Aircraft fault detection method based on random projection distance prediction

Abstract

The invention discloses an aircraft fault detection method based on random projection distance prediction, and relates to the field of aircraft intelligent fault detection, and the method comprises the steps: S1, collecting air flow, air pressure and temperature data through a pressure adjustment stop valve sensor system installed on an aircraft; s2, preprocessing the data acquired in the step S1; s3, constructing an unsupervised anomaly detection model; s4, training and optimizing an unsupervised anomaly detection model; and S5, testing and optimizing the unsupervised anomaly detection model. The abnormal state of the pressure regulating stop valve (PRSOV) is efficiently and accurately detected in an aviation health monitoring system (HUMS), so that the safety of aircraft operation is improved, the maintenance cost is reduced, and unplanned stop events are reduced.

Description

Technical Field

[0001] The present invention relates to aircraft intelligent fault detection, and particularly to an aircraft fault detection method based on random projection distance prediction. Background Art

[0002] Air transportation is the most convenient means of transportation in modern society, which not only greatly promotes global economic activities but also strengthens the connection between people. However, during flight, an aircraft may experience drastic environmental changes, such as fluctuations in speed, temperature, humidity, and air pressure. These changes can have a negative impact on on-board equipment, especially the performance of core components may decline accordingly, or even anomalies may occur, posing severe challenges to the operation and maintenance of the aircraft. Among them, the pressure regulating shutoff valve (PRSOV) is a key component of the aircraft environmental control system (ECS). They are mainly used to regulate gas flow to ensure the supply of appropriate hot air to the cabin under different flight phases and external environmental conditions. If the PRSOV fails, it may lead to cabin depressurization and abnormal temperature, and in severe cases, may cause aviation safety accidents. Therefore, a reliable and accurate prognostics and health management (PHM) system is crucial for reducing unplanned maintenance, improving flight safety, and reducing operation risks. Currently, a large number of sensors (possibly up to hundreds or thousands) are equipped on aircraft to monitor different operating states, and these sensors generate a vast amount of data. To extract effective information from these data and conduct health status monitoring, data-driven algorithms have become key technologies and are widely applied to the health monitoring and fault diagnosis tasks of key aviation components.

[0003] In early aircraft health monitoring research, researchers mainly relied on experience to identify and evaluate abnormal states by setting thresholds or logical rules. However, this method has limited efficiency when dealing with complex systems and large-scale data, and it is difficult to accurately judge abnormal states. To handle the increasing complexity and data scale, feature engineering has become a key step in algorithm development. Researchers have proposed various feature extraction algorithms that utilize domain knowledge and data characteristics to extract representative fault features from high-dimensional data. For example, Fourier transform and wavelet transform are widely used in signal processing to extract frequency features. However, these methods still have limitations, mainly in two aspects: on the one hand, manual feature engineering heavily relies on expert experience, is prone to introducing biases, and cannot fully capture the dynamic characteristics of aircraft systems; on the other hand, traditional machine learning algorithms have limited non-linear modeling capabilities and perform poorly when facing high-dimensional, large-scale data, making it difficult to adapt to the complex operating environment of aircraft.

[0004] In recent years, deep learning technology, as a powerful data-driven method, has become an effective means for aircraft fault detection and diagnosis due to its excellent automatic feature learning ability. In particular, PRSOV, as a key component, has received increasing attention from researchers. However, the success of deep learning in this field mainly relies on a large amount of labeled abnormal data. However, in actual operation, due to the rarity and diversity of aircraft fault events, it is difficult to collect a large-scale labeled data covering all fault types. In addition, accurately labeling all types of fault data is both time-consuming and expensive. Therefore, unsupervised anomaly detection (UAD) has received extensive attention. The core idea of this method is to train only based on normal data and automatically extract representative features through self-supervised proxy tasks. The UAD method learns the distribution characteristics and latent structure of normal data without relying on abnormal data, so it has great application potential in aviation application scenarios. However, our analysis found that in the field of PRSOV, unsupervised anomaly detection still faces three key challenges.

