility enables the system to adapt to the fault management needs in different scenarios.

[0074] The present invention provides a basis for dynamic adjustment of the fault warning system by setting an initial percentage threshold and monitoring the occurrence of faults in real time. This step uses statistical analysis of historical fault data to set an initial threshold based on the average fault occurrence rate and its standard deviation, thereby laying a scientific basis for fault warning. The system can automatically update the historical database when recording new fault data and recalculate the percentage of fault occurrence, reflecting the adaptive learning ability. In addition, allowing users to manually adjust the threshold enhances the flexibility of the system to meet the needs of specific business scenarios. This design not only improves the accuracy and reliability of fault warning, but also optimizes the fault management strategy of the flight control system, thereby improving overall safety and operational efficiency.

[0075] In a more specific solution, step S22 further includes:

[0076] S221. Based on the real-time monitoring and analysis results, the system can adjust the frequency threshold. If the frequency of the current associated fault is significantly higher than the historical average, the system will automatically increase the frequency threshold to reduce the false alarm rate; otherwise, the frequency threshold can be lowered to increase sensitivity.

[0077] Through real-time monitoring and analysis, the system can automatically adjust the percentage threshold to adapt to changes in the current fault frequency. When the fault frequency is significantly higher than the historical average, the system will increase the threshold to reduce the false alarm rate; when the fault frequency is low, the threshold may be lowered to increase the sensitivity of fault detection. This dynamic adjustment mechanism not only enhances the accuracy and reliability of fault warning and effectively reduces unnecessary interference, but also improves the system's ability to respond to potential faults, ensuring efficient fault management under different operating conditions.

[0078] S3. According to the preset frequency threshold, filter out the label pairs with significant correlation, i.e., the fault pairs. This process will help focus on those combinations that show strong correlation in the occurrence of faults.

[0079] In a more specific solution, step S3 further includes:

[0080] S31. Calculate the correlation between label pairs: Use statistical methods (such as Pearson correlation coefficient or chi-square test) to calculate the frequency and correlation of each pair of labels in the fault data.

[0081] S32. Record the number of simultaneous occurrences of each pair of tags and the number of separate occurrences of each pair of tags, so as to calculate the true correlation index.

[0082] S33, comparing the calculated correlation with a preset frequency threshold, and retaining only those tag pairs whose correlation is higher than the frequency threshold, thereby focusing on combinations showing stronger correlation.

[0083] S34. Arrange the screened label pairs into a set to provide a basis for subsequent fault analysis, prediction model training or decision support.

[0084] S35. Dynamically adjust the frequency threshold: Regularly evaluate and adjust the frequency threshold based on actual application effects and data changes to optimize the selection of tag pairs and improve the sensitivity of fault detection.

[0085] The present invention calculates the correlation of label pairs, and the system uses statistical methods to analyze the frequency and correlation of labels in fault data to identify label combinations that show strong correlation. This process includes recording the number of times a label appears at the same time and comparing it with the frequency of its individual appearance to derive a true correlation index. Retaining only label pairs above a preset threshold helps focus on the most informative combinations, laying the foundation for subsequent fault analysis, predictive model training, and decision support. The option to dynamically adjust thresholds further enhances the flexibility of the system, enabling it to adapt to data changes, thereby optimizing the selection of label pairs, improving the sensitivity and accuracy of fault detection, and ultimately improving the overall efficiency of fault management.

[0086] S4. Apply the nearest neighbor principle to extract the time difference between the label pairs to analyze the occurrence sequence and time interval of these fault events, thereby revealing the potential causal relationship of the fault occurrence.

[0087] In a more specific solution, step S4 further includes:

[0088] First, identify specific label pairs, such as AB or BA, from the event sequence;

[0089] Then, the time difference between the tag pairs is calculated according to the order of their identifiers. If the identifier A is in the front, the time difference between each A and the nearest B after it is calculated;

[0090] If the identifier B comes first, calculate the time difference from each B to the nearest A.

