A dynamic stress calibration method and system for a target aircraft with real-time feedback

Abstract

The present invention relates to the technical field of stress calibration, and particularly to a dynamic stress calibration method and system for a target aircraft with real-time feedback. The method includes: Step 1: Divide the dynamic stress sensitive area based on the three-dimensional dynamic model of the target aircraft; Step 2: Synchronously collect the original stress signals output by the multi-axis stress sensor array to generate preprocessed stress signals; Step 3: Input the preprocessed stress signals into the dynamic stress compensation model; Step 4: In the controlled loading test of the target aircraft, compare the calibrated stress data with the measurement data of the reference measurement device, and reversely correct the neural network weight parameters of the dynamic stress compensation model according to the comparison difference value, and update the weight allocation strategy of the weighted least squares method to complete the dynamic iteration of the calibration parameters. By combining the multi-axis stress sensor array with the dynamic stress compensation model, the accurate measurement and compensation of the dynamic stress of the target aircraft are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of stress calibration, and particularly to a method and system for dynamic stress calibration of a target aircraft with real-time feedback. Background Art

[0002] In the prior art, the arrangement of sensors is usually based on experience or simple geometric distribution, without fully considering the dynamic stress transmission path and stress concentration area of the target aircraft. This arrangement method may lead to insufficient stress measurement accuracy in key areas, while there are too many sensors arranged in low-stress areas, resulting in waste of resources;

[0003] In addition, in the prior art, the stress signal acquisition system of the target aircraft usually adopts a fixed sampling frequency and cannot dynamically adjust according to the real-time motion state of the target aircraft.

[0004] In summary, there are many deficiencies in the prior art in the field of dynamic stress calibration, and there is an urgent need for a calibration method and system that can provide real-time feedback and dynamically optimize to improve the structural safety and measurement accuracy of the target aircraft. Summary of the Invention

[0005] Based on the above purpose, the present invention provides a method and system for dynamic stress calibration of a target aircraft with real-time feedback. The method includes the following steps:

[0006] Step 1: Divide the dynamic stress sensitive area based on the three-dimensional dynamic model of the target aircraft, and arrange a multi-axis stress sensor array in the dynamic stress sensitive area according to a preset rule, where the preset rule includes a method for determining the sensor spacing and a method for adjusting the density in the stress concentration area;

[0007] Step 2: Synchronously collect the original stress signals output by the multi-axis stress sensor array, dynamically adjust the signal sampling frequency according to the real-time motion acceleration value of the target aircraft, and perform adaptive filtering processing on the original stress signals based on a noise feature library to generate preprocessed stress signals;

[0008] Step 3: Input the preprocessed stress signals into a dynamic stress compensation model. The dynamic stress compensation model takes the target aircraft motion acceleration value and the target aircraft angular velocity value as input parameters, predicts the theoretical stress values at the positions of each sensor through a pre-trained neural network, calculates the dynamic error distribution matrix between the measured stress values and the theoretical stress values, iteratively optimizes the compensation coefficients using the weighted least squares method, and triggers a closed-loop feedback mechanism according to the change rate of the compensation coefficients;

[0009] Step 4: In the controlled loading test of the target aircraft, compare the calibrated stress data with the measurement data of the reference measurement device, reversely correct the neural network weight parameters of the dynamic stress compensation model according to the comparison difference value, and update the weight distribution strategy of the weighted least squares method to complete the dynamic iteration of the calibration parameters.

[0010] Preferably, the method for determining the sensor spacing in step 1 includes:

[0011] a. Collect stress attenuation data on the main stress transmission path through the target drone's no-load motion test, and calculate the stress attenuation gradient at different positions;

[0012] b. Determine the ratio of the equal ratio spacing according to the stress attenuation gradient, so that the adjacent sensor spacing increases according to the ratio as the stress attenuation gradient increases;

[0013] The method for adjusting the density of the stress concentration area includes:

[0014] a. Conduct finite element simulation on the connection part and the cross-section mutation area of the target drone, and extract the maximum stress gradient value and the average stress gradient value of the area;

[0015] b. Use the ratio of the maximum stress gradient value to the average stress gradient value as the multiple of the sensor density increase, and arrange redundant sensors in the area according to the increase multiple.

[0016] Preferably, the process of dynamically adjusting the signal sampling frequency in step 2 includes:

[0017] Obtain the critical instability acceleration value corresponding to the maximum design speed of the target drone through the target drone's power calibration test, and use a preset percentage of the critical instability acceleration value as the acceleration threshold;

[0018] When the real-time motion acceleration value exceeds the acceleration threshold, adopt the high-frequency sampling mode. The frequency of the high-frequency sampling is an integer multiple of the structural resonance frequency of the target drone, and the value of the integer multiple is determined by analyzing the relationship between the resonance frequency and the signal aliasing effect;

[0019] When the real-time motion acceleration value is lower than or equal to the acceleration threshold, adopt the low-frequency sampling mode. The frequency of the low-frequency sampling is the frequency division value of the high-frequency sampling frequency, and the frequency division value is calculated according to the reciprocal of the product of the current motion speed value of the target drone and the structural resonance frequency.

[0020] Preferably, the adaptive filtering process in step 2 includes:

[0021] Collect the baseline noise signals at the positions of each sensor in the no-load stationary state of the target drone, perform fast Fourier transform on the baseline noise signals, extract the noise spectrum characteristics at the positions of each sensor, and establish the noise feature library;

[0022] Match the corresponding noise spectrum characteristics from the noise feature library according to the current sensor position, and select the filter type with the highest correlation with the noise spectrum characteristics in the filter bank;

[0023] The original stress signal is subjected to time-frequency analysis using a sliding window method. The length of the sliding window is dynamically adjusted according to the decay time of the autocorrelation function of the current moving speed value of the target aircraft. When the energy ratio of the signal within a preset frequency band exceeds a first threshold, it is determined as an effective stress component; otherwise, it is determined as environmental vibration noise and filtered out.

[0024] Preferably, the process of establishing the dynamic stress compensation model in step 3 includes:

[0025] The training data of the pre-trained neural network is obtained through a calibration test of the target aircraft. The calibration test includes: pasting reference strain gauges on the surface of the target aircraft, and synchronously recording the moving acceleration value, angular velocity value, temperature value of the target aircraft, and the strain gauge measurement values at the positions of each sensor under various preset motion modes;

[0026] The calculation process of the dynamic error distribution matrix includes: taking the difference between the measured stress value and the theoretical stress value at the position of each sensor as the initial error value, and performing spatial interpolation according to the stress influence factor at the position of each sensor. The stress influence factor is obtained through finite element simulation calculation;

[0027] The weight distribution strategy of the weighted least squares method includes: statistically calculating the standard deviation of the historical error data at the position of each sensor, and assigning a small weight value to the sensor position where the standard deviation exceeds a second threshold.