[0005] First, although the UAD method has been successful in many industrial application fields, the current research on unsupervised anomaly detection for PRSOV is still very limited. Most of the existing methods focus on general industrial equipment, and there are few studies on the optimization and adaptability of PRSOV. Second, there are also huge challenges in directly applying existing UAD methods to PRSOV. For example, the latest Transformer-based unsupervised anomaly detection models usually require a lot of computing resources, which far exceeds the computing power of typical avionics processors. Therefore, it is particularly critical to develop an efficient and computationally inexpensive UAD method. Third, most UAD methods lack theoretical guarantees, and aviation systems have extremely high reliability requirements, which makes methods that lack theoretical support face greater uncertainty in practical applications. Therefore, how to balance computing resource constraints, theoretical interpretability, and high detection accuracy in the framework of unsupervised anomaly detection has become an urgent problem to be solved in current research. Summary of the invention

[0006] The present invention provides an aircraft fault detection method based on random projection distance prediction, which realizes real-time and high-precision anomaly detection.

[0007] According to one aspect of the present disclosure, there is provided an aircraft fault detection method based on random projection distance prediction, the method comprising: S1, collects air flow, air pressure and temperature data through the pressure regulating stop valve sensor system installed on the aircraft; S2, preprocessing the data collected in S1; In order to build an efficient unsupervised anomaly detection model, this study proposed RPDP-AD (unsupervised PRSOV multi-fault anomaly detection algorithm based on random projection distance prediction). Its core technology S3 builds an unsupervised anomaly detection model, including: Random projection dimensionality reduction: Use a random projection matrix to project high-dimensional input data into a low-dimensional space, and preserve the relative distance relationship between data points as much as possible; according to the Johnson-Lindenstrauss theorem, ensure that the projected data still retains the distribution information of the original data; Supervision signal construction: By calculating the distance between data points in the projection space, a supervision signal is constructed so that the model can learn the pattern of normal data in an unsupervised environment; Lightweight MLP multi-layer perceptron network: Due to the limited computing resources of avionics equipment, MLP networks are used to replace computationally intensive models to achieve efficient reasoning; Distribution difference measure: by minimizing the distribution difference between the learnable mapping and the random mapping, it ensures that the model's projection in the low-dimensional space remains stable and interpretable; The finally constructed RPDP-AD model can learn the normal PRSOV operation mode in the absence of labeled data and identify abnormal signals deviating from the normal mode during the detection process.

[0008] S4. Train and optimize the unsupervised anomaly detection model; S5. Test and optimize the unsupervised anomaly detection model.

[0009] The data mainly comes from the pressure regulating shut-off valve (PRSOV) sensor system installed on the aircraft. These sensors are used to monitor the operating status of the PRSOV at different flight stages. Data acquisition involves multiple sensor positions and measurement parameters, including: air flow and pressure parameters: monitoring the air flow, air pressure, and temperature inside the PRSOV to identify abnormal flow fluctuations or airtightness problems. Fault type data: If available, label the sensor data corresponding to known fault modes (such as leakage, blockage, mechanical wear) for subsequent performance evaluation.

[0010] The collected data is in time series format, and timestamps are recorded with a high-precision clock to ensure the synchronization of multi-sensor data for subsequent data fusion and analysis.

[0011] In a possible implementation, the method further includes: S6. Aircraft actual verification and deployment, including: After laboratory testing and optimization, finally deploy the model to the actual aircraft health monitoring system to achieve real-time fault detection: On-board verification: During flight missions, collect PRSOV sensor data in real time and perform online detection through the model; compare the model detection results with the manual inspection results to verify its accuracy and stability; Embedded deployment: Due to limited computing resources of avionics equipment, after optimizing the model calculation efficiency, deploy it to the on-board computing platform or cloud monitoring system to achieve real-time analysis and remote monitoring; Real-time monitoring and maintenance optimization: Integrate this model into the aircraft health monitoring system to achieve all-weather online monitoring, provide early warnings to maintenance personnel, optimize the maintenance plan, and reduce unplanned maintenance.

[0012] The invention can provide efficient, accurate, and low-computation-overhead anomaly detection in the actual aircraft operating environment, significantly improve the operating safety of the PRSOV, and reduce the aviation operation cost.

[0013] In a possible implementation, the pressure regulating shut-off valve sensor system monitors the air flow, air pressure, and temperature inside the pressure regulating shut-off valve sensor to identify abnormal flow fluctuations or airtightness problems; The collected data are all in time series format, and timestamps are recorded with a high-precision clock to ensure the synchronization of multi-sensor data for subsequent data fusion and analysis.

