Logistics unmanned aerial vehicle push rod mounting seat loosening fault intelligent monitoring method and system

Abstract

The application relates to the technical field of unmanned aerial vehicles, and provides a method and system for intelligently monitoring loosening faults of a push rod mounting seat of a logistics unmanned aerial vehicle. The method comprises the following steps: collecting vibration data output by a vibration sensor installed on the push rod mounting seat, strain data output by a strain gauge, driving current data of an electric push rod, and action frequency data of the electric push rod, and inputting a pre-trained intelligent identification model; analyzing and processing the input data through the intelligent identification model to identify the bolt loosening degree, structural deformation degree and fatigue damage degree of the push rod mounting seat; determining the mounting state and structural health degree grade of the push rod mounting seat according to the identified bolt loosening degree, structural deformation degree and fatigue damage degree; and outputting alarm information and maintenance suggestions corresponding to the grade according to the determined mounting state and structural health degree grade. The method realizes identification of three types of faults, i.e. bolt loosening, structural deformation and fatigue damage, and comprehensively covers the safety risk points of the push rod mounting seat.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) technology, and in particular to an intelligent monitoring method and system for loosening faults of push rod mounting bases on logistics UAVs. Background Technology

[0002] Logistics drones are widely used in last-mile delivery scenarios, and the reliability of the electric actuator mount, as a core actuator, directly determines flight safety. Current technologies only optimize the structure of the actuator mount and lack effective online monitoring methods. Some monitoring solutions rely solely on a single vibration sensor to detect loose bolts, failing to identify structural deformation and fatigue damage; furthermore, they do not incorporate the electric actuator's own operational data, making them highly susceptible to environmental interference, resulting in low fault identification accuracy and an inability to provide early warnings of potential risks. This can easily lead to actuator detachment during flight, causing safety accidents. Summary of the Invention

[0003] This application provides an intelligent monitoring method and system for loosening faults of push rod mounting bases for logistics drones, aiming to solve the problem that existing technologies only optimize the structural design of push rod mounting bases and lack effective online monitoring methods.

[0004] In a first aspect, embodiments of this application provide an intelligent monitoring method for loose mounting bracket faults of logistics drone push rods, the method comprising: The vibration data, strain data, drive current data, and number of actuations of the electric actuator are collected from the vibration sensor installed on the actuator mounting base. The collected vibration data, strain data, drive current data, and number of actions data are input into a pre-trained intelligent recognition model; the intelligent recognition model analyzes and processes the input data to identify the degree of bolt loosening, structural deformation, and fatigue damage of the push rod mounting base; Based on the identified bolt loosening degree, structural deformation degree, and fatigue damage degree, the installation status and structural health level of the push rod mounting base are determined. Based on the determined installation status and structural health level, corresponding alarm information and maintenance suggestions are output. It also includes: associating and storing each fault identification result, alarm information, maintenance record, and corresponding flight data to form a fault knowledge base, periodically performing statistical analysis on the data in the fault knowledge base, mining the fault occurrence patterns under different flight conditions and different environmental conditions, optimizing the parameters and structure of the intelligent identification model based on the mined fault occurrence patterns, and pushing the optimized intelligent identification model to the UAV's onboard computing unit for online updates.

[0005] In some embodiments, the acquisition of vibration data output by the vibration sensor mounted on the push rod mounting base, strain data output by the strain gauge, drive current data of the electric push rod, and data on the number of actuations of the electric push rod includes: synchronously acquiring vibration data, strain data, drive current data, and actuation count data during the pre-flight self-check phase, during each actuation of the electric push rod in flight, and after the drone stops. The vibration data is acquired at a first preset frequency, the strain data at a second preset frequency, and the drive current data and actuation count data at a third preset frequency. The first preset frequency is higher than the second preset frequency, and the second preset frequency is higher than the third preset frequency.

[0006] In some embodiments, before inputting the collected vibration data, strain data, drive current data, and number of actions data into the pre-trained intelligent recognition model, the method further includes: removing outliers in the collected data that exceed a preset threshold; aligning the remaining data in segments according to each action cycle of the electric actuator; performing moving average filtering on the vibration data to remove environmental noise; performing time synchronization calibration on the data collected by different sensors to unify the time reference; normalizing all data to eliminate dimensional differences; and compensating for airflow interference in the vibration data by combining the real-time attitude data of the UAV.

[0007] In some embodiments, before inputting the collected vibration data, strain data, drive current data, and number of actions data into the pre-trained intelligent recognition model, the method further includes: constructing a training dataset containing corresponding data for normal state, different bolt loosening degrees, different structural deformation degrees, different fatigue damage degrees, different flight loads, and different ambient temperatures; dividing the training dataset into a training subset, a validation subset, and a test subset; constructing a multi-input multi-output neural network model; training the neural network model using the training subset; adjusting the parameters of the neural network model using the validation subset; verifying the accuracy of the neural network model using the test subset; pre-training the neural network model using historical fault data of similar UAVs and then fine-tuning it using data from the current UAV; and periodically performing online incremental training of the neural network model using newly collected labeled data.

[0008] In some embodiments, the step of analyzing and processing the input data through an intelligent recognition model to identify the bolt loosening degree, structural deformation degree, and fatigue damage degree of the push rod mounting seat includes: the intelligent recognition model extracting multi-dimensional features from the preprocessed input data, fusing the extracted vibration features, strain features, current features, and action frequency features to obtain a comprehensive feature vector; inputting the comprehensive feature vector into three parallel classification branches, which output probability distributions for bolt loosening degree, structural deformation degree, and fatigue damage degree respectively; taking the category with the highest probability in each probability distribution as the identification result of the corresponding fault degree; calculating the confidence level of each identification result; and triggering a secondary data acquisition and identification process when the confidence level is lower than a preset threshold.

[0009] In some embodiments, determining the installation status and structural health level of the push rod mounting base based on the identified bolt loosening degree, structural deformation degree, and fatigue damage degree includes: pre-setting the bolt loosening degree, structural deformation degree, and fatigue damage degree and their corresponding weighting coefficients; multiplying the identified bolt loosening degree, structural deformation degree, and fatigue damage degree values ​​by their respective weighting coefficients and summing them to obtain a comprehensive health score; classifying the corresponding health level according to the preset interval in which the comprehensive health score falls; and directly determining the health level as the highest risk level when any fault degree reaches a preset threshold.

