Intelligent optimization method for arrangement of host island equipment based on three-dimensional modeling

Abstract

The invention provides a host island equipment arrangement intelligent optimization method based on three-dimensional modeling, which relates to the technical field of three-dimensional modeling, and comprises the steps of extracting global feature data of equipment through feature recognition and semantic segmentation, constructing a multi-physics field coupled equipment relation graph, determining an equipment arrangement sequence by using a graph attention network, and optimizing the equipment arrangement sequence. An octree structure is adopted to discretize the arrangement space, reinforcement learning is applied to optimize the position and posture of the equipment, and finally three-dimensional dynamics simulation verification is carried out. Intelligent arrangement of host island equipment is achieved, the space utilization rate and maintenance convenience are improved, and thermodynamic interference between the equipment is reduced.

Description

Technical Field

[0001] The present invention relates to three-dimensional modeling technology, and particularly to an intelligent optimization method for the layout of main engine island equipment based on three-dimensional modeling. Background Art

[0002] Currently, the main engine island equipment layout technology mainly includes rule-based layout methods, optimization algorithm-based layout methods, and artificial intelligence-assisted layout methods, etc. The existing main engine island equipment layout technology has the following defects and deficiencies: It is difficult for the existing technology to comprehensively consider the multi-physical field coupling relationship between equipment, especially the mutual influence in aspects such as hydrodynamics and thermodynamics, resulting in the layout plan may have low operating efficiency or safety hazards. The physical relationships such as thermodynamics and fluid mechanics between equipment are complex and mutually influential, while traditional methods often only consider geometric constraints and simple physical connection relationships, ignoring the deep physical coupling effects.

[0003] The expression and optimization method of the equipment layout space are not fine and flexible enough. The existing technology usually uses uniform grid division or simplified geometric expression methods, which are difficult to adapt to the accurate expression of equipment of different scales and the efficient utilization of complex spaces. Especially for equipment and spaces with irregular shapes, the limitations of this expression method are more obvious.

[0004] There is a lack of a dynamic evaluation mechanism for the requirements of equipment maintenance operations. Most of the existing layout methods only consider static spacing requirements and are difficult to simulate and evaluate the dynamic operation space requirements during equipment maintenance, such as the disassembly path of large equipment and the working space of maintenance personnel, etc. This leads to difficulties in later equipment maintenance and increases the operation cost and safety risks. Summary of the Invention

[0005] The embodiments of the present invention provide an intelligent optimization method for the layout of main engine island equipment based on three-dimensional modeling, which can solve the problems in the existing technology.

[0006] In the first aspect of the embodiments of the present invention, an intelligent optimization method for the layout of main engine island equipment based on three-dimensional modeling is provided, including: Establish a three-dimensional model of the main engine island, and apply feature recognition and semantic segmentation to extract the global feature data of the equipment. The global feature data includes the geometric features, physical features of the equipment, and the constraint relationships between the equipment; Construct a multi-physical field-coupled equipment relationship graph based on the global feature data, where the equipment is used as nodes, the physical associations between the equipment are used as edges, and the edge attributes include fluid characteristics and heat transfer characteristics. Use a graph attention network to analyze the influence weights between the equipment to obtain the equipment layout order; The octree structure is adopted to adaptively discretize the equipment layout space into a multi-resolution three-dimensional grid. A reinforcement learning method with a multi-scale optimization strategy is applied to optimize the spatial position of the equipment. The translation and rotation parameters of the equipment are used as the action space, and the optimization objectives are to avoid collisions, maintain clearances, and ensure accessibility for maintenance. The optimal position and attitude parameters are obtained through iterative calculations. Taking the equipment layout sequence, optimal position, and attitude parameters as the layout plan, it is verified through three-dimensional dynamic simulation to evaluate the spatio-temporal envelope requirements for equipment maintenance operations. After passing the verification, it is confirmed as the final equipment layout plan.

[0007] In an alternative embodiment, Based on the global feature data, a multi-physics field coupled equipment relationship graph is constructed, where the equipment serves as nodes, the physical associations between equipment are edges, and the edge attributes include fluid characteristics and heat transfer characteristics. Using a graph attention network to analyze the influence weights between equipment, the equipment layout sequence is obtained, including: According to the geometric and physical characteristics, a node feature vector is constructed, which includes the spatial dimension parameters, position and attitude parameters, and operating parameters of the equipment. Based on the constraint relationships between equipment, the physical associations between equipment are determined, and the pipeline connection relationship and support structure relationship are used as edges to establish the topological connection between equipment. Based on the fluid characteristics and heat transfer characteristics between equipment, edge attributes are constructed, including: constructing a pressure loss matrix of fluid characteristics based on the Darcy-Weisbach equation, constructing a thermal resistance matrix of heat transfer characteristics based on the thermal conductivity and Nusselt number, calculating the physical field intensities of the fluid field and thermal field according to the equipment operating parameters, normalizing the pressure loss matrix and thermal resistance matrix, and then weighted fusion according to the physical field intensity ratio to obtain a comprehensive feature matrix; performing a consistency check on the comprehensive feature matrix based on mass conservation and energy conservation, and extracting the feature vectors of the edges to form an edge attribute feature set. Using a graph attention network to calculate the influence weights between equipment, integrating the node feature vector, topological connection, and edge attribute feature set into an edge type perception mechanism and gated fusion to obtain the equipment layout sequence.

[0008] In an alternative embodiment, Using a graph attention network to calculate the influence weights between equipment, integrating the node feature vector, topological connection, and edge attribute feature set into an edge type perception mechanism and gated fusion, the equipment layout sequence obtained includes: The equipment is divided according to functional types to construct a type embedding matrix, which is concatenated with the node feature vector to obtain an equipment feature vector. Construct an edge type parameter matrix for the physical associations between different types of devices. Based on the topological connections, calculate the attention scores for edge type perception by combining the device feature vectors and the edge type parameter matrix, and weight them with the edge attribute feature vectors to obtain edge type features; Construct a relationship transformation matrix to perform a non-linear transformation on the edge type features to obtain transformed features. Train a multi-layer perceptron based on the transformed features and the device feature vectors of adjacent devices to obtain gating coefficients, and adaptively fuse different types of transformed features using the gating coefficients to obtain the influence weights between devices; Construct a node-level attention heat map based on the influence weights between devices, and perform perturbation analysis on the device feature vectors to obtain feature sensitivities; Determine the device layout order based on the influence weights between devices, the influence degree indicators of the node-level attention heat map, and the feature sensitivities.

[0009] In an alternative embodiment, Adopt an octree structure to adaptively discretize the device layout space into multi-resolution three-dimensional grids, and apply a reinforcement learning method with a multi-scale optimization strategy to optimize the device spatial positions. Use the translation and rotation parameters of the devices as the action space, and take avoiding collisions, maintaining clearances, and ensuring accessibility for maintenance as the optimization objectives. Iteratively calculate the optimal position and attitude parameters, including: Use an octree structure to adaptively discretize the device layout space, and recursively subdivide the device boundary regions and device clearance regions to obtain multi-resolution three-dimensional grids; Construct the state space of the reinforcement learning based on the multi-resolution three-dimensional grids. The state space includes position and attitude parameter states, grid occupancy states, and constraint states. The action space includes the translation and rotation parameters of the devices, and the movement step size of the action space is proportional to the grid size of the corresponding grid level; Apply the reinforcement learning method on the multi-resolution three-dimensional grids to optimize the device positions. The reinforcement learning method adopts a multi-scale optimization strategy. At the coarse grid level, perform a global position search through a reward function that includes a collision penalty term and a clearance reward term to obtain the initial layout positions of the devices. At the fine grid level, perform local fine optimization by adding a maintenance accessibility evaluation term to the reward function, and iteratively obtain the initial optimal position and attitude parameters; Perform constraint verification on the multi-resolution three-dimensional grids based on the initial optimal position and attitude parameters, including: checking pipeline connection constraints, device spacing constraints, and support structure constraints; verifying the accessibility of maintenance channels, lifting paths, and operating spaces. After passing the verification, generate the final optimal position and attitude parameters.

