A system and method for unmanned aerial vehicle inspection of wind turbine blades in high-cold environments

Abstract

This invention relates to the field of industrial automatic control systems, and discloses a system and method for inspecting wind turbine blades using a drone in cold environments. The system includes a control module configured to establish a dual closed-loop control rule: constructing an icing criterion based on blade surface environmental data to distinguish between icing and structural damage in image data, and generating inspection result data; and monitoring the drone's own icing status in real time, executing a flight path replanning command when a preset flight safety threshold is triggered. This invention solves the problem of insufficient robustness of existing control systems in cold, highly disturbed environments due to model failure and rigid control logic. By decoupling the environment, platform state, and task objectives collaboratively, the control system can proactively adapt to environmental disturbances and platform state changes, ensuring the execution of industrial inspection tasks.

Description

Technical Field

[0001] This invention relates to a system and method for inspecting wind turbine blades using unmanned aerial vehicles (UAVs) in cold environments, belonging to the field of industrial automatic control system technology. Background Technology

[0002] Currently, using automated inspection systems to monitor the condition of large equipment, such as wind turbine blades, has become a routine technique. Compared with traditional manual climbing operations, this method has significant advantages in terms of efficiency and safety. Its core technology lies in providing an automatic control system that can drive the carrier platform, such as an unmanned aerial vehicle, to accurately execute predetermined trajectories and collect data. However, when such conventional control systems are deployed in complex working conditions of high altitude and cold weather, the premise of a stable environment and deterministic model on which their design depends no longer applies. Strong gusts, unpredictable wind shear, and icing of the body and blades due to low temperatures are not just simple external disturbances. They will change the aerodynamic characteristics of the flight platform itself and the response characteristics of the actuators in real time and nonlinearly, causing the mathematical model on which the control system depends to fail rapidly during operation.

[0003] Faced with this situation of model failure and strong disturbances coexisting, existing control logic typically adopts a rigid countermeasure strategy. The system attempts to forcibly correct the error deviating from the predetermined flight path through high-frequency, large-amplitude control outputs. This not only rapidly consumes valuable onboard energy but is also more likely to cause control saturation or even system instability under extreme disturbances. Simply enhancing the platform's cold resistance or power redundancy only raises the countermeasure threshold and does not solve the fundamental limitations of control logic in the face of model dynamic uncertainty. However, existing technical discussions focus more on the hardware and simple software control problems of how to acquire high-quality image data, while neglecting the deficiencies of control strategies and systemic risk management in maintaining platform survivability and mission robustness in high-altitude and complex environments. For example, Chinese invention patent CN115529407A discloses an image acquisition method, system, device, and storage medium for non-stop inspection. This solution proposes a method to control the movement of an aircraft carrying a camera. The technology of moving the camera to a preset position and keeping its rotational angular velocity consistent with that of the wind turbine blades to achieve image acquisition without stopping the machine relies on mechanical synchronization to stabilize the subject and solve the problem of image acquisition when the blades are rotating at high speed. However, the control logic of this technology is still static and rigid: its focus is only on how to keep the camera and the blades relatively stationary to achieve high-definition shooting. In essence, it is still a kind of precise tracking of deterministic motion trajectory. It does not take into account the real-time icing risk that the flight platform itself will experience in high-altitude and cold environments, the strong wind shear disturbances brought by the environment, and how to incorporate these dynamic nonlinear state changes into flight safety decisions and mission priority planning. In actual high-altitude and cold working conditions, once the UAV body begins to ice up or encounters a sudden strong wind, this traditional control system with image acquisition as the sole objective will immediately lose its robustness due to model mismatch and may even face serious safety risks of platform loss of control.

[0004] Therefore, the technical problem to be solved by this invention is how to provide a new control system and method that can collaboratively decouple and dynamically plan the environmental platform state and task objectives in a cold and highly disturbed environment, avoid rigid dependence on precise trajectories, and achieve robust execution of industrial inspection tasks while ensuring system survival. Summary of the Invention

[0005] This invention provides a system and method for inspecting wind turbine blades using unmanned aerial vehicles (UAVs) in cold environments. Its main purpose is to solve the problem that existing control systems, when facing cold and highly disturbed environments, suffer from insufficient robustness in inspection operations due to model failure and rigid control logic, making it difficult to coordinate the relationship between the environmental platform and the task.

[0006] To achieve the above objectives, the present invention provides a UAV inspection system for wind turbine blades in high-altitude and cold environments, the system comprising: The drone platform is equipped with a blade inspection unit and an environmental monitoring unit for monitoring the real-time status of the blade surface and the drone platform itself. The control module is configured to establish and execute the following operating rules: acquire blade surface image data collected by the blade inspection unit and simultaneously acquire real-time environmental data of the blade surface collected by the environmental monitoring unit; construct a preset icing criterion based on the real-time environmental data, and use the icing criterion to distinguish between icing areas and structural damage areas in the blade surface image data to generate inspection result data; acquire real-time icing status data of key components of the UAV platform collected by the environmental monitoring unit; continuously compare the real-time icing status data with a preset flight safety status threshold; and when the real-time icing status data triggers the preset flight safety status threshold, prioritize the execution of a flight path replanning command to adjust the subsequent flight path of the UAV platform until the real-time icing status data returns to within the preset flight safety status threshold.

[0007] Preferably, when constructing the icing criterion based on real-time environmental data, the control module is further configured to: calculate a context risk coefficient based on the real-time environmental data; and use the context risk coefficient to dynamically adjust the discrimination threshold used in the icing criterion through a preset adjustment function relationship.

[0008] Preferably, the control module is further configured to: acquire inspection result data, and execute risk model self-calibration logic based on the inspection result data; wherein when the inspection result data indicates that no abnormality was found, the control module automatically reduces the risk accumulation coefficient used to evaluate the future risk points of the preset blade while clearing the risk point count of the corresponding blade to zero.

[0009] Preferably, the risk model self-calibration logic is further configured such that when the inspection result data indicates that a serious risk has been found, the control module is configured to prevent the risk point count from being cleared to zero, but instead forcibly set it to a locked maximum value and generate a manual intervention alarm until the control module receives an external manual repair confirmation signal before the locked maximum value is released.

[0010] Preferably, the control module further includes a risk accumulator module, which is used to dynamically accumulate the number of risk points of each blade in the entire field based on preset risk accounting rules and in response to real-time operation events of the wind turbine, thereby constructing an operation and maintenance risk exposure model.

[0011] Preferably, the risk accounting rules include: a basic fatigue accumulation rule based on normal operating time; and at least one event-triggered accumulation rule, which is configured to respond to highly deterministic wind turbine operating data, including emergency shutdown event data and wind turbine icing status data identified by power curves.

[0012] Preferably, the system further includes a risk velocity factor calculation module, which is used to: periodically calculate and store historical change data of risk points, and generate a risk velocity factor Vf for each blade in the entire field based on the historical change data; where Vf=1+(Pcurrent−Phistory) / Pbase, where Pcurrent is the current number of risk points of the blade, Phistory is the number of risk points of the blade before a preset historical time point, and Pbase is a preset baseline value of risk points used for normalization; the control module uses the product of the number of risk points and the risk velocity factor as the priority basis for decision-making, and the system further includes: an opportunity identifier module, which is used to obtain the future safe operation and maintenance window in the high-altitude cold environment as a system constraint; and an optimal settlement controller module, which is used to solve an optimal control problem based on the operation and maintenance risk exposure model and the safe operation and maintenance window constraint with the control objective of maximizing the total amount of risk point settlement, and generate a specific UAV operation instruction sequence.

