Railway unmanned inspection safety control method, system and electronic device

Abstract

The present application relates to the field of artificial intelligence and rail transit technology, and provides a railway unmanned inspection safety control method, system and electronic equipment, comprising: constructing a dynamic space-time risk field model based on multi-source environment perception data along the railway; generating an adaptive safety boundary based on the risk equipotential surface of the dynamic space-time risk field model; planning a risk avoidance flight path for the unmanned aerial vehicle based on the adaptive safety boundary, and obtaining real-time state data of the unmanned aerial vehicle during flight along the risk avoidance flight path; performing trajectory prediction based on the real-time state data, and triggering deterministic failure safety control when the predicted trajectory and the no-fly boundary meet a preset intersection condition, so as to convert the unmanned aerial vehicle into a falling state with a predictable ballistic trajectory. The present application generates an adaptive safety boundary through a dynamic space-time risk field model, and triggers failure safety control in combination with trajectory prediction, thereby realizing the transformation from passive response to active intervention, ensuring railway operation safety while taking into account inspection efficiency.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and rail transit technology, and in particular to a method, system and electronic equipment for unmanned railway inspection safety control. Background Technology

[0002] Traditional manual railway inspection methods face problems such as low efficiency, high cost, and significant safety risks to operators. Automated inspection using drones has become an important alternative. However, in the high-speed railway environment, it is crucial to ensure that drone operations do not introduce new risks to normal railway operations. Currently, drone safety measures in related technologies mostly employ passive or isolated protective methods, such as setting up simple static geofencing or returning to base after signal loss.

[0003] However, these measures cannot dynamically adapt to the complexity and real-time changes of the railway environment, and are insufficient to effectively address dynamic risk factors such as transient aerodynamic effects of trains, variable micro-weather conditions, and electromagnetic interference. Furthermore, when a drone experiences extreme malfunctions such as flight control or power failures, the system cannot actively intervene in its descent trajectory, making it difficult to ensure that drone debris does not intrude into railway clearances, including trains and overhead contact lines, potentially posing an irreversible threat to railway traffic safety. Summary of the Invention

[0004] This invention provides a railway unmanned inspection safety control method, system, and electronic equipment to address the shortcomings of existing UAV safety protection measures, which cannot dynamically adapt to changes in the complex railway environment and lack predictive failure control capabilities under extreme conditions, making it difficult to ensure railway operation safety.

[0005] This invention provides a method for unmanned railway inspection safety control, comprising: Acquire multi-source environmental sensing data along the railway line, and construct a dynamic spatiotemporal risk field model coupled with multi-physics fields based on the multi-source environmental sensing data. The dynamic spatiotemporal risk field model is used to represent the risk potential energy of a spatial point in the time dimension. Based on the risk equipotential surface of the dynamic spatiotemporal risk field model, an adaptive safety boundary is generated, which includes a no-fly zone boundary. Based on the adaptive safety boundary, a risk avoidance trajectory is planned for the UAV, and real-time status data of the UAV during flight along the risk avoidance trajectory is obtained; Based on the real-time status data, trajectory prediction is performed to generate the predicted trajectory of the UAV. When the predicted trajectory and the no-fly zone meet the preset intersection conditions, deterministic failure safety control is triggered to convert the UAV into a crash state with a predictable ballistic trajectory.

[0006] According to the present invention, a railway unmanned inspection safety control method is provided, wherein the construction of a dynamic spatiotemporal risk field model based on the multi-source environmental perception data includes: Based on the multi-source environmental perception data, multiple sub-risk marginal distribution functions are constructed respectively. The multiple sub-risk marginal distribution functions include at least a static geographical risk function, a train dynamic aerodynamic risk function, a micro-meteorological risk function, and a communication link risk function. Obtain the dependency parameters obtained by calibrating a predefined multivariate dataset using maximum likelihood estimation; Based on the aforementioned dependency parameters, a multidimensional Archimedes connection function is used to perform nonlinear coupling calculations on the multiple sub-risk marginal distribution functions to generate the dynamic spatiotemporal risk field model.

[0007] According to the present invention, a railway unmanned inspection safety control method is provided, wherein multiple sub-risk marginal distribution functions are constructed based on the multi-source environmental perception data, including: Based on the railway 3D map model data in the multi-source environmental perception data, the distance from spatial points to the obstacle surface is extracted, and the static geographic risk function is constructed according to the preset warning distance, the distance from the spatial points to the obstacle surface, and the risk growth rate coefficient determined by the virtual collision energy model. The train's real-time position and speed are extracted from the multi-source environmental perception data, and the train's dynamic aerodynamic risk function is constructed based on the train's real-time position and speed. Based on the meteorological grid data along the route from the multi-source environmental perception data, wind speed, gust factor and rainfall intensity are extracted, and the micro-meteorological risk function is constructed according to the wind speed, the gust factor, the rainfall intensity and the preset maximum wind resistance level and maximum rain resistance level. The real-time channel quality information of the communication base station is obtained from the multi-source environmental perception data, and the communication link risk function is constructed based on the real-time channel quality information and the minimum signal-to-noise ratio threshold required to maintain stable control.

[0008] According to the present invention, a railway unmanned inspection safety control method is provided, wherein constructing a dynamic aerodynamic risk function of the train based on the real-time position and speed of the train includes: The real-time position and speed of the train are input into a pre-trained deep learning proxy model to infer and output the pressure gradient and curl of the fluid velocity field at the target spatial point. The deep learning proxy model is trained based on a three-dimensional flow field distribution dataset. The train dynamic aerodynamic risk function is constructed based on the pressure gradient, the curl of the fluid velocity field, and the calibration coefficient.

[0009] According to the present invention, a railway unmanned inspection safety control method includes an adaptive safety boundary comprising multi-level risk regions. The method for generating the adaptive safety boundary based on the risk equipotential surface of the dynamic spatiotemporal risk field model includes: Configure a first risk probability threshold, a second risk probability threshold, and a third risk probability threshold, wherein the first risk probability threshold is greater than the second risk probability threshold, and the second risk probability threshold is greater than the third risk probability threshold; The risk equipotential surface in the dynamic spatiotemporal risk field model whose risk value is equal to the first risk probability threshold is defined as the no-fly boundary; The risk equipotential surface in the dynamic spatiotemporal risk field model in which the risk value is equal to the second risk probability threshold is defined as the prudent operation boundary. The risk equipotential surface in the dynamic spatiotemporal risk field model whose risk value is equal to the third risk probability threshold is defined as the conventional inspection boundary.

[0010] According to the present invention, a railway unmanned inspection safety control method is provided, wherein planning a risk avoidance trajectory for the UAV based on the adaptive safety boundary includes: The trajectory planning problem of the UAV is modeled as an optimal control problem, and an objective function for the optimal control problem is constructed. The objective function is a performance index functional that minimizes the flight time and the risk aversion coefficient, and the risk aversion coefficient is associated with the adaptive safety boundary. A fast traversal algorithm based on the level set method is adopted to transform the optimal control problem, which aims to minimize the objective function, into solving the corresponding functional equation. Solve the equation in the gridded dynamic spatiotemporal risk field model to obtain the optimal flight path that minimizes the objective function and satisfies the initial and final boundary conditions, and use the optimal flight path as the risk avoidance track.

[0011] According to the present invention, a railway unmanned inspection safety control method includes the following steps: trajectory prediction is performed based on the real-time status data to generate a predicted trajectory for the unmanned aerial vehicle (UAV); and deterministic failure safety control is triggered when the predicted trajectory and the no-fly zone meet a preset intersection condition. Nonlinear state estimation is performed on the real-time state data, and the posterior probability distribution of the state is approximated by the sampling points of the unscented transformation to obtain the state vector sequence and the corresponding covariance matrix sequence of the UAV in a future preset time period. Based on the state vector sequence, a predicted trajectory of the UAV is generated, and based on the covariance matrix sequence, a confidence ellipsoid of the predicted trajectory is generated. Calculate the probability of overlap between the confidence ellipsoid and the spatial region enclosed by the no-fly zone boundary; If the overlap probability is greater than a preset probability threshold, the preset intersection condition is determined to be met, and a trigger command is generated to execute the deterministic failure safety control.

[0012] According to the present invention, a railway unmanned inspection safety control method is provided, wherein the execution of the deterministic failure safety control includes: In response to the trigger command, the physical connection between the main power supply and the rotor of the UAV is instantly cut off by a miniature explosive device within a preset response time to perform physical isolation; The electromagnetic braking device is controlled to lock the UAV's arms in a preset horizontal deployment posture with target air resistance, thereby forming the fall state with predictable ballistic characteristics and a preset drag coefficient.