[0010] In an alternative embodiment, The methods for adjusting the weights of each item in the reward function include: Design a hierarchical adaptive reward mechanism to determine the weights of the reward function according to the degree of constraint violation during the optimization process: when the collision volume between devices is greater than the first preset threshold, adjust the weight of the collision penalty term to a preset multiple of the current weight; when the device gap is less than the second preset threshold, adjust the weight of the gap reward term to a preset multiple of the current weight; when the number of accessible paths in the maintenance channel is less than the third preset threshold, adjust the weight of the maintenance accessibility evaluation term to a preset multiple of the current weight; when the degree of violation of each constraint is reduced to less than a preset ratio of the corresponding preset threshold, restore the corresponding weight to the initial setting value.

[0011] In an alternative implementation, Take the device layout order, optimal position, and attitude parameters as the layout plan, and verify through three-dimensional dynamic simulation. The evaluation of the spatio-temporal envelope requirements for device maintenance operations includes: Construct a three-dimensional geometric structure of the device layout scenario based on the device layout order, optimal position, and attitude parameters, where the optimal position parameter is the coordinate value of the device in three-dimensional space, and the attitude parameter is the rotation matrix of the device in three-dimensional space; Obtain the device maintenance operation trajectory, divide the device maintenance operation trajectory into operation sequences, and establish a set of human joint motion parameters and a set of tool operation parameters based on the operation sequences; Calculate the motion trajectory during the maintenance operation based on the set of human joint motion parameters and the set of tool operation parameters, and generate a spatio-temporal envelope for the maintenance operation; Perform interference detection between the spatio-temporal envelope of the maintenance operation and the three-dimensional geometric structure, and calculate the minimum distance value and collision volume value during the maintenance operation; Determine the accessibility of the maintenance operation based on the minimum distance value and the collision volume value, and generate a feasibility evaluation result for the device layout plan.

[0012] In an alternative implementation, Calculating the motion trajectory during the maintenance operation based on the set of human joint motion parameters and the set of tool operation parameters to generate a spatio-temporal envelope for the maintenance operation includes: Construct a velocity field function of the maintenance operation trajectory according to the set of human joint motion parameters and the set of tool operation parameters. The velocity field function includes a maintenance difficulty function, a curvature term, and a global constraint term. The maintenance difficulty function is related to the joint motion range and the tool operation torque; Drive the level set evolution based on the velocity field function to generate an initial spatio-temporal envelope; Establish a multi-scale uncertainty distribution of the operation trajectory, including: establishing a Gaussian distribution based on the standard deviation of the human joint positions and the standard deviation of the tool positions, calculating the joint probability distribution at each sampling moment, where the joint probability distribution characterizes the spatial distribution of the human-tool system at that moment, and accumulating in the time dimension to obtain the uncertainty characteristics of the complete operation trajectory; Dynamically adjust the expansion coefficient of the initial spatio-temporal envelope surface according to the uncertainty characteristics, where the expansion coefficient consists of a reference value and an uncertainty correction value, and the uncertainty correction value is the product of the larger value of the standard deviation of the joint positions and the standard deviation of the tool positions and a proportional coefficient related to the operation difficulty, to generate the final spatio-temporal envelope surface for maintenance operations.

[0013] In a second aspect of the embodiments of the present invention, there is provided an electronic device, including: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0014] In a third aspect of the embodiments of the present invention, there is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0015] The present invention realizes the efficient automation of equipment layout by proposing an intelligent optimization method for the layout of main engine island equipment based on three-dimensional modeling, and effectively solves many technical problems existing in traditional equipment layout.

[0016] By constructing a device relationship graph with multi-physical field coupling and applying a graph attention network to analyze the influence weights between devices, the present invention can scientifically and reasonably determine the equipment layout order, fully consider the physical relevance between devices, avoid the problem of unreasonable layout caused by subjective experience, and improve the equipment operation efficiency and safety.

[0017] The present invention adopts an adaptive multi-resolution three-dimensional grid with an octree structure and a reinforcement learning method to realize the intelligent optimization of the equipment spatial position. It not only ensures that the equipment layout meets multiple constraint conditions such as collision avoidance, clearance maintenance, and accessibility for maintenance, but also evaluates the spatio-temporal envelope surface of maintenance operations through a three-dimensional dynamic simulation verification mechanism, greatly improving the reliability and practicality of the equipment layout plan. Description of the Drawings

[0018] Figure 1 It is a flowchart of the intelligent optimization method for the layout of main engine island equipment based on three-dimensional modeling according to the embodiments of the present invention; Figure 2 It is a comparison diagram of the equipment layout effects of different layout order methods; Figure 3 It is a detection result diagram of the spatial envelope interference during the maintenance operation of the equipment layout plan. Specific implementation manners

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part rather than all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0021] Figure 1 It is a schematic flowchart of an intelligent optimization method for the equipment layout of the main engine island based on three-dimensional modeling in an embodiment of the present invention. As Figure 1 shown, the method includes: Establish a three-dimensional model of the main engine island, and apply feature recognition and semantic segmentation to extract the global feature data of the equipment. The global feature data includes the geometric features, physical features of the equipment, and the constraint relationships between the equipment; Construct a multi-physical-field-coupled equipment relationship graph based on the global feature data, where the equipment is used as nodes, the physical associations between the equipment are used as edges, and the edge attributes include fluid characteristics and heat transfer characteristics. Use a graph attention network to analyze the influence weights between the equipment to obtain the equipment layout order; Adopt an octree structure to adaptively discretize the equipment layout space into multi-resolution three-dimensional grids, and apply a reinforcement learning method with a multi-scale optimization strategy to optimize the spatial position of the equipment. Use the translation and rotation parameters of the equipment as the action space, and use collision avoidance, gap maintenance, and maintenance accessibility as the optimization objectives, and iteratively calculate to obtain the optimal position and attitude parameters; Use the equipment layout order, optimal position, and attitude parameters as the layout plan, verify through three-dimensional dynamic simulation, and evaluate the requirements of the spatial envelope during the equipment maintenance operation; after passing the verification, confirm it as the final equipment layout plan.

[0022] In an optional implementation manner, constructing a multi-physical-field-coupled equipment relationship graph based on the global feature data, where the equipment is used as nodes, the physical associations between the equipment are used as edges, and the edge attributes include fluid characteristics and heat transfer characteristics. Using a graph attention network to analyze the influence weights between the equipment to obtain the equipment layout order includes: Construct a node feature vector based on the geometric and physical characteristics, where the node feature vector includes the spatial dimension parameters, position and attitude parameters, and operating parameters of the device; Determine the physical association between devices based on the constraint relationships between devices, and establish the topological connection between devices by taking the pipeline connection relationship and the support structure relationship as edges; Construct edge attributes based on the fluid characteristics and heat transfer characteristics between devices, including: constructing a pressure loss matrix of fluid characteristics based on the Darcy-Weisbach equation, constructing a thermal resistance matrix of heat transfer characteristics based on the thermal conductivity and Nusselt number, calculating the physical field intensities of the fluid field and the thermal field according to the device operating parameters, and performing normalization processing on the pressure loss matrix and the thermal resistance matrix, and then weighted fusion according to the physical field intensity ratio to obtain a comprehensive feature matrix; performing consistency verification on the comprehensive feature matrix based on mass conservation and energy conservation, and extracting the feature vectors of the edges to form an edge attribute feature set; Calculate the influence weights between devices using a graph attention network, and integrate the node feature vector, topological connection, and edge attribute feature set into an edge type perception mechanism and gated fusion to obtain the device layout order.

[0023] Exemplarily, an accurate three-dimensional model of the main engine island is established by combining BIM technology and three-dimensional laser scanning. The PointNet++ deep learning framework is used to perform feature recognition on the point cloud data, and a three-dimensional semantic segmentation network based on the U-Net architecture is combined to extract the geometric features of the device, including dimensions, shapes, and spatial topological relationships. Import the device CAD model and extract the physical features of the device, including parameters such as weight. Automatically identify and extract connection constraints such as pipeline connection points and cable interfaces, as well as operation constraints such as maintenance space and installation distance.

