Unmanned aerial vehicle flight log recording method and system based on multi-sensor fusion

Abstract

The invention belongs to the technical field of unmanned aerial vehicle flight logs, and provides an unmanned aerial vehicle flight log recording method based on multi-sensor fusion, which comprises the following steps: embedding double accelerometers at two sides of a flight control mainboard, installing double magnetic compasses, double barometers and a Beidou module, and calibrating the time of each sensor by using a precise time protocol; increasing the sampling frequency when the unmanned aerial vehicle is suspected to be out of control; the data storage adopts an anti-damage design, and the shell of the black box meets a specific standard; when an out-of-control accident of the unmanned aerial vehicle occurs, time axis alignment, analysis processing and cross-sensor correlation analysis are performed on data of each sensor, and finally, intelligent diagnosis and root cause tracing of out-of-control reasons are realized through fusion of multi-source sensor data and a big data model, so that the existing record limitation is changed; and key data and reliable support are provided for out-of-control analysis of the unmanned aerial vehicle.

Description

Technical Field

[0001] The invention belongs to the technical field of unmanned aerial vehicle flight logs, and in particular to a method for recording unmanned aerial vehicle flight logs based on multi-sensor fusion. Background Art

[0002] With the widespread application of drones in various fields, the importance of flight safety and data recording analysis has become increasingly prominent. At present, there are many problems with the flight data recording of most drones. On the one hand, flight data recording is usually integrated with the flight control mainboard. When the drone encounters an accident such as a crash, the flight control is very likely to be damaged by a short circuit or collision, causing the device recording the log to suffer irreversible damage, resulting in the loss of key data and making subsequent fault analysis impossible. On the other hand, existing drone black box recorders have defects in recording functions. The shell is made of injection molding material, which has low structural strength and is not resistant to high temperatures. The shell and circuit board are easily burned in high temperature environments. At the same time, the recorded data is limited, and can only record basic information such as flight time, flight route, take-off and landing coordinates. It is difficult to conduct in-depth technical analysis of complex situations during the flight, such as loss of control, and cannot meet the needs of comprehensive monitoring and accurate analysis of drone flights; Therefore, there is an urgent need for a recording system that is independent of the flight control system and can comprehensively record flight parameters.

[0003] To this end, the present invention provides a method and system for recording a UAV flight log based on multi-sensor fusion. Summary of the invention

[0004] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.

[0005] The technical solution adopted by the present invention to solve the technical problem is: In a first aspect, the present invention provides a method for recording a UAV flight log based on multi-sensor fusion, comprising: Step 1: Arrangement of multi-source sensors; Step 2: Data storage anti-destruction design and drone black box recovery; Step 3: Based on the recovered drone black box, obtain the state data of the drone at the time of the loss of control accident, and analyze and process the state data; the analysis and processing includes but is not limited to: multi-source sensor time axis alignment, dual accelerometer data fusion, heading angle offset compensation, and dual barometer data fusion; Step 4: Use the processed status data to perform cross-sensor correlation analysis; Step 5: Based on the processed status data and cross-sensor correlation analysis, the multi-source sensor data is integrated with the LSTM big data model to achieve intelligent diagnosis and root cause tracing of the cause of the out-of-control.

[0006] In a second aspect, the present invention provides a UAV flight log recording system based on multi-sensor fusion, comprising: Multi-source sensor module: Arrangement of multi-source sensors; Data storage module: data storage anti-destruction design; Data processing module: obtains the state data of the UAV when it loses control, and performs multi-source sensor time axis alignment, dual accelerometer data fusion, heading angle offset compensation and dual barometer data fusion processing on the state data; Data analysis module: performs cross-sensor correlation analysis based on the processed status data; Big data model diagnosis module: Based on the processed status data and cross-sensor correlation analysis; through the fusion of multi-source sensor data and LSTM big data model, intelligent diagnosis and root cause tracing of the cause of out-of-control can be achieved.

[0007] The beneficial effects of the present invention are as follows: 1. Reasonable arrangement of multi-source sensors changes the limitation of existing technology that can only record the flight time, route and coordinate position of drones. The collected data provides key information for in-depth analysis of drone out-of-control events, such as multi-dimensional information such as acceleration, air pressure, and heading during the flight of drones, providing a richer data basis for subsequent analysis; 2. Through the fusion of multi-source sensor data and big data models, intelligent diagnosis and root cause tracing of the cause of loss of control can be achieved. It can extract data features, capture time series dependencies, quantify the intensity of physical contradictions, and provide strong data support and analysis basis for UAV flight loss of control. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The present invention will be further described below in conjunction with the accompanying drawings.