[0014] Since the actually collected data usually contains noise, missing values, and irregular variations, data preprocessing is required to improve data quality and enhance the learning effect of the model. The main preprocessing steps include: In a possible implementation, S2, preprocess the data collected in S1, including: Denoising: Use a low-pass filter to remove high-frequency noise to ensure the smoothness of the vibration signal; process the sensor data through Kalman filtering to make it more stable and reduce the influence of environmental noise; use wavelet transform denoising to extract main features and reduce the interference caused by sensor errors; Normalization: Use Z-score standardization or Min-Max normalization to ensure that all data is kept within the same numerical range to reduce the influence of scale differences between features; Time series slicing: Use a sliding window method to divide the data into fixed time windows for input into the model for analysis; Data augmentation: Apply data augmentation techniques, including: time warping, adding noise, signal mixing, to improve the generalization ability of the model.

[0015] After the model is constructed, training and optimization are required to ensure that the model can effectively learn the distribution characteristics of normal data.

[0016] In a possible implementation, S4, train and optimize an unsupervised anomaly detection model, including: Unsupervised training: Only use normal data to train the model to minimize the distance prediction error in the random projection space, so that the model can learn the structural characteristics of the normal operating state of the pressure regulating shut-off valve PRSOV; Loss function optimization: Adopt a distribution difference metric loss to ensure that the projected low-dimensional data can still maintain the spatial structure of the high-dimensional data; Hyperparameter tuning: Conduct a grid search to adjust hyperparameters such as the learning rate, projection dimension, and number of MLP layers to improve the stability and performance of the model; Regularization: Adopt L2 regularization to prevent overfitting so that the model can generalize to different PRSOV operating conditions.

[0017] After the model training is completed, it needs to be tested to evaluate its detection ability under different fault modes and operating environments.

[0018] In one possible implementation, S5, testing and optimizing the unsupervised anomaly detection model includes: Offline testing: Test on the real PRSOV sensor dataset, calculate key indicators such as AUC, accuracy, recall, F1 score, and evaluate the detection performance of the model; verify the contribution of different modules to the model performance through ablation experiments and optimize the overall architecture; Noise sensitivity analysis: adding different levels of noise to the test data to evaluate the robustness of the model in practical applications; Multiple fault mode detection: Using data from 7 known fault modes, we test whether the model can effectively distinguish different fault types; fine-tune the model based on the test results to ensure its stability under various flight conditions. Using data from 7 known fault modes, we test whether the model can effectively distinguish different fault types.

[0019] The model is fine-tuned based on the test results to ensure its stability under various flight conditions.

[0020] Compared with the prior art, the present invention has the following beneficial effects: The present invention discloses an aircraft fault detection method based on random projection distance prediction in an embodiment. In order to overcome the above challenges, Random Projection Theory provides an effective solution. Random projection is a dimensionality reduction technique that uses a random projection matrix to project high-dimensional data into a low-dimensional space while retaining the distance relationship between data points as much as possible. In a high-dimensional space, the relative position and distance between data points reflect their similarities and differences, and the random projection method retains the structural characteristics of the original data by maintaining these distance relationships. In recent years, researchers have tried to combine random projection with deep learning to improve the feature extraction capability of the model. However, the current research on industrial unsupervised anomaly detection is still in the theoretical exploration stage, and a systematic method system has not yet been formed.

[0021] To address the applicability issue of the UAD method in the PRSOV field and further explore the potential of the deep learning-based random projection theory in industrial anomaly detection, this study proposes an unsupervised PRSOV multi-fault anomaly detection algorithm based on random projection distance prediction (RPDP-AD). RPDP-AD constructs a supervision signal by leveraging the distance information in the random projection space, enabling the model to automatically encode the data structure and distribution pattern of normal samples in an unsupervised environment. Additionally, we introduce a distribution difference measurement method to minimize the difference between the learnable mapping and the random mapping in the low-dimensional space. Different from existing deep learning methods, our method maintains theoretical interpretability through a fixed random mapping while achieving adaptive feature learning through a trainable projection head. To our knowledge, this study is the first to combine the random projection theory with a deep neural network for application in industrial unsupervised anomaly detection, especially in the field of aircraft component monitoring. Furthermore, to provide theoretical support, we analyze the inner product preservation property based on the Johnson-Lindenstrauss theorem, further demonstrating the feasibility of the combination of random projection and deep learning.

[0022] Notably, different from other computationally intensive models, RPDP-AD adopts a simple MLP (multi-layer perceptron) network structure, thus achieving real-time and high-precision anomaly detection with limited computational resources and effectively meeting the application requirements in the aviation field. Experiments conducted on a real PRSOV vibration dataset (containing 7 different types of faults) show that RPDP-AD outperforms existing mainstream UAD methods in multiple fault types. Additionally, we also conduct a sensitivity analysis to verify the stability of RPDP-AD under different random projection matrices and noise conditions. This study not only provides a new technical solution for aircraft health monitoring but also further expands the application of the random projection theory in deep learning, laying an important foundation for future unsupervised anomaly detection research.