[0013] Preferably, when solving the optimal control problem, the optimal clearing controller module is further configured to use the energy model or range model of the UAV platform as a second constraint.

[0014] Preferably, the risk accounting rule further includes a manual calibration interface, which is configured to allow maintenance personnel to set a risk accumulation coefficient for preset blades.

[0015] A method for inspecting wind turbine blades using a drone in cold environments includes the following steps: Step a: Collect image data of the blade surface through the blade inspection unit, and simultaneously collect real-time environmental data of the blade surface through the environmental monitoring unit. Step b: Construct a preset icing criterion based on real-time environmental data; Step c: Use the icing criterion to distinguish between the icing area and the structural damage area in the blade surface image data in order to generate inspection result data; Step d: Real-time icing status data of key components of the UAV platform are obtained through the environmental monitoring unit. Step e involves continuously comparing real-time icing status data with preset flight safety status thresholds. Step f, and when the real-time icing status data triggers the preset flight safety status threshold, the route replanning command is executed first to adjust the subsequent flight path of the UAV platform. Step g continues until the real-time icing status data returns to within the preset flight safety status threshold.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This solution constructs a technical path that couples environmental disturbance parameters with the control commands of the inspection actuator. By utilizing real-time sensed information on changes in the external environment, such as temperature and wind speed, it directly participates in the closed-loop calculation process of flight attitude and trajectory. This allows the control system to no longer passively resist environmental disturbances, but to use environmental factors as part of the control model for feedforward compensation and dynamic adjustment. This avoids the problem of control response lag or divergence caused by model mismatch when encountering sudden environmental disturbances in traditional control methods.

[0017] 2. By establishing a real-time linkage mechanism between the flight platform's own state parameters and the inspection task planning, the problem of rigid execution of the operation process in extreme environments is solved. For example, when the system detects that the blades or the airframe are icing or other changes that alter the model's characteristics by monitoring the deviation between the execution power consumption and attitude feedback, the control logic will autonomously adjust the preset inspection waypoints, hovering safety distances, and image acquisition strategies. This enables the industrial inspection process itself to adapt to the dynamic changes of the controlled object and the execution platform, rather than simply maintaining the stability of the flight platform.

[0018] 3. This solution separates the task objective of the inspection operation from the attitude maintenance of the flight platform in the control logic. Under conditions such as strong wind shear, the core objective of the system control becomes maintaining the relative position and attitude stability of the inspection probe and the target area of ​​the blade, while allowing the flight platform to adapt to disturbances within a certain range. This flexible control strategy frees it from the dilemma of consuming a lot of energy or causing control overload in order to maintain absolute spatial coordinate stability under harsh conditions, and ensures the continuity and data acquisition quality of core industrial inspection tasks under complex disturbances. Attached Figure Description

[0019] Figure 1 This is a logic block diagram of the dual closed-loop adaptive inspection system of the present invention; Figure 2 This is a comparison chart showing the impact of the context adjustment logic of this invention on risk assessment in different temperature ranges; Figure 3 This is a diagram illustrating the four key factors contributing to the insufficient robustness of existing technologies for high-altitude and cold-weather inspections. Detailed Implementation

[0020] The present invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the scope of protection of the present invention.

[0021] This invention discloses a system and method for inspecting wind turbine blades using a drone in cold environments. Technically, it constructs an industrial process adaptive control system. Physically, this system mainly includes a drone platform as a sensing and execution terminal, and a control module as the control core. The drone platform is configured according to conventional technology in the field, carrying a blade inspection unit for acquiring blade status and an environmental monitoring unit for monitoring the operating environment and the platform's own status. The control module, as the system's brain, communicates with the drone platform via a data link. Logically, the control module is configured to execute a real-time inner-loop control logic and a medium- to long-term outer-loop adaptive control logic. The inner-loop control logic aims to ensure the drone platform performs its operations effectively. The outer loop adaptive control logic is responsible for modeling, optimizing, and calibrating the operation and maintenance risks of the entire wind farm industrial process, while ensuring the survivability and effectiveness of data acquisition during the mission. This outer loop logic further includes a risk accumulator module, an opportunity identifier module, an optimal settlement controller module, and a risk model self-calibration logic. The two work together to achieve robustness and optimal resource allocation in the industrial inspection process under high-altitude and cold environments. In the inner loop control logic, the primary challenge for the control module is to ensure the reliability of its sensors and actuators under strong disturbances. To this end, the control module is configured to establish and execute a dual closed-loop operation rule. The first rule aims to solve the data contamination problem of the inspection sensors, namely icing and structural damage in high-altitude and cold environments. Given the high similarity of damage patterns in images, the control module, when executing this rule, first acquires image data of the blade surface collected by the blade inspection unit via a drone platform, and simultaneously acquires real-time environmental data of the blade surface collected by the environmental monitoring unit, such as surface temperature and ambient humidity. The objective obstacle faced by the control system is that image data alone cannot make a highly reliable distinction between icing and cracks. To solve this problem, the control module of this solution constructs a preset icing criterion based on real-time environmental data. The establishment procedure for this criterion is a deterministic engineering calibration process. For example, a physical model is established using offline experimental data, defining a condition where the blade surface temperature is below 0℃ and the ambient air relative humidity is above 95% simultaneously. This constitutes a context state with a high probability of icing. Furthermore, the control module uses icing criteria to distinguish abnormal regions identified in the blade surface image data through image processing algorithms, such as edge detection or deep learning segmentation. The distinction logic is set as follows: if an abnormal image feature point appears during a context state with a high probability of icing, the control module marks it as an icing area in the inspection result data; conversely, if a similar image anomaly is detected in a non-icing environment, such as 5°C, it is marked as a structural damage area. The second rule aims to address the platform survivability issue of the inspection actuator. The control module acquires real-time icing status data of key components of the UAV platform, such as the wing leading edge or rotor, collected by the environmental monitoring unit.The control module continuously compares the real-time icing status data with a preset flight safety threshold, which is also obtained through deterministic calibration procedures. For example, wind tunnel testing may be used to define a trigger threshold as a decrease in the lift characteristics of the wing or rotor due to icing exceeding 20%.