[0028] Preferably, the specific process of iteratively optimizing the compensation coefficient in step 3 includes:

[0029] The compensation coefficient is dynamically smoothed using the exponentially weighted moving average method. The smoothing coefficient of the exponential weighting is adjusted according to the mutation detection result of the motion state of the target aircraft, including:

[0030] When it is detected that the change rate of the moving acceleration exceeds a third threshold, the smoothing coefficient is reduced to accelerate the response speed. When the motion state is stable, the smoothing coefficient is increased to improve the noise resistance ability;

[0031] The triggering conditions of the closed-loop feedback mechanism include: the absolute value of the change rate of the compensation coefficient is less than a fourth threshold in N consecutive iterations. The values of N and the fourth threshold are determined by statistically calculating the number of stress fluctuations and amplitudes within a typical motion cycle of the target aircraft.

[0032] Preferably, the process of reversely correcting the neural network weight parameters in step 4 includes:

[0033] The calculation method of the comparison difference value is: calculating the root mean square error between the calibrated stress data and the data of the reference measurement device at the same timestamp;

[0034] The correction amount of the neural network weight parameters is determined according to a non-linear function of the root mean square error, and the slope value of the non-linear function is dynamically adjusted by analyzing the relationship between the error convergence speed and stability during the historical correction process;

[0035] The update process of the weight allocation strategy includes: for a group of sensors with spatial correlation, synchronously adjusting their weight values according to the error consistency index of the sensors within the group, and the spatial correlation is obtained by clustering and analyzing the historical error distribution pattern.

[0036] Preferably, the specific method for detecting the mutation of the motion state includes:

[0037] Calculating the difference sequence of the target drone's motion acceleration value in real time. When the number of times the absolute value of the difference sequence continuously exceeds the fifth threshold reaches a preset value, it is determined that the motion state has mutated;

[0038] The method for determining the fifth threshold is: statistically calculating the standard deviation of the acceleration difference sequence in the historical motion data of the target drone, taking three times the standard deviation as the initial value of the fifth threshold, and dynamically scaling it according to the real-time motion mode.

[0039] Preferably, during the execution of steps 1 to 4:

[0040] The calculation results of all dynamically adjusted parameters are stored in the parameter update queue, and the parameter update queue adopts a double-buffer mechanism to ensure data consistency during the parameter update process;

[0041] The update mechanism of the key decision threshold includes: regularly collecting the operation data of the target drone, calculating the statistical distribution characteristics of the operation data, and triggering the threshold recalibration process when the distribution characteristics deviate from the initial calibration value by more than the sixth threshold.

[0042] Correspondingly, the embodiment of the present invention further provides a target drone dynamic stress calibration system with real-time feedback, which is used to run a target drone dynamic stress calibration method with real-time feedback according to the embodiment of the present invention, including:

[0043] A dynamic stress sensitive area division module, which divides the dynamic stress sensitive area based on the three-dimensional dynamic model of the target drone and generates sensor layout planning information. Its input includes the three-dimensional dynamic model of the target drone, material property parameters, and motion condition data, and the output is a dynamic stress sensitive area distribution map and sensor layout planning information. This module is connected to the sensor array deployment module and transmits the sensor layout planning information to the subsequent module;

[0044] Sensor array deployment module. The sensor array deployment module deploys a multi-axis stress sensor array in a specified area according to the planning information provided by the dynamic stress sensitive area division module. It follows a preset rule. The input is the distribution map of the dynamic stress sensitive area and the sensor layout planning information, and the output is the topological structure data of the sensor array and the sensor position identification. This module is connected to the signal acquisition module to provide basic data support for signal acquisition;

[0045] Signal acquisition module. The signal acquisition module synchronously acquires the original stress signals output by the multi-axis stress sensor array and dynamically adjusts the signal sampling frequency according to the real-time motion acceleration value of the target aircraft. This module includes a sampling frequency adjustment unit and a signal buffer unit to ensure the efficiency and accuracy of signal acquisition. The input includes the topological structure data of the sensor array and the real-time motion acceleration value of the target aircraft, and the output is the original stress signal data and the sampling frequency adjustment parameters. The signal acquisition module is connected to the signal processing module to provide the original data for subsequent signal processing;

[0046] Signal processing module. The signal processing module performs adaptive filtering processing on the original stress signals based on the noise feature library to generate preprocessed stress signals. The module is divided into a noise feature library construction unit, a filter selection unit, and a time-frequency analysis unit, which are responsible for noise feature extraction, filter selection, and signal filtering respectively. The input is the original stress signal data and the current motion speed value of the target aircraft, and the output is the preprocessed stress signal data and the filter type identification. This module is connected to the dynamic stress compensation module to provide the processed signals for stress compensation;

[0047] Dynamic stress compensation module. The dynamic stress compensation module predicts the theoretical stress value through a pre-trained neural network, calculates the dynamic error distribution matrix, and uses the weighted least squares method to iteratively optimize the compensation coefficients. The module includes a neural network prediction unit, an error calculation unit, a compensation coefficient optimization unit, and a closed-loop feedback trigger unit to achieve precise calibration of stress data. The input is the preprocessed stress signal data, the motion acceleration value, and the angular velocity value of the target aircraft, and the output is the calibrated stress data, the compensation coefficient optimization parameters, and the closed-loop feedback trigger signal. This module is connected to the calibration verification module and the closed-loop feedback module to form a complete feedback control link;

[0048] Calibration verification module. The calibration verification module compares the calibrated stress data with the measurement data of the reference measurement device and reversely corrects the parameters of the dynamic stress compensation model. The comparison analysis unit, weight correction unit, and weight distribution update unit in the module work together to ensure the continuous optimization of the calibration parameters. The input is the calibrated stress data and the data of the reference measurement device, and the output is the corrected neural network weight parameters and the updated weight distribution strategy. The calibration verification module is connected to the dynamic stress compensation module to achieve closed-loop correction of the parameters;

[0049] Closed-loop feedback module. The closed-loop feedback module triggers the closed-loop feedback mechanism according to the change rate of the compensation coefficient, and dynamically adjusts the system parameters to adapt to the real-time motion state of the target aircraft. The mutation detection unit, parameter update queue management unit, and threshold calibration unit in the module are responsible for motion state monitoring, parameter storage, and threshold update respectively. The inputs are the change rate of the compensation coefficient and the real-time motion state data of the target aircraft, and the outputs are the smoothing coefficient adjustment parameters and the updated decision threshold. This module is connected to the dynamic stress compensation module and the signal acquisition module to ensure the real-time optimization of the system parameters;

[0050] System control module. The input is the status feedback information of each module, and the output is the communication instructions and task scheduling signals between each module. The system control module is connected to all modules to ensure the efficient operation of the system.

