Axle breakage protection method and system for circuitous wheel of aerial passenger device

Abstract

The invention provides a broken shaft protection method for a circuitous wheel of an aerial passenger device, and relates to the field of coal mine safety monitoring. The problems that in the prior art, space-time alignment errors of multi-source data are large, a physical field is distorted, and real-time performance is insufficient are solved. Through dynamic graph structure modeling, heterogeneous sensors such as a vibration sensor, a temperature sensor, a displacement sensor and the like are mapped into space-time association graph nodes, edge connection is constructed based on spatial proximity and time synchronism, and continuity characteristics of an equipment structure are coded. And generating a three-dimensional continuous field and dynamically visualizing the three-dimensional continuous field through a ray casting volume rendering technology. According to the scheme, high-precision and high-real-time technical support is provided for broken shaft fault early warning.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of coal mine safety monitoring, in particular to a detour wheel broken shaft protection method of an aerial personnel device. BACKGROUND

[0002] The aerial personnel device is an important equipment for transporting personnel in coal mines, and the detour wheel as a key transmission component will cause equipment downtime or even personnel casualties once a broken shaft failure occurs. At present, the state monitoring of the detour wheel mainly relies on multi-source sensing devices such as vibration sensors, temperature sensors, displacement sensors, etc. However, there are the following pain points in actual application:

[0003] (1) The sensor types, precision and sampling frequency differ greatly, leading to difficulties in data space-time alignment, such as inconsistent space-time coverage between vibration data and temperature data;

[0004] (2) Sensor installation position drift, environmental interference and equipment aging lead to frequent abnormal data;

[0005] (3) The traditional analysis method fails to fully combine the equipment dynamics characteristics, leading to inaccurate fault feature extraction and insufficient real-time early warning capability.

[0006] The existing technology mainly adopts threshold alarm or simple spectrum analysis, which fails to effectively fuse multi-source sensing information and lacks dynamic three-dimensional visualization support for the running state of the equipment, making it difficult to realize early and accurate broken shaft warning. SUMMARY

[0007] In view of the needs proposed in the above background art, the present application embodiment provides a detour wheel broken shaft protection method of an aerial personnel device, which aims to solve the technical problems of large space-time alignment error of multi-source sensor data, distortion of physical field continuity, abnormal missed detection caused by multi-parameter decoupling and insufficient real-time of traditional algorithms in fault equipment health state evaluation through dynamic graph structure construction, isometric graph neural network, semantic information gain fusion and dynamics-data joint modeling technology.

[0008] A detour wheel broken shaft protection method of an aerial personnel device, the specific steps include:

[0009] The multi-source sensor data of vibration sensors, temperature sensors, displacement sensors, etc. distributed in the state of coal mine equipment are uniformly modeled as a dynamic graph structure to capture the spatio-temporal correlation. Each sensor is regarded as a node in the graph, and the node features include its spatial coordinates, collection timestamp, and measured parameters (such as vibration amplitude, temperature, displacement). The connection edges between sensors are established according to the spatio-temporal proximity: if two sensors collect data simultaneously within a preset spatial radius (such as 100 meters) and time window (such as 10 seconds), an edge is formed. The edge features include spatial distance, time difference, and data similarity (for example, the smaller the temperature difference measured by two sensors, the higher the similarity). This step converts discrete sensor data into a graph network with topological relationships, providing structured input for subsequent analysis while preserving the spatio-temporal dynamic characteristics of the data.

[0010] The spatio-temporal calibration and dynamic modeling of sensor data are realized through an isometric graph neural network. The isometric message passing mechanism ensures that the model maintains the geometric consistency of physical quantities (such as velocity, force field) under any rotation or translation transformation. Each sensor node sends "messages" to its neighboring nodes, which contain position offset correction and velocity adjustment suggestions. These suggestions are dynamically calculated based on the relative motion and data difference between sensors. For example, if a temperature sensor detects an abnormal running speed, its neighboring vibration sensor node will adjust its data according to the message, simulating the transmission process of the motor driving force. Through multiple rounds of message passing, the data of all nodes gradually converges to a unified spatio-temporal coordinate system, eliminating errors caused by sensor drift or clock desynchronization. Finally, the position and velocity of the nodes are updated using the Euler integration method to ensure the continuity of the dynamic model in the time dimension.