[0010] In some embodiments, the step of outputting alarm information and maintenance suggestions of corresponding levels according to the determined installation status and structural health level includes: outputting no alarm information and normal operation suggestions when the health level is normal; outputting a level 1 alarm information and maintenance suggestions for pre-flight checks when the health level is slightly abnormal; outputting a level 2 alarm information and maintenance suggestions for immediately stopping flight and tightening components when the health level is moderately abnormal; and outputting a level 3 alarm information and maintenance suggestions for immediately performing an emergency landing when the health level is severely abnormal. All alarm information and maintenance suggestions are simultaneously sent to the UAV ground control station and the operation and maintenance management platform.

[0011] In some embodiments, after outputting alarm information and maintenance recommendations corresponding to the determined installation status and structural health level, the method further includes: collecting historical fault identification results and corresponding maintenance records to establish a fault prediction dataset; training a remaining service life prediction model based on the fault prediction dataset; inputting the current installation status and structural health data into the remaining service life prediction model to obtain the remaining service life prediction value of the push rod mounting base; generating a preventive maintenance plan containing maintenance time and maintenance content based on the remaining service life prediction value; and sending the preventive maintenance plan to the operation and maintenance management platform.

[0012] Secondly, this application provides an intelligent monitoring system for loose mounting bracket faults of logistics drone push rods, the system comprising: The data acquisition unit is used to acquire vibration data output by the vibration sensor installed on the push rod mounting base, strain data output by the strain gauge, drive current data of the electric push rod, and data on the number of times the electric push rod has been activated. The model input unit is used to input the collected vibration data, strain data, drive current data, and number of actions into the pre-trained intelligent recognition model; the intelligent recognition model analyzes and processes the input data to identify the degree of bolt loosening, structural deformation, and fatigue damage of the push rod mounting base; The degree recognition unit is used to determine the installation status and structural health level of the push rod mounting base based on the identified bolt loosening degree, structural deformation degree, and fatigue damage degree; based on the determined installation status and structural health level, it outputs alarm information and maintenance suggestions of the corresponding level.

[0013] This application significantly improves the accuracy and anti-interference capability of fault identification by employing multi-source data fusion technology, avoiding false alarms and missed alarms. It simultaneously identifies three types of faults: bolt loosening, structural deformation, and fatigue damage, comprehensively covering the safety risk points of the push rod mounting base. By monitoring the installation status and structural health in real time, it outputs tiered alarms and maintenance suggestions, effectively preventing push rod detachment accidents during flight. It eliminates the need for regular manual inspections, reducing the operation and maintenance costs of logistics drones and improving the reliability and continuity of system operation. Furthermore, it includes: linking and storing each fault identification result, alarm information, maintenance record, and corresponding flight data to form a fault knowledge base; periodically performing statistical analysis on the data in the fault knowledge base to uncover fault occurrence patterns under different flight conditions and environmental conditions; optimizing the parameters and structure of the intelligent identification model based on the discovered fault occurrence patterns; and pushing the optimized intelligent identification model to the drone's onboard computing unit for online updates.

[0014] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic flowchart illustrating the steps of an intelligent monitoring method for loose push rod mounting bracket faults of a logistics drone, provided in one embodiment of this application. Figure 2 This is a first-view structural schematic diagram of a push rod mounting base for a logistics drone provided in an embodiment of this application; Figure 3 This is a second-view structural schematic diagram of a push rod mounting base for a logistics drone provided in one embodiment of this application; Figure 4 This is a schematic block diagram of the structure of an intelligent monitoring system for loose push rod mounting bracket faults of a logistics drone provided in one embodiment of this application; Figure 5 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application.

[0017] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0020] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0021] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0022] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0023] Logistics drones utilize electric actuators to perform core functions such as cargo compartment opening and closing, and landing gear retraction and extension. The actuator mount is a critical load-bearing component connecting the electric actuator to the drone's fuselage. Existing actuator mounts often rely on empirical design or single static strength checks, failing to comprehensively consider the alternating loads during actuator push-pull processes, vibrations and impacts during drone flight, and complex stress conditions under different mounting postures. This results in mounts commonly exhibiting defects such as over-design leading to increased weight or insufficient strength, making them prone to fatigue damage. Furthermore, existing mount optimization methods often employ single-objective optimization, failing to achieve an optimal balance between lightweight design and high rigidity, thus hindering the long endurance and high reliability requirements of logistics drones.

[0024] To solve the above problem, please refer to Figure 1 This application provides an intelligent monitoring method for loose push rod mounting bracket faults in logistics drones, applied to computer equipment. The computer equipment can be deployed on a single server or server cluster. It can also be deployed on handheld terminals, laptops, wearable devices, or robots, etc.

[0025] The push rod mounting structure addressed in this invention is as follows: Figure 2 and Figure 3 As shown, the push rod mounting base is integrally bent from stainless steel sheet, including a U-shaped upright plate and a rectangular base plate. The base plate has three sets of equidistant mounting through holes for fixing the mounting base to the UAV fuselage with bolts; the upper part of the upright plate has push rod mounting through holes for hinged connection of the end of the electric push rod; a rounded corner transition structure is provided at the connection between the upright plate and the base plate to reduce stress concentration; boss limiting structures are provided at both ends of the base plate for quick positioning and limiting the lateral displacement of the mounting base during installation.

[0026] The provided intelligent monitoring method for loose mounting bracket faults on logistics drones includes steps S101 to S103. Details are as follows: Step S101. Collect vibration data output by the vibration sensor installed on the push rod mounting base, strain data output by the strain gauge, drive current data of the electric push rod, and data on the number of times the electric push rod is activated.

[0027] Specifically, this step simultaneously collects four types of complementary data, covering the structural status of the mounting base and the operational status of the electric actuator, providing comprehensive input information for subsequent fault identification.

[0028] Vibration data acquisition involves installing a triaxial vibration sensor at the top center of the U-shaped vertical plate of the push rod mounting base to collect vibration acceleration data of the mounting base in the X, Y, and Z directions. This position is closest to the push rod hinge point and can most sensitively reflect vibration anomalies caused by bolt loosening.

[0029] Strain data acquisition involves attaching a resistance strain gauge to each of the two rounded transition points at the connection between the push rod mounting plate and the base plate, and collecting dynamic strain data at these locations. The rounded transition points are stress concentration areas of the mounting base, and can reflect structural deformation and fatigue damage earliest.

[0030] Drive current data acquisition involves connecting a current sensor in series in the power supply circuit of the electric linear actuator to collect real-time current data during its operation. Changes in the drive current reflect changes in the actuator load and indirectly indicate the loosening or deformation of the mounting base.

[0031] The data on the number of actuations is collected through the controller of the electric linear actuator, which acquires the cumulative number of actuations and the start and end times of each actuation. The number of actuations is a core indicator for assessing the degree of fatigue damage to the mounting base.

