JP7563990B2 - Frame shape estimation device and frame shape estimation method - Google Patents

Description

The present invention relates to a frame shape estimation device and a frame shape estimation method.

In the past, in the structural design of buildings, the cross-section settings and approximate work-rates of the building were determined while incorporating engineering judgment by referring to the design data of existing buildings with similar frame shapes. The technology described in Patent Document 1 is known as a technology for supporting building design.

JP 2019-117603 A

However, similarity judgments to determine whether the structural shapes are similar were performed, for example, by comparing a three-dimensional drawing of a building created based on component data, etc., with a three-dimensional drawing of an existing building to determine whether they were visually similar. This required creating a three-dimensional drawing of the building under construction in order to perform the similarity judgment, which was a time-consuming process.

The present invention was made to solve the above problems, and its purpose is to provide a frame shape estimation device and frame shape estimation method that can determine whether the frame shapes of buildings are similar or not without creating a three-dimensional drawing of the building.

In order to solve the above problem, one aspect of the present invention is a frame shape estimation device that estimates the frame shape of a building based on joint information indicating the state of joints at joints that join the columns and beams of the building, and is characterized by comprising: a storage unit that stores model information indicating a frame shape estimation model constructed based on learning data having a plurality of sets of data in which the joint information and the frame shape of the building are associated; and a frame shape estimation unit that estimates the frame shape of the target building based on an output result obtained by inputting the joint information of a target building to be estimated into the frame shape estimation model stored in the storage unit.

Another aspect of the present invention is a frame shape estimation method for a frame shape estimation device that includes a frame shape estimation model that estimates the frame shape of a building based on joint information indicating the state of joints at joints that join columns and beams of the building, the frame shape estimation method including a memory unit that stores model information indicating a frame shape estimation model constructed based on learning data having a plurality of sets of data including at least the joint information and the frame shape of the building, the frame shape estimation method including an estimation step of estimating the frame shape of the target building based on an output result obtained by inputting the joint information of a target building to be estimated into the frame shape estimation model stored in the memory unit.

According to the present invention, it is possible to determine whether the structural shapes of buildings are similar without creating a three-dimensional drawing of the building.

FIG. 1 is a block diagram showing an example of a frame shape estimation device according to an embodiment. FIG. 4 is a diagram illustrating bonding information according to an embodiment. FIG. 4 is a diagram illustrating bonding information according to an embodiment. FIG. 4 is a diagram illustrating bonding information according to an embodiment. FIG. 4 is a diagram illustrating bonding information according to an embodiment. FIG. 11 is a diagram illustrating building connection information in the embodiment. FIG. 11 is a diagram illustrating building connection information in the embodiment. FIG. 10 is a diagram illustrating learning data according to an embodiment. 11 is a diagram illustrating an example in which joint information 134 is used as learning data for a neural network according to an embodiment. FIG. FIG. 2 is a diagram illustrating a frame shape estimation model in the embodiment. FIG. 2 is a diagram illustrating a frame shape estimation model in the embodiment. 11A to 11C are diagrams illustrating a process of estimating a frame shape in the embodiment. 11A to 11C are diagrams illustrating a process of estimating a frame shape in the embodiment. 1 is a flowchart showing a flow of processing performed by a frame shape estimation device according to an embodiment. 1 is a flowchart showing a flow of processing performed by a frame shape estimation device according to an embodiment.

The following describes a frame shape estimation device and a frame shape estimation method according to an embodiment of the present invention with reference to the drawings.

Fig. 1 is a block diagram showing an example of a frame shape estimation device 1 according to an embodiment. As shown in Fig. 1, the frame shape estimation device 1 is a computer device such as a personal computer or a server device. The frame shape estimation device 1 acquires the skeletal form of a building, for example, the state of a joint at a joint between a column and a beam, as input data, estimates the frame shape of the building that is most similar to the input data, and outputs the frame shape. The frame shape estimation device 1 includes, for example, an input unit 11, a display unit 12, a storage unit 13, and a control unit 14.

The input unit 11 is, for example, an input device such as a keyboard or a mouse, and receives various information such as input data and outputs it to the control unit 14. The display unit 12 is, for example, a display device such as a liquid crystal display, and displays various information such as the structural shape of a building, which is output data (output information). The display unit 12 is an example of an output unit that outputs output information.

The storage unit 13 stores various information used by the frame shape estimation device 1. The storage unit 13 stores learning data 131, model information 132, input information 133, connection information 134, and frame shape type 135.

The learning data 131 is learning data for constructing a structural shape estimation model, which will be described later. The learning data 131 is, for example, learning data having multiple sets of data that associate connection information 134 with structural shape types 135 in buildings that have been constructed in the past.