[0022] When real-time icing data triggers a preset flight safety threshold, the control module's control law immediately switches the system's highest priority from task execution to ensuring platform survival. It prioritizes executing a route replanning command, which is deterministically generated to stop the current operation, climb to a preset safe altitude, and return to the takeoff and landing point along the shortest unobstructed path. This process continues until the real-time icing data returns to within the preset flight safety threshold after de-icing on the ground. Only then does the control system allow the next operation. The construction of the icing criterion and the determination of the flight safety threshold are based on deterministic engineering calibration procedures for high-altitude, cold environments to ensure the reliability of the inner-loop control logic when executing platform survival tasks. The construction procedure for the differentiation threshold used in the icing criterion is based on offline regression analysis of historical environmental data (blade surface temperature, relative humidity, etc.) and actual blade icing records. This analysis outputs a weighted average of two or more key environmental parameters. The system determines that it has entered a high icing probability state when the value of the discrimination function exceeds the preset threshold ThresholdI. This criterion is a prerequisite for distinguishing whether the abnormal area in the image is icing or structural damage. The flight safety state threshold of the UAV platform itself under icing conditions, such as the triggering of a drop in lift characteristics of more than 20% due to icing, is indirectly achieved in actual flight by real-time monitoring of motor power consumption deviation rate: the system monitors in real time the deviation rate ΔP = (Pactual − Pbase) / Pbase between the real-time power consumption Pactual of the motor output and the reference power consumption Pbase under the given hovering attitude command. The power consumption deviation rate ΔPThreshold corresponding to the 20% lift drop threshold has been pre-calibrated through wind tunnel testing and stored in the control module. Once ΔP triggers ΔPThreshold, the flight path replanning command is immediately executed first.

[0023] In the outer-loop adaptive control logic, the control module overcomes the technical dilemma of relying on fragile physical prediction models by constructing a simplified and robust industrial process model based on risk accounting. This model is built and maintained by the risk accumulator module. For each blade asset in the entire system, taking blade B-8 as an example, the risk accumulator module establishes and maintains a risk point accumulator in the control system's database. This accumulator does not perform complex physical predictions but responds to real-time turbine operation events by dynamically accumulating risk points based on preset, transparent risk accounting rules, thereby constructing a quantifiable and understandable operational risk exposure model. The risk accounting rules are deterministically configured to include: a basic fatigue accumulation rule based on normal operating time, for example, for every hour of safe operation of blade B-8, its risk point accumulation increases by 1; and at least one event-triggered accumulation rule, which is configured to only respond to highly deterministic turbine operations that the system can obtain with high confidence. For example, when the SCADA system reports an emergency shutdown event, the control module determines that this is a high-load impact event and immediately accumulates +1000 points for blade B-8. Alternatively, when the control system identifies that the turbine is in an icing state by analyzing the robust operating indicator of the SCADA power-wind speed curve, it initiates a higher accumulation rate, such as +50 points per hour. In this way, the model upon which the control system relies is anchored to simplified, robust engineering facts. To further improve the engineering compliance of this simplified model, the risk accounting rules further include a manual adjustment interface. This interface is configured to allow an authorized maintenance engineer to set a risk accumulation coefficient, for example, 1.5, for a preset blade, such as blade B-8, which is known to have early fatigue issues. Subsequently, when calculating the points for blade B-8, the risk accumulator module will automatically multiply all accumulated values ​​by this coefficient of 1.5.

[0024] To address the secondary bottleneck of rigid rules in the aforementioned basic model—namely, the lack of consideration for the context of risk events—the control module of this solution is further configured to synchronously initiate a context adjustment logic when constructing icing criteria based on real-time environmental data or when executing event-triggered cumulative rules. Specifically, when responding to an event, such as an emergency shutdown, the control module not only captures the event itself but also synchronously captures the real-time environmental data at the instant the event occurs, such as a temperature of -30°C. The control module calculates a context risk coefficient based on this real-time environmental data. This calculation process relies on a preset adjustment function relationship, which in engineering implementation is typically a simplified, engineer-configurable two-dimensional lookup. The lookup table, for example, specifies: IF event = emergency shutdown AND context (temperature) < -25°C THEN coefficient = 3.0; IF event = emergency shutdown AND context (temperature) > -5°C THEN coefficient = 0.8. Subsequently, the control module uses the context risk coefficient to dynamically adjust the discrimination threshold used in the icing criterion, or to dynamically adjust the baseline number of the event trigger accumulation rule, i.e., 1000 points, so that shutdown events at -30°C ultimately accumulate +3000 points (1000 × 3.0), while shutdown events at -5°C only accumulate +800 points (1000 × 0.8), thus making the model (i.e., the number of risk points) more faithfully reflect the situation. Real-world engineering risks; to address another secondary bottleneck in this model—its short-sightedness, focusing only on existing risks rather than new ones—this solution further includes a risk velocity factor calculation module. This module periodically (e.g., every hour) calculates and stores historical change data on the number of risk points. Based on this historical change data, the module generates a risk velocity factor Vf for each blade across the entire field. The calculation procedure for this factor is deterministically defined as Vf = 1 + (Pcurrent − Phistory) / Pbase, where Pcurrent is the current number of risk points on the blade, and Phistory is the number of risk points on the blade at a preset historical time point (e.g., 24 hours ago). The risk point is the number of points in advance, and Pbase is a preset baseline value for the risk point number used for normalization, for example, it can be calibrated to 10,000 points; the numerical derivation example is: leaf C (risk stable), Pcurrent=5000, Phistory=5000, then Vf=1+(5000−5000) / 10000=1.0; leaf D (risk surge), Pcurrent=4900, Phistory=3700 (value 24 hours ago), then Vf=1+(4900−3700) / 10000=1.12; this Vf factor enables the control system to obtain the ability to perceive the first derivative of risk (i.e. trend).

[0025] In the outer loop execution phase of the control system, the control module calculates the product of the number of risk points and the risk speed factor (for example, the decision value of blade C is 5000 × 1.0 = 5000, and the decision value of blade D is 4900 × 1).The system uses 12≈5488 as the priority basis for decision-making; it further includes an opportunity identifier module, which is a standard meteorological data API interface in engineering implementation, used to obtain the future safe operation and maintenance window in high-altitude and cold environments (e.g., the time period within the next 48 hours when the wind speed is less than 10m / s, the visibility is greater than 2000m, and the temperature is higher than -20℃) as the system constraint; the optimal settlement controller module is responsible for solving the control law, which is used to solve an optimal control problem with the control objective of maximizing the total settlement amount (number of risk points × risk speed factor), based on the operation and maintenance risk exposure model (i.e., the decision value list of each blade) and the safe operation and maintenance window constraint; when solving this optimal control problem, the optimal settlement controller module is further configured to use the energy mode of the UAV platform The system uses a model or flight path, for example, a maximum operating time of 30 minutes per flight as a second constraint. The optimal control problem is constructed within the control domain as a path optimization problem with time windows and capacity constraints. The control module solves this problem using a standard heuristic algorithm, such as simulated annealing or a genetic algorithm, and ultimately generates a specific UAV operation instruction sequence. This instruction sequence is, for example,: UAV No. 1 takes off at 14:00, inspects in the order D->C, and is expected to return at 14:28. During the execution phase, the control problem solved by the optimal liquidation controller module is constructed as a path optimization problem with time windows and capacity constraints to ensure that limited maintenance resources are allocated to the blades with the highest risk level and fastest growth. The objective function is Maximize ∑i∈Assets (RiskScorei×Vfi)×δi aims to quantitatively maximize the total risk that can be cleared in this inspection operation. Here, RiskScorei is the current number of risk points for blade i at the time the optimal clearing controller module solves the problem; Vfi is the risk speed factor for blade i; and δi is a binary decision variable, which is 1 if and only if blade i is successfully assigned and completes inspection and clearing in the generated UAV operation instruction sequence, otherwise δi is 0. This solution process follows two mandatory constraints: one is the safety operation and maintenance window constraint, meaning the total duration of all UAV operations must not exceed the total duration of the safety operation and maintenance window determined by the opportunity identifier module. Secondly, there are constraints on the drone energy model, meaning the energy consumed by a single drone flight mission sequence must not exceed the safe threshold of the drone's battery capacity, such as maintaining 20% ​​battery power. The normalized baseline value Pbase for the risk speed factor Vf is determined based on retrospective statistical analysis of historical operational data. Specifically, the peak distribution of the risk point accumulation process for all blades between two periods of zeroing out risk points is statistically analyzed, and the 90th percentile of this distribution is used as the set value for Pbase. This value is set to 10,000 points by the engineering team to ensure that Vf is effectively amplified when the risk point count grows rapidly, rather than being triggered only after the risk point count reaches an extreme peak.