[0013] The present invention also provides a railway unmanned inspection safety control system, comprising: The multi-source environmental sensing module is used to acquire multi-source environmental sensing data along the railway line; A risk field modeling and calculation server is used to construct a dynamic spatiotemporal risk field model coupled with multiple physics fields based on the multi-source environmental perception data. The dynamic spatiotemporal risk field model is used to represent the risk potential energy of a spatial point in the time dimension. The risk field modeling and calculation server is also used to generate an adaptive safety boundary based on the risk equipotential surface of the dynamic spatiotemporal risk field model, and to plan a risk avoidance trajectory for the UAV based on the adaptive safety boundary, wherein the adaptive safety boundary includes a no-fly boundary. An autonomous flight control system for unmanned aerial vehicles (UAVs) is used to execute flight tasks corresponding to the risk avoidance flight path and to acquire real-time status data of the UAV during flight along the risk avoidance flight path. The deterministic failure safety module is used to perform trajectory prediction based on the real-time status data, generate the predicted trajectory of the UAV, and trigger deterministic failure safety control when the predicted trajectory and the no-fly zone meet a preset intersection condition, so as to convert the UAV into a crash state with a predictable ballistic trajectory.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the railway unmanned inspection safety control method as described above.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the railway unmanned inspection safety control method as described above.

[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the railway unmanned inspection safety control method as described above.

[0017] The railway unmanned inspection safety control method, system, and electronic equipment provided by this invention overcomes the limitations of traditional static and isolated defense by constructing a dynamic spatiotemporal risk field model coupled with multi-physics fields using multi-source environmental perception data and generating an adaptive safety boundary that includes no-fly zones. This allows for real-time and accurate quantification and adaptation to the extremely complex dynamic risk environment along railway lines. By planning risk avoidance trajectories for UAVs based on the adaptive safety boundary, the system can dynamically find the optimal flight path while avoiding triggering high-risk areas, effectively balancing safety and inspection efficiency. Furthermore, the system proactively intervenes by predicting the trajectory of real-time status data. Once a risk of intersection between the predicted trajectory and the no-fly zone is detected, deterministic failure safety control is immediately triggered, transforming the originally random and uncontrolled UAV into a crash state with a predictable ballistic trajectory. This prevents UAV debris from intruding into railway clearance and ensures railway operational safety. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the railway unmanned inspection safety control method provided by the present invention; Figure 2 This is a flowchart illustrating the railway unmanned inspection safety control method based on a dynamic spatiotemporal risk field provided by the present invention. Figure 3 This is a schematic diagram of the railway unmanned inspection safety control system provided by the present invention; Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] As a critical national infrastructure, the safe operation of railways is of paramount importance. Traditional manual railway inspection methods face numerous challenges, including low efficiency, high costs, significant safety risks to personnel, and unstructured data. In recent years, unmanned aerial vehicle (UAV) automated inspection technology has offered a solution. However, when introducing UAVs into the high-speed railway environment, the flight safety of the UAVs themselves becomes the primary prerequisite. It must be ensured that, under no circumstances will UAV inspection operations introduce any new safety risks to the normal operation of the railway.

[0022] In related technologies, safety measures for drone inspections are mostly passive or isolated. For example, simple static geofencing is often used to restrict flight areas, or conventional strategies such as signal loss triggering a return-to-home mechanism. However, the railway environment has strong dynamic spatiotemporal characteristics, and existing static, isolated protective measures cannot dynamically adapt to the complexity and real-time changes of the railway environment. These include the transient strong aerodynamic effects of high-speed trains, the variable micro-meteorological environment along the line, and complex situations such as GNSS (Global Navigation Satellite System) signal drift or communication interruptions caused by electromagnetic interference. Especially in extreme cases of drone flight control system failure or power system malfunction, ensuring that drone debris does not intrude into railway clearances, including trains and overhead contact lines, remains a significant technical challenge.

[0023] Therefore, there is an urgent need for a railway unmanned inspection safety control method that can dynamically quantify, predict, and proactively avoid risks, so as to ensure the efficiency of drone inspections while eliminating potential hazards to railway operation safety under extreme out-of-control conditions.

[0024] Figure 1 This is a flowchart illustrating the railway unmanned inspection safety control method provided by the present invention, as shown below. Figure 1 As shown, the method includes: Step S10: Obtain multi-source environmental perception data along the railway line, and construct a dynamic spatiotemporal risk field model with multi-physics coupling based on the multi-source environmental perception data. The dynamic spatiotemporal risk field model is used to represent the risk potential energy of a spatial point in the time dimension.

[0025] Specifically, in the automated inspection of railway infrastructure using unmanned aerial vehicles (UAVs), the flight environment of the UAVs is highly complex and volatile. In this embodiment of the invention, multi-source environmental perception data refers to a collection of various information reflecting the UAV's operating space and the surrounding railway's operational status. Its sources can encompass various sensors and information systems deployed at UAV automated airports and along railway lines. For example, multi-source environmental perception data may include train position and speed data based on train timetables and signaling systems; meteorological monitoring data such as wind speed, gusts, precipitation, and lightning based on gridded meteorological station data; electromagnetic and communication environment data acquired by GNSS signal quality monitors and 5G / LTE communication signal strength monitors; and high-precision railway BIM (Building Information Modeling) and 3D map data, etc.

[0026] After acquiring the aforementioned multi-source data, the system needs to construct a dynamic spatiotemporal risk field model coupled with multiple physics fields. To accurately quantify security threats in complex environments, this embodiment of the invention introduces the concept of a field in physics, establishing a continuous mathematical expression along the railway line in three-dimensional space and time. Specifically, this dynamic spatiotemporal risk field model abandons the traditional approach of simply linearly superimposing the probabilities of individual risk events. Instead, it nonlinearly couples various heterogeneous risks (such as geographical static risks, train aerodynamic risks, micro-meteorological risks, and communication link risks) as interconnected physical fields in three-dimensional space and time.

[0027] It is understandable that the dynamic spatiotemporal risk field model can be represented as a normalized risk function. Its range is [0,1], and it is used to represent the time point. spatial point The potential energy of risk, including spatial points Can be used by coordinates The risk potential energy is represented by the normalized risk function value calculated at each coordinate point in space at a specific time point. In other words, the risk potential energy is the numerical representation of the dynamic spatiotemporal risk field model at each specific spatiotemporal point. It is used to quantitatively characterize the potential probability and severity of an event threatening the flight safety of unmanned aerial vehicles or the operational safety of railways at a specific coordinate point in three-dimensional space at a specific time. The higher the value of the risk potential energy, the greater the safety threat presented at that spatiotemporal point in the risk field. Through this multiphysics coupling modeling, the system can accurately capture the risk amplification effect when extreme events occur concurrently (such as strong crosswinds accompanying high-speed train encounters and simultaneous communication signal attenuation).

[0028] Step S20: Based on the risk equipotential surface of the dynamic spatiotemporal risk field model, generate an adaptive safety boundary, which includes a no-fly zone boundary.

[0029] Specifically, risk equipotential surfaces refer to spatial surfaces formed by connecting three-dimensional spatial points with the same risk potential energy value in a constructed dynamic spatiotemporal risk field model. Since risk potential energy changes dynamically with time (such as the movement of a train or changes in gusts of wind), these equipotential surfaces are also changing in real time.

[0030] By setting different risk probability thresholds, the system can extract corresponding risk equipotential surfaces from a continuously changing risk field, and then generate an adaptive safety boundary. Compared to traditional static fences, the shape and extent of the adaptive safety boundary can flexibly shrink or expand according to changes in current multi-source environmental sensing data. For example, when no trains are passing and the microclimate is favorable, the safety boundary can be closer to the railway facilities to facilitate close-range, detailed inspection; while when a high-speed train approaches, the boundary will rapidly expand outward due to the sharp increase in aerodynamic risk.

[0031] The aforementioned adaptive safety boundary includes at least a no-fly boundary, which represents an absolutely inviolable red line. For example, the space surrounded by an equipotential surface with extremely high risk potential energy close to 1 can be extracted as the no-fly boundary. Under no circumstances can a drone intrude into the inner side of this boundary, thereby ensuring that drone debris will never intrude into the railway clearance, including trains, overhead contact lines, etc.

[0032] Step S30: Based on the adaptive safety boundary, plan a risk avoidance trajectory for the UAV and obtain real-time status data of the UAV during flight along the risk avoidance trajectory.

[0033] Specifically, after defining the adaptive safety boundary that dynamically changes over time, the system will plan risk-avoidance trajectories for the UAV. The essence of trajectory planning is, under the premise of satisfying the UAV's dynamic constraints and origin-end boundary conditions, to find a path in a dynamic risk field that avoids high-risk areas, especially absolutely avoids entering no-fly zones, while also optimizing flight time or energy consumption. This trajectory planning strategy can utilize low-risk windows within the risk field to maximize the UAV's inspection coverage and execution efficiency, avoiding the efficiency losses caused by traditional static no-fly zones.