The model information 132 is information indicating a frame shape estimation model that estimates the frame shape of a building. The frame shape estimation model is a model that estimates the type of frame shape from the building's joint information 134. The model information 132 is a model that learns the relationship between the building's joint information 134 and the frame shape type 135 based on the learning data 131 stored in the storage unit 13 described above. The frame shape estimation model estimates the type of the building's frame shape from the building's joint information 134 that is the estimation target, and outputs a label (type information) that indicates the estimated type.

The input information 133 is information that is input to the frame shape estimation model when estimating the frame shape, and in this case, it is information indicating whether or not there is a column and beam connection for each joint in the building to be estimated.

The joint information 134 is information that indicates whether or not columns and beams are connected at the joints of the building. The specific contents of the joint information 134 will be explained in detail later.

The structure shape type 135 is information indicating the type of structure shape in a building. The structure shape type 135 is, for example, a database of structure shapes, and is information in which the above-mentioned labels (type information) are associated with structure shapes. The structure shapes in the structure shape type 135 are, for example, structure shapes used to construct a structure shape estimation model, structure shapes of buildings designed in the past, etc. The structure shape type 135 may be configured so that structure shapes of buildings designed in the past are added as needed.

The joint information 134 will now be described with reference to Figures 2A to 2D. Figures 2A to 2D are diagrams for explaining the joint information 134 in an embodiment. Figures 2A to 2D show the relationship between the state of joints at the joints of a building and the joint information 134. In Figures 2A to 2D, the directions parallel to the bottom surface (horizontal surface) of the building are the X direction and the Y direction, and the X direction and the Y direction are orthogonal to each other. The direction orthogonal to the bottom surface (horizontal surface) of the building is the Z direction. The frame shape here refers to a structure in which columns and beams are joined at the joints in any of the X, Y, and Z directions.

2A to 2D, for example, the joint information 134 is information indicating whether or not a pillar or beam is joined in any of the X, Y, and Z directions at the joint. In the example of this figure, the joint information 134 is a value obtained by converting a binary number obtained by sequentially arranging a binary number (first information) indicating whether or not a pillar or beam is joined in the +Z direction, a binary number (second information) indicating whether or not a pillar or beam is joined in the +Y direction, and a binary number (third information) indicating whether or not a pillar or beam is joined in the +X direction into a decimal number. In the following description, the "binary number indicating whether or not a pillar or beam is joined" is set to "1" when a pillar or beam is joined, and "0 (zero)" when a pillar and a beam are not joined. However, this is not limited to this. The "binary number indicating whether or not a pillar or beam is joined" may be set to any two values determined arbitrarily.

Figure 2A shows an example of a joint on the first floor of a building. In this joint, a column is joined in the +Z direction, and a beam is not joined in the +Y or +Z direction. In this case, the first information in the joint information 134 is "1", the second information is "0", and the third information is "0". Therefore, when the sequence "100" of the first information to the third information is converted into a binary number, the joint information 134 is the value obtained by converting the binary number "100" into a decimal number, that is, "4".

Figure 2B shows an example of a joint such as the rooftop of a building. In this joint, the column is not joined in the +Z direction, and the beams are joined in the +Y and +X directions. In this case, the first information in the joint information 134 is "0", the second information is "1", and the third information is "1". Therefore, when the sequence "011" of the first information to the third information is converted into a binary number, the joint information 134 is the value obtained by converting the binary number "011" into a decimal number, that is, "3".

Figure 2C shows an example of a joint such as an outer perimeter frame part in the X direction of an intermediate floor of a building. In this joint, a column is joined in the +Z direction, no beam is joined in the +Y direction, and a beam is joined in the +X direction. In this case, the first information in the joint information 134 is "1", the second information is "1", and the third information is "0". Therefore, when the sequence "110" of the first information to the third information is converted into a binary number, the joint information 134 is the value obtained by converting the binary number "110" into a decimal number, that is, "6".

Figure 2D shows an example of a joint such as the Y-direction outer perimeter frame part of an intermediate floor of a building. In this joint, a column is joined in the +Z direction, a beam is joined in the +Y direction, and no beam is joined in the +X direction. In this case, the first information in joint information 134 is "1", the second information is "0", and the third information is "1". Therefore, when the sequence "101" obtained by arranging the first information to the third information is converted into a binary number, joint information 134 is the value obtained by converting the binary number "101" into a decimal number, that is, "5".

As shown in Figures 2A to 2D, the joint information 134 is information that expresses the state of the joints at the joints of a building using eight binary numbers from 000 to 111. Here, a value expressed in binary can be uniquely associated with a value expressed in decimal. Therefore, the joint information 134 can be expressed using eight decimal numbers from 0 to 7. In this embodiment, one of the eight decimal numbers from 0 to 7 is assigned to each of the joints in the building as the joint information 134, and the structural shape of the building can be expressed by a combination of the numbers (joint information 134) assigned to each of the joints.

