A method and system for analyzing the aerodynamic stability of axial compressors based on deterministic learning

Abstract

This invention belongs to the technical field of compressor aerodynamic stability analysis. It proposes a deterministic learning-based method and system for analyzing the aerodynamic stability of axial compressors. Using pulsating pressure data during the aerodynamic instability development process of an axial compressor, a deterministic learning identification algorithm based on radial basis function neural networks is used to model the nonlinear dynamics of the compressor's internal flow field, thereby obtaining dynamic diagrams of normal and instability precursor states. Dynamic indices are extracted from these diagrams to construct dynamic feature vectors describing the compressor's aerodynamic stability. Based on the dynamic feature vectors corresponding to different operating states, the dynamic classification boundary between normal and instability precursor states is determined. The distance between the dynamic feature vector of the compressor data under test and the dynamic classification boundary is calculated to determine the compressor's operating state. This invention enables early warning of aerodynamic instability and has good interpretability and engineering application value.

Description

Technical Field

[0001] This invention belongs to the technical field of compressor aerodynamic stability analysis, and particularly relates to a method and system for aerodynamic stability analysis of axial flow compressors based on deterministic learning. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Predictive and Health Management (PHM) for aero-engines is a key technology for reducing engine maintenance costs and ensuring safe and stable operation. It primarily utilizes real-time information on aero-engine vibration, gas path, lubricating oil, and lifespan to achieve functions such as condition monitoring, fault diagnosis, trend analysis, and lifespan management. It provides early warnings of situations that may affect safe operation, thereby improving the safety, reliability, and maintainability of aero-engines.

[0004] Aerodynamic stability characterizes the compressor's ability to operate stably and is a crucial component of aero-engine dynamic performance (PHM). Aerodynamic stability is generally quantified using surge margin (or stability margin), which refers to the margin between the compressor component's operating point and the surge boundary. During aero-engine operation, insufficient surge margin can lead to compressor instability (including rotating stall and surge). These instabilities can cause reduced engine thrust, structural damage, and even in-flight engine shutdown, posing a significant threat to flight safety. Therefore, a series of online engine surge detection devices have been developed both domestically and internationally, most of which determine whether engine surge has occurred based on the relative change in pressure fluctuation amplitude.

[0005] Unlike surge detection and early warning, compressor aerodynamic stability monitoring refers to real-time monitoring of the stability margin. In aero-engine PHM systems, the degree of degradation in compressor aerodynamic stability under different conditions can be provided: timely warnings can be given when the stability margin approaches or reaches 0, and predictions of stability margin changes can also be given. However, limited by the development of compressor aerodynamic instability theory, existing aero-engine PHM technology cannot monitor changes in the stability margin. The fundamental reason is that most existing aerodynamic stability monitoring methods are mainly based on time-domain or frequency-domain processing techniques for flow field data, lacking dynamic modeling and analysis of the nonlinear evolution of the flow field during aerodynamic instability. Therefore, the extracted data features are difficult to directly correlate with the stability margin. Summary of the Invention

[0006] To overcome the shortcomings of the prior art, this invention provides a method and system for aerodynamic stability analysis of axial compressors based on deterministic learning. It constructs an aerodynamic stability dynamic diagram based on complex measurement data, which can intuitively reflect the evolution process of aerodynamic stability from a dynamic perspective. It can also realize early warning of aerodynamic instability by rapidly detecting the instability precursor operating state, and has good interpretability and engineering application value.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for analyzing the aerodynamic stability of an axial compressor based on deterministic learning, comprising: For the pulsating pressure data in the aerodynamic instability development process of axial compressor, a deterministic learning identification algorithm based on radial basis function neural network is used to perform neural network modeling of the nonlinear dynamic system of the internal flow field of the compressor. Based on the converged neural network weights, a constant neural network expression of the compressor dynamics is constructed, thereby obtaining the dynamic diagrams of normal operation state and instability precursor state. Dynamic indices are extracted from the aerodynamic stability dynamic diagram, and dynamic feature vectors describing the aerodynamic stability of the compressor are constructed. Based on the dynamic feature vectors corresponding to different operating states, the dynamic classification boundary between the normal state and the instability precursor state is determined. Calculate the distance between the dynamic feature vector of the compressor under test and the dynamic classification boundary, and determine the operating state of the compressor based on the calculation results.