[0011] Based on the principle of maximizing information gain in semantic space, high-quality data is selected and noise is removed. First, a semantic label graph is constructed, with temperature and other parameters as label nodes. The edge weights are defined by calculating the physical correlation strength between labels (such as the empirical correlation between temperature and salinity). Each sensor data point is assigned an information quantity in the semantic graph based on its measurement quality (such as signal-to-noise ratio) and the label it belongs to. The information quantity is propagated through the correlation between labels: for example, a high-precision measurement enhances the credibility of neighboring deep labels. Finally, a greedy algorithm is used to iteratively select data points with the maximum information gain, prioritizing the preservation of observation results in critical areas (such as data in the thermal fatigue area), while filtering out abnormal values affected by oil contamination, sensor aging, loose installation, or equipment failure. This process significantly improves the representativeness and reliability of the data set, providing optimized input for three-dimensional reconstruction.

[0012] The fused data is converted into a dynamic three-dimensional hydrological field, and interactive visualization is realized. Based on a mass-damping-stiffness model, the device is discretized into grid cells, and the dynamic behavior of each cell is driven by adjacent sensor data: for example, external motor driving forces act on the grid boundary, and internal cells transfer energy through spring damping effects, simulating turbulent diffusion processes. Using volume rendering technology, temperature, salinity and other parameters are mapped to color and transparency, and a three-dimensional voxel model is generated through a ray casting algorithm. Users can adjust the time axis in real time to observe the evolution of the thermal-mechanical coupling field distribution under the impact of the load, or focus on a specific profile to analyze the fatigue layer and damage gradient zone of the material. In addition, the anomaly detection module automatically marks features such as crack propagation or stress concentration by comparing the reconstructed field with historical patterns, providing support for early warning.

[0013] Further: An overhead passenger device detour wheel shaft protection system, comprising:

[0014] A multi-source sensing module for collecting device state parameters such as vibration, temperature, displacement, etc.

[0015] A graph construction module for constructing a dynamic spatiotemporal correlation graph model.

[0016] An isometric alignment module for realizing data spatiotemporal alignment and dynamic modeling based on a geometric isometric neural network.

[0017] A semantic fusion module for constructing a semantic correlation network and realizing data filtering and fusion.

[0018] A three-dimensional reconstruction and visualization module for generating a dynamic three-dimensional device state field and providing an interactive analysis interface.

[0019] The present application has the following advantages: Based on the continuum hypothesis of the device structure, the present application correlates the sensor network topology with the continuity of the physical field, breaking through the limitations of traditional independent interpolation that ignores local correlation. The method of the present application significantly improves the early warning accuracy and real-time performance of the detour wheel shaft fault through dynamic graph structure modeling and multi-source data fusion, providing a reliable technical guarantee for the safe operation of coal mines. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0021] Figure 1 A flowchart of the method of the present application is shown. DETAILED DESCRIPTION

[0022] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of description and illustration, and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in the present application show the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowcharts can not be implemented in sequence, and the steps without logical context relationship can be reversed in sequence or implemented simultaneously. In addition, one or more other operations can be added to the flowcharts or one or more operations can be removed from the flowcharts under the guidance of the content of the present application.

[0023] In addition, the embodiments described in the present application are only some of the embodiments of the present application, not all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative labor belong to the scope of protection of the present application.