The building connection information 134 will now be described with reference to Figures 3A and 3B. In Figures 3A and 3B, connection information 134 corresponding to the building shown on the right is shown on the left. Specifically, in Figure 3A, connection information 134-1 for building G-1 shown on the right is shown on the left. In Figure 3B, connection information 134-2 for building G-2 shown on the right is shown on the left. Figure 3A shows an example of building G-1 in which the second and third floors have the same number of columns and beams. Figure 3B shows an example of building G-2 in which the third floor has fewer columns and beams than the second floor.

As shown in Figures 3A and 3B, the connection information 134 is shown in a 12 row x 3 column matrix format, with the connection information at each of the three connection points (examples of connection parts) along the Y-axis direction on the first floor as matrix elements.

In the example of FIG. 3A, the junction information 134-1 is shown in a 12 row x 3 column matrix format with the sequence {4, 4, 4} corresponding to the junction information 134-1A, 134-1B, 134-1C at the junction points G-1A, G-1B, G-1C, respectively. In the example of this figure, it is shown that the junction information 134-1D in the junction point group G-1D adjacent to each of the junction points G-1A, G-1B, G-1C in the +Z direction is {6, 6, 4}. It is also shown that the junction information 134-1E in the junction point group G-1E adjacent to the junction point group G-1D in the -X direction is {7, 7, 5}. It is also shown that the junction information 134-1F in the junction point group G-1F adjacent to the junction point group G-1E in the +Z direction is {7, 7, 5}. It is also shown that the junction information 134-1H in junction group G-1H, which is adjacent to junction group G-1F in the +Z direction, is {3, 3, 1}.

In the example of FIG. 3B, for junction points that exist in FIG. 3A but do not exist in FIG. 3B, the junction information 134-1 is shown in a matrix format of 12 rows by 3 columns, similar to FIG. 3A, with the junction information being 0 (zero). In the example of this figure, it is shown that the junction information 134-2A in the junction point group G-2A is {6, 4, 0}. It is also shown that the junction information 134-2B in the junction point group G-2B adjacent to the junction point group G-2A in the - (minus) X direction is {7, 7, 4}. It is also shown that the junction information 134-2C in the junction point group G-2C adjacent to the junction point group G-2B in the +Z direction is {7, 7, 0}. It is also shown that the junction information 134-2D in the junction point group G-2D adjacent to the junction point group G-2C in the +Z direction is {2, 0, 0}.

Now, the learning data 131 will be described with reference to FIG. 4. FIG. 4 shows an example of the building frame shapes corresponding to each type when the frame shapes of buildings are classified, and their connection information. Any method may be used to classify the frame shapes, and for example, the frame shapes are classified using clustering processing or the like. In the example of this figure, an example is shown in which buildings are classified into 10 types according to their frame shapes. Also shown are buildings G-T1 to G-T10 corresponding to each of the 10 types, and their connection information 134-T1 to 134-T10.

The learning data 131 is a set of data in which each of the connection information 134 for multiple buildings is associated with a structural shape type 135. In the example shown in this figure, the learning data 131 is information in which a label (e.g., label A) corresponding to the type of building G-T1 is associated with the connection information 134-T1. The learning data 131 is information in which a label (e.g., label B) corresponding to the type of building G-T2 is associated with the connection information 134-T2. The learning data 131 is information in which a label (e.g., label C) corresponding to the type of building G-T3 is associated with the connection information 134-T3. Similarly, the learning data 131 is information in which the connection information 134 corresponding to each of the 10 types is associated with a label corresponding to that type.

Returning to the explanation of FIG. 1, the control unit 14 is a processor including, for example, a CPU (Central Processing Unit) and the like, and controls the frame shape estimation device 1 in an overall manner. The control unit 14 mainly executes the construction process of a frame shape estimation model and the estimation process of the frame shape. The control unit 14 includes a model construction unit 141, a data acquisition unit 142, a joint information generation unit 143, a frame shape estimation unit 144, and a display control unit 145.

The model construction unit 141 performs machine learning based on the learning data 131 stored in the storage unit 13 to construct a structural shape estimation model. The model construction unit 141 stores the constructed structural shape estimation model in the storage unit 13 as model information 132.

Here, we will explain the construction process of the frame shape estimation model by the model construction unit 141. The model construction unit 141 constructs the frame shape estimation model using, for example, a neural network (NN).

