Artificial intelligence-based method for analyzing reliability of aviation equipment

Abstract

This invention provides an artificial intelligence-based reliability analysis method for aviation equipment, comprising the following steps: S1, analyzing the function and structural composition of intelligent aviation equipment; S2, constructing a reliability index system and parameter model for intelligent aviation equipment; S3, based on the reliability parameters and index definitions of intelligent aviation equipment, determining the index dimensions, data sources, and data types for data collection; S4, conducting failure mode, cause, and mechanism analysis based on the data obtained in step S3 and recording relevant data; S5, using the reliability parameter index model constructed in step S2 to perform reliability analysis on the intelligent aviation equipment and the data obtained in step S4. This method provides a reliability analysis process and framework for novel intelligent aviation equipment, ensuring the reliability of intelligent aviation equipment and providing quantitative standards for reliability analysis of intelligent aviation equipment.

Description

Technical Field

[0001] This invention relates to the field of intelligent aviation equipment reliability engineering, and specifically to an artificial intelligence-based method for aviation equipment reliability analysis. Background Technology

[0002] With the development of technologies such as artificial intelligence and network communication, AI equipment is widely used in autonomous vehicles, industrial robots, aircraft autopilot systems, and drones. As the concept of system-of-systems warfare deepens, intelligent equipment has brought profound changes to combat missions, formations, and combat styles, giving rise to swarm and wolfpack combat formations. However, deep neural network algorithms are highly dependent on data; functions such as intelligence gathering, target recognition, and autopilot are easily affected by noise, leading to unpredictable errors and potentially causing serious consequences. This poses a significant challenge to the reliability of intelligent equipment.

[0003] An intelligent system is an automaton system that integrates intelligent modules and uses one or more functions such as perception and recognition, knowledge and learning, prediction and decision-making, and action to predict, suggest, or make decisions about target content.

[0004] Unlike traditional equipment reliability design analysis, AI-based aerospace equipment, due to its intelligence and autonomy, is highly dependent on data and AI algorithms, and therefore has the following drawbacks:

[0005] 1) Data dependence, uncertainty in intelligent results, and complex fault mechanisms;

[0006] 2) Different external data input into the same artificial intelligence model can lead to behavioral uncertainty due to differences in data type or knowledge base;

[0007] 3) The existing parameters are limited to the artificial intelligence algorithm itself and do not consider the impact of the artificial intelligence algorithm on the system;

[0008] 4) Traditional reliability parameters lack a comprehensive consideration of intelligent systems and are difficult to describe the intelligence, autonomy, and coordination of intelligent systems;

[0009] 5) The variability of behavioral states leads to complex coupling relationships, which are difficult to describe using existing reliability methods;

[0010] 6) Artificial intelligence modules are related to the usage method and have little to do with temperature, vibration, etc. Therefore, traditional reliability tests such as accelerated stress and acceleration factor are not applicable to the reliability analysis of intelligent aviation equipment. There is an urgent need to study a new reliability analysis method for intelligent aviation equipment. Summary of the Invention

[0011] In view of the shortcomings of the prior art, the present invention provides an artificial intelligence-based reliability analysis method for aviation equipment, which can perform reliability analysis on aviation equipment and products with artificial intelligence. It proposes a set of methods for reliability analysis of intelligent aviation equipment from the dimensions of intelligent system function and structure composition, failure mechanism, parameter system, modeling evaluation and testing methods, which can accurately predict and analyze the reliability of intelligent aviation equipment and provide quantitative standards for reliability analysis.

[0012] Specifically, this invention provides an artificial intelligence-based reliability analysis method for aviation equipment, which includes the following steps:

[0013] S1. Analyze the functions and structural composition of intelligent aviation equipment;

[0014] S2. Construct a reliability index system and parameter model for intelligent aviation equipment. The reliability index parameter model includes a system overall mission reliability index model, an operational stability index model, a data quality index model, and an artificial intelligence learning module reliability index model. Specifically, it includes the following sub-steps:

[0015] S21. Construct a system overall task reliability index model, as shown below:

[0016]

[0017] Where R refers to the overall system task reliability index, and T off The duration of system downtime due to a failure within a specified time period; T refers to the total specified operating time of the system.

[0018] S22. Construct an operational stability index model, which includes a hardware reliability model, an instantaneous reliability model, and a communication reliability model.

[0019] The hardware reliability model is as follows:

[0020]

[0021] Where MTBF is the hardware reliability index, T is the hardware uptime, and A is the total number of faults actually detected.

[0022] The instantaneous reliability model is as follows:

[0023]

[0024] Among them, T d The average downtime is T, the uptime of the infrastructure is A, and the total number of faults actually detected is A.

[0025] The communication reliability model includes a connectivity reliability model and a time reliability model;

[0026] The connectivity reliability model is as follows:

[0027]

[0028] Among them, R C (t) is the connectivity reliability index, N C (t) represents the number of times the simulation continues up to time t while the connection remains intact, N. CM This represents the total number of connections.

[0029] The time reliability model is as follows:

[0030]

[0031] Among them, R T N is a time reliability indicator. D N is the number of data frames whose transmission delay is less than a given threshold. DM The total number of data frames transmitted;

[0032] S23. Construct a data quality indicator model, which includes a data sample quality model and an input data quality model;

[0033] S24. Construct a reliability index model for the artificial intelligence learning module, wherein the artificial intelligence learning reliability index model includes perception and recognition reliability, knowledge reliability, learning reliability, prediction reliability, and decision reliability.

[0034] S3. Based on the reliability parameters and indicator definitions of intelligent aviation equipment, determine the indicator dimensions, data sources and data types for data collection, and collect data from the intelligent aviation equipment to be analyzed.