[0032] Step S102. Input the collected vibration data, strain data, drive current data, and number of actions data into the pre-trained intelligent recognition model; analyze and process the input data through the intelligent recognition model to identify the degree of bolt loosening, structural deformation, and fatigue damage of the push rod mounting base.

[0033] Specifically, this step involves inputting the collected multi-source data into a pre-trained intelligent recognition model. Through the model's automatic analysis and processing, it outputs the degree recognition results for the three types of faults.

[0034] The pre-processed vibration data, strain data, drive current data, and number of actions data are input into the intelligent recognition model according to a preset format.

[0035] The intelligent recognition model first extracts multi-dimensional features from the input data, namely the time and frequency domain features of the vibration data, the dynamic features of the strain data, the change features of the driving current, and the number of actions. The extracted features are then fused to form a comprehensive feature vector that can fully reflect the state of the mounting base.

[0036] Fault Classification: The comprehensive feature vector is input into three parallel classification branches. These branches classify and identify the degree of bolt loosening, structural deformation, and fatigue damage, respectively, and output the probability distribution for each category. The category with the highest probability in each probability distribution is taken as the final identification result for the corresponding fault degree, and the confidence level of this result is also output.

[0037] Step S103. Based on the identified bolt loosening degree, structural deformation degree, and fatigue damage degree, determine the installation status and structural health level of the push rod mounting base; based on the determined installation status and structural health level, output corresponding alarm information and maintenance suggestions, and also include: associating and storing each fault identification result, alarm information, maintenance record, and corresponding flight data to form a fault knowledge base, periodically performing statistical analysis on the data in the fault knowledge base, mining the fault occurrence patterns under different flight conditions and different environmental conditions, optimizing the parameters and structure of the intelligent identification model based on the mined fault occurrence patterns, and pushing the optimized intelligent identification model to the UAV onboard computing unit to complete online updates.

[0038] Specifically, this step comprehensively assesses the overall health status of the mounting base based on the three types of fault identification results output by the model, and outputs alarm information and maintenance suggestions of the corresponding level.

[0039] Based on the pre-set weighting coefficients, the values ​​of the three types of fault severity are weighted and summed to obtain the overall health score of the mounting base.

[0040] Based on the preset range of the comprehensive health score, the health status of the mounting base is divided into different levels. A single fault threshold is also set; when any type of fault reaches the threshold, it is directly classified as the highest risk level.

[0041] Based on the determined health level, corresponding alarm information and targeted maintenance suggestions are output. The alarm information is also sent to the UAV ground control station and the operation and maintenance management platform to ensure that relevant personnel can obtain fault information in a timely manner.

[0042] By associating and storing each fault identification result, alarm information, maintenance record and corresponding flight data to form a fault knowledge base, the data in the fault knowledge base is statistically analyzed regularly to explore the fault occurrence patterns under different flight conditions and environmental conditions. Based on the discovered fault occurrence patterns, the parameters and structure of the intelligent identification model are optimized, and the optimized intelligent identification model is pushed to the UAV's onboard computing unit to complete online updates.

[0043] The fault knowledge base is constructed by including a unique fault ID, fault occurrence time, UAV number, flight conditions (flight altitude, flight speed, payload weight), environmental conditions (ambient temperature, ambient humidity, wind speed), fault identification result, alarm level, maintenance measures taken, and post-maintenance status for each fault record.

[0044] Statistical analysis and pattern mining involve conducting monthly statistical analysis on the data in the fault knowledge base and using association rule mining algorithms to uncover the correlations between different flight conditions, environmental conditions, fault types, and fault occurrence rates.

[0045] Model optimization adjusts the fault identification threshold of the intelligent identification model under different working conditions and environmental conditions based on the fault patterns obtained from mining. For working conditions with a high fault occurrence rate, training data for the corresponding working conditions is added, and the model is retrained to improve the model's identification accuracy under that working condition.

[0046] Online updates are performed via the drone's 4G / 5G communication module, pushing optimized model parameters to the drone's onboard computing unit. The model update is completed while the drone is stationary, without affecting its normal flight operations. After the update, automatic model verification is performed to ensure the model is functioning correctly.

[0047] In some embodiments, the acquisition of vibration data output by the vibration sensor mounted on the push rod mounting base, strain data output by the strain gauge, drive current data of the electric push rod, and data on the number of actuations of the electric push rod includes: synchronously acquiring vibration data, strain data, drive current data, and actuation count data during the pre-flight self-check phase, during each actuation of the electric push rod in flight, and after the drone stops. The vibration data is acquired at a first preset frequency, the strain data at a second preset frequency, and the drive current data and actuation count data at a third preset frequency. The first preset frequency is higher than the second preset frequency, and the second preset frequency is higher than the third preset frequency.

[0048] Vibration data, strain data, drive current data, and number of actions data are collected synchronously during the self-check phase before takeoff, during each action of the electric push rod in flight, and during the shutdown phase after flight.

[0049] Data is continuously collected for 30 seconds from the moment the drone is powered on until the propellers start. During this period, the drone is stationary, allowing for the collection of baseline state data free from flight interference, which is used to calibrate sensors and establish a reference state.

[0050] The time interval from when the electric linear actuator receives the drive signal to when the action is completed (lasting 2 seconds) is the period during which the electric linear actuator is in working condition, the mounting base bears dynamic loads, and this phase is the most effective time to detect loose bolts and structural deformation.

[0051] Data was continuously collected for 30 seconds from the moment the propellers completely stopped until the drone was powered off. This phase was used to compare the changes in the drone's condition before and after flight, and to identify cumulative damage that occurred during the flight.

[0052] Meanwhile, differentiated acquisition frequencies are set according to the characteristics of different types of data: vibration data is acquired at a first preset frequency of 1000 Hz to capture high-frequency vibration signals; strain data is acquired at a second preset frequency of 100 Hz to accurately reflect the dynamic changes in strain; and drive current data and actuation frequency data are acquired at a third preset frequency of 10 Hz to meet data accuracy requirements while reducing computational resource consumption. The first preset frequency is higher than the second preset frequency, and the second preset frequency is higher than the third preset frequency.

[0053] In some embodiments, before inputting the collected vibration data, strain data, drive current data, and number of actions data into the pre-trained intelligent recognition model, the method further includes: removing outliers in the collected data that exceed a preset threshold; aligning the remaining data in segments according to each action cycle of the electric actuator; performing moving average filtering on the vibration data to remove environmental noise; performing time synchronization calibration on the data collected by different sensors to unify the time reference; normalizing all data to eliminate dimensional differences; and compensating for airflow interference in the vibration data by combining the real-time attitude data of the UAV.