In the field of machine learning, image recognition technology using neural networks has been developed. In such image recognition using neural networks, training data is a set of data in which color information for each pixel in an image (e.g., RGB values for each pixel) is associated with the type of image, such as an image of a cat or an image of a dog. By training a neural network using such training data, it is possible to construct a trained model that has learned the relationship between the color information in an image and the type of that image. The trained model constructed in this way makes it possible to estimate the type of an image to be estimated from the color information in that image.

In this embodiment, this type of image recognition technology using neural networks, which has already been developed, is applied. Specifically, as described above, the junction information 134 at each junction of the building is expressed as color information at each pixel point of the image.

In this embodiment, the information that associates the connection information 134 converted into a vector with the type of structural shape is used as learning data. By using such learning data, the NN can learn the relationship between the building connection information 134 and the type of structural shape, just like the relationship between color information in an image and the type. In image recognition technology, if a trained model that has learned the relationship between color information at each pixel point of an image and the type of the image can accurately estimate the type of the image, it can be expected that the structural shape estimation model in this embodiment that has learned the relationship between the building connection information 134 and the type of structural shape will be a model that accurately estimates the type of structural shape.

For example, consider creating an image with the joint information 134 as RGB. In this case, for example, the binary number in the first information is the R value of RGB, the binary number in the second information is the G value of RGB, and the binary number in the third information is the B value of RGB. Also, "1" in binary is made to correspond to "255" in RGB, and "0" in binary is made to correspond to "0" in RGB. By creating an image in this way, an image with simple colors can be created. In the field of image recognition technology, it is considered that it is not so difficult to estimate the similarity between images with such simple colors. Therefore, if a method is used in which a NN is trained using the joint information 134 as if it were color information, as in this embodiment, it is considered that existing AI technology can also be applied to recognition of structural shapes.

The model construction unit 141 trains the NN to learn the relationship between the building connection information 134 in the learning data and the various categories (type information) into which the buildings are classified according to their structural shape. The model construction unit 141 adjusts the internal parameters of the NN (weight values and bias values) so that the type output from the building connection information 134 in the learning data approaches the type of the building in the learning data. The model construction unit 141 builds a structural shape estimation model that estimates the type of the building from the connection information 134 of the building to be estimated by repeatedly adjusting the internal parameters of the NN so that the type output from the NN approaches the type of the building in the learning data for each of all learning data. The model construction unit 141 stores information indicating the constructed structural shape estimation model in the storage unit 13 as model information 132.

Figure 5 is a diagram illustrating an example of applying the joint information 134 in an embodiment as learning data for a neural network. As shown in Figure 5, the model construction unit 141 converts the joint information 134 shown in a 12 row x 3 column matrix format as shown in Figures 3A and 3B, for example, into a representation in a 36 row x 1 column vector format. This makes it possible to represent the joint information 134 at each joint point of a building as if it were color information at each pixel point of an image.

Here, an example of the configuration of the NN used to construct the frame shape estimation model will be described. For example, 36 nodes are set in the input layer of the NN, and the joint information 134 converted into a vector format of 36 rows x 1 column, which is the input of the learning data, is input to the input layer of the NN. Three nodes are set in the second layer of the NN, and a weight matrix of (3 x 36), a bias vector of (3 x 1), and an activation function by a sigmoid function are used in the calculation from the 36 nodes in the input layer to the three nodes in the second layer. Three nodes are set in the third layer of the NN, and a weight matrix of (3 x 3), a bias vector of (3 x 1), and an activation function by a sigmoid function are used in the calculation from the three nodes in the second layer to the three nodes in the third layer. Ten nodes were set in the output layer of the NN, and a weight matrix of (10 x 3), a bias vector of (10 x 1), and an activation function based on a sigmoid function were set to be used in the calculation from the three nodes in the third layer to the ten nodes in the output layer. The Adam method was also adopted to optimize the weight matrix and bias vector, so that the squared error between the output layer value and the correct value in the training data (the label corresponding to the type of structural shape) was minimized.

The learning data used by the NN was for the 2x2 span, three-story building shown in Figure 4, and consisted of group data in which building joint information 134 was associated with a structural shape type indicating one of 10 structural shape types. A learning program was created using Chainer, an open source software library, and the NN was made to learn the learning data. The learning was repeated 100,000 times, and the learning time was several minutes. The NN after learning was used as the structural shape estimation model.

Figures 6 and 7 are diagrams for explaining the frame shape estimation model in the embodiment. Figure 6 shows the output values output from the frame shape estimation model when joint information in the learning data is input. Figure 7 shows the correct answer values (labels corresponding to the types of frame shapes) associated with the joint information in the learning data. Figure 6 shows a 10 row x 1 column output vector output from the output layer when joint information 134 corresponding to each of the 10 types of frame shapes is input to the frame shape estimation model. Figure 7 shows the correct answer vector corresponding to Figure 6.

