Automatic optimization method and system for shipyard steel framed bent scheme

Abstract

The embodiment of the invention provides a shipyard steel framed bent scheme automatic optimization method and system. The method is applied to the technical field of buildings and comprises the following steps: selecting an initial size of a shipyard steel bent column beam section and bent design information based on engineering experience library data, and constructing high-precision input characteristics based on factory design information; based on the high-precision input features, constructing a physical constraint graph neural network, and predicting an initial design scheme meeting the physical constraint; and carrying out multi-objective reinforcement learning optimization on the initial design scheme meeting the physical constraint, and carrying out dynamic adjustment in a virtual simulation environment to obtain optimized steel bent design parameters. In this way, according to the method, the shipyard data are monitored in real time, the graph neural network is constructed for feature extraction and parameter optimization, the optimal steel frame parameters are obtained in dynamic adjustment, and the practical value is high.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of building technology, in particular to a side shipyard steel bent scheme automatic optimization method and system. BACKGROUND

[0002] Shipyard plant building is the main place for shipyard industrial production activities, and steel structure bent is the main structure form of shipyard plant building. Due to the special requirements of process production, shipyards are generally established in coastal locations near rivers, and the wind load is generally large. The plant is generally provided with a row of hoist cars with large lifting capacity, so the column, beam section size and wall thickness of the plant bent are large, which will inevitably result in a large amount of steel used for steel structure. Therefore, the fine design of the column, beam and other main force components of the steel bent will have an important influence on the investment cost of the shipyard.

[0003] At present, the calculation and design of steel bent mainly rely on the experience judgment of designers and are carried out in accordance with the “Steel Structure Design Standard” (GB50017-2017) and the “Building Seismic Design Standard” (GB 50011-2010). In addition, the existing calculation software still needs a large amount of manual processing in modeling and model adjustment. Therefore, it is urgent to solve the problems of low efficiency caused by insufficient experience of designers and manual processing of model adjustment.

[0004] Therefore, there is an urgent need for a method that can effectively improve the design optimization efficiency of shipyard steel bent. SUMMARY

[0005] The present disclosure provides a side shipyard steel bent scheme automatic optimization method and system, which monitors shipyard data in real time, constructs a graph neural network for feature extraction and parameter optimization, and at least solves the technical problems of low efficiency, low scheme accuracy and applicability of the prior art.

[0006] According to a first aspect of the present disclosure, a shipyard steel bent scheme automatic optimization method is provided, comprising the following steps: Based on the engineering experience database data, the initial size of the column and beam section of the shipyard steel bent and the bent design information are selected, and high-precision input features are constructed based on the plant design information; Based on the high-precision input features, a graph neural network with physical constraints is constructed to predict an initial design scheme that meets the physical constraints; The initial design scheme that meets the physical constraints is optimized by multi-objective reinforcement learning, and is dynamically adjusted in a virtual simulation environment to obtain optimized steel bent design parameters.

[0007] According to any possible implementation of the aspect as described above, further provided is an implementation, wherein the process of selecting initial sizes of the shipyard steel bent column and beam sections and bent design information based on the engineering experience database and constructing high-precision input features based on the factory building design information is as follows: According to the experience database, initial interface sizes and thicknesses of the factory building bent columns and beams are selected according to different workshop types, a laser scanner is used to scan the existing factory building to generate three-dimensional point cloud data, real-time stresses, wind pressures and equipment and material live loads of the factory building are collected respectively to obtain bent design information. The bent design information is subjected to outlier and normalization processing to obtain graph structure data and high-precision input features.

[0008] According to any possible implementation of the aspect as described above, further provided is an implementation, wherein the process of constructing a graph neural network with physical constraints based on the high-precision input features to predict an initial design scheme satisfying the physical constraints is as follows: Based on the high-precision input features, graph structure definition is performed to obtain node feature vectors and edge feature vectors. The graph neural network update rule is determined to update the node and edge features, and a loss function is calculated to construct the graph neural network with physical constraints. Based on the neural network, section parameter prediction is performed and node connection modes are constructed to output steel frame section size prediction values and the number of bolts.

[0009] According to any possible implementation of the aspect as described above, further provided is an implementation, wherein the loss function includes a physical constraint loss function and a Hooke loss function, and specifically: wherein, L force is the balance equation loss, F ij is the axial force of the edge e ij , A ij is the cross-sectional area, L ij is the original length of the steel frame, is the length change, L hooke is the Hooke loss, is the network weight.