[0024] The node feature vector contains the spatial dimension parameters, position and attitude parameters, and operating parameters of the device. The spatial dimension parameters include the length, width, and height of the device. For example, the spatial dimensions of the main transformer are 5.6 meters in length, 3.2 meters in width, and 4.8 meters in height. The position parameter is represented as the position of the center point of the device in the three-dimensional coordinate system, such as (x = 10.5, y = 8.2, z = 0.0). The attitude parameter is represented by Euler angles, such as (pitch angle = 5°, yaw angle = 0°, roll angle = 0°). The operating parameters include the rated power, flow rate, pressure, temperature, etc. of the device. For example, the rated power of the pump is 250 kW, the flow rate is 400 cubic meters per hour, the outlet pressure is 0.6 MPa, and the operating temperature is 85°C. These parameters are organized into a 128-dimensional feature vector and normalized to the [0, 1] interval using the MinMaxScaler method to ensure the effective fusion of features with different dimensions.

[0025] The pipeline connection relationship is determined based on the pipe diameter, medium type, and connection method. For example, the DN200 steam pipeline connection between two heat exchangers, with the medium being saturated steam and the connection method being flange connection. The support structure relationship is determined based on the structural mechanics model, including two types: foundation support and suspension support. Such as the fixed support between the pump and the foundation, and the elastic hanger support between the pipeline and the structural beam. The connection relationship is identified through the interface point matching of the 3D model, and then the constraint type is determined based on the pipeline layout specification and structural force calculation. The spatial tree index structure is used to accelerate the search for interface points, with the connection tolerance set to 5mm. The support mechanics calculation considers two working conditions: static load and dynamic load, and the load safety factor is 1.2. The topological connection is stored in the form of an adjacency matrix, where 1 indicates the existence of a connection and 0 indicates the absence of a connection.

[0026] For the fluid characteristics, a pressure loss matrix is constructed based on the Darcy - Weisbach equation. For each pair of connected devices, the pressure loss of the pipeline between them is calculated. For example, for the DN150 pipeline connecting the pump and the heat exchanger, with a length of 15 meters, considering the pipeline friction coefficient of 0.018, a flow velocity of 2.5 m / s, and a fluid density of 998 kg / m³, the calculated pressure loss is 35 kPa. For the heat transfer characteristics, a thermal resistance matrix is constructed based on the thermal conductivity and the Nusselt number. For example, the thermal conductivity between two devices is 45 W / (m·K), the convective heat transfer coefficient is 2800 W / (m²·K), considering the pipe wall thickness of 12 mm and the pipeline length of 20 meters, the calculated thermal resistance is 0.12 K / W. The physical field intensities of the fluid field and the thermal field are calculated according to the device operation parameters. For example, the fluid field intensity is calculated as 0.7 based on the flow rate and pressure, and the thermal field intensity is calculated as 0.4 based on the temperature difference and heat flow rate. The pressure loss matrix and the thermal resistance matrix are respectively normalized so that the numerical ranges are both within [0, 1], and then weighted and fused according to the physical field intensity ratio, that is, the fluid field weight of 0.7 and the thermal field weight of 0.4, to obtain the comprehensive characteristic matrix. Each element of this matrix represents the comprehensive physical coupling strength between devices. The consistency of the comprehensive characteristic matrix is verified based on the mass conservation and energy conservation to ensure the correctness of the physical relationship. Specifically, the system verifies the fluid mass balance of each node, requiring that the difference in the total fluid inflow and outflow of all devices is less than 1%; at the same time, an energy balance check is performed on the heat transfer, requiring that the total difference in heat input and output is less than 2%. Through this consistency check, the edge relationships that conform to the physical laws are screened out, and information such as the comprehensive strength value, connection type identifier, and physical parameter difference between each pair of connected devices is extracted to form a 32 - dimensional edge attribute feature vector, thus forming an edge attribute feature set.

[0027] Use the graph attention network to calculate the influence weights between devices and determine the device layout order.

[0028] The present invention realizes the accurate modeling and quantitative analysis of the physical association of the main engine island equipment, overcomes the problem of insufficient expression of the complex coupling relationship between equipment in the traditional method; constructs a device relationship diagram reflecting the multi-physical field coupling, and realizes the intelligent calculation of the equipment influence weight through the graph attention network, can capture the implicit association between equipment, automatically generate a reasonable equipment layout sequence, effectively improves the rationality of equipment layout and the engineering implementation efficiency, and provides intelligent support for the optimal layout of the main engine island equipment.

[0029] In an optional implementation manner, the graph attention network is used to calculate the influence weight between equipment, and the node feature vector, topological connection and edge attribute feature set are integrated into the edge type perception mechanism and gated fusion, and the equipment layout sequence obtained includes: The equipment is divided according to the functional type to construct a type embedding matrix, which is concatenated with the node feature vector to obtain the equipment feature vector; An edge type parameter matrix is constructed for the physical association between different types of equipment, and combined with the topological connection, the attention score of the edge type perception is calculated based on the equipment feature vector and the edge type parameter matrix, and is weighted with the edge attribute feature vector to obtain the edge type feature; A relationship transformation matrix is constructed to perform a non-linear transformation on the edge type feature to obtain a transformation feature, and a multi-layer perceptron is trained based on the transformation feature and the equipment feature vector of the adjacent equipment to obtain a gating coefficient, and the gating coefficient is used to adaptively fuse different types of transformation features to obtain the influence weight between equipment; Construct a node-level attention heat map based on the influence weight between equipment, and perform perturbation analysis on the equipment feature vector to obtain the feature sensitivity; Determine the equipment layout sequence based on the influence weight between equipment, the influence degree index of the node-level attention heat map and the feature sensitivity.

[0030] Exemplarily, the main engine island equipment can be divided into 10 functional types, including pump equipment, heat exchange equipment, storage equipment, valve equipment, etc. For each functional type of equipment, a 16-dimensional type embedding vector is constructed. For example, the type embedding vector of the pump equipment is [0.8, 0.2, 0.3,..., 0.5], and the type embedding vector of the heat exchange equipment is [0.2, 0.9, 0.1,..., 0.4]. The type embedding vector is obtained through a pre-trained equipment type classification network, which adopts a three-layer fully connected layer structure, and the number of hidden layer units is 64, 32, and 16 respectively, and the ReLU activation function is used, and is obtained through the joint training of the equipment CAD model features and parameter labels. For example, for the safety valve in the main steam pipeline system, its node feature vector is 128-dimensional and the type embedding vector is 16-dimensional, and the two are concatenated to obtain a 144-dimensional equipment feature vector. The concatenation operation adopts the vector connection method to ensure the effective fusion of the type information and the inherent features of the equipment.

[0031] In the host island system, the physical associations between devices mainly include four types: fluid connection, thermal connection, mechanical connection, and electrical connection. An 8×8 parameter matrix is constructed for each association type to capture the characteristic interaction patterns of different association types. For example, the parameter matrix of fluid connection focuses on flow rate and pressure characteristics, while the parameter matrix of thermal connection emphasizes temperature and heat flow characteristics. Specifically, for an edge e connecting two devices, the device feature vectors h_i and h_j of the two end devices are taken, the edge type is t, and the corresponding edge type parameter matrix is W_t. The attention score a_ij perceived by the edge type is calculated through matrix multiplication as a_ij = (h_i·W_t·transpose of h_j), and the Softmax function is used to normalize the attention scores of all edges connected to device i. Then, the normalized attention score is multiplied by the edge attribute feature vector to obtain the edge type feature. For example, for the fluid connection edge between the main pump and the heat exchanger, the calculated attention score is 0.28, which is multiplied by the 32-dimensional edge attribute feature vector of this edge to obtain the weighted edge type feature. In the actual system, a total of 4 edge type parameter matrices are constructed, occupying 256 parameters, which can effectively distinguish the influence patterns of different types of edges.

[0032] The relationship transformation matrix designs a 32×32 transformation matrix for each edge type. Through matrix multiplication, the edge type feature is transformed into a new feature space, and non-linearity is introduced through the LeakyReLU activation function to obtain the transformed feature. For example, after the relationship transformation of the fluid connection edge type feature, the pressure gradient and flow balance characteristics are highlighted. For the transformed feature and the feature vector of adjacent devices, a two-layer multi-layer perceptron is trained, with 16 hidden units and 1 output unit, to learn the importance of different types of transformed features. The output of the multi-layer perceptron is processed by the Sigmoid function to obtain the gating coefficient, and the value range is [0, 1]. For example, the gating coefficients of the fluid connection and thermal connection between a certain pair of devices are 0.75 and 0.45 respectively, indicating that the fluid connection relationship has a greater impact between devices. The different types of transformed features are weighted and averaged through the gating coefficient to obtain the comprehensive influence weight between devices.