[0091] The present invention reveals the causal relationship and impact sequence between different events in this way, providing an important basis for understanding the causes and consequences of failures. Its characteristic is that through accurate time difference calculation, it can provide quantitative data support for the mutual influence between events, thereby enhancing the accuracy of fault prediction and analysis. At the same time, this analysis helps to identify potential patterns and laws, optimize maintenance strategies, and improve the overall reliability and responsiveness of the system.

[0092] S5. Perform statistical analysis on the time difference distribution between each pair of tags to evaluate their association strength and provide a quantitative basis for establishing a fault association model.

[0093] In a more specific solution, step S5 further includes:

[0094] S51. Evaluate the linear or nonlinear relationship between the time difference and the frequency of the fault event by calculating the Pearson correlation coefficient or the Spearman rank correlation coefficient. This can reveal whether there is a statistically significant association between the two fault events.

[0095] S52. Quantify the strength of association and evaluate the strength of association: Calculate the average of the time difference between each pair of tags to understand the typical time interval of the fault event; calculate the standard deviation of the time difference to evaluate the volatility of the time difference. A smaller standard deviation indicates that the time difference between events is more consistent, while a larger standard deviation may indicate that there are large differences between different events. Indicators such as the median and quantile (such as quartiles) can be used to fully reflect the distribution characteristics of the time difference.

[0096] S53, after evaluating the strength of association, analyzing the obtained statistical indicators (average time difference and standard deviation), and determining a similarity threshold based on these indicators. For example, the average time difference of different fault event pairs can be observed to identify which pairs have significantly lower time differences than other pairs.

[0097] S54. To ensure effective fault prediction, a time difference threshold is set. Any time difference below this time difference threshold is considered to be significantly correlated. A statistical test (such as a t-test or a chi-square test) is used to verify whether there is significance. Any time difference above this threshold is considered to be weakly correlated.

[0098] The present invention evaluates the relationship between time difference and the frequency of fault events by calculating the Pearson correlation coefficient or the Spearman rank correlation coefficient, which can effectively reveal the linear or nonlinear association between the two. Then, a linear regression or multivariate regression model is used to analyze the degree of influence of the time difference on the occurrence of fault events, thereby constructing a regression equation to evaluate the effectiveness of the time difference in fault prediction. This quantitative analysis method can quantify the strength of association, calculate the average time difference and standard deviation between each pair of labels, and gain an in-depth understanding of the typical time intervals and volatility of fault events. For example, a smaller standard deviation indicates that the time difference between events is more consistent, while a larger standard deviation may reflect that there are large differences between different events.

[0099] In addition, by analyzing these statistical indicators and identifying the average time difference that is significantly lower than other pairs, a reasonable threshold can be set to identify significantly associated event pairs. Finally, a statistical test (such as a t-test or a chi-square test) is used to verify whether the time difference is significant, so that the time difference above this threshold is considered to be weakly associated. The significance of this series of steps is to improve the accuracy and reliability of fault prediction, make maintenance strategies more targeted, and thus improve the overall performance and responsiveness of the system, providing solid data support for fault prevention and decision-making.

[0100] S6. If the time difference between the tag pairs meets the time difference threshold, it is determined that there is an association, wherein:

[0101] S61. If the time difference distribution is highly concentrated and the time interval is short, it is determined to be a strong correlation, indicating that there may be a direct causal relationship between this pair of faults.

[0102] S62. If the time difference distribution is loose and lacks regularity, it is judged as a weak correlation, indicating that the relationship between this pair of faults is relatively indirect.

[0103] In a more specific solution, step S6 also includes the following contents:

[0104] S63. Further analysis of strong and weak correlations. Strongly correlated label pairs usually appear at high frequencies. For example, when a fault occurs, the probability of both occurring at the same time is much higher than the random expectation. By constructing a fault tree, the fault events of the system are decomposed layer by layer, the various causes of the fault are identified, and the potential root causes are found.