[0051] Advantages of the present invention:

[0052] By combining the multi-axis stress sensor array with the dynamic stress compensation model, the accurate measurement and compensation of the dynamic stress of the target aircraft are realized. First, the stress-sensitive area is accurately divided and the sampling frequency is dynamically adjusted according to the motion acceleration to ensure the high-quality acquisition of signals. Second, the pre-trained neural network and the weighted least squares method are used to iteratively optimize the compensation coefficient, improving the accuracy and response speed of stress compensation. Finally, the neural network weights are corrected by comparing with the reference device, ensuring the continuous optimization and dynamic adaptability of the compensation model, enhancing the real-time performance and accuracy of the stress calibration process of the target aircraft, and improving the stability and reliability of the system. Description of the drawings

[0053] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0054] Figure 1 It is the flowchart of the steps of the method of the present invention;

[0055] Figure 2 It is the flowchart of the method for determining the sensor spacing in step 1 of the method of the present invention;

[0056] Figure 3 It is the structural block diagram of the system of the present invention. Detailed implementation manners

[0057] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0058] Please refer to Figures 1 - 3 , an embodiment of the present invention provides a method for calibrating the dynamic stress of a target aircraft with real-time feedback. First, a dynamic stress sensitive area is divided through the three-dimensional dynamic model of the target aircraft. This area is mainly concentrated in the parts of the target aircraft that are easily affected by large stress concentration or changes. According to the stress distribution characteristics of this area, a multi-axis stress sensor array is arranged according to a preset rule. The spacing and density of the sensors will be dynamically adjusted according to the stress distribution of each part of the target aircraft, especially increasing the sensor density in the stress concentration area to more accurately capture the stress signal. This step lays the foundation for subsequent signal acquisition and data processing, ensuring that the stress areas that may change during the movement of the target aircraft can be covered.

[0059] Based on step 1, the sensor array starts to synchronously collect the original stress signals output by each sensor on the surface of the target aircraft. In order to adapt to the different accelerations generated by the target aircraft in different motion states, the signal sampling frequency will be dynamically adjusted according to the real-time motion acceleration value of the target aircraft. When accelerating, the sampling frequency is increased to ensure that higher-frequency stress changes can be captured; while in a lower speed or stable state, the sampling frequency is reduced to save computing resources. In addition, the collected original stress signals are subjected to adaptive filtering processing based on the noise feature library, which can effectively remove environmental noise and ensure the purity and effectiveness of the signals. This step ensures the data quality of subsequent signal processing.

[0060] The preprocessed stress signals are input into the dynamic stress compensation model. This model takes the motion acceleration value and angular velocity value of the target aircraft as inputs, and predicts the theoretical stress values at the positions of each sensor through a pre-trained neural network. By calculating the error between the actually measured stress value and the theoretical stress value, a dynamic error distribution matrix is obtained. The weighted least squares method is used to iteratively optimize the compensation coefficients, and at the same time, the closed-loop feedback mechanism is triggered by the change rate of the compensation coefficients. This mechanism can ensure that the compensation coefficients are adjusted in real time as the motion state changes, enabling the system to respond to the stress changes of the target aircraft in different motion modes in real time.

[0061] In the controlled loading test of the target aircraft, the calibrated stress data is compared with the data of the reference measurement equipment. By calculating the difference between the two, the neural network weight parameters in the dynamic stress compensation model are corrected inversely. Further, the weight distribution strategy is adjusted by the weighted least squares method, enabling the system to improve the accuracy of the model during continuous iteration. This step not only optimizes the compensation coefficient but also ensures that the system can perform accurate stress measurement stably in the long term through dynamic correction.

[0062] Through the effective connection and interaction of these four steps, this method can achieve precise real-time calibration of the dynamic stress data of the target aircraft, with significant advantages in accuracy, real-time performance, and robustness, and is applicable to high-precision target aircraft testing and applications.

[0063] Among them, the three-dimensional dynamic model of the target aircraft is established by constructing a mathematical model that accurately reflects the relationship between the forces and motion states of the target aircraft during operation, facilitating the calibration and compensation of the dynamic stress of the target aircraft. The realization of this model not only involves the structural analysis and motion simulation of the target aircraft but also includes the prediction and adjustment of real-time acceleration, angular velocity, and their corresponding stress distributions.

[0064] The three-dimensional dynamic model of the target aircraft is based on the basic principles of physical mechanics, combined with the geometric structure, mass distribution, and dynamic characteristics of the target aircraft. The model consists of the following key components:

[0065] Rigid body dynamics: The rigid body motion of the target aircraft is described using Newton-Euler equations or Lagrange equations. The model uses these equations to represent physical quantities such as the displacement, velocity, and acceleration of the target aircraft. Specifically, the acceleration and angular velocity of the target aircraft are closely related to the force conditions of the target aircraft and can affect the dynamic response and stress distribution of the target aircraft.

[0066] Mass and inertia matrix: The mass distribution and inertia matrix of the target aircraft are important components of the dynamic model. By detailed modeling of the mass and inertia characteristics of each component of the target aircraft, the response behavior of each component when subjected to external forces can be determined, and then the motion state of the target aircraft can be predicted. The inertia matrix can reflect the rotational inertia of the target aircraft and is the key to describing rotational motion.

[0067] External load and force analysis: The target aircraft is subjected to different types of external loads (such as air resistance, gravity, etc.) during motion. These external loads act on various parts of the target aircraft, resulting in stress distributions in different parts. Through finite element analysis (FEM) or other numerical simulation methods, the stress states of different regions on the surface of the target aircraft can be calculated.

[0068] In a possible implementation, first, a no-load motion test of the target drone is conducted, that is, the target drone is moved without an external load to simulate its motion state during actual use. The key to this process is to collect stress attenuation data on the main stress transfer path of the target drone. Stress attenuation data can reflect the stress reduction at different positions, thereby helping to understand how stress decays with distance. By calculating the stress attenuation gradient at different positions, the changing law of stress during the transfer from the center of the target drone to the outside can be obtained.

[0069] By calculating the stress attenuation gradient, the stress change rate at different locations is obtained. Locations with larger stress attenuation gradients indicate that stress changes faster at that location, so a higher sensor density is required to monitor the stress state in real time; in areas with smaller stress changes, the sensor spacing can be appropriately increased. Based on these attenuation gradient values, the ratio of the geometric spacing is determined so that the spacing between adjacent sensors increases according to the set ratio as the stress attenuation gradient increases. This method can minimize unnecessary sensor layout, optimize monitoring efficiency, and ensure that there are enough sensors in high stress areas to capture key data.

[0070] The joints and cross-sectional mutation areas of the target aircraft are usually places where stress is concentrated or where stress changes greatly. These areas are potential risk points for stress overload during the use of the target aircraft, so they require special attention. Through finite element simulation, the stress distribution of each component of the target aircraft during movement can be simulated, especially at the joints and cross-sectional mutation areas. This simulation step can extract the maximum stress gradient value and the average stress gradient value in the area, helping to evaluate the change amplitude and distribution law of stress in the area.

[0071] By calculating the ratio of the maximum stress gradient value to the average stress gradient value, the intensity of stress change in the area can be obtained. This ratio, as a multiple of the increase in sensor density, means that in areas with large stress gradients (i.e., more drastic stress changes), the sensor layout density needs to be increased. The purpose of this is to provide more redundant sensors in key areas of stress concentration to ensure the accuracy and reliability of the data. The arrangement of redundant sensors can avoid monitoring failures due to sensor failures or data errors, thereby improving the robustness of the entire target drone dynamic stress monitoring system.

