CN114880741B - Building structure component size design method and device embedded with domain knowledge - Google Patents
Building structure component size design method and device embedded with domain knowledge Download PDFInfo
- Publication number
- CN114880741B CN114880741B CN202210444469.9A CN202210444469A CN114880741B CN 114880741 B CN114880741 B CN 114880741B CN 202210444469 A CN202210444469 A CN 202210444469A CN 114880741 B CN114880741 B CN 114880741B
- Authority
- CN
- China
- Prior art keywords
- design
- building structure
- loss
- size
- type
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/20—Configuration CAD, e.g. designing by assembling or positioning modules selected from libraries of predesigned modules
Abstract
The invention relates to a building structure member dimension design method and a device embedded with domain knowledge, comprising the following steps: respectively inputting a structural arrangement diagram and design conditions of a target building structure into a pre-stored size design model corresponding to each type of component to obtain the design size of each type of component in the target building structure; generating a component dimension map for the target building structure based on the design dimensions of all types of components in the target building structure; the size design model is optimally trained by taking the minimum comprehensive design loss of a design feature tensor of a building structure member size diagram as a target; the integrated design loss is a weighted sum of the image loss and the domain knowledge loss corresponding to each type of component. The invention can quickly and reliably complete the intelligent design of the size of the building structural member which accords with the knowledge of the related field of structural design.
Description
Technical Field
The invention relates to the field of building structure design based on artificial intelligence, in particular to a building structure member size design method and device embedded with domain knowledge.
Background
In the design stage of the scheme of the building structure, in order to ensure the safety and the economy of the structural design scheme, the size design of the structural members needs to be carried out quickly and reasonably on the basis of the structural arrangement scheme and under the constraint of structural design conditions.
However, the traditional structural member size design scheme mainly depends on personal experiences of architects and structural engineers, the manual design mode is time-consuming and labor-consuming, the design process efficiency is low, and the existing design experience is difficult to inherit. The size design scheme of the emerging intelligent structural component can not consider the domain knowledge in the structural design, so that the generated design scheme is against the basic common knowledge of structural engineers, and the design effect is difficult to further promote.
Disclosure of Invention
The invention aims to provide a method and a device for designing the size of a building structural member with embedded domain knowledge, which are used for solving the problems that the manual design of the size of the structural member is time-consuming and labor-consuming, the efficiency of the design process is low and the existing design experience is difficult to inherit, solving the problem that the design scheme often violates the basic common knowledge of structural engineers when the size of the structural member is intelligently designed, and realizing the efficient and accurate design of the size of the structural member and considering the domain knowledge related to the structural design.
In a first aspect, the present invention provides a method of building structural element dimensioning incorporating domain knowledge, comprising:
respectively inputting the structural arrangement diagram and the design condition of the target building structure into a pre-stored size design model corresponding to each type of component to obtain the design size of each type of component in the target building structure;
generating a component dimension map for the target building structure based on the design dimensions of all types of components in the target building structure;
the size design model is optimally trained by taking the minimum comprehensive design loss of a design feature tensor of a building structure member size diagram as a target;
the comprehensive design loss is a weighted sum of image loss and field knowledge loss corresponding to each type of component;
the image loss is the difference between the corresponding design feature tensor and the corresponding ideal feature tensor;
the design conditions comprise: aseismic design conditions, wind-resistant design conditions and height design conditions.
In a second aspect, the present invention also provides an apparatus for dimensioning an architectural structural member incorporating domain knowledge, the apparatus comprising:
the input module is used for respectively inputting the structural arrangement drawing and the design condition of the target building structure into a pre-stored size design model corresponding to each type of component to obtain the design size of each type of component in the target building structure;
a generating module for generating a component dimension map of the target building structure based on the design dimensions of all types of components in the target building structure;
the size design model is optimally trained by taking the minimum comprehensive design loss of the design characteristic tensor of the building structural member size diagram as a target;
the comprehensive design loss is a weighted sum of image loss and field knowledge loss corresponding to each type of component;
the image loss is the difference between the corresponding design feature tensor and the corresponding ideal feature tensor;
the design conditions comprise: seismic design conditions, wind resistance design conditions and height design conditions.
In a third aspect, the present invention also provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the method for building structural member dimension design with embedded domain knowledge as described in any one of the above.
The invention provides a building structure component size design method and device embedded with domain knowledge, which are characterized in that a size design model taking minimum weighted combination between image loss and domain knowledge loss of corresponding types of components as an optimization target is trained aiming at different types of components in a building structure in advance, the difference between a design characteristic tensor corresponding to a component size diagram and an ideal characteristic tensor is taken as image loss in the model, the component size diagram mapped by the design characteristic tensor is ensured to be proper by introducing the image loss, and the size design of the corresponding types of components is ensured to be optimal by introducing the domain knowledge loss of the corresponding types of components. Secondly, inputting the structural arrangement diagram and the design condition of the target building structure into a size design model corresponding to each type of component respectively to obtain the design size of each type of component in the target building structure; generating a component dimension map for the target building structure based on the design dimensions of all types of components in the target building structure; the invention can quickly and reliably complete the intelligent design of the size of the building structural member which accords with the knowledge of the related field of structural design.
Drawings
In order to more clearly illustrate the technical solutions of the present invention or the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a flow chart of a method for dimensioning building structural members incorporating domain knowledge provided by the present invention;
FIG. 2 is a schematic diagram of a neural network building process for embedding domain knowledge provided by the present invention;
FIG. 3 is a schematic diagram of the weight change of the image loss and the domain knowledge loss under each type of component selected by the optimizer provided by the present invention;
FIG. 4 is a network architecture of a sizing model provided by the present invention;
FIG. 5 is a schematic diagram of a sizing model training and application process provided by the present invention;
FIG. 6 is a schematic diagram of the pre-processor implementation provided by the present invention;
FIG. 7 is a schematic diagram of the post-processor implementation provided by the present invention;
FIG. 8 is a diagram of matrix orientation definition provided by the present invention;
FIG. 9 is a block diagram of a domain knowledge embedded building structure component sizing device provided by the present invention;
FIG. 10 is a schematic structural diagram of an electronic device implementing a domain knowledge embedded architectural structural member sizing method provided by the present invention;
reference numerals:
a, weight change of image loss; b, weight change of the x-th theoretical equal design under each type of component; c, weight change of z-th theoretical unequal design under each type of component.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The present invention, embedded with domain knowledge, building structure component sizing method and apparatus is described below in conjunction with fig. 1-10.
In a first aspect, as shown in fig. 1, to solve the defects that the size design method of the building structural member in the prior art is inefficient and cannot consider knowledge in the related field of structural design, the invention provides a size design method of a building structural member embedded with the knowledge in the field, comprising:
s11, respectively inputting a structural arrangement diagram and design conditions of a target building structure into a pre-stored size design model corresponding to each type of component to obtain the design size of each type of component in the target building structure; it will be appreciated that the present invention trains corresponding domain knowledge embedded sizing models for different component types.
