Element automatic arrangement system and element automatic arrangement method

Abstract

To provide an element automatic arrangement system and an element automatic arrangement method with which the necessary and sufficient number of elements can be arranged for a framing plan for construction.SOLUTION: Provided is an element automatic arrangement system 100 for arranging elements at appropriate positions in a framing plan for construction, and the system comprises: reading means 1 that reads a framing plan for arrangement D in which a target member P on which the elements E are to be arranged is written to obtain arrangement target data T; and operation means 3 that uses a learning dataset L obtained through machine learning of the characteristics of the target member in another framing plan in which the elements are already arranged on the target member. The operation means 3 has means that automatically arranges the elements E on the target member P through operation based on the learning dataset L according to the characteristics of the target member P in the arrangement target data T.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to an automatic element placement system and an automatic element placement method for placing elements such as stakes and columns required for architectural design in appropriate positions on various framing plans created in the design stage of a building. [Background technology]

[0002] During the design stage of a building, various floor plans are created. A floor plan is a plan showing each structural element, such as the foundation, floor, and ceiling, from above or below. Depending on the element (architectural element) involved, there are various types of floor plans, such as foundation plan, pile plan, floor plan, beam plan, roof plan, and ceiling plan.

[0003] These framing plans are interrelated. For example, the pile framing plan places the appropriate number of piles in the appropriate positions under the foundation according to the shape of the foundation described in the foundation framing plan. Similarly, the rafter framing plan places the rafters and rafters while considering the positions of the beams and columns described in the beam framing plan.

[0004] It is desirable to place the necessary and sufficient number of piles, beams, columns, and other elements in optimal positions based on structural calculations. Conventionally, technology has been developed that calculates the number of piles required from information such as the weight of the building and the properties of the ground, and then automatically places the piles based on floor plan information.

[0005] Patent Document 1 discloses a technology that calculates the number of piles required to support the entire building in advance and automatically places piles at the intersections of exterior wall corners and walls. With the technology in Patent Document 1, if there are piles that are spaced apart more than the preset interval after the piles are automatically placed, the number of piles placed is adjusted so that the total number of piles approaches the required number.

[0006] Furthermore, Patent Document 2 discloses a technology for automatically placing piles at corners, T-shaped sections, cross sections, directly below columns with a predetermined axial force, etc. of the foundation on the foundation plan. With the technology in Patent Document 2, if the spacing between placed piles on the foundation is equal to or greater than a predetermined interval, piles are placed at positions that divide the spacing equally. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2020-148013 [Patent Document 2] Japanese Patent Application Laid-Open No. 2006-127186 Summary of the Invention [Problem to be solved by the invention]

[0008] However, the technology of Patent Document 1 is a technology aimed at providing highly accurate estimates, and it automatically places piles at wall entrance / exit corners and wall intersections based on a floor plan. If the number of piles placed is less than the number calculated from the weight of the building, a process is performed to adjust the placement so that the number of piles approaches a predetermined number. Specifically, it employs a method in which piles are added between piles when the distance between piles is equal to or greater than a predetermined interval. It is also designed to shift the positions of the piles when the distance between piles is equal to or less than a limit value.

[0009] However, with this method, if the total number of stakes is less than the calculated number and there are parts where the distance between stakes is greater than a predetermined interval, the stakes will always (forcibly) be placed so that the distance is less than the predetermined interval. As a result, there is a possibility that stakes will be placed in places where they do not actually need to be added. Conversely, if the total number of automatically placed stakes exceeds the calculated number, there is a possibility that stakes will not be added in places where they are needed. Therefore, the technique of Patent Document 1 has the problem that the pile arrangement may become unbalanced.

[0010] On the other hand, the technology in Patent Document 2 involves inputting beams and columns into the roof plan and floor plan, and then automatically placing piles at corners, T-shaped sections, and cross sections of the foundation, taking into account the axial force of the columns, in relation to the foundation plan. Patent Document 2 also shares with the technology in Patent Document 1 the point that piles are added when the distance between piles is equal to or greater than a predetermined value, but is characterized by the point that piles are added so as to divide the distance between piles equally.