[0033] The determination of the equipment layout sequence adopts a weighted sorting method, comprehensively considering three factors: the influence weight between equipment, the node-level attention value, and the feature sensitivity. For each equipment, its comprehensive score is calculated: Score = 0.5 × Node-level attention value + 0.3 × Maximum influence weight + 0.2 × Weighted sum of feature sensitivity. The influence degree index of the node-level attention heat map, that is, the node-level attention value, is a numerical index generated by aggregating the influence weights of all edges connected to this node, indicating the comprehensive importance of the equipment in the entire system network. For example, the node-level attention value of the main pump is 4.28, indicating its high connection importance in the entire equipment system. The maximum influence weight refers to the maximum correlation strength value between the current equipment and any other equipment, reflecting the key connection relationship of the equipment. For example, the influence weight between the main heat exchanger and the main pump is 0.82, which is higher than the connection weights with other equipment, so its maximum influence weight is taken as 0.82. The weighted sum of feature sensitivity considers the influence degree of each parameter of the equipment on the system, and sums up the sensitivities of each feature dimension of the equipment multiplied by the corresponding weight coefficients. After scoring 10 equipment in a certain main engine island according to this method, the comprehensive score of the main pump is 3.92, the main heat exchanger is 3.61, and the pressurizer is 3.22. Sorting according to the comprehensive score from high to low, the final equipment layout sequence is obtained to ensure that key equipment and closely related equipment can be given priority, reducing the constraint conflicts in subsequent layout.

[0034] Figure 2 Figure 4 is a comparison chart of the equipment layout effects of different layout sequence methods. The horizontal axis represents the layout steps (1 - 9), and the vertical axis represents the cumulative number of constraint conflicts (0 - 60). The four methods in the figure include: the present invention (edge type-aware graph attention network), represented by circular data points and solid lines; the degree centrality heuristic method, represented by square data points and solid lines; the expert experience rule method, represented by triangular data points and solid lines; and the conventional topological sorting method, represented by square data points and dashed lines. It can be clearly seen from the chart results that after 9 layout steps are completed, the method of the present invention only generates 27 constraint conflicts, which is significantly better than the other three methods. The degree centrality heuristic method generates 42 conflicts, the expert experience rule method generates 47 conflicts, and the conventional topological sorting method generates the most 51 conflicts. The chart shows that the method of the present invention maintains a low conflict growth rate from the first step, and as the layout steps increase, the gap with other methods gradually widens. Especially in the later stage of layout (steps 7 - 9), the method of the present invention shows obvious advantages, with the conflict quantity increasing relatively slowly, while other methods show a steeper growth trend. This indicates that the equipment layout sequence optimization method based on the edge type-aware graph attention network proposed by the present invention can more effectively consider various physical associations between equipment, thereby generating a more reasonable layout sequence and significantly reducing constraint conflicts.

[0035] The present invention effectively solves the problem of modeling multi-type physical associations in the layout of host island equipment by integrating a graph attention network and an edge type perception mechanism; by introducing type embeddings and relationship transformation matrices, it enhances the ability to express the functional characteristics and association types of equipment; adopts a gated fusion mechanism to achieve the adaptive integration of different physical associations, overcoming the limitation of single physical relationship expression in traditional methods; combines node attention and feature sensitivity analysis to make the layout sequence take into account both the importance of the system structure and parameter sensitivity.

[0036] In an alternative embodiment, an octree structure is used to adaptively discretize the equipment layout space into a multi-resolution three-dimensional grid, and a reinforcement learning method with a multi-scale optimization strategy is applied to optimize the spatial position of the equipment. Using the translation and rotation parameters of the equipment as the action space, with the objectives of avoiding collisions, maintaining clearances, and ensuring accessibility for maintenance and repair, the optimal position and attitude parameters are obtained through iterative calculation, including: The octree structure is used to adaptively discretize the equipment layout space, and recursive subdivision is performed in the equipment boundary region and the equipment clearance region to obtain a multi-resolution three-dimensional grid; Based on the multi-resolution three-dimensional grid, a state space for reinforcement learning is constructed. The state space includes position and attitude parameter states, grid occupancy states, and constraint states. The action space includes the translation and rotation parameters of the equipment, and the movement step size of the action space is proportional to the grid size of the corresponding grid level; The reinforcement learning method is applied to optimize the equipment position on the multi-resolution three-dimensional grid. The reinforcement learning method adopts a multi-scale optimization strategy. At the coarse grid level, a global position search is performed through a reward function that includes a collision penalty term and a clearance reward term to obtain the initial layout position of the equipment. At the fine grid level, local fine optimization is performed by adding a maintenance accessibility evaluation term to the reward function, and the initial optimal position and attitude parameters are obtained through iteration; Based on the initial optimal position and attitude parameters, constraint verification is performed on the multi-resolution three-dimensional grid, including: checking pipeline connection constraints, equipment spacing constraints, and support structure constraints; verifying the accessibility of maintenance channels, lifting paths, and operating spaces, and generating the final optimal position and attitude parameters after verification.

[0037] Exemplarily, taking a device layout space with a length of 10 meters, a width of 8 meters, and a height of 6 meters as an example, initially the entire space is divided into a grid of layer 0 with a size of 10×8×6 meters. Subsequently, adaptive subdivision is carried out according to the preset subdivision criteria. The subdivision criteria can be set as follows: if the current grid intersects with the device boundary, or is located between two devices and the distance is less than the preset threshold, then the grid is subdivided into 8 equal sub-grids. For example, for the grid of layer 0 containing the device boundary, the first subdivision is performed to obtain 8 sub-grids of layer 1 with a size of 5×4×3 meters; for the grids of layer 1 that still contain the device boundary or the device gap area, continue to subdivide to obtain the sub-grids of layer 2 with a size of 2.5×2×1.5 meters. In practical applications, for complex devices, up to 5 levels of subdivision can be performed, and the minimum grid size can reach 1 / 32 of the initial grid size, ensuring a sufficiently fine description of the key areas.

[0038] Construct the state space of reinforcement learning based on the multi-resolution three-dimensional grid, where the position and pose parameter states record the current position coordinates (x, y, z) and rotation angles (α, β, γ) of each device; the grid occupancy state records the occupancy of each grid cell in the multi-resolution grid, with a value of 0 indicating free, and a positive integer value indicating occupied by a certain device, and the occupancy value is equal to the device number; the constraint state records the satisfaction of various constraints, including distance constraints, connection constraints, and support constraints, etc. The action space includes the translation and rotation parameters of the device. The translation action is defined as the movement along the x, y, and z coordinate axes, and the movement step size is proportional to the grid size of the current grid level. For example, at the grid level of layer 2, the movement step size in the x direction is 2.5 meters, in the y direction is 2 meters, and in the z direction is 1.5 meters; the rotation action is defined as the rotation around the x, y, and z coordinate axes, and the rotation step size is 30 degrees at the coarse grid level and 10 degrees at the fine grid level.

[0039] Apply the reinforcement learning method on the multi-resolution three-dimensional grid for device position optimization. The reinforcement learning method adopts a multi-scale optimization strategy, which is divided into two stages: coarse optimization and fine optimization. Global position search is carried out at the coarse grid level (layers 0 - 2), and the optimization process is guided by a reward function that includes a collision penalty term and a gap reward term. The collision penalty term is set as: when it is detected that a collision occurs between devices or between a device and the boundary, a penalty value of -100 is given; the gap reward term is set as: when the distance between devices is greater than the minimum safety distance (set to 0.5 meters) but less than the maximum allowable distance (set to 2 meters), a positive reward related to the distance is given, up to +50. The Q-learning algorithm is used for training, with the learning rate set to 0.1 and the discount factor to 0.9. After 500 iterations in the coarse optimization stage, a globally optimal initial device layout position is obtained.