[0008] Secondly, the present invention provides a method for analyzing the aerodynamic stability of an axial compressor based on deterministic learning, including: The modeling module is configured to: for the pulsating pressure data during the aerodynamic instability development process of the axial compressor, use a deterministic learning identification algorithm based on radial basis function neural network to perform neural network modeling of the nonlinear dynamic system of the internal flow field of the compressor, construct a constant neural network expression of the compressor dynamics based on the converged neural network weights, and then obtain the dynamic diagrams of the normal operation state and the instability precursor state. The dynamic classification boundary module is configured to: extract dynamic indices from the aerodynamic stability dynamic diagram, construct a dynamic feature vector describing the aerodynamic stability of the compressor, and determine the dynamic classification boundary between the normal state and the instability precursor state based on the dynamic feature vectors corresponding to different operating states. The stability analysis module is configured to calculate the distance between the dynamic feature vector of the compressor under test and the dynamic classification boundary, and determine the operating state of the compressor based on the calculation results.

[0009] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0010] Fourthly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in the first aspect.

[0011] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0012] The above one or more technical solutions have the following beneficial effects: In this invention, for pulsating pressure data during the development of compressor aerodynamic instability, a deterministic learning identification algorithm based on radial basis function neural networks is used to model the nonlinear dynamic system of the compressor's internal flow field using a neural network. Based on the converged neural network weights, a constant neural network expression of the compressor dynamics is constructed, thereby obtaining dynamic diagrams of normal operation and instability precursor states. This approach is more feasible than establishing a parameterized model and can be used for aerodynamic stability monitoring under actual complex service conditions. The distance between the feature vector of the tested compressor and the classification boundary is used to characterize the trend of stability margin changes. By real-time detection of precursor patterns between normal and instability states, progressive trend prediction and early warning of aerodynamic instability can be achieved, thus providing more sufficient response time for the control system and contributing to the digital and intelligent implementation of aero-engine aerodynamic stability monitoring.

[0013] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0014] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0015] Figures 1(a) and 1(b) show the spatial trajectories of normal samples and instability precursor samples in the embodiments of the present invention; Figure 2 This is a graph showing the convergence curve of the neural network weights in an embodiment of the present invention. Figures 3(a) and 3(b) show the compressor dynamics imaging corresponding to the normal mode and the instability precursor mode in the embodiments of the present invention; Figure 4 This is a schematic diagram of the classification boundary and region division based on dynamic indicators in an embodiment of the present invention; Figure 5 This is an example diagram of the online monitoring process and early warning in an embodiment of the present invention. Detailed Implementation

[0016] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0017] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0018] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0019] Example 1 This embodiment discloses a method for analyzing the aerodynamic stability of an axial compressor based on deterministic learning, including: For the pulsating pressure data in the aerodynamic instability development process of axial compressor, a deterministic learning identification algorithm based on radial basis function neural network is used to perform neural network modeling of the nonlinear dynamic system of the internal flow field of the compressor. Based on the converged neural network weights, a constant neural network expression of the compressor dynamics is constructed, thereby obtaining the dynamic diagrams of normal operation state and instability precursor state. Dynamic indices are extracted from the aerodynamic stability dynamic diagram to construct a dynamic feature vector describing the aerodynamic stability of the compressor; Based on the dynamic feature vectors corresponding to different operating states, determine the dynamic classification boundary between the normal state and the instability precursor state; Calculate the distance between the dynamic feature vector of the compressor under test and the dynamic classification boundary, and determine the operating state of the compressor based on the calculation results.

