An LVC virtual-real fusion simulation training method for unmanned aerial vehicle equipment support

Abstract

The application discloses a kind of LVC virtual-real fusion simulation training methods for unmanned aerial vehicle equipment support, it is related to unmanned aerial vehicle equipment support training and computer simulation technical field, solve the problem that the degree of virtual-real fusion is insufficient in existing simulation training, fault injection lacks logical correlation, training evaluation lacks explainability.The application builds simulation resource library containing live, virtual and constructed simulation resources and establishes interoperation channel, collects continuous operation characteristic data stream to solve virtual response curve and generates tactile driving instruction to drive tactile feedback device, calculates each node fault trigger probability based on unmanned aerial vehicle system fault conduction model and generates cascade failure along directed edge, and compares continuous operation characteristic data stream and tactile feedback with standard support process causal diagram to identify deviation operation step and determine cascade influence, realize high-fidelity virtual-real mutual driving training for unmanned aerial vehicle equipment support, dynamic cascade fault injection and explainability evaluation.

Description

Technical Field

[0001] This invention belongs to the field of UAV equipment support training and computer simulation technology, specifically a virtual-real fusion simulation training method for UAV equipment support. Background Technology

[0002] Unmanned aerial vehicle (UAV) equipment support is a crucial link in the UAV combat system, encompassing multiple core tasks such as detection, maintenance, troubleshooting, ammunition loading, and support command. The operational proficiency and emergency troubleshooting capabilities of support personnel directly impact the UAV's availability and mission success rate. However, UAV equipment support training has long faced prominent challenges, including a scarcity of real-world equipment resources, high training costs, and the difficulty in reproducing high-risk failure scenarios. Real-world training relies on actual UAVs and support equipment, and the characteristics of multiple models and small batches of equipment make it difficult to comprehensively cover all training resources. Furthermore, some critical failures are destructive and irreversible, making repeated drills on real equipment impossible. Therefore, simulation training has become an important approach to developing UAV equipment support capabilities.

[0003] LVC (Live-Vertical-Vertical) simulation is a technological system that organically integrates three types of simulation resources: live-action simulation, virtual simulation, and constructive simulation. Live-action simulation uses real personnel and equipment to provide the most realistic operational feedback. Virtual simulation uses real personnel operating virtual systems to simulate equipment operation processes through digital prototypes and virtual reality technology. Constructive simulation uses computer-generated virtual forces or environments to provide large-scale, complex mission scenarios and simulation capabilities. While LVC has been widely applied in flight training and combat simulations, in the field of UAV equipment support training, achieving high-fidelity real-time interaction between support equipment operation and virtual equipment response, constructing a logically related dynamic fault injection mechanism, and providing interpretable quantitative evaluation of the training process remain critical technical challenges that need to be overcome.

[0004] The following problems exist in the existing technology:

[0005] In existing methods, the interaction between the physical operation and the virtual model mostly adopts a discrete state synchronization method, which simplifies the operator's continuous operation into a Boolean state of operation completion or incompleteness, resulting in a large deviation between the virtual response and the real operation at the process level. At the same time, the tactile feedback mapping of the virtual response to the physical end lacks a detailed description of the changes in mechanical characteristics throughout the operation process, and the operator cannot obtain a real feel consistent with the physical state of the virtual object.

[0006] In existing methods, most of the faults in the training scenarios are pre-set. The timing and type of the faults are unrelated to the actual operation behavior of the operators, and the faults are isolated from each other. They cannot simulate the cascading failure effect caused by improper operation and system vulnerability in real equipment systems. This makes the training scenarios lack suddenness and uncertainty, and it is difficult to cultivate the operators' causal reasoning and emergency troubleshooting capabilities.

[0007] Existing methods often use a weighted summation of terminal indicators such as operation time and number of errors for training and evaluation, which cannot trace the causal relationship between a specific operational deviation and subsequent failure events; the evaluation results are only a single score or grade, lacking specific guidance for operators on improvement directions. Summary of the Invention

[0008] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes an LVC virtual-real fusion simulation training method for unmanned aerial vehicle (UAV) equipment support, which is used to solve the above-mentioned technical problem.

[0009] The first aspect of this invention provides an LVC virtual-real fusion simulation training method for unmanned aerial vehicle (UAV) equipment support, comprising the following steps:

[0010] S1: Construct a simulation resource library for UAV equipment support. The simulation resource library includes real-world simulation resources based on physical UAV support equipment, virtual simulation resources based on UAV digital prototypes, and constructed simulation resources based on UAV system fault propagation models and fault event sets. Establish interoperability channels between real-world simulation resources, virtual simulation resources, and constructed simulation resources.

[0011] S2: During the simulation training process, a continuous operation feature data stream generated by the interaction between the operator and the physical UAV support equipment is collected; a virtual response curve is obtained by the virtual simulation resource, and a tactile driving command is generated based on the virtual response curve; the tactile driving command is sent to the tactile feedback device integrated in the physical UAV support equipment through the interoperability channel, driving the tactile feedback device to generate tactile feedback.

[0012] S3: Obtain the real-time state vector of each node in the UAV system fault propagation model; calculate the potential field gradient intensity of each node in the UAV system fault propagation model based on the continuous operation feature data stream; calculate the fault trigger probability of each node based on the potential field gradient intensity and the real-time state vector of each node; when the fault trigger probability exceeds the preset probability threshold, select and trigger the fault event corresponding to the node from the fault event set, and propagate the fault event along the directed edge of the UAV system fault propagation model to generate a cascaded fault.

[0013] S4: After the simulation training is completed, the time-series causal inference method is used to compare the continuous operation feature data stream and the tactile feedback with the preset standard assurance process causal graph to identify the operation steps that deviate from the causal nodes in the standard assurance process causal graph; based on the cascade fault, the cascade impact corresponding to the identified deviation operation steps is determined, and an evaluation report containing the deviation operation steps and their cascade impact is generated.

[0014] Preferably, step S1 includes the following steps:

[0015] Construct a real-world simulation resource that includes a physical drone support equipment integrated with operational feature sensors and haptic feedback devices. The operational feature sensors are used to collect continuous operational feature data streams generated when operators operate the physical drone support equipment.

[0016] A virtual simulation resource is constructed, in which a digital prototype of a UAV is included, containing a simplified finite element model for real-time calculation of mechanical response.

[0017] The simulation resources are constructed, and the fault propagation model of the UAV system in the simulation resources adopts a directed graph structure, which contains multiple nodes and multiple directed edges. Each node corresponds to a functional component of the UAV and is accompanied by a nominal state vector. Each directed edge represents the causal propagation path of the fault from one node to another and is accompanied by a propagation condition. Each fault event in the fault event set corresponds to a node.

[0018] At the same time, an interoperability channel is established to connect live simulation resources with virtual simulation resources, live simulation resources with construction simulation resources, and virtual simulation resources with construction simulation resources.

[0019] Preferably, in step S2, obtaining the virtual response curve from the virtual simulation resources includes the following steps:

[0020] The continuous operation feature data stream is used as a boundary condition input to the finite element simplified model of key components in the UAV digital prototype. The continuous physical response of the virtual component under the force generated by the operator's manipulation of the UAV support equipment is calculated in real time to obtain the initial virtual response curve. Simultaneously, a theoretical physical model is established based on the material physical properties of the key components. The expected response of the operation at the current time t is calculated using the continuous operation feature data stream and the theoretical physical model. ;

[0021] By calculating the distortion of the virtual-real mapping The cumulative deviation of the rate of change between the initial virtual response curve and the expected response over the entire operation process is calculated using the following formula:

[0022]

[0023] in, This represents the virtual response state corresponding to the initial virtual response curve. The time window for a complete operation task;

[0024] by With the goal of minimization, the parameters of the simplified finite element model are dynamically adjusted until the cumulative deviation between the rate of change of the initial virtual response curve and the rate of change of the desired response throughout the entire operation process meets the preset accuracy requirements, thus obtaining the virtual response curve.

