An intelligent laying method and system for aircraft system pipes

Abstract

This invention discloses an intelligent piping laying method and system for aircraft systems, which acquires three-dimensional data and positioning information of aircraft system piping and generates optimal piping laying paths. The system detects piping positions in real time and overlays the optimal piping laying path onto the detected positions, while also projecting the best installation position for the piping clamps in real time. Based on advanced technologies such as depth vision, intelligent path planning, and augmented reality guidance, this invention achieves intelligent and automated laying of aircraft system piping. This invention demonstrates significant advantages in improving efficiency, enhancing accuracy, reducing costs, and improving user experience, providing a comprehensive solution for the intelligent upgrading of the aviation manufacturing industry, and possesses broad application prospects and significant practical value.

Description

Technical Field

[0001] This invention belongs to the technical field of aircraft system piping installation, specifically relating to an intelligent method and system for aircraft system piping installation. Background Technology

[0002] With the rapid development of the aviation manufacturing industry, modern aircraft structures are becoming increasingly complex. Among them, the aircraft system piping system, as the "blood vessels" and "nerves" of the aircraft, plays a crucial role in ensuring the normal operation of the aircraft. Traditional methods of laying aircraft system piping mainly rely on manual operation. This method is not only time-consuming and labor-intensive, but also prone to assembly errors, making it difficult to meet the high requirements of modern aviation manufacturing for assembly precision and quality.

[0003] In recent years, semi-automated pipeline laying systems have been gradually applied in the aerospace manufacturing industry. These systems typically use robotic arms in conjunction with simple vision systems for pipeline positioning and installation. However, this method has poor adaptability to complex environments and struggles to handle the confined and winding spaces within aircraft cabins. Furthermore, due to a lack of intelligent path planning capabilities, these systems often require manual path pre-setting and cannot dynamically adjust according to actual conditions, resulting in suboptimal pipeline layouts and potential interference risks. On the other hand, although 3D modeling technology is widely used in the aircraft design phase, translating virtual design into actual installation remains a significant challenge. While existing augmented reality technology can assist the assembly process to some extent, its guidance effect is limited due to a lack of real-time environmental perception and intelligent decision-making capabilities, making it difficult to adapt to complex and ever-changing actual assembly environments. In addition, existing pipeline laying systems generally suffer from low system integration and poor coordination between modules. For example, the path planning module and the actual installation guidance module are often disconnected, making real-time information interaction and feedback difficult. This leads to a disconnect between theoretical design and actual installation, affecting overall assembly quality.

[0004] For example, existing Chinese patent CN117763770A discloses a method for laying aero-engine pipelines, including the following steps: constructing a spatial mathematical model for laying aero-engine pipelines, generating a laying map based on the mathematical model; searching for an initial path using the A* algorithm based on the laying map, where the A* algorithm takes the starting point as the initial node and uses an evaluation function to evaluate candidate points around the node, with the candidate point with the smallest evaluation function value being the next node; and optimizing the initial path using the PSO optimization algorithm to make the curved parts of the pipeline corresponding to the path manufacturable.

[0005] For example, Chinese patent CN117113603A discloses a method and system for simulating the laying of external pipes for an aero-engine. The method includes the following steps: generating an engine casing model and a model of obstacles on the engine casing; determining the starting and ending points of the external pipes of the engine casing; generating a pipe laying map; calculating the pipe path using a pipe path planning algorithm; performing post-processing on the generated pipe path, the post-processing including: removing redundant points, removing redundant turns, stretching the pipe to make it machinable, and attaching the installation edge; generating spline curve nodes based on the path nodes obtained after post-processing; and generating the centerline and pipe based on the spline curve nodes. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent method and system for laying aircraft system pipelines, aiming to solve the problems of insufficient assembly accuracy, low efficiency, and high risk of interference in existing aircraft system pipeline laying.

[0007] This invention is mainly achieved through the following technical solutions:

[0008] A method for intelligently laying aircraft system piping includes the following steps:

[0009] Step S1: Obtain the three-dimensional data and positioning information of the aircraft system piping;

[0010] Step S11: Collect three-dimensional data of the aircraft system piping; simultaneously, acquire the spatial pose information of the airborne inertial sensors;

[0011] Step S12: Based on spatial pose information, calculate the three-dimensional data of the aircraft system piping in the airborne inertial sensor coordinate system;

[0012] Step S13: Convert the 3D data in the airborne inertial sensor coordinate system into point cloud position information in the world coordinate system;

[0013] Step S2: Generate the optimal pipeline laying path;

[0014] Step S21: Based on the point cloud location information, a continuous radiation field is generated using a radiation field generation algorithm to construct a three-dimensional radiation field model; the three-dimensional space is divided into multiple grids, and the probability density of each grid is calculated; based on the probability density, combined with the gray value G(x, i) of the point, the grid feature vector F(i) is calculated and input into the graph neural network.

[0015] Step S22: Based on the graph neural network, extract the grid features and calculate the probability of the pipeline endpoint;

[0016] Step S23: Use the deep deterministic gradient algorithm to perform path planning and generate the pipeline laying path;

[0017] Step S24: Based on pipeline interference constraints, obtain the optimal pipeline laying path;

[0018] Step S3: Real-time detection of pipeline location, superimposing the optimal pipeline laying path onto the real-time detected pipeline location, and real-time projection of the optimal installation position of the pipeline clamp.

[0019] To better realize the present invention, further, in step S13, a Kalman filter algorithm is used for coordinate transformation, wherein the state equation and observation equation are:

[0020] X k =F k X k-1 +B k U k +w k ;

[0021] Z k =H k X k +v k ;

[0022] Where: X k It is a state vector;

[0023] X k-1 This is the system's state vector at time k-1, containing the position, velocity, and attitude information measured by the inertial measurement unit;

[0024] F k This is the state transition matrix;

[0025] B k To control the input matrix;

[0026] U k For control vectors;

[0027] w k This is process noise;

[0028] Z k For observation vectors;

[0029] H k The observation matrix;

[0030] v k To observe noise.

[0031] To better implement this invention, in step S21, the radiation field generation algorithm is further as follows:

[0032]

[0033] Where F(x, y, z) represents the radiation field intensity at the spatial point (x, y, z);

[0034] (x i y i , z i () represents the coordinates of the i-th sampling point;

[0035] N is the number of sampling points;

[0036] w i The weight of the i-th sampling point;

[0037] φ() is a radial basis function;

[0038] Calculate the probability density for each grid cell:

[0039]

[0040] Where: x is a point in the point cloud.