[0054] By removing outliers exceeding a preset threshold from the collected data, aligning the remaining data in segments according to each action cycle of the electric actuator, applying a moving average filter to the vibration data to remove environmental noise, calibrating the data collected by different sensors to unify the time reference, normalizing all data to eliminate dimensional differences, and compensating for airflow interference in the vibration data by combining real-time attitude data of the UAV.

[0055] Outliers are identified using the 3σ principle. The mean and standard deviation of each data sequence are calculated, and data that exceed three times the standard deviation of the mean are marked as outliers and deleted.

[0056] Using the rising edge of the electric actuator drive signal as the starting point of each action cycle, all data are segmented according to the action cycle to ensure that different types of data within the same cycle are aligned in time.

[0057] The vibration data were filtered using a moving average with a window size of 10 sampling points to smooth high-frequency noise and retain valid vibration signals.

[0058] All sensors and devices use a unified timestamp synchronized with GPS to time-synchronize and calibrate the collected data, ensuring that the time error of data from different sources is less than 1 millisecond.

[0059] The min-max normalization method is used to map all data to the range of 0 to 1, eliminating the influence of different units on model training and inference.

[0060] Based on the airspeed, pitch angle, and roll angle data of the UAV, a mapping relationship between airflow interference and vibration data is established. The airflow interference component is subtracted from the original vibration data to improve the signal-to-noise ratio of the vibration data.

[0061] In some embodiments, before inputting the collected vibration data, strain data, drive current data, and number of actions data into the pre-trained intelligent recognition model, the method further includes: constructing a training dataset containing corresponding data for normal state, different bolt loosening degrees, different structural deformation degrees, different fatigue damage degrees, different flight loads, and different ambient temperatures; dividing the training dataset into a training subset, a validation subset, and a test subset; constructing a multi-input multi-output neural network model; training the neural network model using the training subset; adjusting the parameters of the neural network model using the validation subset; verifying the accuracy of the neural network model using the test subset; pre-training the neural network model using historical fault data of similar UAVs and then fine-tuning it using data from the current UAV; and periodically performing online incremental training of the neural network model using newly collected labeled data.

[0062] By constructing a training dataset containing data corresponding to normal conditions, different bolt loosening degrees, different structural deformation degrees, different fatigue damage degrees, different flight loads, and different ambient temperatures, the training dataset is divided into training subsets, validation subsets, and test subsets. A multi-input multi-output neural network model is constructed. The neural network model is trained using the training subset, the parameters of the neural network model are adjusted using the validation subset, and the accuracy of the neural network model is verified using the test subset. The neural network model is pre-trained using historical fault data of similar UAVs and then fine-tuned using data from this UAV. The neural network model is also periodically incrementally trained online using newly collected labeled data.

[0063] 1000 sets of normal state data were collected; bolt loosening was classified into four levels: no loosening, slight loosening, moderate loosening, and severe loosening, with 500 sets of data collected for each level; structural deformation was classified into three levels: no deformation, slight deformation, and severe deformation, with 500 sets of data collected for each level; fatigue damage was classified into three levels: no damage, slight damage, and severe damage, with 500 sets of data collected for each level. Simultaneously, it covered five levels of flight load from no load to full load, and seven levels of ambient temperature from -20 degrees Celsius to 50 degrees Celsius.

[0064] The training dataset was divided into a training subset, a validation subset, and a test subset in a ratio of 7:2:1.

[0065] A multi-input multi-output (MIMO) model combining convolutional neural networks (CNNs) and fully connected networks is constructed. A two-layer CNN is used to extract temporal features from vibration and strain data, while a two-layer fully connected network is used to extract features from driving current and action frequency data. All features are concatenated and fused before being input into three parallel fully connected classification branches, each corresponding to the identification of a specific type of fault.

[0066] Training was performed using the Adam optimizer with a learning rate of 0.001, a batch size of 32, and 100 training epochs. Training was stopped early when the validation set loss stopped decreasing for 10 consecutive epochs.

[0067] The model was pre-trained using 100,000 historical fault data points from 100 similar drones, and then fine-tuned using 1,000 normal state data points from the drone itself to adapt the model to the characteristics of the drone.

[0068] Every 100 hours of flight, the model is incrementally trained using 500 newly collected, manually labeled data points to continuously improve its recognition accuracy.

[0069] In some embodiments, the step of analyzing and processing the input data through an intelligent recognition model to identify the bolt loosening degree, structural deformation degree, and fatigue damage degree of the push rod mounting seat includes: the intelligent recognition model extracting multi-dimensional features from the preprocessed input data, fusing the extracted vibration features, strain features, current features, and action frequency features to obtain a comprehensive feature vector; inputting the comprehensive feature vector into three parallel classification branches, which output probability distributions for bolt loosening degree, structural deformation degree, and fatigue damage degree respectively; taking the category with the highest probability in each probability distribution as the identification result of the corresponding fault degree; calculating the confidence level of each identification result; and triggering a secondary data acquisition and identification process when the confidence level is lower than a preset threshold.

[0070] The intelligent recognition model extracts multi-dimensional features from the preprocessed input data. It fuses the extracted vibration features, strain features, current features, and number of actions features to obtain a comprehensive feature vector. The comprehensive feature vector is then input into three parallel classification branches. The three classification branches output probability distributions for bolt loosening degree, structural deformation degree, and fatigue damage degree, respectively. The category with the highest probability in each probability distribution is taken as the recognition result of the corresponding fault degree. The confidence level of each recognition result is calculated. When the confidence level is lower than a preset threshold, a secondary data acquisition and recognition process is triggered.

[0071] Multi-dimensional feature extraction includes: Vibration characteristics: Extract time-domain features including mean, variance, peak value, kurtosis, and margin factor; extract frequency-domain features including dominant frequency, secondary dominant frequency, and amplitude of each harmonic component.

[0072] Strain characteristics: Extract the maximum strain value, minimum strain value, strain change rate, strain cycle number, and average strain value.

[0073] Current characteristics: Extract the peak starting current, average operating current, current fluctuation coefficient, current rise time, and current fall time.

[0074] Action frequency characteristics: Extract cumulative action frequency, action frequency per unit time, and average interval time of the most recent 100 actions.

[0075] A concatenation and fusion approach is used to sequentially concatenate all extracted features into a one-dimensional comprehensive feature vector. All three classification branches employ a three-layer fully connected network, with the output layer of each branch using the Softmax activation function to output the probability distribution of the corresponding fault category. The maximum probability value in each probability distribution is taken as the confidence level of the identification result. A preset confidence threshold of 0.8 is used. When the confidence level of any identification result falls below 0.8, a second data acquisition is triggered for 5 seconds, followed by a re-identification. If the confidence level of the second identification is still below 0.8, an unknown fault alarm is output, prompting maintenance personnel to conduct a manual inspection.

