Propeller blade flutter identification model training method, identification method and system

Abstract

The invention discloses a training method, a recognition method and a system of a propeller blade flutter recognition model, and belongs to the technical field of wing propeller blade abnormal flutter recognition. The method comprises the following steps: constructing a source domain sample based on a public data set, and taking vibration signals in different flutter states and corresponding identifiers as target domain samples; features of vibration signals of different depth layers of the source domain sample and the target domain sample are obtained, the maximum mean value difference is determined according to the features, and then a joint loss function of a loss vector and task classification loss is constructed; a deep convolutional neural network model is pre-trained according to a source domain sample, transfer learning of a pre-trained model from a source domain feature to a target domain feature is realized according to a joint loss function, a propeller blade flutter recognition model is obtained, a vibration signal is input into the propeller blade flutter recognition model, and the flutter state and type of a wing can be obtained. And real-time and accurate flutter monitoring and identification are realized, so that the operation safety and reliability of an aircraft are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of abnormal flutter identification of wing propeller blades, and specifically provides a training method, an identification method, and a system for a propeller blade flutter identification model. Background Art

[0002] As a key propulsion device on an aircraft, the normal operation of a wing propeller is crucial for the flight performance of the aircraft. However, during actual operation, due to the dynamic coupling effect between the airflow, the propeller blades, and the wing structure, the propeller blades are prone to flutter. This flutter not only reduces the propulsion efficiency of the propeller, affects the flight speed of the aircraft, but may also generate noise pollution, affecting the comfort of the cockpit and the passenger cabin. More seriously, long-term flutter may lead to fatigue fracture of the blades and even cause structural damage to the entire propeller drive system, seriously threatening the operational reliability and safety of the aircraft.

[0003] To effectively address the problems caused by propeller blade flutter, it is particularly important to monitor and accurately identify the flutter phenomenon. By real-time monitoring the vibration signals of the blades, the dynamic characteristics of the blades can be precisely adjusted, such as changing their stiffness, mass distribution, etc., to reduce the probability of flutter occurrence. At the same time, accurate flutter identification can also help optimize the maintenance cycle of the propeller, avoiding unnecessary maintenance costs. In addition, for potential problems found during monitoring, the mechanical structure of the propeller device can be optimized to improve its anti-flutter ability and overall performance. However, during the actual operation of the wing propeller, due to the high-speed rotational movement of the blades relative to the aircraft fuselage, traditional wired monitoring devices are difficult to obtain the vibration signals of the blades in real time, especially for low-intensity flutter phenomena, which poses a great challenge to the monitoring and identification of flutter.

[0004] In view of the limitations of traditional wired monitoring devices in the flutter monitoring of propeller blades, it is particularly important to develop a wireless monitoring and identification method to achieve real-time and accurate flutter monitoring and identification. Summary of the Invention

[0005] Aiming at the problems existing in the prior art, the present invention provides a training method, an identification method, and a system for a propeller blade flutter identification model, which can achieve real-time and accurate flutter monitoring and identification, thereby improving the operational safety and reliability of the aircraft, reducing maintenance costs, and reducing noise pollution, and having significant economic benefits and practical value.

[0006] The present invention is realized through the following technical solutions: First invention, the present application provides a training method for a propeller blade flutter identification model, including the following steps: Step 1: Construct source domain samples based on the blade flutter dataset. According to the vibration signals corresponding to different flutter states of the target wing to be constructed and their identification, the vibration signals of different flutter states and the corresponding identifications are used as target domain samples. Step 2: Use a deep convolutional neural network model to extract the source domain samples and target domain samples, obtain the features of the vibration signals at different depth layers, determine the maximum mean difference according to the features of different depth layers, determine the loss vector according to the maximum mean difference, and then construct a joint loss function of the loss vector and the task classification loss. Step 3: Pre-train the deep convolutional neural network model according to the source domain samples, and realize the transfer learning of the pre-trained model from the source domain features to the target domain features according to the joint loss function to obtain a propeller blade flutter recognition model.

[0007] Preferably, constructing the source domain samples based on the blade flutter dataset in Step 1 includes: Generate the time-frequency domain diagram of blade vibration according to the blade flutter dataset, and construct source domain samples according to the time-frequency domain diagram of blade vibration.