[0105] S64, Failure Mode and Effects Analysis (FMEA): Evaluate the impact, frequency and detectability of each failure mode through correlation strength to help formulate targeted prevention and repair measures. The final analysis report will include possible cause-and-effect diagrams, root cause analysis and recommended improvement measures.

[0106] S65. Define weak correlation. Weakly correlated label pairs appear less frequently in fault data, or their probability of co-occurrence is close to the random level, but there is still a certain correlation. Collect data on environmental variables, operating conditions, or other external factors related to the occurrence of faults in order to fully understand how these factors affect fault events. Use statistical tools such as correlation analysis and regression analysis to explore the impact of these external factors and environmental variables on weakly correlated faults.

[0107] S66. Store the analysis results in a fault association database, including strongly associated and weakly associated label pairs, analysis results, related environmental data, etc. This database will provide a basis for subsequent data mining and support model updating and decision making.

[0108] S67. Regularly review new fault data, analyze the consistency between actual faults and judgment results, and adjust the judgment threshold in a timely manner, such as increasing the threshold to filter noise, or lowering the threshold to capture more potential correlations.

[0109] The present invention conducts in-depth analysis of strongly correlated and weakly correlated fault label pairs, which can significantly improve the effectiveness of fault management and preventive measures. Strongly correlated label pairs usually show significant frequency and patterns when faults occur, which enables us to decompose fault events layer by layer by building a fault tree, identify various potential causes, and thus find the root cause. This process not only helps us understand the specific mechanism of the fault, but also provides a basis for formulating effective repair plans.

[0110] At the same time, although weakly correlated label pairs appear less frequently in fault data and their probability of co-occurrence is close to the random level, they still have important analytical value. By collecting data on environmental variables, operating conditions and other external factors related to the occurrence of faults, researchers can fully understand how these factors affect the occurrence of fault events. Using statistical tools such as correlation analysis and regression analysis, we can deeply explore the impact of these external factors on weakly correlated faults, thereby providing a new perspective for fault prevention.

[0111] These analysis results are stored in the fault association database, which includes not only strongly associated and weakly associated label pairs, but also analysis results, related environmental data and other information. This database provides a solid foundation for subsequent data mining and supports the continuous updating of models and decision-making. By regularly reviewing new fault data, researchers can analyze the consistency between actual faults and judgment results, and adjust the judgment threshold in a timely manner, such as increasing the threshold to filter noise, or lowering the threshold to capture more potential associations, thereby improving the accuracy of fault prediction and system responsiveness.

[0112] S7. Compare and analyze the time difference distribution of fault pairs between different batches, and observe the distribution characteristics of the time difference, such as periodicity, symmetry and clustering, to determine whether there is a stable association.

[0113] In a more specific embodiment, step S7 further includes the following steps:

[0114] S71. Observe the distribution characteristics of the time difference, extract fault pairs between different batches from the fault data set, and record the occurrence time of each fault in the fault pair. Calculate the time difference between the fault pairs (such as the interval between the occurrence of the faults). This process can be achieved through the difference of the timestamp to form a time difference data set.

[0115] S72. Periodicity analysis: Use spectrum analysis techniques such as Fourier transform to process time difference data and identify periodic patterns: By analyzing the main frequency in the frequency domain, it can be determined whether the fault pair exhibits periodic characteristics. The autocorrelation function (ACF) and partial autocorrelation function (PACF) are used to evaluate the periodicity of time difference data to determine whether there are periodic fault pairs, thereby providing a basis for the fault warning system.

[0116] S73. Symmetry measurement: By drawing a distribution graph of the time difference (such as a histogram or density curve), observe its morphological characteristics and analyze the symmetry between the left and right sides. Use statistical methods (such as the Kolmogorov-Smirnov test) to quantify the symmetry and evaluate whether the distribution of the time difference conforms to the normal distribution or other known distribution.

[0117] S74. Clustering phenomenon analysis: Use clustering analysis techniques (such as K-means clustering, DBSCAN, etc.) to group time difference data to find similarities and differences between different batches. Clustering can identify high-incidence and low-incidence time periods for fault pairs. Use visualization tools such as scatter plots and heat maps to display clustering results, helping analysts to more intuitively understand the distribution characteristics of fault pairs and support decision making.