[0010] According to any possible implementation of the aspect as described above, further provided is an implementation, wherein the process of performing multi-objective reinforcement learning optimization on the initial design scheme satisfying the physical constraints and dynamically adjusting in a virtual simulation environment is as follows: Based on the initial design scheme satisfying the physical constraints, a dynamically simulative virtual steel frame model is constructed to calculate real-time stress distribution and displacement field. calculate an initial total steel amount, determine a state space based on the initial total steel amount, a real-time stress distribution and a displacement field, and calculate a reward function, and determine an action space based on the design parameters; combine the state space, the action space and the reward function to construct a state-action-reward tuple, and use a proximal policy optimization training method to optimize to obtain optimized cross-section parameters.

[0011] According to the above-mentioned aspect and any possible implementation manner, further provided is an implementation manner, and the process of constructing a dynamically simulative virtual steel frame model and calculating a real-time stress distribution and a displacement field based on the initial design scheme satisfying the physical constraints is: perform finite element modeling based on the cross-section size prediction value to generate a stiffness matrix; calculate a dynamic load based on a real-time monitored live load and wind load of the factory building, and combine the stiffness matrix to construct a virtual steel frame model; based on the virtual steel frame model, calculate a steel frame node displacement and a stress distribution through a multi-threaded solver configuration.

[0012] According to the above-mentioned aspect and any possible implementation manner, further provided is an implementation manner, and the process of calculating an initial total steel amount, determining a state space based on the initial total steel amount, a real-time stress distribution and a displacement field is: calculate an initial total steel amount based on steel frame design parameters; calculate a member stress ratio and a node displacement exceeding rate based on an implemented stress distribution and displacement field of the member; perform normalization processing on the initial total steel amount, the stress ratio and the displacement exceeding rate respectively, and perform vector splicing to construct a state space.

[0013] According to the above-mentioned aspect and any possible implementation manner, further provided is an implementation manner, and the reward function is specifically: wherein R e is a member stress ratio, M(t) is a real-time material amount of a bolt, M0 is an initial material amount, D(t) is a node displacement exceeding rate, is a change amount of a Euclidean distance of an action vector.

[0014] According to a second aspect of the present disclosure, a shipyard steel frame scheme automatic optimization system is provided, comprising a factory building data acquisition and processing module, a steel frame scheme design module and a design parameter optimization module. The factory data acquisition and processing module is used for selecting initial sizes of a shipyard steel bent column beam section and bent design information based on engineering experience library data, and constructing high-precision input features based on factory design information. The steel bent scheme design module is used for constructing a physically constrained graph neural network based on the high-precision input features, and predicting an initial design scheme meeting the physical constraints. The design parameter optimization module is used for multi-objective reinforcement learning optimization of the initial design scheme meeting the physical constraints, and dynamic adjustment in a virtual simulation environment to obtain optimized steel bent design parameters.

[0015] Compared with the prior art, the present application has the following technical effects: The present application can obtain the best steel frame parameters through real-time monitoring of shipyard data and constructing a graph neural network for feature extraction and parameter optimization, and dynamic adjustment. Meanwhile, the present application can simultaneously optimize and adjust the steel frame parameters and the number of steel frames to the best and most economic benefit based on multi-objective reinforcement learning optimization, and can obtain an optimization method with economic benefit while meeting the mechanical properties of the steel frame, which has strong practical value.

[0016] It should be understood that the content described in the summary section is not intended to limit or important features of the embodiments of the present disclosure, nor to limit the scope of the present disclosure. Other features of the present disclosure will become apparent through the following description. BRIEF DESCRIPTION OF DRAWINGS

[0017] The above and other features, advantages, and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are intended to better understand the present disclosure, and do not limit the present disclosure. In the drawings, the same or similar reference numerals refer to the same or similar elements, wherein: Figure 1 A flowchart of an automatic optimization method for a side shipyard steel bent scheme is shown according to an embodiment of the present disclosure; Figure 2 A structure diagram of an automatic optimization system for a side shipyard steel bent scheme is shown according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0018] To make the purpose, technical scheme and advantages of the embodiments of the present disclosure clearer, the technical scheme in the embodiments of the present disclosure will be described clearly and completely below with reference to the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present disclosure.

[0019] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0020] Referring to Figure 1 The present embodiment provides a shipyard steel rack scheme automatic optimization method based on genetic algorithm, including the following steps: S101, based on the engineering experience database data, selecting the initial size of the shipyard steel rack column and beam section and the rack design information, and constructing high-precision input features based on the factory design information.

