Predictive maintenance method for drone climb process, electronic device, and storage medium

Abstract

The application discloses a predictive maintenance method for a UAV climbing process, an electronic device and a storage medium, and belongs to the technical field of UAV control. In order to improve the operation ability of the UAV, the application comprises the following steps: collecting input parameters for performance prediction of the UAV climbing process and performing normalization processing, including power system input parameters, motion state input parameters and environmental input parameters; constructing a main body layer of a performance prediction model for the UAV climbing process, adopting a three-layer architecture of a power characteristic layer, a state response layer and an interlayer interaction layer; constructing output indexes of the model, including a dynamic characteristic index group and a climbing ability index group, and adopting a full connection mode to establish an association relationship between the model main body layer and the output indexes; constructing a loss function of the performance prediction model for the UAV climbing process; and based on a model prediction result, restoring the actual physical quantity through reverse normalization, and then performing a predictive maintenance calculation for the UAV climbing process. The application guarantees the timeliness and reliability of maintenance.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle (UAV) control, and in particular relates to a predictive maintenance method, electronic equipment, and storage medium for a UAV's climbing process. Background Art

[0002] The drone's ascent process is the most critical step in the entire flight mission, overcoming multiple factors such as gravity and air resistance, placing extremely high demands on the powertrain's performance. With the widespread application of drones in logistics, emergency rescue, and other fields, the reliability and safety of the ascent process directly impact mission effectiveness. Especially in complex weather conditions, the coupling effect between powertrain performance, flight attitude, and environmental factors becomes more pronounced, placing even higher demands on predictive maintenance during the ascent process.

[0003] Existing predictive maintenance methods for drones during climb primarily focus on monitoring and evaluating single powertrain parameters, such as motor temperature and battery voltage. This approach overlooks the synergistic effects of multiple subsystems during the climb process, including thrust output, attitude adjustment, and energy conversion. This approach fails to accurately assess performance changes and fault development trends during the climb process, resulting in a lack of systematic and forward-looking maintenance decisions.

[0004] In terms of climbing performance prediction, existing methods often use simple linear models or statistical models, which make it difficult to accurately describe the nonlinear dynamic characteristics of the climbing process. Especially in steep climbing and strong disturbance environments, the sudden changes and uncertainty of system performance are more obvious, and existing prediction models have poor adaptability to these complex working conditions, and the prediction results are difficult to support real-time maintenance decisions. In terms of maintenance strategies for the climbing process, existing methods mostly use fixed-cycle or single-threshold-triggered maintenance modes. This mechanical maintenance method cannot dynamically adjust the maintenance strategy according to actual climbing performance and environmental conditions, which can easily lead to waste of maintenance resources or delay of maintenance timing, affecting the safety and reliability of the climbing process.

[0005] The existence of these problems makes it easy for drones to experience failures such as insufficient power, attitude instability, and reduced energy efficiency during the climb process, seriously affecting flight safety; maintenance decisions are delayed or excessive, resulting in increased maintenance costs and low efficiency; system performance cannot be optimized in a timely manner, affecting the climbing effect; especially in severe weather and complex mission scenarios, these problems are more prominent, restricting the improvement of drone operational capabilities. Summary of the Invention

[0006] The problem to be solved by the present invention is to improve the operational capability of UAVs, and a predictive maintenance method, electronic equipment and storage medium for the climbing process of UAVs are proposed.

[0007] To achieve the above object, the present invention is implemented through the following technical solutions:

[0008] A predictive maintenance method for a UAV during a climbing process comprises the following steps:

[0009] S1. Collect and normalize the input parameters for the performance prediction of the UAV during the climb process, including the power system input parameters, motion state input parameters, and environmental input parameters;

[0010] S2. Construct a performance prediction model for the UAV's climbing process, using a three-layer architecture consisting of a dynamic characteristics layer, a state response layer, and an inter-layer interaction layer.

[0011] S3. Based on the main layer of the performance prediction model for the UAV climbing process obtained in step S2, construct the model output indicators, including a dynamic characteristic indicator group and a climbing ability indicator group, and establish the association relationship between the main layer of the model and the output indicators using a fully connected approach;

[0012] S4. Construct a performance prediction model for a UAV during climb, including a loss function consisting of a base loss, a power dynamic mapping loss, a state transition loss, and a response feature loss. The model is then optimized using the integrated mapping loss.

[0013] S5. Based on the input parameter data of the performance prediction of the drone climbing process determined in step S1 and the normalized output index data obtained in step S3, construct a training set, a validation set, and a test set;

[0014] S6. Setting parameters for the performance prediction model of a drone climbing process constructed in steps S2-S4, then training the performance prediction model of a drone climbing process using the training set obtained in step S5, validating it using the validation set, and testing the validated performance prediction model of a drone climbing process using the test set;

[0015] S7. Using the drone monitoring and management platform, obtain input parameters for predicting the performance of the drone's climb process. These parameters are then input into a drone climb process performance prediction model trained in step S6. The resulting prediction results are then restored to actual physical quantities through denormalization, and a predictive maintenance calculation for the drone's climb process is then performed.

[0016] Furthermore, the input parameters for the performance prediction of the UAV climbing process in step S1 are as follows:

[0017] The power system input parameters include motor input voltage A1, motor input current A2, propeller speed A3, measured thrust A4, motor surface temperature A5, motor remaining capacity A6, motor power factor A7 and motor battery internal resistance A8;

[0018] The motion state input parameters include vertical speed A9, horizontal speed A10 , instantaneous height A 11 , pitch angle A 12 Total weight A 13 , vertical acceleration A 14 , dynamic pressure A 15 and lift coefficient A 16 ;

[0019] Environmental input parameters include atmospheric pressure A 17 、Ambient temperature A 18 , relative humidity A 19 , horizontal wind speed A 20 , vertical wind speed A 21 、Air density A 22 , gravitational acceleration A 23 , air viscosity A 24 , and Reynolds number A 25 .

[0020] Furthermore, the specific implementation method of step S2 includes the following steps:

[0021] S2.1. Construct a power characteristics layer. Model the four dimensions of motor input, thrust-to-weight characteristics, energy efficiency characteristics, and thrust output, incorporating weight coefficients and bias terms. The motor input sublayer focuses on basic power output capability; the thrust-to-weight characteristics sublayer focuses on power distribution; the energy efficiency characteristics sublayer reflects energy conversion efficiency; and the thrust output sublayer reflects the final power output.