[0025] Preferably, in step S2, the step of generating tactile driving commands based on the virtual response curve and sending the tactile driving commands to the tactile feedback device integrated into the physical UAV support equipment via the interoperability channel to drive the tactile feedback device to generate tactile feedback includes the following steps:

[0026] Extract the mechanical feature points corresponding to the elastic deformation stage, plastic deformation stage and failure critical point in the virtual response curve, and convert the mechanical feature points into at least one of force command sequence and torque command sequence to generate haptic drive commands;

[0027] After receiving a tactile driving command, the tactile feedback device drives the actual UAV support equipment to generate at least one of force feedback and torque feedback corresponding to the mechanical state of the virtual response curve in the elastic deformation stage, plastic deformation stage, and failure critical point.

[0028] Preferably, in step S3, the calculation of the fault triggering probability of each node based on the potential field gradient field strength and the real-time state vector of each node includes the following steps:

[0029] The continuous operation feature data stream is mapped to the state space of the constructed simulation resource, and the potential gradient field intensity generated by operation behavior B on each node in the state space is calculated. ;

[0030] Get each node The real-time state vector at the current time t Deviating from its preset nominal state vector The degree;

[0031] Calculate the failure trigger probability of each node. The calculation formula is:

[0032]

[0033] Where τ is the duration of the operation;

[0034] Let be the time-varying risk intensity function, i.e. ,in This is the normalization coefficient.

[0035] Preferably, in step S3, when the probability of a fault triggering exceeds a preset probability threshold, selecting and triggering the fault event corresponding to the node from the fault event set, and propagating the fault event along the directed edge of the UAV system fault propagation model to generate a cascading fault, includes the following steps:

[0036] When the probability of a fault triggering a certain node exceeds a preset probability threshold, a fault event corresponding to that node is selected from the fault event set and triggered.

[0037] The triggered fault event will propagate to subsequent nodes along all directed edges originating from that node in the fault propagation model of the UAV system.

[0038] For each directed edge with attached propagation conditions, determine whether the real-time state vector of the node satisfies the propagation conditions. When the propagation conditions are satisfied, trigger the fault event corresponding to the successor node.

[0039] By propagating step by step along the directed edges, a cascading fault is generated, consisting of the initial fault node and its subsequent fault nodes.

[0040] Preferably, in step S4, the step of employing a temporal causal inference method to compare the continuous operation feature data stream and the tactile feedback with a preset standard assurance process causal graph to identify operation steps that deviate from the causal nodes in the standard assurance process causal graph includes the following steps:

[0041] Obtain a pre-built standard assurance process causal graph. The standard assurance process causal graph adopts a directed graph structure, containing multiple causal nodes and multiple directed edges. Each causal node corresponds to a standard assurance operation step, and each directed edge represents the causal dependency relationship between two causal nodes.

[0042] The operation actions and their occurrence times reflected in the continuous operation feature data stream are matched with the preset execution sequence of the standard support operation steps corresponding to each causal node in the standard support process causal diagram; at the same time, the operator's response actions to tactile feedback and their occurrence times reflected in the tactile feedback data are matched with the standard response sequence corresponding to each causal node in the standard support process causal diagram.

[0043] During the matching process, the operation steps in the continuous operation feature data stream and the tactile feedback data that deviate from the causal nodes in the standard assurance process causal graph are identified. The deviations include at least one of the following: missing operation steps, reversed execution order of operation steps, operation step execution time exceeding a preset time threshold, and tactile feedback response actions not conforming to preset response specifications.

[0044] Preferably, in step S4, determining the cascading impact corresponding to the identified deviation operation step based on the cascading fault includes the following steps:

[0045] The occurrence time of the identified deviation operation step is compared with the trigger time of each fault event in the cascaded fault to determine the fault event that is temporally associated with the deviation operation step.

[0046] Based on the fault propagation path information contained in the cascaded faults, trace the path through which the deviation operation step triggers subsequent cascaded faults via fault transmission relationships, and determine the cascaded impact corresponding to the deviation operation step.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] This invention collects continuous operation feature data streams and optimizes the simplified finite element model online with the goal of minimizing distortion. At the same time, it generates tactile driving commands based on the stage mechanical characteristics of the virtual response curve to drive the tactile feedback device, realizing continuous mapping of virtual and real states and high-fidelity tactile feedback throughout the operation process, so that the operator can obtain a tactile sensation consistent with the physical state of the virtual object.

[0049] This invention simulates the implicit degradation of a system by using the directed graph structure and dynamic drift of node states in a fault propagation model of an unmanned aerial vehicle system. The fault trigger probability is calculated by the potential gradient field strength of each node and the deviation of the node's real-time state from the operational behavior. The fault is then propagated along the directed edges to generate cascaded faults, thus realizing the dynamic cascaded fault injection that couples operational behavior with system vulnerability. This makes the training scenario have suddenness and causal logic.

[0050] This invention identifies various types of deviations from the operational process by comparing operational data and tactile feedback with a standard assurance process causal graph. It also compares the deviations with cascading failures in a time sequence to trace the failure propagation path and generates an evaluation report that includes the deviations, deviation types, and cascading effects. This enables a causal traceability-based interpretable evaluation of the training process. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0052] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0053] Please see Figure 1 This invention provides a virtual-real fusion simulation training method for unmanned aerial vehicle (UAV) equipment support, comprising the following steps:

[0054] S1: Construct a simulation resource library for UAV equipment support. The simulation resource library includes real-world simulation resources based on physical UAV support equipment, virtual simulation resources based on UAV digital prototypes, and constructed simulation resources based on UAV system fault propagation models and fault event sets. Establish interoperability channels between real-world simulation resources, virtual simulation resources, and constructed simulation resources.

[0055] S2: During the simulation training process, a continuous operation feature data stream generated by the interaction between the operator and the physical UAV support equipment is collected; a virtual response curve is obtained by the virtual simulation resource, and a tactile driving command is generated based on the virtual response curve; the tactile driving command is sent to the tactile feedback device integrated in the physical UAV support equipment through the interoperability channel, driving the tactile feedback device to generate tactile feedback.

[0056] S3: Obtain the real-time state vector of each node in the UAV system fault propagation model; calculate the potential field gradient intensity of each node in the UAV system fault propagation model based on the continuous operation feature data stream; calculate the fault trigger probability of each node based on the potential field gradient intensity and the real-time state vector of each node; when the fault trigger probability exceeds the preset probability threshold, select and trigger the fault event corresponding to the node from the fault event set, and propagate the fault event along the directed edge of the UAV system fault propagation model to generate a cascaded fault.

[0057] S4: After the simulation training is completed, the time-series causal inference method is used to compare the continuous operation feature data stream and the tactile feedback with the preset standard assurance process causal graph to identify the operation steps that deviate from the causal nodes in the standard assurance process causal graph; based on the cascade fault, the cascade impact corresponding to the identified deviation operation steps is determined, and an evaluation report containing the deviation operation steps and their cascade impact is generated.

[0058] Specifically, a simulation resource library for UAV equipment support is constructed. This library includes real-world simulation resources based on physical UAV support equipment, virtual simulation resources based on digital UAV prototypes, and constructed simulation resources based on UAV system fault propagation models and fault event sets. An interoperability channel is established between these three types of simulation resources. During simulation training, continuous operational characteristic data streams generated by the interaction between operators and the physical UAV support equipment are collected. Virtual response curves are calculated using the virtual simulation resources, and tactile driving commands are generated based on these curves. These commands are then sent to the tactile feedback device integrated into the physical UAV support equipment to generate tactile feedback. Simultaneously with the virtual-real interaction, the real-time state vectors of each node in the UAV system fault propagation model are acquired. The potential field gradient intensity of each node is calculated from the continuous operational characteristic data stream. The fault trigger probability of each node is calculated based on the potential field gradient intensity and the real-time state vector. When the fault trigger probability exceeds a preset probability threshold, a fault event is selected from the fault event set and triggered. The fault event is then propagated along directed edges to generate cascading faults. After the simulation training is completed, the continuous operation feature data stream and tactile feedback are compared with the preset standard assurance process cause-effect diagram to identify deviation operation steps. Based on the cascade faults, the cascade effects corresponding to the deviation operation steps are determined, and an evaluation report containing the deviation operation steps and their cascade effects is generated.