[0009] Figure 1 It is a flowchart of the steps of a method for recording a UAV flight log based on multi-sensor fusion according to the present invention; Figure 2 It is a system module diagram of a UAV flight log recording system based on multi-sensor fusion of the present invention. DETAILED DESCRIPTION

[0010] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods. Example 1

[0011] like Figure 1 As shown, a method for recording a UAV flight log based on multi-sensor fusion according to an embodiment of the present invention comprises the following steps: Step 1: Arrangement of multi-source sensors; First, specifically, the arrangement of the multi-source sensor includes: installing dual accelerometers, dual magnetic compasses, dual barometers, and Beidou modules; Second, specifically, dual accelerometers are respectively embedded at appropriate positions on both sides of the flight control mainboard; the dual accelerometers can be installed on both sides of the flight control mainboard, at a distance of ±8cm from the axis of symmetry of the center of gravity, and a silicone shock-absorbing bracket is used to isolate the motor vibration, eliminate the influence of the motor on the accelerometer, and avoid the same-phase interference of the motor vibration; the parameters of the accelerometer are set as: sampling rate 500Hz, resolution 0.001g; Install a dual magnetic compass. Use a magnetic shielding design for the installation of the dual magnetic compass. Wrap the outer shell of the dual magnetic compass with 1mm thick Mu-Metal alloy. In addition, ensure that the installation of the dual magnetic compass is away from the end of the drone's arm and more than 15cm away from the motor to eliminate the impact of the motor current on the dual magnetic compass. The dual magnetic compass automatically performs hard iron / soft iron compensation before each takeoff to ensure that the heading angle error is <0.5°. A dual barometer is installed, wherein the static pressure pipeline adopts an L-shaped stainless steel conduit with a polytetrafluoroethylene filter installed at the front end to eliminate the influence of turbulence; the inner diameter of the L-shaped stainless steel conduit is 2 mm and the length is 3 cm; The dual barometer integrates a DS18B20 digital temperature sensor, which can correct the air pressure value in real time with an accuracy of ±0.1hPa; Install the Beidou module, where the Beidou module adopts an antenna layout, with a mushroom-shaped omnidirectional antenna on the top and a patch antenna at the bottom, which is anti-blocking and spare, and supports B1C / B2a dual-frequency points; in addition, the Beidou module is connected to the Beidou-3 PPP-B2b signal, and the positioning accuracy can be achieved to the centimeter level; The sensors use the precise time protocol and the Coordinated Universal Time (UTC) provided by the BeiDou module to calibrate the local clock offset, thereby achieving synchronization and alignment of all sensor times. Specifically, the time of the Beidou module, dual barometers, dual magnetic compasses, and dual accelerometers are uniformly calibrated to ensure that each sensor is strictly synchronized in the time dimension, providing a solid time reference guarantee for the precise operation of the system and accurate data collection; At the same time, when the drone is suspected of being out of control, the sampling frequency of each sensor will be increased from the normal level to the maximum value to collect data more intensively and accurately; under normal working conditions, each sensor will sample at the preset frequency, and once suspected signs of loss of control are detected, the sampling frequency will be immediately increased; when a drone out of control event occurs, the system will retain a large amount of data to avoid data loss, and provide detailed data records for subsequent in-depth analysis of the cause of the drone out of control, helping to comprehensively and accurately troubleshoot the root cause of the fault; Step 2: Data storage anti-destruction design and drone black box recovery; Specifically, the black box shell is made of 3mm thick titanium alloy with silicon carbide high temperature resistant coating. At the same time, the structure is designed as a bionic honeycomb structure to meet the MIL-STD-810H standard. In terms of data storage and retention, dual storage media plus shard storage make data storage more complete. Utilize 3DXPoint technology to record all sensor raw data in real time. In addition, use atomic write technology to ensure that a single write operation cannot be interrupted, thus avoiding data loss caused by file system crash. It uses dual-partition rolling storage: Partition A stores current mission data, and Partition B stores historical mission summaries. Partition A is automatically locked when the drone loses control and crashes. At the same time, the drone is equipped with an advanced high-precision scanner that