S12, generating a component size diagram of the target building structure based on the design sizes of all types of components in the target building structure;
the size design model is optimally trained by taking the minimum comprehensive design loss of the design characteristic tensor of the building structural member size diagram as a target; the comprehensive design loss is a weighted sum of image loss and field knowledge loss corresponding to each type of component; the image loss is the difference between the corresponding design feature tensor and the corresponding ideal feature tensor; the design conditions comprise: seismic design conditions, wind resistance design conditions and height design conditions.
The invention provides a building structure component size design method embedded with domain knowledge, which is characterized in that a size design model taking minimum weighted combination between image loss and domain knowledge loss of corresponding types of components as an optimization target is trained aiming at different types of components in a building structure in advance, the difference between a design characteristic tensor corresponding to a component size diagram and an ideal characteristic tensor is taken as the image loss in the model, the component size diagram mapped by the design characteristic tensor is ensured to be proper by introducing the image loss, and the size design of the corresponding types of components is ensured to be optimal by introducing the domain knowledge loss of the corresponding types of components. Then, respectively inputting the structural arrangement drawing and the design condition of the target building structure into a size design model corresponding to each type of component to obtain the design size of each type of component in the target building structure; generating a component dimension map of the target building structure based on the design dimensions of all types of components in the target building structure; the invention can quickly and reliably complete the intelligent design of the size of the building structural member which accords with the knowledge of the related field of structural design.
On the basis of the above embodiments, as an alternative embodiment, the size design model includes a pre-processor, a neural network embedded with domain knowledge corresponding to each type of component, and a post-processor; the step of respectively inputting the structural arrangement diagram and the design condition of the target building structure into a pre-stored size design model corresponding to each type of component to obtain the design size of each type of component in the target building structure comprises the following steps:
based on a pre-processor, coding a structural arrangement diagram and design conditions of the target building structure, and stacking to obtain an input feature tensor of the target building structure;
the method comprises the steps of coding a structural layout diagram and a design condition by utilizing a preprocessor, and coding a combination of the structural layout diagram and any design condition/component type into a two-dimensional feature matrix, wherein the position of a non-zero element in the two-dimensional feature matrix represents structural arrangement, and the value of the non-zero element represents the design condition or the component type; stacking the two-dimensional feature matrixes in a third dimension to obtain an input feature tensor in a tensor form;
extracting high-dimensional features in the input feature tensor based on a neural network of embedded domain knowledge corresponding to each type of component, and performing up-sampling on the high-dimensional features to obtain a design feature tensor of the target building structure component size diagram;
and decoding the design feature tensor of the target building structure component size graph based on a post processor to obtain the design size corresponding to each type of component in the target building structure.
The size design model can be used for rapidly and reliably completing the intelligent design of the sizes of corresponding components of the building structure according with the domain knowledge.
On the basis of the above embodiments, as an alternative embodiment, as shown in fig. 2, the neural network embedded with domain knowledge is obtained by training as follows:
constructing a convolutional neural network for extracting high-dimensional features in an input feature tensor of the building structure;
constructing a deconvolution neural network for up-sampling high-dimensional features output by the convolutional neural network to obtain a design feature tensor of a building structure member size diagram; wherein the ideal feature tensor for the building structure element dimension map is derived based on the ideal element dimension map for the building structure;
connecting the convolutional neural network and the deconvolution neural network to obtain a neural network structure to be trained;
it can be understood that the invention performs feature extraction on the input feature tensor through the convolutional neural network, abandons useless features, and determines high-dimensional features related to the size of the component, for example, 1 × 1 × 2000 high-dimensional features are extracted from the input feature tensor of 256 × 256 × 5. The deconvolution neural network generates a design feature tensor of the component dimension map, for example, a design feature tensor of 256 × 256 × 1, on the basis of the high-dimensional features. The convolution neural network and the deconvolution neural network can be selected as the U-net. The concatenated convolutional and deconvolved neural networks form a neural network that embeds domain knowledge.
Constructing a domain knowledge evaluator used for determining the loss of the domain knowledge corresponding to each type of component reflected by the design feature tensor of the building structure component size diagram based on the prestored domain knowledge corresponding to each type of component; wherein the domain knowledge corresponding to each type of component includes: theoretical equal design and theoretical unequal design under each type of component;
the domain knowledge evaluator of the embodiment introduces domain knowledge corresponding to each type of member to calculate the domain knowledge loss corresponding to each type of member reflected by the design feature tensor, and evaluates the rationality of the design feature tensor according to the domain knowledge loss condition.
Constructing an optimizer for performing parameter optimization on the neural network structure to be trained by taking the minimum comprehensive design loss of the design feature tensor of the building structure dimension diagram as a target in a training stage; wherein the image loss in the integrated design loss is determined based on an image loss function and a difference between an ideal feature tensor and a design feature tensor of a dimensional map of the architectural structural member;
in the training stage, the design conditions, the structural arrangement diagram and the corresponding component dimension diagram in the training sample are all input into the dimension design model, and the optimizer performs weighted combination on the domain knowledge loss and the image loss corresponding to each type of component to obtain comprehensive design loss; then, aiming at minimum comprehensive design loss, optimizing parameters of the neural network embedded with the domain knowledge; in the evaluation stage, the design conditions, the structural arrangement diagram and the corresponding component dimension diagram in the test sample are all input into the dimension design model, the optimizer performs weighted combination on the domain knowledge loss and the image loss corresponding to each type of component to obtain comprehensive design loss, parameters of a neural network embedded with domain knowledge are not optimized, the training effect of the network architecture is evaluated based on the comprehensive design loss, and the network architecture can be applied when the evaluation is qualified.
It can be understood that different size design models are constructed for different types of components, and the field knowledge estimator in the different size design models selects the field knowledge loss corresponding to the component type; in colloquial, corresponding domain knowledge loss is selected according to component types, and a corresponding neural network embedded with domain knowledge is trained. The present embodiment selects the optimizer as Adam with the impulse set to 0.5. Training lasts for 600 rounds.
The comprehensive design loss in the size design model is calculated by the following formula:
Loss all ＝Loss img +Loss knwl
Loss img ＝ω img ×λ img ×Loss img )
in the above formula, loss img For image Loss, loss knwl The domain knowledge loss corresponding to each type of component is composed of multiple domain knowledge losses under each type of component; loss knwl，k For the k-th field knowledge Loss under various types of components, the Loss is taken as required knwl，1 、Loss knwl，2 、Loss knwl，3 Or Loss knwl，4 ；λ img And λ knwl，k The weights of the kth domain knowledge under the image and each type of component are respectively; omega img And ω knwl，k The method is used for adjusting weight change of the k domain knowledge under the images and the various types of components in the training process.
It should be noted that the weight of the kth domain knowledge under the image and each type of component is selected by trial calculation; the weight of image loss is kept constant along with the training, and the weight of the kth domain knowledge under each type of component is increased along with the training; i.e. omega img Remains unchanged, omega knwl，k Increasing as training progresses.
For example: the weights for determining image loss and loss of knowledge in various domains under each type of component are shown in table 1 after trial calculation.
TABLE 1
Correspondingly, fig. 3 illustrates the weight change of image loss and the loss of multiple kinds of domain knowledge under each type of component in the training process, wherein a represents the weight change of image loss, and b represents the weight change of the xth theoretical equivalent design under each type of component; c represents the weight change of the z-th theoretical unequal design under each type of component.