[0020] This embodiment utilizes a deterministic learning algorithm to model the dynamics of the internal flow field data of the compressor, and visualizes the modeling results into an aerodynamic stability dynamic diagram. Dynamic feature vectors reflecting the compressor's instability evolution are extracted from the dynamic diagram. A classification boundary is constructed within the dynamic feature space between normal operation and instability precursor states. The distance between the measured data feature vector and the classification boundary is calculated in real time and used as a dynamic metric for compressor stability margin, for aerodynamic stability monitoring and instability early warning. This method visualizes complex flow field measurement data into an aerodynamic stability dynamic diagram, intuitively reflecting the evolution of aerodynamic stability from a dynamic perspective. It also enables early warning of aerodynamic instability through rapid detection of instability precursor states, demonstrating good interpretability and engineering application value. It is expected to provide an effective real-time monitoring method for the safe and stable operation of compressor systems.

[0021] Rotating stall and surge of the axial compressor in a turbofan engine are essentially complex dynamic phenomena arising from an infinite-dimensional distributed parameter system described by partial differential equations. The model shown below provides a finite-dimensional approximation:

[0022] in, This indicates that the compressor is evenly distributed around its circumference. The flow coefficient state vector corresponding to each sensor This represents the average pressure rise. Parameters , and These represent the compressor geometric constants, throttling parameters, and Greitzer stability parameters, respectively. Nonlinear functions. For compressor characteristic functions, , , and It is a coefficient matrix.

[0023] By appropriately selecting system parameters and compressor characteristic functions, the model can simulate the evolution of various aerodynamic instabilities and maintain good consistency with experimental data. In the first... At each measurement point The intrinsic dynamics can be rewritten as:

[0024] in,

[0025] Indicates by the first Each measurement point and its A state vector consisting of the flow coefficients at adjacent points; To remove Subvectors after; express The main dynamics; Indicates being and Secondary dynamics of influence. and These are system parameters. To model uncertainties, including external disturbances and modeling errors.

[0026] In actual experiments, the measured discrete-time data This can be approximated using the following Euler discrete model:

[0027] Where the sampling period T>0 is a sufficiently small constant.

[0028] Step 1: Obtain the pulsating pressure data of the axial compressor.

[0029] First, a dataset of aerodynamic instability pulsating pressures is established, encompassing different operating states (normal, instability precursors, and instability). This embodiment uses an axial compressor test bench as an example, arranging three Kulite sensors circumferentially around the compressor to capture pulsating pressure signals during the aerodynamic instability development process (i.e.,...). By gradually closing the throttle valve, the complete development process of a real compressor from normal operation to unstable operation is simulated. Pulsating pressure signals of the dynamic development process of aerodynamic stability are collected in parallel and synchronously to construct an aerodynamic stability dataset for axial compressors covering typical complex service conditions. Normal operation samples and instability precursor samples are extracted from the measured signals, and state-space trajectories are constructed using samples processed by low-pass filtering. Figures 1(a) and 1(b) show the three-dimensional spatial trajectories formed by normal state samples and precursor state samples in a certain experiment.

[0030] Step 2: For the three-dimensional spatial trajectory formed by the pulsating pressure data during the development of aerodynamic instability in the axial compressor, a deterministic learning identification algorithm based on radial basis function neural network is used to perform neural network modeling of the nonlinear dynamic system of the internal flow field of the compressor. Based on the converged neural network weights, a constant neural network expression of the compressor dynamics is constructed, thereby obtaining the dynamic diagrams of normal operation state and instability precursor state.

[0031] For the aerodynamic stability dataset in step 1, a deterministic learning and identification algorithm based on radial basis function neural network and Lyapunov stability is designed to model the nonlinear dynamic system of the compressor internal flow field using a neural network. Based on the converged neural network weights, a constant neural network expression of the compressor dynamics is calculated, as shown in Figures 3(a)-3(b). Dynamic diagrams for normal operation and instability precursor states are generated through three-dimensional visualization and stored in the dynamic knowledge base. These diagrams contain the state trajectory information of the compressor signal and the dynamic information along the state trajectory, characterizing the dynamic features of the compressor aerodynamic stability in space and time. This is a holographic expression of the aerodynamic stability development process.