integrates optical and electronic sensing technology to obtain accurate data information in complex environments; when the drone loses control and crashes, the system will automatically switch to passive mode in an instant; S1: In this passive mode, the high-precision scanner starts up quickly and transmits a specific scanning frequency band signal; this signal can minimize environmental noise interference and ensure effective signal transmission and reception; related equipment in the surrounding environment, such as ground base stations with signal response functions and other aircraft with communication modules, will respond to the signal according to the built-in program once receiving the specific scanning frequency band signal; S2: After receiving these reply signals, the scanner immediately enters the data analysis process; by using signal processing algorithms, the key parameters of the reply signal strength, phase, and arrival time are deeply analyzed; For example, when searching for an out-of-control drone, the scanner will first roughly outline the area with stronger signals based on the distribution of received signal strength, and preliminarily determine the area where the drone may exist, because the closer the distance, the higher the signal strength; S3: Then, based on the phase information of the signal, the scanner can determine the relative position relationship of different signal sources; by building a phase difference model, the phase of each reply signal is compared and analyzed to further refine the search area; for example, if the phase difference of the reply signals of two adjacent devices shows a specific pattern, combined with the known distance between the devices, the spatial position relationship between the drone and these devices can be more accurately calculated; S4: The scanner uses the time of flight (ToF) principle to calculate the arrival time of the signal. Knowing the signal propagation speed, the distance the signal propagates can be calculated based on the time difference between the scanning signal and the reply signal received by different devices; Exemplarily, based on the distance data of multiple devices, a triangulation method or a multilateral positioning method is used to construct a spatial coordinate model, thereby determining the specific position of the drone in three-dimensional space; Three points are known. It should be noted that the three points refer to the location coordinates of the device with signal response function and the distance between the uncontrolled crashed drone and these three points. For example, in a two-dimensional plane, suppose there are three points , , ; The point where the drone crashed out of control The distances to these three points are F1, F2, and F3 respectively; according to the distance formula between two points ; ; ; Combine the above calculation formulas to obtain the point of the uncontrolled crashed drone ;Thereby achieving accurate search for the drone’s black box; S5: In terms of search path planning, the scanner adopts a strategy of gradually expanding from near to far. With the point with the strongest signal strength as the center, the scanner scans piece by piece from the inside to the outside in a grid or fan-shaped area. In each scanning area, the scanner flexibly adjusts the scanning accuracy and frequency according to the density of device distribution. For areas with dense devices, the scanning frequency is increased to ensure that no signal feedback is missed. In areas with sparse devices, the scanning range is appropriately expanded to improve search efficiency. From setting the boundaries of the scanning area to detailed inspections point by point, every link has been optimized to ensure that no drone location information is missed; ultimately, through multi-dimensional data fusion and positioning calculation, the drone's position is accurately located, providing accurate coordinates for subsequent recovery of the drone's black box for out-of-control status analysis, while effectively recovering unnecessary property losses that may be caused by the drone's out-of-control; Step 3: Based on the recovered drone black box, obtain the state data of the drone at the time of the loss of control accident, and analyze and process the state data; the analysis and processing includes but is not limited to: multi-source sensor time axis alignment, dual accelerometer data fusion, heading angle offset compensation, and dual barometer data fusion; First, align the time axis information of the accelerometer, magnetic compass, barometer, and Beidou module; and achieve synchronization in hardware; Specifically, find out the start and end time of the data of the accelerometer, magnetic compass, barometer, and Beidou module, build a unified time axis, and use the cubic spline interpolation method on the unified time axis to obtain new interpolation data for the data of each sensor; Specifically, suppose n+1 nodes are known ,in, ;in,