[0008] Preferably, the construction method of the target domain samples in Step 1 is as follows: Select the target wing. Use an experimental method to measure the vibration signals corresponding to different flutter states of the target wing under actual working conditions and classify and identify them. Obtain the time-frequency domain diagram of the vibration signal, and use the time-frequency domain diagram and the corresponding classification identification as the target domain samples.

[0009] Preferably, the deep convolutional neural network model in Step 2 is a ResNet model; Input the source domain samples and target domain samples into the ResNet model. The ResNet model extracts the shallow features and deep features of the source domain samples and target domain samples, determines the maximum mean difference between the shallow features and deep features, and constructs a joint loss function according to the maximum mean difference.

[0010] Preferably, realizing the transfer learning of the source domain features to the target domain features according to the joint loss function to obtain a propeller blade flutter recognition model includes: Input the source domain samples into the deep convolutional neural network model, and optimize the deep convolutional neural network model using the classification loss to obtain a pre-trained model; Input the target domain samples into the pre-trained model to obtain the multi-level features of the source domain and the target domain; Calculate the maximum mean difference between the source domain and target domain features layer by layer; Construct a joint loss function according to the maximum mean difference to train the model, update the model parameters through backpropagation, minimize the joint loss, and realize the transfer alignment of the source domain features to the target domain; Freeze the underlying parameters of the model, optimize the high-level network parameters using the target domain samples and classification loss, minimize the classification loss through backpropagation, update the high-level network parameters, and obtain the propeller blade flutter recognition model.

[0011] Second invention, the present application provides a training system for a propeller blade flutter recognition model, including: A sample construction module for constructing source domain samples based on a blade flutter data set, identifying vibration signals corresponding to different fluttering states of the constructed target wing, and using the vibration signals and corresponding identifications in different fluttering states as target domain samples; A loss module for using a deep convolutional neural network model to extract source domain samples and target domain samples, obtaining the characteristics of vibration signals in different depth layers, determining the maximum mean difference based on the characteristics of different depth layers, determining a loss vector based on the maximum mean difference, and further constructing a joint loss function of the loss vector and the task classification loss; A training module for pre-training a deep convolutional neural network model according to the source domain samples, and realizing transfer learning of the pre-trained model from source domain features to target domain features according to the joint loss function to obtain a propeller blade flutter recognition model.

[0012] Third invention, the present application provides a propeller blade flutter recognition method including the following steps: S1. Collect vibration signals during the operation of the propeller and perform digital sampling on the vibration signals; S2. Add serial numbers to the sampled vibration digital signals and generate corresponding cyclic redundancy check codes; S3. Check the integrity of the vibration digital signals according to the cyclic redundancy check codes and in combination with the serial number mechanism and the cyclic redundancy check mechanism to obtain complete vibration digital signals; S4. Input the vibration digital signals into the propeller blade flutter recognition model trained by the training method described in any one of claims 1-5 to obtain the flutter state and corresponding flutter type of the propeller.

[0013] Preferably, when checking the integrity of the vibration digital signals, if there is data loss in the vibration digital signals, process the vibration digital signals and re-check them.

[0014] Fourth invention, the present application provides a propeller blade flutter recognition system, including: An acquisition module for collecting vibration signals during the operation of the propeller and performing digital sampling on the vibration signals; A sequence module for adding serial numbers to the sampled vibration digital signals and generating corresponding cyclic redundancy check codes; A verification module, which is used to verify the integrity of the vibration digital signal according to the cyclic redundancy check code and by combining the serial number mechanism and the cyclic redundancy check mechanism, so as to obtain the integrity vibration digital signal; An identification module, which is used to input the vibration digital signal into the propeller blade flutter identification model trained by the training method described in any one of claims 1-5, so as to obtain the flutter state of the propeller and the corresponding flutter type.

[0015] The fifth invention, this application provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps of the described propeller blade flutter identification method are implemented.