[0041] c i Indicates the center of grid i.

[0042] c i Indicates the center of grid j.

[0043] σ is an adjustable parameter;

[0044] The grayscale value G(x, i) of the point is:

[0045] G(x,i)=I(x)·P(x|i);

[0046] The feature vector F(i) for each grid cell:

[0047]

[0048] To better implement the present invention, step S22 further includes the following steps:

[0049] Step A1: Calculate the relationship features between each node and its surrounding nodes, and generate node feature vectors;

[0050]

[0051] Where R(p, q) represents the relationship features between each node and its surrounding nodes;

[0052] ||pq||2 represents the Euclidean distance between nodes p and q;

[0053] and These represent the i-th neighboring nodes of nodes p and q, respectively.

[0054] K is the number of adjacent nodes to be considered;

[0055] G(p, q) is the Gaussian kernel function:

[0056]

[0057] Step A2: Calculate the weight relationships between nodes and generate edge feature vectors;

[0058]

[0059] Where: W(p, q) represents the weight relationship between node p and node q;

[0060] β is an adjustable parameter;

[0061] N(p) represents the neighborhood set of node p;

[0062] ||pv||2 is the Euclidean distance between node p and node v;

[0063] Step A3: Aggregate node features and edge features, and update the node representation;

[0064]

[0065] Among them: W (k) and b( k) These are learnable parameters;

[0066] Let u be the feature representation of node u at the k-th layer;

[0067] Let v be the feature representation of node v at the (k+1)th layer;

[0068] α vu Attention coefficient;

[0069] σ() is the activation function;

[0070] N(v) is the neighborhood set of node v;

[0071] Step A4: Calculate the probability that each node will become the end point of the pipeline.

[0072] To better realize the present invention, further, in step S22, the formula for the graph neural network is:

[0073]

[0074] in: This represents the feature representation of node v at the k-th layer.

[0075] σ() is the activation function;

[0076] W (k) and b( k) These are learnable parameters;

[0077] Let u be the feature representation of node u at the (k-1)th layer;

[0078] N(v) is the neighborhood set of node v;

[0079] AGGREGATE (k) For aggregate functions:

[0080]

[0081] Where: a() is the attention scoring function;

[0082] α vu Attention coefficient;

[0083] Let u be a node in the neighborhood, representing the grid position adjacent to the current node. The features of each neighboring node are weighted and aggregated according to attention weights to form a new feature representation of the node. This mechanism allows the network to adaptively aggregate neighboring information, capturing spatial structure more accurately;

[0084] Let v be the feature representation of node v at the (k-1)th layer;

[0085] Let w be the feature vector (or hidden state) of a neighboring node w in the (kl)th layer of the graph neural network.

[0086] To better realize the present invention, step S24 further includes the following steps:

[0087] Step B1: Discretize the pipe surface into a series of point sets {p i Then calculate the minimum distance from each point to the obstacle:

[0088]

[0089] Where o represents the set of obstacle points;

[0090] Step B2: Construct a continuous range field function:

[0091]

[0092] Among them, w i (x) is the weighting function:

[0093]

[0094] Step B2: If D(x) <d threshold If so, it is considered that there is a risk of collision;

[0095] Where: dthreshold This is the preset distance threshold.

[0096] To better realize the present invention, step S3 further includes the following steps:

[0097] Step S31: Use a laser sensor to detect the pipeline position in real time and generate pipeline position data;

[0098] Step S32: Overlay the optimal pipeline laying path onto the pipeline location data, and use an adaptive rendering algorithm to dynamically adjust the rendering parameters to generate an augmented reality scene;

[0099] The adaptive rendering algorithm is as follows:

[0100] I(x,y)=k a I a +k d (L·N)I d +k s (R·V) n I s ;

[0101] Where: I(x, y) represents the final rendering intensity of pixel (x, y);

[0102] k a k d and k s These are the ambient light, diffuse reflection, and specular reflection coefficients, respectively.

[0103] I a I d and I s These represent the intensities of ambient light, diffuse light, and specular light, respectively.

[0104] L is the direction vector of the light source;

[0105] N is the surface normal vector;

[0106] R is the direction vector of the reflected ray;

[0107] V is the line-of-sight vector;

[0108] n is the specular reflection index;

[0109] Step S33: Calculate the optimal installation position of clamps in the optimal pipeline laying path, where the number of clamps used is minimized while satisfying mechanical constraints;

[0110] Define a state transition function f(i) to represent the minimum number of clamps required to get from the starting point to the i-th point:

[0111]

[0112] Among them, D max This refers to the maximum permissible distance between the clamps;

[0113] Any point i(x) on the path D(i,j) i y i , z i ) and point j(x j y j , z j The distance between them;

[0114] Step S34: Using multiple contour line simulation technology, the optimal installation position of the pipeline clamp is projected in real time through AR glasses.

[0115] This invention is mainly achieved through the following technical solutions:

[0116] An intelligent pipeline laying system for aircraft systems, used to implement the above-mentioned intelligent pipeline laying method for aircraft systems, includes a depth vision subsystem, an intelligent path planning subsystem, and an augmented reality guidance subsystem connected sequentially from front to back;

[0117] The depth vision subsystem is used to acquire 3D data and positioning information of aircraft system pipelines, generate 3D models of the pipelines, and determine the point cloud position information in the world coordinate system based on the inertial measurement unit coordinate system.

[0118] The intelligent path planning subsystem is used to generate a three-dimensional radiation field model based on the three-dimensional model and point cloud location information; and to plan the optimal pipeline laying path in the three-dimensional radiation field model using graph neural networks and reinforcement learning algorithms.

[0119] The augmented reality guidance subsystem is used to detect pipeline locations in real time and overlay the optimal pipeline laying path onto the detected pipeline locations; the optimal installation position of pipeline clamps is projected in real time through augmented reality glasses.

[0120] To better realize the present invention, the depth vision subsystem further includes a lidar module, an inertial measurement unit module, a coordinate transformation module, and an optical encoder connected sequentially from front to back. The optical encoder is communicatively connected to the lidar module and the inertial measurement unit module, respectively.

[0121] The lidar module is used to acquire three-dimensional data and positioning information of the aircraft's system piping.

[0122] The inertial measurement unit module is used to acquire the spatial pose information of the airborne inertial sensor and calculate the position of the three-dimensional data of the aircraft system piping in the airborne inertial sensor coordinate system.

[0123] An optical encoder is used to acquire the scanning frequency of the lidar module and record the current timestamp of the inertial measurement unit module.