[0023] The invention can efficiently and accurately detect the abnormal state of the pressure regulating shut-off valve (PRSOV) in the aircraft health monitoring system (HUMS), thereby improving the safety of aircraft operation, reducing maintenance costs, and reducing unplanned downtime events. In practical applications, traditional anomaly detection methods often rely on manually setting thresholds or expert-driven feature engineering, and cannot adapt to complex and high-dimensional sensor data. Existing deep learning methods, on the other hand, rely heavily on a large amount of labeled anomaly data, making it difficult to meet the requirements of the aviation industry. In contrast, the RPDP-AD (unsupervised PRSOV multi-fault anomaly detection algorithm based on random projection distance prediction) proposed in the present invention adopts a completely unsupervised learning method, only relying on normal data for training, without the need to collect expensive and scarce fault samples, which greatly improves the deployability and scalability of the method. In addition, the method also performs excellently in an environment with limited computing resources. Different from the computationally intensive Transformer architecture, RPDP-AD adopts a lightweight MLP network, which can run on the embedded platform of avionics equipment to achieve real-time detection without affecting the execution of other critical tasks. Experimental results show that the method outperforms existing unsupervised anomaly detection algorithms on the real PRSOV sensor dataset, achieving higher detection accuracy in key metrics such as AUC and F1 score, and can maintain stable performance under 7 different fault modes. In addition, sensitivity analysis further verifies the robustness of RPDP-AD under different random projection matrices and noise conditions, which means that the method can adapt to different aircraft models and operating conditions and maintain high-precision detection capabilities. In practical applications, RPDP-AD can be seamlessly integrated into the existing aircraft health monitoring system, and by real-time monitoring the operating state of the PRSOV, it can timely identify potential faults, provide warning information to maintenance personnel, optimize maintenance strategies, and avoid flight interruptions caused by sudden failures. At the same time, the method can be extended to the health monitoring of other key aircraft components (such as hydraulic systems, electrical control systems, etc.), providing reliable technical support for the intelligent maintenance of a wider range of aviation industries. Therefore, the invention not only advances the combination of random projection theory and deep learning in industrial anomaly detection theoretically, but also provides an efficient, low-cost, interpretable and easy-to-deploy solution for aircraft health monitoring in practical applications, greatly improving aviation safety and operational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Schematic diagram of a pressure regulating and shut-off valve showing an embodiment of the present disclosure.

[0025] Figure 2 Examples of PRSOV signals of different categories showing an embodiment of the present disclosure.

[0026] Figure 3 Shows the workflow of the RPDP-AD framework according to an embodiment of the present disclosure.

[0027] Figure 4 Shows an aircraft fault detection algorithm based on random projection distance prediction according to an embodiment of the present disclosure. Detailed implementation manners

[0028] The following will describe in detail various exemplary embodiments, features and aspects of the present disclosure with reference to the accompanying drawings. The same reference numerals in the drawings denote elements having the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0029] The special term "exemplary" herein means "serving as an example, embodiment or illustration". Any embodiment described herein as "exemplary" is not necessarily to be construed as superior or better than other embodiments.

[0030] In addition, for better illustration of the present disclosure, numerous specific details are given in the following detailed implementation manners. Those skilled in the art should understand that the present disclosure can be implemented without some specific details. In some instances, methods, means, elements and circuits well known to those skilled in the art are not described in detail so as to highlight the gist of the present disclosure.

[0031] Theoretical basis: We provide a theoretical basis analysis for the proposed RPDP-AD anomaly detection model. Specifically, we explore the inner product preservation property based on the Johnson-Lindenstrauss lemma, which links the random projection theory with deep learning neural networks.

[0032] Random projection is an efficient dimensionality reduction technique. Given a high-dimensional data matrix , where n is the number of data points and d is the original dimension of the data. Random projection uses a random matrix to project the data into a k-dimensional low-dimensional space (where ) to achieve dimensionality reduction. The Johnson-Lindenstrauss lemma provides a theoretical basis for this algorithm, which shows that by using a random projection matrix, data in a high-dimensional space can be projected into a low-dimensional space and the distance relationship between data points can be maintained to a certain extent, which has also been strictly proven mathematically. The theorem is expressed as follows: Theorem 1. (Johnson-Lindenstrauss) Let . Let be a set containing n data points, . Then there exists a Lipshcitz mapping such that for all : (1) Subsequently, Theorem 2 clearly shows the norm-preserving property of random projection, which is expressed as follows.

[0033] Theorem 2. (Norm Preservation) Let . Assume that All elements in are independently sampled from N (0, 1). Then, (2) Theorem 2 requires that the random projection matrix A is independently sampled from a Gaussian distribution. However, Lemma 1 shows that using a Gaussian distribution is not strictly necessary. Many other distributions with unit variance and satisfying certain boundedness conditions (or high-order moment conditions) are equally sufficient.

[0034] Lemma 1. Assume that for ， Each element is uniformly distributed in . Then for any vector (3) (4) Based on the above theorem and lemma, a simple corollary can be obtained, that is, the inner product is preserved under random projection.

[0035] Corollary 1. Let and . Let where A is a matrix, and each element in the matrix is independently and identically distributed sampled from the Gaussian distribution N(0, 1) (or from U(-1, 1)). Then there is (5) The above theory confirms that random projection can preserve the structure, features, and inner product of the original data. This further provides a solid theoretical basis for the proposed RPDP-AD anomaly detection method.

[0036] II. Random Projection Distance Estimation As can be seen from Corollary 1, the inner product in the random projection space preserves the potential class results and feature information of the original data. For this reason, we train a learnable model to predict the inner product in the random projection to force it to represent the potential information in the data. This can provide a strong and effective supervision signal for the training of the model without any manual labels, thus realizing the feature representation and training of large-scale data without labels. The proposed RPDP-AD model is shown as Figure 3As shown. Assume that the PRSOV dataset is obtained from a real scenario or an aircraft simulation model , , where N is the total number of samples. Randomly select a sample pair D from the dataset , and then send them into a weight - shared siamese network architecture respectively. This network is denoted as , which maps the input samples from the d - dimensional space to a new k - dimensional low - dimensional space. are learnable network parameters. Similarly, we also need to establish a random projection function , which also randomly projects the input samples from the d - dimensional space to a new k - dimensional low - dimensional space. For this purpose, RPDP - AD learns the implicit structural information in the original data space by establishing an inner - product difference loss function, which is expressed as follows: (6) The distance information (inner product) between data points contains important class - structure information. For example, in the feature space, the relative positions and distances between data points reflect their similarities and differences. This is crucial for encoding the distribution boundaries between different classes. In other words, random projection preserves this distance information between data points during the dimensionality - reduction process. For the anomaly - detection task, the model only uses normal samples and aims to establish the distribution boundary of normal samples. RPDP - AD forces the model to characterize the distances and similarities in the data space through the inner - product prediction in the random space, thereby accurately encoding the distribution of normal data patterns. Thus, the model effectively learns the internal structure of normal data, making it easier to detect abnormal data that deviates from the learned distribution.

[0037] III. Distribution Difference Metrics As Figure 3 shown, the proposed RPDP - AD encodes the relationship between class results and sample distances using inner - product prediction in the random space. This section further introduces distribution - difference metrics to enhance the model's recognition performance for abnormal classes, which is expressed as follows: (7) It can be seen that aims to minimize the distribution difference in two mapping spaces (the mapping space of the neural network and the mapping space of random projection), so that the feature mapping of the neural network also has the characteristics of class structure and distance preservation. In addition, after the normal samples go through the above optimization training, the model will obtain a lower distribution - difference metric. When a new abnormal sample is input, the feature structure of the normal samples encoded by the model will no longer be applicable, and there will be a significant increase. This can be a very good supplement to the anomaly - monitoring signal.

[0038] IV. Total Model Loss The multi-fault anomaly monitoring model proposed in this study is based on the Johnson-Lindenstrauss lemma and realizes the distribution encoding of the normal mode through the inner product prediction in the random projection space. The proposed RPDP-AD uses the inner product difference and distribution difference loss functions as the supervision signals to optimize the model parameters, which is expressed as: (8) where and are the weight hyperparameters of the two losses respectively. Since the obtained PRSOV data samples are 201 pressure features, the weight-sharing siamese network in RPDP-AD uses a multi-layer perceptron to avoid overfitting. The Adam optimizer is used with a learning rate of 0.001 and the batch size is set to 64. RPDP-AD is implemented using Pytorch and Python 3.8 and is trained and tested on a workstation with a 3090 GPU.