[0016] Compared with the prior art, the present invention has the following beneficial technical effects: For the training method of a propeller blade flutter identification model provided by this application, firstly, by using a deep convolutional neural network (CNN) model, it can automatically extract multi-level features from vibration signals, including shallow local features and deep global features, so as to achieve high-precision identification of abnormal flutter of propeller blades. Secondly, by introducing the maximum mean discrepancy (MMD) to measure the distribution difference between the source domain and the target domain, and constructing a joint loss function, hierarchical domain adaptation is realized. It effectively reduces the inter-domain difference, enables the model to also show good performance in the target domain, and enhances the generalization ability of the model. In addition, in fields such as aero-engine health monitoring, it is often very difficult to obtain a large amount of labeled data. This method combines transfer learning technology, and realizes the transfer learning of source domain features to target domain features through the joint loss function. Even if the labeled data in the target domain is limited, the model can use the knowledge of the source domain to improve the performance in the target domain, effectively solving the domain shift problem in propeller blade flutter identification; at the same time, through transfer learning, the model can also achieve good identification results with a small amount of target domain data, and has strong practicality and application value. This method effectively solves the domain shift and small sample problems in propeller blade flutter identification through the combination of deep CNN, joint loss function and transfer learning technology, and has the advantages of high precision, strong generalization and engineering practicality, and has important application value in fields such as aero-engine health monitoring.

[0017] This application also proposes a propeller blade flutter identification system and an electronic device, which have all the advantages of the above-mentioned training method of a propeller blade flutter identification model. Brief Description of the Drawings

[0018] To more clearly illustrate the technical solutions of the embodiments of the present application, the accompanying drawings required for the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present application and should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0019] Figure 1 It is a flowchart of the method for identifying propeller blade flutter of the present invention; Figure 2 It is a schematic diagram of the wireless blade flutter monitoring system of the present invention; Figure 3 It is a flowchart of constructing the intelligent identification model for blade flutter of the present invention. Detailed implementation manners

[0020] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Usually, the components of the embodiments of the present application described and shown in the accompanying drawings here can be arranged and designed in various different configurations.

[0021] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed present application, but merely represents the selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts belong to the scope of protection of the present application.

[0022] A training method for a propeller blade flutter identification model includes the following steps: Step 1: Construct source domain samples based on the blade flutter data set, construct vibration signals corresponding to different fluttering states of the target wing according to the source domain samples and label them, and use the vibration signals of different fluttering states and the corresponding labels as target domain samples.

[0023] Construct source domain samples based on the existing blade flutter data set. These samples contain vibration signals under different fluttering states of different wings and label each state.

[0024] Use the vibration signals and labels of the target wing obtained from the source domain samples as target domain samples. The purpose of this step is to provide target domain data for subsequent transfer learning to ensure that the model can perform well in the target domain.

[0025] Step 2: Use a deep convolutional neural network model to extract the features of the source domain samples and the target domain samples for the vibration signals at different depth layers, determine the maximum mean discrepancy based on the features at different depth layers, determine the loss vector based on the maximum mean discrepancy, and then construct a joint loss function of the loss vector and the task classification loss; Deep Convolutional Neural Network (CNN): Use CNN to extract the features of the source domain and target domain samples. CNN can automatically extract multi-level features from vibration signals.

[0026] Maximum Mean Discrepancy (MMD): Measure the distribution difference between two domains by calculating the maximum mean discrepancy between the features of the source domain and target domain samples at different depth layers, which is used to reduce the distribution difference between the source domain and the target domain.

[0027] Deep CNN (such as ResNet or Inception variants) extracts the shallow local features and deep global features of vibration signals, and realizes hierarchical domain adaptation through layer-by-layer MMD calculation.

[0028] Joint loss function: Combine the MMD loss and the task classification loss to construct a joint loss function. The MMD loss is used to reduce the domain difference, and the task classification loss is used to ensure the classification performance of the model on the source domain. By jointly optimizing these two losses, the model can perform better on the target domain.

[0029] Step 3: Train a deep convolutional neural network model according to the source domain samples, and realize the transfer learning of the source domain features to the target domain features according to the joint loss function, so as to obtain a propeller blade flutter recognition model.

[0030] Use the source domain samples to train a deep convolutional neural network model. Through the joint loss function, the model not only learns the classification task of the source domain during training, but also learns how to transfer the source domain features to the target domain.

[0031] Through the joint loss function, the model realizes the transfer learning from the source domain to the target domain. The advantage of transfer learning is that even if the labeled data in the target domain is limited, the model can use the knowledge of the source domain to improve the performance on the target domain.

[0032] This method effectively solves the domain shift and small sample problems in propeller blade flutter recognition by combining deep CNN, joint loss function and transfer learning techniques, and has high accuracy, strong generalization ability and engineering practicability, and has important application value in the fields of aero-engine health monitoring, etc.