[0024] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features. It should also be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. In the description of the present application, it should also be noted that the terms "first", "second", "third" and the like are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0025] The present application will be described in detail below with reference to the drawings in the specification Figure 1 The present application will be described in detail below with reference to the drawings in the specification

[0026] The present application is based on four-step core processes of multi-sensor data graph representation construction, isomorphic message passing and space-time alignment, information gain driven data selection and fusion, and dynamic three-dimensional reconstruction and visualization, fuses the isomorphic graph neural network idea and the semantic space information gain maximization method, and realizes high-precision three-dimensional modeling and dynamic visualization of data. The specific steps are as follows:

[0027] Step one: multi-sensor data graph representation construction

[0028] The goal of this step is to model multi-source sensor (vibration sensor, temperature sensor, displacement sensor, etc.) data into a dynamic graph structure to capture spatio-temporal correlation.

[0029] The dynamic graph structure is constructed in particular by the following method:

[0030] 1.1, First define the nodes of the dynamic graph structure: regard each sensor (such as a temperature sensor or a vibration sensor device) as a node, and the node features include the position coordinates a timestamp t, a data type; the position coordinates include the three-dimensional position (installation position X, Y, Z) of the sensor; the timestamp t represents the specific time of data collection; the data type includes a plurality of different measurement parameters, including temperature T, salinity S, running speed , etc.

[0031] The node set V represents the set of all sensor nodes, and each node v i corresponds to a sensor, and the mathematical expression of the node set V is V = {v i |i∈Sensors}.

[0032] 1.2, Then define the edges of the dynamic graph structure: the edges are used to connect the sensors that affect each other in the same space-time range, representing the potential association between data, and the generation conditions of the edges include spatial proximity and temporal proximity, wherein the spatial proximity includes that the Euclidean distance between the sensors does not exceed a preset radius r, and the temporal proximity includes that the data collection time difference is less than a time window τ; the edge features include the spatial distance the time difference Δt ij and the data similarity For example, the distance between a temperature sensor and a vibration sensor represents the spatial distance; the interval of two measurements is 5 minutes, which represents the time difference;

[0033] The data similarity is calculated by a Gaussian kernel function, and the value range is [0, 1], and the closer the value is to 1, the more similar the data is, wherein d i is the measurement data vector of the sensor v i ; ||d i -d j || represents the Euclidean distance of the data vectors of the two sensors, reflecting the data difference; the data difference is mapped to the similarity by the Gaussian kernel function, quantifying the observation consistency between the sensors. The edges with high similarity have greater weights in subsequent analysis.

[0034] The edge set E represents the set of all edges, and the edge e ij connects the sensors v i and v j , and exists only when the space-time proximity condition is met, and the mathematical expression of the edge set E is wherein represents the sensors v i and vj Euclidean distance between v ij and v i , r represents the spatial proximity threshold (e.g. 500 meters), determines whether the sensors are connected due to spatial correlation, Δt j represents the time difference of data collection between v ij and v i , τ represents the time window threshold (e.g. 10 minutes), determines whether the sensors are connected due to time synchronization, the mathematical expression of the edge set E defines the generation rule of the graph, ensuring that only spatio-temporally proximal sensors will be connected, thereby capturing the relevance of local phenomena.

[0035] 1.3, Dynamic graph structure generation based on nodes and edges: The entire dynamic graph structure G contains a node set V and an edge set E, which is mathematically expressed as G = (V, E).

[0036] Further, the edge generation rule introduces a running speed direction consistency constraint, defining a directional edge weight

[0037]

[0038] where θ ij is the running speed direction angle between v i and v j , σ v is the running speed standard deviation.

[0039] When , even if it exceeds the preset spatial radius, it still establishes a connection to capture stress concentration interaction.

[0040] Further, for the planar characteristics of displacement sensor data, superpixel segmentation is used to generate virtual nodes, and virtual nodes and adjacent temperature sensor nodes are connected through spatial overlap area calculation of edge weight.