[0124] The coordinate transformation module is used to convert 3D data in the airborne inertial sensor coordinate system into point cloud position information in the world coordinate system.

[0125] To better realize the present invention, the intelligent path planning subsystem further includes a radiation field generation module, a graph neural network module, a reinforcement learning module, and a pipeline interference avoidance module connected sequentially from front to back;

[0126] The radiation field generation module is used to divide the three-dimensional space into multiple grids based on the three-dimensional model and point cloud location information; calculate the probability density of each grid and generate a three-dimensional radiation field model.

[0127] The graph neural network module is used to receive the three-dimensional radiation field model, extract the feature vector of each grid, and calculate the probability of each grid being the end point of the pipeline.

[0128] The reinforcement learning module receives probability information from the grid and selects action strategies based on the ε-greedy algorithm; through iterative optimization, it generates pipeline laying paths.

[0129] The pipeline interference avoidance module is used to receive the pipeline laying path and obtain the optimal pipeline laying path based on interference constraints.

[0130] To better realize the present invention, the augmented reality guidance subsystem further includes a laser sensor module, a simulation rendering module, a clamp position calculation module and an AR projection module connected in sequence from front to back;

[0131] The laser sensor module is used to detect the pipeline position in real time and generate pipeline position data;

[0132] The simulation rendering module is used to receive the optimal pipeline laying path and overlay the optimal pipeline laying path onto the pipeline location data to generate an augmented reality scene.

[0133] The clamp position calculation module is used to determine the clamp position in the optimal pipeline laying path based on a preset distance threshold and generate clamp position data.

[0134] The AR projection module is used to receive clamp position data and project the optimal installation position of the pipe clamp in real time through AR glasses.

[0135] The beneficial effects of this invention are as follows:

[0136] This invention, based on advanced technologies such as depth vision, intelligent path planning, and augmented reality guidance, enables the intelligent and automated installation of aircraft system piping. It demonstrates significant advantages in improving efficiency, enhancing accuracy, reducing costs, and improving user experience, providing a comprehensive solution for the intelligent upgrading of the aviation manufacturing industry. It has broad application prospects and significant practical value.

[0137] From an overall architectural perspective, this invention achieves seamless integration of three subsystems: depth vision, intelligent path planning, and augmented reality guidance, forming a closed-loop intelligent installation process. Specifically, the various modules of this invention achieve high synergy and complementarity. For example, the depth vision subsystem provides precise spatial information for path planning, and the results of path planning are fed back to the operator in real time through the AR system. This close information interaction ensures a high degree of consistency between virtual design and actual installation, effectively solving the "design-installation" disconnect problem existing in traditional methods. Specifically, this invention introduces an augmented reality guidance subsystem. By combining real-time environmental perception and high-precision AR projection technology, this system provides operators with intuitive and precise installation guidance. In particular, the use of multiple contour line simulation technology not only provides accurate spatial positioning information but also effectively solves the display problem of traditional AR systems in complex lighting environments. This greatly reduces human error and improves installation accuracy and efficiency. The highly integrated system design of this invention significantly improves the overall efficiency and accuracy of pipeline installation. Compared with traditional methods, this system can increase pipeline installation efficiency by approximately 300%, while controlling installation and assembly at the millimeter level, which is of great significance for precision assembly work in aircraft manufacturing.

[0138] This invention employs a fusion technology of high-precision lidar and inertial measurement unit (IMU), coupled with an innovative coordinate transformation algorithm, to achieve high-precision 3D reconstruction of the complex aircraft cabin environment. This not only provides reliable spatial information for subsequent path planning but also significantly improves the system's adaptability to complex environments. Particularly in the confined and winding space of an aircraft cabin, this system demonstrates excellent spatial perception capabilities, laying the foundation for precise pipeline layout. The improved probability density estimation method used in the 3D radiation field model construction significantly improves computational efficiency while maintaining accuracy. The dynamic programming algorithm used in the clamp position calculation ensures mechanical performance while minimizing material usage, reflecting the system's intelligence and economy.

[0139] This invention innovatively combines graph neural networks and deep reinforcement learning techniques to develop a highly efficient and flexible path planning algorithm. This algorithm can not only quickly generate optimal pipeline laying paths but also dynamically adjust the planning results in real time to adapt to environmental changes. Especially in complex scenarios involving parallel laying of multiple pipelines, the system demonstrates excellent interference avoidance capabilities, effectively reducing the risk of interference between pipelines and between pipelines and other structural components. Attached Figure Description

[0140] Figure 1 This is a schematic diagram of the intelligent pipeline laying system for aircraft systems according to the present invention.

[0141] Figure 2 This is a block diagram of the depth vision subsystem.

[0142] Figure 3 A schematic diagram of the intelligent path planning subsystem;

[0143] Figure 4 A schematic diagram of the principle of the augmented reality guidance subsystem. Detailed Implementation

[0144] Example 1:

[0145] An intelligent piping system for aircraft systems, such as Figures 1-4 As shown, it includes a depth vision subsystem, an intelligent path planning subsystem, and an augmented reality guidance subsystem connected from front to back, which are used to realize the intelligent and efficient laying of aircraft system pipelines.

[0146] The depth vision subsystem is used to acquire three-dimensional data and positioning information of the aircraft system pipelines, generate a three-dimensional model of the pipelines, and determine the position information of the point cloud data in the world coordinate system based on the inertial measurement unit coordinate system.

[0147] The intelligent path planning subsystem is used to receive the 3D model and point cloud position information sent by the depth vision subsystem; generate a high-precision 3D radiation field model based on the 3D model and point cloud position information; and plan the optimal pipeline laying path in the high-precision 3D radiation field model using graph neural networks and reinforcement learning algorithms.

[0148] The augmented reality guidance subsystem is used to receive the optimal pipeline laying path sent by the intelligent path planning subsystem; detect the pipeline position in real time through a laser sensor; overlay the optimal pipeline laying path onto the detected pipeline position in real time; and project the optimal installation position of the pipeline clamps in real time through augmented reality glasses.

[0149] Preferably, the depth vision subsystem includes a LiDAR module, an inertial measurement unit (IMU) module, an optical encoder, and a coordinate transformation module. The IMU module is communicatively connected to the LiDAR module, and the optical encoder is communicatively connected to both the LiDAR module and the IMU module; the coordinate transformation module is communicatively connected to the IMU module.

[0150] The lidar module is used to acquire three-dimensional data and positioning information of the aircraft's system piping.

