Aircraft environment control system fault diagnosis method, device and equipment and storage medium
By combining BP neural networks and fault tree analysis in the aircraft environmental control system, the problem of insufficient sensitivity and accuracy in fault diagnosis in existing technologies has been solved, and the fault diagnosis capabilities of early warning, precise location and self-optimization have been realized.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-15
- Publication Date
- 2026-03-31
AI Technical Summary
Existing technologies for fault diagnosis of aircraft environmental control systems suffer from problems such as low sensitivity and accuracy, high modeling difficulty, high cost, poor generalization ability, lack of interpretability of diagnostic results, and insufficient self-learning ability.
A baseline model based on a BP neural network is adopted, combined with data preprocessing and fault tree analysis. By collecting historical operating parameters to build a database, the neural network structure is optimized to achieve real-time fault warning and root cause localization, and the model is self-optimized through feedback data.
It enables early and accurate diagnosis and root cause localization of aircraft environmental control system faults, possesses self-evolution capabilities, improves the sensitivity and accuracy of diagnosis, and provides a clear fault interpretation path and continuous optimization capabilities.
Smart Images
Figure CN121764026A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of aircraft airborne system design and fault prediction technology, specifically relating to a fault diagnosis method, device, equipment and storage medium for an aircraft environmental control system. Background Technology
[0002] In recent years, the application of various types of aircraft (including but not limited to eVTOL, UAVs, and traditional aircraft) has become increasingly widespread, penetrating key areas such as logistics transportation, urban air traffic, emergency rescue, and agricultural plant protection. With the continuous increase in the power density and system complexity of airborne electronic equipment, the reliability of environmental control systems (EHS) has become one of the key factors affecting flight safety. EHS typically consists of multiple coupled subsystems such as air conditioning, pressurization, filtration, cooling, and icing / fogging prevention, and its failure modes are complex and varied.
[0003] Currently, fault diagnosis for environmental control systems largely relies on threshold alarms or physical model-based diagnostic methods. Threshold alarm methods are simple, but have low sensitivity and accuracy, and cannot predict potential performance degradation. Physical model-based methods require accurate system mechanism equations, which are difficult and costly to model for complex and highly coupled environmental control systems, and have poor generalization ability.
[0004] While existing technologies offer some attempts at fault prediction based on artificial intelligence (such as neural networks), they often suffer from the following shortcomings: 1) Simple data preprocessing procedures are inadequate in handling noise and missing data in aircraft operation data, affecting the quality of model input; 2) Fixed prediction model structure limits generalization ability and accuracy; 3) Diagnostic results lack interpretability, making it difficult for maintenance personnel to quickly pinpoint the root cause based on warnings, leading to low response efficiency; 4) Lack of self-learning mechanisms prevents the continuous optimization of diagnostic models using historical fault data. Therefore, the field requires a comprehensive fault diagnosis solution that can achieve early, accurate, interpretable, and self-evolving capabilities. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, device, equipment and storage medium for fault diagnosis of aircraft environmental control systems, which can realize early warning, accurate diagnosis and root cause location of environmental control system faults, and has the ability to self-optimize through feedback data.
[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for diagnosing faults in an aircraft environmental control system, comprising the following steps: Collect historical operating parameters of each subsystem of the environmental control system under normal operating conditions and perform data preprocessing to build a historical database; Based on the historical database, a baseline model based on a BP neural network is established for each subsystem. The operating parameters of each subsystem are used as inputs, and the state parameters representing the operating state of the system are used as outputs. The structural parameters of each BP neural network are optimized using the minimization of the mean absolute deviation as the evaluation index to obtain the optimized baseline model of each subsystem. The real-time operating parameters of the environmental control system are collected and input into the optimized baseline model to obtain the predicted values of the state parameters of each subsystem's operating status; the predicted values are compared with the measured values, and if the difference exceeds a preset threshold, the corresponding subsystem is determined to have a fault and a fault warning signal is generated; In response to the fault warning signal, a fault tree analysis process corresponding to the faulty subsystem and fault phenomenon is initiated to locate the root cause of the fault.
[0007] In one embodiment, the method further includes: collecting feedback data after each fault handling, retraining and optimizing the parameters of the optimized baseline model based on the feedback data; and establishing a fault case library, which is used to record fault phenomena, fault causes, handling processes, and handling results.