[0076] In some embodiments, determining the installation status and structural health level of the push rod mounting base based on the identified bolt loosening degree, structural deformation degree, and fatigue damage degree includes: pre-setting the bolt loosening degree, structural deformation degree, and fatigue damage degree and their corresponding weighting coefficients; multiplying the identified bolt loosening degree, structural deformation degree, and fatigue damage degree values ​​by their respective weighting coefficients and summing them to obtain a comprehensive health score; classifying the corresponding health level according to the preset interval in which the comprehensive health score falls; and directly determining the health level as the highest risk level when any fault degree reaches a preset threshold.

[0077] By pre-setting the weight coefficients corresponding to the degree of bolt loosening, structural deformation and fatigue damage, the values ​​of the three fault degrees identified are multiplied by their respective weight coefficients and then summed to obtain a comprehensive health score. The corresponding health level is divided according to the preset interval of the comprehensive health score. When any fault degree reaches the preset threshold, the health level is directly determined to be the highest risk level.

[0078] The weighting coefficients are set according to the degree of impact of the fault on flight safety. The weighting coefficient for the degree of bolt loosening is set to 0.5, the weighting coefficient for the degree of structural deformation is set to 0.3, and the weighting coefficient for the degree of fatigue damage is set to 0.2.

[0079] The degree of failure is quantified by assigning four levels of bolt loosening degree to values ​​of 0, 1, 2, and 3 respectively; three levels of structural deformation degree to values ​​of 0, 2, and 3 respectively; and three levels of fatigue damage degree to values ​​of 0, 2, and 3 respectively.

[0080] The overall health score is calculated as follows: Overall health score = 100 - (bolt looseness value × 0.5 + structural deformation value × 0.3 + fatigue damage value × 0.2) × 25. The score range is from 0 to 100, and the higher the score, the better the health status.

[0081] Health level classification: 90-100 points: Normal condition; 70-89 points: Minor abnormality; 40-69 points: Moderately abnormal; 0-39 points: Severe abnormality; Critical value determination: When the degree of bolt loosening reaches severe loosening, the degree of structural deformation reaches severe deformation, or the degree of fatigue damage reaches severe damage, regardless of the comprehensive health score, the health level is directly determined to be severely abnormal.

[0082] In some embodiments, the step of outputting alarm information and maintenance suggestions of corresponding levels according to the determined installation status and structural health level includes: outputting no alarm information and normal operation suggestions when the health level is normal; outputting a level 1 alarm information and maintenance suggestions for pre-flight checks when the health level is slightly abnormal; outputting a level 2 alarm information and maintenance suggestions for immediately stopping flight and tightening components when the health level is moderately abnormal; and outputting a level 3 alarm information and maintenance suggestions for immediately performing an emergency landing when the health level is severely abnormal. All alarm information and maintenance suggestions are simultaneously sent to the UAV ground control station and the operation and maintenance management platform.

[0083] When the health level is normal, no alarm information and normal operation suggestions are output. When the health level is slightly abnormal, a level 1 alarm information and maintenance suggestions for pre-flight checks are output. When the health level is moderately abnormal, a level 2 alarm information and maintenance suggestions for immediately stopping flight and tightening components are output. When the health level is severely abnormal, a level 3 alarm information and maintenance suggestions for immediately performing an emergency landing are output. All alarm information and maintenance suggestions are sent to the UAV ground control station and operation and maintenance management platform at the same time.

[0084] Level 1 Alarm: The ground control station displays a yellow warning message, which includes the fault type, fault severity and recommended maintenance time; the operation and maintenance management platform sends an SMS notification to the operation and maintenance personnel.

[0085] Level 2 Alarm: The ground control station displays an orange warning message and issues a voice alarm; the drone automatically cancels subsequent flight missions and executes the return-to-home procedure; the operation and maintenance management platform sends a dual notification to the operation and maintenance manager via telephone and SMS.

[0086] Level 3 Alarm: The ground control station displays a red warning and issues a continuous, urgent voice alarm; the drone immediately aborts its current mission, searches for the nearest safe landing point, and executes an emergency landing procedure; the operation and maintenance management platform sends an emergency notification to all relevant personnel.

[0087] Maintenance recommendations include: Normal operation recommendation: Continue flight missions and perform checks according to the routine maintenance plan.

[0088] Recommended pre-flight inspection: Before the next flight, use a torque wrench to check the torque of all mounting bolts to ensure they meet the specified requirements.

[0089] Immediately stop flight and tighten components. Recommendation: After the drone lands, immediately calibrate the torque of all mounting bolts, check for any obvious deformation of the mounting base, and only continue flying after confirming that there are no abnormalities.

[0090] Immediately implement emergency landing recommendations: After the drone makes an emergency landing, it is forbidden to take off again. Remove the mounting base for a comprehensive inspection and replace it with a new mounting base if necessary.

[0091] In some embodiments, after outputting alarm information and maintenance recommendations corresponding to the determined installation status and structural health level, the method further includes: collecting historical fault identification results and corresponding maintenance records to establish a fault prediction dataset; training a remaining service life prediction model based on the fault prediction dataset; inputting the current installation status and structural health data into the remaining service life prediction model to obtain the remaining service life prediction value of the push rod mounting base; generating a preventive maintenance plan containing maintenance time and maintenance content based on the remaining service life prediction value; and sending the preventive maintenance plan to the operation and maintenance management platform.

[0092] A fault prediction dataset is established by collecting historical fault identification results and corresponding maintenance records. A remaining service life prediction model is trained based on the fault prediction dataset. The current installation status and structural health data are input into the remaining service life prediction model to obtain the remaining service life prediction value of the push rod mounting base. A preventive maintenance plan containing maintenance time and maintenance content is generated based on the remaining service life prediction value and sent to the operation and maintenance management platform.

[0093] The fault prediction dataset is constructed by including the following for each data record: mount number, cumulative flight time, cumulative number of actions, average flight load, historical fault records, maintenance records, and actual service life.

[0094] The remaining service life prediction model adopts a long short-term memory network model. The input features include the current bolt loosening degree, structural deformation degree, fatigue damage degree, cumulative flight time, cumulative number of actions, and average flight load. The output is the predicted value of the remaining service life in flight hours.