As shown in Figures 6 and 7, the frame shape estimation model outputs an output vector with an accuracy such that the error from 1 or 0 (zero), which is shown as the correct value in the correct vector, is around 0.001. As a result, the internal parameters of the frame shape estimation model are adjusted to approach the correct value, and it is considered that the frame shape of the building can be estimated with high accuracy from the connection information 134.

Returning to the explanation of FIG. 1, the data acquisition unit 142 acquires input data (input information) when estimating the frame shape. For example, the data acquisition unit 142 acquires information indicating the presence or absence of a joint between a column and a beam for each joint in a building as input data via the input unit 11. The data acquisition unit 142 stores the acquired input data in the memory unit 13 as input information 133.

The connection information generation unit 143 generates connection information 134 based on the input data acquired by the data acquisition unit 142. The connection information generation unit 143 determines the state of the connection at the joint, i.e., whether or not a column and a beam are joined to the joint, based on the input information 133 stored in the memory unit 13, and generates connection information using the determination result. The connection information generation unit 143 stores the generated connection information in the memory unit 13 as connection information 134.

The frame shape estimation unit 144 estimates the frame shape of the building based on the frame shape estimation model stored in the storage unit 13 as the model information 132 and the connection information 134 of the building to be estimated. The frame shape estimation unit 144 estimates the frame shape of the building based on the output obtained by inputting the connection information 134 of the building to be estimated into the frame shape estimation model, for example.

Specifically, the frame shape estimation unit 144 acquires output values corresponding to each type of assumed frame shape from the frame shape estimation model as type information. For example, if there are 10 types of assumed frame shapes, the frame shape estimation unit 144 acquires an output vector of 10 rows x 1 column output from the frame shape estimation model. The 10 rows x 1 column output vector indicates values close to the correct value for each of the 10 types of frame shape types, i.e., the degree of similarity to that frame shape type (similarity degree).

The frame shape estimation unit 144 estimates the type of frame shape that is most similar to the frame shape of the building to be estimated based on the output values corresponding to each type of frame shape output from the frame shape estimation model. For example, the frame shape estimation unit 144 estimates the type of frame shape that corresponds to the largest output value among the output values corresponding to each type of frame shape output from the frame shape estimation model, that is, the type of frame shape that corresponds to the largest degree of similarity (degree of similarity) to the type of frame shape, as the type of frame shape that is most similar to the frame shape of the building to be estimated. For example, the frame shape estimation unit 144 acquires the frame shape for the estimated type from the frame shape type 135, and outputs the acquired frame shape as the frame shape of the building to be estimated.

Figures 8 and 9 are diagrams explaining the process of estimating the frame shape in the embodiment. In Figures 8 and 9, the left side shows the joint information 134 and frame shape of the building to be estimated, and the right side shows the joint information 134 and frame shape of the building estimated based on the output of the frame shape estimation model.

As shown in Figures 8 and 9, even when unlearned connection information 134 different from the learned data is input to the frame shape estimation model, it can be seen that a relatively similar frame shape is estimated as the frame shape of the building to be estimated. The frame shape estimation unit 144 estimates the type with the highest degree of similarity among the output vectors output from the frame shape estimation model as the frame shape of the building to be estimated. From this, it can be seen that when unlearned connection information 134 is input to the frame shape estimation model, the output vector is output so that the output value corresponding to the type of relatively similar frame shape is a large value.

Returning to the explanation of FIG. 1, the display control unit 145 (an example of an output control unit) outputs output information indicating the structural shape of the building estimated by the structural shape estimation unit 144 to the display unit 12 (output unit). That is, the display control unit 145 causes, for example, the display unit 12 to display a three-dimensional image of the structural shape. The display control unit 145 causes the display unit 12 to display a plurality of structural shapes belonging to a type corresponding to the label (type information) estimated by the structural shape estimation unit 144. In this case, the display control unit 145 may, for example, prioritize and cause the display unit 12 to display a structural shape that is close to a reference structural shape for the estimated type (for example, a structural shape for each type used as learning data).

Next, the operation of the frame shape estimation device 1 according to this embodiment will be described with reference to Figs. 10 and 11. Fig. 10 is a flowchart showing the flow of the process (process of constructing a frame shape estimation model) performed by the frame shape estimation device 1 according to this embodiment. Fig. 11 is a flowchart showing the flow of the process (process of estimating the frame shape of the building to be estimated) performed by the frame shape estimation device 1 according to this embodiment.