[0118] S75, stability assessment, combined with the results of time difference analysis, through the Apriori algorithm or FP-Growth algorithm to mine potential fault association rules, evaluate the strength of association between fault pairs in different batches and the stability of their time distribution. Establish a long-term monitoring mechanism, regularly perform the same time difference analysis on new data to verify previous conclusions and adjust the fault prediction model in a timely manner.

[0119] The present invention can deeply reveal the laws and patterns of fault occurrence between different batches by systematically analyzing the time difference of fault events. This analysis process first calculates the interval time of fault events to form a time difference data set, laying the foundation for subsequent periodic analysis. Using spectrum analysis techniques such as Fourier transform, the periodic characteristics in the time difference data can be identified, and its periodicity can be evaluated by autocorrelation function (ACF) and partial autocorrelation function (PACF), so as to determine whether there are periodic fault events, which provides an important basis for the fault warning system.

[0120] In addition, by plotting the distribution of the time difference and applying statistical methods (such as the Kolmogorov-Smirnov test), the symmetry of the time difference data can be quantified to analyze whether it conforms to a normal distribution or other known distributions. This symmetry measure helps to understand the essential characteristics of the fault event.

[0121] Cluster analysis techniques (such as K-means clustering, DBSCAN, etc.) further group the time difference data to find the similarities and differences between different batches. This process can clearly identify the high-incidence and low-incidence time periods of failure events, and through visualization tools such as scatter plots and heat maps, the analysis results are more intuitive and support decision-making.

[0122] Finally, combined with the results of time difference analysis, the Apriori algorithm or FP-Growth algorithm is used to mine potential fault association rules, which can evaluate the stability of different batches and their association strength, establish a long-term monitoring mechanism, and regularly perform the same time difference analysis on new data to verify previous conclusions and adjust the fault prediction model in a timely manner.

[0123] In summary, the present invention can reveal the potential causal relationship between faults by accurately quantifying the time interval between faults, surpassing the limitations of traditional static association rules. This feature enables researchers to gain insight into the dynamic evolution of fault sequences and identify possible fault propagation paths, thereby providing a theoretical basis for preventive maintenance of the system. In addition, time difference analysis provides strong support for the timeliness assessment of fault modes, making the reliability research of flight control systems in complex environments more in-depth. By analyzing the time distribution of fault events in historical data, this method not only improves the sensitivity of fault detection, but also enhances the ability to predict future fault trends.

[0124] The second embodiment of the present invention:

[0125] like Figure 1 As shown, a fault association rule mining method for fault data in a flight control system includes:

[0126] S1. Extracting a specified batch of fault history data from the flight control system fault database. In this embodiment, the extracted data includes the timestamp of the fault occurrence, the fault code, and other related data. These data will form the basis for subsequent analysis.

[0127] S2. Count the frequency of simultaneous occurrence of two related faults in different batches, and calculate the percentage of the total batches. In this embodiment, after extracting the required fault history data from the flight control fault database, a frequency matrix is ​​created based on the data, where the rows represent batches and the columns represent fault types. For each batch, the number of times the fault pair occurs simultaneously is recorded. Traverse the extracted data and count the number of times the related fault pair occurs simultaneously in each batch.

[0128] like Figure 2 As shown in the figure, if fault X and fault Y occur simultaneously in batch A, the corresponding position in the frequency matrix is ​​increased by 1. The total number of all extracted batches is counted for subsequent percentage calculation. For each batch, the percentage of the frequency of the associated fault pairs occurring simultaneously in the total number of occurrences in the batch is calculated.