[0072] The layout of sensors needs to be dynamically adjusted according to the actual motion state and stress distribution of the target drone. By combining the stress attenuation data and stress gradient analysis obtained in the no-load motion test, it can ensure that the sensor layout is more in line with the dynamic characteristics of the target drone, avoiding over-layout or under-layout. The strategy of increasing the geometric spacing can concentrate sensors in high-stress areas, improving the sensitivity to stress changes and monitoring accuracy.

[0073] By calculating the gradient information of the stress concentration area through finite element simulation and combining with the density adjustment strategy, the stress overload areas that may occur in the target aircraft can be accurately identified. Deploying more redundant sensors in these areas can prevent monitoring failures caused by inaccurate data of individual sensors. In addition, deploying redundant sensors in the cross-section mutation area and the connection area can effectively capture the mutation of stress, avoid the situation where the stress mutation is not detected in time, and thus early warning of possible structural risks.

[0074] In a possible implementation manner, the process of dynamically adjusting the signal sampling frequency mainly involves determining the signal sampling frequency according to the motion acceleration of the target aircraft and the resonant characteristics of the structure. This process adjusts the sampling frequency to adapt to the dynamic stress monitoring requirements of the target aircraft in different motion states, thereby improving the accuracy and real-time performance of the monitoring data.

[0075] Specifically, before starting to dynamically adjust the signal sampling frequency, it is necessary to obtain the critical instability acceleration value at the maximum designed speed of the target aircraft through the power calibration test of the target aircraft. The critical instability acceleration value represents that when the target aircraft reaches a certain speed, the system may have an unstable acceleration change. This value is used as a benchmark for setting the acceleration threshold. The acceleration threshold is usually a certain preset percentage (such as 80%) of the critical instability acceleration and is the basis for dynamically adjusting the sampling frequency. By setting this threshold, it is possible to determine whether the target aircraft enters a high acceleration state during the actual motion process to decide whether to switch to the high-frequency sampling mode.

[0076] Furthermore, when the real-time motion acceleration value of the target aircraft exceeds the set acceleration threshold, it indicates that the target aircraft has entered a high-speed motion or may have a severe dynamic response. In this case, in order to more accurately capture the stress change of the target aircraft in the high acceleration state, it is necessary to adopt the high-frequency sampling mode. The frequency of high-frequency sampling needs to be set as an integer multiple of the resonant frequency of the target aircraft structure. This is because in the high-frequency state, if the relationship between the sampling frequency and the structure resonant frequency is inappropriate, it may lead to the signal aliasing effect, that is, the high-frequency components of the stress signal cannot be correctly recorded. By analyzing the relationship between the resonant frequency and the signal aliasing effect, an appropriate integer multiple coefficient is determined to ensure that the high-frequency sampling can accurately capture the dynamic response of the target aircraft during high-speed motion.

[0077] When the real-time motion acceleration value of the target drone is lower than or equal to the acceleration threshold, it indicates that the target drone is in a state of lower motion speed or acceleration. At this time, the dynamic response of the target drone is relatively stable, and the required sampling frequency can be reduced to save storage space and processing resources. In this case, a low-frequency sampling mode is adopted. The low-frequency sampling frequency is a divided frequency value of the high-frequency sampling frequency. The calculation of this divided frequency value is based on the reciprocal of the product of the current motion speed of the target drone and the structural resonance frequency. Through this calculation, it can be ensured that in the low-acceleration state, the sampling frequency can adapt to the motion changes of the target drone, while avoiding the unnecessary computational and storage burdens brought by too high a frequency.

[0078] In a possible implementation manner, in the no-load stationary state of the target drone, first, the baseline noise signals at various positions of the target drone are collected by each sensor. These baseline signals do not contain stress information, but are environmental noise and the vibration noise of the device itself. Then, a fast Fourier transform (FFT) is performed on these collected baseline noise signals, so as to extract the noise spectral characteristics at the positions of each sensor. Through the FFT, the noise signal can be transformed from the time domain to the frequency domain, and the noise patterns in different frequency ranges can be identified. These noise spectral characteristics will then be stored in the noise feature library as the basis for subsequent filter selection.

[0079] In actual operation, according to the sensor positions in the current motion state of the target drone, the noise spectral characteristics most relevant to the current sensor positions are matched from the pre-established noise feature library. This matching process is carried out according to the sensor positions and the characteristics of the collected signals. When the noise spectral characteristics matching is completed, the system will select the filter type with the highest correlation with the matched noise spectral characteristics. This step is to ensure that the selected filter can most effectively remove the signals similar to the environmental noise, so as to more accurately retain the effective stress signals.

[0080] After the signal filter is selected, the original stress signal will be subjected to time-frequency analysis using the sliding window method. The sliding window method is a commonly used time-frequency analysis technique. By setting a sliding window on the signal, the frequency and time distribution of the signal are analyzed in real time. In this process, the length of the sliding window is dynamically adjusted according to the decay time of the autocorrelation function of the current motion speed value of the target drone. Specifically, when the motion speed of the target drone is high, the signal change speed will also increase, and at this time, the length of the sliding window will be appropriately shortened to maintain sufficient time resolution; on the contrary, in the low-speed motion state, the window length can be appropriately extended.

[0081] When performing time-frequency analysis, the system calculates the energy ratio of the signal in the preset frequency band. If the energy ratio of the signal in these frequency bands exceeds the preset first threshold, the signal is judged as an effective stress component, indicating that it contains useful stress information; otherwise, the signal is judged as environmental vibration noise. Signals judged as noise will be filtered out to avoid interference with the measurement of dynamic stress of the target aircraft.

[0082] In a possible implementation, the acquisition of training data is the basis for establishing a dynamic stress compensation model. These data are obtained through a target drone calibration test. The core of the test is to measure the strain change of the target drone by pasting a reference strain gauge on the surface of the target drone, and simultaneously record the acceleration value, angular velocity value, temperature value of the target drone in different motion modes and the strain gauge measurement value of each sensor position. Different preset motion modes simulate various motion states that the target drone may encounter under actual working conditions, ensuring the diversity and representativeness of the training data.

[0083] This process ensures the comprehensiveness and accuracy of the training data, enabling the neural network to learn the stress change patterns of the target aircraft under various dynamic conditions, thereby improving the accuracy and robustness of the model.

[0084] Based on the neural network training data, the next step is to calculate the dynamic error distribution matrix. The first step of the calculation process is to use the difference between the measured stress value and the theoretical stress value at each sensor location as the initial error value. These differences reflect the measurement error at each sensor location. Next, the error value is spatially interpolated using the stress influence factor at each sensor location. The stress influence factor is obtained through finite element simulation calculation and can accurately describe the stress distribution characteristics of each part of the target drone.

[0085] The stress influence factors obtained through finite element simulation can accurately reflect the stress transfer characteristics of each part of the target aircraft, ensuring that the spatial distribution of the error is highly consistent with the actual motion state of the target aircraft. The interpolation process enables the error value to be reasonably calculated between different sensors, thus providing an accurate basis for dynamic compensation.