[0151] An inertial measurement unit module is used to acquire spatial pose information from airborne inertial sensors; based on the spatial pose information, it calculates the position of the three-dimensional data of the aircraft system pipeline measured by lidar in the airborne inertial sensor coordinate system;

[0152] An optical encoder is used to acquire the scanning frequency of the lidar module and record the current timestamp of the inertial measurement unit module.

[0153] The coordinate transformation module is used to receive the three-dimensional data in the airborne inertial sensor coordinate system sent by the inertial measurement unit module; and to convert the three-dimensional data in the airborne inertial sensor coordinate system into point cloud position information in the world coordinate system.

[0154] Specifically, the lidar module uses a high-precision phase-modulated continuous wave (FMCW) lidar with ranging accuracy down to the millimeter level and a scanning frequency of up to 20Hz, which enables the system to quickly and accurately acquire three-dimensional data of the aircraft system piping.

[0155] The inertial measurement unit (IMU) module employs microelectromechanical systems (MEMS) technology, integrating a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer, providing 6 degrees of freedom attitude information. The module's sampling rate of up to 1000Hz ensures the real-time performance and accuracy of spatial pose information.

[0156] The photoelectric encoder has a resolution of 0.1°, enabling it to accurately capture changes in the scanning frequency of the lidar module. This high-precision time synchronization mechanism is crucial for subsequent data fusion.

[0157] Preferably, the intelligent path planning subsystem includes a radiation field generation module, a graph neural network module, a reinforcement learning module, and a pipeline interference avoidance module; the graph neural network module is communicatively connected to the radiation field generation module, the reinforcement learning module is communicatively connected to the graph neural network module, and the pipeline interference avoidance module is communicatively connected to the reinforcement learning module.

[0158] The radiation field generation module is used to receive the 3D model and point cloud position information sent by the depth vision subsystem; divide the 3D space into multiple grids; calculate the probability density of each grid; and generate a high-precision 3D radiation field model based on the probability density.

[0159] The graph neural network module is used to receive a high-precision three-dimensional radiation field model sent by the radiation field generation module; extract the feature vector of each grid; and calculate the probability of each grid being the end point of the pipeline based on the feature vector.

[0160] The reinforcement learning module is used to receive grid probability information sent by the graph neural network module; select action strategies based on the ε-greedy algorithm; and generate the optimal pipeline laying path through iterative optimization.

[0161] The pipeline interference avoidance module is used to receive the optimal pipeline laying path sent by the reinforcement learning module; define pipeline interference constraints; and optimize the optimal pipeline laying path based on the interference constraints.

[0162] Specifically, the radiation field generation module employs an improved Fast Fourier Transform (FFT) algorithm, dividing the space into a 64x64x64 grid, with each grid having a side length of 5cm. This grid division method significantly improves computational efficiency while maintaining accuracy.

[0163] The graph neural network module employs a 5-layer graph convolutional network structure, with 64, 128, 256, 128, and 64 neurons in each layer, respectively. During training, the Adam optimizer was used with a learning rate of 0.001 and a batch size of 32. Extensive experiments demonstrated that these hyperparameter settings achieved optimal performance in aircraft system piping laying tasks.

[0164] Preferably, the augmented reality guidance subsystem includes a laser sensor module, a simulation rendering module, a clamp position calculation module, and an AR projection module. The simulation rendering module is communicatively connected to the laser sensor module, the clamp position calculation module is communicatively connected to the simulation rendering module, and the AR projection module is communicatively connected to both the simulation rendering module and the clamp position calculation module. These modules work together to provide intuitive and accurate visual guidance for the actual installation process of aircraft system piping.

[0165] The laser sensor module is used to detect the pipeline position in real time and generate pipeline position data.

[0166] The simulation rendering module is used to receive the optimal pipeline laying path sent by the intelligent path planning subsystem; overlay the optimal pipeline laying path onto the pipeline location data; and generate an augmented reality scene.

[0167] The clamp position calculation module is used to determine the clamp position in the optimal pipeline laying path based on a preset distance threshold; and generate clamp position data.

[0168] An AR projection module is used to receive the augmented reality scene and the clamp position data; and to project the optimal installation position of the pipe clamp in real time through AR glasses.

[0169] Specifically, the laser sensor module employs high-precision structured light laser scanning technology, enabling it to capture the spatial position and orientation of the pipeline in real time. The module boasts a scanning frequency of up to 30Hz and a measurement accuracy better than 0.1mm. This high spatiotemporal resolution allows the system to respond promptly to minute changes in the pipeline's position, providing a reliable data foundation for subsequent real-time rendering and guidance.

[0170] The simulation rendering module receives the optimal pipeline laying path from the intelligent path planning subsystem and overlays it onto the pipeline position data detected in real time by the laser sensor module. This module employs an innovative adaptive rendering algorithm that can dynamically adjust rendering parameters based on ambient lighting conditions and pipeline material characteristics.

[0171] The AR projection module is the key interface for interaction between the system and the operator. This module employs multiple contour lines to simulate the clamp's installation location visually in the AR glasses. Compared to traditional single-point marking, this contour-based AR projection method offers the following advantages: First, it provides continuous spatial information, allowing operators to more intuitively understand the clamp's installation location; second, the multi-layered contour lines provide depth information, helping operators accurately determine the spatial position of the installation point; finally, this method has relatively lower requirements for the AR glasses' positioning accuracy and exhibits better robustness.

[0172] The augmented reality guidance subsystem of this invention provides operators with precise and intuitive installation guidance. In practical applications, this intelligent guidance system can improve the efficiency of pipeline installation by more than 50%, while controlling the installation error to within 1mm, significantly improving the quality and efficiency of aircraft system pipeline laying.

[0173] Through the coordinated operation of the aforementioned modules, this invention can quickly and accurately complete pipeline laying tasks, significantly improving work efficiency and installation quality. Compared with traditional methods, this system increases laying speed by approximately 300% and installation accuracy by approximately 50%, greatly reducing assembly errors and workload associated with manual operation. This invention can quickly and accurately plan the optimal pipeline laying path in the complex environment of an aircraft cabin, while effectively avoiding interference between pipelines. This not only greatly improves the efficiency of pipeline laying but also significantly reduces the risk of human error, providing an intelligent and automated solution for pipeline laying in aircraft manufacturing processes.

[0174] Example 2:

[0175] A smart method for laying aircraft system piping, such as Figure 1 As shown, it includes the following steps:

[0176] 1. First, based on the depth vision subsystem, acquire the three-dimensional data and positioning information of the aircraft system pipeline.