[0026] Finally, to achieve the adaptive characteristics of the control system, i.e., the model learns and evolves from its own decision results, the control module is further configured to execute a risk model self-calibration logic after the inspection task is completed. The control module first obtains the inspection result data generated by the inner loop logic; and performs closed-loop feedback calibration of the model based on the inspection result data. Its calibration procedure is deterministically divided into: Procedure 1, when the inspection result data indicates that no abnormalities were found, the system determines that its model's risk assessment of the blade is too pessimistic. At this time, the control module automatically lowers the number of risk points used to assess the future risk points of the preset blade while clearing the risk point count to zero (i.e., clearing is completed). The risk accumulation coefficient, for example, is reduced from 1.5 to 1.35 by multiplying it by a decay factor of 0.9, thus allowing the model to automatically converge to reality. In Procedure Two, when inspection results indicate a serious risk, the system determines that the risk has been confirmed and escalated. At this point, the risk model self-calibration logic is configured to prevent the risk point count from being zeroed out, instead forcibly setting it to a locked maximum value, such as 1,000,000 points, and immediately generating an alarm requiring manual intervention. This locked state will permanently maintain the highest priority for that blade in the control system until the control module receives a response from an external source, such as a maintenance work order (CMMS) system. The maximum value is unlocked only after a manual repair confirmation signal is received, thus forming a closed-loop control system for the industrial control system, from modeling to execution and then to model adaptation. To achieve the adaptive characteristics of the control system, the risk model self-calibration logic automatically lowers the blade risk accumulation coefficient when the inspection result indicates that no abnormality was found. The attenuation factor used in this action, such as 0.9, is an optimized value obtained from simulation backtesting tests on historical data. This test aims to evaluate the convergence speed of model parameters and the stability of inspection priority ranking under different attenuation factor settings: In the simulation, attenuation factors ranging from 0.8 to 0.99 are selected for testing, and the model parameters are set at 3... The evaluation indicators were convergence within 0 days and the minimum volatility of daily inspection priority ranking. Ultimately, 0.9 was selected as the optimal engineering trade-off value. This ensures that the risk accumulation coefficient can converge stably to the true engineering risk level based on actual inspection feedback, avoiding the long-term solidification of manually set risk coefficients (e.g., 1.5) or pessimistic estimates accumulated from historical events (e.g., 4500 points of emergency shutdown). At the same time, even if maintenance personnel set the risk accumulation coefficient through the manual calibration interface, the value range of this coefficient is limited to the engineering range of 0.5 to 3.0, and always follows the decay effect of the self-calibration logic, thereby preventing the failure of the decision-making logic due to manually set extreme values.

[0027] This invention also provides a method for inspecting wind turbine blades using a UAV in cold environments. This method is a flowchart reflecting the operational logic of the aforementioned system and includes the following steps: Step a, collecting image data of the blade surface through the blade inspection unit and simultaneously collecting real-time environmental data of the blade surface through the environmental monitoring unit; Step b, constructing a preset icing criterion based on the real-time environmental data; Step c, using the icing criterion to distinguish between icing areas and structural damage areas in the blade surface image data to generate inspection result data; Step d, acquiring real-time icing status data of key components of the UAV platform through the environmental monitoring unit; Step e, continuously comparing the real-time icing status data with a preset flight safety status threshold; Step f, and when the real-time icing status data triggers the preset flight safety status threshold, prioritizing the execution of a flight path replanning command to adjust the subsequent flight path of the UAV platform; Step g, until the real-time icing status data returns to within the preset flight safety status threshold.

[0028] Example 1: In a high-altitude wind farm where the UAV inspection system for wind turbine blades in a cold environment of the present invention has been deployed, the control system faces a dilemma of industrial process control and resource allocation during a winter cold wave. The wind farm asset list shows that blade A-5 is an old unit, whose risk accumulation coefficient has been manually adjusted to 1.5, and triggered an emergency shutdown event at a low temperature of -30℃ 24 hours ago. Blade B-2 is a stable middle-aged unit, whose risk points mainly come from the basic fatigue accumulation rules. Blade C-8 has experienced an SCA (Supervisory Capacity) event in the last 12 hours. The DA data began continuously displaying wind turbine icing status data; simultaneously, the opportunity identifier module learned that there was only a brief 3-hour safe operation and maintenance window within the next 72 hours. Faced with this control scenario of severe demand-opportunity mismatch, a traditional system relying on calendars or simple threshold alarms would be paralyzed in decision-making, unable to quantitatively rank the three assets with different states: A-5, B-2, and C-8. The control system of this invention then activates its outer-loop adaptive control logic. First, the risk accumulator module updates the operation and maintenance risk exposure model according to the risk accounting rules. For blade A-5, the control... The module not only responded to the emergency shutdown event data accumulation of +1000 points, but also activated the context calibration logic. Given that the event occurred at -30℃, which is below the calibration threshold of -25℃, a context risk coefficient of 3.0 was applied, correcting the accumulated points for the event to +3000 points. This was then multiplied by a manual calibration coefficient of 1.5, resulting in an accumulated +4500 points for this event alone. For blade C-8, the system accumulated +600 points over 12 hours, following a rule of +50 points per hour. Secondly, the risk velocity factor calculation module was activated, using Vf=1+(Pcurrent−Phi) The procedure of story) / Pbase is used for calculation. Blade A-5, due to its point count surging from a previous cumulative value of 2000 points to 6500 points (2000+4500) within 24 hours, has its Vf calculated as a higher value, such as 1.45. In contrast, the point count growth of blades B-2 and C-8 is gradual, and their Vf factors are close to 1.0. Thus, the control system, through the synergistic effect of the context risk coefficient and risk velocity factor, avoids the control contradiction of different risk sources, namely historical, current, and trend, which cannot be uniformly measured, and prioritizes the decision of A-5 (6500×1).(45≈9425) is definitively placed at a position far higher than B-2 or C-8; subsequently, the optimal liquidation controller module, with the control objective of maximizing the total liquidation amount (number of risk points × Vf), solves the problem under the constraints of a 3-hour safe operation and maintenance window and the UAV energy model. Its output UAV operation command sequence is determined to prioritize the inspection of blade A-5; when the UAV platform executes this command and arrives at blade A-5, its inner-loop control logic is activated. Image data collected by the blade inspection unit shows multiple anomalies on the blade surface that appear to be cracks. However, data from the environmental monitoring unit simultaneously acquired by the control module shows that the blade surface temperature is -5℃ and the humidity is 98%, triggering the preset icing criterion. The control module immediately applies this criterion to distinguish features in the image data, classifying all abnormal areas as icing areas, thus avoiding the dilemma of identifying icing and structural damage in high-altitude, cold environments.