[0011] However, even with the method of Patent Document 2, there is still the possibility that piles will be placed in places where they do not actually need to be added, or that piles will not be added in places where they are needed. As such, with conventional technology, the placement of elements such as piles is determined uniformly based on the weight of the building and the set value of the distance between piles, which creates the problem that tacit knowledge, such as empirical knowledge of widening or narrowing the spacing between elements, is not reflected.

[0012] The present invention has been made in consideration of the above-mentioned problems, and its purpose is to provide an automatic element placement system and an automatic element placement method that can place elements as necessary and sufficiently on an architectural framing plan. [Means for solving the problem]

[0013] The means adopted by the present inventors to solve the above problems will be described below. The automatic element placement system of the present invention is a system for placing elements at appropriate positions in an architectural framing plan. Its basic configuration includes a reading means for reading a framing plan showing target components for placement of the elements and creating the placement target data, and a calculation means for using a training dataset obtained by machine learning the characteristics of target components for which elements have already been placed in other framing plans. The calculation means is characterized by being a means for automatically placing the elements according to the characteristics of the target components in the placement target data by calculation based on the training dataset.

[0014] The learning data set used by the calculation means of the present invention can be machine-learned to learn the features of target components whose elements have already been placed in other framing plans, thereby learning the features of the positions of elements that are determined to be optimal in various design cases. Then, based on the learning data set, the target components of the framing plans read by the reading means can be placed.

[0015] In this way, in the present invention, the placement of elements calculated by the calculation means is not a uniform placement based solely on logical calculations, but is placed taking into consideration past examples including tacit knowledge from past designs. Therefore, a rational arrangement of elements can be calculated automatically without redundant or insufficient arrangement.

[0016] The layout plan read by the reading means can be of any type, as long as it shows the target components for element placement. Examples include a foundation plan showing the foundation for pile placement, a beam plan showing the beams for rafter placement, and a floor plan showing the floor beams for axis columns.

[0017] The following means can be used to solve the problem. In the above configuration, the target component is a combination of linear components, and the calculation means includes probability calculation means and arrangement determination means.

[0018] The probability calculation means uses a training data set obtained by machine learning the positional features of elements arranged alongside the straight line portions of the target component in the other framing plan and the positional features of elements arranged at the intersection of the straight line portions, and determines the probability that the element should be arranged at a predetermined position on the target component of the arrangement target data based on the training data set. On the other hand, the placement determination means is means for placing the elements based on the probability calculated by the probability calculation means for each predetermined position on the target component of the placement target data.

[0019] The learning dataset used by the probability calculation means in the calculation means is obtained by machine learning the positional features of elements arranged alongside straight sections of the target component and the positional features of elements arranged at the intersections of the straight sections. Straight sections include not only beam-like sections of the target component, but also peninsula-like sections with one end unconnected. Furthermore, intersections of straight sections include not only corners and T-shaped sections where the target component intersects at right angles, but also diagonal intersections.

[0020] In the above configuration, the straight line sections where each element can be placed and the intersections between the straight line sections are mechanically analyzed to determine the characteristics of the target component on which the elements are placed, such as the length, position, distribution, and shape of the surrounding area, making it possible to learn in more detail the characteristics of element placement in past designs. In addition, the placement determination means places the elements based on the "probability of whether or not the element should be placed" calculated by the probability calculation means, so the density and balance of the placement can be adjusted as desired by selecting a threshold value. Therefore, a more rational arrangement of elements can be calculated automatically.

[0021] As yet another means that can be adopted to solve the problem, in the above configuration, the probability calculation means can use a learning data set obtained by machine learning the first feature in the other framing plan, which is the feature of the position of an element arranged at a corner of the target component, the second feature, which is the feature of the position of two elements arranged side by side on a straight line portion, and the third feature, which is the feature of the position of four elements arranged in a rectangular shape. At this time, the probability that the element should be placed at a predetermined position in the target component of the placement target data is determined based on an overall probability obtained by superimposing the probabilities calculated from the first feature, the second feature, and the third feature.