[0177] Preferably, such as Figure 2 As shown, a lidar module scans the environment at a frequency of 20Hz, while an inertial measurement unit (IMU) module records spatial pose information at a sampling rate of 1000Hz. A photoelectric encoder is used to accurately capture the lidar's scanning frequency, achieving high-precision time synchronization. Then, a coordinate transformation module, based on a Kalman filter algorithm, converts the data in the airborne inertial sensor coordinate system into point cloud position information in the world coordinate system, with the transformation error controlled within 1cm.

[0178] Preferably, the coordinate transformation module employs an improved Kalman filter algorithm, whose state equation and observation equation are as follows:

[0179] X k =F k X k-1 +B k U k +w k ;

[0180] Z k =H k X k +v k ;

[0181] Among them, X k Let F be the state vector. k Let B be the state transition matrix. k To control the input matrix, U k w is the control vector k For process noise, Z k H is the observation vector.k Let v be the observation matrix. k To observe the noise, the noise covariance matrix Q is adjusted. k and R k This invention achieves high-precision conversion from the airborne inertial sensor coordinate system to the world coordinate system, with the conversion error controlled within 1cm.

[0182] 2. Then, based on the intelligent path planning subsystem, the optimal pipeline laying path is generated.

[0183] (1) First, based on the radiation field generation module, a high-precision three-dimensional radiation field model is constructed, and the space is divided into a 64x64x64 grid.

[0184] Preferably, the present invention employs an improved radiation field generation algorithm, the mathematical expression of which is as follows:

[0185]

[0186] Where F(x, y, z) represents the radiation field intensity at the spatial point (x, y, z);

[0187] (x i y i , z i () represents the coordinates of the i-th sampling point;

[0188] N is the number of sampling points;

[0189] w i The weight of the i-th sampling point;

[0190] φ() is a radial basis function.

[0191] The radiation field generation algorithm plays a crucial role in this invention. Its core function is to construct a high-precision three-dimensional spatial representation, serving as the foundation for subsequent intelligent path planning. There is a clear logical relationship between this algorithm and the probability density of the three-dimensional model and the network, as detailed below:

[0192] First, point cloud data acquired from the depth vision subsystem provides initial 3D spatial information, which is then converted into positional information in the world coordinate system. Based on these spatial points, a radiation field generation algorithm constructs a continuous radiation field intensity F(x, y, z); this radiation field intensity is essentially a weighted representation of the spatial point distribution, where each sampling point contributes its own radiation influence, which is spatially diffused through a radial basis function (RBF). The radial basis function uses a Gaussian kernel:

[0193]

[0194] Where r represents the straight-line distance between the target point and the sampling point; β is an adjustable parameter, typically ranging from 0.1 to 10. In one embodiment of the present invention, β is set to 1.5, a value derived through extensive experiments that achieves a good balance between computational efficiency and accuracy.

[0195] Next, the radiation field algorithm discretizes the three-dimensional space into a regular grid (64×64×64) and calculates the probability density of each point belonging to each grid. Preferably, the radiation field generation module employs an improved probability density estimation method to generate a high-precision three-dimensional radiation field model. The core idea of ​​this method is to divide the three-dimensional space into a grid and then calculate the probability density of each point belonging to each grid. Specifically, the probability density function of each point belonging to a certain grid is defined as follows:

[0196]

[0197] Where x represents a point in the point cloud, c i The center of grid i is represented by σ, which is an adjustable parameter. In embodiments of the present invention, the value of σ is typically between 0.1 and 1, and the specific value can be adjusted according to the density and distribution characteristics of the point cloud. For example, when the point cloud is relatively sparse, a larger σ value (such as 0.8) can be selected to increase the smoothing effect; while when the point cloud is dense and rich in detail, a smaller σ value (such as 0.2) can be selected to retain more detailed information.

[0198] This probability density function establishes a soft correspondence between point cloud data and discrete grids, avoiding information loss that might result from hard assignment. Subsequently, by combining the grayscale values ​​(or pixel values) of the points, the feature vector for each grid is calculated:

[0199]

[0200] G(x,i)=I(x)·P(x|i);

[0201] Where: G(x, i) is the gray value of each point in the point cloud belonging to grid i;

[0202] I(x) represents the pixel value of point x in the image. This method combines the color information of a point with its spatial distribution, enabling it to better capture the geometric and visual features of the environment;

[0203] F(i) represents the feature vector corresponding to grid i. This feature extraction method not only considers the spatial distribution of points but also incorporates color information, enabling a more comprehensive description of environmental features and providing more reliable input for subsequent path planning.

[0204] This design incorporates both spatial geometric information and visual features into the radiation field, providing rich input features for the graph neural network. The grid feature vectors of the radiation field model become the initial features of the nodes in the graph neural network, thus establishing a bridge between the radiation field and the graph neural network. This transformation process from continuous space to discrete grids, and then to graph structures, enables the efficient processing and optimization of complex 3D environments.

[0205] (2) Then, based on the graph neural network module, the grid features are extracted and the endpoint probability is calculated. Specifically, a 5-layer graph convolutional network structure is adopted, and the number of neurons in each layer is 64, 128, 256, 128 and 64 respectively.

[0206] Preferably, graph neural networks and reinforcement learning algorithms are also used to plan the optimal pipeline laying path in a high-precision three-dimensional radiation field model. This invention employs an improved graph neural network structure, the core formula of which is as follows:

[0207]

[0208] in, This represents the feature representation of node v at the k-th layer.

[0209] σ() is the activation function;

[0210] W (k) and b (k) These are learnable parameters;

[0211] Let be the feature representation of node u at layer (k-1), where "node" corresponds to a grid cell in three-dimensional space. In the context of pipeline planning, each node represents a location in space, and its feature vector encodes the geometric characteristics of that location, information about the surrounding environment, and its suitability as part of the pipeline path.

[0212] This is the neighborhood set of node v, that is, all nodes directly connected to node v. In a 3D mesh, this usually refers to adjacent mesh cells, corresponding to the possible local movement directions of the pipeline.

[0213] AGGREGATE (k) The aggregation function is the attention mechanism. In this invention, we chose the attention mechanism as the aggregation function to better capture the relationships between nodes.

[0214]

[0215] Where a() is the attention scoring function, and this invention uses a dot product attention mechanism.

[0216] (3) Based on the reinforcement learning module, the Deep Deterministic Policy Gradient (DDPG) algorithm is used for path planning. The Actor network and Critic network both adopt a three-layer fully connected structure with 256 and 128 hidden layer neurons, respectively.