[0033] Refer to Figure 3 , based on the above propeller blade flutter recognition model, this application also provides a propeller blade flutter recognition method, and the specific steps are as follows: S1. Collect the vibration signals during the operation of the propeller and perform digital sampling on the vibration signals; Collecting vibration signals is the basis for identifying flutter because flutter usually manifests as anomalies in vibration signals. The vibration conditions of the propeller during operation are accurately captured by sensors, and then the continuous vibration signals are converted into discrete digital signals using digital sampling technology, which is convenient for subsequent digital processing and analysis. The sampling frequency and accuracy have an important impact on the accuracy of the results.

[0034] S2. Add serial numbers to the sampled vibration digital signals and generate corresponding cyclic redundancy check codes; Adding serial numbers to the sampled vibration digital signals means adding a unique identifier to each data point packet, which is convenient for data tracking and verification and realizes the integrity detection of vibration signals.

[0035] Generating cyclic redundancy check codes is an error detection technology used to ensure data integrity. By calculating the check code of the data, it can be verified at the receiving end whether the data has been tampered with or damaged during transmission.

[0036] S3. According to the cyclic redundancy check codes and combined with the serial number mechanism and cyclic redundancy check mechanism, verify the integrity of the vibration digital signals to obtain integrity vibration digital signals; Verifying data integrity uses cyclic redundancy check codes and serial number mechanisms to check whether the data remains intact during transmission or processing. This is a key step to ensure the accuracy of subsequent analysis; after verification, complete and undamaged vibration digital signals are obtained for subsequent blade flutter analysis.

[0037] S4. Construct source domain samples based on the blade flutter data set, construct the vibration signals corresponding to different fluttering states of the target wing according to the source domain samples and label them, and use the vibration signals and corresponding labels in different fluttering states as target domain samples.

[0038] Source domain samples are a set of vibration signals with known fluttering states and are used to train the model.

[0039] Extract the vibration signals in different fluttering states according to the source domain samples, assign a unique identifier to each state, obtain vibration signals of different categories, and label the vibration signals to form target domain samples. These vibration signals with labels constitute the target domain samples and are used for model testing and verification.

[0040] S5. Use a deep convolutional neural network model to extract the features of the vibration signals in the source domain samples and target domain samples at different depth layers, determine the maximum mean difference according to the features at different depth layers, determine the loss vector according to the maximum mean difference, and then construct a joint loss function of the loss vector and the task classification loss; The deep convolutional neural network model is a powerful feature extraction tool that can automatically learn the deep features of data; through the deep convolutional neural network, the features of the source domain and target domain samples at different depth layers are extracted.

[0041] Measuring the difference between the feature distributions of the source domain and target domain is the key in transfer learning. Combining the loss vector and the task classification loss, a joint loss function for optimizing the model is constructed.

[0042] S6. Pre-train the deep convolutional neural network model according to the source domain samples, and realize the transfer learning of the source domain features to the target domain features according to the joint loss function to obtain the propeller blade flutter recognition model.

[0043] Use the source domain samples to train the deep convolutional neural network. Through the joint loss function, guide the model to learn how to transfer the source domain features to the target domain and improve the generalization ability of the model. After training and optimization, a model that can identify the propeller blade flutter is obtained.

[0044] The training method is as follows: 1) Input the source domain samples into the deep convolutional neural network model, and optimize the deep convolutional neural network model using the classification loss to make the model achieve high classification accuracy on the source domain and obtain the pre-trained model; 2) Input the target domain samples into the pre-trained model to obtain the multi-level features of the source domain and target domain.

[0045] The multi-level features include shallow features (time-domain detail features) and deep features (frequency-domain global features).

[0046] 3) Calculate the maximum mean difference between the source domain and target domain features layer by layer; 4) Train the model according to the maximum mean difference by constructing a joint loss function, update the model parameters through backpropagation, minimize the joint loss, and realize the transfer alignment of the source domain features to the target domain.

[0047] 5) Freeze the underlying parameters of the model, optimize the high-level network parameters using the target domain samples and the classification loss, minimize the classification loss through backpropagation, update the high-level network parameters, and obtain the propeller blade flutter recognition model.

[0048] S7. Input the vibration digital signal obtained in step S3 into the propeller blade flutter recognition model to obtain the flutter state and corresponding flutter type of the propeller blade.

[0049] Input the verified vibration digital signal into the recognition model to obtain the flutter state and type: The model outputs the flutter state of the propeller blade and the corresponding flutter type according to the input vibration signal, providing decision support for subsequent maintenance or repair.