As shown in FIG. 10, in the construction process of the frame shape estimation model, the frame shape estimation device 1 first acquires learning data (step S10). The model construction unit 141 of the frame shape estimation device 1 acquires the learning data 131 stored in the storage unit 13.

Next, the model construction unit 141 constructs a structural shape estimation model from the training data (step S12). The model construction unit 141 constructs a structural shape estimation model by machine learning, for example, by adjusting internal parameters so that the output when the joint information in the training data is input to a neural network approaches the label corresponding to the type of structural shape in the training data.

Next, the model construction unit 141 stores information indicating the constructed model as model information 132 in the storage unit 13 (step S13). After the processing of step S13, the model construction unit 141 ends the processing (processing for constructing a structural shape estimation model).

As shown in FIG. 11, in the process of estimating the frame shape, the frame shape estimation device 1 first acquires input data (step S20). The data acquisition unit 142 of the frame shape estimation device 1 acquires, as input data, information indicating the state of the joints at each joint of the building to be estimated, for example, via the input unit 11. The data acquisition unit 142 stores the acquired input data in the storage unit 13 as input information 133.

Next, the joint information generating unit 143 of the frame shape estimation device 1 generates joint information 134 of the building to be estimated from the input data (step S21). The joint information generating unit 143 determines the state of the joint at the joint based on the input data stored in the storage unit 13 as the input information 133, and generates joint information of the building to be estimated using the determined result. The joint information generating unit 143 stores the generated joint information in the storage unit 13 as the joint information 134.

Next, the frame shape estimation unit 144 of the frame shape estimation device 1 estimates a group label from the frame shape estimation model and the connection information 134 of the building to be estimated (step S22). The frame shape estimation unit 144 estimates a label corresponding to the type of frame shape of the building to be estimated based on the output obtained by inputting the connection information 134 stored in the memory unit 13 into the frame shape estimation model.

Next, the frame shape estimation device 1 outputs a frame shape of a type corresponding to the label output from the frame shape estimation model (step S23). The frame shape estimation unit 144 acquires the frame shape corresponding to the estimated label from the frame shape type 135 stored in the storage unit 13. The display control unit 145 of the frame shape estimation device 1 causes, for example, a schematic diagram or a three-dimensional image of the acquired frame shape to be displayed on the display unit 12. After the processing of step S23, the frame shape estimation device 1 ends the processing (processing to estimate the frame shape).

As described above, the frame shape estimation device 1 according to the embodiment includes a storage unit 13 and a frame shape estimation unit 144. The storage unit 13 stores model information 132 indicating a frame shape estimation model. The frame shape estimation model is a model that estimates the frame shape of a building based on the joint information 134. The joint information 134 is information indicating the state of the joint at the joint that joins the column and the beam of the building. The frame shape estimation unit 144 estimates the frame shape of the building to be estimated based on the frame shape estimation model and the joint information 134 of the building to be estimated (target building). As a result, the frame shape estimation device 1 according to the embodiment can estimate the frame shape from the joint information of the building. Therefore, it is possible to determine whether the frame shape of a building being planned is similar to that of an existing building using the estimated frame shape without creating a three-dimensional drawing of the building.

In the frame shape estimation device 1 of the embodiment, the joint information includes first information, second information, and third information. The first information is binary information indicating whether or not a column is connected in the +Z direction (an example of a first direction perpendicular to the horizontal plane of the building) at the joint. The second information is binary information indicating whether or not a beam is connected in the +Y direction (an example of a second direction parallel to the horizontal plane of the building) at the joint. The third information is binary information indicating whether or not a beam is connected in the +X direction (an example of a third direction parallel to the horizontal plane of the building and perpendicular to the second direction) at the joint. As a result, in the frame shape estimation device 1 of the embodiment, the joint information can be expressed as a combination of three pieces of information including the first information, the second information, and the third information. Therefore, the state in which the column and the beam are connected at the joint can be expressed concisely and appropriately.

In addition, in the embodiment of the frame shape estimation device 1, the joint information is a numerical value obtained by converting a sequence of binary numbers representing the first information, binary numbers representing the second information, and binary numbers representing the third information into a three-digit binary number and then converting the sequence into a three-digit binary/decimal number. As a result, in the embodiment of the frame shape estimation device 1, the joint information 134 at each joint point of the building can be expressed as if it were color information at each pixel point of an image. Therefore, it is possible to construct a frame shape estimation model by applying image recognition technology using NN, which has been developed in advance, and it is also possible to improve the estimation accuracy using the constructed model.