[0029] After the UAV platform returns with the inspection results data, the outer loop logic of the control system initiates the final step: the risk model self-calibration logic. Since the inspection results data, i.e., the data after being differentiated by the icing criterion, indicates that no abnormalities were found, i.e., no structural damage was found, the control system determines that its previous risk assessment of the A-5 blade was too pessimistic. This assessment was influenced by the previous emergency shutdown event and the 1.5 human factor. Therefore, the control module executes the instruction to clear the risk points of the A-5 blade to zero, i.e., complete the clearance, and at the same time, further performs the model self-calibration action, i.e., automatically lowers the risk accumulation coefficient of the blade, for example, from 1.5 to 1.35. This step makes the control system's model, i.e., the risk accumulator module, and the feedback of its decision execution, i.e., the inspection results, form an adaptive closed loop. Through this operation, the control system not only clears the current maximum risk exposure, but also makes the model itself, on which it relies for decision-making, automatically converge towards a direction that is more in line with the reality of the specific asset A-5 project.

[0030] Example 2: To objectively verify the effectiveness of the control system of the present invention in optimizing resource allocation, a comparative experiment based on a simulation platform was conducted. This experiment aimed to evaluate the difference in risk mitigation efficiency between a system employing the complete control logic of the present invention (experimental group) and a simplified control system (control group) that prioritizes risks based solely on the absolute value of the current number of risk points, under typical, resource-constrained, high-altitude wind farm operation and maintenance scenarios. The experimental platform was constructed as a digital twin model containing 50 wind turbines, or 150 blades. This model could simulate wind turbine operation events in high-altitude environments, including emergency shutdowns, icing (triggered by simulated power curve changes), and turbulent winds exceeding design speeds. The frequency and severity of these events were set to random processes consistent with the statistical characteristics of such environments. Simultaneously, the simulation platform incorporated historical weather forecast data to generate a safe operation and maintenance window sequence with realistic constraints. The initial state of the experiment was set to 1000 initial risk points for all blades, followed by a 30-minute simulation. The simulation lasted for 30 days. During the simulation, two control systems, namely the experimental group and the control group, ran in parallel. They received the same simulated SCADA event stream and safety operation and maintenance window information stream. The control module of the experimental group fully implemented the risk accumulation rules, context adjustment logic, and risk speed factor Vf calculation procedure. Pbase was set to 10,000 points, and the goal was to maximize the total amount of liquidation (number of risk points × Vf). Combined with the UAV energy model, the maximum single operation time was set to 45 minutes for optimal liquidation control. The control module of the control group adopted simplified control logic. Its risk accumulation rules were the same as those of the experimental group, but it did not include context adjustment and risk speed factor calculation. Its optimal liquidation controller only made decisions with the goal of maximizing the total amount of liquidation of the current number of risk points. The key performance evaluation indicators of the experiment were set as the total number of risk points liquidated by the UAV successfully dispatched by each system during the 30-day simulation period, and the peak number of risk points reached by the blades not inspected during the simulation period.

[0031] On the 15th day of the simulation, a specific scenario emerged: Blade X experienced an emergency shutdown at -28℃, causing its risk points to jump to 7200 due to context calibration and event triggering accumulation rules, with a Vf factor of 1.52; Blade Y had been in an icing state for 36 consecutive hours, accumulating 6800 risk points, but due to continuous accumulation, its Vf factor was only 1.05; Blade Z belonged to an older unit, with a manual calibration coefficient of 2.0, accumulating 8000 risk points, but with no recent major events, its Vf factor was 1.0; at this point, the opportunity identifier module predicted only one 2-hour safe operation and maintenance window within the next 24 hours; the test group control module calculated the decision priority for each blade: Blade X (7200 × 1.52 ≈ 10944), Blade Y (6800 × 1.05 ≈ 7140), Blade Z ( 8000×1.0×2.0=16000); The control module of the control group only sorts the current risk points by the manual adjustment coefficient: blade Z (8000×2.0=16000), blade X (7200), blade Y (6800); Therefore, the operation instruction sequence generated by the optimal clearing controller module of the experimental group is to prioritize the inspection of blade Z, and then consider blade X, while the control group only regards blade Z as the highest priority; It should be noted that in this scenario, although the current risk points and coefficient product of blade Z are the highest, blade X has a higher urgency of potential risk due to emergency shutdown at low temperature and a higher risk speed factor. After the entire 30-day simulation cycle, the operation data of the two systems were statistically analyzed. Some key results are shown in Table 1. Table 1 is a comparison of the key performance indicators of the experimental group and the control group during the 30-day simulation cycle.

[0032] Table 1: Comparison of key performance indicators between the experimental group and the control group.

[0033]

[0034] Data analysis shows that the experimental group improved the total number of risk points cleared by approximately 33.8% compared to the control group. Simultaneously, the peak risk of uninspected blades under its control decreased by approximately 31.2%, and the number of high-risk lockouts triggered also significantly decreased. Analysis of simulation logs indicates that the difference in the experimental group's operational results mainly stems from its control logic's ability to perceive risk context and speed. This allows the limited maintenance window to be prioritized for blades in harsh environments, such as low temperatures, where events are occurring, or where risks are rapidly accumulating, rather than simply the blade with the highest absolute risk value. For example, in the scenario of day 15, although the experimental group prioritized clearing the blade Z with the highest risk point product, its decision-making based on the Vf factor allowed it to allocate resources to risk growth within subsequent windows. The faster blade X, while the control group may miss opportunities to address other pressing risks by focusing only on the blade with the highest absolute value; in addition, the risk model self-calibration logic of the experimental group was triggered 35 times during the simulation period, and the risk accumulation coefficient was reduced by an average of 8.5%, which shows that the model has the ability to learn from actual feedback and continuously optimize its risk assessment accuracy, thereby improving the utilization efficiency of resources during the window period; the data from this simulation test show that the control system of this invention, which includes context tuning, risk velocity factor and model self-calibration logic, can more effectively allocate operation and maintenance resources in high-altitude and cold environments, prioritize the handling of more urgent risks, and maintain the risk level of the entire industrial process at a low level, compared with the simplified control strategy based only on the current absolute value of risk.

[0035] Example 3: This example combines Figures 1 to 3 This document describes a system and method for inspecting wind turbine blades using unmanned aerial vehicles (UAVs) in cold environments. Figure 1 As shown, the system logically constructs a dual closed loop, including an inner-loop control logic responsible for platform survival and data validity, and an outer-loop adaptive control logic responsible for operational risk modeling and optimization. The inner-loop logic uses the UAV platform as the perception and execution terminal. Its blade inspection unit collects image data of the blade surface, and the environmental monitoring unit collects the blade surface and UAV's own status. This status data is sent to the flight safety status monitoring module to compare the real-time icing status with safety thresholds. When a threshold is triggered, the flight path replanning command, serving as the inner-loop safety instruction, is executed first. Simultaneously, image data and environmental data are fused to construct icing criteria, which are then used to distinguish between icing areas and structural damage, generating inspection result data. The outer-loop logic uses a risk accumulator module to construct an operational risk exposure model. Its output is used by a risk speed factor calculation module to perceive the risk increment trend. The optimal clearing controller module solves the optimal control problem based on this to generate a UAV operation instruction sequence, serving as the outer-loop control instruction. The inspection result data is used as feedback input to the risk model self-calibration logic. This logic adjusts the model based on the inspection results and acts on the risk accumulator module, thus forming a complete control closed loop.