[0034] During the flight of the UAV to perform inspection tasks according to the planned risk avoidance flight path, the system will continuously acquire its real-time status data. This real-time status data includes, but is not limited to, the UAV's current three-dimensional position coordinates, three-axis velocity, acceleration, attitude angles (yaw, pitch, roll), and the speed of each motor, which reflect the current kinematic and dynamic performance of the UAV.

[0035] Step S40: Based on the real-time status data, trajectory prediction is performed to generate the predicted trajectory of the UAV. When the predicted trajectory and the no-fly zone meet the preset intersection conditions, deterministic failure safety control is triggered to convert the UAV into a crash state with a predictable ballistic trajectory.

[0036] Specifically, due to sudden changes in the natural environment or unexpected hardware failures of the UAV (such as power system failure or flight control anomalies), the actual flight trajectory may deviate from the planned path. Therefore, the system needs to predict the trajectory based on real-time state data. This process typically employs high-precision state estimation algorithms to predict the UAV's motion trend over a future period based on its historical and current state, thereby generating a predicted trajectory. Due to environmental noise and model uncertainties, this predicted trajectory is often not a single line but rather a three-dimensional spatial distribution area containing a certain confidence probability.

[0037] When the system determines that the predicted trajectory and the no-fly zone boundary meet the preset intersection conditions, for example, if the probability that the predicted drone will run out of control and intrude into the no-fly zone boundary in the next few seconds exceeds the minimum safety threshold that the system can tolerate, the system will no longer wait for the actual boundary violation to occur, but will instead perform pre-emptive intervention and immediately trigger deterministic failure safety control.

[0038] The deterministic failure safety control mentioned here refers to the system's proactive abandonment of attempts to salvage the drone's normal flight attitude when facing high-risk extreme conditions, instead employing a highly deterministic physical intervention. Its core objective is to transform the drone into a falling state with a predictable ballistic trajectory. Under normal uncontrolled conditions, due to motor failure, rotor damage, or aerodynamic turbulence, the drone's falling trajectory is disordered and extremely divergent, easily leading to a major accident by crashing into railway clearances. This invention, through specific actuators, such as instantly severing the physical power connection and forcibly fixing the fuselage attitude to exhibit maximum wind resistance, eliminates all uncontrollable aerodynamic forces on the drone, forcibly transforming it into a falling object with stable aerodynamic characteristics conforming to classical ballistic physics. Because the falling state becomes predictable, and the timing of the trigger control is based on a pre-predicted trajectory, it ensures that even if the drone crashes, its final landing point will inevitably fall outside the no-fly zone, thereby guaranteeing railway operational safety.

[0039] The method provided in this invention overcomes the limitations of traditional static and isolated defense by constructing a dynamic spatiotemporal risk field model coupled with multi-physics fields using multi-source environmental perception data and generating an adaptive safety boundary that includes no-fly zones. It can quantify and adapt to the extremely complex dynamic risk environment along railway lines in real time and with high accuracy. By planning risk avoidance trajectories for UAVs based on the adaptive safety boundary, it can dynamically find the optimal flight path while avoiding triggering high-risk areas, effectively balancing safety and inspection efficiency. Furthermore, the system intervenes in advance by predicting the trajectory of real-time status data. Once a risk of intersection between the predicted trajectory and the no-fly zone is detected, deterministic failure safety control is immediately and proactively triggered, transforming the originally random and uncontrolled UAV into a crash state with a predictable ballistic trajectory. This prevents the possibility of UAV debris intruding into railway clearance and ensures railway operational safety.

[0040] Based on the above embodiments, step S10, which involves constructing a dynamic spatiotemporal risk field model coupled with multiple physics fields based on the multi-source environmental perception data, includes: Step S11: Based on the multi-source environmental perception data, construct multiple sub-risk marginal distribution functions, which include at least a static geographical risk function, a train dynamic aerodynamic risk function, a micro-meteorological risk function, and a communication link risk function.

[0041] Specifically, before quantifying the comprehensive risk at any point in the three-dimensional space along the railway at a specific time, it is necessary to first model each type of independent risk source separately. The sub-risk marginal distribution function here refers to the independent and normalized (e.g., mapped to the [0,1] interval) risk probability distribution model established for each single risk factor after analyzing the multi-source environmental perception data.

[0042] Specifically, to comprehensively cover the complex threats faced by railway drone inspections, the constructed sub-risk marginal distribution function includes at least a static geographical risk function, a train dynamic aerodynamic risk function, a micro-meteorological risk function, and a communication link risk function. Among these, the static geographical risk function... This primarily reflects the spatial potential energy of a drone colliding with fixed structures such as railway clearances and overhead contact lines; the train's dynamic aerodynamic risk function. This primarily reflects the degree of interference between the transient strong aerodynamic effects (such as strong pressure gradients and vorticity) generated when a high-speed train passes and the flight attitude of the UAV; micro-meteorological risk function. The aim is to quantify the risk of loss of control caused by the variable microclimates along the route (such as gusts, rainfall, and lightning); communication link risk function. It is used to assess the probability of loss of connection or loss of control caused by 5G / LTE signal attenuation or electromagnetic interference.

[0043] Step S12: Obtain the dependency parameters obtained by calibrating the preset multivariate dataset using maximum likelihood estimation.

[0044] Specifically, in real physical environments, risk factors often do not exist in isolation; for example, severe weather is often accompanied by a decline in communication signal quality. Therefore, embodiments of the present invention abandon the traditional assumption of simply linearly superimposing or multiplying the probabilities of independent events, and instead explore the strength of the intrinsic correlation between various risk factors.

[0045] A pre-defined multivariate dataset typically refers to a heterogeneous dataset composed of massive amounts of historical meteorological data, actual train operation data, communication monitoring data, and simulation data. Dependency parameters (such as parameters) The key variable representing the correlation between different risk dimensions is the maximum likelihood estimation (MLE) algorithm. The system uses the maximum likelihood estimation (MLE) algorithm to perform offline calibration on the above-mentioned preset multivariate dataset, thereby accurately extracting the dependency parameters that can quantify the intrinsic correlation strength of each risk factor.

[0046] Step S13: Based on the dependency parameters, a multidimensional Archimedes connection function is used to perform nonlinear coupling calculations on the multiple sub-risk marginal distribution functions to generate the dynamic spatiotemporal risk field model.

[0047] Specifically, after obtaining the independent sub-risk marginal distribution functions and their dependency parameters, the system uses a multidimensional Archimedean connection (Copula) function to perform final model fusion, generating a normalized dynamic spatiotemporal risk field model. The calculation formula is as follows: in, It is a multidimensional Archimedes Copula function, preferably Gumbel Copula, because it can effectively capture the positive correlation between variables at the tail of the distribution, that is, extreme risk events tend to occur simultaneously.

[0048] Understandably, compared to simple linear superposition, this embodiment of the invention employs a multidimensional Archimedes connection function for nonlinear coupling calculation, which can extremely effectively capture the positive correlation between multiple variables at the tails of the distribution. In actual inspection scenarios, this means that the function can capture the objective physical law that extreme risk events tend to occur simultaneously. Through this nonlinear coupling, when the values ​​of multiple sub-risk factors simultaneously approach the high-risk range, the comprehensive risk potential energy output by the dynamic spatiotemporal risk field model will be nonlinearly amplified, thereby accurately characterizing the risk state when multiple physical fields are superimposed.

[0049] The method provided in this invention achieves full coverage of risk sources by establishing four sub-risk models: static geography, train aerodynamics, micrometeorology, and communication links. By introducing multidimensional Archimedes connection functions and maximum likelihood estimation, the complex correlations between various risk sources can be accurately quantified, thereby improving the accuracy of comprehensive risk assessment and the system's ability to predict and adapt to extremely complex scenarios.

[0050] Based on any of the above embodiments, step S11 specifically includes: Step S111: Based on the railway 3D map model data in the multi-source environmental perception data, extract the distance from the spatial point to the obstacle surface, and construct the static geographic risk function according to the preset warning distance, the distance from the spatial point to the obstacle surface, and the risk growth rate coefficient determined by the virtual collision energy model.

[0051] Specifically, when constructing the static geographic risk function, the system first uses high-precision 3D railway map model data (such as railway BIM model) to propose the current spatial point. Distance to the nearest obstacle surface (denoted as ) Based on this, a preset warning distance (denoted as ) was introduced. This serves as the basic setting. Here, the preset warning distance can be set based on the provisions of the Railway Safety Management Regulations regarding safety protection zones (such as not less than 10 meters from the toe of the embankment or the outer edge of the top of the cutting slope), and then finely adjusted in conjunction with the BIM model of the specific line.