In addition, in the frame shape estimation device 1 of the embodiment, the frame shape estimation model outputs a similarity degree indicating the degree of similarity to each of multiple types of the frame shape of the building based on the input joint information. The frame shape estimation unit 144 acquires the similarity degrees output from the frame shape estimation model, and estimates the frame shape belonging to the type with the largest similarity degree among the acquired similarity degrees as the frame shape of the building to be estimated. As a result, the frame shape estimation device 1 of the embodiment can estimate the frame shape belonging to the type with the largest similarity degree as the frame shape of the building to be estimated. Therefore, the frame shape estimation unit 144 can estimate the frame shape most similar to the building to be estimated by the simple process of selecting the largest value from the output vector output from the model.

The frame shape estimation device 1 according to the embodiment includes a data acquisition unit 142, a joint information generation unit 143, and a display control unit 145 (output control unit). The data acquisition unit 142 acquires input data indicating the state of each joint at the joint points of the building to be estimated. The joint information generation unit 143 generates joint information 134 of the building to be estimated based on the input data acquired by the data acquisition unit 142. The display control unit 145 outputs output information indicating the frame shape of the building estimated by the frame shape estimation unit 144 to the display unit 12 (output unit). As a result, the frame shape estimation device 1 according to the embodiment can generate joint information from the input data. Therefore, the user does not need to generate joint information, and convenience can be improved. Furthermore, the display control unit 145 can output output information indicating the estimated frame shape of the building to the display unit 12 (output unit). Therefore, the user can, for example, visually confirm the frame shape, making it easier for the user to select a more appropriate frame shape.

The frame shape estimation device 1 according to this embodiment also includes a model construction unit 141 that constructs a frame shape estimation model based on the learning data. As a result, the frame shape estimation device 1 according to this embodiment can reconstruct a frame shape estimation model with updated learning data, for example, periodically.

The frame shape estimation method according to the present embodiment is a frame shape estimation method of the frame shape estimation device 1 having a storage unit 13 that stores model information 132, and includes a building frame estimation step. The model information 132 is a frame shape estimation model that estimates the frame shape of a building from joint information, and is information indicating the frame shape estimation model constructed based on learning data. The learning data has a plurality of sets of data in which information indicating the type of the frame shape of the building is associated with the joint information 134 of the building. In the building frame estimation step, the frame shape estimation unit 144 in the frame shape estimation device 1 estimates the frame shape of the building from the frame shape estimation model and the joint information 134 of the building to be estimated. As a result, the frame shape estimation method according to the present embodiment has the same effect as the frame shape estimation device 1 described above, and for example, it is possible to determine whether the frame shape of a building being planned is similar to that of an existing building using the estimated frame shape without creating a three-dimensional drawing of the building.

The present invention is not limited to the above embodiment, and can be modified without departing from the spirit of the present invention. For example, in the above embodiment, a configuration example in which the frame shape estimation device 1 includes the model construction unit 141 has been described, but the present invention is not limited to this, and the frame shape estimation model may be constructed outside the frame shape estimation device 1. In this case, for example, the model construction device may include the model construction unit 141 and construct the frame shape estimation model.

In the above embodiment, the structural shape estimation device 1 is configured as a single device, but the present invention is not limited to this and may be configured as a plurality of devices (e.g., a plurality of server devices). In addition, the structural shape estimation device 1 may be configured such that a part or the whole of the memory unit 13 is provided externally.

In addition, in the above embodiment, an example was described in which the output unit was the display unit 12, but this is not limited to this, and the output unit may be another output unit, such as a printer or a storage device such as a file server.

In this way, in the above embodiment, a method was proposed to express the state of the joint at the joint point in binary, and to assign the numbers 0 to 7, which are the values obtained by converting the binary number to decimal, to the joint points to express the frame shape by combining the joint information 134. In addition, the joint information 134 obtained in this way is vectorized, and the association between the joint information 134 and the frame shape is used as training data, which is a set of data, to make the NN learn, thereby demonstrating the possibility of extracting a type of frame shape that is similar in a rational manner. In the above embodiment, an example in which the joint information 134 is input to the frame shape estimation model has been described, but this is not limited to this. For example, span length data in the X direction and Y direction, floor height data, building volume data, etc. may be input to the frame shape estimation model. It is expected that by using the span length data, floor height data, building volume data, etc. in combination, it will be possible to estimate a more similar frame shape. In this case, the learning data includes not only the building joint information, but also multiple sets of data in which the type of the building's structural shape is associated with information such as the building's X- and Y-direction span length and floor height distribution. Such learning data is trained by a neural network to create a structural shape estimation model.

In the above embodiment, an example of learning data through a NN has been described, but the present invention is not limited to this. For example, a CNN (Convolutional CNN) may be used instead of a NN. When a CNN is used, it may be possible to create a network that can grasp the planar or elevational patterns of the above numbers 0 to 7 by manipulating filters and pooling provided in the CNN. Also, instead of a NN or CNN, a model that is applied to existing machine learning, such as a decision tree, hierarchical Bayes, or SVM (Support Vector Machine), may be used.