[0022] S2.2. Construct a state response layer to describe the UAV's response characteristics during the climb process, using the velocity response sublayer, acceleration response sublayer, attitude response sublayer, and energy efficiency sublayer. An LSTM structure is introduced to capture temporal evolution patterns.

[0023] S2.3. The inter-layer interaction layer first extracts environmental impact features, dynamic state features, and motion features, and then performs feature fusion.

[0024] Furthermore, the indicators in the dynamic characteristic index group in step S3 include thrust response time D1, attitude adjustment rate D2, speed tracking error D3, power fluctuation rate D4, thrust-to-weight ratio dynamic value D5 and attitude stability D6, and the indicators in the climbing ability index group include average climb rate D7, energy efficiency index D8, track keeping accuracy D9, power margin D10, and so on. 10 , lift-to-drag ratio D 11 and climb stability index D 12 .

[0025] Furthermore, the specific implementation method of step S4 includes the following steps:

[0026] S4.1. Constructing base losses;

[0027] The basic loss function adopts the form of mean square error, that is, the mean square error between the predicted value and the measured value of all output indicators is calculated, which is the basic loss ;

[0028] S4.2. Constructing power dynamic mapping loss;

[0029] During the climbing process of the UAV, the input voltage and current are converted electrically and mechanically to generate thrust output, accompanied by temperature changes and the dynamic evolution of thrust-to-weight characteristics. The power dynamic mapping loss is characterized and the expression is obtained as follows:

[0030] ;

[0031] in, Dynamic mapping loss for power; is the power deviation coefficient, is the temperature response coefficient, is the thrust-to-weight ratio coefficient;

[0032] S4.3. Constructing state transition loss;

[0033] The climbing process involves state space migration characteristics, including the coupling of multi-dimensional forces and the dynamic changes of velocity and acceleration. It is characterized by state migration loss and the expression is:

[0034] ;

[0035] in, is the state transition loss, is the cosine function, is the characteristic area, is the air speed, is the characteristic time interval, is the dynamic balance coefficient, is the speed characteristic coefficient;

[0036] S4.4. Constructing Response Feature Loss;

[0037] The system response characteristics of the UAV during climbing are characterized by a high-order nonlinear dynamic process, including the coupling of attitude dynamic characteristics, efficiency response characteristics, and stability characteristics. The expression is obtained by characterizing the response characteristic loss:

[0038] ;

[0039] in, For response feature loss; is the characteristic parameter, is the transfer coefficient; 、 、 are the weight coefficients of attitude dynamic characteristics, efficiency response characteristics and stability characteristics respectively; is the angle and acceleration conversion coefficient; is the stability and lift-to-drag ratio conversion coefficient;

[0040] S4.5. Constructing a comprehensive mapping loss.

[0041] Based on the unified description of dynamic characteristics, state characteristics and response characteristics, a comprehensive mapping loss function is constructed , the expression is:

[0042] ;

[0043] in, 、 、 、 are the weight coefficients of basic loss, power dynamic mapping loss, state transition loss and response feature loss respectively.

[0044] Furthermore, step S5 divides the data into a training set, a validation set, and a test set in a ratio of 8:1:1.

[0045] Furthermore, the model training strategy of step S6 is as follows:

[0046] The model is trained using the backpropagation algorithm. During training, the model calculates the predicted value through forward propagation, substitutes the predicted value and the true value into the loss function to calculate the loss, and then updates the network parameters through backpropagation. To prevent overfitting, an early stopping strategy is adopted. Training is stopped when the validation set loss does not decrease for 10 consecutive epochs.

[0047] After the model training is completed, the relationship between input and output is established, and the output result is restored to the actual physical quantity through denormalization.

[0048] Furthermore, the specific implementation method of step S7 includes the following steps:

[0049] S7.1. Construct a comprehensive health index (CHI) by comprehensively analyzing power performance indicators, energy efficiency performance indicators, and stability performance indicators:

[0050] ;

[0051] in, 、 、 are the weight coefficients corresponding to the power performance index, energy efficiency performance index, and stability performance index respectively; 、 、 They are power performance index, energy efficiency performance index and stability performance index respectively;

[0052] Combining the power system output capacity and power fluctuation characteristics, the power performance index is constructed, and the step length correction is introduced through the exponential function to reflect the performance status of the power system. The expression is:

[0053] ;

[0054] in, 、 are respectively the first sensitivity adjustment parameter and the second sensitivity adjustment parameter; is the reference value of power margin, is the base value of power fluctuation rate;

[0055] Based on the energy conversion efficiency and motion characteristic parameters, energy efficiency performance indicators are established. Through the combination of ratio relationship and correction terms, the energy utilization level of the system is evaluated. Energy efficiency performance indicators The expression is:

[0056] ;

[0057] in, is the third sensitivity adjustment parameter, is the benchmark value of energy efficiency index, is the maximum value allowed for the speed tracking error,

[0058] Integrate posture control and motion stability parameters to construct stability evaluation indicators , the expression is:

[0059] ;

[0060] in, 、 They are the reference values ​​of attitude stability and climb stability index respectively;

[0061] S7.2. Establish a dual exponential decay model that considers the influence of environmental factors. Combined with the comprehensive health index, dynamically predict the performance degradation trend and obtain the system performance retention R(t) at time t, expressed as:

[0062] ;

[0063] ;

[0064] in, is the characteristic life parameter, is the reference life, obtained from the design specification; 、 are the temperature coefficient and humidity coefficient, is the shape parameter, is the benchmark value of the comprehensive health index. is the health impact coefficient, obtained from expert experience; The reference temperature of the motor surface is determined by design specifications or expert experience; Reference relative humidity, determined by design specifications or expert experience;

[0065] S7.3. Performance margin assessment: By normalizing and weighting multiple performance parameters, a quantitative indicator of the system's operational safety margin is constructed to reflect the overall operational status of the system, resulting in the dynamic margin index (DMI):

[0066] ;

[0067] in, is the index weight of the single performance of the j-th output parameter of the model, j = 1~12, determined by expert experience; is the performance value of the j-th output parameter at time t; 、 is the minimum and maximum allowable values ​​of the j-th output parameter, determined by design specifications or expert experience;

[0068] S7.4. Maintenance timing optimization: Determine the optimal maintenance interval by minimizing the cost function. Combine preventive maintenance and breakdown maintenance costs, consider the impact of system performance retention, and establish the total maintenance cost per unit time. :

[0069] ;