[0022] The placement determination means calculates a rank based on the overall probability calculated by the probability calculation means for each predetermined position in the target component of the placement target data, and is characterized in that it is a means for placing the element at a position of a predetermined rank or higher.

[0023] In this configuration, in addition to two points on the straight line when the target component is viewed as a line and four points on the rectangle when the target component is viewed as a surface, the corners are also used as points where it is known in advance that elements are likely to be located, and machine learning is used to learn characteristics such as the length and position of the target component where the elements are located, and the shape of the surrounding area, to create a learning dataset. This allows the system to learn not only about the characteristics of the arrangement as microscopic points (corners) and lines (straight lines) that support the structure, but also about the characteristics of the arrangement as surfaces (rectangular shapes) that work together with the structure of surrounding target components to support the structure.

[0024] The probability calculation means also superimposes the calculated three probability distributions for points, lines, and surfaces. In other words, the overall probability can be calculated at a given position by multiplying the calculated probabilities. Based on the probabilities, a ranking is assigned to each given position on the target component, and elements are placed at positions where the rank is equal to or greater than the threshold value. By carrying out the above processing, it is possible to achieve a more balanced and rational layout when viewed as the building as a whole.

[0025] As yet another means that can be adopted to solve the problem, the placement determination means may include means for deleting an element at a predetermined position or adding an element at a predetermined position based on the relationship between the probability for each position calculated by the probability calculation means and the distance between adjacent elements, for the position where it has been decided to place the element.

[0026] In the probability for each element position calculated by the probability calculation means, when the elements are arranged closer than a set value, if the arrangement probability of the adjacent element is low, the element is deleted. Also, if the distance between the adjacent elements is greater than a set value, a new element is added between the elements.

[0027] With the above configuration, rather than simply deleting elements when the distance between elements is close, the placement probability of adjacent elements is taken into account when determining whether to delete an element, so there is no risk of deleting a necessary element. Also, by adding an element when the distance between adjacent elements is greater than a set value, even if a placement omission occurs on rare occasions due to a problem specific to statistical processing, adding an element allows for a design that is structurally safe.

[0028] In the above-described automatic element placement system, the placement plan is set as a foundation plan, the elements are set as piles, and the target components are set as a foundation, and the learning data set can be obtained by machine learning the characteristics of foundations for which piles have already been placed in other pile plans.

[0029] The above configuration can be a system that automatically places piles on a foundation in a foundation plan, using piles as elements. Pile placement depends on a wide range of parameters, including not only the shape of the building but also the properties of the ground and the type of piles. Therefore, it is difficult to determine a uniform, necessary and sufficient placement through logical calculations. With the above configuration, by using a learning dataset that has been machine-learned to learn the characteristics of the shape of the foundation where piles have already been placed in the pile plan, rational placement can be automatically performed on a foundation plan, which is a placement plan.

[0030] As yet another means that can be adopted to solve the problem, it is also possible to adopt a means as an automatic element placement method for placing elements at appropriate positions on a building plan. This method is configured to include a reading step of reading a placement plan showing the target component on which the element is to be placed and converting it into placement target data, an image analysis step of grasping the characteristics of the target component from the placement target data obtained by the reading step, and a calculation step of using a learning dataset obtained by machine learning data from other placement plans in which elements have already been placed relative to the target component. The calculation step is a step of automatically arranging the elements according to the characteristics of the target components in the placement target data by calculation based on the learning data set.

[0031] Even with this configuration, the arrangement of elements calculated by the calculation step is not a uniform arrangement based solely on logical calculations, but rather an arrangement calculated taking into account tacit knowledge from past designs. Therefore, a rational arrangement of elements can be calculated automatically without redundant or insufficient arrangement.