[0032] The dynamic system is a nonlinear differential equation that describes the evolution of compressor pulsating pressure data over time, representing the physical essence of aerodynamic instability phenomena such as compressor rotational stall and surge.

[0033] This step utilizes a radial basis function neural network to perform dynamic modeling on the pulsating pressure data obtained in step 1. First, in the trajectory... Within the distributed area, the following RBF neural network nodes are uniformly arranged:

[0034] in, Represents a neuron node. q For neuron indexing, Q For the number of neurons, n This represents a positive integer value. In this embodiment, it takes... n =1, which means modeling the compressor dynamics in three-dimensional space.

[0035] The following dynamics identifier is used to learn the first... Dominant dynamics at each measurement point :

[0036] in, For the state of the identifier, For the gain of the identifier, For the estimated RBF weight vector, T The sampling period is The regression vector can be represented by the following formula:

[0037] neuron nodes As the center, the expression for the q-th element is:

[0038] in, Indicates the width of the receptive field; Indicates by the first Each measurement point and its The state vector is composed of the flow coefficients at adjacent points.

[0039] The initial weights of the neural network are set to zero. For RBF neural network nodes in the neighborhood of the spatial trajectory formed by the compressor pulsating pressure signal, the corresponding sub-vectors satisfy the continuous excitation condition and the exponent converges to the optimal value during the weight adjustment process. For neural network nodes far from the trajectory, the weights remain at zero.

[0040] weight vector Update as follows:

[0041] in, This indicates the learning gain.

[0042] Based on the above design, compressor dynamics This can be represented by a constant RBF neural network. According to deterministic learning theory, when the partial continuous excitation condition is met, the neural network weights near the system trajectory will converge exponentially to a small neighborhood of the ideal weights. This means that the neural network accurately captures the local flow field dynamics characteristics of the compressor under this operating condition. Figure 2 This demonstrates the convergence of the neural network weights during the learning process.

[0043] Calculating the weights of a constant neural network :

[0044] in, , This is the index of the weight iteration step. Using the obtained constant neural network... To approximate the nonlinear dynamics of the compressor, a library of aerodynamic instability modes is constructed:

[0045] in, Represents normal mode. This represents a precursory instability mode. Combining the state trajectory formed by the pulsating pressure signal, the dynamic modeling results are visualized in three dimensions, and the generated compressor aerodynamic stability dynamic diagram is denoted as... It not only contains the state trajectory information (spatial location) of the signal, but also the dynamic information distributed along the trajectory, thus characterizing the holographic features of the compressor's aerodynamic stability in both space and time.

[0046] As shown in Figure 1, the state trajectory is the filtered result of the pulsating pressure at three measuring points plotted in a three-dimensional coordinate system. The dynamics diagram is the deterministic learning modeling result of the dynamics corresponding to the pulsating pressure signals at the three measuring points plotted in a three-dimensional coordinate system.

[0047] Step 3: Extract dynamic indices from the aerodynamic stability dynamic diagram and construct a dynamic feature vector describing the aerodynamic stability of the compressor; determine the dynamic classification boundary between the normal state and the instability precursor state based on the dynamic feature vectors corresponding to different operating states.

[0048] For the aerodynamic stability dynamics diagram in step 2, the estimator residuals, temporal dispersion, and spatial dispersion are extracted to construct feature vectors describing the evolution of compressor aerodynamic stability. Using classification models such as support vector machines, within the dynamic feature space, the feature vectors are used as input to the classifier model, and the actual compressor operating state is used as the model output to train and obtain the classification boundary between normal and instability precursor states.