[0059] In one embodiment of the present invention, step S1 includes the following steps:

[0060] Construct a real-world simulation resource that includes a physical drone support equipment integrated with operational feature sensors and haptic feedback devices. The operational feature sensors are used to collect continuous operational feature data streams generated when operators operate the physical drone support equipment.

[0061] A virtual simulation resource is constructed, in which a digital prototype of a UAV is included, containing a simplified finite element model for real-time calculation of mechanical response.

[0062] The simulation resources are constructed, and the fault propagation model of the UAV system in the simulation resources adopts a directed graph structure, which contains multiple nodes and multiple directed edges. Each node corresponds to a functional component of the UAV and is accompanied by a nominal state vector. Each directed edge represents the causal propagation path of the fault from one node to another and is accompanied by a propagation condition. Each fault event in the fault event set corresponds to a node.

[0063] At the same time, an interoperability channel is established to connect live simulation resources with virtual simulation resources, live simulation resources with construction simulation resources, and virtual simulation resources with construction simulation resources.

[0064] Specifically, the simulation resources use actual UAV support equipment as the object. The actual support equipment used in UAV equipment support is selected as the physical carrier of the simulation resources; this equipment includes ground power vehicles, integrated testing instruments, torque wrenches, and aviation connector assembly / disassembly tools, etc. Operational characteristic sensors are integrated into the selected UAV support equipment. These sensors continuously collect multi-dimensional physical quantities generated when operators manipulate the equipment during simulation training. The collected quantities include torque, rotational angular velocity, axial pressure, and contact time. For example, a strain gauge torque sensor and a gyroscope are embedded inside the handle of the torque wrench to measure the torque applied by the operator and the angular velocity of the wrench's rotation, respectively. A pressure sensor is installed on the wrench head to measure the axial pressure. The sensor sampling frequency is 500 Hz. Tactile feedback devices are also integrated into the selected UAV support equipment. The tactile feedback device adopts an active torque feedback mechanism based on a servo motor. The servo motor operates in a torque closed-loop control mode and outputs a corresponding reverse torque according to the received torque command value. Taking a torque wrench as an example, a servo motor is installed at the rotation axis of the wrench. The servo motor receives the torque command sequence generated by the virtual simulation resource calculation and outputs a torque in real time that is opposite to the direction of the force applied by the operator, so that the operator can feel the damping sensation corresponding to the mechanical state of the virtual response curve when operating the wrench.

[0065] The virtual simulation resources use a digital prototype of an unmanned aerial vehicle (UAV) as the object. High-precision 3D geometric models of the entire UAV and its subsystems are constructed. Subsystems include at least a flight control system, a power system, a data link system, and a power supply system. The 3D geometric models are created using CAD software at a 1:1 scale based on the actual dimensions and structure of the UAV. Material physical properties are assigned to key components in the 3D geometric models. Key components include bolted connectors, aviation plugs, engine mounts, and wing spars; the assigned physical properties include material density, elastic modulus, Poisson's ratio, coefficient of friction, and yield strength. Taking bolted connectors as an example, if 7075 aviation aluminum alloy is used, its elastic modulus is 71.7 GPa, its Poisson's ratio is 0.33, its yield strength is 503 MPa, and its density is 2.81 g / cm³. A simplified finite element model is constructed for the key components. The simplified finite element model uses a reduced-order modeling method, compressing the computational degrees of freedom of the full-size finite element model from tens of thousands to hundreds. The input to this simplified finite element model is a continuous operation feature data stream, and the output is the mechanical response of the virtual component under the operation, including stress distribution, strain distribution, and displacement.

[0066] The simulation resources are constructed using a fault propagation model and a set of fault events for an unmanned aerial vehicle (UAV) system. The UAV system fault propagation model adopts a directed graph structure, denoted as G=(N,E), where N is the set of nodes and E is the set of directed edges. Each node in the node set N corresponds to a functional component of the UAV; taking a certain type of medium-altitude long-endurance UAV as an example, the node set must at least include a power module node. Flight control core board node Data link transceiver node GPS module node and servo drive module node Each node is accompanied by a pre-defined nominal state vector. The nominal state vector is composed of the key parameters of the functional component under normal operating conditions; taking the power module node as an example, its It includes a nominal output voltage of 28 volts, a nominal operating temperature of 40 degrees Celsius, and a nominal output current of 10 amps. Each directed edge in the set E represents a causal propagation path of a fault from one node to another; each directed edge is accompanied by a propagation condition, which is the threshold range of each parameter in the real-time state vector of the node originating from that directed edge; [The text abruptly ends here, likely due to an incomplete sentence or missing information.] Pointing to the flight control core board node Taking a directed edge as an example, the propagation condition is that the real-time output voltage of the power module node is below 22 volts or above 32 volts. When this condition is met, the fault propagates from the power module to the flight control core board, triggering a self-test anomaly fault on the flight control core board. The fault event set stores the fault events corresponding to each node in the UAV system fault propagation model; each fault event includes the fault characterization, fault injection parameters, and fault propagation triggering conditions when the node fails. The UAV system fault propagation model and the fault event set together constitute the simulation resources.

[0067] An interoperability channel is constructed between real-world simulation resources, virtual simulation resources, and structural simulation resources based on a data distribution service system. The interoperability channel connects three pairs of resources, each pair being a bidirectional data transmission channel. The first pair connects the real-world simulation resource and the virtual simulation resource; this connection transmits continuous operational feature data streams to the virtual simulation resource and transmits haptic drive commands generated by the virtual simulation resource to the haptic feedback device. The second pair connects the real-world simulation resource and the structural simulation resource; this connection transmits continuous operational feature data streams to the structural simulation resource and transmits fault events and cascading faults generated by the structural simulation resource to the real-world simulation resource, enabling the instrument displays or alarm devices of the UAV support equipment in the real-world simulation resource to synchronously display the corresponding fault characteristics. The third pair connects the virtual simulation resource and the structural simulation resource; this connection transmits fault events and cascading faults to the virtual simulation resource, enabling the UAV digital prototype in the virtual simulation resource to synchronously display the abnormal state corresponding to the fault, including changes in virtual instrument readings and color-changing highlighting of virtual components.

[0068] In one embodiment of the present invention, step S2, which involves obtaining the virtual response curve from the virtual simulation resources, includes the following steps:

[0069] The continuous operation feature data stream is used as a boundary condition input to the finite element simplified model of key components in the UAV digital prototype. The continuous physical response of the virtual component under the force generated by the operator's manipulation of the UAV support equipment is calculated in real time to obtain the initial virtual response curve. Simultaneously, a theoretical physical model is established based on the material physical properties of the key components. The expected response of the operation at the current time t is calculated using the continuous operation feature data stream and the theoretical physical model. ;

[0070] By calculating the distortion of the virtual-real mapping The cumulative deviation of the rate of change between the initial virtual response curve and the expected response over the entire operation process is calculated using the following formula:

[0071]

[0072] in, This represents the virtual response state corresponding to the initial virtual response curve. The time window for a complete operation task;

[0073] by With the goal of minimization, the parameters of the simplified finite element model are dynamically adjusted until the cumulative deviation between the rate of change of the initial virtual response curve and the rate of change of the desired response throughout the entire operation process meets the preset accuracy requirements, thus obtaining the virtual response curve.

[0074] Specifically, this embodiment uses tightening an aviation plug with a torque wrench as an example. The aviation plug uses an M10 threaded connection and is made of 7075-T6 aluminum alloy. This operation was chosen as an example because torque control of threaded fasteners is one of the most frequent operations in UAV equipment support, and its mechanical process fully includes the three stages of elasticity, plasticity, and failure. This method is applicable to operational training on various UAV support equipment.