[0086] After the error distribution matrix is calculated, the next step is to apply the weighted least squares method for dynamic stress compensation. The core of the weighted least squares method is to assign weights based on the historical error data of each sensor position. By counting the standard deviation of the historical error data of each sensor position, the sensor positions with large error fluctuations are identified and assigned small weight values, while the sensor positions with small error fluctuations and stable measurements are assigned large weight values. This ensures that during the compensation process, more attention is paid to stable and reliable sensor data, reducing the deviation introduced by large errors of individual sensors.

[0087] The weighted least squares method can effectively suppress the influence of noise and abnormal data during the dynamic stress compensation process by assigning higher weights to sensors with smaller errors, improving the accuracy and stability of the compensation results. At the same time, through the evaluation of the standard deviation of historical error data, the weights of each sensor can be dynamically adjusted to adapt to the changes of the target aircraft in different working environments.

[0088] By integrating technologies such as calibration tests, finite element simulations, and the weighted least squares method, an accurate dynamic stress compensation model has been effectively established, which can significantly improve the accuracy and real-time performance of the stress calibration of the target aircraft.

[0089] In a possible implementation, the iterative optimization of the compensation coefficient is first dynamically smoothed by the exponentially weighted moving average method (EWMA). This method uses past and current compensation coefficients to calculate a new smoothed value, with more recent values being given greater weight. The exponentially weighted smoothing coefficient determines the influence of historical data on the current calculation.

[0090] When the target aircraft is in motion and the change rate of the detected motion acceleration exceeds the third threshold, it means that the motion state of the target aircraft has changed rapidly or mutated. At this time, in order to respond more quickly to the change in the motion state, the system reduces the smoothing coefficient and assigns more weight to new data, thereby accelerating the adjustment of the compensation coefficient and improving the response speed.

[0091] When the motion state of the target aircraft becomes stable and the change rate of the motion acceleration is lower than the third threshold, it means that the motion state tends to be stable. At this time, the system increases the smoothing coefficient, increasing the influence of historical data on the current compensation coefficient, thereby improving the anti-noise ability and reducing fluctuations caused by noise interference.

[0092] Through this dynamic adjustment mechanism, the system can quickly respond when the motion of the target aircraft mutates, timely correct the compensation coefficient, and when the motion state is stable, reduce over-response and enhance the system's ability to suppress noise. This flexible adjustment of the smoothing coefficient not only improves the real-time responsiveness of the system but also ensures smoothness and accuracy in the stable state.

[0093] The optimization of the compensation coefficient also relies on a closed-loop feedback mechanism. In each iteration, the system will evaluate whether to trigger the feedback mechanism based on the change of the compensation coefficient. The specific triggering condition is: if the absolute value of the change rate of the compensation coefficient is less than the fourth threshold in N consecutive iterations, it means that the compensation coefficient has tended to be stable, and the system will consider that the current compensation coefficient has reached the optimized state and stop further adjustment.

[0094] Among them, the determination methods of the N value and the fourth threshold are:

[0095] N value: The N value represents the number of consecutive iteration times, which is determined by counting the number of stress fluctuations of the target aircraft within a typical motion cycle and can be set according to the specific motion characteristics of the target aircraft.

[0096] Fourth threshold: The fourth threshold is the maximum allowable amplitude of the change rate of the compensation coefficient, and its value is usually determined by counting the stress fluctuation amplitude of the target aircraft within a typical motion cycle to ensure that the change of the compensation coefficient will not be too large and guarantee the system stability.

[0097] Through the closed-loop feedback mechanism, the system can automatically determine when to stop the iterative optimization of the compensation coefficient, avoid unnecessary calculations and over-adjustment, and ensure the stable operation of the system within an appropriate time. This mechanism can improve the efficiency of the compensation process, reduce the unnecessary calculation burden, and improve the accuracy and response speed of the overall system.

[0098] This method of iteratively optimizing the compensation coefficient can optimize the response speed and noise immunity while ensuring high-precision compensation by combining the exponentially weighted moving average method and the closed-loop feedback mechanism, thus improving the overall performance of the system.

[0099] In a possible implementation manner, the system evaluates the current calibration effect by calculating the root mean square error between the calibrated stress data and the data of the reference measurement device. The root mean square error is a commonly used method to measure the difference between two sets of data and can reflect the overall deviation between the stress data and the measured values of the reference device.

[0100] Specifically, first, ensure that the calculation is carried out at the same time stamp, that is, the time points corresponding to the calibrated stress data and the data of the reference device are the same. This can ensure the comparability of the data and avoid errors caused by time differences. Further, by calculating the root mean square error between the calibrated data and the reference data, an error value is obtained, which is used to evaluate the accuracy of the current neural network output and reflects the degree to which the system needs to be adjusted. By quantifying the difference between the data, it helps the system identify the errors in the current stress calibration process, thereby providing a basis for subsequent weight correction.

[0101] Further, according to the calculated root mean square error, the weight parameters of the neural network will be corrected. The correction amount is not simply linearly adjusted, but determined by a non-linear function.

[0102] Specifically, the root mean square error is mapped through a non - linear function, which determines the amount of weight correction of the neural network. This can more finely control the influence of the error on the weight, avoiding over - adjustment or under - adjustment. The slope value of the non - linear function (i.e., the sensitivity of the error to the correction amount) is dynamically adjusted. By analyzing the convergence speed and stability of the error in the historical correction process, the system can intelligently adjust the slope to make the weight correction more efficient and stable. For example, if the error convergence speed is slow, the slope can be appropriately increased to speed up the correction speed. The non - linear function makes the correction process more flexible and precise, and can adjust the weights of the neural network according to the characteristics of the current error, thereby improving the calibration accuracy and efficiency of the system.

[0103] The update of the weight allocation strategy mainly involves sensor groups with spatial correlation. Spatial correlation reflects the mutual relationship between sensors in terms of physical location and the correlation of their measurement data.

[0104] Specifically, by analyzing the error consistency index within the sensor group, it can be judged whether the error distributions of the sensors within the group are consistent. If the errors of some sensors are small and the errors of other sensors are relatively large, the system will dynamically adjust the weights of these sensors to ensure the accuracy of the overall measurement.

[0105] By clustering and analyzing the historical error distribution patterns, the system can identify which sensor groups have high spatial correlation. Clustering analysis can judge the spatial correlation between sensors based on the error fluctuation trends of the sensors, and then adjust the weights of the sensors within the group.

[0106] The update of the weight allocation strategy enables the system to optimize according to the characteristics of different sensor groups, reducing the impact on the overall calibration effect caused by large errors of individual sensors. In addition, this strategy can effectively increase the weights of sensor groups with strong spatial correlation, further improving the calibration accuracy.

[0107] Dynamically optimize the dynamic stress calibration process of the target aircraft by comparing difference value calculation, non - linear weight correction, and spatial correlation analysis, so as to achieve efficient and accurate real - time feedback calibration.