[0008] Secondly, the present invention provides a fault diagnosis device for an aircraft environmental control system, comprising: The data preprocessing module is used to collect historical operating parameters of each subsystem of the environmental control system under normal operating conditions and perform data preprocessing in order to build a historical database; The baseline model management module is used to establish and optimize baseline models based on BP neural networks for each subsystem based on the historical database. The real-time monitoring and early warning module is used to collect the real-time operating parameters of the environmental control system and input them into the optimized baseline model to obtain the predicted values of the state parameters of each subsystem's operating status. The predicted values are then compared with the measured values, and if the difference exceeds a preset threshold, a fault early warning signal is generated. The fault analysis module is used to initiate a fault tree analysis process corresponding to the faulty subsystem and fault phenomenon in response to the fault warning signal.
[0009] Thirdly, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described in the first aspect.
[0010] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.
[0011] The beneficial effects of this invention include: This invention utilizes a deeply optimized BP neural network baseline model, which can keenly capture subtle abnormal changes in system parameters, enabling early warning and high-precision diagnosis of faults. By combining data-driven neural networks with model-driven fault tree analysis, the neural network is responsible for discovering problems, while the fault tree, with its clear logical hierarchy, explains the problems and locates the root causes, overcoming the drawbacks of the opacity of neural networks. By retraining the model based on feedback data and building a fault case library, the system can continuously learn from new operational data and fault handling experience, enabling the diagnostic model and knowledge base to continuously evolve and possess lifelong learning capabilities. This invention covers the entire process from data preprocessing, model construction and optimization, real-time diagnosis, root cause analysis to knowledge accumulation and self-evolution, forming a complete, closed-loop intelligent fault diagnosis system. Attached Figure Description
[0012] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0013] Figure 1 This is a schematic diagram of the fault diagnosis method for the aircraft environmental control system according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the principle of the BP neural network (BPNN) in an embodiment of the present invention; Figure 3 This is a structural block diagram of the aircraft environmental control system fault diagnosis device according to an embodiment of the present invention. Detailed Implementation
[0014] The embodiments of this application will now be described in detail with reference to the accompanying drawings.
[0015] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0016] The fault diagnosis method for aircraft environmental control systems described in this invention has universal principles. This embodiment uses an unmanned aerial vehicle (UAV) environmental control system as an example for detailed explanation, but it should be understood that this method is also applicable to other types of aircraft, requiring only adaptive adjustments based on the specific aircraft model's environmental control system structure and parameters.
[0017] Reference Figure 1 The fault diagnosis method for an aircraft environmental control system provided in this embodiment of the invention mainly includes the following steps: S101: Historical Data Construction and Preprocessing This step involves collecting historical operating parameters of each subsystem of the environmental control system under normal operating conditions and performing data preprocessing to build a historical database. Details are as follows: First, historical data is collected, including data from each subsystem of the environmental control system during normal operation and fault-free periods. The subsystems of the environmental control system include the air conditioning system, pressurization system, filtration system, cooling system, and icing and anti-fogging system.
[0018] For the air conditioning system: collect parameters such as flow rate, pressure measurement points, heat exchange efficiency, dehumidification rate, and condensate discharge at various points in the system to determine its output operating status parameters as cabin temperature and humidity.
[0019] For pressurization systems: parameters such as pressure difference between the inside and outside of the cabin, air flow, cabin pressure altitude, and system leakage rate are collected to determine the output operating status parameter as the cabin pressure.
[0020] For filtration systems: collect parameters such as filtration efficiency, pressure loss, and filter life to determine the output operating status parameter as the pollutant concentration.
[0021] For the cooling system: collect parameters such as refrigerant flow rate, temperature, pressure, and heat exchange efficiency at various points to determine the output operating status parameters as system energy consumption.
[0022] For the icing and anti-fog system: collect parameters such as the power of each heating element in the system, the temperature and humidity inside the cabin, and the meteorological data outside the cabin to determine the icing rate as the output operating status parameter.