[0075] The torque wrench integrates operating characteristic sensors including: a strain gauge torque sensor with a range of 0 to 50 N·m and an accuracy class of 0.5, used to measure the torque applied by the operator. MEMS gyroscope, with a range of ±2000 degrees per second, is used to measure the angular velocity of a wrench. Piezoresistive pressure sensor, with a range of 0 to 500 Newtons, is used to measure axial pressure along the axis of a wrench. The three sensors operate using a synchronous triggering method, with a unified sampling frequency of 500 Hz. When using other types of UAV support equipment, the configuration of the operational characteristic sensors is adjusted accordingly. For example, when using a comprehensive testing instrument, a probe contact pressure sensor, voltage sensor, and current sensor are integrated; when using a ground power vehicle, a power supply voltage sensor, power supply current sensor, and cable insertion / removal force sensor are integrated; when using an aviation plug assembly / disassembly tool, an insertion / removal force sensor and a displacement sensor are integrated. At each sampling moment, each sensor simultaneously generates a data value, collectively forming a set of multi-dimensional data. Taking a torque wrench as an example, at the sampling moment... Obtain three data elements As the operation progresses from start to finish, all data sets at each sampling moment are arranged sequentially in ascending order of timestamp, with a fixed interval of 2 milliseconds between adjacent sampling moments, thus forming a continuous operation characteristic data stream.

[0076] A simplified finite element model was established for the threaded connector of the aviation plug involved in this simulation training. First, a full-size finite element model was constructed, using hexahedral linear elements for mesh generation. Face-to-face contact elements were used for the threaded contact surface. The element size was determined through mesh convergence analysis. The final full-size model contains approximately 52,000 nodal degrees of freedom. Material parameters were taken as typical values ​​for 7075-T6 aluminum alloy according to GB / T 3880-2012 standard: elastic modulus... =71.7 GPa, Poisson's ratio v=0.33, yield strength =503MPa; Initial friction coefficient of threaded contact surface =0.15 (This is the recommended sliding friction coefficient of aluminum alloy-aluminum alloy under unlubricated conditions in the mechanical design manual). Next, the full-size model is reduced in order using the Craig-Bampton substructure modal synthesis method. Nodes at the bolt head and threaded contact surface are selected as the principal degrees of freedom, and retained and constrained modes are extracted, condensing the full-size model into a reduced-order model with approximately 300 principal degrees of freedom. All modes with natural frequencies below 5kHz are retained during the reduction process. The reduced stiffness and mass matrices are stored in the virtual simulation resources, and the solution step size is 2ms. Two adjustable parameters, the threaded contact surface friction coefficient μ and the axial stiffness reduction correction factor β, are reserved in the simplified finite element model. The initial value of μ... =0.15; initial value of β =1.0, where β is the ratio of the axial stiffness of the bolt after the order reduction to the axial stiffness of the full-size model. The initial value of 1.0 indicates that the axial stiffness of the full-size model is directly adopted in the initial state of the order reduction model.

[0077] A theoretical physical model of the same aviation plug threaded connector is established simultaneously. The theoretical physical model includes calculation rules for the elastic stage, as well as the plastic and failure stages, and its output is the desired response. The elastic stage employs the classical thread fastener mechanics formula. Where d=10mm is the nominal diameter of the thread. The desired response is represented by a nut coefficient K = 0.2, taken as the median of the recommended range of 0.18 to 0.23 for unlubricated aluminum alloy-aluminum alloy threaded pairs in ASME PCC-1 standard. The Ramberg-Osgood constitutive model is used for the plastic and failure stages. ,in, To respond realistically, For actual stress, =71.7 GPa, strength coefficient H=800 MPa, strain hardening exponent n=0.11; H and n are taken from the Ramberg-Osgood fitting results of the room temperature tensile curve of 7075-T6 aluminum alloy in MIL-HDBK-5J standard. Bolt stress area This refers to the nominal stress area of ​​the M10 coarse thread according to ISO 898-1 standard, with a clamping length L=20mm.

[0078] At each solution step, the current sampling time is... torque This is converted into an equivalent tangential force applied to the threaded contact surface. The conversion is based on the principle of torque balance, taking the nominal thread diameter d as the lever arm, and the equivalent tangential force... Axial pressure is applied evenly to each node of the threaded contact surface. The forces are applied directly to the bolt head nodes. Using these two forces as boundary conditions, a nonlinear static analysis is performed using the Newton-Raphson iterative solver. The convergence criterion is that the norm of the nodal force residuals is less than [value missing]. After solving, the nodal displacement vectors are obtained.

[0079] Post-processing extracts three outputs from the nodal displacement vector: the axial tightening force is obtained by weighting and summing the axial stresses at each node on the bolt cross-section according to the element area. The unit is Newton; extract the Mises equivalent stress values ​​of all nodes in the preset stress concentration area at the first engagement point of the thread root, and take the maximum value among them. This maximum value is defined as the sampling time. Mises equivalent stress at the root of the thread The unit is megapascal; the axial elongation of the bolt is obtained by taking the difference in axial displacement between the two ends of the bolt. The unit is millimeters. Axial tightening force directly determines the reliability and anti-loosening performance of threaded connections and is a core quality indicator of tightening operations. Therefore, axial tightening force is selected as the main response variable, thus obtaining the initial virtual response. .

[0080] Simultaneously with the finite element solution, the desired response is calculated from the same continuous operational characteristic data stream using a theoretical physical model. .according to Determine the material condition. When When the pressure is less than 503 MPa, it is in the elastic stage, and thus... .when When the stress reaches ≥503MPa, the system enters the plastic and failure stages, and the current stress will be... Substituting into the Ramberg-Osgood constitutive model, the true strain is obtained. ;Calculate the current cross-sectional area based on the assumption of constant volume during plastic deformation. Plastic elongation True stress ,but ;when Reaching yield fastening force When the failure threshold is reached, it is determined that the failure has been reached. The value decreases with further loading, simulating thread stripping failure characteristics. K=0.2 in the theoretical physical model remains constant during training, serving as an unadjustable reference. μ and β in the simplified finite element model are optimized and adjusted in subsequent steps.

[0081] An initial virtual response curve is formed as it changes over time. A desired response curve is formed as the time changes, defining the virtual-to-real mapping distortion. To quantitatively assess the dynamic approximation of these two response curves throughout the entire operation. In the calculation formula, The time window for a complete operation task, based on statistics from 20 skilled operators, shows that the average operation time for tightening an M10 aviation connector is 2.8 seconds. This embodiment takes... =3 seconds; This represents the maximum absolute value of the expected rate of change of response. This value is pre-calculated by the theoretical physical model before operation. Using a typical operating torque curve (torque increases linearly from 0 to 30 N·m over 3 seconds) as input, the theoretical physical model is run once, and the maximum absolute value of the expected rate of change of response is taken as the output. In this example, this value is approximately... Cows per second.

[0082] In real-time computing, for the current sampling time Define the sliding window distortion within the sliding window at the end time.

[0083] ,in , , =2 milliseconds. Sliding window points =500, corresponding to a time span of 1 second.

[0084] when Exceeding the preset accuracy threshold At that time, the gradient descent method was initiated to optimize two adjustable parameters in the simplified finite element model online. The initial value of the thread contact surface friction coefficient μ was... =0.15, learning rate =0.005, perturbation step size =0.001. Axial stiffness reduction correction factor β, initial value. =1.0, learning rate =0.01, perturbation step size =0.002. Parameter updates are performed every 0.1 seconds, with this cycle corresponding to 50 sampling points. m is the iteration number, m=0,1,2,…, where m=0 corresponds to the initial value; This represents the friction coefficient value after the m-th parameter update. = =0.15; This is the friction coefficient value after the (m+1)th parameter update. and These are the order reduction correction factor values ​​after the m-th and (m+1)-th updates, respectively. = =1.0. The update uses gradient descent; taking μ as an example, The gradient is calculated using the central difference method: within the current evaluation period, μ is temporarily set to a value in the simplified finite element model. Complete the fast solution and calculate the corresponding sliding window distortion. Then temporarily set μ to Solve and calculate Finally, update according to the above formula. Apply the same rule to β, i.e. Each time the parameters are updated, subsequent calculations use the new parameter values.

[0085] The convergence condition for parameter optimization is five consecutive evaluation periods. All less than Precision threshold The basis is: It is a dimensionless measure of the difference in rate of change; when When <0.05, the difference in the rate of change does not exceed 5% of the maximum ideal rate of change, and the maximum ideal rate of change is used. The calculation shows that the corresponding rate of change difference is about 600 N / s; within a 3-second operation window, the total clamping force deviation caused by this difference is about 1800 N; the yield clamping force is 29.2 kN, with a relative deviation of about 6.2%, which is just within the boundary of human tactile perception of force difference, and can ensure the realism of training while optimizing convergence efficiency.