[0041] Step two: isometric message passing and spatio-temporal alignment

[0042] The purpose of this step is to dynamically calibrate the dynamic graph structure constructed in step one, eliminate spatio-temporal deviations caused by sensor drift, clock desynchronization or measurement errors, and align all data to a unified spatio-temporal coordinate system, providing a physically consistent data basis for subsequent three-dimensional reconstruction. Specifically, isometric graph neural network is used to realize spatio-temporal alignment and dynamic modeling of sensor data.

[0043] Since the sensor network can move or rotate as a whole due to motor driving force, equipment drift or human adjustment, how to ensure that the model-predicted physical quantities (such as temperature gradient, running speed direction) do not become false due to changes in the coordinate system is the main purpose of this step, so in this step, the spatio-temporal alignment and dynamic modeling of sensor data are realized through the isometric graph neural network. The significance of isometry is that no matter how the sensor network as a whole rotates or translates, the model-predicted physical quantities will be transformed synchronously, maintaining geometric consistency.

[0044] For example, if the entire sensor network translates 100 meters east, the predicted value of the position of all nodes in the model will automatically increase by 100 meters in the east direction, while the scalar parameters such as temperature and salinity remain unchanged.

[0045] Message passing is the core mechanism of graph neural networks, which refers to the process in which nodes in a graph exchange information with their neighbor nodes through edges and collaboratively correct their own states. In the fault monitoring scenario, this process simulates the data negotiation and physical quantity propagation between sensors.

[0046] Data negotiation between sensors: each sensor (node) adjusts its own data according to the observation results of its neighbors to eliminate errors (such as position drift, clock bias).

[0047] Physical quantity propagation: the influence of fault phenomena (such as motor driving force, thermal fatigue area) is transmitted through edges.

[0048] In this step, the specific process of message passing includes:

[0049] 2.1, Message generation: taking edge features and node features as input, the generated message content includes position correction suggestions and speed correction suggestions, for example, position correction suggestions are based on relative position differences, suggesting nodes to adjust coordinates, while speed correction suggestions are based on speed differences; the message The mathematical expression of message generation is shown in equation (1):

[0050]

[0051] In equation (1), ε ij is the edge embedding, which combines spatial distance, time difference and data similarity; θ k is the neural network, which learns how to generate correction quantities based on edge features; represents the speed difference vector.

[0052] For example, the difference between temperature sensor v1 and sonar v2 is S1-S2=0.2‰, and the model generates a message including:

[0053] Position correction: meters, suggesting v1 to move west;

[0054] Velocity correction: Suggest v1 increase eastward running speed.

[0055] The neural network θ k is a 3-layer fully connected structure, the input layer contains edge features ε ij and node type encoding, the hidden layer uses 128-dimensional Swish activation function, and the output layer generates scaling coefficients for position and velocity correction. During training, a semi-supervised loss function is used:

[0056]

[0057] where the second term forces the velocity field to satisfy mass conservation, ensuring physical consistency.

[0058] 2.2, Message aggregation: each node collects all adjacent edge messages and performs weighted summation, which can be mathematically expressed as formula (2):

[0059]

[0060] In formula (2), edges with high data similarity are given greater weight, such as edges have a stronger impact than edges; based on this, high-confidence sensors dominate the correction process, and low-quality data impact is suppressed.

[0061] Assume that temperature sensor v1 receives three pieces of information, respectively:

[0062] from the vibration sensor

[0063] from the displacement sensor

[0064] from the temperature sensor

[0065] Then the aggregation result is: meters, so the correction result is that the temperature sensor v1 corrects 0.17 meters west.

[0066] 2.3, Euler integration update: the purpose is to apply the correction amount of the message passing to the node state and advance the time step, and the mathematical expression of this update process is shown in formula (3):

[0067]

[0068] Through the update process of formula (3), the position can be updated with the average speed, ensuring smooth motion.

[0069] For example, during the calibration of the sensor network during transients such as device start-stop, load mutation, etc., in the initial state, the temperature sensor array is affected by typhoon, part of the node position drifts (maximum deviation 200 meters), and the noise of the salinity data increases.