[0129] The calculation formula is:

[0130] Percentage = (number of occurrences / total number of batches) × 100%

[0131] S3. According to the preset percentage threshold, filter out the label pairs with significant association. In this embodiment, we filter out the label pairs with significant association according to the preset percentage threshold (for example, 10%). First, count the frequency of all label pairs appearing at the same time, and calculate the ratio of the number of occurrences of each pair of labels A and B to the total number of label occurrences. Then, compare these frequencies with the preset threshold. If the frequency of a label pair exceeds the threshold, the label pair is considered to have a significant association. Finally, all label pairs that meet the conditions are recorded for subsequent analysis and application, so as to identify potential failure modes or influencing factors.

[0132] S4. Apply the nearest neighbor principle to extract the time difference between the tag pairs. In this embodiment, the nearest neighbor principle is applied to extract the time difference between the tag pairs. The specific steps are as follows: First, for each tag pair A and B, record the timestamp of their first appearance. Then, calculate the difference between the two timestamps to obtain the time difference. Subsequently, the time differences of all tag pairs are summarized.

[0133] S5. Perform statistical analysis on the distribution of time differences between each pair of labels. In this embodiment, statistical methods such as histograms or density plots are used to display the distribution of time differences so as to visually analyze their characteristics. Then, key statistical indicators, including mean, median, standard deviation, and percentiles, are calculated to quantify the central tendency and dispersion of the time differences. Finally, through hypothesis testing or correlation analysis, it is evaluated whether there is a significant difference in the time difference between different label pairs, thereby providing data support for understanding the relationship between labels and their potential influencing factors.

[0134] S6. If the time difference between the tag pairs meets the preset value, it is determined that there is an association. In this embodiment, if the time difference between the tag pairs meets the preset value, it is determined that there is an association. Specifically, if the time difference distribution is highly concentrated and the time interval is short, it is determined to be a strong association, indicating that there may be a direct causal relationship between the fault pairs; conversely, if the time difference distribution is loose and lacks regularity, it is determined to be a weak association, indicating that the relationship between the fault pairs is relatively indirect.

[0135] S7. When comparing different batches, observe the distribution characteristics of the time difference, such as periodicity, symmetry and clustering. Figure 3 As shown, in this embodiment, a spectrum analysis method such as Fourier transform is used to detect whether there are periodic features in the time difference data to determine the regularity of fault occurrence and the stability of the time interval. A statistical test method is used to evaluate the symmetry of the time difference to determine whether its distribution presents obvious symmetric characteristics, thereby identifying potential systematic fault modes. A clustering algorithm (such as K-means or DBSCAN) is applied to classify the time difference data to identify the degree of association and similarity between different fault types, further revealing the potential connection between faults.

[0136] The above is only a specific embodiment of the present invention, and the present invention is described in detail. The unexplained part is a conventional technology. However, the protection scope of the present invention is not limited to this. Any changes or substitutions that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for mining fault association rules for fault data in a flight control system, characterized in that: The following steps are involved: S1. Extracting designated fault history data of different batches from the flight control failure database; S2. Count the frequency of a pair of related faults occurring simultaneously in different batches, and calculate the frequency of occurrence of the pair of related faults; S3. According to a preset frequency threshold, filter out label pairs with significant association, i.e., fault pairs; S4, extracting the time difference between the tag pairs; S5, performing statistical analysis on the distribution of the time difference between each pair of tags to evaluate their association strength; S6. Compare the time difference between the tag pairs with the time difference threshold to obtain the correlation strength between the fault pairs; S7. Compare and analyze the time difference distribution of fault pairs between different batches, observe the distribution characteristics of the time difference, and determine whether there is a stable correlation.

2. The fault association rule mining method for fault data in a flight control system according to claim 1, characterized in that: Step 1 specifically includes the following steps: S11, reading the fault type code, timestamp and fault description information from the flight control fault record database; S12. Remove invalid or duplicate data and fill in missing values ​​to ensure data integrity.