[0050] Correspondingly, the present application further provides a training system for a propeller blade flutter recognition model, which may include: A sample construction module, configured to construct source domain samples based on a blade flutter data set, construct vibration signals corresponding to different flutter states according to the source domain samples and identify them, and use the vibration signals of different flutter states and the corresponding identifications as target domain samples.

[0051] A loss module, configured to use a deep convolutional neural network model to extract the features of the vibration signals of the source domain samples and the target domain samples at different depth layers, determine the maximum mean difference according to the features of different depth layers, determine a loss vector according to the maximum mean difference, and further construct a joint loss function of the loss vector and the task classification loss; A transfer training module, configured to train a deep convolutional neural network model according to the source domain samples, and implement transfer learning of the source domain features to the target domain features according to the joint loss function to obtain a propeller blade flutter recognition model.

[0052] Embodiment 1 As Figure 1 shown, a recognition method for a propeller blade flutter recognition device includes the following steps: S100: Construct a wireless blade flutter monitoring system as Figure 2 shown.

[0053] The wireless blade flutter monitoring system consists of two parts: a blade end monitoring device and a wing end recognition device. The blade end monitoring device includes a vibration acceleration sensor, a signal acquisition module, an analog-to-digital conversion module, a power supply module, a data storage module, and a wireless data transmission module; the wing end recognition device includes a wireless data reception module, a power supply module, a data processing module, and an intelligent analysis module.

[0054] S200: The power supply module at the blade end monitoring device uses a 12V lithium battery as the power input, isolates multiple branches through a 0-ohm resistor, and uses a linear voltage regulator chip to adjust the power supply voltage of each branch to meet the power supply voltage requirements of the vibration acceleration sensor and each module of the blade end monitoring device; S300: Fix the vibration acceleration sensor on the leeward side surface of the propeller blade to collect the vibration signal during the operation of the propeller in real time. Use the signal acquisition module to perform preprocessing operations such as noise reduction, DC component removal, low-pass filtering, and anti-aliasing filtering on the collected vibration signal, and then use the analog-to-digital conversion module to perform digital sampling on the vibration signal after the preprocessing operation.

[0055] S400: Add a serial number to each group of sampled vibration digital signals and generate the corresponding cyclic redundancy check code, save the vibration digital signals and the corresponding serial numbers and cyclic redundancy check codes to the data storage module, and then send the vibration digital signals and the corresponding serial numbers and cyclic redundancy check codes to the wireless data receiving module at the wing-end identification device through the wireless data sending module.

[0056] S500: After the wireless data receiving module at the wing-end identification device receives the vibration digital signal, use the serial number mechanism and cyclic redundancy check mechanism to verify the integrity of the transmitted vibration digital signal. When it is identified that data is missing due to fluctuations during the wireless data transmission process, relevant data can be re-extracted from the data storage module and the wireless data transmission can be carried out again to improve the fault tolerance of the wireless blade flutter monitoring system.

[0057] In this embodiment, the wireless data sending and receiving module can select wireless data transmission methods such as infrared, Bluetooth, and WLAN, or other wireless data transmission methods that can execute the solution in this specification, considering the mechanical structure and size characteristics of the wing and propeller blades, and comprehensively considering the transmission rate, transmission reliability, module size, operating power consumption, and obstacle penetration ability.

[0058] S600: According to Figure 3 the intelligent recognition model construction flowchart shown, generate the vibration time-frequency domain diagram of the blade based on the public dataset of the blade flutter experiment as the source domain sample for training the blade flutter type classification model. Then, for the wing propeller blade of the target model, experimentally measure the vibration signals corresponding to different types of flutter that may occur in the actual working condition environment, generate the corresponding vibration time-frequency domain diagram and classify and label it according to Table 1 as the target domain sample for training the blade flutter type classification model.

[0059] Table 1 Classification labels for propeller blade flutter states

[0060] S700: Input the generated time-frequency domain diagrams into the ResNet model, extract the features of different layer depths of the source domain samples and target domain samples respectively, calculate the maximum mean discrepancy (MMD) of the features of different layers, and obtain the MMD-deep loss vector. Design a joint loss function that includes the task classification loss and the MMD-deep loss vector. On this basis, pre-train the ResNet model with the source domain data, and then use the designed joint loss function to realize the transfer learning of the source domain features to the target domain features, establish an intelligent recognition model for propeller blade flutter, and input the model into the intelligent analysis module at the wing-end identification device.