[0035] S4. Based on the data obtained in step S3, conduct fault mode, cause and mechanism analysis, obtain effective data and record the analysis results;

[0036] S5. Input the valid data obtained in step S4 into the reliability index parameter model of the intelligent aviation equipment constructed in step S2 to perform reliability analysis on the intelligent aviation equipment to be analyzed.

[0037] Preferably, the data sample quality model in step S2 is:

[0038]

[0039] Where K is the peak stability rate of the data sample, x iLet u be the mean of the i-th sample, σ be the fourth standard deviation, K = 3 be the normal distribution, K > 3 be the waveform flat, K < 3 be the waveform abrupt, and the smaller the peak stability rate K of the data sample, the higher the reliability of the data sample quality.

[0040] The input data quality model is as follows:

[0041]

[0042] Where PSI is the population stability rate of the data sample, i is the i-th group of the data, B is the number of groups of the data, and p i The proportion of interference data in the i-th group of the intelligent system, q i The PSI value represents the proportion of normal data in the i-th group; the smaller the PSI value, the higher the reliability of the data input to the intelligent system.

[0043] Preferably, the reliability of the perception and recognition includes the duty cycle and the Davies-Bouldin index, wherein the duty cycle is:

[0044]

[0045] Where, n obc K represents the total number of obstacles within the range, K represents the number of clusters within the current local search radius, and DR ranges from... The larger the DR range, the more obstacles are identified and perceived.

[0046] The Davies-Bouldin index is:

[0047]

[0048] Where k is the number of clusters divided by the current AI model, and S i Let d be the scattering value within the i-th cluster. ij denoted by DB, represents the Euclidean distance between cluster centers i and j. A smaller DB index indicates higher reliability of the clusters generated by the AI ​​model.

[0049] Preferably, the reliability of the knowledge includes recall and deduplication rate.

[0050] Recall rate P re The calculation formula is:

[0051]

[0052] Where TP is the number of samples predicted as true positives, and FN is the number of samples predicted as false negatives. 0 ≤ P re ≤1, P re The higher the value, the higher the reliability;

[0053] Precision Ppr The calculation formula is:

[0054]

[0055] Where TP is the number of samples predicted as true positives, FP is the number of samples predicted as false positives, and 0 ≤ P pr ≤1, P pr The higher the value, the higher the reliability.

[0056] Preferably, the learning reliability includes the F1 score and training intensity.

[0057] The F1 score is the harmonic average of the model precision and the model recall, and its calculation formula is as follows:

[0058]

[0059] Among them, P pr P represents the precision of the current AI model. re F1 represents the recall rate of the current AI model, where 0 ≤ F1 and a higher F1 value indicates higher reliability.

[0060] The training intensity describes the number of times the model needs to continuously retrieve data from the experience replay pool for training during the learning process. It represents the average number of samples retrieved per sample for training. The formula for training intensity is as follows:

[0061]

[0062] Where BatchSize represents the number of data or samples passed to the program for learning in a single batch. N is the total number of training iterations, and T... step This represents the difference between the total time step and the time nodes.

[0063] Preferably, the prediction reliability includes the mean reciprocal rank and the KL divergence. The mean reciprocal rank (MRR) is calculated by taking the average reciprocal rank of all predicted recommendation results of the model, and its calculation formula is as follows:

[0064]

[0065] Wherein, the total number of ranking results for the |U| parameter, rank i Let MMR be the rank of the i-th sample in the recommendation list. The larger the MMR, the higher the reliability of the prediction effect.

[0066] KL divergence D KL The formula for calculating the difference in class distribution between the categorical labeled data and the predicted data is as follows:

[0067]

[0068] Among them, P i T is the number of predicted samples in the i-th category. i D is the total number of labels for all prediction hypotheses in the i-th category, where n is the total number of categories. KL The smaller the value, the closer the predicted classification is to the actual classification, and the higher the reliability of the prediction effect.

[0069] Preferably, decision reliability includes AUC value and goodness of fit.

[0070]

[0071] Where, {(x1,y1),(x2,y2),...,(x m ,y m )} represents the coordinates of the ROC curve, 0≤AUC≤1, and the larger the AUC, the higher the reliability of the effect;

[0072] The goodness of fit is calculated using the following formula:

[0073] P f =|p1-p2|

[0074] in

[0075] Where p1 represents the model fit before noise and outliers, p2 represents the model fit after adding outliers, and w k Let c represent the set of the k-th cluster. j Let P represent the j-th category, N represent the total number of samples, p1 > 0, p2 > 0, and P' ..."""""'" f >0, P f The smaller the value, the higher the reliability of the effect.

[0076] Preferably, the samples are training samples, interference samples, and verification test samples.

[0077] Preferably, the functions in step S1 include perception, recognition, prediction, decision-making, and execution functions, and the structural components in step S1 include intelligent equipment hardware and intelligent equipment software.

[0078] Preferably, the reliability index system in step S2 includes primary reliability indexes, secondary reliability indexes, and tertiary reliability indexes.

[0079] Preferably, the method further includes step S6: conducting reliability tests and experiments: After the reliability assessment is completed, the intelligent system is subjected to reliability tests and experiments to verify the accuracy of the assessment results. The specific experimental methods include the following sub-steps:

[0080] S61. Determine the test modules: Based on the reliability verification requirements of intelligent aviation equipment, determine the objects and modules for reliability testing;

[0081] S62. Determine the testing principles: Based on the functional and structural characteristics, failure modes and failure mechanisms of intelligent aviation equipment, determine the test samples and failure judgment principles.

[0082] S63. Determine the test method: Based on the failure mode and failure mechanism, determine the test method, test tooling, equipment, software, data and test environment for the test process;

[0083] S64. Conduct testing: Conduct reliability testing experiments according to the experimental operation procedures;

[0084] S65. Analyze test data: Collect experimental data, remove outliers from the data, and conduct statistical analysis of the data;

[0085] S66. Verify reliability results: Verify and analyze the test results.