[0086] Once the convergence condition is met, fix the current values ​​of μ and β, for example, until convergence to... =0.18, =0.92. The model then uses this set of parameters for all subsequent calculations, and the output fastening force curve is the final virtual response curve, containing the complete mechanical characteristics of the elastic segment, plastic segment, and failure critical point.

[0087] In one embodiment of the present invention, in step S2, generating tactile driving commands based on virtual response curves and sending the tactile driving commands via the interoperability channel to the tactile feedback device integrated into the physical UAV support equipment to drive the tactile feedback device to generate tactile feedback includes the following steps:

[0088] Extract the mechanical feature points corresponding to the elastic deformation stage, plastic deformation stage and failure critical point in the virtual response curve, and convert the mechanical feature points into at least one of force command sequence and torque command sequence to generate haptic drive commands;

[0089] After receiving a tactile driving command, the tactile feedback device drives the actual UAV support equipment to generate at least one of force feedback and torque feedback corresponding to the mechanical state of the virtual response curve in the elastic deformation stage, plastic deformation stage, and failure critical point.

[0090] Specifically, the virtual response curve is a continuous curve with time as the horizontal axis and axial clamping force as the vertical axis, fully experiencing the elastic deformation stage, the plastic deformation stage, and the failure critical point. To generate haptic driving commands, the mechanical feature points corresponding to the three stages are extracted from this virtual response curve.

[0091] The characteristic point of the elastic deformation stage is the location where the slope of the virtual response curve first changes significantly in the initial rising segment. In the elastic stage, the axial clamping force and torque have a linear relationship, and the slope of the curve is approximately constant. When the slope of the virtual response curve decreases by more than 10% from the initial slope, it is determined that the plastic deformation stage has begun. This inflection point is marked as the characteristic point of the elastic end, denoted as... Its coordinates are .

[0092] The characteristic point of the plastic deformation stage is the point in the curve segment between the elastic end and the failure critical point where the slope changes most drastically. The second derivative of each sampling point on this curve segment is calculated using the central difference method, and the point with the largest absolute value of the second derivative is taken as the representative characteristic point of the plastic deformation stage, denoted as [reference needed]. Its coordinates are .

[0093] The failure threshold is when the virtual response curve reaches the yield clamping force. The point at which the axial tightening force reaches its peak and then begins to decrease is marked as the critical failure characteristic point, denoted as [missing information]. Its coordinates are .

[0094] If the operator stops operation before the failure threshold is reached, then No generation, only extraction and .

[0095] After extracting the aforementioned mechanical characteristic points, the axial tightening force values ​​of each characteristic point are converted into corresponding torque commands. For rotary tools like torque wrenches, tactile feedback is manifested in the rotational resistance torque felt by the operator's hand. Therefore, the axial tightening force characteristic values ​​need to be back-calculated into equivalent resistance torques. This is based on classical thread tightening mechanics. Reverse torque ,in For feature points The corresponding axial clamping force value. From this, the target torque values ​​corresponding to the three feature points are obtained. , , .

[0096] The generated haptic actuation command is a sequence of torque commands. During the elastic phase, the command sequence synchronously generates continuously varying torque commands according to the real-time axial clamping force value of the virtual response curve. and At this point, a transient change term for the stress torque value is inserted into the instruction to highlight the transition in physical state. The instruction sets the location from... A rapid drop to zero torque sequence simulates the sensation of resistance disappearing the instant a thread slips. For other types of UAV support equipment, the form of the tactile drive commands is adjusted accordingly; for example, the tactile drive commands on aviation plug assembly / disassembly tools are force command sequences, while those on integrated testing instruments are vibration command sequences. In this embodiment, a torque command sequence is used as the specific form of the tactile drive command.

[0097] Tactile drive commands are sent from the virtual simulation resource to the tactile feedback device in the live simulation resource via the first pair of connections in the interoperability channel. The tactile feedback device integrated on the physical UAV support equipment in the live simulation resource receives the tactile drive commands and generates corresponding tactile feedback. Taking a torque wrench as an example, the tactile feedback device is an active torque feedback mechanism based on a servo motor integrated into the wrench's rotation axis. After receiving the torque command sequence, the servo motor uses the torque value at each moment in the sequence as the target setpoint for torque closed-loop control, outputting the corresponding reverse torque through the servo driver's current loop. This generates resistance feedback at the operator's hand, corresponding to the mechanical state of the virtual response curve. In the elastic stage, the rotational damping felt by the operator increases linearly with the axial tightening force. In the plastic stage, the rate of increase in damping slows down, and the operator feels a slower increase in resistance. At the failure threshold, the damping disappears instantaneously, and the operator experiences the mechanical characteristics of slippage, i.e., the resistance suddenly drops to zero. The tactile feedback generated by the tactile feedback device is at least one of force feedback and torque feedback; for a torque wrench, the tactile feedback is torque feedback; for an aviation plug assembly / disassembly tool, the tactile feedback is force feedback; for a comprehensive testing instrument, the tactile feedback is vibration feedback. Operators perceive changes in the virtual component's state in real time based on the tactile feedback they receive, and adjust their subsequent operational behavior accordingly.

[0098] In one embodiment of the present invention, step S3, which involves calculating the fault trigger probability of each node based on the potential field gradient intensity and the real-time state vector of each node, includes the following steps:

[0099] The continuous operation feature data stream is mapped to the state space of the constructed simulation resource, and the potential gradient field intensity generated by operation behavior B on each node in the state space is calculated. ;

[0100] Get each node The real-time state vector at the current time t Deviating from its preset nominal state vector The degree;

[0101] Calculate the failure trigger probability of each node. The calculation formula is:

[0102]

[0103] Where τ is the duration of the operation;

[0104] Let be the time-varying risk intensity function, i.e. ,in This is the normalization coefficient.

[0105] Specifically, at the start of simulation training, a fault propagation model of the UAV system is retrieved from the constructed simulation resources. This model adopts a directed graph structure, denoted as G=(N,E), where N is the set of nodes and E is the set of directed edges. Each node in the node set N corresponds to a functional component of the UAV, u is the node number (u=1,2,…,U), and U is the total number of nodes; in this embodiment, U=5. Each node... Includes a preset nominal state vector The nominal state vector is composed of the key parameters of the functional component under normal operating conditions. Each node It has a real-time state vector , for The actual working state at the current time t Dimensions and Similarly, the physical parameter types corresponding to each dimension correspond one-to-one with the nominal state vector.

[0106] Each node The drift rate for each state dimension is preset. Its dimension is the same as the nominal state vector, determined by the typical aging characteristics of this functional component, and has been stored in the UAV system fault propagation model when constructing simulation resources. Taking the power module node as an example... For example, its drift rate vector Includes output voltage drift rate Volts per minute, operating temperature drift rate =0.05 degrees Celsius per minute, output current drift rate =0 amperes per minute.

[0107] At the start of the simulation training, i.e., at time t=0, each node The real-time state vector is initialized to its nominal state vector, i.e. During simulation training, the real-time state vectors of each node drift naturally over time to simulate the latent degradation that occurs in the UAV system after long-term operation. Drift amount Is with For vectors of the same dimension, the drift in the d-th dimension is determined by a preset drift rate. And the current time t is determined, that is , where t is the current time, measured in minutes, from the start of the simulation training. At each evaluation time synchronized with S2 (every 0.1 seconds), the constructed simulation resources perform drift calculations and real-time state vector updates.

[0108] Operational behavior B is the torque T(t) and angular velocity applied to the drone by the operator through the drone support equipment, which are determined by the torque T(t) and angular velocity at the current time t. An operating state defined jointly by T(t) and axial pressure F(t). The three physical quantities F(t) are mapped to a three-dimensional operational state space, where each dimension corresponds to torque, angular velocity, and axial pressure, respectively. The operational state point at the current time t is defined as... P(t) is the mathematical representation of operation B in the operation state space.

[0109] For each node A normal operation reference point is preset in the operation state space. This reference point represents the... Ideal operating condition with no damage. The reference point coordinates were pre-set in the UAV system fault propagation model when constructing the simulation resources. (Based on the power module node...) For example, its reference point value is... =5 N·m =0.5 radians per second, =20 Newtons.