[0040] Subsequently, local fine optimization is carried out at the fine grid level (layers 3 - 5), and an overhaul accessibility evaluation item is added to the reward function. The overhaul accessibility evaluation uses the A* path planning algorithm to check the feasible paths from the entrance point to each equipment maintenance point. If a path exists and its length is less than the preset threshold, a positive reward is given, and the reward value is inversely proportional to the path length, up to +80 at most; if no feasible path exists, a penalty of -50 is given. For example, for a valve equipment that needs regular overhaul, its maintenance point is set on the north side of the equipment, and it is required that the path length from the entrance to the maintenance point does not exceed 15 meters, and the path width is not less than 0.8 meters. By setting the moving step size of the fine grid level to 0.3 meters and the rotation step size to 5 degrees, fine adjustment is carried out, and a total of 1000 iterations are performed, and finally the initial optimal position and attitude parameters are obtained.

[0041] Based on the initial optimal position and attitude parameters, constraint verification is carried out on the multi - resolution three - dimensional grid. The constraint verification includes two aspects: functional constraint verification and accessibility verification. Functional constraint verification includes: checking pipeline connection constraints, equipment spacing constraints, and support structure constraints. The pipeline connection constraint requires that pipelines can be arranged between the interface points of connected equipment, and the number of pipeline bends does not exceed the preset value of 3 times; the equipment spacing constraint requires that the minimum distance between equipment is not less than the safety distance of 0.5 meters; the support structure constraint requires that there is sufficient support area (at least 80% of the equipment bottom area) at the bottom of heavy equipment (weight exceeding 500 kg). Accessibility verification includes: verifying the accessibility of maintenance channels, lifting paths, and operation spaces. The maintenance channel verification requires that there is a channel with a width not less than 0.8 meters from the entrance point to the maintenance points of each equipment; the lifting path verification requires that there is an unobstructed vertical lifting space above large equipment (any dimension exceeding 1.5 meters); the operation space verification requires that a operation space not less than 1 m × 1 m is reserved around the equipment that needs manual operation.

[0042] After the above verification, if it is found that a certain constraint is not met, return to the third step for local adjustment; if all constraints are met, generate the final optimal position and attitude parameters to complete the equipment layout optimization process.

[0043] The present invention realizes the adaptive discretization of the equipment layout space through an octree structure, creates a multi - resolution three - dimensional grid, and enables the computing resources to be concentrated in key areas; combines a multi - scale optimization strategy based on reinforcement learning to achieve global position search at the coarse grid level and fine optimization at the fine grid level, effectively balancing the optimization scope and accuracy. At the same time, it integrates multi - objective constraints such as collision avoidance, clearance maintenance, and overhaul accessibility, and guides the layout optimization through a hierarchical reward function; overcomes the problems of low computational efficiency and difficulty in handling complex constraints of traditional layout methods, significantly improves the layout efficiency while ensuring the rationality of the layout, reduces manual intervention, and provides a new idea for the intelligent layout of industrial equipment.

[0044] In an alternative embodiment, the method for adjusting the weights of the terms in the reward function includes: Design a hierarchical adaptive reward mechanism to determine the weights of the reward function according to the degree of constraint violation during the optimization process: when the collision volume between devices is greater than a first preset threshold, adjust the weight of the collision penalty term to a preset multiple of the current weight; when the gap between devices is less than a second preset threshold, adjust the weight of the gap reward term to a preset multiple of the current weight; when the number of accessible paths in the maintenance channel is less than a third preset threshold, adjust the weight of the maintenance accessibility evaluation term to a preset multiple of the current weight; when the degree of violation of each constraint is reduced to a preset percentage below the corresponding preset threshold, restore the corresponding weight to the initial setting value.

[0045] Exemplarily, in the initial stage, initial weights are set for the terms in the reward function. Specifically, the initial weight of the collision penalty term can be set to 10.0, the initial weight of the device gap reward term can be set to 5.0, and the initial weight of the maintenance accessibility evaluation term can be set to 8.0. These initial weight values can be appropriately adjusted according to the specific application scenario to reflect the relative importance of different constraint conditions.

[0046] The hierarchical adaptive weight adjustment mechanism mainly includes the following specific steps: For the weight adjustment of the collision penalty term between devices: After each layout evaluation, calculate the total collision volume between all pairs of devices in the current layout plan. If the total collision volume is greater than a first preset threshold (e.g., 100 cubic centimeters), then adjust the weight of the collision penalty term to a preset multiple of the current weight. This preset multiple can be set to 2.0, that is, when a serious collision is detected, the weight of the collision penalty term will increase from the current value to 2 times of it. For example, if the current weight is 10.0, it will become 20.0 after adjustment, thereby enhancing the algorithm's tendency to avoid device collisions.

[0047] For the weight adjustment of the device gap reward term: When evaluating the current layout plan, calculate the minimum gap distance between all pairs of adjacent devices. If the minimum gap distance is less than a second preset threshold (e.g., 50 millimeters), then adjust the weight of the gap reward term to a preset multiple of the current weight. This preset multiple can be set to 2.0.

[0048] For the weight adjustment of the maintenance accessibility evaluation term: Count the number of accessible paths in the maintenance channel in the current layout plan. If the number of accessible paths is less than a third preset threshold (e.g., for a system with 10 key maintenance points, the threshold can be set to 7), then adjust the weight of the maintenance accessibility evaluation term to a preset multiple of the current weight. This preset multiple can be set to 2.0.

[0049] In the actual implementation process, it is also necessary to define the judgment criteria for reducing the degree of constraint violation. When the degree of violation of each constraint is reduced to less than the preset ratio of the corresponding preset threshold, the corresponding weight is restored to the initial set value. Specifically: When the collision volume between devices is reduced to less than 30% of the first preset threshold (100 cubic centimeters), that is, less than 30 cubic centimeters, the weight of the collision penalty term is restored to the initial value of 10.0; When the minimum clearance between devices increases to more than 150% of the second preset threshold (50 millimeters), that is, more than 75 millimeters, the weight of the clearance reward term is restored to the initial value of 5.0; When the number of accessible paths in the maintenance passage increases to more than 120% of the third preset threshold (7 paths), that is, greater than or equal to 9 paths, the weight of the maintenance accessibility evaluation term is restored to the initial value of 8.0.

[0050] To ensure the robustness of the algorithm, upper and lower limits should also be set during the weight adjustment process. For example, the upper limit of the weight of the collision penalty term can be set to 5 times the initial weight (i.e., 50.0), and the upper limits of the weights of the clearance reward term and the maintenance accessibility evaluation term can be set to 3 times the initial weight (i.e., 15.0 and 24.0) respectively. At the same time, the lower limit of all weights is not lower than their initial values.

[0051] Through the hierarchical adaptive reward mechanism of the present invention, the optimization algorithm can dynamically adjust the weights of various evaluation indicators according to the constraint violation conditions in different stages, effectively improving the layout optimization efficiency and result quality; it is applicable to the equipment layout optimization problem in a complex engineering environment, especially suitable for scenarios that need to meet multiple constraint conditions simultaneously.

[0052] In an alternative implementation, taking the equipment arrangement order, optimal position, and attitude parameters as the arrangement plan, through three-dimensional dynamic simulation verification, the evaluation of the spatio-temporal envelope requirements for equipment maintenance operations includes: Construct a three-dimensional geometric structure of the equipment arrangement scenario based on the equipment arrangement order, optimal position, and attitude parameters, where the optimal position parameter is the coordinate value of the equipment in three-dimensional space, and the attitude parameter is the rotation matrix of the equipment in three-dimensional space; Obtain the equipment maintenance operation trajectory, divide the equipment maintenance operation trajectory into operation sequences, and establish a set of human joint movement parameters and a set of tool operation parameters based on the operation sequences; Calculate the movement trajectory during the maintenance operation based on the set of human joint movement parameters and the set of tool operation parameters, and generate a spatio-temporal envelope for the maintenance operation; Perform interference detection between the spatio-temporal envelope of the maintenance operation and the three-dimensional geometric structure, and calculate the minimum distance value and collision volume value during the maintenance operation; Determine the accessibility of the maintenance operation based on the minimum distance value and the collision volume value, and generate a feasibility evaluation result of the equipment layout plan.