[0052] To make the risk value transition more consistent with physical laws, this embodiment of the invention also introduces a risk growth rate coefficient (denoted as ) determined by a virtual collision energy model. This coefficient is based on the possible intrusion speed of the drone (denoted as ). ) and the collision energy threshold that may cause structural damage (denoted as The calibration formula, derived from joint calibration, ensures that the risk can quickly saturate to 1 during high-speed intrusion. .

[0053] Combining the parameters mentioned above, the static geographic risk function is typically constructed as a non-linear Sigmoid function, as shown below: The above construction method ensures that when the spatial point is outside the safe warning distance, the risk value smoothly and slowly decays to an extremely low level; while when the drone approaches or even crosses the warning distance, the risk value will increase rapidly and exponentially with the risk growth rate coefficient and approach the saturation value of 1, thus accurately characterizing the nonlinear mutation characteristics of physical collision risk.

[0054] Step S112: Extract the real-time position and speed of the train from the multi-source environmental perception data, and construct the dynamic aerodynamic risk function of the train based on the real-time position and speed of the train.

[0055] Specifically, high-speed trains generate strong aerodynamic flow fields during operation, which is a unique risk source for railway inspection. The system extracts the real-time location and speed of the target train through sensing units or signaling systems along the line, uses this traffic operation data as input, and converts it into the probability of microscopic aerodynamic interference at the location of the UAV, thereby constructing a dynamic aerodynamic risk function for the train.

[0056] Further, in step S112, constructing the train dynamic aerodynamic risk function based on the train's real-time position and its operating speed includes: The real-time position and speed of the train are input into a pre-trained deep learning proxy model to infer and output the pressure gradient and curl of the fluid velocity field at the target spatial point. The deep learning proxy model is trained based on a three-dimensional flow field distribution dataset. The train dynamic aerodynamic risk function is constructed based on the pressure gradient, the curl of the fluid velocity field, and the calibration coefficients.

[0057] It should be noted that, in order to meet the requirements of millisecond-level real-time early warning, this embodiment of the invention introduces a deep learning proxy model trained on a three-dimensional flow field distribution dataset.

[0058] Specifically, the system runs professional Computational Fluid Dynamics (CFD) software (such as ANSYS Fluent) offline to generate massive amounts of fluid simulation data for different vehicle models (such as CR400AF, CRH380A, etc.), different speed ranges (such as 50-350 km / h), and different intersection conditions, thereby establishing a huge three-dimensional flow field distribution dataset. The deep learning proxy model (for example, a network architecture with powerful spatial feature extraction capabilities such as U-Net) is trained and converged based on this huge dataset.

[0059] During real-time online inspection, the system only needs to input the train type, real-time train location, and train speed into the deep learning agent model. The model can then complete the nonlinear mapping in an extremely short time (milliseconds) and directly infer and output the pressure gradient and the curl (i.e., vorticity) of the fluid velocity field at the target spatial point. These two physical quantities are the core fluid dynamic parameters that determine the UAV's attitude tumbling or stall.

[0060] After obtaining the current hydrodynamic parameters of a spatial point, these parameters need to be converted into normalized risk values. The system sets calibration coefficients corresponding to the pressure gradient and velocity field curl. These calibration coefficients are determined by combining real physical flow field data obtained from particle image velocimetry experiments on scaled-down train and UAV models in a wind tunnel environment with CFD simulation data through multiple linear regression analysis, ensuring a strong correlation between the model's output risk value and the actual physical effects.

[0061] The system constructs the train dynamic aerodynamic risk function based on the pressure gradient, the curl of the fluid velocity field, and the calibration coefficients, using the following proportional mathematical model: in, It is a pressure gradient. It is the curl (i.e., vorticity) of the fluid velocity field. This is the calibration coefficient. The aerodynamic risk value calculated from this can reflect both the drastic pressure change caused by the repulsive airflow at the front of the high-speed train and the suction effect generated by the wake vortex at the rear of the train.

[0062] Step S113: Based on the meteorological grid data along the route in the multi-source environmental perception data, extract wind speed, gust factor and rainfall intensity, and construct the micro-meteorological risk function according to the wind speed, the gust factor, the rainfall intensity and the preset maximum wind resistance level and maximum rain resistance level.

[0063] Specifically, the micro-meteorological environment along railway lines is complex and variable. After acquiring discrete meteorological data through small automatic weather stations, the system can use spatial interpolation algorithms such as Kriging interpolation to generate continuous meteorological grid data along the railway line, from which wind speed can be extracted. Gust factor and rainfall intensity Key meteorological indicators, etc.

[0064] To ensure these indicators have a consistent meaning in risk measurement, the system incorporates the maximum wind resistance level preset in the UAV flight control manual. and maximum rain resistance level Using this as the denominator, we construct the micro-meteorological risk function in the following form: In addition, as a fallback mechanism for extreme situations, when thunderstorms or severe convective weather are predicted, the micro-meteorological risk function value for the region will be directly set to 1.

[0065] Step S114: Obtain real-time channel quality information of the communication base station from the multi-source environmental perception data, and construct the communication link risk function based on the real-time channel quality information and the minimum signal-to-noise ratio threshold required to maintain stable control.

[0066] Specifically, drone loss of control often stems from the disconnection of control signals. The system continuously monitors and acquires real-time channel quality information from the communication base station, such as RSRP (Reference Signal Receiving Power) and SINR (Signal to Interference plus Noise Ratio). The system incorporates a minimum signal-to-noise ratio threshold, calibrated through extensive field testing, required for stable control. The communication link risk function can also employ smoothing mapping functions such as the Sigmoid function to quantify the probability that the current channel quality will drop below the aforementioned minimum threshold. The specific formula is as follows: in, The minimum signal-to-noise ratio threshold required to maintain stable control. The risk growth coefficients are all calibrated using a large amount of field test data. This function represents real-time channel quality information. It comprehensively considers signal strength attenuation, multipath effects, and potential electromagnetic interference sources such as substations, intuitively transforming the abstract communication signal strength into the probability risk of the drone losing contact or control.

[0067] Based on any of the above embodiments, the adaptive security boundary includes multi-level risk areas, and correspondingly, step S20 specifically includes: Step S21: Configure a first risk probability threshold, a second risk probability threshold, and a third risk probability threshold, wherein the first risk probability threshold is greater than the second risk probability threshold, and the second risk probability threshold is greater than the third risk probability threshold; Step S22: Define the risk equipotential surface in the dynamic spatiotemporal risk field model whose risk value is equal to the first risk probability threshold as the no-fly boundary; Step S23: Define the risk equipotential surface in the dynamic spatiotemporal risk field model whose risk value is equal to the second risk probability threshold as the prudent operation boundary; Step S24: Define the risk equipotential surface in the dynamic spatiotemporal risk field model whose risk value is equal to the third risk probability threshold as the conventional inspection boundary.

[0068] Specifically, in the dynamic spatiotemporal risk field model, the risk potential energy is a continuously changing value. To facilitate the decision-making and execution of the flight control system, the system discretizes the continuous risk field into multi-level risk regions by configuring a first risk probability threshold, a second risk probability threshold, and a third risk probability threshold (e.g., set to 0.9, 0.5, and 0.1 respectively). These three thresholds, which decrease in degree, represent the system's gradient perception of safety threats from highly dangerous to relatively safe.

[0069] The system extracts a three-dimensional spatial surface where the risk value equals the first risk probability threshold (e.g., 0.9, representing an extremely high-risk critical point). This risk equipotential surface in the dynamic spatiotemporal risk field model, where the risk value equals the first risk probability threshold, is defined as a no-fly boundary. In the system's visualization interface, this is typically marked as a red absolute no-fly boundary, as shown below: in, Indicates no-fly zone boundaries. This represents the first risk probability threshold. The area inside this boundary is a high-risk intersection zone. Regardless of the drone's operational status, it must never cross this equipotential surface under any circumstances, thus serving as the final virtual barrier to ensure the absolute safety of the railway clearance.

[0070] Furthermore, the system defines the risk equipotential surface in the dynamic spatiotemporal risk field model, where the risk value equals the second risk probability threshold, as the prudent operating boundary, which is typically marked in yellow in the system's visualization interface. Specifically, it can be represented as follows: in, Indicates the boundaries of prudent operations. This represents the second risk probability threshold.

[0071] When a drone crosses the boundary from the safe zone into the cautious operation zone for inspection purposes, the flight control system will automatically take preventative measures, such as increasing the priority of the drone's flight stability control algorithm, adjusting the attitude response gain, and simultaneously increasing the frequency of transmitting video and telemetry data back to the ground station to ensure that the operation in this moderately risky area is safe and under control.