[0095] The preventative maintenance plan generation includes the following steps: When the predicted remaining service life is greater than 100 flight hours, a routine maintenance plan is generated, and maintenance is performed according to the original plan. When the predicted remaining service life is between 50 and 100 flight hours, a Level 1 preventative maintenance plan is generated, recommending a focused inspection of the mount during the next scheduled maintenance. When the predicted remaining service life is between 20 and 50 flight hours, a Level 2 preventative maintenance plan is generated, recommending a comprehensive inspection and maintenance of the mount within 3 flight days. When the predicted remaining service life is less than 20 flight hours, a Level 3 preventative maintenance plan is generated, recommending immediate cessation of flight operations and replacement of the mount with a new one.

[0096] Please see Figure 4 As shown, Figure 4 This is a schematic diagram of the intelligent monitoring system 200 for loose mounting brackets of logistics drones provided in this application embodiment. The intelligent monitoring system 200 is used to execute the steps of the intelligent monitoring method for loose mounting brackets of logistics drones shown in the above embodiments. The intelligent monitoring system 200 can be a single server or a server cluster, or it can be a terminal, such as a handheld terminal, a laptop computer, a wearable device, or a robot.

[0097] like Figure 4 As shown, the intelligent monitoring system 200 for loosening faults of the push rod mounting base of the logistics drone includes: The data acquisition unit 201 is used to acquire vibration data output by the vibration sensor installed on the push rod mounting base, strain data output by the strain gauge, drive current data of the electric push rod, and data on the number of times the electric push rod is operated. The model input unit 202 is used to input the collected vibration data, strain data, drive current data and action number data into the pre-trained intelligent recognition model; the intelligent recognition model analyzes and processes the input data to identify the degree of bolt loosening, structural deformation and fatigue damage of the push rod mounting seat; The degree recognition unit 203 is used to determine the installation status and structural health level of the push rod mounting base based on the identified bolt loosening degree, structural deformation degree, and fatigue damage degree; based on the determined installation status and structural health level, it outputs corresponding alarm information and maintenance suggestions; it also includes: associating and storing each fault identification result, alarm information, maintenance record, and corresponding flight data to form a fault knowledge base; periodically performing statistical analysis on the data in the fault knowledge base to mine the fault occurrence patterns under different flight conditions and different environmental conditions; optimizing the parameters and structure of the intelligent recognition model based on the mined fault occurrence patterns; and pushing the optimized intelligent recognition model to the UAV onboard computing unit for online updates.

[0098] In some embodiments, the acquisition of vibration data output by the vibration sensor mounted on the push rod mounting base, strain data output by the strain gauge, drive current data of the electric push rod, and data on the number of actuations of the electric push rod includes: synchronously acquiring vibration data, strain data, drive current data, and actuation count data during the pre-flight self-check phase, during each actuation of the electric push rod in flight, and after the drone stops. The vibration data is acquired at a first preset frequency, the strain data at a second preset frequency, and the drive current data and actuation count data at a third preset frequency. The first preset frequency is higher than the second preset frequency, and the second preset frequency is higher than the third preset frequency.

[0099] In some embodiments, before inputting the collected vibration data, strain data, drive current data, and number of actions data into the pre-trained intelligent recognition model, the method further includes: removing outliers in the collected data that exceed a preset threshold; aligning the remaining data in segments according to each action cycle of the electric actuator; performing moving average filtering on the vibration data to remove environmental noise; performing time synchronization calibration on the data collected by different sensors to unify the time reference; normalizing all data to eliminate dimensional differences; and compensating for airflow interference in the vibration data by combining the real-time attitude data of the UAV.

[0100] In some embodiments, before inputting the collected vibration data, strain data, drive current data, and number of actions data into the pre-trained intelligent recognition model, the method further includes: constructing a training dataset containing corresponding data for normal state, different bolt loosening degrees, different structural deformation degrees, different fatigue damage degrees, different flight loads, and different ambient temperatures; dividing the training dataset into a training subset, a validation subset, and a test subset; constructing a multi-input multi-output neural network model; training the neural network model using the training subset; adjusting the parameters of the neural network model using the validation subset; verifying the accuracy of the neural network model using the test subset; pre-training the neural network model using historical fault data of similar UAVs and then fine-tuning it using data from the current UAV; and periodically performing online incremental training of the neural network model using newly collected labeled data.

[0101] In some embodiments, the step of analyzing and processing the input data through an intelligent recognition model to identify the bolt loosening degree, structural deformation degree, and fatigue damage degree of the push rod mounting seat includes: the intelligent recognition model extracting multi-dimensional features from the preprocessed input data, fusing the extracted vibration features, strain features, current features, and action frequency features to obtain a comprehensive feature vector; inputting the comprehensive feature vector into three parallel classification branches, which output probability distributions for bolt loosening degree, structural deformation degree, and fatigue damage degree respectively; taking the category with the highest probability in each probability distribution as the identification result of the corresponding fault degree; calculating the confidence level of each identification result; and triggering a secondary data acquisition and identification process when the confidence level is lower than a preset threshold.

[0102] In some embodiments, determining the installation status and structural health level of the push rod mounting base based on the identified bolt loosening degree, structural deformation degree, and fatigue damage degree includes: pre-setting the bolt loosening degree, structural deformation degree, and fatigue damage degree and their corresponding weighting coefficients; multiplying the identified bolt loosening degree, structural deformation degree, and fatigue damage degree values ​​by their respective weighting coefficients and summing them to obtain a comprehensive health score; classifying the corresponding health level according to the preset interval in which the comprehensive health score falls; and directly determining the health level as the highest risk level when any fault degree reaches a preset threshold.

[0103] In some embodiments, the step of outputting alarm information and maintenance suggestions of corresponding levels according to the determined installation status and structural health level includes: outputting no alarm information and normal operation suggestions when the health level is normal; outputting a level 1 alarm information and maintenance suggestions for pre-flight checks when the health level is slightly abnormal; outputting a level 2 alarm information and maintenance suggestions for immediately stopping flight and tightening components when the health level is moderately abnormal; and outputting a level 3 alarm information and maintenance suggestions for immediately performing an emergency landing when the health level is severely abnormal. All alarm information and maintenance suggestions are simultaneously sent to the UAV ground control station and the operation and maintenance management platform.

[0104] In some embodiments, after outputting alarm information and maintenance recommendations corresponding to the determined installation status and structural health level, the method further includes: collecting historical fault identification results and corresponding maintenance records to establish a fault prediction dataset; training a remaining service life prediction model based on the fault prediction dataset; inputting the current installation status and structural health data into the remaining service life prediction model to obtain the remaining service life prediction value of the push rod mounting base; generating a preventive maintenance plan containing maintenance time and maintenance content based on the remaining service life prediction value; and sending the preventive maintenance plan to the operation and maintenance management platform.

[0105] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the intelligent monitoring system for loosening faults of the push rod mounting base of the logistics drone described above and its various modules can be referred to the corresponding content in the various embodiments of the intelligent monitoring method for loosening faults of the push rod mounting base of the logistics drone, and will not be repeated here.