[0108] In a possible implementation, first, calculate the motion acceleration value of the target aircraft in real - time, and generate a difference sequence by calculating the difference of the acceleration sequence. The difference sequence represents the rate of change of acceleration at adjacent moments. Calculating the acceleration difference sequence helps to capture sudden changes in the motion state, that is, when the motion mode of the target aircraft changes violently.

[0109] The acceleration difference sequence can reflect the instantaneous motion change of the target aircraft. When the acceleration mutates, the value of the difference sequence will increase significantly, providing a data basis for subsequent motion state determination.

[0110] When the number of times the absolute value of the difference sequence continuously exceeds the set fifth threshold reaches the preset value, the system will determine that a sudden change in the motion state has occurred. Specifically, for threshold determination, it is necessary to detect whether there are points in the acceleration difference sequence that continuously exceed a specific threshold, and when the number of these points reaches a certain preset value, it will be determined that a sudden change in the motion state has occurred.

[0111] This process detects the difference sequence one by one, compares its relationship with the fifth threshold, and then determines whether a sudden change in the motion state has occurred. If the change in the difference sequence exceeds the normal fluctuation range (i.e., exceeds the fifth threshold), and this change lasts for a certain period of time (exceeds the preset number of times), it can be considered that a drastic change in the motion state of the target drone has occurred.

[0112] The method for determining the fifth threshold is achieved by statistically calculating the standard deviation of the acceleration difference sequence in the historical motion data of the target drone. The standard deviation measures the amplitude of acceleration changes in the historical motion data. Taking three times the standard deviation as the initial value of the fifth threshold means that under normal circumstances, the fluctuation range of the difference sequence should be within three times the standard deviation.

[0113] With the change of the real-time motion mode, the fifth threshold will be dynamically scaled according to the current motion mode. For example, when the motion of the target drone is more intense, the threshold may be appropriately increased to avoid false judgments caused by drastic changes; when the motion is stable, the threshold may be decreased to more sensitively detect small state mutations.

[0114] The initial value of the fifth threshold is set based on the standard deviation of the historical data, which can ensure that the threshold is reasonable within the fluctuation range of the historical data. At the same time, this threshold is adjusted in real time to enable it to adaptively adapt to the current motion mode of the target drone, thereby ensuring that the system's response to state mutations is both sensitive and not overly sensitive.

[0115] By calculating the acceleration difference sequence in real time, dynamically determining the fifth threshold, and determining mutations based on the threshold, it is possible to accurately and flexibly detect sudden changes in the motion state of the target drone, and thus provide accurate data support for dynamic stress calibration with real-time feedback.

[0116] In a possible implementation, during the execution of steps 1 to 4, all calculation results of dynamically adjusted parameters will be stored in the parameter update queue. To ensure data consistency and effectiveness during parameter update, a double-buffer mechanism is adopted. The double-buffer mechanism means that there are two buffers in the system, one buffer is used to store the data currently in use, and the other buffer is used to store the data being updated. When new data is ready, the system will swap the current buffer and the update buffer, so that the updated data can seamlessly replace the old data while maintaining data consistency.

[0117] The introduction of the dual - buffer mechanism ensures that during the dynamic adjustment process, the system will not experience temporary unavailability or inconsistent states due to data updates. In this way, every time the parameters are adjusted, whether it is the calibration data with real - time feedback or key information such as the acceleration of the target aircraft's movement, they can be synchronized and updated to maintain consistency, thereby improving the stability and reliability of the system.

[0118] During the dynamic stress calibration of the target aircraft, the key determination thresholds are used to judge whether the movement state of the target aircraft has mutated, and then determine whether to trigger the stress calibration process. To ensure that these thresholds always reflect the actual operating state of the target aircraft, the system regularly collects the operating data of the target aircraft and evaluates the accuracy of the thresholds based on the statistical distribution characteristics of these data. If the statistical distribution characteristics of the operating data deviate from the initial calibration value by more than the set sixth threshold, the system will trigger the threshold recalibration process. This process recalculates and adjusts the thresholds to ensure that the target aircraft can accurately reflect the changes in the movement state in real - time under different operating conditions.

[0119] This update mechanism is closely combined with the aforementioned dual - buffer mechanism. The dual - buffer can ensure that during the threshold recalibration, the switching between the old threshold and the new threshold in the parameter update queue will not affect the real - time response of the system. During the threshold recalibration process, the new threshold is calculated and stored in the update queue, and will be switched to the new value during the next update to ensure the stable operation of the entire system.

[0120] Correspondingly, the embodiment of the present invention also provides a target aircraft dynamic stress calibration system with real - time feedback, which is used to run the target aircraft dynamic stress calibration method with real - time feedback described in the embodiment of the present invention, including:

[0121] A dynamic stress sensitive area division module that divides the dynamic stress sensitive area based on the three - dimensional dynamic model of the target aircraft and generates sensor layout planning information. Its input includes the three - dimensional dynamic model of the target aircraft, material property parameters, and motion condition data, and the output is the dynamic stress sensitive area distribution map and sensor layout planning information. This module is connected to the sensor array deployment module and transmits the sensor layout planning information to the subsequent module;

[0122] A sensor array deployment module that, according to the planning information provided by the dynamic stress sensitive area division module, deploys a multi - axis stress sensor array in the specified area according to preset rules. The input is the dynamic stress sensitive area distribution map and sensor layout planning information, and the output is the topological structure data of the sensor array and the sensor position identifier. This module is connected to the signal acquisition module to provide basic data support for signal acquisition;

[0123] Signal acquisition module: The signal acquisition module synchronously acquires the original stress signals output by the multi-axis stress sensor array and dynamically adjusts the signal sampling frequency according to the real-time motion acceleration value of the target aircraft. This module includes a sampling frequency adjustment unit and a signal buffer unit to ensure the efficiency and accuracy of signal acquisition. The inputs include the sensor array topology structure data and the real-time motion acceleration value of the target aircraft, and the outputs are the original stress signal data and the sampling frequency adjustment parameters. The signal acquisition module is connected to the signal processing module to provide the original data for subsequent signal processing;

[0124] Signal processing module: The signal processing module performs adaptive filtering processing on the original stress signals based on the noise feature library to generate preprocessed stress signals. The internal parts of the module are divided into a noise feature library construction unit, a filter selection unit, and a time-frequency analysis unit, which are responsible for noise feature extraction, filter selection, and signal filtering respectively. The inputs are the original stress signal data and the current motion speed value of the target aircraft, and the outputs are the preprocessed stress signal data and the filter type identifier. This module is connected to the dynamic stress compensation module to provide the processed signals for stress compensation;

[0125] Dynamic stress compensation module: The dynamic stress compensation module predicts the theoretical stress value through a pre-trained neural network, calculates the dynamic error distribution matrix, and uses the weighted least squares method to iteratively optimize the compensation coefficients. The module includes a neural network prediction unit, an error calculation unit, a compensation coefficient optimization unit, and a closed-loop feedback trigger unit to achieve precise calibration of stress data. The inputs are the preprocessed stress signal data, the motion acceleration value, and the angular velocity value of the target aircraft, and the outputs are the calibrated stress data, the compensation coefficient optimization parameters, and the closed-loop feedback trigger signal. This module is connected to the calibration verification module and the closed-loop feedback module to form a complete feedback control link;