[0072] Finally, the system defines the risk equipotential surface in the dynamic spatiotemporal risk field model, where the risk value equals the third risk probability threshold, as the regular inspection boundary, which is typically marked in green in the system visualization interface. Specifically, it can be represented as follows: in, Indicates the boundaries of routine inspections. This represents the third risk probability threshold.

[0073] The area outside this equipotential surface is considered a low-risk safety zone, within which drones can freely perform routine automated inspection tasks according to a predetermined plan. It should be understood that this third risk probability threshold is typically set in the system's underlying logic through large-scale Monte Carlo simulations to ensure that even in the worst-case scenario of a failure at this boundary, the drone's crash trajectory will be confined outside the red no-fly zone.

[0074] The method provided in this invention introduces multi-level risk probability thresholds to transform a continuous spatiotemporal risk field into an intuitive and executable three-layer dynamic safety equipotential surface boundary. This step-like safety space division mechanism enables drones to both adhere to the absolute no-fly zone with the highest safety standards and achieve efficient and precise close-range inspections within a prudent area by enhancing flight control mechanisms, thus achieving a balance between safety and operational efficiency.

[0075] Based on any of the above embodiments, in step S30, the step of planning a risk avoidance trajectory for the UAV based on the adaptive safety boundary includes: Step S31: Model the UAV's trajectory planning problem as an optimal control problem, and construct the objective function of the optimal control problem. The objective function is a performance index functional that minimizes flight time and risk aversion coefficient, and the risk aversion coefficient is associated with the adaptive safety boundary. Step S32: Using a fast traversal algorithm based on the level set method, the optimal control problem with minimizing the objective function as the optimization objective is transformed into solving the corresponding functional equation. Step S33: Solve the equation in the gridded dynamic spatiotemporal risk field model to obtain the optimal flight path that minimizes the objective function and satisfies the initial boundary conditions and the termination boundary conditions, and use the optimal flight path as the risk avoidance track.

[0076] It should be noted that traditional UAV trajectory planning often employs simple rule-based obstacle avoidance algorithms (such as A...). Algorithms, etc., are insufficient to handle the risks of multiphysics environments that are subject to highly dynamic changes. Therefore, this invention models the UAV trajectory planning problem as an optimal control problem.

[0077] Specifically, the goal of this optimal control problem is to find a control input. (such as acceleration), which affects the drone's state. While satisfying the boundary conditions (start point and end point), the system minimizes flight time and risk. To solve this optimal control problem, the system constructs an objective function for the optimal control problem. This objective function is a performance index functional that minimizes flight time and risk aversion coefficients. Specifically, it is expressed as follows: in, This is the risk aversion coefficient. The integral term of this functional includes both a constant term representing flight time and a penalty term composed of the dynamic risk field potential energy and the risk aversion coefficient. The risk aversion coefficient is an adjustable parameter and is related to the adaptive safety boundary. When this coefficient is set larger, it means that the system has a lower tolerance for risk. The objective function will be driven by a greater risk penalty during optimization, and the planned path will tend to be further away from high-risk no-fly zones and prudent operation boundaries. It will choose to take a longer route and increase flight time in order to obtain an absolute safety margin.

[0078] The aforementioned optimal control problem theoretically corresponds to the complex Hamilton-Jacobi-Bellman (HJB) equations. Due to the complexity of the UAV's state space and the dynamic nature of the risk field, these equations typically lack analytical solutions, making direct solution extremely difficult. To achieve millisecond-level real-time trajectory planning, this embodiment of the invention employs a Fast Marching Method based on the Level Set Method. This algorithm transforms the optimal control problem, which aims to minimize the objective function, into solving the corresponding functional equations (i.e., the Eikonal equations).

[0079] Specifically, the system solves the Eikonal equation within a gridded dynamic spatiotemporal risk field model. Starting from the UAV's current position (initial boundary condition), it rapidly extrapolates the arrival cost to each grid point outwards until the target position is reached (termination boundary condition). During this extrapolation, high-risk areas act as high-drag media, advancing slowly, while low-risk areas advance rapidly. Through this mechanism, the algorithm efficiently calculates the globally or locally optimal flight path, thereby obtaining the optimal flight path that minimizes the objective function and satisfies both the initial and termination boundary conditions. Finally, this optimal flight path is issued to the UAV as a risk-avoidance trajectory for execution.

[0080] The method provided in this invention models trajectory planning as minimizing a functional objective that balances time efficiency and risk. By introducing a fast travel algorithm based on the level set method, the complex solution process is transformed into efficient grid numerical computation of the functional equation. This method is particularly suitable for handling complex scenarios where adaptive safety boundaries dynamically change due to train movement or sudden weather changes, and achieves efficient planning of real-time globally optimal safe paths in complex physical fields.

[0081] Based on any of the above embodiments, step S40 specifically includes: Step S41: Perform nonlinear state estimation on the real-time state data, approximate the posterior probability distribution of the state through sampling points of unscented transformation, and obtain the state vector sequence and corresponding covariance matrix sequence of the UAV in a future preset time period. Step S42: Based on the state vector sequence, generate the predicted trajectory of the UAV, and based on the covariance matrix sequence, generate the confidence ellipsoid of the predicted trajectory. Step S43: Calculate the probability of overlap between the confidence ellipsoid and the spatial region enclosed by the no-fly boundary; Step S44: If the overlap probability is greater than a preset probability threshold, determine that the preset intersection condition is met, and generate a trigger command to execute the deterministic failure safety control.

[0082] Specifically, the flight dynamics model of a UAV exhibits strong nonlinear characteristics. Especially when encountering gusts of wind or partial power failure, the traditional Extended Kalman Filter (EKF) based on a first-order linearized approximation is prone to producing large estimation errors. Therefore, this embodiment of the invention employs the Unscented Kalman Filter (UKF) algorithm to perform nonlinear state estimation on real-time state data.

[0083] In this process, the system approximates the posterior probability distribution of the state through sampling points obtained from an unscented transformation. These sampling points (often called Sigma points) are a set of discrete points selected according to specific deterministic rules near the current state estimate. These points accurately capture the mean and covariance of the UAV's true nonlinear state distribution. By substituting these Sigma points into the nonlinear state transition equation for propagation, the system can obtain the UAV's state vector sequence and corresponding covariance matrix sequence within a preset future time period (e.g., the next 3-5 seconds) without needing to perform derivative calculations (Jacobi matrix calculations) on the complex model. Indicates time The state vector, Indicates time The covariance matrix is ​​given by the following: The state vector sequence represents the expected predicted position and velocity of the UAV at various future time points; the covariance matrix sequence represents the degree of uncertainty of these predicted values ​​in three-dimensional space.

[0084] After obtaining the above sequence, the system generates the predicted trajectory of the UAV based on the state vector sequence, which is a three-dimensional spatial curve connecting discrete expected future locations. Simultaneously, due to prediction errors, the system generates a confidence ellipsoid of the predicted trajectory based on the covariance matrix sequence. Here, the confidence ellipsoid refers to a three-dimensional spatial probability envelope constructed with a point on the predicted trajectory as its center, based on a specific confidence level. It intuitively represents the boundary of all possible spatial locations where the UAV may appear at a specific future time, after considering all environmental disturbances and model errors.

[0085] After generating a confidence ellipsoid containing error probability boundaries, the system needs to compare its spatial topological relationship with the defined absolute safety red line (i.e., no-fly zone boundary) to calculate the overlap probability of the spatial region enclosed by the confidence ellipsoid and the no-fly zone boundary (such as a red equipotential surface). This overlap probability is typically calculated by comparing the volume of the three-dimensional confidence ellipsoid with that of the red no-fly zone boundary. The volume enclosed It is obtained by integrating the intersection of three-dimensional space.

[0086] When the overlap probability is greater than a preset probability threshold (e.g., a system-set tolerance limit) When this occurs, it means that not only the single-point predicted trajectory of the UAV, but also its potential error fluctuation range, has substantially constituted a serious threat of intruding into the railway clearance. At this point, the system immediately determines that a preset intersection condition is met and generates a trigger command to execute deterministic failure safety control. This trigger condition can be specifically represented as follows: in, Indicates the probability of overlap. This represents the predicted state vector of the drone at a future time, where Representing the current moment, This represents the predicted future time step. Indicates the time up to the current moment. The set of all measurements constitutes the prior data basis for trajectory prediction, including historical and current flight status data (such as position, speed, attitude) transmitted back by UAV sensors, as well as multi-source environmental perception data (such as wind speed, wind direction, etc.).

[0087] The method provided in this invention overcomes the errors caused by linearization of traditional nonlinear models by introducing unscented Kalman filtering (UKF), and improves the accuracy of state estimation and trajectory prediction of UAVs under extreme conditions by using Sigma points to accurately approximate the state distribution. At the same time, it achieves the accurate quantification of potential crash violation risks by using the three-dimensional spatial overlap probability of the confidence ellipsoid and the no-fly zone boundary for trigger determination, thus transforming safety control from post-event response to pre-event intervention.