[0106] The aforementioned intelligent monitoring method for loose mounting bracket faults on logistics drones can be implemented as a computer program, which can, for example... Figure 4 It runs on the device shown.

[0107] Please see Figure 5 , Figure 5 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application. The computer device includes a processor, a memory, and a network interface connected via a device bus, wherein the memory may include a storage medium and internal memory.

[0108] The storage medium can store operating devices and computer programs. The computer program includes program instructions that, when executed, cause the processor to perform any intelligent monitoring method for loose push rod mounting brackets on logistics drones.

[0109] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0110] The internal memory provides an environment for the execution of computer programs in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to execute any intelligent monitoring method for loose push rod mounting bracket faults in logistics drones.

[0111] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the terminal to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0112] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0113] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps: The vibration data, strain data, drive current data, and number of actuations of the electric actuator are collected from the vibration sensor installed on the actuator mounting base. The collected vibration data, strain data, drive current data, and number of actions data are input into a pre-trained intelligent recognition model; the intelligent recognition model analyzes and processes the input data to identify the degree of bolt loosening, structural deformation, and fatigue damage of the push rod mounting base; Based on the identified bolt loosening degree, structural deformation degree, and fatigue damage degree, the installation status and structural health level of the push rod mounting base are determined. Based on the determined installation status and structural health level, corresponding alarm information and maintenance suggestions are output. It also includes: associating and storing each fault identification result, alarm information, maintenance record, and corresponding flight data to form a fault knowledge base, periodically performing statistical analysis on the data in the fault knowledge base, mining the fault occurrence patterns under different flight conditions and different environmental conditions, optimizing the parameters and structure of the intelligent identification model based on the mined fault occurrence patterns, and pushing the optimized intelligent identification model to the UAV's onboard computing unit for online updates.

[0114] In some embodiments, the acquisition of vibration data output by the vibration sensor mounted on the push rod mounting base, strain data output by the strain gauge, drive current data of the electric push rod, and data on the number of actuations of the electric push rod includes: synchronously acquiring vibration data, strain data, drive current data, and actuation count data during the pre-flight self-check phase, during each actuation of the electric push rod in flight, and after the drone stops. The vibration data is acquired at a first preset frequency, the strain data at a second preset frequency, and the drive current data and actuation count data at a third preset frequency. The first preset frequency is higher than the second preset frequency, and the second preset frequency is higher than the third preset frequency.

[0115] In some embodiments, before inputting the collected vibration data, strain data, drive current data, and number of actions data into the pre-trained intelligent recognition model, the method further includes: removing outliers in the collected data that exceed a preset threshold; aligning the remaining data in segments according to each action cycle of the electric actuator; performing moving average filtering on the vibration data to remove environmental noise; performing time synchronization calibration on the data collected by different sensors to unify the time reference; normalizing all data to eliminate dimensional differences; and compensating for airflow interference in the vibration data by combining the real-time attitude data of the UAV.

[0116] In some embodiments, before inputting the collected vibration data, strain data, drive current data, and number of actions data into the pre-trained intelligent recognition model, the method further includes: constructing a training dataset containing corresponding data for normal state, different bolt loosening degrees, different structural deformation degrees, different fatigue damage degrees, different flight loads, and different ambient temperatures; dividing the training dataset into a training subset, a validation subset, and a test subset; constructing a multi-input multi-output neural network model; training the neural network model using the training subset; adjusting the parameters of the neural network model using the validation subset; verifying the accuracy of the neural network model using the test subset; pre-training the neural network model using historical fault data of similar UAVs and then fine-tuning it using data from the current UAV; and periodically performing online incremental training of the neural network model using newly collected labeled data.

[0117] In some embodiments, the step of analyzing and processing the input data through an intelligent recognition model to identify the bolt loosening degree, structural deformation degree, and fatigue damage degree of the push rod mounting seat includes: the intelligent recognition model extracting multi-dimensional features from the preprocessed input data, fusing the extracted vibration features, strain features, current features, and action frequency features to obtain a comprehensive feature vector; inputting the comprehensive feature vector into three parallel classification branches, which output probability distributions for bolt loosening degree, structural deformation degree, and fatigue damage degree respectively; taking the category with the highest probability in each probability distribution as the identification result of the corresponding fault degree; calculating the confidence level of each identification result; and triggering a secondary data acquisition and identification process when the confidence level is lower than a preset threshold.

[0118] In some embodiments, determining the installation status and structural health level of the push rod mounting base based on the identified bolt loosening degree, structural deformation degree, and fatigue damage degree includes: pre-setting the bolt loosening degree, structural deformation degree, and fatigue damage degree and their corresponding weighting coefficients; multiplying the identified bolt loosening degree, structural deformation degree, and fatigue damage degree values ​​by their respective weighting coefficients and summing them to obtain a comprehensive health score; classifying the corresponding health level according to the preset interval in which the comprehensive health score falls; and directly determining the health level as the highest risk level when any fault degree reaches a preset threshold.

[0119] In some embodiments, the step of outputting alarm information and maintenance suggestions of corresponding levels according to the determined installation status and structural health level includes: outputting no alarm information and normal operation suggestions when the health level is normal; outputting a level 1 alarm information and maintenance suggestions for pre-flight checks when the health level is slightly abnormal; outputting a level 2 alarm information and maintenance suggestions for immediately stopping flight and tightening components when the health level is moderately abnormal; and outputting a level 3 alarm information and maintenance suggestions for immediately performing an emergency landing when the health level is severely abnormal. All alarm information and maintenance suggestions are simultaneously sent to the UAV ground control station and the operation and maintenance management platform.

[0120] In some embodiments, after outputting alarm information and maintenance recommendations corresponding to the determined installation status and structural health level, the method further includes: collecting historical fault identification results and corresponding maintenance records to establish a fault prediction dataset; training a remaining service life prediction model based on the fault prediction dataset; inputting the current installation status and structural health data into the remaining service life prediction model to obtain the remaining service life prediction value of the push rod mounting base; generating a preventive maintenance plan containing maintenance time and maintenance content based on the remaining service life prediction value; and sending the preventive maintenance plan to the operation and maintenance management platform.

[0121] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the steps of the intelligent monitoring method for loose mounting bracket faults of logistics drone push rods provided in any embodiment of this application.