[0110] B is Potential gradient field strength generated at the location , where β=1.0; .

[0111] of Deviating from it The degree of deviation is determined by the amount of deviation. Quantization. Let the nominal state vector be... Real-time state vector Where W is the dimension of the state vector, and w is the dimension index (w=1,2,…,W); calculate ,in and They represent The nominal and real-time values ​​in the w-th dimension. (Based on power module nodes) For example, its nominal state vector is (28, 40, 10), with dimension W=3. Its real-time state vector at the 10th minute of simulation training is (27.9, 40.5, 10). The deviation is then... ≈0.51.

[0112] Define the time-varying risk intensity function ,in =0.01 is the normalization coefficient. The cumulative risk is obtained by integrating the time-varying risk intensity function over the operation duration [0,t]. .

[0113] In the discrete implementation, it is updated once at each evaluation time, and the evaluation time interval is... =0.1 seconds, q is the evaluation time number (q=1,2,…), and the q-th evaluation time is ,satisfy Cumulative risk is calculated using rectangular integrals. ,in, for The cumulative risk value at the q-th assessment time. for The cumulative risk value at the (q-1)th evaluation time, the initial cumulative risk . The time-varying risk intensity function value at the q-th evaluation time is derived from the operating state point at that time. and real-time state vector The result is obtained by substituting into the time-varying risk intensity function.

[0114] Finally, the fault trigger probability is obtained. The probability value ranges from [0,1), and tends to 1 as the cumulative risk increases.

[0115] In one embodiment of the present invention, step S3, when the probability of a fault triggering exceeds a preset probability threshold, involves selecting and triggering the fault event corresponding to the node from the fault event set, and propagating the fault event along the directed edge of the UAV system fault propagation model to generate a cascading fault, including the following steps:

[0116] When the probability of a fault triggering a certain node exceeds a preset probability threshold, a fault event corresponding to that node is selected from the fault event set and triggered.

[0117] The triggered fault event will propagate to subsequent nodes along all directed edges originating from that node in the fault propagation model of the UAV system.

[0118] For each directed edge with attached propagation conditions, determine whether the real-time state vector of the node satisfies the propagation conditions. When the propagation conditions are satisfied, trigger the fault event corresponding to the successor node.

[0119] By propagating step by step along the directed edges, a cascading fault is generated, consisting of the initial fault node and its subsequent fault nodes.

[0120] Specifically, in this embodiment, a preset probability threshold is used. =0.7. The rationale for setting the probability threshold to 0.7 is as follows: Setting it too high would result in very few faults being triggered, making the training scenarios unchallenging; setting it too low would lead to excessively frequent faults, causing the training to lose its discriminative power. Through multiple simulation training pre-tests, a threshold of 0.7 was found to trigger faults when operational behavior significantly deviates from normal operation and the system is highly vulnerable, while simultaneously preventing frequent fault triggers due to minor operational fluctuations. At each evaluation time... For each node Judge each one individually. When The failure trigger probability satisfies When this happens, it is determined that the node should trigger a fault event.

[0121] For satisfying nodes The fault event corresponding to the node is selected from the fault event set. Each fault event in the fault event set includes the fault characterization, fault injection parameters, and fault propagation triggering conditions when the node fails; taking the power module node as an example. Taking a corresponding fault event as an example, its fault characteristic is an abnormal drop in output voltage, the fault injection parameter is an output voltage drop to 18 volts, and the fault propagation trigger condition is an output voltage drop duration exceeding 2 seconds. After selecting a fault event, it is marked as triggered, and its fault injection parameters are written into the real-time state vector of that node in the constructed simulation resources. Taking the power module node as an example, after its fault event is triggered, the constructed simulation resources will... The output voltage component in the real-time state vector is forcibly set to 18 volts from the current real-time value to simulate a fault characterization of abnormal output voltage drop.

[0122] The triggered fault event propagates along the directed edges of the UAV system fault propagation model. The fault propagation process is as follows: when a node... After a failure event is triggered, retrieve all directed edges originating from that node in the UAV system failure propagation model; The endpoint of a directed edge starting at point A is... Where x is the number of the endpoint node, x∈{1,2,…,U} and x≠u; simultaneously, obtain the propagation conditions attached to this directed edge; the propagation conditions specify the starting node. Real-time state vector The threshold range that each parameter needs to meet, when When the actual parameter value falls within this threshold range, the fault will occur. spread to From the power module node Pointing to the flight control core board node Taking a directed edge as an example, the propagation condition is that the real-time output voltage of the power module node is below 22 volts or above 32 volts; from Pointing to the data link transceiver node Taking a directed edge as an example, the propagation condition attached to it is that the real-time main frequency of the flight control core board node is less than 100 MHz.

[0123] The propagation conditions for each directed edge are determined one by one. The starting node is obtained from the constructed simulation resources. Current The pre-defined propagation conditions on the directed edge specify the parameters. The permissible range is usually 100%. or In the form of, The lower threshold is... This is the upper limit threshold. The current actual values ​​of each parameter are compared one by one with the threshold values ​​specified in the propagation conditions. When If the actual value of any parameter satisfies the threshold range specified in the propagation condition, the propagation condition is deemed satisfied, and the endpoint node of the directed edge is selected. Marked as a fault-triggered state, and selected from the fault event set... The corresponding fault event is triggered. When When the actual values ​​of all parameters in the path do not meet the propagation conditions, the directed edge will not propagate a fault.

[0124] Through hierarchical propagation along directed edges, the initially triggered node failure event leads to a series of subsequent node failure events. Organizing the initial failure node and all subsequent failure nodes and their corresponding failure events triggered by propagation along directed edges into a whole according to the triggering time sequence and propagation path constitutes a cascading failure.

[0125] Cascading faults provide a complete record of the entire process, from initial fault triggering, propagation through the causal transmission path, to the eventual failure of multiple functional components. Taking a power module node fault as an example, the propagation path of a cascading fault is as follows: Power module node... An abnormal drop in output voltage triggers a node on the flight control core board after the propagation condition is met. Self-test abnormality, flight control core board node After the main frequency decreases to further satisfy the propagation conditions, the data link transceiver node is triggered. Signal attenuation eventually leads to a cascading fault chain involving three nodes. .

[0126] The generated cascading faults are synchronously injected into both virtual simulation resources and real-world simulation resources. The injection method corresponds to the second and third pairs of connections in the interoperability channel. Through the second pair of connections, the faults are transmitted to the real-world simulation resources, causing the instrument displays or alarm devices of the UAV support equipment to synchronously present the corresponding fault characteristics. Through the third pair of connections, the faults are transmitted to the virtual simulation resources, causing the UAV digital prototype to synchronously present the abnormal state corresponding to the fault, including changes in virtual instrument readings and color-changing highlighting of virtual components.

[0127] In one embodiment of the present invention, in step S4, the use of a temporal causal inference method to compare the continuous operation feature data stream and the tactile feedback with a preset standard assurance process causal graph to identify operation steps that deviate from the causal nodes in the standard assurance process causal graph includes the following steps:

[0128] Obtain a pre-built standard assurance process causal graph. The standard assurance process causal graph adopts a directed graph structure, containing multiple causal nodes and multiple directed edges. Each causal node corresponds to a standard assurance operation step, and each directed edge represents the causal dependency relationship between two causal nodes.

[0129] The operation actions and their occurrence times reflected in the continuous operation feature data stream are matched with the preset execution sequence of the standard support operation steps corresponding to each causal node in the standard support process causal diagram; at the same time, the operator's response actions to tactile feedback and their occurrence times reflected in the tactile feedback data are matched with the standard response sequence corresponding to each causal node in the standard support process causal diagram.

[0130] During the matching process, the operation steps in the continuous operation feature data stream and the tactile feedback data that deviate from the causal nodes in the standard assurance process causal graph are identified. The deviations include at least one of the following: missing operation steps, reversed execution order of operation steps, operation step execution time exceeding a preset time threshold, and tactile feedback response actions not conforming to preset response specifications.