The adjustment coefficient alpha of the image loss in the training process can be seen from the graph img The adjustment coefficient omega of the k-th domain knowledge loss under each type of component is kept constant along with the training knwl,k Increasing as training progresses.
And training and evaluating the neural network structure to be trained by utilizing a building structure sample, the domain knowledge evaluator and the optimizer to obtain the neural network embedded with the domain knowledge. Wherein the architectural structure sample comprises a structural layout drawing, a design condition and an ideal component dimension drawing of the corresponding architectural structure;
the invention relates to a data set construction process for neural network training and evaluation embedded with domain knowledge, which comprises the following steps:
the first step is as follows: design data of building structures is widely collected from engineering projects.
The building structure includes frame core tube structure, shear wall structure and frame structure etc. this embodiment takes frame core tube structure as an example to explain:
the frame core tube structure is a form of a frame shear wall structure, and is a building structure with various component types and high design difficulty.
The frame core tube structure is characterized in that a central core tube is formed by enclosing an elevator shaft, a stair, a ventilation shaft, a cable shaft, a public washroom and partial equipment rooms in the central part of a building, an outer frame inner tube structure is formed by the central core tube and a peripheral frame, and the central core tube structure is poured by reinforced concrete. The structure is very favorable for structural stress, has excellent shock resistance and is a mainstream structural form widely adopted by super high-rise buildings.
The design data of the building structure comprises a CAD structural design drawing in DWG format and design conditions in text format.
Each standard layer is provided with a corresponding design drawing and a corresponding design condition.
For a design drawing, due to the existence of a large amount of redundant information, key element extraction is required. Specifically, irrelevant information such as axes, labels, drawing frames and the like in the design drawing needs to be removed, and only the frame columns and the shear walls are reserved to obtain a structural arrangement diagram (with component category attributes) and a component dimension diagram (with section dimension attributes, considered as an ideal component dimension diagram).
The design conditions in the frame core tube structure can be further divided into three types of earthquake-resistant design conditions, wind-resistant design conditions and height design conditions, and are further divided into 4 design conditions (including earthquake influence coefficient alpha, wind influence coefficient w and total structural height h) shown in table 2 s And standard layer feature height h r )。
The design features are the same for the standard floors of the same building except for the standard floor feature height.
TABLE 2
The calculation formula of the earthquake influence coefficient alpha is as follows:
in the above formula, α max Is the maximum value of the seismic influence coefficient, T g Is a site characteristic period(s), and T is a structure basic natural vibration period;
the calculation formula of the basic natural vibration period T of the structure is as follows:
in the above formula, h s Is the structure height and b is the structure width.
As shown below, table 3 illustrates the maximum value of the seismic influence coefficient α max Table 4 illustrates the determination of the field characteristic period(s).
TABLE 3
TABLE 4
The calculation formula of the wind influence coefficient w is as follows:
w＝μ a w 0
in the above formula, h i And mu zi The length and the wind pressure height change coefficient of the ith ground clearance section are respectively, and n is the number of the ground clearance sections. The section of the floor height to be lifted is divided in advance, for example, the section of the floor height to be lifted is divided into 10 sections (0 to 10, 10 to 20, 20 to 30, etc.) at intervals of 10 meters, and 10 sections of the floor height are obtained.
Total height h of the structure s I.e. the total height of the main structure of the building above the ground.
Standard layer characteristic height h f The calculation formula of (a) is as follows:
h r ＝(h lower +h upper )/2
in the above formula, h lower And h upper Respectively is the lower elevation limit and the upper elevation limit of the standard layer.
The second step is that: and generating an architectural structure sample containing the structural arrangement diagram, the design condition and the ideal component size diagram, and constructing a training data set and a test data set by using the architectural structure sample.
FIG. 4 illustrates a network architecture containing a pre-processor, a domain knowledge embedded neural network corresponding to each type of component, a domain knowledge evaluator, an optimizer, and a sizing model of a post-processor.
FIG. 5 illustrates a schematic diagram of a sizing model training and application process, and as can be appreciated in connection with FIG. 5, the present invention first collects architectural structure samples including structural layout diagrams, design conditions, and ideal component sizing diagrams, constructs training data sets and test data sets; then, in the training stage, the neural network embedded with the domain knowledge is subjected to supervised learning by using a training data set, so that the neural network can master the domain knowledge; in the evaluation stage, the test data set is used for carrying out performance evaluation on the neural network embedded with the domain knowledge obtained in the training stage, and the network architecture qualified in evaluation can be put into application.
In the application stage, inputting a structural arrangement diagram and design conditions of the building structure to be designed into a preprocessor to obtain an input characteristic tensor of the building structure to be designed; then inputting the input feature tensor of the building structure to be designed into a neural network which is qualified in evaluation and embedded with domain knowledge, and obtaining a design feature tensor of a structural member dimension diagram of the building structure to be designed; and then inputting the design feature tensor of the structural member dimension diagram of the building to be designed into a post processor to obtain the optimal dimension design of the structural member of the corresponding type of the building to be designed in consideration of the domain knowledge, and further combining to obtain the structural member dimension diagram of the building to be designed, and further performing computer-aided deepened design based on the interaction of an engineer and structural design software (such as PKPM and ETABS).
The invention embeds the neural network of the domain knowledge, and integrates the domain knowledge corresponding to each type of component in the training process, thereby ensuring that the intelligent design of the component size conforms to the domain knowledge.
On the basis of the foregoing embodiments, as an optional embodiment, the encoding a structural layout diagram and design conditions of the target building structure, and stacking to obtain an input feature tensor of the target building structure includes:
representing a structural layout map of the target building structure as a component location matrix of the target building structure; wherein a non-zero element position in the member positioning matrix refers to the presence of a member, and a value at the non-zero element position is filled with a pending code;
replacing the to-be-determined codes in the member positioning matrix with member type codes of members at the positions by referring to the structural arrangement diagram of the target building structure and the pre-stored member type codes to obtain a member type matrix of the target building structure;
replacing the undetermined code in the member positioning matrix with any design condition value to obtain any design condition characteristic matrix of the target building structure;
normalizing and stacking all design condition feature matrices and member category matrices of the target building structure to obtain an input feature tensor of the target building structure;
and all the design condition feature matrix, the member category matrix and the member positioning matrix of the target building structure are two-dimensional feature matrices.
FIG. 6 illustrates operations performed by the pre-processor, for example, in a frame core barrel configuration, including:
1) And constructing a standardized grid, and determining the unit size and the number of the grid according to the minimum spacing of the members of the building structure and the maximum width of the structure. For this embodiment, since the maximum width of the structure is 63.0m and the minimum spacing of the members is greater than 0.25m, a standardized grid of 64m × 64m is constructed with cell sizes of 0.25m × 0.25m.
2) The structural member is positioned in the standardized grid. The building structural member is mapped to the nearest grid point/line according to the structural layout of the building structure with the building structural member center position referring to the building structural member position. For example: the middle point position of the frame column is the geometric center of the frame column, and the central position of the shear wall is the central line of the shear wall.