[0086] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0087] (1) This invention proposes an artificial intelligence-based reliability analysis method for aviation equipment. It proposes a set of reliability analysis methods for intelligent aviation equipment from the dimensions of intelligent system function and structure composition, failure mechanism, index parameter system, modeling evaluation and testing methods. It can be directly called when analyzing intelligent aviation equipment, and can perform reliability analysis on intelligent aviation equipment to ensure the reliability of intelligent aviation equipment. It provides a quantitative standard for the reliability analysis of intelligent aviation equipment and fills the gap in the existing technology.

[0088] (2) The index parameter model of the present invention includes a system overall task reliability index model, an operational stability index model, a data quality index model, and an artificial intelligence learning module index model, which can perform reliability analysis of intelligent aviation equipment in a multi-dimensional and comprehensive manner, and ensure the accuracy of the analysis results.

[0089] (3) Based on the data obtained in step S3, the present invention conducts fault mode, cause and mechanism analysis, obtains effective data and records the analysis results, and can delete obviously inconsistent or erroneous data to ensure the accuracy of the model input data, thereby preventing large errors in the analysis results and ensuring the accuracy of reliability analysis. Attached Figure Description

[0090] Figure 1 This is a schematic diagram of the process of the present invention;

[0091] Figure 2 This is a flowchart of an embodiment of the present invention;

[0092] Figure 3 This is a schematic diagram of the indicator system in an embodiment of the present invention. Detailed Implementation

[0093] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0094] Specifically, this invention provides an artificial intelligence-based reliability analysis method for aviation equipment, such as... Figure 1 As shown, it includes the following steps:

[0095] S1. Analyze the functions and structural composition of intelligent aviation equipment. This includes performing functional and structural analyses of the intelligent aviation equipment sequentially, and recording the results of the functional and structural analyses in tabular form.

[0096] S2. Construct a reliability index system parameter model for intelligent aviation equipment. The reliability index parameter model includes a system overall mission reliability index model, an operational stability index model, a data quality index model, and an artificial intelligence learning module index model.

[0097] The overall system task reliability index model is shown below:

[0098]

[0099] Where R refers to the overall system task reliability index, and T off The reliability index refers to the duration during which the system stops operating due to a failure within a specified time. T refers to the total specified operating time of the system. The reliability index system is divided into primary, secondary, and tertiary indicators.

[0100] The operational stability index models include hardware reliability models, instantaneous reliability models, and communication reliability models. The calculation methods for these three models are shown below:

[0101] The hardware reliability model is as follows:

[0102]

[0103] Where MTBF is the hardware reliability, T is the hardware uptime, and A is the total number of faults actually detected.

[0104] The instantaneous reliability model is as follows:

[0105]

[0106] Among them, T d The average downtime is T, the uptime of the infrastructure is A, and the total number of faults actually detected is A.

[0107] Communication reliability includes connectivity reliability models and time reliability models; connectivity reliability models and time reliability models are used to predict the reliability of communication and are more suitable for intelligent aviation equipment.

[0108] The connectivity reliability model is as follows:

[0109]

[0110] Among them, R C (t) represents the connectivity reliability, N C (t) represents the number of times the simulation continues up to time t while the connection remains intact, N. CM This represents the total number of connections.

[0111] The time reliability model is as follows:

[0112]

[0113] Among them, R T For time reliability, N D N is the number of data frames whose transmission delay is less than a given threshold. DM This represents the total number of data frames transmitted.

[0114] The data quality indicator model includes a data sample quality model and an input data quality model. The calculation methods for the two models are shown below:

[0115] The data sample quality model is as follows:

[0116]

[0117] Where K is the peak stability rate of the data sample, x i Let u be the mean of the i-th sample, σ be the fourth standard deviation, K = 3 be the normal distribution, K > 3 be the waveform flat, K < 3 be the waveform abrupt, and the smaller the peak stability rate K of the data sample, the higher the reliability of the data sample quality.

[0118] The input data quality model is:

[0119]

[0120] Where PSI is the population stability rate of the data sample, i is the i-th group of the data, B is the number of groups of the data, and p i The proportion of interference data in the i-th group of the intelligent system, q i The PSI value represents the proportion of normal data in the i-th group; the smaller the PSI value, the higher the reliability of the data input to the intelligent system.

[0121] The AI ​​learning module's performance indicators include perceptual reliability, knowledge reliability, learning reliability, prediction reliability, and decision reliability. The calculation methods for perceptual reliability, knowledge reliability, learning reliability, prediction reliability, and decision reliability are as follows:

[0122] Perception and recognition reliability includes duty cycle (DR) and the Davies-Bouldin index. The duty cycle is:

[0123]

[0124] Where, n obc K represents the total number of obstacles within the range, K represents the number of clusters within the current local search radius, and DR ranges from... The larger the DR range, the more obstacles are identified and perceived.

[0125] The formula for calculating the Davies-Bouldin index is:

[0126]

[0127] Where k is the number of clusters divided by the current AI model, and S i Let d be the scattering value within the i-th cluster. ij denoted by DB, represents the Euclidean distance between cluster centers i and j. A smaller DB index indicates higher reliability of the clusters generated by the AI ​​model.

[0128] Knowledge reliability includes recall and duplication rate, and the calculation formulas are as follows:

[0129] Recall rate P re The calculation formula is:

[0130]

[0131] Where TP is the number of samples predicted as true positives, and FN is the number of samples predicted as false negatives. 0 ≤ P re ≤1, P re The higher the value, the higher the reliability;

[0132] Precision P pr The calculation formula is:

[0133]

[0134] Where TP is the number of samples predicted as true positives, FP is the number of samples predicted as false positives, and 0 ≤ P pr ≤1, P pr The higher the value, the higher the reliability.