[0122] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0123] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for intelligent monitoring of loose mounting bracket faults on a logistics drone push rod, characterized in that, include: The vibration data, strain data, drive current data, and number of actuations of the electric actuator are collected from the vibration sensor installed on the actuator mounting base. The collected vibration data, strain data, drive current data, and number of actions data are input into a pre-trained intelligent recognition model; the intelligent recognition model analyzes and processes the input data to identify the degree of bolt loosening, structural deformation, and fatigue damage of the push rod mounting base; Based on the identified bolt loosening degree, structural deformation degree, and fatigue damage degree, the installation status and structural health level of the push rod mounting base are determined. Based on the determined installation status and structural health level, corresponding alarm information and maintenance suggestions are output. This also includes: linking and storing each fault identification result, alarm information, maintenance record and corresponding flight data to form a fault knowledge base; periodically performing statistical analysis on the data in the fault knowledge base to explore the fault occurrence patterns under different flight conditions and environmental conditions; optimizing the parameters and structure of the intelligent identification model based on the discovered fault occurrence patterns; and pushing the optimized intelligent identification model to the UAV's onboard computing unit for online updates.

2. The method according to claim 1, characterized in that, The data collected includes vibration data from the vibration sensor mounted on the push rod mounting base, strain data from the strain gauge, drive current data of the electric push rod, and data on the number of actuations of the electric push rod. Vibration data, strain data, drive current data, and number of actions data are collected synchronously during the self-check phase before takeoff, during each action of the electric push rod in flight, and during the shutdown phase after flight. Vibration data is collected at a first preset frequency, strain data is collected at a second preset frequency, and drive current data and number of actions are collected at a third preset frequency. The first preset frequency is higher than the second preset frequency, and the second preset frequency is higher than the third preset frequency.

3. The method according to claim 1, characterized in that, Before inputting the collected vibration data, strain data, driving current data, and number of actions data into the pre-trained intelligent recognition model, the process also includes: Remove outliers in the collected data that exceed the preset threshold, segment and align the remaining data according to each action cycle of the electric actuator, and perform moving average filtering on the vibration data to remove environmental noise; Time synchronization calibration is performed on data collected from different sensors to unify the time reference. All data are normalized to eliminate dimensional differences. Airflow interference compensation is performed on vibration data in conjunction with real-time attitude data of the UAV.

4. The method according to claim 1, characterized in that, Before inputting the collected vibration data, strain data, driving current data, and number of actions data into the pre-trained intelligent recognition model, the process also includes: A training dataset containing data under normal conditions, different bolt loosening degrees, different structural deformation degrees, different fatigue damage degrees, different flight loads, and different ambient temperatures is constructed. The training dataset is divided into a training subset, a validation subset, and a test subset, and a multi-input multi-output neural network model is constructed. The neural network model is trained using a training subset, its parameters are adjusted using a validation subset, and its accuracy is verified using a test subset. The neural network model is pre-trained using historical fault data from similar UAVs and then fine-tuned using data from the current UAV. Newly collected labeled data is used periodically for online incremental training of the neural network model.

5. The method according to claim 1, characterized in that, The process of analyzing and processing input data using an intelligent recognition model to identify the degree of bolt loosening, structural deformation, and fatigue damage of the push rod mounting base includes: The intelligent recognition model extracts multi-dimensional features from the preprocessed input data and fuses the extracted vibration features, strain features, current features and action frequency features to obtain a comprehensive feature vector. The comprehensive feature vector is input into three parallel classification branches. The three classification branches output the probability distributions of bolt loosening degree, structural deformation degree and fatigue damage degree, respectively. The category with the highest probability in each probability distribution is taken as the identification result of the corresponding fault degree. The confidence level of each identification result is calculated. When the confidence level is lower than the preset threshold, a secondary data acquisition and identification process is triggered.

6. The method according to claim 1, characterized in that, The process of determining the installation status and structural health level of the push rod mounting base based on the identified bolt loosening degree, structural deformation degree, and fatigue damage degree includes: The pre-set bolt loosening degree, structural deformation degree, and fatigue damage degree, along with their corresponding weighting coefficients, are used to obtain a comprehensive health score by multiplying the identified bolt loosening degree, structural deformation degree, and fatigue damage degree values ​​by their respective weighting coefficients and summing them. Based on the preset interval in which the comprehensive health score falls, the corresponding health level is divided. When any fault degree reaches the preset threshold, the health level is directly determined to be the highest risk level.

7. The method according to claim 1, characterized in that, Based on the determined installation status and structural health level, the system outputs corresponding alarm information and maintenance suggestions, including: When the health level is normal, no alarm information and normal operation suggestions are output. When the health level is slightly abnormal, a level 1 alarm information and maintenance suggestions for pre-flight checks are output. When the health level is moderately abnormal, a level 2 alarm information and maintenance suggestions for immediately stopping flight and tightening components are output. When the health level is severely abnormal, a level 3 alarm information and maintenance suggestions for immediately performing an emergency landing are output. All alarm information and maintenance suggestions are sent to the UAV ground control station and operation and maintenance management platform at the same time.

8. The method according to claim 1, characterized in that, After outputting alarm information and maintenance suggestions of the corresponding level based on the determined installation status and structural health level, the following is also included: Collect historical fault identification results and corresponding maintenance records to establish a fault prediction dataset. Train a remaining service life prediction model based on the fault prediction dataset. Input the current installation status and structural health data into the remaining service life prediction model to obtain the remaining service life prediction value of the push rod mounting base. Generate a preventive maintenance plan containing maintenance time and maintenance content based on the remaining service life prediction value, and send the preventive maintenance plan to the operation and maintenance management platform.

9. A smart monitoring system for loose mounting bracket faults of logistics drone push rods, used to implement the method as described in any one of claims 1-8, characterized in that, include: The data acquisition unit is used to acquire vibration data output by the vibration sensor installed on the push rod mounting base, strain data output by the strain gauge, drive current data of the electric push rod, and data on the number of times the electric push rod has been activated. The model input unit is used to input the collected vibration data, strain data, drive current data, and number of actions into the pre-trained intelligent recognition model; the intelligent recognition model analyzes and processes the input data to identify the degree of bolt loosening, structural deformation, and fatigue damage of the push rod mounting base; The degree identification unit is used to determine the installation status and structural health level of the push rod mounting base based on the identified bolt loosening degree, structural deformation degree, and fatigue damage degree. Based on the determined installation status and structural health level, corresponding alarm information and maintenance suggestions are output. This also includes: linking and storing each fault identification result, alarm information, maintenance record and corresponding flight data to form a fault knowledge base; periodically performing statistical analysis on the data in the fault knowledge base to explore the fault occurrence patterns under different flight conditions and environmental conditions; optimizing the parameters and structure of the intelligent identification model based on the discovered fault occurrence patterns; and pushing the optimized intelligent identification model to the UAV's onboard computing unit for online updates.

Images