[0049] To quantify the evolution of aerodynamic stability, three key dynamic indices are extracted as feature vectors based on the modeling results from step 2. : 1) Estimator residuals .

[0050] Test data It can be approximated by the following Euler system:

[0051] Among them, superscript Used to identify test data, all other symbols are consistent with the above content.

[0052] Using dynamics library The constant RBF approximation model in Construct the following dynamic estimator :

[0053] in, For the estimator state, These are the estimator parameters.

[0054] By comparing the two equations above, the following residual system can be obtained. :

[0055] Among them, China

[0056] Use a length of A sliding window is used to calculate the estimator residuals for each residual system:

[0057] in, k This represents the index of the data point at the current time. The Estimator residual index (ERI) is defined as the difference between the minimum residuals associated with the normal mode and the instability precursor mode:

[0058] ERI is used to characterize actual compressor dynamics functions and model libraries. The degree of matching between different training modes. Specifically, when the compressor is in normal operating condition, its dynamic function is more consistent with the normal mode in the model library, thus the residuals of the normal estimator are... Smaller, while the residuals of the precursor estimator Relatively large, at this time As the compressor gradually approaches the precursory instability state, the system dynamics begin to deviate from the normal mode and evolve towards the precursory mode, manifested as the residuals of the normal estimator. Increase, precursor estimator residuals The value decreases, causing ERI to gradually decrease and become negative.

[0059] 2) Spatial dispersion.

[0060] To measure the compressor dynamics diagram To assess the spatial dispersion, the Lyapunov exponent from nonlinear time series analysis is introduced. Consider the first data point of the kinetic plot. and another located in Points outside the neighborhood (Right now ,in Propagate these two points forward. After the first step, the first growth rate indicator is defined as:

[0061] Similarly, the first The growth rate indicator is defined as:

[0062] in , Let be the total length of the dynamic graph, and Located at point of Step outside the neighboring domain.

[0063] The spatial heterogeneity index (SHI) is defined as follows:

[0064] The Lyapunov index (SHI) is used to measure the spatial dispersion (i.e., chaos) of the compressor dynamics diagram. Under normal operating conditions, the state trajectory formed by the three measurement points is spherical, with insignificant differences in dynamic characteristics across dimensions. However, in the instability precursor stage, the state trajectory exhibits a clear directional structure, reflecting the evolution of compressor dynamics from random fluctuations to ordered instability. Therefore, the Lyapunov index-based measurement method can quantify the changes in the spatial distribution structure of the dynamics during instability development.

[0065] 3) Time dispersion.

[0066] To describe the dynamic diagram Evolutionary characteristics at the frequency domain level are defined by synthesizing the spectral characteristics of each dimension in the dynamic diagram, and the temporal heterogeneity index (THI) is defined.

[0067] Specifically, for dynamic diagrams Let any dimension be The maximum spectral amplitude is represented by a family of negative exponential functions to approximate the frequency domain information. :

[0068] in, Indicates frequency, This represents the decay control factor of the negative exponential function. This represents the interval coefficient of the family of functions.

[0069] No. The quantization parameters for each dimension are defined as follows:

[0070] The time heterogeneity index is calculated by aggregating the quantitative parameters across all dimensions as follows:

[0071] Compared to normal operation, aerodynamic instability signals are typically dominated by low-frequency components, with their energy concentrated in a specific low-frequency range. These characteristics are also reflected in the dynamic diagram. The spectral distribution of each component. Through comprehensive analysis. By analyzing the spectral characteristics of each component, THI can characterize the trend of compressor dynamics evolving towards an unstable state from a frequency domain perspective.