[0135] Learning reliability includes the F1 score and training intensity. The F1 score is the harmonic average of model precision and model recall, and its calculation formula is as follows:

[0136]

[0137] Among them, P pr P represents the precision of the current AI model. re F1 represents the recall rate of the current AI model, where 0 ≤ F1 and the higher the F1 value, the higher the reliability.

[0138] Training intensity describes the number of times data needs to be drawn from the experience replay pool for training during the model learning process. It represents the average number of samples to be drawn for training. The formula for training intensity is as follows:

[0139]

[0140] Where BatchSize represents the number of data or samples passed to the program for learning in a single batch. N is the total number of training iterations, and T... step This represents the difference between the total time step and the time nodes.

[0141] Prediction reliability includes mean reciprocal rank and KL divergence. The mean reciprocal rank (MRR) is calculated by taking the average reciprocal rank of all predictions from the model, and its formula is shown below:

[0142]

[0143] Wherein, the total number of ranking results for the |U| parameter, rank i Let MMR be the rank of the i-th sample in the recommendation list. The larger the MMR, the higher the reliability of the prediction effect.

[0144] KL divergence D KL The formula for calculating the difference in class distribution between the categorical labeled data and the predicted data is as follows:

[0145]

[0146] Among them, P i T is the number of predicted samples in the i-th category. i D is the total number of labels for all prediction hypotheses in the i-th category, where n is the total number of categories. KL The smaller the value, the closer the predicted classification is to the actual classification, and the higher the reliability of the prediction. The above samples include training samples, interference samples, and validation test samples.

[0147] Preferably, decision reliability includes AUC value and goodness of fit.

[0148]

[0149] Where, {(x1,y1),(x2,y2),...,(x m ,y m )} represents the coordinates of the ROC curve, 0≤AUC≤1, and the larger the AUC, the higher the reliability of the effect;

[0150] The goodness of fit is calculated using the following formula:

[0151] P f =|p1-p2|

[0152] in

[0153] Where p1 represents the model fit before noise and outliers, p2 represents the model fit after adding outliers, and w k Let c represent the set of the k-th cluster. j Let P represent the j-th category, N represent the total number of samples, p1 > 0, p2 > 0, and P' ..."""""'" f >0, P f The smaller the value, the higher the reliability of the effect.

[0154] S3. Based on the reliability parameters and indicator definitions of intelligent aviation equipment, determine the indicator dimensions, data sources, and data types for data collection.

[0155] S4. Based on the data obtained in step S3, conduct fault mode, cause and mechanism analysis and record relevant data.

[0156] S5. Input the valid data obtained in step S4 into the reliability index parameter model of the intelligent aviation equipment constructed in step S2 to perform reliability analysis on the intelligent aviation equipment to be analyzed.

[0157] The method of the present invention will be further described below with reference to specific embodiments: In this embodiment, it is specifically used for the analysis of intelligent unmanned aerial vehicle equipment, such as... Figure 2 As shown, the specific steps are as follows:

[0158] S1. First, conduct a functional and structural composition analysis of the intelligent unmanned aerial vehicle system, which mainly includes two aspects: functional analysis and structural composition analysis.

[0159] Unmanned Aerial Vehicle (UAV) System Functional Analysis: Generally, the functions are analyzed from five aspects: perception, identification, prediction, decision-making, and execution. The general definitions of these functions are as follows:

[0160] Perception: The system collects image information of the route and ground location information through sensors and cameras, and transmits it to the flight control module.

[0161] Recognition: Cluster analysis of perceived data, including location information recognition and obstacle recognition.

[0162] Prediction: Path planning analysis is performed based on obstacles and fault locations.

[0163] Decision-making: Based on the mission, fault status, and remaining range, decide on the next task for the drone (such as whether to handle the problem on-site or return to base).

[0164] Execution: Based on the decision, drive the actuation system to achieve flight control.

[0165] Based on the initial definition of functions, the capability list of the unmanned aerial vehicle (UAV) system is determined and described, as shown in the table below:

[0166]

[0167] Analysis of the structural composition of unmanned aerial vehicles (UAVs):

[0168] The structure of intelligent aviation equipment consists of hardware and software components, as defined below:

[0169] Hardware components: Typical hardware in the product, such as actuators and bearings; hardware used for artificial intelligence computing, including central processing units (CPUs), application-specific integrated circuits (ASICs), graphics processing units (GPUs), tensor processing units (TPUs), and intelligent processing units (IPUs). Various types of cameras, sensors, and other devices used to collect images, sound, and other data formats input to the AI ​​system, as well as other communication infrastructure hardware.

[0170] Software component:

[0171] First, the core of the software component consists of machine learning / deep learning (ML / DL) based algorithms and other rule-based algorithms. These algorithms include image and speech recognition, computer vision, natural language processing (NLP), and classification.

[0172] Secondly, it also includes data collection, processing, and decision-making components: by analyzing the system's structural composition, a list of the system's structural components is obtained, as shown in the table below.

[0173] Serial Number structure Composition Description Corresponding functions

[0174] S2. Construct a reliability index parameter model for intelligent drones. The reliability index parameter model includes a system overall task reliability index model, an operational stability index model, a data quality index model, and an artificial intelligence learning module index model.

[0175] Operating environment stability covers hardware failures caused by hardware equipment malfunctions, downtime of intelligent systems, etc.

[0176] Data quality covers faults caused by unreasonable design of standard sample data (distribution and type of interference samples and test verification sample data) in intelligent systems;

[0177] The metrics for the artificial intelligence learning module include failures in different types of AI models (perception and recognition, knowledge base, learning, prediction, and decision-making). The overall metric system is as follows: Figure 3 As shown.