[0088] Based on any of the above embodiments, step S44, which involves executing the deterministic failure safety control, includes: In response to the trigger command, the physical connection between the main power supply and the rotor of the UAV is instantly cut off by a miniature explosive device within a preset response time to perform physical isolation; The electromagnetic braking device is controlled to lock the arms of the UAV in a preset horizontal deployment posture with target air resistance, thereby forming the fall state with predictable ballistic characteristics and a preset drag coefficient.

[0089] It should be noted that during the uncontrolled fall of a traditional drone, the motors often rotate uncontrollably, causing chaotic and unpredictable aerodynamic forces on the fuselage, resulting in it drifting in all directions. To eliminate this root cause, this system incorporates a deterministic failsafe module on the drone itself, with an independent power supply and equipped with an independent micro inertial measurement unit (IMU) and microcontroller unit (MCU).

[0090] When this module responds to a trigger command, it requires extremely high timeliness, needing to complete the core physical action within a preset response time (e.g., 50 milliseconds). Specifically, the system instantly severs the physical connection between the drone's main power supply and the rotor by using a miniature explosive device (e.g., a miniature explosive bolt connecting the drone's arm and rotor power unit). This physical isolation action deprives the drone of its ability to generate all uncontrollable aerodynamic thrust, turning it from an out-of-control aircraft into a purely physical falling object governed by gravity and air resistance.

[0091] Simply cutting off power is not enough to guarantee the certainty of the landing point. If the fuselage tumbles in the air, its air resistance coefficient will still change constantly. Therefore, at the moment power is cut off, the module simultaneously controls the electromagnetic braking device (such as an electromagnetic engagement locking structure) to activate, so as to firmly lock the drone's arms in a preset horizontal deployment posture with the target air resistance (i.e., maximum air resistance), such as locking four or six arms in a fully horizontal, T-shape deployment.

[0092] The preset horizontal deployment attitude is designed to maximize the UAV's frontal area, thereby creating a descent state with predictable ballistic characteristics and a preset drag coefficient. Here, predictable ballistic characteristics mean that the UAV's aerodynamic characteristics become stable and predictable under this fixed attitude, and its subsequent descent trajectory conforms to the classic parabolic ballistic physics model. More importantly, this attitude and preset drag coefficient, forcibly fixed in the air through physical means, are then reverse-calibrated by the system in the background through large-scale Monte Carlo simulations (calculating safety boundaries). The physical model used (at that time) maintains a high degree of consistency, thereby ensuring safety in the actual physical environment.

[0093] Based on any of the above embodiments Figure 2 This is a flowchart illustrating the railway unmanned inspection safety control method based on a dynamic spatiotemporal risk field provided by the present invention, as shown below. Figure 2 As shown, in a complete automated railway drone inspection cycle, the system follows... Figure 2 The four modules shown work together, and the specific workflow and simulation verification are as follows: S1, Data acquisition and transmission of the multi-source environmental sensing module After the system starts up, the multi-source environmental perception module begins operation first. Deployed at and along the unmanned aerial vehicle (UAV) automated airport, it continuously collects environmental and equipment data. For example, it acquires wind and rain data through small automatic weather stations; obtains lightning warning data through a lightning positioning and monitoring early warning system; monitors GNSS signal quality and 5G / LTE communication signal strength using specialized equipment (such as GNSS signal quality monitors and 5G / LTE communication signal strength monitors); and simultaneously connects to the railway's backend system to sense train position and speed. This multi-source heterogeneous data is packaged in real time and transmitted to the server.

[0094] S2, Analysis and Modeling of Risk Field Modeling and Computation Server This server is a privately owned high-performance computing unit deployed on edge computing nodes or cloud-based intelligent centers. After receiving multi-source sensing data, it will conduct risk assessments in parallel across four dimensions: calculating static geographical risks using BIM models, calculating train aerodynamic risks based on deep learning agent models, calculating micro-meteorological risks by integrating meteorological grids, and calculating communication link risks. Subsequently, the server uses a nonlinear coupling algorithm based on Copula functions to couple the above four sub-risks to generate a dynamic spatiotemporal risk field, and generates three layers of equipotential surface safety boundaries based on preset probability thresholds: a red no-fly zone boundary, a yellow prudent operation boundary, and a green routine inspection boundary.

[0095] S3, Planning and Execution of Autonomous Flight Control System for Unmanned Aerial Vehicles The UAV autonomous flight control system is equipped with an industrial-grade UAV that has RTK centimeter-level positioning capability, IP55 or higher protection level, wide temperature range operation capability from -30℃ to 50℃, and wind resistance capability of no less than level 7. Its built-in trajectory planning and decision-making unit interacts with the aforementioned computing server in real time with low latency via 5G or fiber optic private network.

[0096] The autonomous flight control system on the UAV acquires risk field model data and dynamic safety boundaries via a dedicated network, and then solves for the optimal trajectory based on the HJB equations in a gridded space. During flight, the system continuously generates and executes flight control commands. Simultaneously, the core algorithm at the flight control level utilizes UKF fusion nonlinear state estimation to obtain a high-precision current state of the UAV and predicts its future state accordingly. The system calculates the confidence ellipsoid based on the time trajectory. At the end of each control loop, the system checks whether the inspection task is complete. If it is, the loop ends directly; otherwise, the inspection task continues.

[0097] S4, Prediction and Execution of Deterministic Failure Safety Modules This module is the ultimate safety device installed on the drone itself. It includes an independent miniature inertial measurement unit (IMU) and microcontroller (MCU), a flight control decision unit, and an attitude-fixing actuator composed of miniature explosive bolts and electromagnetic brakes. Upon receiving a specific command or meeting preset extreme conditions, it can complete the entire process from triggering to attitude fixation within 50 milliseconds, transforming the drone into a falling object with a predictable ballistic trajectory.

[0098] As the final safety baseline, the deterministic failure safety module receives trajectory prediction data in real time. This module operates at high speed, determining that the probability of the confidence ellipsoid intersecting the no-fly zone boundary is greater than... (i.e., whether the preset intersection conditions are met). If the intersection probability does not exceed the limit (No), the system will loop back to the flight control phase to continue normal operation; once the overlap probability is found to exceed the threshold (Yes), the module will no longer wait for the flight control system to rescue it, but will trigger deterministic failure safety control. By detonating a micro explosive device and activating electromagnetic braking, the attitude of the UAV is solidified, the fall can be predicted, and it is ensured that it follows a classical physical ballistic trajectory and finally falls into a safe area to achieve a safe landing, without ever intruding into the railway clearance.

[0099] To verify the effectiveness of the above complete workflow under extreme conditions, the present invention also provides a verification embodiment based on a high-fidelity simulation platform.

[0100] First, the simulation platform uses Unreal Engine to build a high-precision 3D model of the railway scene, with the BIM model achieving centimeter-level accuracy. It also utilizes MATLAB / Simulink to build a UAV dynamics model and complex flight control algorithms. The simulation scene and objects represent a 10-kilometer-long high-speed railway including bridges, tunnels, and stations. The test object is a six-rotor industrial UAV model, whose mass, dimensions, motors, propellers, and other physical parameters are consistent with the actual UAV.

[0101] When the simulated drone performs inspection tasks along the green safety boundary, the system dynamically injects various risk events, such as simulating the strong aerodynamic impact generated by a CR400AF train passing by at a speed of 350km / h; simulating a sudden level 7 lateral gust; simulating the drone encountering a sudden loss of GPS signal at the tunnel entrance, resulting in a sharp increase in positioning error; and simulating the sudden failure of a single motor of the drone.

[0102] The system randomly selected fault points and performed 1000 random fault injection Monte Carlo simulations while the UAV was within the green safety boundary. The core evaluation metric was the number of times the UAV wreckage landing point intruded into the red no-fly zone after the deterministic failure safety module was triggered.

[0103] In the comparison group employing the traditional passive return-to-home strategy for lost contact / loss of control, the UAV is highly prone to uncontrollable tumbling when faced with motor failure or severe GPS drift, with a probability exceeding 30% of its incursion into the red no-fly zone. However, using the dynamic risk field and predictive failure control method provided by this invention, in 1000 extreme simulation experiments, the number of times the UAV wreckage landing point intruded into the red no-fly zone was only 0-3 times, verifying the 99.7% (i.e., 3σ) safety confidence level claimed by this invention. The simulation results clearly demonstrate the breakthrough effect and significant progress of this invention in ensuring absolute railway safety.

[0104] The railway unmanned inspection safety control system provided by the present invention is described below. The railway unmanned inspection safety control system described below can be referred to in correspondence with the railway unmanned inspection safety control method described above.