[0032] In the above configuration, the calculation step may also include a probability calculation step and an arrangement determination step. The probability calculation step is a step that uses a learning dataset obtained by machine learning the first feature in the other framing plan, which is the position feature of an element located at the corner of the target component, the second feature, which is the position feature of two elements located next to each other in a straight line section, and the third feature, which is the position feature of four elements located in a rectangular shape. In addition, in the probability calculation step, the probability that the element should be placed at a predetermined position in the target component of the placement target data is determined by superimposing the probabilities calculated from the first feature, the second feature, and the third feature, respectively.

[0033] On the other hand, the placement determination step is a step of calculating a rank based on the overall probability calculated for each specified position in the target component of the placement target data by the probability calculation step, and is a step of placing the element at a position of a predetermined rank or higher. The step also includes deleting an element from a predetermined position or adding an element to a predetermined position based on the relationship between the probability for each position calculated by the probability calculation step and the distance between adjacent elements, for the position where it has been decided to place the element.

[0034] With the above configuration, in the probability calculation step, it is possible to learn not only the characteristics of the arrangement as microscopic points (corners) and lines (straight lines) that support the structure, but also the characteristics of the arrangement as surfaces (rectangular shapes) that support the structure in cooperation with the structure of surrounding target components. In addition, in the placement determination step, the three probability distributions of points, lines, and surfaces calculated by the probability calculation means are superimposed. In other words, at any position, the overall probability can be calculated by multiplying the calculated probabilities. By carrying out the above processing, it is possible to achieve a more balanced and rational layout when viewed as the building as a whole.

[0035] Furthermore, in the above configuration, the layout plan can be a foundation plan, the elements can be piles, the target members can be foundations, and the learning dataset can be obtained by machine learning the characteristics of foundations where piles have already been placed in other pile plans. In this configuration, by using a learning dataset that has been machine-learned to learn the shapes of foundations where piles have already been placed in pile plans, rational placement can be automatically performed on the foundation plans. [Effects of the Invention]

[0036] The automatic element placement system of the present invention includes a reading means for reading a placement framing plan and a calculation means for using a learning data set obtained by machine learning the characteristics of target components for which elements have already been placed in other framing plans. The calculation means is a means for automatically placing elements for target components in the placement target data based on the learning data set, and is therefore able to perform calculations that take into account past examples including tacit knowledge from past designs, rather than a uniform placement based solely on logical calculations. This has the effect of allowing elements to be placed as necessary and as needed on the architectural plan. [Brief explanation of the drawings]

[0037] [Figure 1] 1 is an explanatory diagram illustrating a configuration of an automatic element placement system according to the present invention. [Figure 2] FIG. 10 is an explanatory diagram showing the details of advance machine learning in the automatic element placement system of the present invention. [Figure 3] 1 is a flowchart showing a processing flow of the automatic element placement system of the present invention. [Figure 4] FIG. 10 is an explanatory diagram showing a state in which a drawing is read in the automatic element placement system of the present invention. [Figure 5] FIG. 10 is an explanatory diagram showing a state in which dimensions are automatically detected in the automatic element placement system of the present invention. [Figure 6] FIG. 10 is an explanatory diagram showing the state in which the placement probability of stakes is displayed in the automatic element placement system of the present invention. [Figure 7] FIG. 10 is an explanatory diagram showing a state in which piles are automatically placed in the automatic element placement system of the present invention. [Figure 8] FIG. 10 is an explanatory diagram showing the state in which the excess or shortage of piles has been adjusted in the automatic element placement system of the present invention. [Figure 9] FIG. 10 is an explanatory diagram illustrating an automatic element placement system according to a first modified example of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0038] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to Figures 1 to 8. In the following description, each figure is depicted schematically for the sake of simplicity. The automatic element arrangement system (hereinafter simply referred to as the system) 100 of the present invention is a system that operates on the web, as shown in Fig. 1, and a customer terminal C connects to the system 100 via the internet. Although the system shown in Fig. 1 is a web system, it may also be a system that is installed on the customer terminal C and operates as a stand-alone system, and the system's form is not limited thereto.