3) Representing a normalized grid as a component localization matrix M 0 . The grid points where the structural member exists are represented by pending codes, and otherwise, are represented by 0. Subject matter based on the foregoingThe matrix size obtained by the normalized grid is 256 × 256, and the pending code in this embodiment is represented by 1.
4) Positioning matrix M based on component 0 And component type encoding to obtain component category matrix M c . In order to facilitate the introduction of subsequent domain knowledge, the vertical members of the frame core tube structure are divided into m + n +2 types, and are represented by different codes, namely frame columns, core tube outer walls (i is more than or equal to 1 and is less than or equal to m), core tube inner walls (j is more than or equal to m +1 and is less than or equal to m + n) and other members. The direction 1 to the direction m are not parallel, the direction m +1 to the direction m + n are not parallel, m is the total number of the common directions of the outer walls of the core cylinders, and n is the total number of the common directions of the inner walls of the core cylinders. For a conventional core barrel wall arranged orthogonally in the X and Y directions, there is m = n =2.
5) Based on component positioning matrix M 0 And earthquake influence coefficient alpha, wind influence coefficient w and total structural height h s And standard layer feature height h f Respectively obtain the earthquake-resistant feature matrix M e Wind resistance feature matrix M w Overall height matrix M h And a standard layer height matrix M f 。
6) Component class matrix M c Seismic characteristic matrix M e Wind resistance feature matrix M w A total height matrix M h And a standard layer height matrix M f And respectively carrying out normalization processing, and finally stacking the matrixes in a third dimension to obtain an input feature tensor. The input feature tensor is 256 × 256 × 5.
In order to facilitate the introduction of subsequent domain knowledge, when converting the design drawing (mainly referring to the structural layout drawing) into a matrix, it should be ensured that: 1) The symmetry axis of the structure coincides with the symmetry axis of the matrix; 2) The collinear shear walls can be positioned in the same row through rotation in the matrix; 3) The vertically extending members are located at the same position in the designed matrix of the standard layer.
The method integrates the structural arrangement diagram and the design condition of the target building structure into a tensor (input feature tensor) in a standardized grid construction, digital coding and stacking mode so as to conveniently mine and analyze the tensor to generate the design feature tensor of the size diagram of the target building structure member.
On the basis of the foregoing embodiments, as an optional embodiment, decoding a design feature tensor of the target building structure component size diagram to obtain design sizes corresponding to the types of components in the target building structure, includes:
decoding the design feature tensor of the target building structure member size map according to the mapping relation between a preset building structure member size map and the design feature tensor of the building structure member size map to obtain a corresponding member size map; wherein the dimensions corresponding to each type of element in the respective element dimension map are the mean of all elements corresponding to each type of element in the tensor of design features of the target architectural structural element dimension map;
and merging the sizes corresponding to the various types of components in the corresponding component size diagram and a plurality of pre-stored size modules corresponding to the various types of components by adopting a recent principle to obtain the design sizes corresponding to the various types of components in the target building structure.
FIG. 7 illustrates operations performed by the post-processor, for example, in a frame core barrel configuration, including:
(1) Defining the mapping relationship between the building structure element dimension map and the design feature tensor for the building structure element dimension map is equivalent to defining the feature dimension of the element and the element dimension matrix M s The mapping relationship between the codes (design feature tensors) of (c); the mapping relationship is as follows:
a c ＝s c /(10mm)
in the above formula, a c Is a component size matrix M s Code of (a), s c Is a characteristic dimension (mm) of the structural member;
in the above formula, t w Is the shear wall thickness, A c Is the cross-sectional area of the frame post.
(2) Based on component class matrix M c And carrying out bidirectional mapping. In the training phase, the post-processor maps the ideal component dimension map into a component dimension matrix M s Regularization is then used as the supervised trained label tensor. In the training and application stage, the post-processor inversely regularizes the design feature tensor into a component size matrix M s Then, respectively mapping reversely to obtain corresponding component dimension maps;
(3) And merging the sections. Assume a component dimension matrix M s Derived by a post-processor of the frame post sizing model, the post-processor then maps the component size matrix M s Taking the average value of all elements representing the frame column as the initial section size of the frame column; according to a preset frame column section size modulus, merging the initial section size of the frame column into the final section size of the frame column; the modulus unit of the section size of the shear wall is 20mm, and the modulus unit of the section size of the frame column is 50mm. For example, the initial section size of a certain frame column is 420mm, and the section size modes preset for the frame column are 350mm,400mm,450mm and the like; the final cross-sectional dimension of the frame post is 400mm, merged according to the most recent principles.
The invention ensures that the size design result of each type of member of the target building structure conforms to the convention, thereby providing convenience for construction and construction.
On the basis of the foregoing embodiments, as an alternative embodiment, the determining, based on the prestored domain knowledge corresponding to each type of component, the loss of the domain knowledge corresponding to each type of component reflected by the design feature tensor of the dimension map of the architectural structural component includes:
determining building structure positioning masks corresponding to various types of components based on the component category matrix of the building structure; in the building structure positioning mask code, the position corresponding to each type of component is marked as 1, otherwise, the position is marked as 0;
based on the component category matrix, the invention can obtain the building structure positioning mask corresponding to various components: for example: frame column positioning Mask C The positioning Mask of the core tube outer wall in the ith direction WE,i (1 ≦ i ≦ m), and the positioning of the inner wall of the core barrel in the jth directionBit Mask WI,j (1 ≦ j ≦ n); taking the frame column positioning mask as an example, a position mark 1 of the frame column exists, and a position mark 0 of the frame column does not exist;
determining the loss of the design feature tensor of the building structure member size diagram relative to each theoretically equal design and each theoretically unequal design under each type of members based on the building structure positioning mask, the design feature tensor of the building structure member size diagram and the loss calculation formula corresponding to each theoretically equal design and each theoretically unequal design under each type of members;
the method comprises the steps of setting a loss calculation formula corresponding to each theoretical equal design and each theoretical unequal design under each type of member, and carrying out tensor calculation on a positioning mask corresponding to each type of member and a design feature tensor Pred of a building structure member size diagram by using the loss calculation formula so as to obtain the loss of Pred relative to each theoretical equal design and each theoretical unequal design under each type of member;
and determining the loss of the domain knowledge corresponding to each type of member reflected by the design feature tensor of the building structure member size diagram by utilizing the loss of the design feature tensor of the building structure member size diagram relative to each theoretically equivalent design and each theoretically unequal design under each type of member and the dynamic weight corresponding to each theoretically equivalent design and each theoretically unequal design under each type of member.
The invention gives the loss calculation formula corresponding to each theoretical equal design and each theoretical unequal design under each type of component, thereby conveniently calculating the loss of Pred corresponding to each theoretical equal design and each theoretical unequal design under each type of component by using the positioning mask and Pred corresponding to each type of component, and providing an implementation mode for calculating the domain knowledge loss of Pred corresponding to each type of component.