[0126] Calibration verification module: The calibration verification module compares the calibrated stress data with the measurement data of the reference measurement device and reversely corrects the parameters of the dynamic stress compensation model. The comparison analysis unit, weight correction unit, and weight allocation update unit in the module work together to ensure the continuous optimization of the calibration parameters. The inputs are the calibrated stress data and the reference measurement device data, and the outputs are the corrected neural network weight parameters and the updated weight allocation strategy. The calibration verification module is connected to the dynamic stress compensation module to achieve closed-loop correction of the parameters;

[0127] Closed-loop feedback module. The closed-loop feedback module triggers the closed-loop feedback mechanism according to the change rate of the compensation coefficient, and dynamically adjusts the system parameters to adapt to the real-time motion state of the target aircraft. The mutation detection unit, parameter update queue management unit, and threshold calibration unit in the module are responsible for motion state monitoring, parameter storage, and threshold update respectively. The inputs are the change rate of the compensation coefficient and the real-time motion state data of the target aircraft, and the outputs are the smoothing coefficient adjustment parameter and the updated determination threshold. This module is connected to the dynamic stress compensation module and the signal acquisition module to ensure the real-time optimization of the system parameters;

[0128] System control module. The input is the status feedback information of each module, and the output is the communication instruction and task scheduling signal between each module. The system control module is connected to all modules to ensure the efficient operation of the system.

[0129] This invention covers any alternatives, modifications, equivalent methods, and solutions made within the essence and scope of this invention. To enable the public to have a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments of this invention. However, those skilled in the art can fully understand this invention without the description of these details. Additionally, well-known methods, processes, procedures, components, and circuits, etc., are not described in detail to avoid unnecessary confusion to the essence of this invention.

[0130] The above description is only a preferred implementation manner of this invention. It should be noted that for those of ordinary skill in the art, several improvements and refinements can be made without departing from the principle of this invention, and these improvements and refinements should also be regarded as the protection scope of this invention.

Claims

1. A dynamic stress calibration method for a target drone with real-time feedback, characterized in that, It includes the following steps: Step 1: Divide the dynamic stress sensitive area based on the three-dimensional dynamic model of the target aircraft, and arrange a multi-axis stress sensor array within the dynamic stress sensitive area according to a preset rule, where the preset rule includes a method for determining the sensor spacing and a method for adjusting the density of the stress concentration area; Step 2: Synchronously collect the original stress signals output by the multi-axis stress sensor array, dynamically adjust the signal sampling frequency according to the real-time motion acceleration value of the target aircraft, and perform adaptive filtering processing on the original stress signals based on the noise feature library to generate preprocessed stress signals; Step 3: Input the preprocessed stress signals into the dynamic stress compensation model. The dynamic stress compensation model takes the motion acceleration value and angular velocity value of the target aircraft as input parameters, predicts the theoretical stress values at the positions of each sensor through a pre-trained neural network, calculates the dynamic error distribution matrix of the measured stress values and the theoretical stress values, iteratively optimizes the compensation coefficients using the weighted least squares method, and triggers a closed-loop feedback mechanism according to the change rate of the compensation coefficients; Step 4: In the controlled loading test of the target aircraft, compare the calibrated stress data with the measurement data of the reference measurement device, reversely correct the neural network weight parameters of the dynamic stress compensation model according to the comparison difference value, and update the weight distribution strategy of the weighted least squares method to complete the dynamic iteration of the calibration parameters; The method for determining the sensor spacing in Step 1 includes: a. Collect the stress attenuation data on the principal stress transmission path through the no-load motion test of the target aircraft, and calculate the stress attenuation gradient at different positions; b. Determine the ratio of the equal ratio spacing according to the stress attenuation gradient, so that the adjacent sensor spacing increases according to the ratio as the stress attenuation gradient increases; The method for adjusting the density of the stress concentration area includes: a. Perform finite element simulation on the connection parts and cross-section mutation areas of the target aircraft, and extract the maximum stress gradient value and the average stress gradient value of the areas; b. Take the ratio of the maximum stress gradient value to the average stress gradient value as the multiple of the sensor density increase, and arrange redundant sensors in the area according to the increase multiple; 2. A dynamic stress calibration method for a target drone with real-time feedback according to claim 1, characterized in that, The process of dynamically adjusting the signal sampling frequency in Step 2 includes: Obtain the critical instability acceleration value corresponding to the maximum design speed of the target aircraft through the dynamic calibration test of the target aircraft, and take a preset percentage of the critical instability acceleration value as the acceleration threshold; When the real-time motion acceleration value exceeds the acceleration threshold, adopt the high-frequency sampling mode. The frequency of the high-frequency sampling is an integer multiple of the structural resonance frequency of the target aircraft, and the value of the integer multiple is determined by analyzing the relationship between the resonance frequency and the signal aliasing effect; When the real-time motion acceleration value is lower than or equal to the acceleration threshold, adopt the low-frequency sampling mode. The frequency of the low-frequency sampling is a frequency division value of the high-frequency sampling frequency, and the frequency division value is calculated according to the reciprocal of the product of the current motion speed value of the target aircraft and the structural resonance frequency; 3. A real-time feedback target drone dynamic stress calibration method according to claim 2, characterized in that, The adaptive filtering processing in Step 2 includes: Collect the baseline noise signals at the positions of each sensor in the no-load stationary state of the target aircraft, perform fast Fourier transform on the baseline noise signals, extract the noise spectrum characteristics at the positions of each sensor, and establish the noise feature library; Match the corresponding noise spectrum feature from the noise feature library according to the current sensor position, and select the filter type in the filter bank with the highest correlation with the noise spectrum feature; Perform time-frequency analysis on the original stress signal using the sliding window method. The length of the sliding window is dynamically adjusted according to the decay time of the autocorrelation function of the current motion speed value of the target aircraft. When the energy ratio of the signal within the preset frequency band exceeds the first threshold, it is determined as an effective stress component; otherwise, it is determined as environmental vibration noise and filtered out.

4. A method for calibrating the dynamic stress of a target aircraft with real-time feedback according to claim 3, characterized in that The establishment process of the dynamic stress compensation model in step 3 includes: The training data of the pre-trained neural network is obtained through the calibration test of the target aircraft. The calibration test includes: pasting reference strain gauges on the surface of the target aircraft, and synchronously recording the motion acceleration value, angular velocity value, temperature value of the target aircraft and the strain gauge measurement values at the positions of each sensor under multiple preset motion modes; The calculation process of the dynamic error distribution matrix includes: using the difference between the measured stress value and the theoretical stress value at each sensor position as the initial error value, and performing spatial interpolation according to the stress influence factor at the position of each sensor. The stress influence factor is obtained through finite element simulation calculation; The weight distribution strategy of the weighted least squares method includes: statistically calculating the standard deviation of the historical error data at each sensor position, and assigning a small weight value to the sensor position where the standard deviation exceeds the second threshold.