[0036] like Figure 2 As shown in the figure, a bar chart compares the cumulative risk points of context-adjusted and non-context-adjusted modes in different environmental temperature ranges. The horizontal axis includes the cumulative risk points at <-25℃, -25℃ to -5℃, and >-5℃, while the vertical axis ranges from 0 to 3000 points. The data shows that in the <-25℃ range, the number of points with context adjustment (3000 points) is significantly higher than the number of points without context adjustment (1000 points). In the >-5℃ range, the number of points with context adjustment (800 points) is lower than the number of points without context adjustment (1000 points). In the -25℃ to -5℃ range, the two are basically the same at 1000 points, thus confirming that the adjustment logic can dynamically adjust the risk assessment according to the environmental context. Figure 3 As shown, the existing system lacks robustness in high-altitude and cold environments. The fundamental reason for this is the combined effect of four factors: environmental disturbances such as strong gusts and wind shear, low temperatures leading to icing, nonlinear changes in platform aerodynamic characteristics, model failure such as failure of the premise of relying on a stable environment, lack of environmental disturbance as feedback input, rapid failure of mathematical models during the process, rigid control logic such as the adoption of rigid countermeasure strategies, single and fixed control target of tracking absolute position, high-frequency and large-amplitude control output to correct errors, and static task planning such as static task planning that cannot be flexibly adjusted according to the actual state of the platform such as abnormal energy consumption.

[0037] Example 4: To further illustrate the contribution of specific technical features of the present invention, such as the risk velocity factor Vf and context tuning logic, to improving control performance, the following comparative experiment was conducted: Comparative Example 1: This comparative example uses the exact same simulation test platform, initialization conditions, simulated SCADA event flow, and safety operation and maintenance window information flow settings as Example 2; the only difference between it and the test group of Example 2, i.e., the system using the complete technical solution of the present invention, is that the control system used in this comparative example removes the calculation and application of the risk velocity factor Vf when the optimal liquidation controller module makes a decision, that is, its control objective function is simplified. The simulation is designed to maximize the total amount of risk points to be cleared while retaining the risk accumulation rules, context adjustment logic, and risk model self-calibration logic. This setting aims to simulate a control strategy that only considers the current absolute value of risk but lacks awareness of the risk development trend, in order to illustrate the effect of the introduction of the risk velocity factor Vf on the efficiency of the control system and resource allocation. On the 15th day of the simulation, the scenario in Example 2 is also faced: Blade X accumulates 7200 risk points due to emergency shutdown at low temperature (after context adjustment), Blade Y accumulates 6800 points due to continuous icing, and Blade Z accumulates 8000 points due to the old unit coefficient (multiplied by a coefficient of 2).The equivalent risk after 0 is 16000); at this time, since the control logic of this comparative model does not calculate the risk speed factor Vf, its optimal liquidation controller module only sorts the current number of risk points and the manual adjustment coefficient, and determines that blade Z (16000) has the highest priority, blade X (7200) is second, and blade Y (6800) is the lowest; therefore, when facing a safety operation and maintenance window of only 2 hours in the next 24 hours, the UAV operation instruction sequence generated by the control system of this comparative model is to prioritize the inspection of blade Z. After the inspection of blade Z is completed, the remaining window time may not be enough to complete the inspection of blade X or blade Y based on the range and energy constraints; compared with the decision of the experimental group in Example 2, which considered the Vf factor and placed blade X (decision priority 10944) before blade Y (7140), the control system of this comparative model failed to identify the urgency of the risk of blade X; after the end of the entire 30-day simulation cycle, the operation data of this comparative model control system was statistically analyzed, and its key The key performance index values ​​are basically consistent with those of the control group in Example 2, as shown in Table 1. Data shows that the total number of risk points cleared was 185,600, lower than the 248,300 in the experimental group of Example 2; the peak risk point count of uninspected blades reached 21,500, higher than the 14,800 in the experimental group; the number of times the maximum risk point lockout was triggered was 5, more than the 2 in the experimental group; the window period resource utilization rate, i.e., the effective operating time ratio, was 78%, lower than the 91% in the experimental group. Log analysis further indicates that, due to the lack of consideration for the speed of risk, the control system in this comparative model tends to allocate resources to blades with high absolute risk point counts but slow growth. For blades whose risk increases rapidly due to sudden events or continuous adverse operating conditions, the response priority is relatively lagging, resulting in missed opportunities for early intervention. This makes the risk of these blades more likely to accumulate to a higher level, even triggering lockout, thereby reducing the overall efficiency of risk clearing and increasing the system's risk peak.

[0038] Comparative Example 2: This comparative example also uses the exact same simulation test platform, initialization conditions, simulated SCADA event flow, and safety operation and maintenance window information flow settings as Example 2. The only difference between this comparative example and Example 2 is that the control system used in this comparative example removes the context calibration logic when the risk accumulator module executes the risk accounting rules. That is, all event-triggered accumulation rules use a fixed number of reference points that do not change with the environmental context for accumulation, while retaining the calculation and application of the risk velocity factor Vf and the risk model self-calibration logic. This setting aims to simulate a control strategy that can sense risk trends but cannot distinguish the severity differences of the same event under different environments, in order to compare and illustrate the role of context calibration logic in improving the fidelity of risk assessment. On the 15th day of the simulation, the scenario of Example 2 is faced again: blade X experiences a low temperature of -28℃. During an emergency shutdown, blade Y continues to freeze, while blade Z is an older unit. Because this comparative control system removed the context calibration logic, the emergency shutdown event for blade X only accumulates a fixed +1000 points, multiplied by a manual coefficient of 1.5, totaling +1500 points. Assuming its previous accumulated value was 2000 points, the current risk points are 3500. Blade Y continues to freeze, accumulating to 6800 points; blade Z accumulates to 8000 points (equivalent risk 16000). Then, the risk speed factor Vf is calculated: Vf for blade X = 1 + (3500 − 2000) / 10000 = 1.15; Vf for blade Y is 1.05; Vf for blade Z is 1.0. At this point, the comparative control system calculates the decision priority for each blade: Blade X (3500 × 1.15 ≈ 4025), Blade Y (6800 × 1.05 ≈ 7140), Blade Z (8000 × 1.0 × 2.05 ≈ 7140).(0=16000); In this case, the optimal clearing controller module determines that blade Z has the highest priority, followed by blade Y, and blade X has the lowest priority. Compared to the experimental group in Example 2, which was able to identify the high risk of emergency shutdown at low temperatures (decision priority 10944) through context tuning and place blade X before blade Y, the control system of this comparative version, due to its lack of context awareness, underestimated the risk of blade X, causing its inspection priority to be ranked after the continuously icing blade Y. After the entire 30-day simulation cycle, the operating data of this comparative version control system were statistically analyzed. The results show that, compared to the experimental group in Example 2, its risk is lower. The total risk points were low, while the peak risk of uninspected blades was high, triggering risk lockouts more frequently. Log analysis revealed that the lack of context calibration prevented the control system from assessing the differences in risk increments for the same event under different environmental conditions, resulting in a fidelity discrepancy between the number of risk points and the engineering risks they represent. Particularly in extremely cold environments, the impact of low temperatures on material brittleness makes the risk of load events at low temperatures far higher than at room temperature. This comparative control system failed to reflect this difference, leading to distorted decision-making and a failure to prioritize maintenance windows for risk events in high-risk contexts, thus impacting the overall risk control effectiveness.