[0070] in, It is the reference time, obtained from design specifications or expert experience; It is the preventive maintenance cost, determined by market research or maintenance history; The cost of maintenance for failures is determined by market research or maintenance history;

[0071] The optimal maintenance time interval OMI is determined by solving the minimum value of the maintenance cost function:

[0072] ;

[0073] S7.5. Based on the comprehensive health index, system performance retention, dynamic margin index, and optimal maintenance time interval obtained in steps S7.1-S7.4, set 、 They are The corresponding first threshold and second threshold are determined by expert experience; 、 They are The corresponding first threshold and second threshold are determined by expert experience; 、 They are The corresponding first and second thresholds are determined by expert experience to construct the following maintenance decision scheme:

[0074] Emergency maintenance: When or or , it is necessary to shut down the machine immediately for maintenance, conduct key performance indicator tests, and replace parts if necessary;

[0075] Condition monitoring: When and and , it is necessary to conduct regular data collection and performance trend analysis according to the optimal maintenance interval (OMI), and formulate a regular maintenance plan;

[0076] Planned maintenance: When 、 、 When the standards for emergency maintenance and condition monitoring are not met, preventive inspections and adjustments are required within the optimal maintenance interval (OMI), the condition of damaged parts is assessed, and corresponding adjustments are required if problems are encountered.

[0077] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of a predictive maintenance method for a climbing process of a drone when executing the computer program.

[0078] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements a predictive maintenance method for a UAV climbing process.

[0079] Beneficial effects of the present invention:

[0080] The predictive maintenance method for the UAV climbing process described in the present invention constructs a system health assessment model, a power system degradation prediction model, a performance margin assessment model, and a maintenance timing optimization model for the UAV climbing process. On this basis, a maintenance decision optimization model is constructed to achieve scientific decision-making and system optimization for predictive maintenance of the UAV climbing process.

[0081] The predictive maintenance method for drones during their ascent, described in this paper, establishes a comprehensive evaluation system encompassing power performance, energy efficiency, and stability. Incorporating a performance prediction model that factors in environmental factors, it enables accurate assessment of system status. Through real-time dynamic margin monitoring and a tiered maintenance mechanism based on cost optimization, this method ensures timely and reliable maintenance while achieving efficient resource utilization, effectively safeguarding the safe operation of drones during ascent. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 This is a flow chart of a predictive maintenance method for a UAV climbing process according to the present invention;

[0083] Figure 2 Graph showing the CHI calculation results under different working conditions of the present invention;

[0084] Figure 3 This is a diagram of the DMI calculation results under different working conditions of the present invention. DETAILED DESCRIPTION

[0085] In order to make the objectives, technical solutions, and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present invention and are not intended to limit the present invention. That is, the specific embodiments described herein are only some embodiments of the present invention, not all embodiments. Generally, the components of the specific embodiments of the present invention described and illustrated in the drawings herein can be arranged and designed in various different configurations, and the present invention can also have other embodiments.

[0086] Therefore, the following detailed description of the specific embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but is merely representative of selected specific embodiments of the present invention. All other specific embodiments obtained by those skilled in the art based on the specific embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0087] In order to further understand the content, features and effects of the present invention, the following specific embodiments are given as examples, and the attached Figure 1 -Attached Figure 3 The detailed instructions are as follows:

[0088] Example 1:

[0089] A predictive maintenance method for a UAV during a climbing process comprises the following steps:

[0090] S1. Collect and normalize the input parameters for the performance prediction of the UAV during the climb process, including the power system input parameters, motion state input parameters, and environmental input parameters;

[0091] Furthermore, the input parameters for the performance prediction of the UAV climbing process in step S1 are as follows:

[0092] The power system input parameters include motor input voltage A1, motor input current A2, propeller speed A3, measured thrust A4, motor surface temperature A5, motor remaining capacity A6, motor power factor A7 and motor battery internal resistance A8;

[0093] The motion state input parameters include vertical speed A9, horizontal speed A 10 , instantaneous height A 11 , pitch angle A 12 Total weight A 13 , vertical acceleration A 14 , dynamic pressure A 15 and lift coefficient A 16 ;

[0094] Environmental input parameters include atmospheric pressure A 17 、Ambient temperature A 18 , relative humidity A 19 , horizontal wind speed A 20 , vertical wind speed A 21 、Air density A 22 , gravitational acceleration A 23 , air viscosity A 24 , and Reynolds number A 25 .

[0095] S2. Construct a performance prediction model for the UAV's climbing process, using a three-layer architecture consisting of a dynamic characteristics layer, a state response layer, and an inter-layer interaction layer.

[0096] Furthermore, the specific implementation method of step S2 includes the following steps:

[0097] S2.1. Construct a power characteristics layer. Model the four dimensions of motor input, thrust-to-weight characteristics, energy efficiency characteristics, and thrust output, incorporating weight coefficients and bias terms. The motor input sublayer focuses on basic power output capability; the thrust-to-weight characteristics sublayer focuses on power distribution; the energy efficiency characteristics sublayer reflects energy conversion efficiency; and the thrust output sublayer reflects the final power output.

[0098] S2.1.1. The design of the motor input sublayer uses the ReLU activation function for nonlinear mapping. It multiplies all the power system input parameters and environmental parameters, including atmospheric pressure, ambient temperature, and relative humidity, by their corresponding weight coefficients, and then adds a bias term. Finally, the output value of the motor input sublayer is obtained through the ReLU function. ;

[0099] S2.1.2. The thrust-weight characteristic sublayer uses the ReLU activation function to focus on the relationship between thrust and weight. Characterize the thrust-to-weight ratio characteristics, Characterize the relationship between power consumption and battery capacity, Characterize the nonlinear effect of wind speed on thrust-weight characteristics and obtain the output value of the thrust-weight characteristic sublayer , the expression is:

[0100] ;

[0101] in, is the bias term for the feature sub-layer, is the thrust-to-weight ratio influence weight coefficient, is the power consumption rate weight coefficient, is the weight coefficient of vertical acceleration to thrust-weight characteristic, is the weight coefficient of dynamic pressure on thrust-weight characteristics, is the wind field interference weight coefficient, is the weight coefficient of air pressure on thrust-weight characteristics, is the weight coefficient of gravitational acceleration on thrust-weight characteristics;

[0102] S2.1.3. The energy efficiency feature sublayer uses the sigmoid activation function to focus on energy conversion efficiency, using Characterize the effect of temperature gradient on energy conversion efficiency, Characterize the impact of aerodynamic effects on energy conversion and obtain the output value of the energy efficiency characteristic sublayer , the expression is:

[0103] ;

[0104] in, is the bias term of the energy efficiency characteristic sublayer; is the temperature difference loss weight coefficient, is the weight coefficient of the influence of the motor's remaining capacity on energy efficiency, is the weight coefficient of the influence of power factor on energy efficiency, is the weight coefficient of the influence of the motor battery internal resistance on energy efficiency, is the dynamic energy consumption weight coefficient, Aerodynamic efficiency weight coefficient, is the density correction weight coefficient;

[0105] S2.1.4. The thrust output sublayer uses the tanh function to focus on the final output effect, using Characterize the coupling effect of air and gravity loads and obtain the output value of the thrust output sublayer , the expression is:

[0106] ;

[0107] in, is the bias term of the inference output sublayer; is the weight coefficient of the influence of speed on thrust output, is the influence weight coefficient of the measured thrust on the thrust output, is the weight coefficient of the vertical speed on the thrust output, is the weight coefficient of the influence of horizontal velocity on thrust output, The weight coefficient of the influence of altitude on thrust output, is the weight coefficient of the effect of pitch angle on thrust output, is the weight coefficient of the influence of lift coefficient on thrust output, is the weight coefficient of air-gravity coupling effect;

[0108] S2.1.5. Constructing the output of dynamical feature layer fusion , the expression is:

[0109] ;

[0110] in, 、 、 、 are the weight coefficients corresponding to the motor input sublayer, thrust-weight characteristic sublayer, energy efficiency characteristic sublayer, and thrust output sublayer respectively;

[0111] S2.2. Construct a state response layer to describe the UAV's response characteristics during the climb process, using the velocity response sublayer, acceleration response sublayer, attitude response sublayer, and energy efficiency sublayer. An LSTM structure is introduced to capture temporal evolution patterns.

[0112] S2.2.1. The speed response sublayer integrates motor operating parameters, flight state parameters, and environmental parameters, and uses the ReLU activation function to capture the dynamic characteristics of the speed change process. Reflects the direct relationship between the motor input power and the speed response, and obtains the output value of the speed response sublayer , the expression is:

[0113] ;

[0114] in, is the bias term of the speed response sublayer; is the influence weight coefficient of vertical velocity, is the influence weight coefficient of horizontal velocity, is the weight coefficient of the influence of altitude on speed response, is the influence weight coefficient of horizontal wind speed on speed response, is the weight coefficient of the vertical wind speed on the speed response, is the weight coefficient of the influence of input power on speed response, is the influence weight coefficient of the velocity response of the thrust, is the weight coefficient of the influence of air density on speed response;

[0115] S2.2.2. The acceleration response sublayer uses the tanh function to describe the dynamic characteristics of the acceleration process, comprehensively considering the thrust system parameters and aerodynamic characteristics to achieve accurate modeling of the acceleration performance. Considering the impact of load changes on acceleration characteristics, the acceleration capability of the UAV is reflected and the output value of the acceleration response sublayer is obtained. , the expression is:

[0116] ;

[0117] in, is the bias term of the acceleration response sublayer; 、 、 、 、 、 、 are the weight coefficients of vertical acceleration, thrust-to-weight ratio, gravity acceleration, aerodynamic pressure, propeller speed, lift coefficient and air viscosity on acceleration characteristics;

[0118] S2.2.3. The attitude response sublayer uses the sigmoid function to characterize the attitude adjustment process, couples the flight attitude parameters with the influence of environmental factors, and constructs an attitude dynamic response model. Characterizes the intensity of horizontal wind disturbances experienced by the aircraft during vertical motion, which directly affects attitude stability; Reflects the interference intensity of rising or falling airflow on the horizontal motion of the aircraft, affects the attitude control requirements, and obtains the output value of the attitude response sublayer , the expression is:

[0119] ;

[0120] in, is the bias term of the attitude response sublayer; 、 、 、 、 、 are the weight coefficients of the influence of pitch angle, dynamic pressure, lateral interference, longitudinal interference, Reynolds number and atmospheric pressure on the attitude of the UAV;

[0121] S2.2.4. The energy efficiency sublayer establishes an energy efficiency model through the ReLU function, combines the motor characteristics, thrust characteristics and environmental parameters, reflects the efficiency change law in the energy conversion process, considers the energy conversion efficiency under environmental disturbance, and uses Reflects the effectiveness of thrust output under given wind conditions, that is, the ability of thrust output to counteract air kinetic energy loss, and obtains the output value of the energy efficiency sublayer , the expression is:

[0122] ;

[0123] in, is the bias term of the energy efficiency sublayer; 、 、 、 、 、 These are the weight coefficients of lift coefficient, kinetic energy compensation, motor remaining capacity, motor power factor, motor battery internal resistance, and relative humidity on energy efficiency;

[0124] S2.2.5. Use the LSTM network to process the output sequences of the speed response sublayer, acceleration response sublayer, posture response sublayer, and energy efficiency sublayer, and conduct in-depth mining of the state response time series features to obtain the output value of the time series feature embedding layer. , the expression is:

[0125] ;

[0126] in, It is a long short-term memory neural network module;

[0127] S2.2.6. Constructing the output of state-response layer fusion , the expression is:

[0128] ;

[0129] in, is the weight coefficient of the output value of the temporal feature embedding layer, is the weight coefficient for the feature fusion of the velocity response sublayer, acceleration response sublayer, attitude response sublayer and energy efficiency sublayer;

[0130] S2.3. The inter-layer interaction layer first extracts environmental impact features, dynamic state features, and motion features, and then performs feature fusion.

[0131] S2.3.1. The environmental impact feature sublayer assigns corresponding weight coefficients to the environmental input parameters and processes the sum of the weighted environmental input parameters through the ReLU function to obtain the output value E of the environmental impact feature sublayer.

[0132] S2.3.2. The power state feature sublayer assigns corresponding weight coefficients to the power system input parameters and processes the sum of the weighted power system input parameters through the ReLU function to obtain the output value F of the power state feature sublayer.

[0133] S2.3.3. The motion feature sublayer assigns corresponding weight coefficients to the motion state input parameters and processes the sum of the weighted motion state input parameters through the ReLU function to obtain the output value G of the motion feature sublayer.