[0053] Exemplarily, construct a three-dimensional geometric structure of the equipment layout scenario based on the equipment layout sequence, optimal position, and attitude parameters. The equipment layout sequence is arranged from high to low according to the equipment importance. For example, in the main engine island, the equipment is arranged in the order of "main pump → main heat exchanger → pressurizer → condenser → feed water pump" in sequence. The optimal position parameter of each equipment is determined by the position of the equipment center point in the three-dimensional coordinate system, including three components (x, y, z), with the unit of meter. Taking the main pump as an example, its optimal position parameter is (12.45, 8.72, 1.50), indicating that the center of the main pump is located at the corresponding position in the coordinate system with the lower left corner of the layout area as the origin. The attitude parameter uses a rotation matrix to represent the orientation of the equipment, which is determined by the rotation angles α, β, γ around the three coordinate axes. The rotation matrix adopts the 3×3 matrix form and is obtained by rotating α angle around the z-axis, β angle around the y-axis, and γ angle around the x-axis in sequence. For example, the rotation angle of the main pump is (0, 0, 45°), indicating that it rotates only 45 degrees around the z-axis. Based on these parameters, use CAD modeling software or virtual reality engine to construct a complete three-dimensional scene model. This model includes the geometric shapes of all equipment, relative position relationships, and pipeline connections, with an accuracy up to the millimeter level. Technologically, a parametric modeling method is adopted, taking the geometric dimensions, position coordinates, and rotation angles of the equipment as input parameters to automatically generate the three-dimensional geometric structure. The model includes the equipment body structure, maintenance space, and connecting pipelines, detailed enough to reflect key parts such as the operation panel and maintenance interface on the equipment.

[0054] The maintenance operation trajectory is obtained through two methods: one is to capture the actions of on-site maintenance personnel during the operation process, using inertial sensors or optical tracking systems to record the action sequence of maintenance personnel completing specific maintenance tasks; the other is to construct a standardized maintenance action model by human factors engineering experts based on the standard operation procedures specified in the maintenance manual. Taking the routine maintenance of a certain heat exchanger as an example, the complete maintenance operation includes: approaching the equipment → opening the inspection door → checking the internal components → replacing the sealing ring → closing the inspection door → leaving the equipment. Divide this process into 25 basic operation actions to form an operation sequence. For each operation action, establish a set of human joint motion parameters, including the position and angle information of 23 key joint points of the human body to reflect the posture changes of the maintenance personnel. For example, during the "opening the inspection door" operation, record the parameter changes of the wrist, elbow, and shoulder joint angles of the maintenance personnel, as well as the hand trajectory, etc. At the same time, establish a set of tool operation parameters to record the position, attitude, and motion trajectory of the maintenance tool during the operation process. For example, record the position coordinates, rotation angle, and force direction of the wrench when using the wrench. These parameters are collected by the motion capture system with a sampling frequency of 60Hz, and the data accuracy is better than 1 centimeter and 1 degree to ensure accurately reflecting the actual operation characteristics.

[0055] Merge the human body movement trajectory and the tool movement trajectory to generate a complete set of maintenance operation movement trajectories. Based on the movement trajectory set, generate the maintenance operation spatio-temporal envelope surface by calculating the spatial envelope of the trajectory points.

[0056] Perform interference detection on the maintenance operation spatio-temporal envelope surface and the three-dimensional geometric structure. The interference detection uses a hierarchical collision detection algorithm, including two stages: rough detection and fine detection. In the rough detection stage, the axial bounding box technique is used to simplify the equipment and the spatio-temporal envelope surface into axial bounding boxes, and impossible collision situations are quickly excluded by comparing the overlapping relationships of the bounding boxes. In the fine detection stage, the separating axis theorem or the GJK algorithm is used for accurate collision detection to calculate the interference situation between the spatio-temporal envelope surface and the three-dimensional model of the equipment. Two types of key data are recorded during the calculation process: the minimum distance value and the collision volume value. The minimum distance value refers to the minimum clearance distance between the maintenance operation spatio-temporal envelope surface and the surrounding equipment, with the unit of meter. The collision volume value refers to the overlapping volume when the spatio-temporal envelope surface interferes with the equipment model, with the unit of cubic meter. Taking the maintenance operation of a certain main pump as an example, it is detected that the minimum distance between the spatio-temporal envelope surface and the adjacent pipe support is 0.08 meters, which is less than the safety clearance standard; it interferes with a certain valve control box, and the collision volume is 0.015 cubic meters. These data intuitively reflect the degree of spatial limitation of the maintenance operation and the potential collision risk.

[0057] The reachability determination of the maintenance operation is based on preset evaluation criteria, including the minimum safety clearance requirement and the maximum allowable collision volume. Generally, the minimum safety clearance requirement is set to 0.15 meters, considering the error of human body movement and operation comfort; the maximum allowable collision volume is set to 0 cubic meters, that is, theoretically no collision is allowed. In practical applications, these criteria can be appropriately adjusted according to project requirements. When it is detected that the minimum distance value is less than the minimum safety clearance requirement, or the collision volume value is greater than the maximum allowable collision volume, it is determined that the current maintenance operation is unreachable and the equipment layout plan is infeasible. The system will generate a detailed evaluation report, including: the identification of the problem equipment, the specific location of the collision or insufficient clearance, the severity score, etc. When the evaluation result does not meet the preset requirements, the system returns to the equipment position optimization step for re-iteration. The specific method is to adjust the position or attitude parameters of the equipment in the collision area, increase the weight of the maintenance space in the optimization target, and then re-run the position optimization algorithm. For example, for the problems found in the above-mentioned main pump maintenance operation, the system automatically adjusts the pipe support outwards by 0.1 meters and raises the installation height of the valve control box by 0.3 meters, and then re-verifies. When the evaluation result meets the preset requirements, confirm and output the final equipment layout plan, and generate a complete layout drawing and three-dimensional model including the equipment position, attitude, connection relationship and maintenance space.

[0058] Figure 3It is a detection result graph of the interference of the space envelope surface during the maintenance operation of the equipment layout plan, showing the performance comparison of three different methods (the present invention, the traditional geometric constraint method, and the manual layout method) during the equipment maintenance operation. The graph is divided into two parts on the left and right: the left side shows the "minimum interference distance of the maintenance operation" of each method, and the right side shows the "collision volume of the maintenance operation". The left graph shows that the method of the present invention has achieved a greater interference distance for all five types of equipment (main pump, heat exchanger, pressurizer, valve group, and storage tank). For the minimum interference distance of the main pump maintenance operation, the present invention reaches 37.5 cm, the traditional geometric constraint method is 26.8 cm, and the manual layout method is only 19.3 cm. The most obvious is the storage tank equipment. The interference distance achieved by the method of the present invention is 44.2 cm, which is 16.1 cm higher than the manual layout method. The right graph shows the comparison of the collision volumes of each method. The method of the present invention has achieved the minimum collision volume for all equipment. Especially for the pressurizer equipment, the collision volume of the method of the present invention is only 378 cubic centimeters, while the traditional method is 1123 cubic centimeters, and the manual layout method is as high as 1758 cubic centimeters. For the storage tank equipment, the collision volume of the method of the present invention drops to 158 cubic centimeters, which is 78.8% less than the traditional method and 88.5% less than the manual layout method. Generally speaking, the method of the present invention has reduced the collision volume by an average of 78.6%, significantly improving the safety and convenience of the maintenance operation. These results indicate that the three-dimensional dynamic simulation verification method proposed by the present invention based on the equipment layout sequence, optimal position, and attitude parameters can effectively improve the feasibility of the equipment maintenance operation and reduce the interference risk.

[0059] Through three-dimensional dynamic simulation and space envelope surface analysis, the present invention realizes the accurate evaluation of the feasibility of equipment maintenance operations. Compared with the traditional static space analysis method, this solution takes into account the space requirements during the dynamic operation of maintenance personnel, improving the accuracy and reliability of the evaluation. Through the quantitative analysis of the minimum distance value and the collision volume value, it provides a clear improvement direction for equipment layout optimization, effectively solving the problem of insufficient maintenance space in the dense layout environment of industrial equipment; the present invention not only improves the rationality of equipment layout, but also significantly reduces the safety risk and time cost of maintenance operations, providing strong support for the full-life cycle design of complex industrial systems.