[0072] Finally, the three indicators are combined into the feature vector of the dynamic graph. .exist Figure 4 In the three-dimensional feature space, each data point represents a feature vector at a specific time. This embodiment uses a linear support vector machine (Linear SVM) to train the feature vectors, and uses the feature vectors... As input, normal and unstable precursor states are used as output labels. Through training with a large amount of experimental data, the optimal classification hyperplane is determined as the classification boundary, and its linear equation is expressed as:

[0073] in, For the normal vector weights, These are bias terms, both obtained through support vector machine training. For example... Figure 4 As shown, the compressor's normal operating state and instability precursor states are clearly separated by this classification boundary. In subsequent calculations and monitoring, the real-time extracted feature vectors are substituted into the decision function. In this way, the current operating status of the compressor can be directly determined based on the sign of the calculation result.

[0074] Step 4: Calculate the distance between the dynamic feature vector of the compressor data under test and the dynamic classification boundary, and determine the operating state of the compressor based on the calculation results.

[0075] For the measured data, the distance between its dynamic feature vector and the classification boundary is calculated in real time, and this distance is used as a dynamic metric for the compressor stability margin. When the dynamic metric is positive, it means that the compressor is operating normally. When the dynamic metric gradually decreases, it indicates that the engine aerodynamic stability is gradually deteriorating. When the dynamic metric decreases to 0, a warning signal is given, indicating that the compressor is in a state of instability precursor. The key feature is that the change in the dynamic metric reflects the evolution characteristics of the measured compressor stability margin, and can give an instability warning in a timely manner when the dynamic margin reaches 0, thus achieving the goal of monitoring the compressor aerodynamic stability.

[0076] For the flow field data of the compressor under test, calculate the current dynamic characteristic vector in real time. Obtain the Euclidean distance between the classification boundary obtained in step 3 and the classification boundary. This is used as a dynamic quantitative indicator of compressor stability margin for online monitoring of aerodynamic stability and early warning of instability. Specifically: When the dynamic eigenvector When far from the safety boundary and located in the healthy zone (i.e. This indicates that the system is in a healthy operating state. when As we approach the classification boundary (i.e.) The decrease indicates that the aerodynamic stability of the compressor is gradually deteriorating; when When crossing the classification boundary into the instability precursor region (i.e. A change from 0 to a negative value indicates that the compressor is in a state of impending instability. At this time, an early warning signal is given, which can provide early warning of subsequent aerodynamic instability.

[0077] like Figure 5 As shown, this method can quickly detect precursor patterns between normal and unstable states and provide an early warning signal 0.707s before instability occurs, thus allowing the control system more sufficient response time.

[0078] The aerodynamic stability imaging model established in this embodiment is more feasible than establishing a parameterized model. It utilizes pulsating pressure data during the development of compressor aerodynamic instability to model and express the holographic dynamic characteristics of the compressor. Furthermore, the dynamic knowledge base can be expanded through continuous learning of new data, thus enabling it to handle aerodynamic stability monitoring tasks under complex actual service conditions. In contrast, axial compressor aerodynamic instability is essentially a complex dynamic phenomenon generated by an infinite-dimensional distributed parameter system described by partial differential equations. Establishing a parameterized aerodynamic stability model is extremely difficult. Even if a model can be established, its physical assumptions and simplified conditions cannot guarantee a complete match with actual operating conditions, making it difficult to directly use for aerodynamic stability monitoring under complex actual service conditions.

[0079] Unlike existing time-domain or frequency-domain processing techniques based on flow field data, the compressor aerodynamic stability monitoring method based on dynamic imaging utilizes neural networks to model and image the nonlinear dynamic system of aerodynamic stability modes such as normal operation and instability precursors. The imaging results are a visualization of the nonlinear evolution dynamics of the flow field during aerodynamic instability, including not only the state trajectory information formed by signals from different measuring points in the compressor system, but also the dynamic information along the state trajectory. It characterizes the dynamic features of compressor aerodynamic stability in both space and time, providing a holographic representation of the aerodynamic stability development process.

[0080] Most existing aerodynamic instability detection devices can only provide detection results after instability occurs. However, this embodiment constructs a dynamic metric for stability margin and uses the distance between the feature vector of the tested compressor and the classification boundary to characterize the trend of stability margin change. By real-time detection of precursor patterns between normal and unstable states, it can achieve progressive trend prediction and early warning of aerodynamic instability, thereby providing more sufficient response time for the control system.

[0081] Within the spatial trajectory neighborhood formed by the compressor flow field pulsating pressure data, these neural network models accurately approximate the nonlinear dynamic functions characterizing the dynamic evolution of flow field state variables. The dynamic features extracted from the imaging results can be correlated with the stability margin, thus improving the reliability of aerodynamic stability monitoring and early warning. Compared to traditional black-box models based on neural networks or deep learning, the neural network model in this embodiment has explicit physical meaning.

[0082] This embodiment utilizes an RBF neural network within a sampled data framework to build a library of aerodynamic instability patterns containing rich dynamic information. It automatically delineates safe zones using classification boundaries, eliminating the need for manual experience to set complex alarm thresholds. By calculating the distance between dynamic indicators and safe boundaries, the system can make rapid decisions in a parallel, real-time manner, facilitating the digital and intelligent implementation of aero-engine aerodynamic stability monitoring.

[0083] This embodiment transforms complex flow field data into a visualized dynamic diagram by determining the learning theory, and establishes a dynamic quantification of the stability margin of the axial compressor and an early warning of aerodynamic instability by utilizing the distance relationship between dynamic indicators and classification boundaries.

[0084] Example 2 The purpose of this embodiment is to provide a method for analyzing the aerodynamic stability of axial compressors based on deterministic learning, including: The modeling module is configured to: use the pulsating pressure data during the aerodynamic instability development process of the axial compressor, based on the deterministic learning and identification algorithm of the radial basis function neural network, to perform neural network modeling of the nonlinear dynamic system of the internal flow field of the compressor, and construct a constant neural network expression of the compressor dynamics based on the converged neural network weights, thereby obtaining the dynamic diagrams of the normal operation state and the instability precursor state; The dynamic classification boundary module is configured to: extract dynamic indices from the aerodynamic stability dynamic diagram, construct a dynamic feature vector describing the aerodynamic stability of the compressor, and determine the dynamic classification boundary between the normal state and the instability precursor state based on the dynamic feature vectors corresponding to different operating states. The stability analysis module is configured to calculate the distance between the dynamic feature vector of the compressor under test and the dynamic classification boundary, and determine the operating state of the compressor based on the calculation results. In further embodiments, the following is also provided: An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When executed by the processor, the computer instructions perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0085] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0086] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0087] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in Embodiment 1.

[0088] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0089] A computer program product includes a computer program that, when executed by a processor, implements the method described in Embodiment 1.

[0090] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.

[0091] The computer program code used to implement the methods of the present invention may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0092] In the context of this invention, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0093] Those skilled in the art will recognize that the units and algorithm steps described in conjunction with the embodiments herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0094] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for analyzing the aerodynamic stability of an axial compressor based on deterministic learning, characterized in that, include: For the pulsating pressure data during the aerodynamic instability development process of an axial compressor, a deterministic learning identification algorithm based on radial basis function neural network is used to model the nonlinear dynamic system of the internal flow field of the compressor using neural network. Based on the converged neural network weights, a constant neural network expression of the compressor dynamics is constructed, thereby obtaining the dynamic diagrams of normal operation state and instability precursor state. Dynamic indices are extracted from the aerodynamic stability dynamic diagram, and dynamic feature vectors describing the aerodynamic stability of the compressor are constructed. Based on the dynamic feature vectors corresponding to different operating states, the dynamic classification boundary between the normal state and the instability precursor state is determined. Calculate the distance between the dynamic feature vector of the compressor under test and the dynamic classification boundary, and determine the operating state of the compressor based on the calculation results.

2. The aerodynamic stability analysis method for axial compressors based on deterministic learning as described in claim 1, characterized in that, For pulsating pressure data during the aerodynamic instability development of an axial compressor, a deterministic learning identification algorithm based on radial basis function neural networks is used to model the nonlinear dynamics of the internal flow field of the compressor using neural networks. A constant neural network representation of the compressor dynamics is then constructed based on the converged neural network weights. Specifically: For RBF neural network nodes within the neighborhood of the spatial trajectory formed by the compressor pulsating pressure data, the corresponding sub-vectors satisfy the continuous excitation condition, and the corresponding neural network weight exponent converges to the optimal value; RBF neural network nodes far from the spatial trajectory are not excited, and the corresponding neural network weights remain zero.

3. The aerodynamic stability analysis method for axial compressors based on deterministic learning as described in claim 1, characterized in that, The estimator residual, temporal dispersion, and spatial dispersion are extracted from the aerodynamic stability dynamics diagram to construct a dynamic feature vector describing the aerodynamic stability of the compressor. The estimator residual is the difference between the minimum residuals associated with the normal mode and the instability precursor mode. The spatial dispersion is used to measure the degree of spatial dispersion of the compressor dynamics diagram. The temporal dispersion is determined by comprehensively analyzing the spectral characteristics of each dimension of the dynamics diagram.

4. The aerodynamic stability analysis method for axial compressors based on deterministic learning as described in claim 3, characterized in that, Consider the first data point on the compressor dynamics diagram, and another data point located at... Step on points outside the neighborhood, and propagate the two points forward. After step, calculate the first step. A growth rate index; the spatial dispersion is determined based on the sum of the calculated growth rate indices; wherein, ; For any dimension of the compressor dynamics diagram, let To represent the maximum spectral amplitude, a family of negative exponential functions is used to approximate the frequency domain information to determine the first... The quantization parameters of all dimensions are aggregated to obtain the time discreteness.

5. The aerodynamic stability analysis method for axial compressors based on deterministic learning as described in claim 3, characterized in that, Based on the dynamic feature vectors corresponding to different operating states, the dynamic classification boundary between the normal state and the instability precursor state is determined. Specifically, the dynamic feature vectors describing the aerodynamic stability of the compressor are used as input data, and the corresponding normal state or instability precursor state is used as output labels. The support vector machine model is trained, and the optimal classification hyperplane is solved as the dynamic classification boundary.

6. The aerodynamic stability analysis method for axial compressors based on deterministic learning as described in claim 1, characterized in that, Calculate the distance between the dynamic feature vector of the compressor under test and the dynamic classification boundary, and determine the operating state of the compressor based on the calculation results, specifically: When the distance between the dynamic feature vector of the compressor under test and the dynamic classification boundary is greater than zero, it indicates that the compressor is in a healthy operating state. When the distance between the dynamic feature vector of the compressor under test and the dynamic classification boundary decreases, it indicates that the aerodynamic stability of the compressor gradually deteriorates. When the distance between the dynamic feature vector of the compressor under test and the dynamic classification boundary changes from zero to a negative value, it indicates that the compressor is in a state of impending instability.

7. A method for analyzing the aerodynamic stability of an axial compressor based on deterministic learning, characterized in that, include: The modeling module is configured to: use the pulsating pressure data during the aerodynamic instability development process of the axial compressor, based on the deterministic learning and identification algorithm of the radial basis function neural network, to perform neural network modeling of the nonlinear dynamic system of the internal flow field of the compressor, and construct a constant neural network expression of the compressor dynamics based on the converged neural network weights, thereby obtaining the dynamic diagrams of the normal operation state and the instability precursor state; The dynamic classification boundary module is configured to: extract dynamic indices from the aerodynamic stability dynamic diagram, construct a dynamic feature vector describing the aerodynamic stability of the compressor, and determine the dynamic classification boundary between the normal state and the instability precursor state based on the dynamic feature vectors corresponding to different operating states. The stability analysis module is configured to calculate the distance between the dynamic feature vector of the compressor under test and the dynamic classification boundary, and determine the operating state of the compressor based on the calculation results.

8. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the method described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the method described in any one of claims 1-6.

Images