[0012] In each small area Above, cubic spline function is a cubic polynomial, specifically: ;in, ; To determine the coefficients of the cubic spline function, the following conditions must be met: The function value is continuous: ;in, ; The first-order derivative is continuous: ;in, ; The second-order derivative is continuous: ;in, ; Boundary conditions: Usually there are natural boundary conditions ; Next, define the time nodes and time intervals. Let the time nodes of the sensor data be The corresponding measurement value is ;in, ; Time interval ; Construct a system of linear equations to solve the second-order derivative ; According to the continuity of the second-order derivative and the boundary conditions, a linear equation system can be obtained. By solving the linear equation system, we can get ;in, ; Then, calculate the coefficients of the cubic polynomial, specifically, in the interval On the third degree polynomial The coefficients are: ; ; ; ; Then interpolation calculation is performed, for any time point on the same time axis ,like ; The interpolation result is:

[0013] It should be noted that in the above content, the subscript i does not refer to the one-to-one correspondence between different sensors, but is a numbering of the data points of each sensor; Exemplarily, for different sensors, including but not limited to: accelerometer, magnetic compass, barometer, Beidou module, when performing cubic spline interpolation processing on their respective data, there are each independent set of data point sequences with i as the subscript; Interpolate the data of each sensor separately to optimize them in their independent time series, and finally align the data of different sensors to a unified time axis, thus completing the time axis alignment operation; Dual accelerometer data fusion: During the flight of the drone, the use of dual accelerometers can provide additional protection for data reliability; the three-axis difference of the dual accelerometers is set to be less than 0.3g as the consistency standard; From a physical perspective, the value of 0.3g corresponds to the maximum load that a drone can withstand when performing conventional maneuvers. For example, when a drone performs an emergency obstacle avoidance operation, it needs to quickly change its flight direction and speed in a short period of time, and the fuselage will be subjected to a large acceleration. In the high-speed climbing stage, in order to overcome gravity and achieve rapid ascent, the drone will also generate a large acceleration. In these extreme but common flight scenarios, the acceleration experienced by the drone approaches or reaches 0.3g. Therefore, using 0.3g as the limit for judging the consistency of dual accelerometer data has strong practical application value. Real-time acquisition of three-dimensional acceleration data measured by dual accelerometers and , calculate the Euclidean norm difference of the two accelerometer data, and then use the calculated , use the dynamic weight formula to calculate the weight W; introduce the dynamic weight formula: , where e is a natural constant; to deal with the data fusion problem of dual accelerometers; when When the g-score is less than 0.3g, as the value of the Euclidean norm difference gradually decreases, the value of the dynamic weight approaches 1. This means that when the difference in the dual accelerometer measurement data is small, the data of the primary sensor has a higher credibility and weight. Because in this case, the measurement results of the two accelerometers are relatively consistent, and the data of the primary sensor can accurately reflect the actual acceleration of the drone, so the primary sensor data is used first in data fusion. when When the acceleration is greater than or equal to 0.3g, the data of the two sensors are weighted and fused according to the weight W to obtain the final output acceleration data. The formula is: In this way, the accelerometer data can be cleaned and optimized, providing a reliable data basis for the accurate judgment and control of the subsequent UAV flight status; It should be noted that the primary sensor is a pre-designated one from the two sensors; Compensation for heading angle deviation: During the flight of the drone, the motor of the drone will generate a certain electromagnetic field when it is working. The electromagnetic field will affect the measurement of the magnetic compass, causing the heading to deviate. In order to compensate for the heading deviation caused by the motor current, the following dynamic correction method is used: ;in, is the corrected heading angle, The heading angle provided by the Beidou module, is the motor current, is the offset coefficient, and its value is , for every 1A increase in motor current, the heading angle will shift by 0.03 degrees accordingly; Dual barometer data fusion: During drone flight, the use of dual barometers can improve the reliability of altitude measurement. The altitude difference between the dual barometers is set to 3m as the consistency standard. Ideally, the two barometers should have similar responses to the same atmospheric pressure environment, and the altitude measurement difference should be kept within a very small range. If the difference is less than 3m, the altitude measured by the barometer is directly used as the altitude data of the drone; If the difference exceeds 3m, the acceleration integral compensation principle is used to obtain a more accurate UAV altitude, providing reliable data support for flight control and ensuring that the UAV can stably and accurately control its flight altitude in various complex environments; Specifically, use the formula: Calculate the altitude of the drone; where: The altitude value directly measured by the barometer at the current moment; it reflects the altitude calculated based on the current atmospheric pressure. is the fused height value of the previous moment; it serves as the starting reference for calculating the height change at the current moment, indicating Time to present At time , the height change obtained by integrating the vertical velocity; Step 4: Use the processed status data to perform cross-sensor correlation analysis; For example, during the flight of a drone, the altitude and