[0217] Preferably, the reinforcement learning module uses the Double Delay Deep Q-Network (DQN) algorithm, and its Q-value update formula is as follows:

[0218] Q(s t a t )←Q(s t a t )+α[r t +γQ(s t+1 argmax a Q(s t+1 ,a))-Q(s t a t )];

[0219] Among them, s t Let 'a' represent the current state, 'a' represent the learning rate, 'γ' represent the discount factor, and 'r' represent the immediate reward. In this invention, 'a' is set to 0.01, and 'γ' is set to 0.99. These parameter values ​​have been repeatedly optimized to achieve rapid convergence and stable performance in aircraft system piping laying tasks.

[0220] (4) Finally, based on the pipeline interference avoidance module, a collision detection method based on the distance field is applied to ensure that the generated path will not interfere with other structures.

[0221] Preferably, the pipeline interference avoidance module employs an improved collision detection algorithm. Its core idea is to simplify the pipeline into a series of spheres and then use the distance between the spheres to determine whether interference exists. Specifically, for two pipeline segments i and j, we calculate the minimum distance between them:

[0222]

[0223] Among them, S i and S j Let d represent the sets of spheres for pipe segments i and j, respectively. ij If the value is less than a preset threshold (set to 5 mm in this invention), interference is considered to exist, and path adjustment is required.

[0224] Preferably, in the implementation of this invention, the pipeline interference avoidance module employs a collision detection method based on a distance field. First, the pipeline surface is discretized into a series of point sets {p}. i Then calculate the minimum distance from each point to the obstacle:

[0225]

[0226] Here, o represents the set of obstacle points. Based on these distance values, a continuous distance field function can be constructed:

[0227]

[0228] Among them, w i (x) is the weight function, usually using inverse distance weighting:

[0229]

[0230] Using this distance field function, it is possible to quickly detect whether the pipeline has collided with an obstacle.

[0231] If D(x) < d threshold (In this invention, d) threshold If the value is set to 5mm, then a collision risk is considered to exist, and the path needs to be adjusted.

[0232] Compared to traditional geometric collision detection algorithms, this distance field-based method has the advantages of high computational efficiency and strong adaptability to complex shapes, making it particularly suitable for interference detection tasks in complex environments such as aircraft system piping.

[0233] 3. Finally, the augmented reality guidance subsystem guides the pipeline laying operation.

[0234] Preferably, a laser sensor module is used to detect the pipeline position in real time at a frequency of 30Hz, with a measurement accuracy better than 0.1mm. A simulation rendering module overlays the optimal pipeline laying path onto the real-time detected pipeline position, and an adaptive rendering algorithm dynamically adjusts the rendering parameters. A clamp position calculation module uses a dynamic programming algorithm to determine the optimal clamp installation position; for 10mm diameter aluminum alloy pipelines, the maximum spacing between adjacent clamps is set to 500mm. Based on an AR projection module, multiple contour line simulation technology is used, typically selecting 3 to 5 contour lines with corresponding distances of 10mm, 20mm, 30mm, etc., to project the optimal installation position of the pipeline clamps in real time through AR glasses.

[0235] Preferably, specifically, the core formula of the rendering algorithm is as follows:

[0236] I(x,y)=k a I a +k d (L·N)I d +k s (R·V) n I s ;

[0237] Where I(x, y) represents the final rendering intensity of pixel (x, y), and k a k d and k s These are the ambient light, diffuse reflection, and specular reflection coefficients, respectively, I. a I d and I s , respectively, represent the intensity of ambient light, diffuse light, and specular light; L is the light source direction vector; N is the surface normal vector; R is the reflected ray direction vector; V is the line-of-sight direction vector; and n is the specular reflection index.

[0238] In practical applications of this invention, these parameters are dynamically adjusted based on real-time collected environmental data. For example, in dimly lit environments, the system appropriately increases the k_a value to improve overall brightness; while on highly reflective pipe surfaces, the system increases the k_s value to enhance specular reflection. This adaptive rendering technology greatly improves the visual quality and recognizability of augmented reality images, enabling operators to clearly identify pipe laying paths in various complex environments.

[0239] Preferably, the clamp position calculation module is a key innovation of this invention. This module uses a dynamic programming algorithm to determine the optimal installation position of the clamps based on the optimal pipeline laying path. The core idea of ​​the algorithm is to minimize the number of clamps used while satisfying mechanical constraints. Specifically, the distance D(i, j) between any two points i and j on the path is calculated as follows:

[0240]

[0241] Then, define a state transition function f(i) to represent the minimum number of clamps required to get from the starting point to the i-th point:

[0242]

[0243] Among them, D max This is the maximum permissible distance between clamps, a value typically determined based on factors such as the pipe's material, diameter, and the pressure it withstands. In one embodiment of the invention, for an aluminum alloy pipe with a diameter of 10mm, D... max The setting is 500mm. This value was derived through extensive experiments and finite element analysis, ensuring pipeline stability while minimizing the number of clamps required.

[0244] By solving the state transition equations using dynamic programming, the optimal distribution of clamps along the entire path can be obtained. This method not only ensures the mechanical stability of the pipeline installation but also minimizes material usage and installation workload, resulting in significant economic benefits.

[0245] Preferably, the process of generating contour lines can be represented as follows:

[0246] L(k)={(x,y,z)|f(x,y,z)=k};

[0247] Here, f(x, y, z) represents the distance function to the center of the clamp, and k represents different distance values. In practical applications, 3 to 5 contour lines are typically selected, corresponding to k values ​​of 10mm, 20mm, 30mm, etc. This multi-layered visualization method provides operators with precise spatial positioning information, greatly improving the accuracy of clamp installation.

[0248] The process of projecting contour lines onto AR glasses can be represented as:

[0249] P(L(k))=T·L(k);

[0250] Here, T is the projection transformation matrix, which comprehensively considers the optical characteristics of the AR glasses, the operator's head posture, and the geometry of the environment. In this invention, the T matrix is ​​continuously updated through a real-time calibration algorithm to ensure that the projected image is always precisely aligned with the actual environment.

[0251] Preferably, the graph neural network module includes:

[0252] The node feature extraction unit is used to calculate the relationship features between each node and its surrounding nodes; and generate node feature vectors.

[0253] An edge feature extraction unit, communicatively connected to the node feature extraction unit, is used to calculate the weight relationships between nodes and generate edge feature vectors.

[0254] The graph convolutional layer, communicatively connected to the node feature extraction unit and the edge feature extraction unit, is used to aggregate node features and edge features; and update node representations.