[0134] S2.3.4. Constructing the output of inter-layer interaction fusion , the expression is:

[0135] ;

[0136] in, 、 、 、 、 They are the weight coefficients corresponding to the fusion output of the dynamic characteristic layer, the fusion output of the state response layer, the output of the environmental impact feature sub-layer, the output of the dynamic state feature sub-layer and the output of the motion feature sub-layer.

[0137] S3. Based on the main layer of the performance prediction model for the UAV climbing process obtained in step S2, construct the model output indicators, including a dynamic characteristic indicator group and a climbing ability indicator group, and establish the association relationship between the main layer of the model and the output indicators using a fully connected approach;

[0138] Furthermore, the indicators in the dynamic characteristic index group in step S3 include thrust response time D1, attitude adjustment rate D2, speed tracking error D3, power fluctuation rate D4, thrust-to-weight ratio dynamic value D5 and attitude stability D6, and the indicators in the climbing ability index group include average climb rate D7, energy efficiency index D8, track keeping accuracy D9, power margin D10, and so on. 10 , lift-to-drag ratio D 11 and climb stability index D 12 , the specific implementation method includes the following steps:

[0139] S3.1. Establish a correlation between the model main layer and the indicators in the dynamic characteristic indicator group;

[0140] ;

[0141] ;

[0142] ;

[0143] ;

[0144] ;

[0145] ;

[0146] in, 、 、 、 、 、 They are the full connections corresponding to thrust response time, attitude adjustment rate, speed tracking error, power fluctuation rate, thrust-to-weight ratio dynamic value and attitude stability;

[0147] S3.2. Establish a correlation between the model's main layer and the indicators in the climb capability indicator group;

[0148] ;

[0149] ;

[0150] ;

[0151] ;

[0152] ;

[0153] ;

[0154] in, 、 、 、 、 、 They are the full connections corresponding to the average climb rate, energy efficiency index, track keeping accuracy, power margin, lift-to-drag ratio and climb stability index.

[0155] S4. Construct a performance prediction model for a UAV during climb, including a loss function consisting of a base loss, a power dynamic mapping loss, a state transition loss, and a response feature loss. The model is then optimized using the integrated mapping loss.

[0156] Furthermore, the specific implementation method of step S4 includes the following steps:

[0157] S4.1. Constructing base losses;

[0158] The basic loss function adopts the form of mean square error, that is, the mean square error between the predicted value and the measured value of all output indicators is calculated, which is the basic loss ;

[0159] S4.2. Constructing power dynamic mapping loss;

[0160] During the climbing process of the UAV, the input voltage and current are converted electrically and mechanically to generate thrust output, accompanied by temperature changes and the dynamic evolution of thrust-to-weight characteristics. The power dynamic mapping loss is characterized and the expression is obtained as follows:

[0161] ;

[0162] in, Dynamic mapping loss for power; is the power deviation coefficient, is the temperature response coefficient, is the thrust-to-weight ratio coefficient;

[0163] S4.3. Constructing state transition loss;

[0164] The climbing process involves state space migration characteristics, including the coupling of multi-dimensional forces and the dynamic changes of velocity and acceleration. It is characterized by state migration loss and the expression is:

[0165] ;

[0166] in, is the state transition loss, is the cosine function, is the characteristic area, is the air speed, is the characteristic time interval, is the dynamic balance coefficient, is the speed characteristic coefficient;

[0167] S4.4. Constructing Response Feature Loss;

[0168] The system response characteristics of the UAV during climbing are characterized by a high-order nonlinear dynamic process, including the coupling of attitude dynamic characteristics, efficiency response characteristics, and stability characteristics. The expression is obtained by characterizing the response characteristic loss:

[0169] ;

[0170] in, For response feature loss; is the characteristic parameter, is the transfer coefficient; 、 、 are the weight coefficients of attitude dynamic characteristics, efficiency response characteristics and stability characteristics respectively; is the angle and acceleration conversion coefficient; is the stability and lift-to-drag ratio conversion coefficient;

[0171] S4.5. Constructing a comprehensive mapping loss.

[0172] Based on the unified description of dynamic characteristics, state characteristics and response characteristics, a comprehensive mapping loss function is constructed , the expression is:

[0173] ;

[0174] in, 、 、 、 are the weight coefficients of basic loss, power dynamic mapping loss, state transition loss and response feature loss respectively.

[0175] S5. Based on the input parameter data of the performance prediction of the drone climbing process determined in step S1 and the normalized output index data obtained in step S3, construct a training set, a validation set, and a test set;

[0176] Furthermore, step S5 divides the data into a training set, a validation set, and a test set in a ratio of 8:1:1.

[0177] S6. Setting parameters for the performance prediction model of a drone climbing process constructed in steps S2-S4, then training the performance prediction model of a drone climbing process using the training set obtained in step S5, validating it using the validation set, and testing the validated performance prediction model of a drone climbing process using the test set;

[0178] Furthermore, the input layer is constructed: it contains 25 neuron nodes, which are divided into power system parameters, motion state parameters, and environmental parameters;

[0179] The main layer of the model is constructed, including a three-layer architecture of dynamic characteristics layer, state response layer and inter-layer interaction;

[0180] Output layer construction: contains 12 neuron nodes, divided into dynamic characteristic index parameters and climbing performance index parameters;

[0181] Loss function construction: It includes three parts: power dynamic mapping loss, state transition loss, and response feature loss.

[0182] Parameter setting: Hyperparameters that need to be set manually based on expert experience include: network structure parameters, number of neurons in each layer, and LSTM hidden layer dimensions;

[0183] Hyperparameters that need to be manually set during model training include: learning rate, batch size, training rounds, and weight coefficients in the loss function;

[0184] Parameters that need to be determined by the model during training: weight coefficients and bias terms of each layer, and weight coefficients of each fusion layer.

[0185] Furthermore, the model training strategy of step S6 is as follows:

[0186] The model is trained using the backpropagation algorithm. During training, the model calculates the predicted value through forward propagation, substitutes the predicted value and the true value into the loss function to calculate the loss, and then updates the network parameters through backpropagation. To prevent overfitting, an early stopping strategy is adopted. Training is stopped when the validation set loss does not decrease for 10 consecutive epochs.

[0187] After the model training is completed, the relationship between input and output is established, and the output result is restored to the actual physical quantity through denormalization.

[0188] S7. Using the drone monitoring and management platform, obtain input parameters for predicting the performance of the drone's climb process. These parameters are then input into a drone climb process performance prediction model trained in step S6. The resulting prediction results are then restored to actual physical quantities through denormalization, and a predictive maintenance calculation for the drone's climb process is then performed.

[0189] Furthermore, the specific implementation method of step S7 includes the following steps:

[0190] S7.1. Construct a comprehensive health index (CHI) by comprehensively analyzing power performance indicators, energy efficiency performance indicators, and stability performance indicators:

[0191] ;

[0192] in, 、 、 are the weight coefficients corresponding to the power performance index, energy efficiency performance index, and stability performance index respectively; 、 、 They are power performance index, energy efficiency performance index and stability performance index respectively;

[0193] Combining the power system output capacity and power fluctuation characteristics, the power performance index is constructed, and the step length correction is introduced through the exponential function to reflect the performance status of the power system. The expression is:

[0194] ;

[0195] in, 、 are respectively the first sensitivity adjustment parameter and the second sensitivity adjustment parameter; is the reference value of power margin, is the base value of power fluctuation rate;

[0196] Based on the energy conversion efficiency and motion characteristic parameters, energy efficiency performance indicators are established. Through the combination of ratio relationship and correction terms, the energy utilization level of the system is evaluated. Energy efficiency performance indicators The expression is:

[0197] ;

[0198] in, is the third sensitivity adjustment parameter, is the benchmark value of energy efficiency index, is the maximum value allowed for the speed tracking error,

[0199] Integrate posture control and motion stability parameters to construct stability evaluation indicators , the expression is:

[0200] ;

[0201] in, 、 They are the reference values ​​of attitude stability and climb stability index respectively;

[0202] System health assessment is the foundation of maintenance decisions. The Comprehensive Health Index (CHI) establishes a weighted evaluation system by comprehensively analyzing key performance indicators to quantitatively represent the overall system status. The CHI considers the coupling effects between indicators and employs nonlinear mapping to more accurately reflect system status.

[0203] S7.2. Establish a dual exponential decay model that considers the influence of environmental factors. Combined with the comprehensive health index, dynamically predict the performance degradation trend and obtain the system performance retention R(t) at time t, expressed as:

[0204] ;

[0205] ;

[0206] in, is the characteristic life parameter, is the reference life, obtained from the design specification; 、 are the temperature coefficient and humidity coefficient, is the shape parameter, is the benchmark value of the comprehensive health index. is the health impact coefficient, obtained from expert experience; The reference temperature of the motor surface is determined by design specifications or expert experience; Reference relative humidity, determined by design specifications or expert experience;

[0207] The power system degradation prediction process combines environmental stress factors to build system performance retention. This performance retention specifically focuses on the impact of temperature and humidity on system life, providing time-series support for maintenance decisions.

[0208] S7.3. Performance margin assessment: By normalizing and weighting multiple performance parameters, a quantitative indicator of the system's operational safety margin is constructed to reflect the overall operational status of the system, resulting in the dynamic margin index (DMI):

[0209] ;

[0210] in, is the index weight of the single performance of the j-th output parameter of the model, j = 1~12, determined by expert experience; is the performance value of the j-th output parameter at time t; 、 is the minimum and maximum allowable values ​​of the j-th output parameter, determined by design specifications or expert experience;

[0211] Performance margin assessment examines system operating status from multiple dimensions and quantifies the system's operational safety margins using the dynamic margin index (DMI). The DMI incorporates time and fluctuation sensitivity corrections to accurately describe the system's dynamic characteristics.

[0212] S7.4. Maintenance timing optimization: Determine the optimal maintenance interval by minimizing the cost function. Combine preventive maintenance and breakdown maintenance costs, consider the impact of system performance retention, and establish the total maintenance cost per unit time. :

[0213] ;

[0214] in, It is the reference time, obtained from design specifications or expert experience; It is the preventive maintenance cost, determined by market research or maintenance history; The cost of maintenance for failures is determined by market research or maintenance history;

[0215] The optimal maintenance time interval OMI is determined by solving the minimum value of the maintenance cost function:

[0216] ;

[0217] Maintenance timing optimization determines the optimal maintenance interval by minimizing a cost function. The optimal maintenance interval takes into account maintenance costs, failure risk, and performance degradation, achieving an economically optimized maintenance strategy.

[0218] S7.5. Based on the comprehensive health index, system performance retention, dynamic margin index, and optimal maintenance time interval obtained in steps S7.1-S7.4, set 、 They are The corresponding first threshold and second threshold are determined by expert experience; 、 They are The corresponding first threshold and second threshold are determined by expert experience; 、 They are The corresponding first and second thresholds are determined by expert experience to construct the following maintenance decision scheme:

[0219] Emergency maintenance: When or or , it is necessary to shut down the machine immediately for maintenance, conduct key performance indicator tests, and replace parts if necessary;

[0220] Condition monitoring: When and and , it is necessary to conduct regular data collection and performance trend analysis according to the optimal maintenance interval (OMI), and formulate a regular maintenance plan;

[0221] Planned maintenance: When 、 、 When the standards for emergency maintenance and condition monitoring are not met, preventive inspections and adjustments are required within the optimal maintenance interval (OMI), the condition of damaged parts is assessed, and corresponding adjustments are required if problems are encountered. Figure 2 and Figure 3 4 and 5 are the calculation result diagrams of CHI and DMI under different working conditions of this embodiment.

[0222] Example 2:

[0223] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the predictive maintenance method for the climbing process of a drone described in Example 1 are implemented.

[0224] The computer device of the present invention may include a processor and memory, such as a single-chip microcomputer including a central processing unit (CPU). Furthermore, the processor is configured to execute a computer program stored in the memory to implement the steps of the aforementioned predictive maintenance method for a drone's climb process. The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor. The memory may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as sound playback or image playback); the data storage area may store data generated based on the use of the mobile phone (such as audio data, a phone book, etc.). In addition, the memory may include a high-speed random access memory and may also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0225] Example 3:

[0226] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the predictive maintenance method for the climbing process of a drone described in Example 1.

[0227] The computer-readable storage medium of the present invention can be any form of storage medium readable by a processor of a computer device, including but not limited to non-volatile memory, volatile memory, ferroelectric memory, etc. The computer-readable storage medium stores a computer program. When the processor of the computer device reads and executes the computer program stored in the memory, the steps of the above-mentioned predictive maintenance method for a drone's climbing process can be implemented. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard drive, a magnetic disk, an optical disk, computer memory, read-only memory (ROM), random access memory (RAM), an electrical carrier signal, a telecommunications signal, and software distribution media.