[0131] Specifically, a standard support process causal graph, pre-constructed for this simulation training subject, is obtained. This standard support process causal graph is pre-established based on the technical specifications for UAV equipment support. The standard support process causal graph adopts a directed graph structure, containing multiple causal nodes and multiple directed edges. Each causal node corresponds to a standard support operation step or inspection point. Taking the tightening operation of an aviation plug as an example, the causal nodes in the standard support process causal graph sequentially include the following operation steps: Step 1, visually inspect the appearance of the aviation plug; Step 2, manually screw in the thread until initial engagement; Step 3, tighten using a torque wrench to the specified torque; Step 4, check if the tightening torque indication value is within the specified range; Step 5, mark the tightened state. Each directed edge represents the causal dependency between two causal nodes, meaning that the subsequent operation step can only be executed after the preceding operation step is completed. For example, the directed edge from Step 1 to Step 2 indicates that the manual screwing operation must be performed after the visual inspection, and the directed edge from Step 3 to Step 4 indicates that the torque indication value must be checked after tightening.

[0132] Each causal node in the standard assurance process cause-effect diagram also corresponds to a preset execution sequence. The preset execution sequence specifies the normal execution time window for that operation step in the standard operation process. Taking step three as an example, its preset execution sequence specifies that the operation time for tightening with a torque wrench should be within the range of 2 to 4 seconds.

[0133] Each causal node in the standard assurance process cause-effect diagram also corresponds to a preset standard response sequence. The standard response sequence specifies the standard response action and time window that the operator should take after experiencing specific tactile feedback. Taking step three as an example, when the operator feels the torque feedback gradually increasing during tightening, they should stop applying force within 0.3 seconds after the torque reaches the specified value. Taking the failure threshold in step three as an example, when the operator feels the tactile feedback of slipped threads suddenly disappearing, they should stop operation and check the thread condition within 0.5 seconds.

[0134] The continuous operation feature data stream, reflecting various operational actions and their occurrence times, is matched with the preset execution sequence of standard support operation steps corresponding to each causal node in the standard support process cause-effect diagram. The continuous operation feature data stream is segmented along the time axis, identifying the type of operational action corresponding to each operation stage. Taking the tightening operation of an aviation plug as an example, torque is used... The system identifies the start of the operation by observing the inflection point from zero to a point where a significant increase occurs; the failure threshold is identified by the decreasing trend of torque after reaching its peak; and the operation ends when torque drops to zero. Each identified operation and its occurrence time is then compared one by one with the preset execution sequence of each causal node in the standard assurance process cause-effect diagram.

[0135] The operator's response to tactile feedback, along with the timing of its occurrence, is matched against the standard response sequence corresponding to each causal node in the standard assurance process cause-effect diagram. The moments when significant changes in tactile feedback are extracted from the data. Taking a torque wrench as an example, the tactile feedback data is a torque feedback sequence; the turning point where the torque value changes from linear growth to a slowdown, and the moment when the torque value suddenly drops, are extracted. These moments correspond to the elastic end feature points in the virtual response curve. and failure threshold The actual time of the operator's actions after the aforementioned time point is compared with the response time window specified in the standard response sequence.

[0136] During the comparison process, operational steps that deviate from the causal nodes in the causal graph of the standard assurance process are identified in the continuous operation feature data stream and tactile feedback. The types of deviation include at least one of four. The first type is a missing operational step; when the corresponding operational action for a causal node in the standard assurance process causal graph is not detected in the continuous operation feature data stream, the operational step is determined to be missing. The second type is an inverted execution order of operational steps; when the order of operational actions detected in the continuous operation feature data stream is inconsistent with the causal dependency relationship specified by the directed edges in the standard assurance process causal graph, the execution order of operational steps is determined to be inverted. The third type is an operational step execution time exceeding a preset time threshold; when the actual execution time of an operational step exceeds the time window specified by the preset execution sequence of the causal node, the execution time of the operational step is determined to be excessive. The fourth type is a tactile feedback response action that does not conform to the preset response specification; when the operator's response action after receiving tactile feedback and the timing of its occurrence do not conform to the standard response sequence, the tactile feedback response action is determined to not conform to the preset response specification.

[0137] In one embodiment of the present invention, step S4, determining the cascading impact corresponding to the identified deviation operation step based on the cascading fault, includes the following steps:

[0138] The occurrence time of the identified deviation operation step is compared with the trigger time of each fault event in the cascaded fault to determine the fault event that is temporally associated with the deviation operation step.

[0139] Based on the fault propagation path information contained in the cascaded faults, trace the path through which the deviation operation step triggers subsequent cascaded faults via fault transmission relationships, and determine the cascaded impact corresponding to the deviation operation step.

[0140] Specifically, the timing of the identified deviation from the operating procedure is compared with the timing of the triggering of each fault event in the cascading fault.

[0141] The occurrence time of deviations from the operation steps is extracted from the continuous operation feature data stream. For deviations of the operation step missing type, the median time of the preset execution sequence of the step in the standard assurance process cause-effect graph is taken as the reference occurrence time. For deviations of the operation step execution order reversed type, the start time of the actually detected operation action is taken as the occurrence time. For deviations of the operation step execution time exceeding the preset time threshold type, the actual start time of the operation step is taken as the occurrence time. For deviations of the haptic feedback response action not conforming to the preset response specification type, the moment when the haptic feedback changes significantly is taken as the occurrence time.

[0142] In a cascading fault, the triggering times of each fault event are recorded by constructed simulation resources during simulation training. When the fault triggering probability of a node exceeds a preset probability threshold, the constructed simulation resources trigger the fault event of that node and record the triggering time. When a fault propagates along a directed edge and triggers fault events of subsequent nodes, the constructed simulation resources similarly record the triggering times of each subsequent fault event.

[0143] A cascading failure consists of an initial fault node and all subsequent fault nodes, with the trigger time recorded for each fault node's failure event. This is the basis for a cascading failure chain. For example, power module node The fault event is triggered at the time of Flight control core board node The fault event is triggered at the time of Data link transceiver node The fault event is triggered at the time of And satisfy .

[0144] The occurrence time of each deviation operation step is compared with the trigger time of each fault event in chronological order. If the occurrence time of the deviation operation step is earlier than the trigger time of a fault event, and the time interval between the two is less than a preset temporal correlation threshold, then the deviation operation step and the fault event are determined to be temporally correlated. The preset temporal correlation threshold is 5 seconds. This value is based on the fact that the typical delay time for fault propagation between adjacent nodes in the fault propagation model of the UAV system is 1 to 3 seconds. The 5-second threshold can cover the normal propagation delay range and leave a certain margin.

[0145] Taking a power module node failure as an example, if the execution time of step three (tightening with a torque wrench) exceeds the limit, the time of occurrence is... And satisfy and If the deviation occurs within seconds, the deviation from the operating procedure is determined to be related to the power module node. The fault events are temporally correlated.

[0146] For fault events that are temporally related to the deviated operation steps, the complete path of subsequent cascading faults is traced based on the fault propagation path information contained in the cascading faults. Starting from the initial fault node that is temporally related to the deviated operation step, the fault propagation model of the UAV system is traced downstream level by level to determine all subsequent fault nodes caused by the initial fault node. This is the cascading fault chain. For example, if the operation steps deviate from the power module node If the failure events are temporally correlated, then along the directed edge... Tracing back to the flight control core board node , and then along the directed edge Tracing back to the data link transceiver node Therefore, the cascading effects of this deviation in operation were determined as follows: the deviation first caused an abnormal drop in the power module output voltage, which then, through a causal propagation path, sequentially triggered a self-test anomaly on the flight control core board and signal attenuation in the data link transceiver. During the tracing process, for each subsequent fault node in the cascading fault chain, its fault characteristics and fault trigger time were recorded. The cascading effects include the initial fault directly associated with the deviation in operation, and all subsequent faults triggered step by step by this initial fault through the causal propagation path.

[0147] The identified deviation steps and their types, the determined time-series associated failure events, and the traced cascading effects are integrated to generate an interpretability assessment report. The interpretability assessment report includes: the operation name, deviation type, actual occurrence time, and standard execution sequence of the corresponding causal node in the standard assurance process cause-effect diagram for each deviation step; for each deviation step, it indicates whether it belongs to the category of missing operation step, reversed operation step execution order, operation step execution time exceeding a preset time threshold, or tactile feedback response action not conforming to preset response specifications; and it demonstrates how the deviation steps trigger subsequent failures cascadingly through the failure propagation chain.