[0070] When the message is transmitted, the high-weight position correction message is sent based on the vibration sensor, and the drifted temperature sensor is gradually pulled back. The high-confidence temperature sensor adjusts the running speed prediction of the adjacent nodes through the speed correction message.

[0071] At this time, the spatio-temporal alignment result is that the coordinate error of all temperature sensors converges to within 10 meters, and the standard deviation of the salinity data decreases from 0.5‰ to 0.2‰.

[0072] Step three: information gain driven data selection and fusion

[0073] The purpose of this step is to select the observation results with the largest amount of information and the highest quality from the massive multi-source sensor data, eliminate noise or redundant data, and fuse different parameters (such as vibration amplitude, temperature, displacement) through semantic association to construct a high-confidence three-dimensional fault field semantic label graph.

[0074] 3.1, Construction of semantic label graph: For the physical meaning of the semantic label graph, the specific definition of its label is to abstract the fault parameters as "semantic labels", for example: temperature (L1), salinity (L2), running speed (L3), depth (L4).

[0075] There is an association between labels, and the edge weight (γ pq ) between labels reflects the physical correlation, and the association includes:

[0076] Empirical correlation: such as temperature and salinity in the thermal fatigue area (γ 12 = -0.7).

[0077] Dynamic correlation: such as running speed gradient and temperature gradient in the stress concentration area (γ 23 = 0.9).

[0078] The edge weight γ pq between the labels is dynamically calculated based on the covariance in the sliding time window, and its mathematical expression is shown in equation (4):

[0079]

[0080] The Lazy Greedy algorithm is used for data selection, the upper bound of information gain is pre-calculated, and the candidate set is maintained through the maximum heap, so that the time complexity is reduced.

[0081] 3.2, Data point information propagation: Its essence is to combine the credibility (quality score) of sensor data with the physical relevance (such as the interaction of temperature and salinity) through the semantic label graph, dynamically adjusting the amount of information of each data point on different parameters (labels). This process simulates the natural coupling relationship of different physical quantities in a fault system.

[0082] The data point information propagation process first needs to perform data quality scoring (s i ) initialization, the purpose of initialization is to evaluate the reliability of sensor data, the initialization process mainly includes anchoring vibration sensors (s i = 0.95) > temperature sensors (s i = 0.85) > displacement sensors (s i = 0.75) based on sensor type and historical performance, while calculating the signal-to-noise ratio in real time to dynamically adjust, for example, a temperature sensor due to oil contamination, sensor aging, loose installation, resulting in increased salinity noise, s i From 0.85 to 0.6.

[0083] Subsequently, information propagation is carried out, and the mathematical expression of the propagation process is shown in equations (5) and (6):

[0084]

[0085] In equation (5), represents the propagated information vector, indicating the integrated influence of data point i in the semantic label graph; A represents the propagation matrix, which describes the information propagation strength between different labels, which is defined by equation (6); v i The original information vector of data point i encodes its measured values or features for each label (such as temperature, salinity); through the form of information propagation, the original information of the data point (such as temperature, salinity) is weighted and diffused according to its quality score and label relevance.

[0086]

[0087] Equation (6) defines the propagation matrix, A pq represents the information propagation strength from label p to label q; α represents a hyperparameter that controls the influence degree of label relevance; γ pq represents the association weight between label p and label q, based on physical laws (such as temperature-salinity negative correlation); ∑ k≠p γ pk represents the sum of the association weights of label p with all other labels.

[0088] In the entire equation (6), αγ pq is the numerator, directly reflecting the influence strength of label p on q, the stronger the relevance (γ pqThe more significant the propagation is; 1 + a∑k ≠ pγ pk is the denominator, which represents the normalization term to prevent label p from excessively affecting other labels due to the association with too many labels; the purpose of designing the propagation matrix is to balance the propagation and adjust the flexibility, avoid some labels (such as temperature) dominating the entire network due to the association with multiple parameters (salinity, running speed, depth), and control the global propagation strength through a, a = 1 when the association has a significant impact, and a = 0 when the information does not propagate.