On the basis of the above embodiments, as an alternative embodiment, the earthquake-resistant design condition includes an earthquake influence coefficient; the wind resistance design condition comprises a wind influence coefficient; the height design condition comprises the total height of the building structure and the characteristic height of a standard layer of the building structure;
the various types of components comprise: frame column, support, shear wall, core tube outer wall and core tube inner wall type components;
for a theoretical equivalent design, the theoretical equivalent design of the frame column/support comprises: symmetrically arranged frame posts/supports have the same cross-sectional dimensions;
the theoretical equal design of a core barrel outer wall comprises: the outer walls of the core barrels arranged in the same direction have the same cross-sectional size;
the theoretical equal design of the core tube inner wall/shear wall comprises the following steps: the core tube inner walls/shear walls which are arranged in a collinear way have the same cross-sectional dimension;
for the theoretical unequal design, the theoretical unequal design of the frame column, the core tube outer wall, the core tube inner wall or the shear wall comprises the following steps: the section size of the frame column, the core tube outer wall, the core tube inner wall or the shear wall on the higher standard layer is not larger than that on the lower standard layer;
the design that the theory of shear force wall varies still includes: the cross-sectional dimension of the shear wall arranged at the periphery should be not smaller than the cross-sectional dimension of the shear wall arranged at the inside.
This example illustrates the domain knowledge of conventional architectural structural members (theoretically equivalent design and theoretically unequal design) in order to better understand the present invention.
On the basis of the above embodiments, as an alternative embodiment, the loss calculation formula corresponding to "symmetrically arranged frame columns have the same cross-sectional dimension" is designed by the frame column theory equality, and specifically is as follows:
Loss knwl，1 ＝Loss C，X +Loss C，Y
Pred d，lr ＝Pred r -Pred l，f
Pred l，f ＝Fliplr(Pred l )
in the above formula, loss knwl，1 Design feature tensor representing dimensional graph of architectural structural member with respect to theoretical equal design of frame column "symmetrically arranged frame columns have same section size" weidu Loss, loss C，X And Loss C，Y The symmetry loss of the frame columns corresponding to the X symmetry axis and the Y symmetry axis respectively,
and
are respectively Mask C Left and right half tensors of, pred l And Pred r Left and right half tensors, mask, of Pred, respectively C Positioning masks corresponding to frame columns in the building structure, pred is a design feature tensor of a building structure member dimension graph, and Sum is carried out on all elements in the tensor; abs (·) takes absolute values of tensor elements; when the number of the tensor is equal to the number of the Hadamard product, the tensor is inverted left and right by the Fliplr (·); FIG. 8 illustrates a matrix orientation definition diagram, and it can be seen that Loss C，X And Loss C，Y Are calculated in an equivalent manner.
The core tube outer wall theory equally designs a loss calculation formula corresponding to the fact that the core tube outer walls arranged in the same direction have the same section size, and the loss calculation formula is as follows:
Pred WE，i ＝Pred⊙Mask WE，i
in the above formula, loss knwl，2 Loss of "co-located core tube outer walls having the same cross-sectional dimension" for theoretical equal design feature tensor for building structural element dimensional map versus core tube outer wall, loss WE，i Is the loss of the core tube outer wall in the ith direction, m is the total number of the directions shared by the core tube outer walls, mask WE，i A positioning mask code of the outer wall of the core barrel in the ith direction in the building structure;
the core tube inner wall theory is designed with equal loss calculation formula corresponding to the 'collinear core tube inner walls with same section size', and the formula is as follows:
Pred WIX，j ＝Pred X，j ⊙Mask WIX，j
Pred X，j ＝Rotate(Pred，θ j )
Mask WIX，j ＝Rotate(Mask WI，j ，θ j )
in the above formula, loss knwl，3 Loss of 'collinear arranged core tube interior walls having the same cross-sectional dimension', mask for theoretical equal design feature tensor for building structural element dimensional map versus core tube interior wall WI，j Is the loss of the inner wall of the core tube in the jth direction, n is the total number of the directions shared by the inner walls of the core tube, rotate (·, theta) is the counterclockwise rotation of the tensor by theta degrees, theta j Is a counterclockwise included angle between the inner wall of the core tube in the jth direction and the X direction; element-by-element division of the tensor, sumX (-) sums the tensor along the X direction, and epsilon is a small quantity to avoid 0;
the frame column, the core tube outer wall, the core tube inner wall or the shear wall are designed in different theories
The loss calculation formula corresponding to the section size of the frame column, the core tube outer wall, the core tube inner wall or the shear wall on the higher standard layer is not larger than the section size of the frame column, the core tube outer wall, the core tube inner wall or the shear wall on the lower standard layer is as follows:
Loss knwl，4 ＝Loss HM +Loss ML
Loss HM ＝Sum(ReLU(Pred H -Pred M ))
Loss ML ＝Sum(ReLU(Pred M -Pred L ))
in the above formula, reLU (. Cndot.) is the activation function, x is the argument, pred H 、Pred M And Pred L Respectively representing the previous layer of the current standard layer, the next layer of the current standard layer and the current standard layer, lost knwl，4 The loss of designing the 'section size of the frame column, the core tube outer wall, the core tube inner wall or the shear wall on the higher standard layer is not larger than the section size on the lower standard layer' for the theoretical unequal design of the design feature tensor of the building structural member dimension map relative to the frame column, the core tube outer wall, the core tube inner wall or the shear wall.
The present invention compares building structural member sizing embedded with domain knowledge to engineer member sizing.
The dimensional design result of the building structural member embedded with the domain knowledge is very close to the design result of an engineer and meets the domain knowledge. A comparison of the building structural member dimensional design embedded with domain knowledge and the engineer's design in terms of material usage and time consumption is shown in table 5. In terms of concrete usage, the building structure member size design embedding domain knowledge uses 3.18% less on frame columns, 1.28% less on shear walls, and 2.06% less on the total of the two, compared to the engineer's design. It can be seen that the building structural member dimensioning embedded with domain knowledge is more material-saving. Furthermore, a skilled engineer typically takes more than 30 minutes to design a cross-sectional dimension once, whereas a building structural member incorporating domain knowledge typically takes only 12 seconds to design a cross-sectional dimension, which increases efficiency by 150 times.
TABLE 5
In a second aspect, a field knowledge embedded building structural member size designing apparatus according to the present invention is described, and the field knowledge embedded building structural member size designing apparatus described below and the field knowledge embedded building structural member size designing method described above are referred to in correspondence with each other. Fig. 9 illustrates a schematic structural view of a domain knowledge embedded building structure member sizing device, as shown in fig. 9, comprising: an input module 21 and a generation module 22; the substituting module 21 is configured to input the structural layout drawing and the design condition of the target building structure into a pre-stored size design model corresponding to each type of component, respectively, to obtain a design size of each type of component in the target building structure; a generating module 22 for generating a component dimension map of the target building structure based on the design dimensions of all types of components in the target building structure; the size design model is optimally trained by taking the minimum comprehensive design loss of the design characteristic tensor of the building structural member size diagram as a target; the comprehensive design loss is a weighted sum of image loss and field knowledge loss corresponding to each type of component; the image loss is the difference between the corresponding design feature tensor and the corresponding ideal feature tensor; the design conditions comprise: seismic design conditions, wind resistance design conditions and height design conditions.
The device for designing a size of a building structural element with embedded knowledge according to the embodiments of the present invention specifically executes the process of the method for designing a size of a building structural element with embedded knowledge in each field, and for details, please refer to the contents of the method for designing a size of a building structural element with embedded knowledge in each field, which are not described herein again.