[0060] In an alternative embodiment, calculating the motion trajectory during the maintenance operation based on the set of human joint motion parameters and the set of tool operation parameters, and generating the space-time envelope surface of the maintenance operation includes: Constructing a velocity field function of the maintenance operation trajectory according to the set of human joint motion parameters and the set of tool operation parameters, the velocity field function includes a maintenance difficulty function, a curvature term, and a global constraint term, and the maintenance difficulty function is related to the joint motion range and the tool operation torque; Drive the level set evolution based on the velocity field function to generate an initial spatio-temporal envelope surface; Establish a multi-scale uncertainty distribution of the operation trajectory, including: establishing a Gaussian distribution based on the standard deviation of the human joint positions and the standard deviation of the tool positions, calculating the joint probability distribution at each sampling moment, where the joint probability distribution characterizes the spatial distribution of the human-tool system at that moment, and accumulating over the time dimension to obtain the uncertainty characteristics of the complete operation trajectory; Dynamically adjust the expansion coefficient of the initial spatio-temporal envelope surface according to the uncertainty characteristics, where the expansion coefficient consists of a reference value and an uncertainty correction value, and the uncertainty correction value is the product of the larger value of the standard deviation of the joint positions and the standard deviation of the tool positions and a proportional coefficient related to the operation difficulty, to generate the final spatio-temporal envelope surface for maintenance operations.

[0061] Exemplarily, construct a velocity field function according to the human joint motion parameter set and the tool operation parameter set, and this function describes the motion trend of the maintenance operation in three-dimensional space. The velocity field function consists of three parts: a maintenance difficulty function, a curvature term, and a global constraint term. The maintenance difficulty function is directly related to the joint motion range and the tool operation torque, and is calculated as the weighted sum of the degree of deviation of the joint angle from the comfortable position and the corresponding joint importance weight, plus the weighted sum of the tool operation torque and the tool importance weight. For example, the weight of the shoulder joint is set to 0.3, the elbow is 0.25, and the wrist is 0.2; when the shoulder joint angle deviates from the comfortable angle by 45 degrees, the difficulty value of this joint is 0.3×(45 / 90)=0.15. The curvature term is used to smooth the trajectory and prevent sharp changes, and its value is the curvature of the trajectory curve at each point multiplied by a smoothing coefficient. The global constraint term considers the limitations of the equipment layout environment, such as narrow channels or fixed obstacles, and is represented by a distance field. The closer to the obstacle, the greater the constraint value. The three items are combined according to the weight ratio of 6:3:1 to obtain the complete velocity field function.

[0062] Generate an initial spatio-temporal envelope surface based on the velocity field function to drive the level set evolution. The level set method is an implicit surface representation technique that realizes the evolution of the surface by solving partial differential equations. Technically, first discretize the maintenance operation space into a three-dimensional grid with a grid resolution of 10 cm, covering the maintenance area. For example, the maintenance area of a certain heat exchanger is set as a space range of 3 m × 2 m × 2.5 m. Define an initial level set function on this grid. Usually, select a spherical surface centered at the starting position of the maintenance personnel as the initial level set, with a radius of 0.5 m, representing the space most basically occupied by the human body. Then, based on the previously constructed velocity field function, solve the level set evolution equation in an iterative manner. During the iterative process, the level set function expands along the direction of the velocity field, expanding slowly in areas with high maintenance difficulty and quickly in areas with low difficulty. Set the iteration step size to 0.1 and the maximum number of iterations to 200. By visualizing the zero isosurface of the level set function, obtain the initial spatio-temporal envelope surface, which reflects the approximate range of the maintenance operation in space but does not yet consider the uncertainty of the operation. The calculation results for the maintenance operation of a certain main pump show that the initial spatio-temporal envelope surface presents an irregular shape, shrinking in narrow spaces and expanding in areas with sufficient operation space, with a total volume of approximately 7.5 cubic meters.

[0063] Establish a multi-scale uncertainty distribution of the operation trajectory. The sources of uncertainty mainly include the natural variability of human movement, the accuracy limitations of tool operations, and the influence of environmental factors. For each joint point, calculate the standard deviation of its position by analyzing historical operation data or the differences in multiple repeated operations. For example, the standard deviation of the shoulder joint position is 2 cm, the elbow is 3 cm, the wrist is 4 cm, and the finger is 5 cm. The standard deviation increases with the increase in joint flexibility. For the tool position, determine the standard deviation according to the tool type and operation accuracy requirements. For example, the position standard deviation of precision operation tools is 1 cm, ordinary maintenance tools are 3 cm, and heavy tools are 5 cm. Based on these standard deviation values, establish a Gaussian distribution model, assuming that the variations of the positions of each joint and tool follow a normal distribution. For each sampling moment, calculate the joint probability distribution of the human joint and tool positions, characterizing the possible distribution of the human-tool system in space at that moment. Technically, use the Monte Carlo method to generate 1000 random samples that conform to the joint distribution for each moment to simulate possible position variations. In the time dimension, accumulate and superimpose the spatial distributions of all sampling moments to obtain the uncertainty characteristics of the complete operation trajectory. This accumulation uses the maximum value strategy, that is, for each point in space, take the maximum value of the probability distributions at all moments as the final uncertainty value of that point.

[0064] The expansion coefficient determines the extent of the expansion of the spatio-temporal envelope surface relative to the initial trajectory, directly affecting the size and shape of the envelope surface. The expansion coefficient consists of a reference value and an uncertainty correction value. The reference value is set according to the safety level of the maintenance operation. For example, for general maintenance operations, it is set to 1.2, indicating a basic expansion of 20%; for precision maintenance operations, it is set to 1.1; for heavy or high-risk maintenance operations, it is set to 1.3. The uncertainty correction value is the product of the larger value of the standard deviation of the joint position and the standard deviation of the tool position and the proportional coefficient related to the operation difficulty. The proportional coefficient related to the operation difficulty is determined according to the average value of the maintenance difficulty function. The greater the difficulty, the higher the coefficient, and the range is usually between 0.5 and 2.0. For example, for a precision valve adjustment operation, due to the high requirement for operation accuracy, the standard deviation of the tool position is 1 cm, which is less than the standard deviation of the finger joint position of 4 cm. Therefore, the larger value of 4 cm is taken; the operation difficulty is evaluated as 0.7 (medium difficulty), and the corresponding proportional coefficient related to the difficulty is 1.2; then the uncertainty correction value is 4 cm × 1.2 = 4.8 cm. The final expansion coefficient is the reference value of 1.1 plus the uncertainty correction value of 4.8 cm, which is applied to the initial spatio-temporal envelope surface. For each point on the initial envelope surface, it is expanded along the normal direction according to the expansion coefficient, and the expansion distance is proportional to the uncertainty characteristics of that point. More expansion is performed in the high-uncertainty area (such as the hand operation area), and less expansion is performed in the low-uncertainty area (such as the standing area). Through this dynamic adjustment, the generated spatio-temporal envelope surface for the maintenance operation more accurately reflects the actual maintenance requirements, avoiding space waste or safety risks.

[0065] The present invention combines ergonomic principles with computational geometry methods, and realizes the precise generation of the spatio-temporal envelope surface for maintenance operations through constructing a velocity field function, level set evolution, and uncertainty analysis. Compared with the traditional static space reservation method, the present invention takes into account the dynamic operation characteristics of the human body and uncertainty factors, and can more accurately reflect the actual maintenance requirements. By introducing a maintenance difficulty function and a dynamic expansion coefficient, the spatio-temporal envelope surface provides sufficient space guarantee in the key operation area, and avoids excessive space reservation in the non-key area, achieving a balance between space utilization and maintenance safety.

[0066] In the second aspect of the embodiments of the present invention, there is provided an electronic device, including: a processor; a memory for storing instructions executable by the processor; wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0067] In the third aspect of the embodiments of the present invention, there is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0068] The present invention may be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having thereon computer-readable program instructions for performing various aspects of the present invention.

[0069] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent optimization method for the layout of main engine island equipment based on 3D modeling, characterized in that, Including: Establish a 3D model of the main engine island, and apply feature recognition and semantic segmentation to extract the global feature data of the equipment. The global feature data includes the geometric features, physical features of the equipment, and the constraint relationships between the equipment; Based on the global feature data, construct a multi-physical-field-coupled equipment relationship graph, where the equipment serves as nodes, the physical associations between the equipment serve as edges, and the edge attributes include fluid characteristics and heat transfer characteristics. Use the graph attention network to analyze the influence weights between the equipment to obtain the equipment layout order; Adopt an octree structure to adaptively discretize the equipment layout space into a multi-resolution 3D grid, and apply a reinforcement learning method with a multi-scale optimization strategy to optimize the spatial position of the equipment. Use the translation and rotation parameters of the equipment as the action space, and take collision avoidance, gap maintenance, and maintenance accessibility as the optimization objectives, and iteratively calculate to obtain the optimal position and attitude parameters; Take the equipment layout order, optimal position, and attitude parameters as the layout plan, verify through 3D dynamic simulation, and evaluate the spatio-temporal envelope requirements for equipment maintenance operations; After passing the verification, confirm it as the final equipment layout plan.

2. The method according to claim 1, wherein Based on the global feature data, construct a multi-physical-field-coupled equipment relationship graph, where the equipment serves as nodes, the physical associations between the equipment serve as edges, and the edge attributes include fluid characteristics and heat transfer characteristics. Using the graph attention network to analyze the influence weights between the equipment to obtain the equipment layout order includes: Construct a node feature vector according to the geometric features and physical features. The node feature vector includes the spatial dimension parameters, position and attitude parameters, and operating parameters of the equipment; Determine the physical associations between the equipment based on the constraint relationships between the equipment, and establish the topological connections between the equipment by taking the pipeline connection relationship and the support structure relationship as edges; Construct edge attributes based on the fluid characteristics and heat transfer characteristics between the equipment, including: constructing a pressure loss matrix of fluid characteristics based on the Darcy-Weisbach equation, constructing a thermal resistance matrix of heat transfer characteristics based on the thermal conductivity and Nusselt number, calculating the physical field intensities of the fluid field and the thermal field according to the equipment operating parameters, and performing normalization processing on the pressure loss matrix and the thermal resistance matrix, and then weighted fusion according to the physical field intensity ratio to obtain a comprehensive feature matrix; Conduct a consistency test on the comprehensive feature matrix based on mass conservation and energy conservation, and extract the feature vectors of the edges to form an edge attribute feature set; Use the graph attention network to calculate the influence weights between the equipment, and integrate the node feature vector, topological connection, and edge attribute feature set into the edge type perception mechanism and gated fusion to obtain the equipment layout order.

3. The method according to claim 2, wherein Use the graph attention network to calculate the influence weights between the equipment, and integrate the node feature vector, topological connection, and edge attribute feature set into the edge type perception mechanism and gated fusion to obtain the equipment layout order includes: Divide the equipment according to the functional type to construct a type embedding matrix, and splice it with the node feature vector to obtain the equipment feature vector; Construct an edge type parameter matrix for the physical associations between different types of equipment, and calculate the attention scores of edge type perception based on the equipment feature vector and the edge type parameter matrix in combination with the topological connection, and perform weighting with the edge attribute feature vector to obtain the edge type feature; Construct a relationship conversion matrix to perform non-linear conversion on the edge type features to obtain conversion features, train a multi-layer perceptron based on the conversion features and the device feature vectors of adjacent devices to obtain gating coefficients, and use the gating coefficients to adaptively fuse different types of conversion features to obtain the influence weights between devices; Construct a node-level attention heat map based on the influence weights between devices, and perform perturbation analysis on the device feature vectors to obtain feature sensitivities; Determine the device layout order based on the influence weights between devices, the influence degree index of the node-level attention heat map, and the feature sensitivities.

4. The method according to claim 1, characterized in that, Adopt an octree structure to adaptively discretize the device layout space into multi-resolution three-dimensional grids, and apply a reinforcement learning method with a multi-scale optimization strategy to optimize the device spatial positions. Use the translation and rotation parameters of the devices as the action space, and take avoiding collisions, maintaining clearances, and ensuring accessibility for maintenance as the optimization objectives, and iteratively calculate to obtain the optimal position and attitude parameters, including: Adopt an octree structure to perform adaptive discretization on the device layout space, and perform recursive subdivision in the device boundary regions and device clearance regions to obtain multi-resolution three-dimensional grids; Construct the state space of reinforcement learning based on the multi-resolution three-dimensional grids. The state space includes position and attitude parameter states, grid occupancy states, and constraint states. The action space includes the translation and rotation parameters of the devices, and the movement step size of the action space is proportional to the grid size of the corresponding grid level; Apply the reinforcement learning method on the multi-resolution three-dimensional grids to optimize the device positions. The reinforcement learning method adopts a multi-scale optimization strategy. At the coarse grid level, perform global position search through a reward function containing a collision penalty term and a clearance reward term to obtain the initial layout positions of the devices. At the fine grid level, perform local fine optimization by adding an accessibility evaluation term for maintenance in the reward function, and iteratively obtain the initial optimal position and attitude parameters; Perform constraint verification on the multi-resolution three-dimensional grids based on the initial optimal position and attitude parameters, including: checking pipeline connection constraints, device spacing constraints, and support structure constraints; verifying the accessibility of maintenance channels, lifting paths, and operation spaces, and generate the final optimal position and attitude parameters after passing the verification.

5. The method according to claim 4, characterized in that The methods for adjusting the weights of each item in the reward function include: Design a hierarchical adaptive reward mechanism, and determine the reward function weights according to the degree of constraint violation during the optimization process: when the collision volume between devices is greater than the first preset threshold, adjust the weight of the collision penalty term to a preset multiple of the current weight; when the device clearance is less than the second preset threshold, adjust the weight of the clearance reward term to a preset multiple of the current weight; when the number of accessible paths for the maintenance channel is less than the third preset threshold, adjust the weight of the accessibility evaluation term for maintenance to a preset multiple of the current weight; when the degree of violation of each constraint is reduced to less than a preset ratio of the corresponding preset threshold, restore the corresponding weight to the initial set value.

6. The method according to claim 1, characterized in that, Take the device layout order, optimal position, and attitude parameters as the layout plan, verify through three-dimensional dynamic simulation, and evaluate the spatio-temporal envelope requirements for device maintenance operations, including: Construct a three-dimensional geometric structure of the equipment layout scenario based on the equipment layout sequence, optimal position, and attitude parameters, where the optimal position parameters are the coordinate values of the equipment in three-dimensional space, and the attitude parameters are the rotation matrices of the equipment in three-dimensional space; Obtain the equipment maintenance operation trajectory, divide the equipment maintenance operation trajectory into operation sequences, and establish a human joint motion parameter set and a tool operation parameter set based on the operation sequences; Calculate the motion trajectory during the maintenance operation based on the human joint motion parameter set and the tool operation parameter set, and generate a maintenance operation spatio-temporal envelope surface; Perform interference detection between the maintenance operation spatio-temporal envelope surface and the three-dimensional geometric structure, and calculate the minimum distance value and the collision volume value during the maintenance operation; Determine the reachability of the maintenance operation according to the minimum distance value and the collision volume value, and generate a feasibility evaluation result of the equipment layout plan.

7. The method according to claim 6, wherein Calculating the motion trajectory during the maintenance operation based on the human joint motion parameter set and the tool operation parameter set to generate a maintenance operation spatio-temporal envelope surface includes: Construct a velocity field function of the maintenance operation trajectory according to the human joint motion parameter set and the tool operation parameter set. The velocity field function includes a maintenance difficulty function, a curvature term, and a global constraint term. The maintenance difficulty function is related to the joint motion range and the tool operation torque; Drive the level set evolution based on the velocity field function to generate an initial spatio-temporal envelope surface; Establish a multi-scale uncertainty distribution of the operation trajectory, including: establishing a Gaussian distribution based on the standard deviation of the human joint position and the standard deviation of the tool position, calculating the joint probability distribution at each sampling moment, the joint probability distribution characterizing the spatial distribution of the human-tool system at that moment, and accumulating in the time dimension to obtain the uncertainty characteristics of the complete operation trajectory; Dynamically adjust the expansion coefficient of the initial spatio-temporal envelope surface according to the uncertainty characteristics. The expansion coefficient consists of a reference value and an uncertainty correction value. The uncertainty correction value is the product of the larger value of the joint position standard deviation and the tool position standard deviation and a proportional coefficient related to the operation difficulty, and generate the final maintenance operation spatio-temporal envelope surface.

8. An electronic device, characterized in that, Including: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

Images