[0089] 3.3, Information gain maximization sampling: information gain measures the contribution of a data point to the overall amount of information, in fault monitoring, data points with high information gain usually have high data quality, spatial or temporal uniqueness, and multi-parameter association, etc. characteristics, then selecting the data point with the maximum information gain is equivalent to preferentially retaining the observation results that contribute most to the explanation of the current fault phenomenon.

[0090] The mathematical expression of the information gain calculation is shown in equation (7):

[0091]

[0092] D S represents the selected data point set; d represents the candidate data point; represents the information amount of data point i on label l after propagation; φ' represents the information gain function (such as logarithmic function or linear decay function); equation (7) calculates the total information amount of the selected data set D S on all labels, then assumes that the candidate data point d is added, the total information amount is recalculated, and the difference between the two is the information gain G(d) of d.

[0093] The greedy algorithm is used to select the data point with the maximum information gain step by step to realize the iteration of maximum sampling. Specifically, first, the initialization is needed, through the empty set D S , the candidate set D P contains all sensor data; the iterative selection process calculates the information gain of each candidate data point d ∈ D P , calculates G(d); selects the point with the maximum gain , adds it to D S and removes it from D P ; when the preset data amount (such as retaining 50% data) or the information gain increment is lower than the threshold (G(d) < 0.1), the iteration is terminated.

[0094] Further, the selection of the information gain function φ' can be a logarithmic function or a linear decay function:

[0095] Logarithmic function: φ'(x) = log(1+x), which can suppress the excessive dominance of high information amount areas;

[0096] Linear decay function: φ'(x) = x e -βx which is used to control the marginal diminishing returns.

[0097] 3.4, Data fusion: The purpose is to generate a three-dimensional field by weighted fusion of filtered data, and the weighted fusion formula is shown in equation (8):

[0098]

[0099] Fused(x,y,z,t) represents the physical quantity value after multi-sensor data fusion at position (x,y,z) and time t, W i is the total information of data point i in the semantic graph

[0100] For example, in a certain grid cell, the temperature sensor d1 (salinity S1 = 34.5‰, s1 = 0.9) and the displacement sensor d2 (salinity S2 = 34.8‰, s2 = 0.7), then the data fusion result is S Fused ≈34.6‰; Obviously, in this step, high weight is given to high information temperature sensor data, and displacement sensor data plays a supplementary role.

[0101] Through the information gain driven strategy, the system has "intelligent attention mechanism" - when the resources are limited (such as insufficient real-time transmission bandwidth), the data with the strongest explanation power for the current fault phenomenon (such as the data of the stress concentration area) is preferentially transmitted, while redundant or low-quality observations (such as repeated temperature records in calm sea areas) are suppressed. This way of data selection and fusion significantly improves the efficiency and accuracy of three-dimensional reconstruction, providing reliable support for extreme weather fault monitoring.

[0102] Step four: dynamic three-dimensional reconstruction and visualization

[0103] This step four is to convert the fused multi-source sensor data into a dynamic three-dimensional fault field (such as temperature field, salinity field, running speed field), and realize interactive analysis through visualization technology, so that researchers can intuitively observe the spatio-temporal evolution of fault phenomena (such as crack propagation and movement, thermal fatigue area fluctuation), and support anomaly detection and early warning.

[0104] The physical modeling of three-dimensional field reconstruction includes the following steps:

[0105] 4.1, Grid discretization: The purpose is to divide the continuous sea area into regular three-dimensional grid cells (voxels) for numerical calculation and visualization. Specifically, the target sea area is divided into three-dimensional grid cells (voxels), each cell contains position (x,y,z) and physical parameters (such as temperature T, salinity S).

[0106] Example: South China Sea region (110°E-120°E, 10°N-20°N, 0-1000m depth) is divided into 1km x 1km x 10m voxels, a total of 106106 units.