vertical acceleration information provided by the barometer and accelerometer should confirm each other; when the barometer shows an increase in altitude, it means that the drone is flying upward; under normal circumstances, the vertical acceleration should be positive or at least zero to support the upward movement; however, if the barometer shows an increase in altitude, but the vertical acceleration is less than 0, this obviously violates normal physical logic, because the negative vertical acceleration indicates that the drone is accelerating downward, which contradicts the increase in altitude shown by the barometer; This contradiction is caused by the barometer being disturbed by local airflow; in complex meteorological environments, such as valleys and near buildings, local airflow may become turbulent; these turbulent airflows will interfere with the barometer's accurate measurement of atmospheric pressure, causing it to give erroneous altitude information; Once such an acceleration-air pressure contradiction is detected, the system will immediately initiate an emergency altitude correction; the emergency altitude correction mechanism will integrate other sensor data, such as the altitude information from the accelerometer and Beidou module, to re-estimate and correct the current altitude to ensure the accuracy of the drone's flight altitude and avoid flight accidents caused by incorrect altitude information; Step 5: Based on the processed status data and cross-sensor correlation analysis, the multi-source sensor data is integrated with the LSTM big data model to achieve intelligent diagnosis and root cause tracing of the cause of the out-of-control; First, the acceleration data is processed using the 1D-CNN network to extract transient features such as high-frequency vibration and impact. The position and altitude data are processed using the LSTM network to capture the temporal dependencies of the flight trajectory. The fully connected layer processes cross-sensor correlation data to quantify the intensity of physical contradictions such as air pressure-acceleration. The second specific one is data collection and window segmentation: the accelerometer collects data at a sampling frequency of 500 Hz; these continuous data streams are segmented into 1-second time windows; each 1-second window contains 500 sampling points. This segmentation method helps to conduct targeted analysis of the acceleration change characteristics in a short period of time later; Data conversion and frequency domain transformation: Logarithmic transformation is performed on the window data every 1 second. Logarithmic transformation can adjust the dynamic range of the data, highlight subtle changes in the data, and make subsequent analysis more sensitive. Subsequently, the converted data is subjected to fast Fourier transformation to convert the acceleration data in the time domain to the frequency domain, generating spectrum data with a frequency range of 0-250Hz. Data feature extraction: Convolutional neural network (CNN) is used to extract data features. The convolution kernel size is set to 5, and local patterns within 5 consecutive sampling points are detected each time. At the same time, 64 filters are used, each of which can identify a specific typical vibration mode, covering vibration modes corresponding to key physical phenomena that may occur during the flight of the drone, such as motor resonance and structural deformation. Max pooling operation: After completing the convolution operation to obtain a large amount of feature information, the max pooling operation is performed; the function of max pooling is to select the maximum value in each local area as the representative feature of the area; in this way, the most significant features of the data are retained while reducing the data dimension; Input data collection: Beidou coordinates and altitude data measured by the barometer are sampled at intervals of 0.1 seconds; Time window setting: 10 consecutive sampling points are taken as a time window. Since the sampling interval is 0.1 seconds, the duration of this time window is 1 second. Memory unit configuration: 128 memory units are set in the hidden layer of the LSTM network; Application of the forget gate mechanism: The forget gate mechanism of the LSTM network plays a key role here. It can automatically identify and ignore irrelevant interference information; In actual flight, due to factors such as satellite signal interference and atmospheric environment changes, Beidou positioning data may experience momentary abnormal fluctuations. The forget gate mechanism can determine whether these instantaneous changes are interference information based on the previous and subsequent associations and characteristics of the data, and filter them out to improve the accuracy and stability of position sequence analysis. When a drone loses control and crashes, the high-precision scanner will emit a specific scanning frequency band signal, which can stimulate related equipment in the surrounding environment to respond with a signal. By analyzing the response signal, a carpet search can be carried out in the area around the drone to accurately locate the crash site of the drone. Obtain the flight data of the crashed drone. Through the above LSTM-based position sequence analysis process, we can deeply explore the time series characteristics in the drone position data, and provide strong data support and analysis basis for drone flight out of control and abnormal situation warning; The technical solution of this embodiment is: through the reasonable arrangement of multiple sensors, the limitation of the existing technology that can only record the flight time, route and coordinate position of the drone is changed; the data collected by the new solution supplements the key information for the subsequent in-depth analysis of the drone out-of-control incident; at the same time, the anti-destruction design is adopted in data storage to further ensure the integrity of the data when the drone loses control; the drone out-of-control accident status information is comprehensively processed to provide real and reliable data support for the subsequent accurate identification of the cause of the drone out-of-control with the help of a large model.