3. The fault association rule mining method for fault data in a flight control system according to claim 2, characterized in that: In step 2, it also includes: According to the statistical results of the fault history data, an initial frequency threshold is set; based on the real-time monitoring and analysis results, the flight control system can adjust the frequency threshold; if the frequency of the current associated fault is significantly higher than the historical average level, the system will automatically increase the frequency threshold to reduce the false alarm rate; otherwise, the frequency threshold will be lowered to increase sensitivity; When new fault data is recorded, the flight control system will automatically update the historical database and recalculate the frequency of faults, allowing the system to continuously learn and adapt to new data patterns; The flight control system allows the user to manually set or adjust the frequency threshold.

4. The fault association rule mining method for fault data in a flight control system according to claim 3, characterized in that: Step 3 also includes the following steps: S31, calculating the relevance of the label pairs: using statistical methods to calculate the frequency and relevance of each pair of labels in the fault data; S32, recording the number of simultaneous appearances of each pair of tags and the number of separate appearances of each pair of tags; S33, comparing the calculated correlation degree with a preset frequency threshold, and retaining only those tag pairs whose correlation degree is higher than the frequency threshold; S34. Arrange the filtered label pairs into a set.

5. The fault association rule mining method for fault data in a flight control system according to claim 4, characterized in that: Step 3 also includes step S35, dynamically adjusting the frequency threshold: regularly evaluating and adjusting the frequency threshold according to actual application effects and data changes to optimize the selection of tag pairs and improve the sensitivity of fault detection.

6. The fault association rule mining method for fault data in a flight control system according to claim 5, characterized in that: Step 4 specifically includes: First, specific labels are identified from the event sequence; Then, the time difference between the tag pairs is calculated based on the sequence of their identifiers.

7. The fault association rule mining method for fault data in a flight control system according to claim 6, characterized in that: S5 performs statistical analysis on the time difference distribution between each pair of tags, including the following analysis methods: S51. Evaluate the linear or nonlinear relationship between the time difference and the frequency of the fault event by calculating the Pearson correlation coefficient or the Spearman rank correlation coefficient; S52, quantifying the association strength and performing association strength evaluation: calculating the average of the time difference between each pair of tags to understand the typical time interval of the fault event; Calculate the standard deviation of the time difference to assess the volatility of the time difference; S53, after performing the association strength evaluation, analyzing the obtained average time difference and standard deviation, and determining a similarity threshold based on the average time difference and the standard deviation; S54. Set a time difference threshold. Any time difference below this time difference threshold is considered a significant association, and statistical tests are used to verify whether there is significance; any time difference above this threshold is considered a weak association.

8. The method for mining fault association rules for fault data in a flight control system according to claim 7, characterized in that: In step S52, a smaller standard deviation indicates that the time difference between events is relatively consistent, while a larger standard deviation may indicate that there are larger differences between different events; the median and quantile are used to fully reflect the distribution characteristics of the time difference.

9. The method for mining fault association rules for fault data in a flight control system according to claim 8, characterized in that: Step S7 also includes the following steps: S71. Observe the distribution characteristics of the time difference, extract fault pairs between different batches from the fault data set, record the occurrence time of each fault in the fault pair, and calculate the time difference between the fault pairs. This step is achieved by the difference in timestamps to form a time difference data set. S72. Periodicity analysis: Use spectrum analysis technology to process the time difference data and identify periodic patterns: By analyzing the main frequency in the frequency domain, it can be determined whether the fault pair presents periodic characteristics. The autocorrelation function and partial autocorrelation function are used to evaluate the periodicity of the time difference data to determine whether there are periodic fault pairs. S73. Symmetry measurement: By plotting the distribution of time differences, observing their morphological characteristics, and analyzing the symmetry between the left and right sides; using statistical methods to quantify the symmetry and evaluate whether the distribution of time differences conforms to the normal distribution or other known distributions; S74. Clustering phenomenon analysis: Use clustering analysis technology to group time difference data to find similarities and differences between different batches, and identify high-incidence and low-incidence time periods of fault pairs through clustering; use visualization tools to display clustering results; S75. Stability evaluation: Combined with the results of time difference analysis, potential fault association rules are mined through Apriori algorithm or FP-Growth algorithm to evaluate the association strength between fault pairs in different batches and the stability of their time distribution.

Images