[0107] 4.2, Establish parameterized field equation: describe the dynamic behavior of the fault field (such as motor driving force movement, temperature diffusion) through physical model, the parameterized field equation is shown in equation (9):

[0108]

[0109] In equation (9), M represents the mass matrix, which is used to represent the inertia of the water body, and the mass of the deep sea region unit is larger; C represents the damping matrix, which is used to simulate the turbulent dissipation and seabed friction; K represents the stiffness matrix, which is used to reflect the compressibility of the transmission system and the density gradient effect; represents external force, such as load impact driven surface wind stress, mechanical vibration.

[0110] The mass matrix M is calculated by the grid cell volume and the density of the transmission system, the damping matrix C is simulated by the Smagorinsky model to simulate turbulent dissipation, and the stiffness matrix K is based on the TEOS-10 state equation to correlate temperature and salinity parameters.

[0111] 4.3, Dynamic update: update the state (position, velocity, temperature, etc.) of each grid through numerical integration, solve the equation at each time step (such as Δt = 1 minute), and update the displacement x and velocity Simulate the dynamics of motor driving force, internal wave, etc., the solving process is shown in equation (10) and equation (11):

[0112]

[0113]

[0114] For example, the initial state: the running speed of a voxel is 0.5m / s, the external force (load impact) is applied, the acceleration is calculated Then update the speed: Update the position:

[0115] 4.4, Data mapping: map data to three-dimensional field, the purpose is to interpolate discrete sensor data to continuous grid and fill in data-free areas.

[0116] Use Kriging interpolation to map discrete sensor data to continuous grid, the Kriging interpolation process is shown in equation (12):

[0117]

[0118] In formula (12), the weight ω i The sensor data quality score s i And the spatial distance, high confidence data has a greater impact.

[0119] 4.5, visualization of dynamic volume rendering: the purpose is to convert three-dimensional field into intuitive graphics, support user interaction analysis;

[0120] 4.51, volume rendering process, including color and transparency mapping:

[0121] Temperature field: red (high temperature) to blue (low temperature) gradient, transparency changes with gradient (stress concentration area is more opaque).

[0122] Salinity field: green (low salt) to purple (high salt) gradient, enhanced contrast in pycnocline area.

[0123] Running speed field: arrow length represents speed size, streamline represents motion direction.

[0124] 4.52, light projection: emit light from user's perspective through three-dimensional field, integrate optical properties (color, transparency) along path, its mathematical expression is shown in formula (13):

[0125]

[0126] T(p(t)) represents the temperature / salinity value of path point p(t).

[0127] σ(p(s)) represents absorption coefficient, controls transparency (such as high salinity water is more "turbid").

[0128] The final output generates two-dimensional pixel color, forms projection image of three-dimensional field.

[0129] Through dynamic three-dimensional reconstruction and visualization, abstract fault data is converted into intuitive spatio-temporal model, so that researchers can "see through" the internal structure of fault, quickly locate abnormal phenomena, and provide visual basis for decision-making of broken shaft fault warning, ecological protection, etc.

[0130] Overall, the method of the application maps multi-source sensors into a spatio-temporal correlation graph node, generates edge connection based on spatial proximity and time synchronization, breaks through the limitations of traditional independent interpolation which ignores the topological relationship of sensor network, adopts geometric covariant message passing mechanism, ensures the geometric consistency of physical quantities (running speed, temperature gradient) under arbitrary coordinate transformation, eliminates sensor drift error, establishes a temperature-salt-flow multi-parameter semantic correlation graph, dynamically allocates data weight through the physical coupling strength between labels, preferentially fuses high information content area, improves the sensitivity of anomaly detection, and finally combines the quality-damping-stiffness physical equation with the ensemble Kalman filter to realize the dynamic assimilation of three-dimensional field under the observation data driving.