[0039] Case Data: The aircraft environmental control system (ECS) is a key subsystem of the aircraft, responsible for regulating and maintaining the temperature, humidity and air flow inside the cabin to ensure passenger comfort and flight safety. The ECS mainly includes functions such as cabin air supply and distribution, pressure control, temperature regulation and humidity management, and its main air source is the pressurized bleed air extracted from the compressor section of the aircraft engine. Faults in the ECS system may lead to cabin depressurization or temperature instability, seriously affecting flight safety. In the ECS, the pressure regulating shut-off valve (PRSOV) is one of the key components, responsible for regulating and distributing the high-temperature bleed air to different subsystems. Since the PRSOV needs to operate in a high-temperature and high-pressure environment, its internal components are vulnerable to damage. Figure 1 shows the internal components of a certain PRSOV structural model.

[0040] It should be noted that the torque motor current is controlled by the system, which drives the movement of the piston pin (pintle) to regulate the air flow into the opening chamber and the discharge orifice (leading to the external environment). The pressure change in the opening chamber further drives the piston movement, thereby adjusting the butterfly valve angle and the downstream pressure. This dynamic regulation mechanism can effectively control the air flow and pressure of the ECS. Faults in the PRSOV may lead to a decrease in the high-temperature air flow regulation efficiency or unstable pressure regulation. For example, charging or discharging faults in the torque motor may seriously affect the precise pressure control ability of the PRSOV.

[0041] Table 1 contains the description information of the PRSOV dataset with different anomaly classes ; The failure of any ECS component will directly affect the safety of the aircraft. Since aviation regulations strictly prohibit the aircraft from continuing to fly in the event of an ECS failure, it is impossible to conduct experiments in a real flight environment. Therefore, in this study, a rigorously verified PRSOV Simulink simulation model is used to generate abnormal data under different working conditions. The generation of each PRSOV sample is carried out under the guidance of valve experts, and different fault states are simulated by adjusting the inherent parameters of the PRSOV (such as the friction coefficient, the degree of charging and discharging blockage, etc.). For each sample, the adjusted pressure data is recorded at each timestamp as the training input. Each training sample contains 201 pressure-related features. The entire dataset contains a total of 8,000 training samples, covering 8 different health states. Table 1 summarizes the various fault types and their sample numbers, while Figure 2 shows an example of the collected signal samples. During the data partitioning process, normal samples are further split into a training set and a test set, while all abnormal types are regarded as test samples to evaluate the abnormal detection ability of the model.

[0042] We conducted a comparative experiment of the RPDP-AD model with other data-driven methods to evaluate its relative performance. During the experimental evaluation process, we adopted the following key metrics: · AUC (Area Under the ROC Curve): Measures the overall detection ability of the model; · AUPRC (Area Under the Precision-Recall Curve): Measures the performance of the model when there is less abnormal data; · Accuracy: Evaluates the overall classification ability of the model; · F1-score: A comprehensive metric that balances precision and recall.

[0043] During the training process, we adopted the Adam optimizer with a learning rate set to 0.001 and a batch size set to 64. All models were implemented based on the PyTorch framework and Python 3.8, and trained for 500 epochs on an RTX 3090 GPU server before being tested. In terms of data preprocessing, all data samples were normalized using Z-Score to ensure consistency across different feature dimensions. For hyperparameter selection and the MLP network structure, we followed the MLP settings in the reference literature and used the LeakyReLU activation function. The random projection matrix was sampled from a standard Gaussian distribution, and the projection dimension was set to 128. However, as mentioned earlier, other probability distributions that satisfy unit variance and certain boundedness can also be used to construct the random projection matrix. The selection of different random projection matrices will be further discussed in the following sensitivity analysis section.

[0044] In this study, we benchmarked RPDP-AD and compared it with other representative data-driven anomaly detection methods. We selected a series of excellent unsupervised machine learning and deep learning anomaly detection models as comparison objects, including RCA, DeepSVDD, GOAD, REPEN, NeuTraL, SLAD, IForest, ROD, ECOD, and KDE. All these models were trained in an unsupervised setting, using only normal signals as training data, and the test results are shown in Table 2. It can be seen from the experimental results that RPDP-AD significantly outperformed other methods in multiple key metrics, including AUC, AUPRC, Accuracy, and F1-score, demonstrating excellent anomaly detection capabilities.