The invention provides a building structure component size design device embedded with domain knowledge, which is characterized in that a size design model taking the minimum weighted combination between image loss and domain knowledge loss of corresponding types of components as an optimization target is trained aiming at different types of components in a building structure in advance, the difference between a design characteristic tensor corresponding to a component size diagram and an ideal characteristic tensor is taken as the image loss in the model, the component size diagram mapped by the design characteristic tensor is ensured to be proper by introducing the image loss, and the size design of the corresponding types of components is ensured to be optimal by introducing the domain knowledge loss of the corresponding types of components. Then, respectively inputting the structural arrangement drawing and the design condition of the target building structure into a size design model corresponding to each type of component to obtain the design size of each type of component in the target building structure; generating a component dimension map for the target building structure based on the design dimensions of all types of components in the target building structure; the invention can quickly and reliably complete the intelligent design of the size of the building structural member which accords with the knowledge of the related field of structural design.
In a third aspect, fig. 10 illustrates a schematic physical structure diagram of an electronic device, and as shown in fig. 10, the electronic device may include: a processor (processor) 1010, a communication Interface (Communications Interface) 1020, a memory (memory) 1030, and a communication bus 1040, wherein the processor 1010, the communication Interface 1020, and the memory 1030 communicate with each other via the communication bus 1040. Processor 1010 may invoke logic instructions in memory 1030 to perform a domain knowledge embedded architectural structural member sizing method comprising: respectively inputting the structural arrangement diagram and the design condition of the target building structure into a pre-stored size design model corresponding to each type of component to obtain the design size of each type of component in the target building structure; generating a component dimension map for the target building structure based on the design dimensions of all types of components in the target building structure; the size design model is optimally trained by taking the minimum comprehensive design loss of the design characteristic tensor of the building structural member size diagram as a target; the comprehensive design loss is a weighted sum of image loss and field knowledge loss corresponding to each type of component; the image loss is the difference between the corresponding design feature tensor and the corresponding ideal feature tensor; the design conditions comprise: seismic design conditions, wind resistance design conditions and height design conditions.
Furthermore, the logic instructions in the memory 1030 can be implemented in software functional units and stored in a computer readable storage medium when the logic instructions are sold or used as independent products. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk, and various media capable of storing program codes.
In a fourth aspect, the present invention also provides a computer program product comprising a computer program, storable on a non-transitory computer readable storage medium, which when executed by a processor, performs a domain knowledge embedded architectural structural member sizing method, the method comprising: respectively inputting a structural arrangement diagram and design conditions of a target building structure into a pre-stored size design model corresponding to each type of component to obtain the design size of each type of component in the target building structure; generating a component dimension map for the target building structure based on the design dimensions of all types of components in the target building structure; the size design model is optimally trained by taking the minimum comprehensive design loss of the design characteristic tensor of the building structural member size diagram as a target; the comprehensive design loss is a weighted sum of image loss and field knowledge loss corresponding to each type of component; the image loss is the difference between the corresponding design feature tensor and the corresponding ideal feature tensor; the design conditions comprise: seismic design conditions, wind resistance design conditions and height design conditions.
In a fifth aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program to perform a method of building structural member sizing embedded with domain knowledge, the method comprising: respectively inputting the structural arrangement diagram and the design condition of the target building structure into a pre-stored size design model corresponding to each type of component to obtain the design size of each type of component in the target building structure; generating a component dimension map for the target building structure based on the design dimensions of all types of components in the target building structure; the size design model is optimally trained by taking the minimum comprehensive design loss of the design characteristic tensor of the building structural member size diagram as a target; the comprehensive design loss is a weighted sum of image loss and field knowledge loss corresponding to each type of component; the image loss is the difference between the corresponding design feature tensor and the corresponding ideal feature tensor; the design conditions comprise: seismic design conditions, wind resistance design conditions and height design conditions.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A domain knowledge embedded building structural member sizing method, the method comprising:
respectively inputting a structural arrangement diagram and design conditions of a target building structure into a pre-stored size design model corresponding to each type of component to obtain the design size of each type of component in the target building structure;
generating a component dimension map for the target building structure based on the design dimensions of all types of components in the target building structure;
the size design model is optimally trained by taking the minimum comprehensive design loss of the design characteristic tensor of the building structural member size diagram as a target;
the comprehensive design loss is a weighted sum of image loss and field knowledge loss corresponding to each type of component;
the image loss is the difference between the corresponding design feature tensor and the corresponding ideal feature tensor;
the design conditions comprise: seismic design conditions, wind resistance design conditions and height design conditions.
2. The domain knowledge embedded architectural structural member sizing method of claim 1, wherein the sizing model comprises a pre-processor, a domain knowledge embedded neural network corresponding to each type of member, and a post-processor; the step of respectively inputting the structural arrangement diagram and the design conditions of the target building structure into a pre-stored size design model corresponding to each type of component to obtain the design size of each type of component in the target building structure comprises the following steps:
based on a pre-processor, coding a structural arrangement diagram and design conditions of the target building structure, and stacking to obtain an input feature tensor of the target building structure;
extracting high-dimensional features in the input feature tensor based on a neural network of embedded domain knowledge corresponding to each type of component, and performing up-sampling on the high-dimensional features to obtain a design feature tensor of the target building structure component size graph;
and decoding the design feature tensor of the target building structure member size graph based on a post processor to obtain the design size corresponding to each type of member in the target building structure.
3. The method of claim 2, wherein the domain knowledge embedded neural network is trained to be obtained by:
constructing a convolutional neural network for extracting high-dimensional features in an input feature tensor of the building structure;
constructing a deconvolution neural network for up-sampling high-dimensional features output by the convolutional neural network to obtain a design feature tensor of a building structure member size diagram; wherein the ideal feature tensor for the building structure element dimension map is derived based on the ideal element dimension map for the building structure;
connecting the convolution neural network with a deconvolution neural network to obtain a neural network structure to be trained;
constructing a domain knowledge evaluator used for determining the domain knowledge loss corresponding to each type of member reflected by the design feature tensor of the building structure member size diagram based on the prestored domain knowledge corresponding to each type of member; wherein the domain knowledge corresponding to each type of component comprises: theoretical equal design and theoretical unequal design under each type of component;
constructing an optimizer for performing parameter optimization on the neural network structure to be trained by taking the minimum comprehensive design loss of the design feature tensor of the building structure dimension diagram as a target in a training stage; wherein the image loss in the integrated design loss is determined based on an image loss function and a difference between an ideal feature tensor and a design feature tensor of a dimensional map of the architectural structural member;
and training and evaluating the neural network structure to be trained by utilizing a building structure sample, the domain knowledge evaluator and the optimizer to obtain the neural network embedded with the domain knowledge.