[0014] Example 2 like Figure 2 As shown, based on Example 1, a drone flight log recording system based on multi-sensor fusion according to an embodiment of the present invention includes the following modules: Multi-source sensor module: Arrangement of multi-source sensors; dual accelerometers are respectively embedded at appropriate positions on both sides of the flight control mainboard; the dual accelerometers can be installed on both sides of the flight control mainboard, at a distance of ±8cm from the symmetry axis of the center of gravity, and a silicone shock-absorbing bracket is used to isolate the motor vibration, eliminate the influence of the motor on the accelerometer, and avoid the same-phase interference of the motor vibration; the parameters of the accelerometer are set as: sampling rate 500Hz, resolution 0.001g; Install a dual magnetic compass. Use a magnetic shielding design for the installation of the dual magnetic compass. Wrap the outer shell of the dual magnetic compass with 1mm thick Mu-Metal alloy. In addition, ensure that the installation of the dual magnetic compass is away from the end of the drone's arm and more than 15cm away from the motor to eliminate the impact of the motor current on the dual magnetic compass. The dual magnetic compass automatically performs hard iron / soft iron compensation before each takeoff to ensure that the heading angle error is <0.5°. A dual barometer is installed, wherein the static pressure pipeline adopts an L-shaped stainless steel conduit with a polytetrafluoroethylene filter installed at the front end to eliminate the influence of turbulence; the inner diameter of the L-shaped stainless steel conduit is 2 mm and the length is 3 cm; The dual barometer integrates a DS18B20 digital temperature sensor, which can correct the air pressure value in real time with an accuracy of ±0.1hPa; Install the Beidou module, where the Beidou module adopts an antenna layout, with a mushroom-shaped omnidirectional antenna on the top and a patch antenna at the bottom, which is anti-blocking and spare, and supports B1C / B2a dual-frequency points; in addition, the Beidou module is connected to the Beidou-3 PPP-B2b signal, and the positioning accuracy can be achieved to the centimeter level; The sensors use the precise time protocol and the Coordinated Universal Time (UTC) provided by the BeiDou module to calibrate the local clock offset, thereby achieving synchronization and alignment of all sensor times. Data storage module: data storage is designed to resist damage; specifically, the black box shell is made of 3mm thick titanium alloy with silicon carbide high temperature resistant coating, and the structure is designed as a bionic honeycomb structure to meet the MIL-STD-810H standard; in terms of data storage and retention, dual storage media plus shard storage make data storage more complete; Data processing module: obtain the state data of the UAV when it loses control, and analyze and process the state data; the analysis and processing includes but is not limited to: multi-source sensor time axis alignment, dual accelerometer data fusion, heading angle offset compensation, dual barometer data fusion; Find the start and end time of the data of the accelerometer, magnetic compass, barometer, and Beidou module, build a unified time axis, and use the cubic spline interpolation method on the unified time axis to obtain new interpolation data for the data of each sensor; Dual accelerometer data fusion: During the flight of the drone, the use of dual accelerometers can provide additional protection for data reliability; the three-axis difference of the dual accelerometers is set to be less than 0.3g as the consistency standard; Real-time acquisition of three-dimensional acceleration data measured by dual accelerometers and , calculate the Euclidean norm difference of the two accelerometer data, and then use the calculated , use the dynamic weight formula to calculate the weight W; introduce the dynamic weight formula: , where e is a natural constant; to deal with the data fusion problem of dual accelerometers; when When the g-score is less than 0.3g, as the value of the Euclidean norm difference gradually decreases, the value of the dynamic weight approaches 1. This means that when the difference in the dual accelerometer measurement data is small, the data of the primary sensor has a higher credibility and weight. Because in this case, the measurement results of the two accelerometers are relatively consistent, and the data of the primary sensor can accurately reflect the actual acceleration of the drone, so the primary sensor data is used first in data fusion. when When the acceleration is greater than or equal to 0.3g, the data of the two sensors are weighted and fused according to the weight W to obtain the final output acceleration data. The formula is: In this way, the accelerometer data can be cleaned and optimized, providing a reliable data basis for the accurate judgment and control of the subsequent UAV flight status; It should be noted that the primary sensor is a pre-designated one from the two sensors; Compensation for heading angle deviation: During the flight of the drone, the electromagnetic field will affect the measurement of the magnetic compass, causing the heading to deviate. In order to compensate for the heading deviation caused by the motor current, the following dynamic correction method is used: ;in, is the corrected heading angle, The heading angle provided by the Beidou module, is the motor current, is the offset coefficient, and its value is , for every 1A increase in motor current, the heading angle will shift by 0.03 degrees accordingly; Dual barometer data fusion: During drone flight, the use of dual barometers can improve the reliability of altitude measurement. The height difference between the dual barometers is set to 3m as the consistency standard. If the difference is less than 3m, the barometer is used directly to measure the altitude as the drone's altitude data. If the difference exceeds 3m, the acceleration integral compensation principle is used to obtain a more accurate UAV altitude, providing reliable data support for flight control and ensuring that the UAV can stably and accurately control its flight altitude in various complex environments; Using the formula: Calculate the altitude of the drone; where: The altitude value directly measured by the barometer at the current moment; it reflects the altitude calculated based on the current atmospheric pressure. is the fused height value of the previous moment; it serves as the starting reference for calculating the height change at the current moment, indicating Time to present At time , the height change obtained by integrating the vertical velocity; Data Analysis Module: The altitude and vertical acceleration information provided by the barometer and accelerometer should confirm each other. When the barometer shows an increase in altitude, it means that the drone is flying upward. Under normal circumstances, the vertical acceleration should be positive or at least zero to support the upward movement. However, if the barometer shows an increase in altitude, but the vertical acceleration is less than 0, this obviously violates normal physical logic, because the negative vertical acceleration indicates that the drone is accelerating downward, which contradicts the increase in altitude shown by the barometer. Once such an acceleration-air pressure contradiction is detected, the system will immediately initiate an emergency altitude correction; the emergency altitude correction mechanism will integrate other sensor data, such as the altitude information from the accelerometer and Beidou module, to re-estimate and correct the current altitude to ensure the accuracy of the drone's flight altitude and avoid flight accidents caused by incorrect altitude information; Big data model diagnosis module: Based on the processed status data and cross-sensor correlation analysis; through the fusion of multi-source sensor data and big data models, intelligent diagnosis and root cause tracing of the cause of out-of-control can be achieved; The acceleration data is processed using the 1D-CNN network to extract transient features such as high-frequency vibration and impact. The position and altitude data are processed using the LSTM network to capture the temporal dependencies of the flight trajectory. The fully connected layer processes cross-sensor correlation data to quantify the intensity of physical contradictions such as air pressure-acceleration. Data collection and window segmentation: The accelerometer collects data at a sampling frequency of 500 Hz; these continuous data streams are segmented into 1-second time windows; each 1-second window contains 500 sampling points. This segmentation method helps to conduct targeted analysis of the acceleration change characteristics in a short period of time; Data conversion and frequency domain transformation: Logarithmic transformation is performed on the window data every 1 second. Logarithmic transformation can adjust the dynamic range of the data, highlight subtle changes in the data, and make subsequent analysis more sensitive. Subsequently, the converted data is subjected to fast Fourier transformation to convert the acceleration data in the time domain to the frequency domain, generating spectrum data with a frequency range of 0-250Hz. Data feature extraction: Convolutional neural network (CNN) is used to extract data features. The convolution kernel size is set to 5, and local patterns within 5 consecutive sampling points are detected each time. At the same time, 64 filters are used, each of which can identify a specific typical vibration mode, covering vibration modes corresponding to key physical phenomena that may occur during the flight of the drone, such as motor resonance and structural deformation. Max pooling operation: After completing the convolution operation to obtain a large amount of feature information, the max pooling operation is performed; the function of max pooling is to select the maximum value in each local area as the representative feature of the area; in this way, the most significant features of the data are retained while reducing the data dimension; Input data collection: Beidou coordinates and altitude data measured by the barometer are sampled at intervals of 0.1 seconds; Time window setting: 10 consecutive sampling points are taken as a time window. Since the sampling interval is 0.1 seconds, the duration of this time window is 1 second. Memory unit configuration: 128 memory units are set in the hidden layer of the LSTM network; Application of the forget gate mechanism: The forget gate mechanism of the LSTM network plays a key role here. It can automatically identify and ignore irrelevant interference information; In actual flight, due to factors such as satellite signal interference and atmospheric environment changes, Beidou positioning data may experience momentary abnormal fluctuations. The forget gate mechanism can determine whether these instantaneous changes are interference information based on the previous and subsequent associations and characteristics of the data, and filter them out to improve the accuracy and stability of position sequence analysis. Through the above LSTM-based position sequence analysis process, we can deeply explore the time series characteristics in the drone position data, and provide strong data support and analysis basis for drone flight loss of control and abnormal situation warning; The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for recording UAV flight logs based on multi-sensor fusion, characterized in that: include: Step 1: Arrangement of multi-source sensors; Step 2: Data storage anti-destruction design and drone black box recovery; Step 3: Based on the recovered drone black box, obtain the state data of the drone at the time of the loss of control accident, and analyze and process the state data; the analysis and processing includes but is not limited to: multi-source sensor time axis alignment, dual accelerometer data fusion, heading angle offset compensation, and dual barometer data fusion; Step 4: Use the processed status data to perform cross-sensor correlation analysis; Step 5: Based on the processed status data and cross-sensor correlation analysis, the multi-source sensor data is integrated with the LSTM big data model to achieve intelligent diagnosis and root cause tracing of the cause of the out-of-control.

2. The method for recording a UAV flight log based on multi-sensor fusion according to claim 1, characterized in that: The arrangement of the multi-sensor is specifically as follows: Install dual accelerometers, dual magnetic compasses, dual barometers, and Beidou module.

3. The method for recording a UAV flight log based on multi-sensor fusion according to claim 1, characterized in that: The data storage anti-corruption design is specifically as follows: The black box shell is made of 3mm thick titanium alloy plus silicon carbide high-temperature resistant coating. It is structurally designed as a bionic honeycomb structure and adopts dual-partition rolling storage: Partition A stores current mission data, and Partition B stores historical mission summaries. Partition A is automatically locked when the drone loses control and crashes.

4. The method for recording a UAV flight log based on multi-sensor fusion according to claim 1, characterized in that: The specific process of aligning the time axis of the multi-source sensors is as follows: Find out the data start and end time of the accelerometer, magnetic compass, barometer, and Beidou module, build a unified time axis, and use the cubic spline interpolation method on the unified time axis to obtain new interpolation data for the data of each sensor.

5. The method for recording a UAV flight log based on multi-sensor fusion according to claim 1, characterized in that: The specific process of dual accelerometer data fusion is as follows: Real-time acquisition of three-dimensional acceleration data measured by dual accelerometers and Calculate the Euclidean norm difference of the two accelerometer data, and then use the calculated The dynamic weight formula is used to calculate the weight W, and the dynamic weight formula is introduced: , where e is a natural constant, is used to deal with the data fusion problem of dual accelerometers.

6. The method for recording a UAV flight log based on multi-sensor fusion according to claim 1, characterized in that: The processing process of the heading angle offset compensation is as follows: When the drone motor is working, it will generate a certain electromagnetic field, which will affect the measurement of the magnetic compass and cause the heading to deviate; In order to compensate for the heading deviation caused by the motor current, a dynamic correction method is used: ;in, is the corrected heading angle, The heading angle provided by the Beidou module, is the motor current, is the offset coefficient, and its value is , for every 1A increase in motor current, the heading angle will shift by 0.03 degrees accordingly.

7. The method for recording a UAV flight log based on multi-sensor fusion according to claim 1, characterized in that: The process of dual barometer data fusion processing is as follows: The use of dual barometers can improve the reliability of altitude measurement; the height difference between the dual barometers is set to 3m as the consistency standard; If the difference is less than 3m, the altitude measured by the barometer is directly used as the altitude data of the drone; If the difference exceeds 3m, the acceleration integral compensation principle is used to obtain the height of the drone using the formula: Calculate the altitude of the drone; where: The altitude value directly measured by the barometer at the current moment; it reflects the altitude calculated based on the current atmospheric pressure. is the fused height value of the previous moment; it serves as the starting reference for calculating the height change at the current moment, indicating the Time to present The change in altitude obtained by integrating the vertical velocity at time .

8. The method for recording a UAV flight log based on multi-sensor fusion according to claim 5, characterized in that: The specific process of processing the data fusion problem of dual accelerometers is as follows: when When the force is less than 0.3g, as the Euclidean norm difference decreases, the value of the dynamic weight approaches 1, and the data of the main sensor has higher credibility and weight; when When the acceleration is greater than or equal to 0.3g, the data of the two sensors are weighted and fused according to the weights to obtain the final output acceleration data; Said The three-dimensional acceleration data measured by the dual accelerometers is used to calculate the Euclidean norm difference between the two accelerometer data. The value of 0.3g corresponds to the maximum load that the drone can withstand when performing conventional maneuvers. The main sensor is a pre-specified one of the two sensors.

9. The method for recording a UAV flight log based on multi-sensor fusion according to claim 1, characterized in that: The process of performing cross-sensor association analysis is as follows: The drone is flying upwards, and the vertical acceleration should be positive or at least zero. If the barometer shows that the altitude is rising, but the vertical acceleration is less than 0, this is inconsistent with the altitude rise shown by the barometer. Immediately initiate emergency altitude correction and recalibrate the current altitude.

10. A UAV flight log recording system based on multi-sensor fusion, characterized by: Includes the following modules: Multi-source sensor module: Arrangement of multi-source sensors; Data storage module: data storage anti-destruction design; Data processing module: obtains the state data of the UAV when it loses control, and performs multi-source sensor time axis alignment, dual accelerometer data fusion, heading angle offset compensation and dual barometer data fusion processing on the state data; Data analysis module: performs cross-sensor correlation analysis based on the processed status data; Big data model diagnosis module: Based on the processed status data and cross-sensor correlation analysis; through the fusion of multi-source sensor data and LSTM big data model, intelligent diagnosis and root cause tracing of the cause of out-of-control can be achieved.

Images