Each component of the frame shape estimation device 1 described above has an internal computer system. A program for realizing the functions of each component of the frame shape estimation device 1 described above may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read into a computer system and executed to perform processing in each component of the frame shape estimation device 1 described above. Here, "reading a program recorded on a recording medium into a computer system and executing it" includes installing the program into a computer system. Here, "computer system" includes hardware such as an OS and peripheral devices.

A "computer system" may also include multiple computer devices connected via a network, including the Internet or communication lines such as a WAN, LAN, or dedicated line. A "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs, and CD-ROMs, as well as storage devices such as hard disks built into a computer system. In this way, the recording medium that stores the program may be a non-transitory recording medium such as a CD-ROM.

The recording medium also includes an internal or external recording medium accessible from a distribution server to distribute the program. The program may be divided into multiple parts, downloaded at different times, and then combined by each component of the structural shape estimation device 1, or each divided program may be distributed by a different distribution server. Furthermore, the term "computer-readable recording medium" includes a recording medium that holds a program for a certain period of time, such as a volatile memory (RAM) in a computer system that becomes a server or client when a program is transmitted over a network. The program may be a recording medium for implementing part of the above-mentioned functions. Furthermore, the program may be a so-called difference file (difference program) that can realize the above-mentioned functions in combination with a program already recorded in the computer system.

In addition, some or all of the above-mentioned functions may be realized as an integrated circuit such as an LSI (Large Scale Integration). Each of the above-mentioned functions may be individually processed, or some or all of the functions may be integrated into a processor. The integrated circuit method is not limited to LSI, and may be realized by a dedicated circuit or a general-purpose processor. Furthermore, if an integrated circuit technology that can replace LSI appears due to advances in semiconductor technology, an integrated circuit based on that technology may be used.

1... structural shape estimation device, 11... input unit, 12... display unit, 13... storage unit, 14... control unit, 131... learning data, 132... model information, 133... input information, 134... connection information, 135... structural shape type, 141... model construction unit, 142... data acquisition unit, 143... connection information generation unit, 144... structural shape estimation unit, 145... display control unit

Claims (7)

a storage unit that stores model information indicating a frame shape estimation model that estimates a frame shape of a building based on joint information indicating a joint state at a joint that joins a column and a beam of the building, the frame shape estimation model being constructed by learning learning data having a plurality of sets of data in which the joint information is associated with the frame shape of the building;
a frame shape estimation unit that estimates a frame shape of the target building based on an output result obtained by inputting the connection information of a target building to be estimated into the frame shape estimation model stored in the storage unit; and
A frame shape estimation device comprising: The joint information includes first binary information indicating whether or not a column is connected in a first direction perpendicular to a horizontal plane of the building at the joint, second binary information indicating whether or not a beam is connected in a second direction parallel to the horizontal plane, and third binary information indicating whether or not a beam is connected in a third direction parallel to the horizontal plane and perpendicular to the first direction.
The frame shape estimation device according to claim 1 . The joint information is a numerical value obtained by converting a sequence of binary numbers representing the first information, binary numbers representing the second information, and binary numbers representing the third information into a three-digit binary number and then converting the three-digit binary number into a decimal number.
The frame shape estimation device according to claim 2. the frame shape estimation model outputs a degree of similarity to each of a plurality of types of the frame shape used in the group data based on the input joint information,
the frame shape estimation unit acquires the similarity degrees output from the frame shape estimation model, and estimates, among the acquired similarity degrees, the frame shape belonging to the type of the frame shape having the largest similarity degree as the frame shape of the target building.
The frame shape estimation device according to any one of claims 1 to 3. A data acquisition unit that acquires information indicating a state of a joint at a joint that joins a column and a beam of the target building;
a connection information generation unit that generates the connection information of the target building based on the connection state of the connection portion of the target building acquired by the data acquisition unit;
an output control unit that outputs output information indicating the frame shape of the building estimated by the frame shape estimation unit to an output unit;
The frame shape estimation device according to claim 1 , further comprising: a model construction unit that constructs the frame shape estimation model based on the learning data,
The frame shape estimation device according to any one of claims 1 to 5. A frame shape estimation method for a frame shape estimation device including a storage unit that stores model information indicating a frame shape estimation model that estimates a frame shape of a building based on joint information indicating a joint state at a joint that joins a column and a beam of the building, the frame shape estimation method comprising:
A structural shape estimation method comprising an estimation step of estimating the structural shape of a target building from the structural shape estimation model stored in the memory unit and the joint information of a target building to be estimated.

Images