[0039] Example 5: When the control system of this invention is first deployed in a specific wind farm, or when the model parameters need to be recalibrated after a period of operation, the engineering problem is how to set a set of initial parameters with engineering basis for the risk accumulator module, context calibration logic, risk velocity factor calculation module, and risk model self-calibration logic. To address this problem, the system adopts the following standardized offline parameter calibration procedure. This procedure utilizes the historical operating data and maintenance records accumulated by the wind farm, and its execution environment is a computing platform with data processing capabilities. The first step of the procedure is data preparation and preprocessing, starting from the wind farm's S... Extract operational data from the CADA database and the computerized maintenance management system (CMMS) for at least the past two years. This data includes hourly operating status, power, wind speed, ambient temperature, emergency shutdown event data, wind turbine icing status data (obtained by setting a deviation threshold between the power curve and the baseline curve; when the deviation exceeds 15% and the temperature is below 2℃, it is marked as icing), records of wind speeds exceeding design speeds (which can be set to continuous wind speeds exceeding 25 m / s), and fault records and maintenance work orders for all blades. The extracted data is then cleaned to remove erroneous or missing records. Alignment is performed by time and turbine / blade identifier; the second step of the procedure is the preliminary quantification of event risk weights, the purpose of which is to determine the basic risk points for the event triggering accumulation rule; blade breakage or the need for major maintenance is selected as the endpoint event, and the statistical correlation between the frequency or cumulative amount of various triggering events, such as emergency shutdown, icing duration, and over-wind speed duration, and the probability of the endpoint event is analyzed within a specific time window before the endpoint event occurs, such as the previous 6 months; using statistical methods such as survival analysis or logistic regression, the correlation between one emergency shutdown event and one hour of normal operating time (basic fatigue) is calculated. The failure rate increase factor corresponding to the cumulative failure rate is denoted as Rstop. Similarly, the failure rate increase factors Ricing and Rturbulence corresponding to one hour of icing and one hour of above-design wind speed are calculated. Based on the basic fatigue accumulation rule, i.e., +1 point per hour, the basic risk points of other events are set to be proportional to their relative risk factors. If Rstop is calculated to be approximately 1000 times the basic risk, then the basic points of emergency shutdown event data are set to +1000 points. Similarly, the basic points are set to +50 points / hour for icing and +10 points / hour for above-design wind speed.

[0040] The third step of the procedure is the calibration of the context risk coefficient, which aims to provide quantitative input for the context calibration logic. For events requiring context calibration, taking emergency shutdown as an example, historical data is grouped according to key context variables, such as ambient temperature, into three intervals: <-25℃, -25℃ to -5℃, and >-5℃. Within each temperature interval, the correlation or risk multiple Rstop,context between the emergency shutdown event and the subsequent blade failure rate is calculated separately. Then, the risk multiple for each interval is compared with the average risk multiple Rstop,avg calculated across all temperatures, which corresponds to the +1000-point baseline wind speed obtained in the second step. The risk multiples are compared to obtain the contextual risk coefficient for each interval, calculated using the formula: Coefficient = Rstop,context / Rstop,avg; where Coefficient is the contextual risk coefficient, Rstop,context is the risk multiple for a specific context interval, and Rstop,avg is the average risk multiple. If the calculated risk multiple in the <-25℃ interval is 3.0 times the average, the contextual risk coefficient for that interval is set to 3.0; if the risk multiple in the >-5℃ interval is only 0.8 times the average, the coefficient is set to 0.8. The coefficients are filled into the context-weighting rule matrix or lookup table; the fourth step of the procedure is to determine the normalized baseline value Pbase for the risk velocity factor; analyze the time series of the accumulation process of risk points for all blades in historical data between two inspections, i.e., when the risk point count is cleared; statistically analyze the numerical distribution when these cumulative sequences reach their peak, and select a statistic of this distribution, such as the 90th percentile or the mean plus twice the standard deviation, as the set value of the normalized baseline value Pbase; this setting aims to ensure that the denominator Pbase in the formula Vf=1+(Pcurrent−Phistory) / Pbase can reflect the risk accumulation of the wind farm blades. The scale of the product allows the Vf factor to have an amplifying effect when the number of risk points grows rapidly but has not yet reached an extreme peak; it can be set to 10,000 points. Here, Vf is the risk velocity factor, Pcurrent is the current number of risk points, Phistory is the number of risk points before the preset historical time point, and Pbase is the normalized baseline value. The fifth step of the procedure is setting the model self-calibration attenuation factor. The selection of the attenuation factor is an engineering consideration in balancing the model's convergence speed and stability. Through simulation, using some historical data as a training set, the system is simulated under different attenuation factor settings, such as from 0.8 to 0.99.01 represents the step size, and the performance is evaluated based on the following metrics: convergence speed of model parameters (i.e., the risk accumulation coefficient), stability of system decisions (i.e., the inspection priority ranking), and risk settlement efficiency. The attenuation factor value (e.g., 0.9) that enables model parameters to converge within a set time, avoids frequent decision fluctuations, and achieves the desired risk control effectiveness on the validation set is selected as the parameter for the risk model's self-calibration logic.

[0041] The sixth step of the procedure involves the initial establishment of the UAV energy model. Based on the performance parameters provided by the UAV manufacturer and the operating environment of the wind farm, such as average wind speed and altitude, an energy consumption estimation model is established. A base power consumption rate can be set, and correction coefficients are introduced according to flight speed, wind speed, and flight state, such as hovering, climbing, and level flight. The energy consumption of a single mission is estimated through integration. Based on the UAV battery capacity and safety margin, such as retaining 20% ​​of the battery, the maximum operating time or maximum range of a single flight, such as 45 minutes, is calculated as the second constraint for the optimal liquidation controller module. This model can be corrected online in subsequent operation based on actual flight data. By executing the above-mentioned standardized offline parameter calibration procedure containing six steps, the control system of this invention can obtain a set of reproducible initial parameter configurations based on historical data and engineering principles during initial deployment or periodic maintenance, thereby providing a calibrated basis for the subsequent online adaptive operation and optimization decision-making of the control system.