[0178] Based on the functional characteristics of the system and the indicator system architecture, the first, second, and third level reliability indicators of the intelligent UAV equipment are determined and recorded in sequence, as shown in the table below.

[0179]

[0180]

[0181] Subsequently, based on the analysis results of the functions and structure of aviation intelligent equipment, the index parameters were determined. The index parameter models include the overall system mission reliability index model, the operational stability index model, the data quality index model, and the artificial intelligence learning module index model.

[0182] S21. The specific steps for constructing the overall system task reliability index model are as follows:

[0183]

[0184] Among them, T off This refers to the duration during which the intelligent system stops operating due to a fault within a specified time. The faults include process-related faults caused by factors such as the stability of the intelligent system's operating environment, data accuracy, and indicators of the artificial intelligence learning module. T refers to the total specified operating time of the intelligent system.

[0185] S22. Construct an operational stability index model: Operational stability refers to the ability to avoid system failure caused by hard and soft faults. Based on the type of fault, the operational environment stability index model can be further divided into hardware reliability, instantaneous reliability, and communication reliability.

[0186] Hardware reliability: Hardware reliability describes the ability of an intelligent system to avoid failure due to the failure or offline of its infrastructure, such as CPU, GPU, distributed server. It can be measured using mean time to failure.

[0187] Mean Time Between Failures (MTBF) refers to the average time between adjacent failures over a certain time period. The specific measurement method is shown in the following formula:

[0188]

[0189] Where T is the infrastructure uptime, and A is the total number of failures actually detected (failures observed during uptime).

[0190] Instantaneous reliability: Instantaneous reliability describes the ability of a system to avoid failure due to a signal exceeding a specified threshold, which would cause the intelligent system to become unresponsive. It can be measured using mean time downtime (MTBF). Mean time downtime T d The formula is shown below:

[0191]

[0192] Where T is the infrastructure uptime, and A is the total number of failures actually detected (failures observed during uptime).

[0193] Communication reliability models include connectivity reliability models and time reliability models;

[0194] The connectivity reliability model is as follows:

[0195]

[0196] Among them, R C (t) is the connectivity reliability index, N C (t) represents the number of times the simulation continues up to time t while the connection remains intact, N. CM This represents the total number of connections.

[0197] The time reliability model is as follows:

[0198]

[0199] Among them, R T N is a time reliability indicator. D N is the number of data frames whose transmission delay is less than a given threshold. DM This represents the total number of data frames transmitted.

[0200] S23. Construct a data quality indicator model, which includes a data sample quality model and an input data quality model.

[0201] The data quality index model refers to the degree to which the correctness of a smart system's operation and results for specified tasks and user objectives is demonstrated after the system has been trained, tested, and validated using standard training / test samples during the training and validation phases. Data correctness encompasses two aspects: data sample quality and (online) data quality.

[0202] Data sample quality model: Data sample quality describes the data volume, precision, noise, missing data, and balanced distribution of training and distractor samples in the sample dataset during the training and testing phases. It can be evaluated using the data sample peak stability rate (KSR), which represents the kurtosis stability of each variable in the standard dataset across the training, validation, and test sets. The data sample peak stability rate K, or the data sample quality model, is shown in the formula:

[0203]

[0204] Where, x i Let be the i-th sample (which can be a training sample, a interference sample, or a validation test sample), u be the sample mean, and σ be the fourth standard deviation. K = 3 indicates a normal distribution, K > 3 indicates a flat waveform, and K < 3 indicates a sharp waveform. The smaller the peak stability rate K of the data samples, the higher the reliability of the data sample quality.

[0205] Input data quality describes the situation of missing data, noise points, and the distribution of normal data points in the input data during the formal use phase of the intelligent system. It can be evaluated using the population stability index (PSI), which is the input data quality model, as shown in the formula:

[0206]

[0207] Where i is the i-th group of data, B is the number of groups of data, and p i The proportion of interference data in the i-th group of the intelligent system, q i This represents the proportion of normal data in the i-th group. The smaller the PSI value, the higher the reliability of the data input to the intelligent system.

[0208] S24. Construct an indicator model for the artificial intelligence learning module. The indicator model for the artificial intelligence learning module includes the reliability of perception and recognition, the reliability of knowledge, the reliability of learning, the reliability of prediction, and the reliability of decision-making.

[0209] Artificial Intelligence Learning Module Indicator Model: Artificial intelligence learning module indicators refer to the collective term for the ability to avoid the failure of intelligent system functions due to the failure of AI model in intelligent system. It can be further divided according to the intelligent functions of intelligent system: perception and recognition reliability, knowledge reliability, learning reliability, prediction reliability, and decision reliability.

[0210] Perception and Recognition Reliability: Perception and recognition reliability refers to the ability of an intelligent system to avoid errors in sensing equipment (such as radar, sonar, visual sensors, etc.), perception and recognition AI module design (algorithm design, program design), and interface data features, labeling, classification, etc., which could lead to incorrect perception and recognition results or perception and recognition function failures. It is evaluated from the following dimensions:

[0211] Duty cycle: Duty cycle DR is the ratio of the number of obstacle classes within the viewport of the AI ​​model used for perception and recognition to the total number of obstacle classes in the environment. The specific calculation method is shown in the formula.

[0212]

[0213] Where, n obc K represents the total number of obstacles within the range, K represents the number of clusters within the current local search radius, and DR ranges from... The larger the DR range, the more obstacles are identified and perceived.

[0214] Davies-Bouldin index.

[0215] The DB index calculates the similarity between each cluster and its most similar cluster, and measures the quality of the overall clustering result by averaging all similarities. The formula for calculating the DB index is as follows:

[0216]

[0217] Where k is the number of clusters divided by the current AI model, and S i Let d be the scattering value within the i-th cluster. ij denoted by DB, represents the Euclidean distance between cluster centers i and j. A smaller DB index indicates higher reliability of the clusters generated by the AI ​​model.