[0105] Based on any of the above embodiments Figure 3 This is a schematic diagram of the unmanned railway inspection safety control system provided by the present invention, as shown below. Figure 3 As shown, the system includes: The multi-source environmental sensing module 310 is used to acquire multi-source environmental sensing data along the railway line; The risk field modeling and calculation server 320 is used to construct a dynamic spatiotemporal risk field model coupled with multiple physics fields based on the multi-source environmental perception data. The dynamic spatiotemporal risk field model is used to represent the risk potential energy of a spatial point in the time dimension. The risk field modeling and calculation server 320 is also used to generate an adaptive safety boundary based on the risk equipotential surface of the dynamic spatiotemporal risk field model, and to plan a risk avoidance trajectory for the UAV based on the adaptive safety boundary, wherein the adaptive safety boundary includes a no-fly boundary. The UAV autonomous flight control system 330 is used to execute the flight mission corresponding to the risk avoidance trajectory and to acquire real-time status data of the UAV during the flight along the risk avoidance trajectory. The deterministic failure safety module 340 is used to perform trajectory prediction based on the real-time status data, generate the predicted trajectory of the UAV, and trigger deterministic failure safety control when the predicted trajectory and the no-fly zone meet a preset intersection condition, so as to convert the UAV into a crash state with a predictable ballistic trajectory.

[0106] The system provided in this invention overcomes the limitations of traditional static and isolated defense by constructing a dynamic spatiotemporal risk field model coupled with multi-physics fields using multi-source environmental perception data and generating an adaptive safety boundary that includes no-fly zones. It can quantify and adapt to the extremely complex dynamic risk environment along railway lines in real time and with high accuracy. By planning risk avoidance trajectories for UAVs based on the adaptive safety boundary, it can dynamically find the optimal flight path while avoiding triggering high-risk areas, effectively balancing safety and inspection efficiency. Furthermore, the system intervenes in advance by predicting the trajectory of real-time status data. Once a risk of intersection between the predicted trajectory and the no-fly zone is detected, deterministic failure safety control is immediately and proactively triggered, transforming the originally random and uncontrolled UAV into a crash state with a predictable ballistic trajectory. This prevents the possibility of UAV debris intruding into railway clearance and ensures railway operational safety.

[0107] Based on any of the above embodiments, the risk field modeling and calculation server includes: The sub-function construction unit is used to construct multiple sub-risk marginal distribution functions based on the multi-source environmental perception data. The multiple sub-risk marginal distribution functions include at least a static geographical risk function, a train dynamic aerodynamic risk function, a micro-meteorological risk function, and a communication link risk function. The parameter acquisition unit is used to acquire the dependency parameters obtained by calibrating a preset multivariate dataset using maximum likelihood estimation. The model building unit is used to perform nonlinear coupling calculations on the multiple sub-risk marginal distribution functions based on the dependent parameters using a multidimensional Archimedes connection function, thereby generating the dynamic spatiotemporal risk field model.

[0108] Based on any of the above embodiments, the sub-function construction unit includes: The static geographic risk function construction subunit is used to extract the distance from a spatial point to the surface of an obstacle based on the railway three-dimensional map model data in the multi-source environmental perception data, and to construct the static geographic risk function according to the preset warning distance, the distance from the spatial point to the surface of the obstacle, and the risk growth rate coefficient determined by the virtual collision energy model. The train dynamic aerodynamic risk function construction subunit is used to extract the real-time position and train speed from the multi-source environmental perception data, and construct the train dynamic aerodynamic risk function based on the real-time position and train speed. The micro-meteorological risk function construction subunit is used to extract wind speed, gust factor and rainfall intensity based on the meteorological grid data along the line in the multi-source environmental perception data, and to construct the micro-meteorological risk function according to the wind speed, the gust factor, the rainfall intensity and the preset maximum wind resistance level and maximum rain resistance level. The communication link risk function construction subunit is used to obtain real-time channel quality information of the communication base station from the multi-source environmental perception data, and construct the communication link risk function based on the real-time channel quality information and the minimum signal-to-noise ratio threshold required to maintain stable control.

[0109] Based on any of the above embodiments, the train dynamic aerodynamic risk function construction subunit is specifically used for: The real-time position and speed of the train are input into a pre-trained deep learning proxy model to infer and output the pressure gradient and curl of the fluid velocity field at the target spatial point. The deep learning proxy model is trained based on a three-dimensional flow field distribution dataset. The train dynamic aerodynamic risk function is constructed based on the pressure gradient, the curl of the fluid velocity field, and the calibration coefficient.

[0110] Based on any of the above embodiments, the risk field modeling and calculation server further includes a security boundary generation unit, which is used for: Configure a first risk probability threshold, a second risk probability threshold, and a third risk probability threshold, wherein the first risk probability threshold is greater than the second risk probability threshold, and the second risk probability threshold is greater than the third risk probability threshold; The risk equipotential surface in the dynamic spatiotemporal risk field model whose risk value is equal to the first risk probability threshold is defined as the no-fly boundary; The risk equipotential surface in the dynamic spatiotemporal risk field model in which the risk value is equal to the second risk probability threshold is defined as the prudent operation boundary. The risk equipotential surface in the dynamic spatiotemporal risk field model whose risk value is equal to the third risk probability threshold is defined as the conventional inspection boundary.

[0111] Based on any of the above embodiments, the risk field modeling and calculation server further includes a risk avoidance trajectory planning unit, which is used for: The trajectory planning problem of the UAV is modeled as an optimal control problem, and an objective function for the optimal control problem is constructed. The objective function is a performance index functional that minimizes the flight time and the risk aversion coefficient, and the risk aversion coefficient is associated with the adaptive safety boundary. A fast traversal algorithm based on the level set method is adopted to transform the optimal control problem, which aims to minimize the objective function, into solving the corresponding functional equation. Solve the equation in the gridded dynamic spatiotemporal risk field model to obtain the optimal flight path that minimizes the objective function and satisfies the initial and final boundary conditions, and use the optimal flight path as the risk avoidance track.

[0112] Based on any of the above embodiments, the deterministic failure safety module is specifically used for: Nonlinear state estimation is performed on the real-time state data, and the posterior probability distribution of the state is approximated by the sampling points of the unscented transformation to obtain the state vector sequence and the corresponding covariance matrix sequence of the UAV in a future preset time period. Based on the state vector sequence, a predicted trajectory of the UAV is generated, and based on the covariance matrix sequence, a confidence ellipsoid of the predicted trajectory is generated. Calculate the probability of overlap between the confidence ellipsoid and the spatial region enclosed by the no-fly zone boundary; If the overlap probability is greater than a preset probability threshold, the preset intersection condition is determined to be met, and a trigger command is generated to execute the deterministic failure safety control.

[0113] Based on any of the above embodiments, the deterministic failure safety module is further configured to: In response to the trigger command, the physical connection between the main power supply and the rotor of the UAV is instantly cut off by a miniature explosive device within a preset response time to perform physical isolation; The electromagnetic braking device is controlled to lock the UAV's arms in a preset horizontal deployment posture with target air resistance, thereby forming the fall state with predictable ballistic characteristics and a preset drag coefficient.

[0114] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 4As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 can call logic instructions in the memory 430 to execute a railway unmanned inspection safety control method. This method includes: acquiring multi-source environmental perception data along the railway line, and constructing a dynamic spatiotemporal risk field model coupled with multiple physics fields based on the multi-source environmental perception data. The dynamic spatiotemporal risk field model represents the risk potential energy of a spatial point in the time dimension. Based on the risk equipotential surface of the dynamic spatiotemporal risk field model, an adaptive safety boundary is generated, including a no-fly zone boundary. Based on the adaptive safety boundary, a risk avoidance trajectory is planned for the UAV, and real-time status data of the UAV during its flight along the risk avoidance trajectory is acquired. Based on the real-time status data, trajectory prediction is performed to generate a predicted trajectory for the UAV. When the predicted trajectory and the no-fly zone boundary satisfy a preset intersection condition, deterministic failure safety control is triggered to convert the UAV into a crash state with a predictable ballistic trajectory.

[0115] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to related technologies, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0116] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the railway unmanned inspection safety control method provided by the above methods. The method includes: acquiring multi-source environmental perception data along the railway line, and constructing a dynamic spatiotemporal risk field model coupled with multiple physics fields based on the multi-source environmental perception data. The dynamic spatiotemporal risk field model is used to represent the risk potential energy of a spatial point in the time dimension. Based on the risk equipotential surface of the dynamic spatiotemporal risk field model, an adaptive safety boundary is generated, which includes a no-fly zone boundary. Based on the adaptive safety boundary, a risk avoidance trajectory is planned for the UAV, and real-time status data of the UAV during its flight along the risk avoidance trajectory is acquired. Based on the real-time status data, trajectory prediction is performed to generate a predicted trajectory of the UAV. When the predicted trajectory and the no-fly zone boundary meet a preset intersection condition, deterministic failure safety control is triggered to convert the UAV into a crash state with a predictable ballistic trajectory.

[0117] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the railway unmanned inspection safety control method provided by the above methods. The method includes: acquiring multi-source environmental perception data along the railway line, and constructing a dynamic spatiotemporal risk field model coupled with multiple physics fields based on the multi-source environmental perception data. The dynamic spatiotemporal risk field model is used to represent the risk potential energy of a spatial point in the time dimension. Based on the risk equipotential surface of the dynamic spatiotemporal risk field model, an adaptive safety boundary is generated, including a no-fly zone boundary. Based on the adaptive safety boundary, a risk avoidance trajectory is planned for the UAV, and real-time status data of the UAV during its flight along the risk avoidance trajectory is acquired. Based on the real-time status data, trajectory prediction is performed to generate a predicted trajectory of the UAV. When the predicted trajectory and the no-fly zone boundary satisfy a preset intersection condition, deterministic failure safety control is triggered to convert the UAV into a crash state with a predictable ballistic trajectory.

[0118] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0119] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for unmanned railway inspection safety control, characterized in that, include: Acquire multi-source environmental sensing data along the railway line, and construct a dynamic spatiotemporal risk field model coupled with multi-physics fields based on the multi-source environmental sensing data. The dynamic spatiotemporal risk field model is used to represent the risk potential energy of a spatial point in the time dimension. Based on the risk equipotential surface of the dynamic spatiotemporal risk field model, an adaptive safety boundary is generated, which includes a no-fly zone boundary. Based on the adaptive safety boundary, a risk avoidance trajectory is planned for the UAV, and real-time status data of the UAV during flight along the risk avoidance trajectory is obtained; Based on the real-time status data, trajectory prediction is performed to generate the predicted trajectory of the UAV. When the predicted trajectory and the no-fly zone meet the preset intersection conditions, deterministic failure safety control is triggered to convert the UAV into a crash state with a predictable ballistic trajectory.

2. The railway unmanned inspection safety control method according to claim 1, characterized in that, The construction of a dynamic spatiotemporal risk field model based on the multi-source environmental perception data includes: Based on the multi-source environmental perception data, multiple sub-risk marginal distribution functions are constructed respectively. The multiple sub-risk marginal distribution functions include at least a static geographical risk function, a train dynamic aerodynamic risk function, a micro-meteorological risk function, and a communication link risk function. Obtain the dependency parameters obtained by calibrating a predefined multivariate dataset using maximum likelihood estimation; Based on the aforementioned dependency parameters, a multidimensional Archimedes connection function is used to perform nonlinear coupling calculations on the multiple sub-risk marginal distribution functions to generate the dynamic spatiotemporal risk field model.

3. The railway unmanned inspection safety control method according to claim 2, characterized in that, Based on the multi-source environmental perception data, multiple sub-risk marginal distribution functions are constructed, including: Based on the railway 3D map model data in the multi-source environmental perception data, the distance from spatial points to the obstacle surface is extracted, and the static geographic risk function is constructed according to the preset warning distance, the distance from the spatial points to the obstacle surface, and the risk growth rate coefficient determined by the virtual collision energy model. The train's real-time position and speed are extracted from the multi-source environmental perception data, and the train's dynamic aerodynamic risk function is constructed based on the train's real-time position and speed. Based on the meteorological grid data along the route from the multi-source environmental perception data, wind speed, gust factor and rainfall intensity are extracted, and the micro-meteorological risk function is constructed according to the wind speed, the gust factor, the rainfall intensity and the preset maximum wind resistance level and maximum rain resistance level. The real-time channel quality information of the communication base station is obtained from the multi-source environmental perception data, and the communication link risk function is constructed based on the real-time channel quality information and the minimum signal-to-noise ratio threshold required to maintain stable control.

4. The railway unmanned inspection safety control method according to claim 3, characterized in that, The process of constructing the train's dynamic aerodynamic risk function based on the train's real-time position and speed includes: The real-time position and speed of the train are input into a pre-trained deep learning proxy model to infer and output the pressure gradient and curl of the fluid velocity field at the target spatial point. The deep learning proxy model is trained based on a three-dimensional flow field distribution dataset. The train dynamic aerodynamic risk function is constructed based on the pressure gradient, the curl of the fluid velocity field, and the calibration coefficient.

5. The railway unmanned inspection safety control method according to claim 1, characterized in that, The adaptive safety boundary includes multi-level risk regions. The generation of the adaptive safety boundary based on the risk equipotential surface of the dynamic spatiotemporal risk field model includes: Configure a first risk probability threshold, a second risk probability threshold, and a third risk probability threshold, wherein the first risk probability threshold is greater than the second risk probability threshold, and the second risk probability threshold is greater than the third risk probability threshold; The risk equipotential surface in the dynamic spatiotemporal risk field model whose risk value is equal to the first risk probability threshold is defined as the no-fly boundary; The risk equipotential surface in the dynamic spatiotemporal risk field model in which the risk value is equal to the second risk probability threshold is defined as the prudent operation boundary. The risk equipotential surface in the dynamic spatiotemporal risk field model whose risk value is equal to the third risk probability threshold is defined as the conventional inspection boundary.

6. The railway unmanned inspection safety control method according to claim 1, characterized in that, The step of planning risk avoidance flight paths for UAVs based on the adaptive safety boundary includes: The trajectory planning problem of the UAV is modeled as an optimal control problem, and an objective function for the optimal control problem is constructed. The objective function is a performance index functional that minimizes the flight time and the risk aversion coefficient, and the risk aversion coefficient is associated with the adaptive safety boundary. A fast traversal algorithm based on the level set method is adopted to transform the optimal control problem, which aims to minimize the objective function, into solving the corresponding functional equation. Solve the equation in the gridded dynamic spatiotemporal risk field model to obtain the optimal flight path that minimizes the objective function and satisfies the initial and final boundary conditions, and use the optimal flight path as the risk avoidance track.

7. The railway unmanned inspection safety control method according to any one of claims 1 to 6, characterized in that, The process of predicting the trajectory based on the real-time status data to generate the predicted trajectory of the UAV, and triggering deterministic failure safety control when the predicted trajectory and the no-fly zone meet a preset intersection condition, includes: Nonlinear state estimation is performed on the real-time state data, and the posterior probability distribution of the state is approximated by the sampling points of the unscented transformation to obtain the state vector sequence and the corresponding covariance matrix sequence of the UAV in a future preset time period. Based on the state vector sequence, a predicted trajectory of the UAV is generated, and based on the covariance matrix sequence, a confidence ellipsoid of the predicted trajectory is generated. Calculate the probability of overlap between the confidence ellipsoid and the spatial region enclosed by the no-fly zone boundary; If the overlap probability is greater than a preset probability threshold, the preset intersection condition is determined to be met, and a trigger command is generated to execute the deterministic failure safety control.

8. The railway unmanned inspection safety control method according to claim 7, characterized in that, The execution of the deterministic failure safety control includes: In response to the trigger command, the physical connection between the main power supply and the rotor of the UAV is instantly cut off by a miniature explosive device within a preset response time to perform physical isolation; The electromagnetic braking device is controlled to lock the UAV's arms in a preset horizontal deployment posture with target air resistance, thereby forming the fall state with predictable ballistic characteristics and a preset drag coefficient.

9. A railway unmanned inspection safety control system, characterized in that, include: The multi-source environmental sensing module is used to acquire multi-source environmental sensing data along the railway line; A risk field modeling and calculation server is used to construct a dynamic spatiotemporal risk field model coupled with multiple physics fields based on the multi-source environmental perception data. The dynamic spatiotemporal risk field model is used to represent the risk potential energy of a spatial point in the time dimension. The risk field modeling and calculation server is also used to generate an adaptive safety boundary based on the risk equipotential surface of the dynamic spatiotemporal risk field model, and to plan a risk avoidance trajectory for the UAV based on the adaptive safety boundary, wherein the adaptive safety boundary includes a no-fly boundary. An autonomous flight control system for unmanned aerial vehicles (UAVs) is used to execute flight tasks corresponding to the risk avoidance flight path and to acquire real-time status data of the UAV during flight along the risk avoidance flight path. The deterministic failure safety module is used to perform trajectory prediction based on the real-time status data, generate the predicted trajectory of the UAV, and trigger deterministic failure safety control when the predicted trajectory and the no-fly zone meet a preset intersection condition, so as to convert the UAV into a crash state with a predictable ballistic trajectory.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the railway unmanned inspection safety control method as described in any one of claims 1 to 8.

Images