4. The method of claim 2, wherein the encoding the structural layout drawing and the design condition of the target building structure and stacking the encoded structural layout drawing and design condition to obtain the input feature tensor of the target building structure comprises:
representing a structural layout map of the target building structure as a component location matrix of the target building structure; wherein a non-zero element position in the member positioning matrix refers to the presence of a member, and a value at the non-zero element position is filled with a pending code;
replacing the to-be-determined codes in the member positioning matrix with member type codes of members at the positions by referring to the structural arrangement diagram of the target building structure and the pre-stored member type codes to obtain a member type matrix of the target building structure;
replacing the undetermined code in the member positioning matrix with any design condition value to obtain any design condition characteristic matrix of the target building structure;
normalizing and stacking all design condition feature matrices and member category matrices of the target building structure to obtain an input feature tensor of the target building structure;
and all the design condition characteristic matrix, the component category matrix and the component positioning matrix of the target building structure are two-dimensional characteristic matrices.
5. The method of designing dimensions of building structural members with embedded domain knowledge according to claim 2, wherein the decoding of the design feature tensor of the target building structural member dimension map to obtain the design dimensions corresponding to each type of member in the target building structure comprises:
decoding the design feature tensor of the target building structure member size diagram according to the mapping relation between a preset building structure member size diagram and the design feature tensor of the building structure member size diagram to obtain a corresponding member size diagram; wherein the dimension corresponding to each type of member in the corresponding member dimension map is a mean of all elements corresponding to each type of member in a design feature tensor of the target architectural structure member dimension map;
and merging the sizes corresponding to the various types of components in the corresponding component size diagram and a plurality of pre-stored size modules corresponding to the various types of components by adopting a recent principle to obtain the design sizes corresponding to the various types of components in the target building structure.
6. The method of claim 3, wherein the determining loss of domain knowledge corresponding to each type of member reflected by the tensor of design features of the dimensional graph of the architectural structural member based on the pre-stored domain knowledge corresponding to each type of member comprises:
determining building structure positioning masks corresponding to various types of components based on the component category matrix of the building structure; in the building structure positioning mask, the position corresponding to each type of component is marked as 1, otherwise, the position is marked as 0;
determining the loss of the design feature tensor of the building structure member size diagram relative to each theoretically equal design and each theoretically unequal design under each type of members based on the building structure positioning mask, the design feature tensor of the building structure member size diagram and the loss calculation formula corresponding to each theoretically equal design and each theoretically unequal design under each type of members;
and determining the loss of the domain knowledge corresponding to each type of member reflected by the design feature tensor of the building structure member size diagram by utilizing the loss of the design feature tensor of the building structure member size diagram relative to each theoretically equivalent design and each theoretically unequal design under each type of member and the dynamic weight corresponding to each theoretically equivalent design and each theoretically unequal design under each type of member.
7. The domain knowledge embedded architectural structural member dimensional design method of claim 6, wherein said seismic design conditions comprise a seismic influence coefficient; the wind resistance design condition comprises a wind influence coefficient; the height design condition comprises the total height of the building structure and the characteristic height of a standard layer of the building structure;
the various types of components comprise: frame column, support, shear wall, core tube outer wall and core tube inner wall type components;
for a theoretical equivalent design, the theoretical equivalent design of the frame column/support comprises: symmetrically arranged frame posts/supports have the same cross-sectional dimensions;
the theoretical equal design of core section of thick bamboo outer wall includes: the outer walls of the core barrels arranged in the same direction have the same cross-sectional size;
the theoretical equal design of the core tube inner wall/shear wall comprises the following steps: the core tube inner walls/shear walls which are arranged in a collinear way have the same cross-sectional dimension;
for the theoretical unequal design, the theoretical unequal design of the frame column, the core tube outer wall, the core tube inner wall or the shear wall comprises the following steps: the section size of the frame column, the core tube outer wall, the core tube inner wall or the shear wall on the higher standard layer is not larger than the section size of the frame column, the core tube outer wall, the core tube inner wall or the shear wall on the lower standard layer;
the design that the theory of shear force wall varies still includes: the cross-sectional dimension of the shear wall arranged at the periphery should be not smaller than the cross-sectional dimension of the shear wall arranged at the inside.
8. The method of designing dimensions of building structural members with embedded knowledge in the art according to claim 7, wherein the frame columns are theoretically equally designed with a formula of calculating loss corresponding to "symmetrically arranged frame columns have the same sectional dimension" as follows:
Loss knwl，1 ＝Loss C，X +Loss C，Y
Pred d，lr ＝Pred r -Pred l，f
Pred l，f ＝Fliplr(Pred l )
in the above formula, loss knwl，1 Loss of design feature tensor representing dimensional graph of architectural structural member "symmetrically arranged frame columns have the same cross-sectional dimension" relative to theoretically equivalent design of frame columns C，X And Loss C，Y The symmetry loss of the frame columns corresponding to the X symmetry axis and the Y symmetry axis respectively,
seed of a plant
Are respectively Mask C Left and right half tensors of, pred l And Pred r Left and right half tensors, mask, of Pred, respectively C Positioning masks corresponding to frame columns in the building structure, pred is a design feature tensor of a building structure member dimension graph, and Sum is carried out on all elements in the tensor; abs (·) takes absolute values of tensor elements; when the number of the tensor is equal to the number of the Hadamard product, the tensor is inverted left and right by the Fliplr (·);
the core tube outer wall theory equal design 'the core tube outer walls arranged in the same direction have the same section size' corresponding loss calculation formula, which is concretely as follows:
Pred WE，i ＝Pred⊙Mask WE，i
in the above formula, loss knwl，2 Loss of "co-located core tube outer walls having the same cross-sectional dimension" for theoretical equal design feature tensor for building structural element dimensional map versus core tube outer wall, loss WE，i For the core tube outer wall at the ith squareUpward loss, m is the total number of directions shared by the outer walls of the core barrel, mask WE，i A positioning mask code of the outer wall of the core barrel in the ith direction in the building structure;
the core tube inner wall theory is designed with equal loss calculation formula corresponding to the 'collinear core tube inner walls with same section size', and the formula is as follows:
Pred WIX，j ＝Pred X，j ⊙Mask WIX，j
Pred X，j ＝Rotate(Pred，θ j )
Mask WIX，j ＝Rotate(Mask WI，j ，θ j )
in the above formula, loss knwl，3 Designing "collinear arrangements" for theoretical equality of design feature tensor for dimensional maps of building structural members relative to core barrel interior wallsCore barrel inner wall having equal cross-sectional dimension loss, mask WI，j Is the loss of the inner wall of the core tube in the jth direction, n is the total number of the common directions of the inner walls of the core tube, and Rotate (·, theta) is the counterclockwise rotation of the tensor by theta degrees j Is a counterclockwise included angle between the inner wall of the core tube in the jth direction and the X direction; element-by-element division of tensor, sumX (·) sums tensors along X direction, and epsilon is a small quantity, avoiding 0;
the theory of the frame column, the core tube outer wall, the core tube inner wall or the shear wall is not equal to design a loss calculation formula corresponding to that the section size of the frame column, the core tube outer wall, the core tube inner wall or the shear wall on a higher standard layer is not larger than that on a lower standard layer, and the loss calculation formula is as follows:
Loss knwl，4 ＝Loss HM +Loss ML
Liss HM ＝Sum(ReLU(Pred H -Pred M ))
Loss ML ＝Sum(ReLU(Pred M -Pred L ))
in the above formula, reLU (. Cndot.) is the activation function,
x is an independent variable, pred H 、Pred M And Pred L Respectively representing the upper layer of the current standard layer, the lower layer of the current standard layer and the current standard layer, loss knwl，4 The loss of designing the 'section size of the frame column, the core tube outer wall, the core tube inner wall or the shear wall on the higher standard layer is not larger than the section size on the lower standard layer' for the theoretical unequal design of the design feature tensor of the building structural member dimension map relative to the frame column, the core tube outer wall, the core tube inner wall or the shear wall.