[0061] In this embodiment, the public dataset of blade flutter experiments discloses experimental data such as Turbomachinery Flutter Databases and EU-funded Aeroelastic Projects, and can also be experimental and simulation data provided in academic research such as Transonic Steam Turbine Blade Flutter Test. In addition, due to the differences between the test objects of the public dataset and the actual target objects, and the different degrees of differences, when designing the combined loss function, it is necessary to optimize the weight parameters and weight hyperparameters of each layer of MMD-deep.

[0062] S800 forwards the real-time vibration data of the propeller blade to the data processing module, and uses the program in the data processing module to generate the time-frequency domain diagram of the real-time vibration signal of the propeller blade. The time-frequency domain diagram of the blade vibration generated by the data processing module is input into the intelligent analysis module to monitor in real time whether there is a flutter phenomenon in the propeller blade and the type of flutter that occurs.

[0063] Regarding the problem of propeller blade flutter monitoring, the real-time monitoring of the propeller blade vibration is realized through a wireless monitoring device. The vibration signal is directly collected at the source of the flutter part of the propeller blade, avoiding the attenuation of the flutter characteristics caused by the mechanical structure, which is beneficial to the accurate identification of the low-intensity blade flutter phenomenon. Using the public dataset of blade flutter as the source domain to train the intelligent identification model of the target propeller blade flutter can greatly reduce the demand for experimental data of the target propeller blade flutter, thus effectively reducing the cost of constructing the intelligent identification model of blade flutter.

[0064] It should be noted that in the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of each module is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules can be combined or integrated into another device, or some features can be ignored or not executed. The modules described as separate components may or may not be physically separated. The components shown as modules can be one physical unit or multiple physical units, that is, they can be located in one place or distributed to multiple different places. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0065] In addition, each module in the various embodiments of the present invention can be integrated in a processing unit, or each module can exist physically alone, or two or more modules can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0066] An electronic device provided by an embodiment of the present application includes a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the steps of a propeller blade flutter identification method described in any of the above embodiments are implemented.

[0067] Another electronic device provided by an embodiment of the present application may further include: an input port connected to the processor, configured to transmit multi-modal data collected by an external acquisition device to the processor; and a display unit connected to the processor, configured to display the processing result of the processor to the outside; a communication module connected to the processor, configured to implement communication between the electronic device and the outside. The display unit may be a display panel, a laser scanning display, etc.; the communication methods adopted by the communication module include but are not limited to Mobile High-Definition Link technology (HML), Universal Serial Bus (USB), High-Definition Multimedia Interface (HDMI), wireless connection (including Wi-Fi technology, Bluetooth communication technology, low-power Bluetooth communication technology, communication technology based on IEEE802.11s).

[0068] A computer-readable storage medium provided by an embodiment of the present application stores a computer program. When the computer program is executed by a processor, the steps of a propeller blade flutter identification method described in any of the above embodiments are implemented.

[0069] For the description of the relevant parts in a propeller blade flutter identification system, an electronic device, and a computer-readable storage medium provided by an embodiment of the present application, please refer to the detailed description of the corresponding parts in a propeller blade flutter identification method provided by an embodiment of the present application, which will not be elaborated here. In addition, the parts of the above technical solutions provided by the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive elaboration.

[0070] The above content is only to illustrate the technical idea of the present invention, and the protection scope of the present invention cannot be limited thereby. Any modification made on the basis of the technical solution according to the technical idea proposed by the present invention falls within the protection scope of the claims of the present invention.

Claims

1. A method for training a propeller blade flutter recognition model, characterized in that: The following steps are involved: Step 1: construct source domain samples based on the blade flutter dataset, construct and identify vibration signals corresponding to different flutter states of the target wing, and use the vibration signals of different flutter states and corresponding identifiers as target domain samples; Step 2: Use a deep convolutional neural network model to extract source domain samples and target domain samples, obtain the characteristics of vibration signals at different depth layers, determine the maximum mean difference based on the characteristics of different depth layers, determine the loss vector based on the maximum mean difference, and then construct a joint loss function of the loss vector and the task classification loss; Step 3: Pre-train a deep convolutional neural network model based on source domain samples, and implement transfer learning of the pre-trained model from source domain features to target domain features based on the joint loss function to obtain a propeller blade flutter recognition model.