[0023] Data Preprocessing: After collecting parameters, the collected raw historical operating parameters are preprocessed, specifically including the identification and handling of missing and outlier values. First, outlier and missing values are identified. In an exemplary embodiment, a traversal method is used to identify missing values. The program sequentially checks each data point in the dataset to determine if it is missing. If the point has no value or is an invalid value, it is marked as a missing value. The Tukey fence method is used to identify outliers. The program marks values outside the normal range as outliers based on the calculation of the quartiles and interquartile ranges of the data. Specifically, first, the 25th percentile of the data to be processed is calculated (…). ) and 75th percentile ( ); then, the interquartile range ( Its value is and The difference, that is = - Based on this, calculate the upper limit of the reasonable range of normal data ( ) and lower limit ( The calculation formula is as follows:
[0024]
[0025] Ultimately, the values in the dataset that are higher than or below The data points are marked as outliers.
[0026] Then, for the identified missing and outlier values, a K-Means clustering algorithm based on Euclidean distance is used to repair them. The specific process is as follows: for each data point to be repaired, it is treated as a multi-dimensional vector, and the Euclidean distance between this point and all other normal data points in the historical database is calculated to measure the similarity between the data. The formula for calculating the Euclidean distance is as follows:
[0027] in, For vectors and Euclidean distance, and They are vectors and The i-th component, k is the total dimension of the vector.
[0028] Subsequently, the N normal data points that are closest to the data point to be repaired in Euclidean distance are selected, and the weighted average of the values of these N data points in the missing or abnormal dimension is taken as the repair value to complete the repair; for example, N=5.
[0029] Discrete-level averaging: To reduce data fluctuations and extract steady-state features, the repaired data is discretized and aggregated according to time windows. Specifically, based on the total duration of the flight mission, the continuous time series data is divided into equal-length discrete time intervals. The length of these intervals can be set to 5 minutes, 10 minutes, or 30 minutes, depending on the data processing requirements. Within each time interval, the arithmetic mean of all normal data points falling within that interval is calculated to smooth out short-term noise.
[0030] Historical database construction: Finally, the standardized data, after missing and outlier repair and discretization averaging by time window, is stored in a structured storage system, thus completing the construction of the historical database.
[0031] S102: Baseline Model Establishment and Optimization This step involves establishing a baseline model for each subsystem based on a backpropagation (BP) neural network, using historical databases as input and state parameters representing the operating state of each subsystem as output. The structural parameters of each BP neural network are then optimized using the minimization of mean absolute deviation (MAD) as the evaluation metric to obtain the optimized baseline model for each subsystem. Specifically: First, the model framework is constructed: Based on the aforementioned historical database, a baseline model based on a BP neural network is established separately for the air conditioning system, pressurization system, filtration system, cooling system, and icing and anti-fogging system under normal operating conditions. The input of each model is the various operating parameters of that subsystem, and the output is the state parameters of the operating status of that subsystem. The principle of the BP neural network is as follows: Figure 2 As shown, it includes an input layer, a hidden layer, and an output layer. For example, taking the training of an air conditioning system as an example, the input layer of the BP neural network model is "parameters such as flow rate, pressure measurement points, heat exchange efficiency, dehumidification rate, and condensate discharge at various points in the system", and the output layer is "cabin temperature and humidity"; the other four systems are similar, and finally five neural network models that can predict the performance of each system are obtained.
[0032] Then, model structure optimization is performed. After obtaining the basic BP neural network model, its structural parameters need to be systematically optimized to improve its prediction accuracy and generalization ability, ultimately forming a reliable optimized baseline model. This embodiment adopts a grid search strategy guided by minimizing the mean absolute deviation (MAD) to independently optimize the structural parameters of the BP neural network for each subsystem. Specifically, this embodiment uses the mean absolute deviation (MAD) as the evaluation index, and searches for the network structure that minimizes MAD by traversing the number of intermediate layer nodes of the BP neural network. MAD can intuitively reflect the average absolute difference between the predicted value and the true value, is insensitive to outliers, and has good robustness. The formula for calculating MAD is as follows:
[0033] in, To verify the actual measured value of the state parameter of the j-th sample in the set, Let m be the predicted value of the state parameter of the j-th sample by the BP neural network model, and m be the total number of sample data in the validation set.
[0034] The direct goal of optimization is to find the model structure that minimizes the MAD value on the validation set.
[0035] The specific method for traversal optimization is as follows: This embodiment finds the optimal configuration by traversing the intermediate (hidden) layers of the BP neural network. The specific optimization method is as follows: First, determine the traversal range for the network structure and the number of nodes. For example, set the number of hidden layers to 1 to 2, and the number of nodes in each layer to between 5 and 50. This range covers common network structures.
[0036] To ensure fairness and comparability in structure optimization, hyperparameters such as learning rate, activation function type, and the ratio of training set / validation set / test set split are set to fixed values.
[0037] Then, a systematic traversal is performed on all possible combinations of the number of hidden layers and the number of nodes in each layer. For each structural combination, the model is trained using the same historical data, and its MAD value is calculated on an independent validation set to objectively evaluate the model performance under that structure.
[0038] Finally, after testing all structural combinations, the combination of hidden layers and nodes that minimizes the MAD value on the validation set is selected as the final network structure of the BP neural network for this subsystem. This structure is considered to have the best prediction accuracy.
[0039] Through the above systematic optimization process, a highly adapted neural network structure was determined for each subsystem, resulting in five high-performance optimized baseline models, which provide accurate comparison benchmarks for subsequent real-time fault diagnosis.
[0040] S103: Real-time fault warning This step involves using the established optimized baseline model to perform online monitoring and health status assessment of the environmental control system, enabling real-time detection and alarm of potential faults. This includes collecting real-time operating parameters of the environmental control system and inputting them into the optimized baseline model to obtain predicted values of the status parameters of each subsystem; comparing the predicted values with the measured values; if the difference exceeds a preset threshold, a fault is determined in the corresponding subsystem, and a fault warning signal is generated. Specifically: During the real-time operation of the aircraft, the operating parameters of each subsystem of the environmental control system are collected synchronously. To ensure consistency with the model training data, the types, sources, and preprocessing procedures of the collected parameters are exactly the same as those used in step S101 when building the historical database, and will not be repeated here.
[0041] The preprocessed real-time operating parameters are used as input vectors and output to the optimized baseline models corresponding to the five subsystems obtained in step S102. Each model performs a forward calculation once and outputs the predicted values of the state parameters of the corresponding subsystem operating state at the next time step (such as the predicted values of cabin temperature and humidity output by the air conditioning system).
[0042] The system synchronously acquires the measured values of the state parameters of each subsystem obtained by the sensors. Then, the predicted values output by the optimized baseline model are compared with the measured values item by item in real time. To quantify the degree of deviation between the two, the relative error is calculated using the following formula: Relative error =
[0043] A preset fault judgment threshold is set for the relative error of each type of state parameter, generally between 15% and 20%. In this embodiment, the threshold is set to 18%. The system continuously monitors the relative error. If the relative error is consistently lower than or equal to the threshold, the corresponding subsystem is determined to be operating in accordance with the normal baseline, and the system is working normally. If the relative error of any subsystem's state parameter consistently or momentarily exceeds the preset threshold, it indicates that the actual operating performance of the subsystem has significantly deviated from its normal behavior pattern. Based on this, the system determines that the subsystem has failed and immediately generates and sends a corresponding fault warning signal.
[0044] This step represents a leap from data monitoring to intelligent judgment, transforming latent performance degradation into explicit and actionable early warning information, providing clear triggering conditions for subsequent root cause analysis.
[0045] S104: Fault Root Cause Analysis This step involves upgrading a simple anomaly alarm to root cause localization after a fault warning is generated in step S103, providing maintenance personnel with a clear troubleshooting path. This includes initiating a fault tree analysis process corresponding to the faulty subsystem and its symptoms in response to the aforementioned fault warning signal, in order to locate the root cause of the fault. Specifically: Upon receiving a fault warning signal from step S103 (such as an abnormal temperature rise caused by a cooling system malfunction), the system automatically triggers a fault tree analysis (FTA) process corresponding to the fault phenomenon. The FTA process uses the specific fault phenomenon as the top event. Based on a pre-set expert knowledge base, the system constructs a multi-level fault tree model from top to bottom, using various factors that may lead to the top event as intermediate and bottom events. Intermediate events are direct but decomposable causes that lead to higher-level events. Bottom events are the most basic fault causes in the fault tree, requiring no further decomposition or decomposability, typically corresponding to specific component failures, abnormal external inputs, or human error.
[0046] For example, the top event, an abnormal rise in cabin temperature, can be directly broken down into two intermediate events: insufficient cooling capacity and excessive heat load. Furthermore, the intermediate event of insufficient cooling capacity can be further broken down into bottom events such as refrigeration compressor failure, poor condenser heat dissipation, expansion valve sticking, and refrigerant leakage.
[0047] In a fault tree, events at different levels are connected by logic gates (AND gates, OR gates, etc.) to clearly represent the causal relationships between these events. When any lower-level event occurs, the higher-level event also occurs, using an OR gate. The higher-level event occurs only when all lower-level events occur simultaneously, using an AND gate. For example, the intermediate event of insufficient cooling capacity can be caused by any one of the following lower-level events: compressor failure, poor condenser heat dissipation, stuck expansion valve, or refrigerant leakage; therefore, these events are related by an OR gate.
[0048] Through this clearly structured and logically rigorous fault tree, the system can link abstract top events (fault phenomena) with specific bottom events (root causes). Maintenance personnel or automated diagnostic programs can use this logic diagram to reason and verify layer by layer downwards from the top event, following the path indicated by the logic gates. This systematically eliminates unrelated factors and quickly locates the most likely root cause of the fault (one or more bottom events), greatly improving troubleshooting efficiency and accuracy.
[0049] S105: Diagnostic Closed Loop and Case Library Establishment This step involves collecting feedback data after each fault handling process, and then retraining and optimizing the parameters of the optimized baseline model based on this data. Simultaneously, a fault case library is established.
[0050] Specifically, the system automatically collects feedback data after each fault handling, forming a dynamically growing dataset. This dataset includes system operating parameters before and after fault confirmation, as well as the final cause of the fault confirmed by maintenance personnel. Based on the feedback data, the system periodically (e.g., monthly) or under specific triggering conditions (e.g., continuous prediction deviations) incorporates samples of the feedback data into the historical database. Subsequently, the system uses the expanded database to retrain and optimize the parameters of the optimized baseline model. This process enables the model to learn the latest normal operating characteristics of the system and newly emerging fault modes, thereby continuously improving its prediction accuracy and generalization ability.
[0051] Simultaneously, the system establishes and maintains a structured fault case library to record and categorize all occurring fault cases in detail. Each complete diagnostic cycle is converted into a standardized case record. The fault case library is used to record in detail the fault phenomenon, the determined fault cause, the specific handling process, and the final handling result.
[0052] Statistical analysis of the case library can identify high-frequency failure modes, thereby optimizing the structure of the Fault Tree Analysis (FTA) process, such as by adding new base events or adjusting logical relationships. Simultaneously, by utilizing backpropagation (BP) neural networks to predict the patterns and trends of failure occurrence, best practice solutions for various types of failures can be summarized, especially handling experience with common failures, providing decision support for subsequent diagnosis and maintenance.
[0053] Those skilled in the art will understand that the above method can be implemented in various forms. For example, the present invention can be embodied as a virtual device or system, in which each module corresponds one-to-one with each step of the above method, and is configured by computer program code, performing corresponding functions when running on a processor.
[0054] Specifically, embodiments of the present invention also provide a fault diagnosis device for an aircraft environmental control system. This device can be implemented through software, hardware, or a combination of both, and deployed on the aircraft's onboard computer or ground control station. Figure 3 As shown, the device mainly includes the following functional modules: The data preprocessing module is used to collect historical and real-time operating parameters of each subsystem of the environmental control system under normal operating conditions, and to preprocess the data to build a historical database. The preprocessing includes using a traversal method to identify missing values, using the Tukey fence method to identify outliers, and using a K-Means clustering algorithm based on Euclidean distance to repair the identified missing and outliers.
[0055] The baseline model management module is used to establish and optimize the baseline model based on the BP neural network for each subsystem based on the historical database. The optimization is based on the mean absolute deviation (MAD) as the evaluation index. By traversing the number of intermediate layer nodes of the BP neural network, the network structure that minimizes the MAD is found, thereby obtaining the optimized baseline model for each subsystem.
[0056] The real-time monitoring and early warning module is used to collect the real-time operating parameters of the environmental control system and input them into the optimized baseline model to obtain the predicted values of the state parameters of each subsystem's operating status; the predicted values are compared with the measured values, and if the difference exceeds a preset threshold, a fault early warning signal is generated.