9. An area-of-knowledge embedded building structural member sizing device, the device comprising:
the input module is used for respectively inputting the structural arrangement drawing and the design condition of the target building structure into a pre-stored size design model corresponding to each type of component to obtain the design size of each type of component in the target building structure;
a generating module for generating a component dimension map of the target building structure based on the design dimensions of all types of components in the target building structure;
the size design model is optimally trained by taking the minimum comprehensive design loss of the design characteristic tensor of the building structural member size diagram as a target;
the comprehensive design loss is a weighted sum of image loss and field knowledge loss corresponding to each type of component;
the image loss is the difference between the corresponding design feature tensor and the corresponding ideal feature tensor;
the design conditions comprise: seismic design conditions, wind resistance design conditions and height design conditions.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor, when executing the program, implements the domain knowledge embedded architectural structural member sizing method of any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210444469.9A CN114880741B (en) | 2022-04-25 | 2022-04-25 | Building structure component size design method and device embedded with domain knowledge |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210444469.9A CN114880741B (en) | 2022-04-25 | 2022-04-25 | Building structure component size design method and device embedded with domain knowledge |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114880741A CN114880741A (en) | 2022-08-09 |
| CN114880741B true CN114880741B (en) | 2023-01-31 |
Family
ID=82671744
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210444469.9A Active CN114880741B (en) | 2022-04-25 | 2022-04-25 | Building structure component size design method and device embedded with domain knowledge |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114880741B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116150839B (en) * | 2022-12-16 | 2023-08-29 | 清华大学 | Steel structure factory building component arrangement design method and system based on field knowledge enhancement |
| CN116680778B (en) * | 2023-04-27 | 2024-03-12 | 清华大学 | Building structure arrangement generation method and system and construction method of generation model |
| CN116226674B (en) * | 2023-05-06 | 2023-09-05 | 中国建筑西南设计研究院有限公司 | Layout model training method, layout method and device for frame beams |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107391804A (en) * | 2017-06-27 | 2017-11-24 | 广东省建筑设计研究院 | Aseismic Behavior of Tall optimization method based on comprehensive cost method |
| CN111339592A (en) * | 2020-02-26 | 2020-06-26 | 广东众图互联网工程设计有限公司 | Automatic design method, device and system for components of building structure model and server |
| CN111666617A (en) * | 2020-05-25 | 2020-09-15 | 清华大学 | Shear wall structure arrangement method and device based on countermeasure generation network |
| CN112966760A (en) * | 2021-03-15 | 2021-06-15 | 清华大学 | Neural network fusing text and image data and design method of building structure thereof |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9830681B2 (en) * | 2014-01-31 | 2017-11-28 | Hover Inc. | Multi-dimensional model dimensioning and scale error correction |
-
2022
- 2022-04-25 CN CN202210444469.9A patent/CN114880741B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107391804A (en) * | 2017-06-27 | 2017-11-24 | 广东省建筑设计研究院 | Aseismic Behavior of Tall optimization method based on comprehensive cost method |
| CN111339592A (en) * | 2020-02-26 | 2020-06-26 | 广东众图互联网工程设计有限公司 | Automatic design method, device and system for components of building structure model and server |
| CN111666617A (en) * | 2020-05-25 | 2020-09-15 | 清华大学 | Shear wall structure arrangement method and device based on countermeasure generation network |
| CN112966760A (en) * | 2021-03-15 | 2021-06-15 | 清华大学 | Neural network fusing text and image data and design method of building structure thereof |
Non-Patent Citations (4)
| Title |
|---|
| Automated Structural Design of Shear Wall Residential Buildings Using GAN-Based Machine Learning Algorithm;Xinzheng Lu;《Research Square》;20211201;全文 * |
| Machine learning applications for building structural design and performance assessment: State-of-the-art review;Han Sun;《Journal of Building Engineering》;20200921;全文 * |
| 基于失效模式的SRC框架-RC核心筒混合结构三水准抗震优化设计;李志强等;《振动与冲击》;20131015(第19期);全文 * |
| 建筑结构主体与建筑部件公差配合研究;王文营等;《城市住宅》;20190525(第05期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114880741A (en) | 2022-08-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114880741B (en) | Building structure component size design method and device embedded with domain knowledge | |
| Amirebrahimi et al. | A BIM-GIS integration method in support of the assessment and 3D visualisation of flood damage to a building | |
| CN113779675B (en) | Physical-data driven intelligent shear wall building structure design method and device | |
| Zhang et al. | Shear wall layout optimization for conceptual design of tall buildings | |
| Lu et al. | Intelligent structural design of shear wall residence using physics‐enhanced generative adversarial networks | |
| CN114491740B (en) | Physical and multi-modal data driven building structure generation method and equipment | |
| Li et al. | Sensitivity and reliability analysis of a self-anchored suspension bridge | |
| CN114417464B (en) | Multi-modal data-driven building structural member generation method and device | |
| Mirrashid et al. | Innovative computational intelligence-based model for vulnerability assessment of RC frames subject to seismic sequence | |
| Su et al. | Shape generation of grid structures by inverse hanging method coupled with multiobjective optimization | |
| CN108984932A (en) | Building structure characteristic rapid extracting method and corresponding loss appraisal calculation method | |
| CN114741766B (en) | Building structure design method and device based on joint structure optimization neural network | |
| Lin et al. | Cluster computing‐aided model updating for a high‐fidelity finite element model of a long‐span cable‐stayed bridge | |
| CN115775116B (en) | BIM-based road and bridge engineering management method and system | |
| Aslay et al. | 3D cost optimization of 3 story RC constructional building using Jaya algorithm | |
| Wdowicki et al. | Analysis of shear wall structures of variable cross section | |
| CN111582634A (en) | Multi-factor safety grading method and system for underground large-space construction | |
| Quinay et al. | Development of simulation-based approach using frame models generated from GIS features and BIM data for application to city seismic response analysis of low-to mid-rise RC structures in Metro Manila | |
| Zhai et al. | A Seismic Hazard Prediction System for Urban Buildings Based on Time-History Analysis | |
| CN114792020B (en) | Quick evaluation method and system for building anti-seismic toughness based on machine learning | |
| Hu et al. | A generalized method for estimating drifts and drift components of tall buildings under lateral loading | |
| He et al. | Correlation between global damage and local damage of RC frame structures under strong earthquakes | |
| Zegard | Structural optimization: from continuum and ground structures to additive manufacturing | |
| Pizarro et al. | Simplified shear wall building model for design optimization | |
| KR102034868B1 (en) | Numerical method for the coupled analysus between the ground deformation analysis due to excavation and groundwater flow analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |