Vacuum equipment leakage fault diagnosis method and system based on mechanism and statistical hybrid model

Abstract

The application discloses an air dynamic test platform vacuum equipment leakage fault diagnosis method and system, aiming at solving the problems that the existing method is difficult to adapt to the complex time-varying characteristics of mixed gas, data is sparse, and lack of interpretability. The method first collects time series pressure data through a conventional pressure sensor, uses the ideal gas state equation as a physical mechanism model, deduces a leakage rate calculation equation, and extracts key characteristics of instantaneous leakage rate and cumulative leakage rate; then a BP neural network is constructed as a data-driven model, taking the leakage rate as input, and outputting normal and abnormal judgments after training with historical data. The application innovatively integrates physical mechanism and machine learning, without the need for special detection equipment, ensuring the interpretability of the diagnosis results, improving the diagnosis accuracy and reliability under small sample and variable working conditions, and realizing high-precision, low-cost air tightness online monitoring and early warning.

Description

Technical Field

[0001] This invention belongs to the field of airtightness leakage fault diagnosis technology for vacuum equipment of aerodynamic testing platforms, specifically involving leakage detection and early warning technology based on the fusion of ideal gas state equation and data-driven model. Background Technology

[0002] Vacuum equipment is an indispensable key subsystem for aerodynamic test platforms to achieve their core function of simulating real flight aerodynamic environments. Its reliability, control precision, and operational stability directly determine the testing capabilities and data quality of the entire aerodynamic test platform system. Therefore, efficient and accurate fault diagnosis of the vacuum equipment itself is a core technical aspect of ensuring the safe, efficient, and reliable operation of the aerodynamic test platform. Failures in the vacuum equipment of an aerodynamic test platform will directly lead to instability in the test flow field and inaccurate simulation parameters, resulting in invalid test data; furthermore, it may cause performance degradation or physical damage to the equipment itself, resulting in high maintenance costs and safety risks; ultimately, it will inevitably lead to test interruptions and a severe decline in the operational efficiency of the aerodynamic test platform.

[0003] Currently, existing fault prediction methods can be mainly divided into three categories: 1) Mechanism-based methods: While the results are accurate, they rely on precise system mathematical models. For complex equipment such as aerodynamic testing platforms and vacuum systems, model construction is extremely difficult, and models with fixed parameters are prone to mismatch and prediction failure due to equipment aging and changes in operating conditions. 2) Big data-driven methods: These do not rely on physical models, but their accuracy requires long-term and complete historical data for model training. Model performance is constrained by data quality, resulting in long training cycles and high computational costs. Furthermore, the generalization ability and interpretability of models for specific equipment (units) are often insufficient. 3) Expert knowledge-based methods: These methods utilize domain experience for reasoning, but their effectiveness depends entirely on the correctness of the knowledge itself and the rationality of the reasoning mechanism. They suffer from bottlenecks such as difficulty in knowledge acquisition, quantification, and updating, and exhibit poor adaptability to new faults.

[0004] While fault diagnosis methods based on mechanistic models have the advantages of clear physical meaning and rigorous theoretical foundation, they have revealed significant limitations in practical applications of aerodynamic test platform systems, especially their core vacuum / gas source systems. The core challenge lies in the significant and unbridgeable gap between the constructed theoretical model and the complex dynamics of actual aerodynamic test platform operation.

[0005] During operation, the vacuum system of an aerodynamic testing platform is often filled with a complex mixture of gases, rather than an ideal single component. The significant differences in molecular weight, specific heat capacity, viscosity, and other physical parameters among these gases result in extremely complex overall thermodynamic and flow characteristics of the system. Establishing a mechanistic model that accurately describes the dynamic behavior of multi-component, non-ideal gas mixtures requires precise perception and modeling of the concentrations and interactions of each component, which is virtually impossible in practice, leading to inherent biases in the model from its very inception.

[0006] The composition and proportion of test gases used in aerodynamic testing platforms may vary depending on the test objectives. Even in a single test, the actual composition of the gas mixture may be a time-varying variable due to factors such as leakage, backflow, or gas decomposition. Traditional static mechanistic models cannot adapt to this variation, resulting in a continuous, time-varying "model mismatch" between the model and the actual object. This mismatch directly leads to distortion in model-based residual calculations, causing false alarms or missed faults, and significantly reducing diagnostic reliability.

[0007] When a system experiences typical faults such as minor leaks, sensor drift, or decreased compressor efficiency, the fault characteristic signals are submerged in the dynamic fluctuations of the system caused by changes in the composition of the mixed gas. Mechanistic models, unable to effectively separate and distinguish between "normal component fluctuations" and "real faults," suffer a significant decrease in sensitivity to early, minor faults. Furthermore, the characteristics of different faults are easily confused, posing a significant challenge to accurately locating the fault source.

[0008] Specialized mechanistic models established for specific gas mixtures require expert re-deriving, verification, and parameter tuning should the aerodynamic testing platform's mission change (gas composition adjustment). This results in high maintenance costs, long update cycles, and a severe lack of flexibility, making it difficult to adapt to the efficient and variable operational requirements of modern aerodynamic testing platforms.

[0009] In summary, in the specific application scenario of aerodynamic testing platforms, the complexity and time-varying nature of the mixed gas composition pose serious challenges to traditional mechanistic model fault diagnosis methods in terms of model accuracy, robustness, sensitivity, and engineering practicality.

[0010] Data-driven approaches do not require in-depth domain expertise and can learn patterns from historical data alone, but they have significant drawbacks: the method is highly dependent on the quality and scale of the data, and its performance drops sharply when data is insufficient or noisy; its "black box" nature makes the decision-making process lack interpretability and makes it difficult to trace the physical mechanisms; at the same time, the model can only reflect the correlations in historical data rather than causal relationships, and its generalization ability is limited, making the prediction results unreliable when the system conditions exceed the historical range.

[0011] Therefore, there is an urgent need in this field for a new fault diagnosis scheme that can overcome the above-mentioned defects, and in particular, adapt to the complex dynamic characteristics of mixed gases. Summary of the Invention

[0012] In view of this, this invention discloses a technology related to leakage fault diagnosis of vacuum equipment in aerodynamic testing platforms. It aims to address the technical problems of existing aerodynamic testing platform vacuum equipment having limited sensor types, only providing basic parameter measurements such as pressure, temperature, and flow rate, making it difficult to comprehensively reflect the complex internal operating state of the equipment, and the difficulty in accurately quantifying leakage rates in scenarios lacking high-precision flow sensors. To this end, a hybrid diagnostic method integrating physical models and data-driven approaches, as well as a gas tightness detection method based on the ideal gas law, are proposed. The hybrid diagnostic method compensates for data sparsity by introducing a mechanistic model based on gas dynamics equations, while simultaneously constructing a lightweight neural network that uses only key feature parameters as input for fault identification. This preserves physical constraints, effectively reduces the "black box" effect, and enhances the model's adaptability to unknown operating conditions, significantly improving the accuracy and reliability of fault diagnosis under small sample sizes and varying operating conditions. This method monitors the temporal changes in gas pressure and temperature within a sealed cavity (the ideal gas law is affected by temperature, although in real-world environments, temperature is assumed to remain constant), and indirectly and accurately calculates the gas mass change based on the thermodynamic principles revealed by the ideal gas law, thereby achieving high-precision quantitative detection of the leakage rate. The specific technical solution is as follows: A method for diagnosing leakage faults in vacuum equipment based on a hybrid mechanistic and statistical model includes the following steps: Step 1) Continuously collect time-series pressure data during the operation of the vacuum equipment using pressure sensors that are conventionally configured in the equipment; Step 2) Using the ideal gas law as the physical mechanism model and combining the mass conservation relationship of the gas before and after leakage, under the constraints of constant temperature and constant volume of the sealed cavity of the vacuum equipment, the system leakage rate calculation equation is directly derived. The instantaneous leakage rate based on the pressure change rate and the cumulative leakage rate based on the pressure deviation are extracted from the collected time-series pressure data as key feature parameters. Step 3) Construct a lightweight neural network as a data-driven model, take the key feature parameters extracted in Step 2) as input, and use the trained data-driven model to realize real-time detection of airtightness leakage and fault status judgment of vacuum equipment.

[0013] Furthermore, the leakage rate calculation equation ,in, This is the absolute pressure before the leak. This is the absolute pressure after leakage. The absolute pressure of the leaked gas in the air. V represents the volume of the leaked gas, and V represents the volume of the sealed cavity of the vacuum equipment.

[0014] Furthermore, the different fault states include different leakage levels and different mixed gas composition conditions.

[0015] Furthermore, the data-driven model includes a backpropagation (bp) neural network.

[0016] Furthermore, the training process of the data-driven model uses historical normal operation data and different fault state data as training sets. The network weights are adjusted through iterative optimization to complete model training and parameter optimization, which is used to improve the model's adaptability to small sample and variable operating conditions.

[0017] A vacuum equipment leakage fault diagnosis system based on a hybrid mechanistic and statistical model includes the following modules: Data acquisition module: Equipped with a pressure sensor, used to continuously acquire and transmit time-series pressure data of the vacuum equipment; Physical mechanism modeling and feature extraction module: used to construct physical mechanism models based on the ideal gas law, derive leakage rate calculation equations by combining the gas mass conservation relationship, and extract key feature parameters of instantaneous leakage rate and cumulative leakage rate; Data-driven modeling and recognition module: It has a built-in lightweight neural network, which is used to identify whether a vacuum device has leaked by taking key feature parameters as input, and completes model training and parameter optimization based on historical data.

[0018] Beneficial effects 1. Combining adaptability and interpretability, overcoming the limitations of traditional methods: By integrating the physical mechanism of the ideal gas law with a data-driven model, it not only eliminates the excessive reliance on precise system parameters in traditional physical modeling and can adapt to the complex and time-varying special working conditions of mixed gas components in aerodynamic testing platforms, solving the core problem of "model mismatch" in traditional mechanistic models, but also makes up for the "black box" defect of pure data-driven methods, giving diagnostic results clear physical interpretability and avoiding unreliable decisions caused by relying solely on data correlation.

[0019] 2. Low cost and high precision, lowering the threshold for engineering implementation: No need to configure expensive special detection equipment (such as high-precision flow sensors). Data can be collected by pressure sensors conventionally configured in vacuum equipment. Based on physical mechanisms, high-precision quantitative detection of leakage rate can be achieved, which greatly reduces the implementation cost and structural complexity of the system. It solves the problem that existing airtightness detection technology is difficult to accurately quantify leakage rate in the absence of special equipment.

[0020] 3. Excellent sensitivity and reliability, ensuring stable equipment operation: By using feature extraction supported by physical laws, the model's adaptability to different test tasks and fluctuating operating parameters is enhanced. At the same time, with the help of a dual verification mechanism of mechanism and data, it effectively distinguishes between "normal component fluctuations" and "real faults", significantly improving the detection sensitivity of early micro-leakage, reducing the false alarm rate and missed alarm rate of faults, solving the problems of low sensitivity and insufficient diagnostic reliability of traditional methods for micro-faults, and ensuring the continuous and safe operation of the aerodynamic test platform. Attached Figure Description

[0021] Figure 1 Flowchart of a method for diagnosing leakage faults in vacuum equipment based on a hybrid model of mechanism and statistics. Detailed Implementation

[0022] The specific implementation process of this method is as follows: First, time-series pressure data during the operation of the vacuum equipment is continuously collected using pressure sensors. Using the ideal gas dynamic equation (PV=nRT) as the physical mechanism model, this model takes the collected time-series pressure values ​​as input and continuously calculates and outputs the leakage rate between the current and previous moments (exhibiting a continuous time-series output characteristic). Based on this leakage rate, the system leakage rate calculation equation is further derived, extracting two types of key feature parameters: instantaneous leakage rate based on pressure change rate and cumulative leakage rate based on pressure deviation. Subsequently, a BP neural network is constructed as a data-driven model suitable for leakage detection. The instantaneous and cumulative leakage rates are used as model inputs, and historical normal operation data of the vacuum equipment and various leakage fault data are used as the training set. The model parameters are tuned through iterative optimization. After training, the model outputs the equipment operating status (normal / abnormal) through a threshold judgment mechanism, ultimately achieving real-time detection and intelligent decision-making regarding the leakage status of the vacuum equipment.

[0023] This innovative method organically integrates traditional physical mechanisms with modern machine learning techniques. It not only relies on the physical essence of the ideal gas dynamic equation to ensure the strong interpretability of the model, but also fully leverages the advantages of BP neural networks in complex fault mode recognition. Furthermore, it eliminates the need for additional dedicated detection equipment, enabling high-precision, low-cost online monitoring and early warning of airtightness using only conventional pressure sensors, thus precisely meeting the practical application needs of vacuum equipment in aerodynamic testing platforms.

[0024] The following section will provide a detailed derivation and explanation of the measurement principle.

[0025] According to the ideal gas law (1) Conclusion: (2) In the formula, P is the absolute pressure of the gas, V is the gas volume, m is the gas mass, M is the molar mass of the gas, R is the gas constant, and T is the absolute temperature.

[0026] Assume the absolute pressure inside the vacuum equipment before leakage is After the leak, the absolute pressure inside the vacuum equipment was The absolute pressure of the gas leaking into the air is ,visible (3) From equation (2), we can see that (4) Assuming the temperature remains constant, and the volume of the vacuum equipment remains unchanged before and after leakage, we can conclude that: (5) After feature extraction is performed using a mechanistic model, a neural network is then used for pattern recognition to achieve airtightness leak detection.

Claims

1. A method for diagnosing leakage faults in vacuum equipment based on a hybrid mechanistic and statistical model, characterized in that: Step 1) Continuously collect time-series pressure data during the operation of the vacuum equipment using pressure sensors that are conventionally configured in the equipment; Step 2) Using the ideal gas law as the physical mechanism model and combining the mass conservation relationship of the gas before and after leakage, under the constraints of constant temperature and constant volume of the sealed cavity of the vacuum equipment, the system leakage rate calculation equation is directly derived. The instantaneous leakage rate based on the pressure change rate and the cumulative leakage rate based on the pressure deviation are extracted from the collected time-series pressure data as key feature parameters. Step 3) Construct a lightweight neural network as a data-driven model, take the key feature parameters extracted in Step 2) as input, and use the trained data-driven model to realize real-time detection of airtightness leakage and fault status judgment of vacuum equipment.

2. The method according to claim 1, characterized in that: The leakage rate calculation equation ,in, This is the absolute pressure before the leak. This is the absolute pressure after leakage. The absolute pressure of the leaked gas in the air. V represents the volume of the leaked gas, and V represents the volume of the sealed cavity of the vacuum equipment.

3. The method according to claim 1, characterized in that: The different fault states include different leakage levels and different mixed gas composition conditions.

4. The method according to claim 1, characterized in that: The data-driven model includes a bp neural network.

5. The method according to any one of claims 1-4, characterized in that: The training process of the data-driven model uses historical normal operation data and data of different fault states as training sets. The network weights are adjusted through iterative optimization to complete model training and parameter optimization, which is used to improve the model's adaptability to small sample and variable operating conditions.

6. A vacuum equipment leakage fault diagnosis system based on a hybrid mechanistic and statistical model, characterized in that: Includes the following modules, Data acquisition module: Equipped with a pressure sensor, used to continuously acquire and transmit time-series pressure data of the vacuum equipment; Physical mechanism modeling and feature extraction module: used to construct physical mechanism models based on the ideal gas law, derive leakage rate calculation equations by combining the gas mass conservation relationship, and extract key feature parameters of instantaneous leakage rate and cumulative leakage rate; Data-driven modeling and recognition module: It has a built-in lightweight neural network, which is used to identify whether a vacuum device has leaked by taking key feature parameters as input, and completes model training and parameter optimization based on historical data.

Images