[0228] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

[0229] Although the present application has been described above with reference to specific embodiments, various modifications may be made thereto and components may be substituted with equivalents without departing from the scope of the present application. In particular, as long as there are no structural conflicts, the various features of the embodiments disclosed herein may be combined with each other in any manner, and the omission of an exhaustive description of these combinations in this specification is solely for the sake of space and resource conservation. Therefore, the present application is not limited to the specific embodiments disclosed herein, but includes all technical solutions within the scope of the claims.

Claims

1. A predictive maintenance method for a UAV climbing process, characterized in that: The steps include: S1. Collect and normalize the input parameters for the performance prediction of the UAV during the climb process, including the power system input parameters, motion state input parameters, and environmental input parameters; The input parameters for the performance prediction of the UAV climbing process in step S1 are as follows: The power system input parameters include motor input voltage A1, motor input current A2, propeller speed A3, measured thrust A4, motor surface temperature A5, motor remaining capacity A6, motor power factor A7 and motor battery internal resistance A8; The motion state input parameters include vertical speed A9, horizontal speed A 10 , instantaneous height A 11 , pitch angle A 12 Total weight A 13 , vertical acceleration A 14 , dynamic pressure A 15 and lift coefficient A 16 ; Environmental input parameters include atmospheric pressure A 17 、Ambient temperature A 18 , relative humidity A 19 , horizontal wind speed A 20 , vertical wind speed A 21 、Air density A 22 , gravitational acceleration A 23 , air viscosity A 24 , and Reynolds number A 25 ; S2. Construct a performance prediction model for a UAV during climbing, using a three-layer architecture consisting of a dynamic characteristics layer, a state response layer, and an inter-layer interaction layer. The specific implementation method of step S2 includes the following steps: S2.

1. Construct a power characteristics layer to model the four dimensions of motor input, thrust-to-weight characteristics, energy efficiency characteristics, and thrust output. Weight coefficients and bias terms are added. The motor input sublayer focuses on basic power output capability. The thrust-weight characteristic sublayer focuses on the power distribution effect; the energy efficiency characteristic sublayer reflects the energy conversion efficiency; The thrust output sublayer reflects the final power effect; S2.1.

1. The design of the motor input sublayer uses the ReLU activation function for nonlinear mapping. It multiplies all the power system input parameters and environmental parameters, including atmospheric pressure, ambient temperature, and relative humidity, by their corresponding weight coefficients, and then adds a bias term. Finally, the output value of the motor input sublayer is obtained through the ReLU function. ; S2.1.

2. The thrust-weight characteristic sublayer uses the ReLU activation function to focus on the relationship between thrust and weight. Characterize the thrust-to-weight ratio characteristics, Characterize the relationship between power consumption and battery capacity, Characterize the nonlinear effect of wind speed on thrust-weight characteristics and obtain the output value of the thrust-weight characteristic sublayer ; S2.1.

3. The energy efficiency feature sublayer uses the sigmoid activation function to focus on energy conversion efficiency, using Characterize the effect of temperature gradient on energy conversion efficiency, Characterize the impact of aerodynamic effects on energy conversion and obtain the output value of the energy efficiency characteristic sublayer ; S2.1.

4. The thrust output sublayer uses the tanh function to focus on the final output effect, using Characterize the coupling effect of air and gravity loads and obtain the output value of the thrust output sublayer ; S2.1.

5. Constructing the output of dynamical feature layer fusion , the expression is: ； in, 、 、 、 are the weight coefficients corresponding to the motor input sublayer, thrust-weight characteristic sublayer, energy efficiency characteristic sublayer, and thrust output sublayer respectively; S2.

2. Construct a state response layer to describe the UAV's response characteristics during the climb process, using the velocity response sublayer, acceleration response sublayer, attitude response sublayer, and energy efficiency sublayer. An LSTM structure is introduced to capture temporal evolution patterns. S2.2.

1. The speed response sublayer integrates motor operating parameters, flight state parameters, and environmental parameters, and uses the ReLU activation function to capture the dynamic characteristics of the speed change process. Reflects the direct relationship between the motor input power and the speed response, and obtains the output value of the speed response sublayer ; S2.2.

2. The acceleration response sublayer uses the tanh function to describe the dynamic characteristics of the acceleration process, comprehensively considering the thrust system parameters and aerodynamic characteristics to achieve accurate modeling of the acceleration performance. Considering the impact of load changes on acceleration characteristics, the acceleration capability of the UAV is reflected and the output value of the acceleration response sublayer is obtained. ; S2.2.

3. The attitude response sublayer uses the sigmoid function to characterize the attitude adjustment process, couples the flight attitude parameters with the influence of environmental factors, and constructs an attitude dynamic response model. Characterizes the intensity of horizontal wind disturbances experienced by the aircraft during vertical motion, which directly affects attitude stability; Reflects the interference intensity of rising or falling airflow on the horizontal motion of the aircraft, affects the attitude control requirements, and obtains the output value of the attitude response sublayer ; S2.2.

4. The energy efficiency sublayer establishes an energy efficiency model through the ReLU function, combines the motor characteristics, thrust characteristics and environmental parameters, reflects the efficiency change law in the energy conversion process, considers the energy conversion efficiency under environmental disturbance, and uses Reflects the effectiveness of thrust output under given wind conditions, that is, the ability of thrust output to counteract air kinetic energy loss, and obtains the output value of the energy efficiency sublayer ; S2.2.

5. Use the LSTM network to process the output sequences of the speed response sublayer, acceleration response sublayer, posture response sublayer, and energy efficiency sublayer, and conduct in-depth mining of the state response time series features to obtain the output value of the time series feature embedding layer. , the expression is: ； in, It is a long short-term memory neural network module; S2.2.

6. Constructing the output of state-response layer fusion , the expression is: ； in, is the weight coefficient of the output value of the temporal feature embedding layer, is the weight coefficient for the feature fusion of the velocity response sublayer, acceleration response sublayer, attitude response sublayer and energy efficiency sublayer; S2.

3. The inter-layer interaction layer first extracts environmental impact features, dynamic state features, and motion features, and then performs feature fusion. S2.3.

1. The environmental impact feature sublayer assigns corresponding weight coefficients to the environmental input parameters and processes the sum of the weighted environmental input parameters through the ReLU function to obtain the output value E of the environmental impact feature sublayer. S2.3.