[0218] Knowledge reliability: Knowledge reliability describes the ability to avoid problems such as sparse or missing data attributes, which could lead to excessively sparse entity relationships in the knowledge base built based on AI models, thus rendering the intelligent system's intelligent relationship query and intelligent search capabilities ineffective.

[0219] Recall rate P re This refers to the ratio of samples predicted as true positives by the model in an intelligent system to samples predicted as true positives for the input data sample. Its formula can be defined as follows:

[0220]

[0221] Where TP is the number of samples predicted as true positives, and FN is the number of samples predicted as false negatives. 0 ≤ P re ≤1, P re The higher the value, the higher the reliability.

[0222] Precision: Precision P pr This refers to the ratio of samples that the model in an intelligent system predicts to be true examples out of all samples that are actually true examples, given the input data samples.

[0223]

[0224] Where TP is the number of samples predicted as true positives, FP is the number of samples predicted as false positives, and 0 ≤ P pr ≤1, P pr The higher the value, the higher the reliability.

[0225] Learning reliability describes the ability of an intelligent system to avoid errors or inefficiencies in its learning process due to factors such as inappropriate settings of parameters like the learning rate of the AI ​​model, interference from extreme environments during the learning process, or incorrect input of information sample data.

[0226] The F1 score is the harmonic average of the model precision and the model recall.

[0227]

[0228] Among them, P pr P represents the precision of the current AI model. re F1 represents the recall rate of the current AI model. 0 ≤ F1, and the larger the F1 value, the higher the reliability.

[0229] Training intensity describes the number of times data needs to be drawn from the experience replay pool for training during the model's learning process. It represents the average number of samples used per model for training. The formula for training intensity is shown below:

[0230]

[0231] Where BatchSize represents the number of data or samples passed to the program for learning in a single batch. N is the total number of training iterations, and T... step This represents the difference between the total time step and the time nodes.

[0232] Predictive reliability: Predictive reliability describes the ability of an intelligent system to avoid predictive errors due to limitations in AI model capabilities, sparse entity relationships, etc., when performing predictive analysis based on user behavior and input information, leading to unsatisfactory predictions. Several factors influence the predictive reliability of intelligent systems:

[0233] Mean Reciprocal Ranking (MRR): The Mean Reciprocal Ranking (MRR) is calculated by taking the average reciprocal ranking of all predicted recommendations from the model.

[0234]

[0235] Wherein, the total number of ranking results for the |U| parameter, rank i Let MMR be the rank of the i-th sample in the recommendation list. The larger the MMR, the more reliable the prediction.

[0236] Kullback-Leibler divergence: KL divergence D KL Calculate the class distribution difference between the categorical labeled data and the predicted data.

[0237]

[0238] Among them, P i T is the number of predicted samples in the i-th category. i is the total number of labels for all prediction hypotheses in the i-th category, where n is the total number of categories. D KL The smaller the value, the closer the predicted classification is to the actual classification, and the higher the reliability of the prediction.

[0239] Decision reliability describes the ability of an intelligent system to avoid errors in decision instructions and ultimately erroneous decision-making behavior due to factors such as incorrect input data, missing input data attributes, or incorrectly set decision constraints.

[0240] The AUC value represents the area under the ROC curve of an AI model, plotted with the true positive rate (TPR) on the y-axis and the false positive rate (FPR) on the x-axis. It is used to measure the predictive performance of an AI model. The formulas for the true positive rate (TPR) and the false positive rate (FPR) are as follows:

[0241]

[0242]

[0243] Where TP is the number of samples predicted as true positives, FP is the number of samples predicted as false positives, and FN is the number of samples predicted as false negatives. AUC can be obtained by summing the areas under the ROC curve, i.e.:

[0244]

[0245] {(x1,y1),(x2,y2),...,(x m ,y m )} represents the coordinates of the ROC curve, 0≤AUC≤1, and the larger the AUC, the higher the reliability of the effect.

[0246] Goodness of fit: This is a relative indicator used to evaluate the robustness of data against interference and to measure the change in the model's goodness of fit to the data.

[0247] P f =|p1-p2|

[0248] in

[0249] Where p1 represents the model fit before noise and outliers are removed, and p2 represents the model fit after outliers are added. k Let c represent the set of the k-th cluster. j Let p1 represent the j-th category, and N represent the total number of samples. p1 > 0, p2 > 0, P f >0, P f The smaller the value, the higher the reliability of the effect.

[0250] S3. Collect and organize the data:

[0251] Based on the reliability parameters and indicator definitions of intelligent aviation equipment, the indicator dimensions, data sources, and data types for data collection are determined. Data collection is performed on the intelligent aviation equipment to be analyzed, and data values ​​are recorded, as shown in the table below:

[0252] Serial Number Indicator Dimensions Data source Data types Data value

[0253] S4. Using the data from step S3 as input, conduct fault mode, cause and mechanism analysis, remove invalid data, obtain valid data and record the analysis results.

[0254] Failure Mode and Cause Analysis: Failure mode and cause analysis includes three categories: hardware system failure, software system failure, and functional failure.

[0255] Hardware system failure mode analysis:

[0256] Typical failure modes of hardware systems mainly include: hardware failures, software failures, and network failures.

[0257] Examples of hard failures include failures in high-performance computing clusters and cloud computing systems, which provide the infrastructure for many AI applications.

[0258] A soft fault (or transient fault) is a failure that occurs only when a signal in the system exceeds a threshold. Network failures can affect communication functions, and networks are a crucial component of many AI systems.

[0259] Software system failure mode analysis: Typical failure modes of software systems mainly include prediction errors, data quality issues, model bias, and adversarial attacks (AA).