[0021] In the present embodiment, the cross-sectional experience database stores the cross-sectional size and plate thickness of the column and beam according to different workshop types, such as workshop height, span, crane lifting weight, etc. According to the design parameter information, the corresponding design information is queried in the experience library, and the initial cross-sectional size and plate thickness of the rack column and beam are selected.

[0022] After obtaining the data, the existing factory is scanned by a laser scanner to generate point cloud data, recording the three-dimensional coordinates (xi, yi, zi) of each node of the steel frame; Subsequently, load data collection is carried out, and strain sensors, anemometers and weighing equipment are installed at different positions of the existing factory, and the real-time stress of the key parts of the steel frame is measured by the strain sensors , specifically: , wherein F(t) is the real-time axial force, and A is the initial cross-sectional size. The wind pressure W(z) at different heights of the steel frame is recorded by the anemometer, and the live load Q(t) of the equipment and materials is monitored by the weighing equipment; Again, the steel rack sample is subjected to tensile test to obtain the yield strength f y and the elastic modulus E t ; Subsequently, the data is subjected to outlier processing and normalization processing, and the graph structure data G=(V,E) is output, wherein the node attributes are: coordinates (x i , y i , z i ), load F i =G i +Q i +W i , wherein G i is the dead load (structure self weight), Q i is the live load, and W i is the wind load, and the edge attributes are: cross-sectional parameters (b ij , h ij , t ij ), respectively, height, width, thickness, and material f y , E.

[0023] S102, based on the high-precision input features, a graph neural network with physical constraints is constructed to predict an initial design scheme that meets the physical constraints.

[0024] In the present embodiment, the joint modeling process based on the graph neural network (GNN) and the physical constraints is as follows: First, a message passing mechanism is constructed: In the network, each layer of GNN updates the node and edge features, specifically: (1) wherein, is the feature vector of the ith node in the lth layer, are trainable weight matrices, is the edge feature, is an edge message function, realized by an MLP multi-layer perceptron, specifically: (2) Subsequently, a loss function is calculated, including a physical constraint loss function and a Hooke's loss function.

[0025] First, the physical constraint loss is constructed, specifically: first, physical constraint embedding is performed, including: balance equation loss: force balance of nodes is forced , , .

[0026] Specifically: (3) wherein, L force is the balance equation loss, F ij is the axial force of edge e ij .

[0027] Secondly, the Hooke's law loss is calculated to constrain the stress-strain relationship , specifically: (4) wherein, A ij is the cross-sectional area, specifically: A ij =b ji ·t ij , L ij is the original length of the steel frame, is the length change, L hooke is the Hooke's loss.

[0028] The prediction result is obtained based on the output of the constructed network, specifically: (5) (6) wherein, , , is a cross-sectional dimension predicted output value, is a cross-sectional predicted width, is a cross-sectional predicted height, is a web predicted thickness, n bolt is the number of bolts.

[0029] S103, multi-objective reinforcement learning optimization is performed on the initial design scheme satisfying the physical constraint, and dynamic adjustment is performed in a virtual simulation environment.

[0030] First, with the initial design scheme satisfying the physical constraint, a dynamically simulative virtual steel frame model is constructed, and the specific process is as follows: Based on the cross-sectional dimension predicted value, a stiffness matrix is generated, and finite element modeling is performed, specifically as follows: (7) wherein, E e is the elastic modulus of the e-th component, A e is the cross-sectional area, and L e is the component length.

[0031] Secondly, based on the real-time monitoring of the live load and wind load of the workshop, the virtual steel frame model is injected, specifically as follows: (8) wherein, F static is the constant load, F wind (t) is the time-varying wind load, F equip (t) is the equipment vibration load, which is obtained by fitting the acceleration sensor data, and F(t) is the dynamic load.

[0032] Finally, the node displacement and stress distribution are obtained by configuring the multi-threaded solver for calculation, specifically as follows: (9) wherein, u i , u j are the displacements of the nodes at both ends of the component.

[0033] After the construction of the steel frame model and the output of the calculation results, the reinforcement learning strategy is designed, and how to adjust the design parameters in the dynamic environment is trained, and the specific process is as follows: First, the state space S is determined, specifically as follows: Based on the steel design parameters, the initial total steel consumption is calculated, specifically as follows: (10) wherein, is the density of steel, , , , respectively, the cross-sectional width, height and thickness of the e-th member, L e is the member length.

[0034] The member stress ratio and the node displacement exceeding rate are calculated based on the stress value and the displacement field of the member implementation, specifically: (11) (12) where u i (t) is the three-dimensional displacement vector of node i at time t, L i is the span of the main beam where node i is located.

[0035] Finally, the stress, displacement and material consumption are normalized to obtain , and , respectively, so as to determine the spatial state S, specifically: (13) where is the vector splicing operation.

[0036] After completing the state space description, the action space is designed, which needs to map the output of the reinforcement learning strategy to the adjustment of the actual engineering parameters (such as cross-sectional height, thickness).

[0037] First, adjust the cross-sectional size, the specific process is: Make incremental decisions on the construction height and adjust the thickness ratio, specifically: (14) (15) where is the adjustment amplitude, is the height adjustment amount, is the thickness adjustment coefficient.

[0038] Second, optimize the node connection, specifically adjust the number of bolts, specifically: (16) where is the bolt number adjustment parameter.

[0039] Finally, the action vector is synthesized, specifically: (17) Subsequently, the reward function is calculated, which can determine the safety characteristics of the member by calculating the reward function, specifically: (18) wherein, is the Euclidean distance change of action vector, used to suppress policy mutation.

[0040] Finally, based on the state space description, action space description and reward function obtained by the component, the state-action-reward tuple (S t , a t , r t ) is constructed, and the proximal policy optimization (PPO) training method is used to obtain the optimized cross-section parameters.

[0041] In the above technical solution, the embodiment realizes multi-dimensional dynamic optimization of steel structure performance by real-time fusion of mechanical feedback of digital twin model and reinforcement learning algorithm: based on real-time monitoring of stress ratio and displacement exceeding rate, the component cross-section parameters and node connection strategy are automatically adjusted, the maximum stress margin of the structure is reduced, and the risk of plastic deformation is significantly inhibited; through dynamic optimization of material consumption and intelligent identification of redundant components, the steel consumption is significantly reduced compared with the initial design under the premise of meeting the safety threshold, directly reducing the engineering cost; at the multi-objective coordination level, the exponential reward function converts the safety, economy and stability indicators into a unified optimization target, avoiding the performance imbalance problem caused by traditional single-objective optimization; at the engineering implementation level, the automatic decision mechanism can quickly respond to sudden load changes (such as extreme wind conditions), and the efficiency is significantly improved compared with manual parameter adjustment, and through closed-loop interaction with measured data and initial design, an intelligent optimization chain is formed throughout the design-construction-operation whole cycle, and the trained strategy model can also be migrated to similar steel structure projects, shortening the optimization period of new projects, and having significant engineering universal value.

[0042] As shown in Figure 2 , the embodiment also provides an automatic optimization system for shipyard steel bent scheme, comprising: a factory data acquisition and processing module 1, a steel bent scheme design module 2, and a design parameter optimization module 3; The factory data acquisition and processing module 1 is used to select the initial size of the steel bent column and beam section and the bent design information based on the engineering experience database data, and to construct high-precision input features based on the factory design information; The steel bent scheme design module 2 is used to construct a physically constrained graph neural network based on the high-precision input features, and to predict an initial design scheme that meets the physical constraints; The design parameter optimization module 3 is used to perform multi-objective reinforcement learning optimization on the initial design scheme that meets the physical constraints, and to dynamically adjust in a virtual simulation environment to obtain optimized steel bent design parameters.

[0043] It should be noted that, for the foregoing method embodiments, the purposes of simplicity and brevity of the description are served, as a series of actions are described, but those skilled in the art should know that the present disclosure is not limited to the order of the actions described, because, according to the present disclosure, certain steps can be performed in other orders or at the same time. Secondly, those skilled in the art should know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to the present disclosure.

[0044] It should be understood that the steps can be reordered, added, or deleted using the various forms of flow shown above. For example, the steps described in the present disclosure can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, which is not limited herein.

[0045] The above detailed description does not constitute a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for automatically optimizing a shipyard steel bent scheme, characterized in that: The following steps are involved: Based on the engineering experience database data, the initial dimensions of the shipyard steel bent column and beam sections and bent design information are selected, and high-precision input features are constructed based on the plant design information; Based on the high-precision input features, a physical constraint graph neural network is constructed to predict an initial design solution that satisfies the physical constraints; The initial design scheme that meets the physical constraints is optimized through multi-objective reinforcement learning and dynamically adjusted in a virtual simulation environment to obtain the optimized steel frame design parameters.