5. A dynamic stress calibration method for a target drone with real-time feedback according to claim 4, characterized in that, The specific process of iteratively optimizing the compensation coefficient in step 3 includes: Perform dynamic smoothing processing on the compensation coefficient using the exponentially weighted moving average method. The smoothing coefficient of the exponential weighting is adjusted according to the mutation detection result of the motion state of the target aircraft, including: When it is detected that the change rate of the motion acceleration exceeds the third threshold, reduce the smoothing coefficient to accelerate the response speed. When the motion state is stable, increase the smoothing coefficient to improve the anti-noise ability; The triggering condition of the closed-loop feedback mechanism includes: the absolute value of the change rate of the compensation coefficient is less than the fourth threshold in N consecutive iterations. The values of N and the fourth threshold are determined by statistically analyzing the stress fluctuation times and amplitudes within the typical motion cycle of the target aircraft.

6. A dynamic stress calibration method for a target drone with real-time feedback according to claim 5, characterized in that The process of reversely correcting the neural network weight parameters of the dynamic stress compensation model in step 4 includes: The calculation method of the comparison difference value is: calculate the root mean square error between the calibrated stress data and the data of the reference measurement device at the same time stamp; The correction amount of the neural network weight parameters is determined according to the non-linear function of the root mean square error. The slope value of the non-linear function is dynamically adjusted by analyzing the relationship between the error convergence speed and stability in the historical correction process; The update process of the weight distribution strategy includes: for sensor groups with spatial correlation, synchronously adjust their weight values according to the error consistency index of the sensors within the group. The spatial correlation is obtained by clustering and analyzing the historical error distribution pattern.

7. A dynamic stress calibration method for a target drone with real-time feedback according to claim 5, characterized in that, The specific method of the mutation detection of the motion state includes: Real-time calculate the difference sequence of the motion acceleration value of the target aircraft. When the number of times that the absolute value of the difference sequence continuously exceeds the fifth threshold reaches the preset value, it is determined that the motion state has mutated; The method for determining the fifth threshold is as follows: statistically analyze the standard deviation of the acceleration difference sequence in the historical motion data of the target aircraft, take three times the standard deviation as the initial value of the fifth threshold, and dynamically scale it according to the real-time motion mode.

8. A dynamic stress calibration method for a target drone with real-time feedback according to claim 1, characterized in that During the execution of steps 1 to 4: The calculation results of all dynamically adjusted parameters are stored in the parameter update queue, and the parameter update queue adopts a double-buffer mechanism to ensure data consistency during the parameter update process; The update mechanism of the key determination threshold includes: regularly collecting the operation data of the target aircraft, calculating the statistical distribution characteristics of the operation data, and triggering the threshold recalibration process when the distribution characteristics deviate from the initial calibration value by more than the sixth threshold.

9. A dynamic stress calibration system for a target drone with real-time feedback, which is used to run a dynamic stress calibration method for a target drone with real-time feedback according to any one of claims 1-8, and is characterized in that, It includes: A dynamic stress-sensitive area division module that divides the dynamic stress-sensitive area based on the three-dimensional dynamic model of the target aircraft and generates sensor layout planning information. Its inputs include the three-dimensional dynamic model of the target aircraft, material property parameters, and motion condition data, and the outputs are the dynamic stress-sensitive area distribution map and sensor layout planning information. This module is connected to the sensor array deployment module and transmits the sensor layout planning information to the subsequent module; A sensor array deployment module that deploys a multi-axis stress sensor array in a specified area according to the planning information provided by the dynamic stress-sensitive area division module. The inputs are the dynamic stress-sensitive area distribution map and sensor layout planning information, and the outputs are the topological structure data of the sensor array and the sensor position identifier. This module is connected to the signal acquisition module to provide basic data support for signal acquisition; A signal acquisition module that synchronously acquires the original stress signals output by the multi-axis stress sensor array and dynamically adjusts the signal sampling frequency according to the real-time motion acceleration value of the target aircraft. This module includes a sampling frequency adjustment unit and a signal buffer unit to ensure the efficiency and accuracy of signal acquisition. The inputs include the sensor array topological structure data and the real-time motion acceleration value of the target aircraft, and the outputs are the original stress signal data and the sampling frequency adjustment parameters. The signal acquisition module is connected to the signal processing module to provide raw data for subsequent signal processing; A signal processing module that performs adaptive filtering processing on the original stress signals based on the noise feature library to generate preprocessed stress signals. The module is divided into a noise feature library construction unit, a filter selection unit, and a time-frequency analysis unit, which are responsible for noise feature extraction, filter selection, and signal filtering respectively. The inputs are the original stress signal data and the current motion speed value of the target aircraft, and the outputs are the preprocessed stress signal data and the filter type identifier. This module is connected to the dynamic stress compensation module to provide processed signals for stress compensation; Dynamic stress compensation module. The dynamic stress compensation module predicts the theoretical stress value through a pre-trained neural network, calculates the dynamic error distribution matrix, and uses the weighted least squares method to iteratively optimize the compensation coefficient. The module includes a neural network prediction unit, an error calculation unit, a compensation coefficient optimization unit, and a closed-loop feedback trigger unit to achieve precise calibration of stress data. The inputs are the preprocessed stress signal data, the target aircraft's motion acceleration value, and the angular velocity value. The outputs are the calibrated stress data, the optimized compensation coefficient parameters, and the closed-loop feedback trigger signal. This module is connected to the calibration verification module and the closed-loop feedback module to form a complete feedback control link; Calibration verification module. The calibration verification module compares the calibrated stress data with the measurement data of the reference measurement device and reversely corrects the parameters of the dynamic stress compensation model. The comparison analysis unit, weight correction unit, and weight assignment update unit in the module work together to ensure the continuous optimization of the calibration parameters. The inputs are the calibrated stress data and the reference measurement device data. The outputs are the corrected neural network weight parameters and the updated weight assignment strategy. The calibration verification module is connected to the dynamic stress compensation module to achieve closed-loop correction of the parameters; Closed-loop feedback module. The closed-loop feedback module triggers the closed-loop feedback mechanism according to the compensation coefficient change rate and dynamically adjusts the system parameters to adapt to the real-time motion state of the target aircraft. The mutation detection unit, parameter update queue management unit, and threshold calibration unit in the module are responsible for motion state monitoring, parameter storage, and threshold update respectively. The inputs are the compensation coefficient change rate and the target aircraft's real-time motion state data. The outputs are the smoothing coefficient adjustment parameters and the updated decision threshold. This module is connected to the dynamic stress compensation module and the signal acquisition module to ensure the real-time optimization of the system parameters; System control module. The input is the status feedback information of each module, and the output is the communication instructions and task scheduling signals between each module. The system control module is connected to all modules to ensure the efficient operation of the system.

Images