[0260] Many prediction errors are caused by distribution shift. Distribution shift usually means that the operating environment is different from the training set environment.

[0261] Data quality can also cause software failures. Failures can occur if the data input to the algorithm is noisy, contaminated, or comes from faulty sensors.

[0262] Model bias can cause inaccurate model predictions, leading to malfunctions.

[0263] Adversarial attacks (AA) are a critical issue for the reliability of complex intelligent systems. Rearranging data in small increments can lead to inaccurate model predictions and consequently, failures.

[0264] Functional failure

[0265] a) Fault detection and identification

[0266] Intelligent complex systems can identify environmental obstacles by inputting environmental information based on map data and extracting the features of environmental obstacles. However, during operation, errors in system components (such as radar, sonar, and visual sensors), AI module design (algorithm design and program design), and data input can lead to incorrect perception and identification results.

[0267] b) Learning Failure

[0268] The essence of intelligent system learning is that the system can improve its performance based on past experience. However, during the operation of intelligent systems, factors such as inappropriate settings of parameters such as the learning rate of the AI ​​system, interference from extreme environments during the learning process, and incorrect input of information sample data can cause errors in the autonomous learning of intelligent systems.

[0269] c) Predicting Faults

[0270] Intelligent systems rely on AI models to perform predictive analysis based on user behavior and input information. However, errors in system components, AI model design, and data input can lead to incorrect prediction results.

[0271] d) Decision-making failure

[0272] Decision-making is a typical application of intelligent systems, which ultimately help users make decisions based on user input. Errors in decision-making instructions from intelligent systems are often caused by factors such as incorrect input data, missing input data attributes, and incorrect constraint settings.

[0273] e) Process failures

[0274] The failure of intelligent system execution results is caused by protection strategies such as redundant backup of hardware and interfaces, fallback measures in software design, and overfitting of AI models to improve recognition, perception, and autonomous decision-making accuracy.

[0275] Analysis of Fault Mechanisms in Intelligent Systems

[0276] Based on the above failure modes of intelligent systems, we categorize the factors affecting the reliability of intelligent systems into three categories: operating environment, data, and model (i.e., algorithm). Operating environment: The normal execution of intelligent functions or tasks by an intelligent system depends on the operating environment provided by the infrastructure layer equipment. Changes in the operating environment may cause the intelligent system to fail to perform tasks or stop operating. Reasons for changes in the operating environment include: equipment failures in infrastructure such as high-performance computing clusters and cloud computing systems, processing unit failures in intelligent functions, and system failures such as bandwidth issues, data buffer overflows, data transmission congestion leading to system response timeouts, and system service interruptions. Data: The intelligent functions provided by an intelligent system largely depend on data samples. For example, during the system model training and validation phases, training samples and validation test samples are needed to train the model and adapt it to the current environment. When the model training is insufficient or the training samples (interference or noise data distribution or proportions are not compliant) are insufficiently prepared, the system may malfunction in executing intelligent tasks when put into use, leading to errors in the execution of intelligent tasks. On the other hand, during the formal use phase of the intelligent system, the degree of task execution of the intelligent system is also greatly affected by the input sample data. If the noise rate of the input data is too high or the data structure is abnormal, such as inconsistent dataset dimensions, it may lead to a decrease in the accuracy of the intelligent system or the intelligent system stopping operation.

[0277] The realization of intelligent functions in intelligent systems relies on their AI models. Depending on the functional characteristics of the AI ​​models, the influencing factors on the model algorithms also vary. For example, in supervised learning algorithms such as classification, which classifies data based on features, if data with ambiguous or easily confused features is not considered, a single data point might be classified into two different labels, making subsequent entity identification difficult. In unsupervised learning algorithms such as clustering, without prior knowledge, an unreasonable number of clusters in the input model can lead to abnormal cluster structures generated by the final model for the target data, causing errors in the intelligent system's task execution.

[0278] The following table shows the analysis of the failure modes, causes, and mechanisms of the intelligent system.

[0279]

[0280] Based on the above analysis, invalid and interfering data can be eliminated, and valid data can be obtained and recorded for reliability analysis.

[0281] S5. Input the valid data obtained in step S4 into the reliability index parameter model of the intelligent aviation equipment constructed in step S2 to perform reliability analysis on the intelligent aviation equipment to be analyzed.

[0282] Based on the valid data collected in step S4, the reliability indicators are substituted into the indicator model in step S2 to conduct reliability indicator analysis and evaluation, including data analysis, indicator solution, and result evaluation. The reliability analysis results are recorded in the table below, and the reliability evaluation model is referenced in the table below.

[0283]

[0284]

[0285] S6. Conduct reliability tests and experiments on the reliability analysis results:

[0286] After the reliability assessment is completed, reliability testing is conducted on the intelligent system to verify the accuracy of the assessment results. The test methods include:

[0287] S61. Determine the test modules: Based on the reliability verification requirements of intelligent aviation equipment, determine the objects and modules for reliability testing;

[0288] S62. Determine the testing principles: Based on the functional and structural characteristics, failure modes and failure mechanisms of intelligent aviation equipment, determine the test samples and failure judgment principles.

[0289] S63. Determine the test method: Based on the failure mode and failure mechanism, determine the test method, test tooling, equipment, software, data and test environment for the test process;

[0290] S64. Conduct testing: Conduct reliability testing experiments according to the experimental operation procedures;

[0291] S65. Analyze test data: Collect experimental data, remove outliers from the data, and conduct statistical analysis of the data;

[0292] S66. Verify reliability results: Verify and analyze the test results.