[0255] The output layer, which is communicatively connected to the graph convolutional layer, is used to calculate the probability that each node becomes the end point of the pipeline.

[0256] Preferably, the graph neural network module employs an innovative node feature extraction method. This method not only considers the features of the node itself but also incorporates the relational information of surrounding nodes. Specifically, for each node p, the relational feature R(p, q) between it and its surrounding nodes q is calculated:

[0257]

[0258] Where ||pq||2 represents the Euclidean distance between nodes p and q. and Let p and q represent the i-th neighboring nodes of nodes p and q respectively, K be the number of neighboring nodes considered (in this invention, K is usually between 5 and 10), and G(p, q) be the Gaussian kernel function.

[0259]

[0260] Here, σ is an adjustable parameter, typically ranging from 0.1 to 1. This design of relational features can capture local structural information between nodes, which helps improve the expressive power of graph neural networks.

[0261] In terms of edge feature extraction, this invention calculates the weight relationships between nodes:

[0262]

[0263] Where β is an adjustable parameter, and N(p) represents the neighborhood set of node p. In practical applications, the value of β is usually between 0.5 and 2. A larger β value will make the weight distribution more concentrated, while a smaller β value will make the weight distribution more uniform. The graph convolutional layer of this invention uses an attention mechanism to aggregate node features and edge features:

[0264]

[0265] in, Let W represent the feature representation of node v at layer k, where σ() is the activation function (ReLU function is used in this invention). (k) and b (k) α is a learnable parameter. vu Attention coefficient:

[0266]

[0267] Here, 'a' is a learnable attention vector, and || denotes the vector concatenation operation. This attention mechanism can adaptively adjust the importance of different neighboring nodes, thereby improving the model's expressive power and generalization performance.

[0268] Preferably, the reinforcement learning module employs the Deep Deterministic Policy Gradient (DDPG) algorithm. This algorithm comprises four main components: an Actor network, a Critic network, an experience replay buffer, and a target network.

[0269] The Actor network is responsible for generating actions based on the current state. Its structure is a three-layer fully connected neural network with 256 and 128 neurons in the hidden layers, respectively. The output layer uses the tanh activation function to restrict actions to the range [-1, 1]. The Actor network is updated via policy gradients.

[0270]

[0271] Where, θ μ Let Q(s, a|θ) represent the parameters of the Actor network. Q ) represents the output of the Critic network.

[0272] The Critic network is used to evaluate the action-value function. Its structure is also a three-layer fully connected neural network with 256 and 128 neurons in the hidden layers, respectively. The Critic network is updated by minimizing the temporal difference error.

[0273]

[0274] Among them, y i =r i +γQ′(s i+1 ,μ′(s i+1 |θ μ′ )|θ Q′ ) represents the target Q value, and Q' and μ' represent the target Critic network and the target Actor network, respectively.

[0275] The experience replay buffer is used to store state transition samples (s) t a t r t s t+1 The buffer size is set to 106. During training, a batch (batch size set to 64) is randomly sampled from the buffer each time for learning. This method can break the correlation between samples and improve the stability of learning.

[0276] The target network is used for the stable learning process, and its parameters θ' are gradually approached to the parameters θ of the online network through soft updates.

[0277] θ'←τθ+(1-τ)θ';

[0278] Wherein, τ is the soft update coefficient, which is set to 0.001 in this invention. This soft update mechanism can effectively reduce the fluctuation of the target value and improve the stability of training.

[0279] This invention enables intelligent and efficient laying of aircraft system pipelines. Compared with traditional manual laying methods, this invention not only improves installation efficiency by more than 3 times, but also controls installation errors to the millimeter level, providing a comprehensive intelligent solution for pipeline laying in the aircraft manufacturing process.

[0280] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications or equivalent changes made to the above embodiments based on the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for intelligent laying of aircraft system pipelines, characterized in that, Includes the following steps: Step S1: Obtain the three-dimensional data and positioning information of the aircraft system piping; Step S11: Collect three-dimensional data of the aircraft system piping, and at the same time, acquire the spatial pose information of the airborne inertial sensors; Step S12: Based on spatial pose information, calculate the three-dimensional data of the aircraft system piping in the airborne inertial sensor coordinate system; Step S13: Convert the 3D data in the airborne inertial sensor coordinate system into point cloud position information in the world coordinate system; Step S2: Generate the optimal pipeline laying path; Step S21: Based on the point cloud location information, a continuous radiation field is generated using a radiation field generation algorithm to construct a three-dimensional radiation field model; the three-dimensional space is divided into multiple grids, and the probability density of each grid is calculated; Based on probability density, the gray value of the junction point G ( x , i ), calculate the grid feature vector F( i ), to input graph neural networks; Step S22: Based on the graph neural network, extract the grid features and calculate the probability of the pipeline endpoint; Step S23: Use the deep deterministic gradient algorithm to perform path planning and generate the pipeline laying path; Step S24: Based on pipeline interference constraints, obtain the optimal pipeline laying path; Step S3: Real-time detection of pipeline location, superimposing the optimal pipeline laying path onto the real-time detected pipeline location, and real-time projection of the optimal installation position of the pipeline clamps; Step S31: Use a laser sensor to detect the pipeline position in real time and generate pipeline position data; Step S32: Overlay the optimal pipeline laying path onto the pipeline location data, and use an adaptive rendering algorithm to dynamically adjust the rendering parameters to generate an augmented reality scene; The adaptive rendering algorithm is as follows: ； in: I (x, y) represents the final rendering intensity of pixel (x, y); k a , k d and k s These are the ambient light, diffuse reflection, and specular reflection coefficients, respectively. I a , I d and I s These represent the intensities of ambient light, diffuse light, and specular light, respectively. L The direction vector of the light source; N It is the surface normal vector; R The direction vector of the reflected ray; V The line-of-sight vector; n The specular reflectance index; Step S33: Calculate the optimal installation position of clamps in the optimal pipeline laying path, where the number of clamps used is minimized while satisfying mechanical constraints; Define a state transition function f ( i ) indicates from the starting point to the th i Minimum number of clamps required per point: ； ； Among them, D max This refers to the maximum permissible distance between the clamps; D(i, j) represents any point i (x, j) on the path. i y i , z i ) and point j (x j y j , z j The distance between them; Step S34: Using multiple contour line simulation technology, the optimal installation position of the pipeline clamp is projected in real time through AR glasses.