[0148] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A virtual-real fusion simulation training method for unmanned aerial vehicle (UAV) equipment support, characterized in that, Includes the following steps: S1: Construct a simulation resource library for UAV equipment support. The simulation resource library includes real-world simulation resources based on physical UAV support equipment, virtual simulation resources based on digital UAV prototypes, and construction simulation resources based on UAV system fault propagation models and fault event sets. Establish interoperability channels between real-world simulation resources, virtual simulation resources, and constructed simulation resources; S2: During the simulation training process, collect continuous operational feature data streams generated by the physical interaction between the operator and the UAV support equipment; The virtual response curve is calculated from the virtual simulation resources, and a tactile driving command is generated based on the virtual response curve. The tactile driving command is sent to the tactile feedback device integrated in the physical UAV support equipment via the interoperability channel to drive the tactile feedback device to generate tactile feedback. S3: Obtain the real-time state vector of each node in the UAV system fault propagation model; calculate the potential field gradient intensity of each node in the UAV system fault propagation model based on the continuous operation feature data stream; calculate the fault trigger probability of each node based on the potential field gradient intensity and the real-time state vector of each node; when the fault trigger probability exceeds the preset probability threshold, select and trigger the fault event corresponding to the node from the fault event set, and propagate the fault event along the directed edge of the UAV system fault propagation model to generate a cascaded fault. S4: After the simulation training is completed, the time-series causal inference method is used to compare the continuous operation feature data stream and the tactile feedback with the preset standard guarantee process causal graph to identify the operation steps that deviate from the causal nodes in the standard guarantee process causal graph. Based on the cascading faults, the cascading effects corresponding to the identified deviations in the operating steps are determined, and an assessment report containing the deviations in the operating steps and their cascading effects is generated.

2. The LVC virtual-real fusion simulation training method for UAV equipment support according to claim 1, characterized in that, S1 includes the following steps: Construct a real-world simulation resource that includes a physical drone support equipment integrated with operational feature sensors and haptic feedback devices. The operational feature sensors are used to collect continuous operational feature data streams generated when operators operate the physical drone support equipment. A virtual simulation resource is constructed, in which a digital prototype of a UAV is included, containing a simplified finite element model for real-time calculation of mechanical response. The simulation resources are constructed, and the fault propagation model of the UAV system in the simulation resources adopts a directed graph structure, which contains multiple nodes and multiple directed edges. Each node corresponds to a functional component of the UAV and is accompanied by a nominal state vector. Each directed edge represents the causal propagation path of the fault from one node to another and is accompanied by a propagation condition. Each fault event in the fault event set corresponds to a node. At the same time, an interoperability channel is established to connect live simulation resources with virtual simulation resources, live simulation resources with construction simulation resources, and virtual simulation resources with construction simulation resources.

3. The LVC virtual-real fusion simulation training method for UAV equipment support according to claim 2, characterized in that, In step S2, obtaining the virtual response curve from the virtual simulation resources includes the following steps: The continuous operation feature data stream is used as a boundary condition input to the finite element simplified model of key components in the UAV digital prototype. The continuous physical response of the virtual component under the force generated by the operator's manipulation of the UAV support equipment is calculated in real time to obtain the initial virtual response curve. Simultaneously, a theoretical physical model is established based on the material physical properties of the key components. The expected response of the operation at the current time t is calculated using the continuous operation feature data stream and the theoretical physical model. ; By calculating the distortion of the virtual-real mapping The cumulative deviation of the rate of change between the initial virtual response curve and the expected response over the entire operation process is calculated using the following formula: in, This represents the virtual response state corresponding to the initial virtual response curve. The time window for a complete operation task; by With the goal of minimization, the parameters of the simplified finite element model are dynamically adjusted until the cumulative deviation between the rate of change of the initial virtual response curve and the rate of change of the desired response throughout the entire operation process meets the preset accuracy requirements, thus obtaining the virtual response curve.

4. The LVC virtual-real fusion simulation training method for UAV equipment support according to claim 3, characterized in that, In step S2, the step of generating tactile driving commands based on the virtual response curve and sending the tactile driving commands to the tactile feedback device integrated into the physical UAV support equipment via the interoperability channel to drive the tactile feedback device to generate tactile feedback includes the following steps: Extract the mechanical feature points corresponding to the elastic deformation stage, plastic deformation stage and failure critical point in the virtual response curve, and convert the mechanical feature points into at least one of force command sequence and torque command sequence to generate haptic drive commands; After receiving a tactile driving command, the tactile feedback device drives the actual UAV support equipment to generate at least one of force feedback and torque feedback corresponding to the mechanical state of the virtual response curve in the elastic deformation stage, plastic deformation stage, and failure critical point.

5. The LVC virtual-real fusion simulation training method for unmanned aerial vehicle (UAV) equipment support according to claim 1, characterized in that, In step S3, the calculation of the fault triggering probability of each node based on the potential field gradient intensity and the real-time state vector of each node includes the following steps: The continuous operation feature data stream is mapped to the state space of the constructed simulation resource, and the potential gradient field intensity generated by operation behavior B on each node in the state space is calculated. ; Get each node The real-time state vector at the current time t Deviating from its preset nominal state vector The degree; Calculate the failure trigger probability of each node. The calculation formula is: Where τ is the duration of the operation; Let be the time-varying risk intensity function, i.e. ,in This is the normalization coefficient.

6. The LVC virtual-real fusion simulation training method for UAV equipment support according to claim 5, characterized in that, In step S3, when the probability of a fault triggering exceeds a preset probability threshold, the fault event corresponding to the node is selected from the fault event set and triggered, and the fault event is propagated along the directed edge of the UAV system fault propagation model to generate a cascading fault. This includes the following steps: When the probability of a fault triggering a certain node exceeds a preset probability threshold, a fault event corresponding to that node is selected from the fault event set and triggered. The triggered fault event will propagate to subsequent nodes along all directed edges originating from that node in the fault propagation model of the UAV system. For each directed edge with attached propagation conditions, determine whether the real-time state vector of the node satisfies the propagation conditions. When the propagation conditions are satisfied, trigger the fault event corresponding to the successor node. By propagating step by step along the directed edges, a cascading fault is generated, consisting of the initial fault node and its subsequent fault nodes.

7. The LVC virtual-real fusion simulation training method for unmanned aerial vehicle (UAV) equipment support according to claim 1, characterized in that, In step S4, the temporal causal inference method is used to compare the continuous operation feature data stream and the tactile feedback with a preset standard assurance process causal graph to identify operation steps that deviate from the causal nodes in the standard assurance process causal graph. This includes the following steps: Obtain a pre-built standard assurance process causal graph. The standard assurance process causal graph adopts a directed graph structure, containing multiple causal nodes and multiple directed edges. Each causal node corresponds to a standard assurance operation step, and each directed edge represents the causal dependency relationship between two causal nodes. The operation actions and their occurrence times reflected in the continuous operation feature data stream are matched with the preset execution sequence of the standard support operation steps corresponding to each causal node in the standard support process causal diagram; at the same time, the operator's response actions to tactile feedback and their occurrence times reflected in the tactile feedback data are matched with the standard response sequence corresponding to each causal node in the standard support process causal diagram. During the matching process, the operation steps in the continuous operation feature data stream and the tactile feedback data that deviate from the causal nodes in the standard assurance process causal graph are identified. The deviations include at least one of the following: missing operation steps, reversed execution order of operation steps, operation step execution time exceeding a preset time threshold, and tactile feedback response actions not conforming to preset response specifications.

8. The LVC virtual-real fusion simulation training method for UAV equipment support according to claim 7, characterized in that, In step S4, determining the cascading impact corresponding to the identified deviation operation step based on the cascading fault includes the following steps: The occurrence time of the identified deviation operation step is compared with the trigger time of each fault event in the cascaded fault to determine the fault event that is temporally associated with the deviation operation step. Based on the fault propagation path information contained in the cascaded faults, trace the path through which the deviation operation step triggers subsequent cascaded faults via fault transmission relationships, and determine the cascaded impact corresponding to the deviation operation step.

Images