[0057] The fault analysis module is used to respond to the fault warning signal and initiate a fault tree analysis process corresponding to the faulty subsystem and fault phenomenon. The fault tree takes the fault phenomenon as the top event, and various factors that may cause the top event as intermediate events and bottom events. Logic gates are used to represent the causal logical relationship between events in order to locate the root cause of the fault.
[0058] In addition, the device may also include a self-learning optimization module (not shown separately in the figure, but may be used as an independent module or integrated into the above modules), which is used to collect feedback data after each fault handling, and to retrain and optimize the parameters of the optimized baseline model based on the feedback data; at the same time, a fault case library is established and maintained to record fault phenomena, fault causes, handling processes and handling results.
[0059] The above modules interact with each other through the system bus to complete the complete fault diagnosis task, from data acquisition, model building, real-time diagnosis, root cause analysis to self-optimization.
[0060] This invention also provides an electronic device, which includes at least one processor (CPU), a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described method steps. The specific steps are as described above and will not be repeated here.
[0061] Furthermore, embodiments of the present invention also provide a computer-readable storage medium, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), flash memory (FLASH), a solid-state drive (SSD), or any other non-transitory storage medium. One or more computer programs (or software units, instructions) are stored on this storage medium. When these programs are executed by a processor of a device (such as the aforementioned electronic device), the device performs the steps of the aforementioned aircraft environmental control system fault diagnosis method, the specific steps of which are as described above and will not be repeated here.
[0062] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for diagnosing faults in an aircraft environmental control system, characterized in that, Includes the following steps: Collect historical operating parameters of each subsystem of the environmental control system under normal operating conditions and perform data preprocessing to build a historical database; Based on the historical database, a baseline model based on a backpropagation neural network is established for each subsystem. The operating parameters of each subsystem are used as inputs, and the state parameters that characterize the operating state of each subsystem are used as outputs. Using the minimization of mean absolute deviation as the evaluation metric, the structural parameters of each BP neural network are optimized to obtain the optimized baseline model of each subsystem. The real-time operating parameters of the environmental control system are collected and input into the optimized baseline model to obtain the predicted values of the state parameters of each subsystem's operating status; the predicted values are compared with the measured values, and if the difference exceeds a preset threshold, the corresponding subsystem is determined to have a fault and a fault warning signal is generated; In response to the fault warning signal, a fault tree analysis process corresponding to the faulty subsystem and fault phenomenon is initiated to locate the root cause of the fault.
2. The method according to claim 1, characterized in that, The data preprocessing includes using a traversal method to identify missing values, using the Tukey fence method to identify outliers, and using a K-Means clustering algorithm based on Euclidean distance to repair the identified missing and outliers.
3. The method according to claim 1, characterized in that, The optimization of the structural parameters of each BP neural network is specifically carried out by using the mean absolute deviation as an evaluation index and searching through the number of intermediate layer nodes of the BP neural network to find the network structure that minimizes the mean absolute deviation.
4. The method according to claim 1, characterized in that, The fault tree analysis process takes the fault phenomenon as the top event, and various factors that may lead to the occurrence of the top event as intermediate events and bottom events, and uses logic gates to represent the causal logical relationship between events.
5. The method according to claim 1, characterized in that, The environmental control system includes subsystems such as an air conditioning system, a pressurization system, a filtration system, a cooling system, and an icing and anti-fogging system.
6. The method according to claim 1, characterized in that, The method also includes collecting feedback data after each fault handling, and retraining and optimizing the parameters of the optimized baseline model based on the feedback data.
7. The method according to claim 1 or 6, characterized in that, The method also includes establishing a fault case library, which is used to record fault phenomena, fault causes, handling processes, and handling results.
8. A fault diagnosis device for an aircraft environmental control system, characterized in that, include: The data preprocessing module is used to collect historical operating parameters of each subsystem of the environmental control system under normal operating conditions and perform data preprocessing in order to build a historical database; The baseline model management module is used to establish and optimize baseline models based on BP neural networks for each subsystem based on the historical database. The real-time monitoring and early warning module is used to collect the real-time operating parameters of the environmental control system and input them into the optimized baseline model to obtain the predicted values of the state parameters of each subsystem's operating status. The predicted values are then compared with the measured values, and if the difference exceeds a preset threshold, a fault early warning signal is generated. The fault analysis module is used to initiate a fault tree analysis process corresponding to the faulty subsystem and fault phenomenon in response to the fault warning signal.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.