[0293] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A reliability analysis method for aviation equipment based on artificial intelligence, characterized in that: It includes the following steps: S1. Analyze the functions and structural composition of intelligent aviation equipment; S2. Construct a reliability index parameter model for intelligent aviation equipment. The reliability index parameter model includes a system overall mission reliability index model, an operational stability index model, a data quality index model, and an artificial intelligence learning module reliability index model. Specifically, it includes the following sub-steps: S21. Construct a system overall task reliability index model, as shown below: ； in, Refers to the overall system task reliability index, This refers to the duration during which the system stops operating due to a fault within a specified time. This refers to the total duration of the system's scheduled operation. S22. Construct an operational stability index model, which includes a hardware reliability model, an instantaneous reliability model, and a communication reliability model. The hardware reliability model is as follows: ； in, For hardware reliability indicators, For hardware runtime, This represents the total number of faults actually detected. The instantaneous reliability model is as follows: ； in, The average downtime is... For infrastructure uptime, This represents the total number of faults actually detected. The communication reliability model includes a connectivity reliability model and a time reliability model; The connectivity reliability model is as follows: ； in, For connectivity reliability indicators, N represents the number of times the connection remains active up to time t. CM This represents the total number of connections. The time reliability model is as follows: ； in, For time reliability indicators, The number of data frames whose transmission latency is less than a given threshold. The total number of data frames transmitted; S23. Construct a data quality indicator model, which includes a data sample quality model and an input data quality model; S24. Construct a reliability index model for the artificial intelligence learning module. The reliability index model for the artificial intelligence learning module includes perception and recognition reliability, knowledge reliability, learning reliability, prediction reliability, and decision reliability. Perception and recognition reliability refers to the ability of the intelligent system to avoid errors in sensing equipment, algorithm design, program design, and interface data features, labeling, and classification, which could lead to errors in the perception and recognition results or failures in the perception and recognition function. Knowledge reliability describes the ability to avoid the problem of sparse entity relationships in the knowledge base built based on AI models due to the sparsity or lack of data attributes, which would render the intelligent system's intelligent relationship query and intelligent search capabilities ineffective. Learning reliability describes the ability of an intelligent system to avoid errors or inefficiencies in its learning process due to factors such as inappropriate learning rate parameter settings, interference from extreme environments during the learning process, or incorrect input of information sample data. Predictive reliability describes the ability of an intelligent system to avoid the inability of the AI ​​model to make predictions based on user behavior and input information when the system makes predictions based on user behavior and input information due to insufficient AI model capabilities and sparse entity relationships. Decision reliability describes the ability of an intelligent system to avoid errors in decision instructions and ultimately erroneous decision-making behavior due to incorrect input data, missing input data attributes, or incorrectly set decision constraints. S3. Based on the reliability parameters and indicator definitions of intelligent aviation equipment, determine the indicator dimensions, data sources and data types for data collection, and collect data from the intelligent aviation equipment to be analyzed. S4. Based on the data obtained in step S3, conduct fault mode, cause and mechanism analysis, obtain effective data and record the analysis results; S5. Input the valid data obtained in step S4 into the reliability index parameter model of the intelligent aviation equipment constructed in step S2 to perform reliability analysis on the intelligent aviation equipment to be analyzed.

2. The artificial intelligence-based reliability analysis method for aviation equipment according to claim 1, characterized in that: The reliability of the knowledge mentioned includes recall and precision, where recall is... The calculation formula is: ； in, To predict the number of samples that are true examples, To predict the number of samples that are false negatives, , The higher the value, the higher the reliability; Precision The calculation formula is: ； in, To predict the number of samples that are true examples, To predict the number of samples that are false positives, , The higher the value, the higher the reliability.

3. The artificial intelligence-based reliability analysis method for aviation equipment according to claim 1, characterized in that: The prediction reliability includes the mean reciprocal rank and the KL divergence. The mean reciprocal rank (MRR) value is used to characterize the average reciprocal rank of all prediction recommendations from the model, and its calculation formula is as follows: ； in, To predict the total number of recommendation results, For the first The ranking of a sample in the recommendation list; the higher the MMR value, the more reliable the prediction effect. KL divergence The formula used to characterize the difference in class distribution between categorical labeled data and predicted data is as follows: ； in, It is the first Number of predicted samples in each category It is the first The total number of labels for all prediction hypotheses in each category. The total number of categories, The smaller the value, the closer the predicted classification is to the actual classification, and the higher the reliability of the prediction effect.

4. The artificial intelligence-based reliability analysis method for aviation equipment according to claim 1, characterized in that: The samples include training samples, interference samples, and validation test samples.

5. The artificial intelligence-based reliability analysis method for aviation equipment according to claim 1, characterized in that: The functions in step S1 include perception, recognition, prediction, decision-making, and execution. The structure of step S1 includes intelligent equipment hardware and intelligent equipment software.

6. The artificial intelligence-based reliability analysis method for aviation equipment according to claim 1, characterized in that: In step S2, the reliability system indicators for intelligent equipment include primary reliability indicators, secondary reliability indicators, and tertiary reliability indicators.

7. The artificial intelligence-based reliability analysis method for aviation equipment according to claim 1, characterized in that: It also includes step S6, conducting reliability tests and experiments: After the reliability assessment is completed, the intelligent system is subjected to reliability tests and experiments to verify the accuracy of the assessment results. The specific test methods include the following sub-steps: S61. Determine the test modules: Based on the reliability verification requirements of intelligent aviation equipment, determine the objects and modules for reliability testing; S62. Determine the testing principles: Based on the functional and structural characteristics, failure modes and failure mechanisms of intelligent aviation equipment, determine the test samples and failure judgment principles. S63. Determine the test method: Based on the failure mode and failure mechanism, determine the test method, test tooling, equipment, software, data and test environment for the test process; S64. Conduct testing: Conduct reliability testing experiments according to the experimental operation procedures; S65. Analyze test data: Collect experimental data, remove outliers from the data, and conduct statistical analysis of the data; S66. Verify reliability results: Verify and analyze the test results.

Images