2. The intelligent laying method for aircraft system piping according to claim 1, characterized in that, In step S13, the Kalman filter algorithm is used for coordinate transformation, where the state equation and observation equation are: ； ； in: X k It is a state vector; X k-1 For the system in k The state vector at time -1; F k This is the state transition matrix; B k To control the input matrix; U k For control vectors; w k This is process noise; Z k For observation vectors; H k The observation matrix; v k To observe noise.

3. The intelligent laying method for aircraft system piping according to claim 1, characterized in that, In step S21, the radiation field generation algorithm is as follows: ； Where F(x, y, z) represents the radiation field intensity at the spatial point (x, y, z); (x i y i , z i ) is the first i The coordinates of each sampling point; N This represents the number of sampling points; w i For the first i The weight of each sampling point; ( ) represents the radial basis functions; Calculate the probability density for each grid cell: ； in: x For points in a point cloud, c i Represents a grid i The center, c i Represents a grid j The center, σ is an adjustable parameter; grayscale value of a dot G ( x , i )for: ； in: I ( x ) represents a point x Pixel values ​​in an image; The feature vector F of each grid i ): 。 4. The intelligent laying method for aircraft system piping according to claim 1, characterized in that, Step S22 includes the following steps: Step A1: Calculate the relationship features between each node and its surrounding nodes, and generate node feature vectors; ； Where R(p, q) represents the relationship features between each node and its surrounding nodes; This represents the Euclidean distance between nodes p and q; and Let p and q represent the first and second nodes respectively. i 1 adjacent node; K The number of adjacent nodes to be considered; G( p , q ) is the Gaussian kernel function: ； Step A2: Calculate the weight relationships between nodes and generate edge feature vectors; ； Among them: W( p , q Let ) represent the weight relationship between nodes p and q; β These are adjustable parameters; N ( p ) represents a node p The neighborhood set; For nodes p and nodes v The Euclidean distance; Step A3: Aggregate node features and edge features, and update the node representation; ； Among them: W (k) and b (k) These are learnable parameters; For nodes u In the k Layer feature representation; For nodes v In the k+ Feature representation at level 1; α vu Attention coefficient; σ() is the activation function; N ( v ) is a node v The neighborhood set; Step A4: Calculate the probability that each node will become the end point of the pipeline.

5. A method for intelligent laying of aircraft system pipelines according to claim 1 or 4, characterized in that, In step S22, the formula for the graph neural network is: ； in: Represents a node v In the k Layer feature representation; σ() is the activation function; W (k) and b (k) These are learnable parameters; For nodes u In the k- Feature representation at level 1; N ( v ) is a node v The neighborhood set; AGGREGATE (k) For aggregate functions: ； ； in: a ( ) represents the attention scoring function; α vu Attention coefficient; u For nodes within the neighborhood; For nodes v In the k- Feature representation at level 1; For the first digit of the graph neural network k In the -l layer, neighboring nodes w eigenvectors.

6. The intelligent laying method for aircraft system piping according to claim 1, characterized in that, Step S24 includes the following steps: Step B1: Discretize the pipe surface into a series of point sets { p i Then calculate the minimum distance from each point to the obstacle: ； in, o Represents the set of obstacle points; Step B2: Construct a continuous range field function: ； in, w i ( x ) is the weighting function: ； Step B2: If If so, it is considered that there is a risk of collision; in: d threshold This is the preset distance threshold.

7. An intelligent piping system for aircraft systems, used to implement the intelligent piping method for aircraft systems as described in any one of claims 1-6, characterized in that, It includes a depth vision subsystem, an intelligent path planning subsystem, and an augmented reality guidance subsystem, which are connected sequentially from front to back; The depth vision subsystem is used to acquire 3D data and positioning information of aircraft system pipelines and generate 3D models of the pipelines. Based on the coordinate system of the inertial measurement unit, determine the point cloud position information in the world coordinate system; The intelligent path planning subsystem is used to generate a three-dimensional radiation field model based on the three-dimensional model and point cloud location information; and to plan the optimal pipeline laying path in the three-dimensional radiation field model using graph neural networks and reinforcement learning algorithms. The augmented reality guidance subsystem is used to detect pipeline locations in real time and overlay the optimal pipeline laying path onto the detected pipeline locations; the optimal installation position of pipeline clamps is projected in real time through augmented reality glasses.

8. The intelligent piping laying system for aircraft systems according to claim 7, characterized in that, The depth vision subsystem includes a lidar module, an inertial measurement unit module, a coordinate transformation module, and an optical encoder connected sequentially from front to back. The optical encoder is communicatively connected to the lidar module and the inertial measurement unit module, respectively. The lidar module is used to acquire three-dimensional data and positioning information of the aircraft's system piping. The inertial measurement unit module is used to acquire the spatial pose information of the airborne inertial sensor and calculate the position of the three-dimensional data of the aircraft system piping in the airborne inertial sensor coordinate system. An optical encoder is used to acquire the scanning frequency of the lidar module and record the current timestamp of the inertial measurement unit module. The coordinate transformation module is used to convert 3D data in the airborne inertial sensor coordinate system into point cloud position information in the world coordinate system.

9. The intelligent piping laying system for aircraft systems according to claim 7, characterized in that, The intelligent path planning subsystem includes a radiation field generation module, a graph neural network module, a reinforcement learning module, and a pipeline interference avoidance module, which are connected sequentially from front to back. The radiation field generation module is used to divide the three-dimensional space into multiple grids based on the three-dimensional model and point cloud location information; calculate the probability density of each grid and generate a three-dimensional radiation field model. The graph neural network module is used to receive the three-dimensional radiation field model, extract the feature vector of each grid, and calculate the probability of each grid being the end point of the pipeline. The reinforcement learning module receives probability information from the grid and selects action strategies based on the ε-greedy algorithm; through iterative optimization, it generates pipeline laying paths. The pipeline interference avoidance module is used to receive the pipeline laying path and obtain the optimal pipeline laying path based on interference constraints.

10. The intelligent piping laying system for aircraft systems according to claim 7, characterized in that, The augmented reality guidance subsystem includes a laser sensor module, a simulation rendering module, a clamp position calculation module, and an AR projection module connected from front to back. The laser sensor module is used to detect the pipeline position in real time and generate pipeline position data; The simulation rendering module is used to receive the optimal pipeline laying path and overlay the optimal pipeline laying path onto the pipeline location data to generate an augmented reality scene. The clamp position calculation module is used to determine the clamp position in the optimal pipeline laying path based on a preset distance threshold and generate clamp position data. The AR projection module is used to receive clamp position data and project the optimal installation position of the pipe clamp in real time through AR glasses.

Images