[0042] Example 6: When the control system of this invention is applied to a wind farm in a high-altitude, cold mountainous area with complex terrain and microclimate characteristics, in order to handle flight safety boundary conditions caused by sudden changes in the local environment, the system performs a preliminary safety procedure verification and parameter adaptation before deployment. This procedure includes using a digital elevation model and historical meteorological data to pre-plan and store multiple emergency return backup routes under different conditions for the site. At the same time, combined with the specific aerodynamic characteristics and anti-icing performance test data of the UAV platform, the flight safety state threshold is adjusted in a targeted manner. For example, for models with weak wind resistance or in areas known to be prone to strong turbulence, the icing threshold for triggering route replanning is set more conservatively, such as triggering it when the lift decreases by 15%. During a specific inspection mission, the UAV platform is flying close to the target blade D-3 according to the instruction sequence generated by the optimal clearing controller module. At this time, the icing status data of the key components of the UAV itself, which are obtained in real time by the environmental monitoring unit, increases sharply due to encountering local freezing rain clouds. Its value exceeds the aforementioned 15% lift decrease threshold within 5 seconds.

[0043] The inner-loop control logic of the control module immediately obtains the highest control priority based on preset safety rules and executes the route replanning command. This command first queries the current UAV position, altitude, attitude, and the latest wind speed and direction data provided by the environmental monitoring unit. Then, it selects an optimal emergency return path from the pre-stored backup route library, or by solving a path optimization problem that considers the current wind field and terrain constraints in real time. For example, it instructs the UAV to abandon the current inspection task, climb to a safe altitude of 300 meters above terrain obstacles, and return to the take-off and landing point along a path with the minimum headwind angle and no obstruction. Meanwhile, the control module transmits this safety trigger event and current flight status information back to the outer loop control logic in real time via the data link. Upon receiving this information, the optimal clearing controller module immediately suspends the current subsequent operation planning based on blade D-3 and updates the UAV's status to "emergency return." Simultaneously, based on the updated available UAV resources and the remaining safety maintenance window, it re-triggers the solution of the optimal control problem to dynamically adjust the operation command sequence of other UAVs or re-plan the priority of subsequent inspection tasks until the returning UAV lands safely and its icing status data returns to within the threshold.

[0044] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0045] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A system for inspecting wind turbine blades using unmanned aerial vehicles (UAVs) in cold environments, characterized in that, The system includes: The drone platform is equipped with a blade inspection unit and an environmental monitoring unit for monitoring the real-time status of the blade surface and the drone platform itself. The control module is configured to establish and execute the following operating rules: acquire image data of the blade surface collected by the blade inspection unit, and simultaneously acquire real-time environmental data of the blade surface collected by the environmental monitoring unit. Based on real-time environmental data, a preset icing criterion is constructed, and the icing criterion is used to distinguish between icing areas and structural damage areas in the blade surface image data to generate inspection result data; real-time icing status data of key components of the UAV platform itself collected by the environmental monitoring unit is acquired. The real-time icing status data is continuously compared with the preset flight safety status threshold. When the real-time icing status data triggers a preset flight safety status threshold, the flight path replanning command is executed first to adjust the subsequent flight path of the UAV platform until the real-time icing status data returns to within the preset flight safety status threshold.

2. The unmanned aerial vehicle (UAV) inspection system for wind turbine blades in high-altitude and cold environments according to claim 1, characterized in that, When constructing icing criteria based on real-time environmental data, the control module is further configured to: calculate a context risk coefficient based on the real-time environmental data; and use the context risk coefficient to dynamically adjust the discrimination threshold used in the icing criteria through a preset adjustment function relationship.

3. The unmanned aerial vehicle (UAV) inspection system for wind turbine blades in high-altitude and cold environments according to claim 1, characterized in that, The control module is further configured to: acquire inspection result data and execute risk model self-calibration logic based on the inspection result data; wherein when the inspection result data indicates that no abnormality was found, the control module automatically reduces the risk accumulation coefficient used to evaluate the future risk points of the preset blade while clearing the risk point count of the corresponding blade to zero.

4. The unmanned aerial vehicle (UAV) inspection system for wind turbine blades in a high-altitude, cold environment according to claim 3, characterized in that, The risk model self-calibration logic is further configured such that when the inspection result data indicates that a serious risk has been found, the control module is configured to prevent the risk point count from being cleared to zero, but instead forcibly set it to a locked maximum value and generate a manual intervention alarm. The locked maximum value is only released when the control module receives an external manual repair confirmation signal.

5. The unmanned aerial vehicle (UAV) inspection system for wind turbine blades in a high-altitude, cold environment according to claim 1, characterized in that, The control module further includes a risk accumulator module, which is used to dynamically accumulate the number of risk points of each blade in the entire field based on preset risk accounting rules and in response to real-time operation events of the wind turbine, thereby constructing an operation and maintenance risk exposure model.

6. The unmanned aerial vehicle (UAV) inspection system for wind turbine blades in a high-altitude, cold environment according to claim 5, characterized in that, The risk accounting rules include: a basic fatigue accumulation rule based on normal operating time; and at least one event-triggered accumulation rule, which is configured to respond to highly deterministic wind turbine operating data, including emergency shutdown event data and wind turbine icing status data identified by power curves.

7. The unmanned aerial vehicle (UAV) inspection system for wind turbine blades in a high-altitude, cold environment according to claim 5, characterized in that, The system further includes a risk velocity factor calculation module, which is used to periodically calculate and store historical change data of risk points, and generate a risk velocity factor Vf for each blade in the entire field based on the historical change data; where Vf=1+(Pcurrent−Phistory) / Pbase, where Pcurrent is the current number of risk points of the blade, Phistory is the number of risk points of the blade before a preset historical time point, and Pbase is a preset baseline value of risk points used for normalization; the control module uses the product of the number of risk points and the risk velocity factor as the priority basis for decision-making. The system further includes an opportunity identifier module, which is used to obtain the future safe operation and maintenance window in the high-altitude and cold environment as the system constraint condition; and an optimal settlement controller module, which is used to solve an optimal control problem based on the operation and maintenance risk exposure model and the safe operation and maintenance window constraint with the control objective of maximizing the total amount of risk point settlement, and generate a specific UAV operation instruction sequence.

8. A UAV inspection system for wind turbine blades in high-altitude and cold environments according to claim 7, characterized in that, When solving the optimal control problem, the optimal liquidation controller module is further configured to use the energy model or range model of the UAV platform as a second constraint.

9. A UAV inspection system for wind turbine blades in high-altitude and cold environments according to claim 5, characterized in that, The risk accounting rules further include a manual calibration interface, which is configured to allow maintenance personnel to set a risk accumulation coefficient for preset blades.

10. A method for inspecting wind turbine blades using a drone in a high-altitude, cold environment, characterized in that... Includes the following steps: Step a: Collect image data of the blade surface through the blade inspection unit, and simultaneously collect real-time environmental data of the blade surface through the environmental monitoring unit. Step b: Construct a preset icing criterion based on real-time environmental data; Step c: Use the icing criterion to distinguish between the icing area and the structural damage area in the blade surface image data in order to generate inspection result data; Step d: Real-time icing status data of key components of the UAV platform are obtained through the environmental monitoring unit. Step e involves continuously comparing real-time icing status data with preset flight safety status thresholds. Step f, and when the real-time icing status data triggers the preset flight safety status threshold, the route replanning command is executed first to adjust the subsequent flight path of the UAV platform. Step g continues until the real-time icing status data returns to within the preset flight safety status threshold.

Images