2. The method for automatic optimization of shipyard steel bent scheme according to claim 1, characterized in that: The process of selecting the initial dimensions of the shipyard steel bent column and beam cross-section and bent design information based on the engineering experience database data, and constructing high-precision input features based on the plant design information is as follows: Based on the empirical database data, the initial interface dimensions and thicknesses of the factory building's frame columns and beams were selected according to different workshop types. A laser scanner was used to scan the existing factory building to generate 3D point cloud data. The real-time stress, wind pressure, and live load of equipment and materials were collected to obtain the frame design information. The rack design information is processed for outliers and normalization, and graph structure data is constructed to obtain high-precision input features.

3. The method for automatic optimization of shipyard steel bent scheme according to claim 1, characterized in that: The process of constructing a graph neural network with physical constraints based on the high-precision input features and predicting an initial design solution that satisfies the physical constraints is as follows: Based on the high-precision input features, a graph structure is defined to construct node feature vectors and edge feature vectors; Determine the graph neural network update rule to update the node and edge features, calculate the loss function, and construct a physically constrained graph neural network; Based on the neural network, cross-section parameters are predicted and a node connection method is constructed, and a predicted value of the steel frame cross-section size and the number of bolts are output.

4. The method for automatically optimizing a shipyard steel bent scheme according to claim 3, characterized in that: The loss function includes a physical constraint loss function and a Hook loss function, specifically: Among them, L force is the loss of the balance equation, F ij For edge e ij Axial force, A ij is the cross-sectional area, L ij is the original length of the steel frame, is the length change, L hooke For Hooker's loss, is the network weight.

5. The method for automatically optimizing a shipyard steel bent scheme according to claim 4, characterized in that: The process of performing multi-objective reinforcement learning optimization on the initial design solution that meets the physical constraints and performing dynamic adjustment in the virtual simulation environment is as follows: Based on the initial design scheme that meets the physical constraints, a virtual steel frame model capable of dynamic simulation is constructed and real-time stress distribution and displacement field are calculated; calculating an initial total steel usage, determining a state space based on the initial total steel usage, the real-time stress distribution, and the displacement field, calculating a reward function, and determining an action space description based on the design parameters; Combining the state space, the action space description and the reward function, a state-action-reward tuple is constructed, and a proximal policy optimization training method is used for optimization to obtain optimized cross-sectional parameters.

6. The method for automatic optimization of shipyard steel bent scheme according to claim 5, characterized in that: The process of constructing a virtual steel frame model capable of dynamic simulation based on the initial design solution that meets the physical constraints and calculating the real-time stress distribution and displacement field is as follows: Performing finite element modeling based on the predicted cross-sectional dimensions to generate a stiffness matrix; The dynamic load is calculated based on the live load and wind load monitored in real time in the plant, and a virtual steel frame model is constructed in combination with the stiffness matrix; Based on the virtual steel frame model, the steel frame node displacement and stress distribution are calculated by configuring a multi-threaded solver.

7. The method for automatic optimization of shipyard steel bent scheme according to claim 5, characterized in that: The process of calculating the initial total steel usage and determining the state space based on the initial total steel usage, the real-time stress distribution and the displacement field is as follows: The initial total steel consumption is calculated based on the steel frame design parameters; Based on the actual stress distribution and displacement field of the component, the component stress ratio and node displacement exceeding standard rate are calculated; The initial total steel usage, stress ratio and displacement exceeding standard rate are respectively normalized and vector splicing is performed to construct a state space.

8. The method for automatic optimization of shipyard steel bent scheme according to claim 5, characterized in that: The reward function is specifically: Among them, R e is the stress ratio of the component, M(t) is the real-time material consumption of the bolt, M0 is the initial material consumption, D(t) is the node displacement exceeding standard rate, is the Euclidean distance change of the motion vector.

9. A shipyard steel bent scheme automatic optimization system, implemented using the shipyard steel bent scheme automatic optimization method according to any one of claims 1 to 8, characterized in that: include: Plant data acquisition and processing module (1), steel rack scheme design module (2) and design parameter optimization module (3); The plant data acquisition and processing module (1) is used to select the initial size of the shipyard steel bent frame column and beam cross section and bent frame design information based on the engineering experience database data, and to construct high-precision input features based on the plant design information; The steel bent scheme design module (2) is used to construct a graph neural network of physical constraints based on the high-precision input features, and predict an initial design scheme that meets the physical constraints; The design parameter optimization module (3) is used to perform multi-objective reinforcement learning optimization on the initial design scheme that meets the physical constraints, and to perform dynamic adjustment in a virtual simulation environment to obtain optimized steel frame design parameters.

Images