[0045] Table 2 shows that RPDP-AD achieved the highest average AUC (0.990), AUPRC (0.953), accuracy (0.986), and F1-score (0.973) among all evaluation methods, indicating that this method has the best comprehensive performance in the PRSOV anomaly detection task. The high AUC value reflects that RPDP-AD can effectively distinguish normal and abnormal data under different types of anomaly patterns, and the highest AUPRC value further verifies that it can still maintain high precision and high recall rate in the class imbalance scenario. In addition, Table 2 further shows that RPDP-AD performs stably under all seven anomaly types, and achieved nearly perfect detection scores (AUC / AUPRC / Accuracy / F1 ≥ 0.990) in Type 5 (Charge & Friction) and Type 6 (Discharge & Friction), indicating that this method has strong robustness in dealing with multi-source anomaly interactions. At the same time, AUC = 0.996 in Type 1 (Charge) and F1-score = 0.979 in Type 2 (Discharge), indicating that RPDP-AD has good adaptability to both independent and mixed fault modes.

[0046] In summary, RPDP-AD always maintains excellent detection performance for all anomaly types of PRSOV. The highest average scores it achieved in terms of AUC, AUPRC, accuracy, and F1-score further verify its advantages in terms of robustness and reliability. RPDP-AD can still maintain a high detection accuracy in the complex anomaly combination scenario, while some existing data-driven methods have shown obvious limitations in the anomaly detection of specific categories. Through this comparative experiment, we can determine that RPDP-AD has excellent adaptability, generalization ability, and industrial application value in the aircraft health monitoring task.

[0047] Comparison of experimental results between Table 2 and other data-driven methods on the PRSOV dataset ; Regarding the problem of anomaly detection for the Pressure Regulating Shut-Off Valve (PRSOV) in the aircraft Environmental Control System (ECS), an efficient, unsupervised, and computationally resource-friendly intelligent diagnosis method is proposed to improve the safety and maintenance efficiency of aircraft. The PRSOV regulates and controls the airflow during flight to ensure the stability of the temperature and pressure inside the cabin. However, under the influence of long-term operation and complex environmental changes, this component may experience wear, leakage, or faults, which can affect the normal operation of the aircraft and even lead to safety accidents. Due to the strict reliability requirements of the aviation industry, ensuring the health status of the PRSOV is crucial for flight safety. Traditional threshold monitoring and rule-based anomaly detection methods have limited performance when dealing with complex data with high dimensions, non-linearity, and multiple fault modes. Existing machine learning methods heavily rely on expert-driven feature engineering and cannot fully capture the dynamic characteristics of the system. At the same time, although deep learning methods perform well in other industrial fields, their training often requires a large amount of labeled anomaly data. However, in actual aviation applications, due to the scarcity, diversity, and high labeling cost of fault data, it is difficult to obtain comprehensive labeled data. Therefore, the goal of this study is to design an unsupervised anomaly detection method that does not require anomaly data, has theoretical interpretability, and has a low computational overhead to overcome the limitations of existing methods in the PRSOV monitoring task. To this end, we propose RPDP-AD (Unsupervised PRSOV Multi-Fault Anomaly Detection Algorithm Based on Random Projection Distance Prediction), which uses the random projection theory to map high-dimensional data into a low-dimensional space and constructs a supervision signal by calculating the distance information after projection, so that the distribution pattern of normal data can be learned without relying on anomaly data. At the same time, we introduce a distribution difference measurement method to minimize the deviation between the mapping learned by the model and the random mapping, ensuring that the model can retain the structural information of the data. In addition, to adapt to the computational resource limitations in the aviation field, RPDP-AD adopts a lightweight MLP (Multi-Layer Perceptron) network, avoiding the high computational requirements of computationally intensive methods such as Transformer, thus achieving real-time and high-precision anomaly detection. More importantly, we provide a theoretical analysis based on the Johnson-Lindenstrauss theorem, proving the feasibility and effectiveness of the random projection method in the deep learning framework. Finally, this method can efficiently detect multiple fault modes at a low computational cost, not only filling the gap in the research of unsupervised anomaly detection in the PRSOV field, but also expanding the application of the random projection theory in industrial deep learning, providing a new accurate, reliable, and interpretable solution for the aircraft health monitoring system.

[0048] The embodiments of the present disclosure have been described above. The above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art in the technical field without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to technologies in the market, or to enable other ordinary skill in the art in the technical field to understand the embodiments disclosed herein.