2. The method for training a propeller blade flutter recognition model according to claim 1, characterized in that: In step 1, source domain samples are constructed based on the blade flutter dataset, including: The blade vibration time-frequency domain diagram is generated according to the blade flutter data set, and the source domain samples are constructed according to the blade vibration time-frequency domain diagram.

3. The method for training a propeller blade flutter recognition model according to claim 1, characterized in that: The method for constructing the target domain sample in step 1 is as follows: Select the target wing; The vibration signals corresponding to different flutter states of the target wing under actual working conditions are measured by experimental methods and classified and identified; Obtain the time-frequency domain diagram of the vibration signal, and use the time-frequency domain diagram and the corresponding classification identifier as target domain samples.

4. The method for training a propeller blade flutter recognition model according to claim 1, characterized in that: The deep convolutional neural network model described in step 2 is a ResNet model; The source domain samples and the target domain samples are input into the ResNet model. The ResNet model extracts the shallow features and deep features of the source domain samples and the target domain samples, determines the maximum mean difference between the shallow features and the deep features, and constructs a joint loss function based on the maximum mean difference.

5. The method for training a propeller blade flutter recognition model according to claim 1, characterized in that: The transfer learning from source domain features to target domain features is realized according to the joint loss function, and the propeller blade flutter recognition model is obtained, including: The source domain samples are input into the deep convolutional neural network model, and the deep convolutional neural network model is optimized using classification loss to obtain a pre-trained model; Input the target domain samples into the pre-trained model to obtain multi-level features of the source domain and the target domain; Calculate the maximum mean difference between source domain and target domain features layer by layer; The joint loss function is constructed based on the maximum mean difference to train the model, and the model parameters are updated through back propagation to minimize the joint loss and achieve the migration alignment of source domain features to the target domain. The underlying parameters of the model are frozen, and the high-level network parameters are optimized using target domain samples and classification loss. The classification loss is minimized through back propagation, and the high-level network parameters are updated to obtain the propeller blade flutter recognition model.

6. A training system for a propeller blade flutter identification model, characterized in that: include: A sample construction module is used to construct source domain samples based on the blade flutter data set, and to construct and identify vibration signals corresponding to different flutter states of the target wing, and to use the vibration signals of different flutter states and corresponding identifiers as target domain samples; A loss module is used to extract source domain samples and target domain samples using a deep convolutional neural network model, obtain features of vibration signals at different depth layers, determine the maximum mean difference according to the features at different depth layers, determine the loss vector according to the maximum mean difference, and then construct a joint loss function of the loss vector and the task classification loss; The training module is used to pre-train a deep convolutional neural network model based on source domain samples, and to achieve transfer learning of the pre-trained model from source domain features to target domain features based on a joint loss function, so as to obtain a propeller blade flutter recognition model.

7. A propeller blade flutter identification method, characterized in that: The following steps are involved: S 1. Collecting vibration signals during the operation of the propeller and digitally sampling the vibration signals; S2, adding a serial number to the sampled vibration digital signal and generating a corresponding cyclic redundancy check code; S3, verifying the integrity of the vibration digital signal according to the cyclic redundancy check code and combining the sequence number mechanism with the cyclic redundancy check mechanism to obtain a vibration digital signal with integrity; S4. Input the vibration digital signal into the propeller blade flutter recognition model trained by the training method described in any one of claims 1 to 5 to obtain the flutter state of the propeller and the corresponding flutter type.

8. A propeller blade flutter identification method according to claim 7, characterized in that: The integrity of the vibration digital signal is verified. When there is data missing in the vibration digital signal, the vibration digital signal is processed and re-verified.

9. A propeller blade flutter identification system, characterized in that: include: The acquisition module is used to collect the vibration signal of the propeller during operation and perform digital sampling on the vibration signal; A sequence module, used to add a sequence number to the sampled vibration digital signal and generate a corresponding cyclic redundancy check code; A verification module is used to verify the integrity of the vibration digital signal according to the cyclic redundancy check code and in combination with the sequence number mechanism and the cyclic redundancy check mechanism to obtain a vibration digital signal with integrity; The recognition module is used to input the vibration digital signal into the propeller blade flutter recognition model trained by the training method described in any of claims 1-5 to obtain the flutter state of the propeller and the corresponding flutter type.

10. An electronic device, characterized in that: The method comprises a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of a propeller blade flutter identification method as claimed in claim 7 or 8 are implemented.

Images