[0131] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for protecting the axle of a bypass wheel of an aerial passenger transport device from breakage, characterized in that, The specific steps include: Based on the multi-source sensors distributed in the status of coal mine equipment, a dynamic spatiotemporal correlation graph model is constructed to uniformly represent heterogeneous data. Subsequently, a geometrically variable neural network was used to achieve spatiotemporal alignment and dynamic modeling of sensor data; Based on the dynamic consistency node states output by spatiotemporal alignment and dynamic modeling, a semantic association network is constructed for data filtering and fusion optimization. Finally, the optimized dataset is used to generate a dynamic 3D fault field and enable interactive analysis.

2. The method according to claim 1, characterized in that, The construction of the dynamic spatiotemporal relational graph model specifically includes defining node features, defining edge connection rules, and generating topology relationships. Defining node features involves abstracting each sensor as a graph node, and the node features include spatial coordinates, time series labels, and multi-dimensional measurement parameters. Defining edge connection rules involves generating connecting edges between nodes based on spatial proximity and time synchronization, and the edge features include spatial displacement, time series deviation, and data similarity measures. Generating topology relationships involves transforming discrete sensor data into a graph network with a dynamic spatiotemporal topology.

3. The method according to claim 1, characterized in that, The method of using geometric equivariant neural networks to achieve spatiotemporal alignment and dynamic modeling of sensor data specifically includes equivariant message passing, dynamic calibration iteration, and dynamic calibration iteration. The equivariant message passing design has a translation- and rotation-invariant message generation mechanism, and the nodes pass messages containing position correction and state adjustment amounts to ensure the geometric consistency of physical quantities such as vibration modes, temperature gradients, and stress distribution. The dynamic calibration iteration gradually eliminates sensor clock deviation and position drift error through multiple rounds of message passing, so that the node data converges to a unified spatiotemporal reference system; The dynamic state update employs an explicit time integration method to update node positions and state parameters, maintaining the physical continuity of the dynamic model in the time dimension.

4. The method according to claim 1, characterized in that, The construction of a semantic association network for data filtering and fusion optimization specifically includes semantic label modeling, information propagation and distribution, and optimization selection strategies. The semantic label modeling involves defining hydrological parameters as semantic nodes, calculating the association strength weights between labels through physical experience correlation, and forming a semantic network topology. The information propagation and distribution are based on the sensor data quality score and its semantic label. The global information contribution of each data point is calculated in the semantic network, and cross-parameter propagation is carried out through label association. The optimization selection strategy employs a greedy iterative algorithm to select a subset of data points with the greatest information gain, prioritizing the retention of high-value observations in key areas while filtering out noise and outliers, and outputting an optimized, high-quality dataset.

5. The method according to claim 1, characterized in that, The process of generating a dynamic 3D fault field and enabling interactive analysis includes discretizing the faulty water body into grid cells, constructing a mass-damping-stiffness dynamic equation, and driving the dynamic behavior of the grid cells through data from nearby sensors. A physical field interpolation algorithm is used to map multi-source data to a continuous grid, and the dynamic equations are combined to constrain the generation of a spatiotemporally continuous multi-parameter fault field. A ray projection volume rendering technique is used to map physical parameters into color and transparency, allowing users to manipulate the timeline, cut profiles, and overlay multi-parameter fields in real time. Based on historical pattern comparisons, abnormal features are automatically marked, providing visual decision support for fault monitoring.

6. A system for protecting the axle of a bypass wheel in an aerial passenger transport device, characterized in that, include: Multi-source sensing module is used to collect equipment status parameters such as vibration, temperature, and displacement; The graph construction module is used to build dynamic spatiotemporal relational graph models; The isotropic alignment module achieves spatiotemporal alignment and dynamic modeling of data based on a geometric isotropic neural network. The semantic fusion module is used to construct a semantic association network and to achieve data filtering and fusion. The 3D reconstruction and visualization module is used to generate dynamic 3D device state fields and provide an interactive analysis interface.

Images