Claims

1. An aircraft fault detection method based on random projection distance prediction, characterized in that, The method includes: S1. Collect air flow, air pressure and temperature data through the pressure regulating stop valve sensor system installed on the aircraft; S2. Preprocess the data collected in S1; S3. Construct an unsupervised anomaly detection model, including: Random projection dimensionality reduction: Use a random projection matrix to project high-dimensional input data into a low-dimensional space and preserve the relative distance relationship between data points as much as possible; According to the Johnson-Lindenstrauss theorem, ensure that the projected data still retains the distribution information of the original data; Supervision signal construction: Construct a supervision signal by calculating the distance between data points in the projection space, enabling the model to learn the patterns of normal data in an unsupervised environment; Lightweight MLP multi-layer perceptron network: Due to limited computing resources of avionics equipment, use an MLP network to replace computationally intensive models to achieve efficient inference; Distribution difference measurement: Ensure the stability and interpretability of the model's projection in the low-dimensional space by minimizing the distribution difference between the learnable mapping and the random mapping; S4. Train and optimize the unsupervised anomaly detection model; S5. Test and optimize the unsupervised anomaly detection model.

2. The aircraft fault detection method based on random projection distance prediction according to claim 1, wherein, The method also includes: S6. Aircraft actual verification and deployment, including: After laboratory testing and optimization, finally deploy the model to the actual aircraft health monitoring system to achieve real-time fault detection: On-board verification: During the flight mission, collect the pressure regulating stop valve sensor data in real time and perform online detection through the model; Compare the model detection results with the manual inspection results to verify its accuracy and stability; Embedded deployment: Due to limited computing resources of avionics equipment, after optimizing the model's computing efficiency, deploy it to the on-board computing platform or cloud monitoring system to achieve real-time analysis and remote monitoring; Real-time monitoring and maintenance optimization: Integrate this model into the aircraft health monitoring system to achieve all-weather online monitoring, provide early warnings to maintenance personnel, optimize the maintenance plan, and reduce unplanned maintenance.

3. The aircraft fault detection method based on random projection distance prediction according to claim 1, wherein The pressure regulating stop valve sensor system monitors the air flow, air pressure and temperature inside the pressure regulating stop valve sensor, and is used to identify abnormal flow fluctuations or airtightness problems; All the collected data are in time series format, and time stamps are recorded with a high-precision clock to ensure the synchronization of multi-sensor data for subsequent data fusion and analysis.

4. A method for aircraft fault detection based on random projection distance prediction according to claim 1, characterized in that, S2. Preprocess the data collected in S1, including: Denoising processing: Use a low-pass filter to remove high-frequency noise and ensure the smoothness of the vibration signal; Process the sensor data through Kalman filtering to make it more stable and reduce the influence of environmental noise; Use wavelet transform denoising to extract the main features and reduce the interference caused by sensor errors; Normalization: Use Z-score standardization or Min-Max normalization to ensure that all data are kept within the same numerical range to reduce the influence of the scale difference between features; Time series slicing: Use the sliding window method to divide the data into fixed time windows for input into the model for analysis; Data augmentation: Apply data augmentation techniques, including: time warping, adding noise, signal mixing, to improve the generalization ability of the model.

5. The aircraft fault detection method based on random projection distance prediction according to claim 1, wherein S4. Train and optimize the unsupervised anomaly detection model, including: Unsupervised training: Only use normal data to train the model to minimize the distance prediction error in the random projection space, enabling the model to learn the structural features of the normal operating state of the pressure regulating shut-off valve (PRSOV). Loss function optimization: Adopt the distribution difference metric loss to ensure that the projected low-dimensional data can still maintain the spatial structure of the high-dimensional data. Hyperparameter tuning: Conduct grid search to tune hyperparameters such as learning rate, projection dimension, number of MLP layers, etc., to improve the stability and performance of the model. Regularization: Adopt L2 regularization to prevent overfitting, enabling the model to generalize to different PRSOV operating conditions.

6. The aircraft fault detection method based on random projection distance prediction according to claim 1, wherein, S5. Test and optimize the unsupervised anomaly detection model, including: Offline testing: Test on the real PRSOV sensor dataset, calculate key metrics such as AUC, accuracy, recall, F1 score, etc., to evaluate the detection performance of the model; verify the contribution of different modules to the model performance through ablation experiments and optimize the overall architecture. Noise sensitivity analysis: Add different levels of noise to the test data to evaluate the robustness of the model in practical applications. Multi-fault mode detection: Use data of 7 known fault modes to test whether the model can effectively distinguish different fault types; fine-tune the model according to the test results to ensure its stability under various flight conditions.

Images