2. The power state feature sublayer assigns corresponding weight coefficients to the power system input parameters and processes the sum of the weighted power system input parameters through the ReLU function to obtain the output value F of the power state feature sublayer. S2.3.

3. The motion feature sublayer assigns corresponding weight coefficients to the motion state input parameters and processes the sum of the weighted motion state input parameters through the ReLU function to obtain the output value G of the motion feature sublayer. S2.3.

4. Constructing the output of inter-layer interaction fusion , the expression is: ； in, 、 、 、 、 are the weight coefficients corresponding to the fusion output of the dynamic characteristic layer, the fusion output of the state response layer, the output of the environmental impact feature sub-layer, the output of the dynamic state feature sub-layer, and the output of the motion feature sub-layer; S3. Based on the main layer of the performance prediction model for the UAV climbing process obtained in step S2, construct the model output indicators, including a dynamic characteristic indicator group and a climbing ability indicator group, and establish the association relationship between the main layer of the model and the output indicators using a fully connected approach; S4. Construct a performance prediction model for a UAV during climb, including a loss function consisting of a base loss, a power dynamic mapping loss, a state transition loss, and a response feature loss. The model is then optimized using the integrated mapping loss. S5. Based on the input parameter data of the performance prediction of the drone climbing process determined in step S1 and the normalized output index data obtained in step S3, construct a training set, a validation set, and a test set; S6. Setting parameters for the performance prediction model of a drone climbing process constructed in steps S2-S4, then training the performance prediction model of a drone climbing process using the training set obtained in step S5, validating it using the validation set, and testing the validated performance prediction model of a drone climbing process using the test set; S7. Using the drone monitoring and management platform, obtain input parameters for predicting the performance of the drone's climb process. These parameters are then input into a drone climb process performance prediction model trained in step S6. The resulting prediction results are then restored to actual physical quantities through denormalization, and a predictive maintenance calculation for the drone's climb process is then performed.

2. The predictive maintenance method for a UAV during a climbing process according to claim 1, characterized in that: The indicators in the dynamic characteristic index group in step S3 include thrust response time D1, attitude adjustment rate D2, speed tracking error D3, power fluctuation rate D4, thrust-to-weight ratio dynamic value D5 and attitude stability D6. The indicators in the climbing ability index group include average climb rate D7, energy efficiency index D8, track keeping accuracy D9, power margin D10, and so on. 10 , lift-to-drag ratio D 11 and climb stability index D 12 .

3. The predictive maintenance method for a UAV climbing process according to claim 2, characterized in that: The specific implementation method of step S4 includes the following steps: S4.

1. Constructing base losses; The basic loss function adopts the form of mean square error, that is, the mean square error between the predicted value and the measured value of all output indicators is calculated, which is the basic loss ; S4.

2. Constructing power dynamic mapping loss; Based on the fact that during the climbing process of the UAV, the input voltage and current are converted electrically and mechanically to generate thrust output, accompanied by temperature changes and dynamic evolution of thrust-weight characteristics, the power dynamic mapping loss is characterized. ; S4.

3. Constructing state transition loss; The climbing process involves state space migration characteristics, including the coupling of multi-dimensional forces and the dynamic changes of velocity and acceleration, which are characterized by state migration loss. ; S4.

4. Constructing Response Feature Loss; The system response characteristics of the UAV during climbing are characterized by a high-order nonlinear dynamic process, including the coupling of attitude dynamic characteristics, efficiency response characteristics and stability characteristics, which is characterized by the response characteristic loss. ; S4.

5. Construct a comprehensive mapping loss function based on the unified description of dynamic characteristics, state characteristics, and response characteristics. .

4. The predictive maintenance method for a UAV during a climbing process according to claim 3, characterized in that: Step S5 divides the data into a training set, a validation set, and a test set in a ratio of 8:1:

1.

5. The predictive maintenance method for a UAV during a climbing process according to claim 4, characterized in that: The model training strategy of step S6 is as follows: The model is trained using the backpropagation algorithm. During training, the model calculates the predicted value through forward propagation, substitutes the predicted value and the true value into the loss function to calculate the loss, and then updates the network parameters through backpropagation. To prevent overfitting, an early stopping strategy is adopted. Training is stopped when the validation set loss does not decrease for 10 consecutive epochs. After the model training is completed, the relationship between input and output is established, and the output result is restored to the actual physical quantity through denormalization.

6. The predictive maintenance method for a UAV climbing process according to claim 5, characterized in that: The specific implementation method of step S7 includes the following steps: S7.

1. Construct a comprehensive health index (CHI) by comprehensively analyzing power performance indicators, energy efficiency performance indicators, and stability performance indicators. S7.

2. Establish a dual exponential decay model that considers environmental factors and combines it with the comprehensive health index to dynamically predict performance degradation trends and obtain the system performance retention R(t) at time t. S7.

3. Performance margin assessment: By normalizing and weighting multiple performance parameters, a quantitative indicator of the system's operational safety margin is constructed to reflect the overall operational status of the system, resulting in the dynamic margin index (DMI). S7.

4. Maintenance timing optimization: Determine the optimal maintenance interval by minimizing the cost function. Combine preventive maintenance and breakdown maintenance costs, consider the impact of system performance retention, and establish the total maintenance cost per unit time. ; The optimal maintenance time interval OMI is determined by solving the minimum value of the maintenance cost function: ； S7.

5. Based on the comprehensive health index, system performance retention, dynamic margin index, and optimal maintenance time interval obtained in steps S7.1-S7.4, set 、 They are The corresponding first threshold and second threshold are determined by expert experience; 、 They are The corresponding first threshold and second threshold are determined by expert experience; 、 They are The corresponding first and second thresholds are determined by expert experience to construct the following maintenance decision scheme: Emergency maintenance: When or or , it is necessary to shut down the machine immediately for maintenance, conduct key performance indicator tests, and replace parts if necessary; Condition monitoring: When and and , it is necessary to conduct regular data collection and performance trend analysis according to the optimal maintenance interval (OMI), and formulate a regular maintenance plan; Planned maintenance: When 、 、 When the standards for emergency maintenance and condition monitoring are not met, preventive inspections and adjustments are required within the optimal maintenance interval (OMI), the condition of damaged parts is assessed, and corresponding adjustments are required if problems are encountered.

7. An electronic device, characterized in that: The invention comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the predictive maintenance method for the climbing process of a drone according to any one of claims 1 to 6 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the predictive maintenance method for the climbing process of a drone as described in any one of claims 1 to 6.

Images