[0039] System Configuration First, the overall configuration of this system 100 will be described with reference to Figure 1. A user U is a user of this system 100, and the reading means 1 of the system 100 is a block that reads a layout plan D, which is a layout plan for construction, from a customer terminal C based on the operation of the user U, and sets the read plan as layout target data T. Various framing plans can be used for the framing plan to be loaded, but it must be appropriate as a layout plan D that shows the target member P where element E should be placed. For example, if element E is a column, the target member P is a floor beam, and the layout plan D will be a floor plan. Also, if element E is a pile, the target member P is a foundation, and the layout plan D will be a foundation plan. In the following explanation, as an example, element E is a pile, layout plan D is a foundation plan, and target member P is a foundation.

[0040] The layout plan D is image data input to or pre-saved on the customer terminal C, and is, for example, vector data such as PDF output from a CAD system, or raster data such as JPEG or PNG (Portable Network Graphics) obtained by scanning a paper drawing with a scanner S. In the case of raster data obtained by scanning a paper drawing, the entire data may be tilted, so a tilt correction function may be installed.

[0041] The system 100 may be provided with an image analysis means 2. The image analysis means 2 performs image analysis on the placement target data T obtained by reading the placement plan D using an OCR (Optical Character Recognition / Reader) function. The OCR function can read and store the overall width and depth dimensions of the foundation, which is the target component P, as needed. It can also calculate the reference pitch of element E from the external dimensions. The specific procedure for reading dimensions will be described later.

[0042] The calculation means 3 is a block that automatically calculates the arrangement of piles, which are elements E, using the learning dataset L. The calculation means 3 uses a learning dataset L obtained by machine learning a large amount of features of foundations where piles have already been placed in pile plans. Various machine learning methods can be used, but supervised learning that identifies in advance the parts of the foundation where piles are placed is preferred, and deep learning using a neural network is even more preferred. Details of machine learning will be described later.

[0043] The calculation means 3 includes a probability calculation means 31 and a placement determination means 32. The probability calculation means 31 uses a learning dataset L obtained by prior machine learning to assign weights according to the characteristics of the foundation in the placement target data T. Then, the probability that a pile should be placed at a predetermined position of the foundation of the placement target data T is calculated based on the weights. The placement determination means 32 determines whether or not to place a stake based on the probability at a predetermined position calculated by the probability calculation means 31. In making the determination, a threshold is set in advance for the index based on the probability, and a stake is placed at a position where the index exceeds the threshold.

[0044] The probability calculation means 31 may delete stakes if the stakes are placed closer than a set value and the placement probability of the adjacent stakes is low in the calculated probability for each predetermined position. Also, if the distance between adjacent stakes is greater than a set value, a new stake may be added between the stakes.

[0045] "Preliminary machine learning" The training dataset L can be obtained by machine learning a large amount of features such as the shape and position of the foundation at the position of the piles placed on the pile plan. Specifically, a human being looks at the pile plan and trains the system on image data of the same position as the piles placed on the original foundation plan, or a specified range including that position. The foundation plan that is the source of the image data to be trained is a foundation plan used in a pile plan that places the piles (element E) on the foundation (target component P), and is preferably selected from those that have been properly designed and constructed without any problems in the past. In the machine learning, features based on the content shown in Figure 2 are trained.

[0046] Specifically, for example, the following three feature amounts can be learned. Note that in Figure 2, the positions of the piles are shown as round markers, but the image data actually learned is image data of a portion of a foundation plan showing only the shape of the foundation without any pile markers. However, even if the pile plan contains pile markers, image data from which the pile markers have been removed by image processing may be used.

[0047] First, as the first feature, we learn the features of the "point" where the foundations intersect, as shown in the square frame in Figure 2(a). For example, we learn features such as the length of the two intersecting foundations, whether it is an exterior corner or an interior corner, and whether it is a T-shaped or L-shaped corner.

[0048] For the second feature, as shown in the square box in Figure 2(b), any one foundation is considered a "line," and the system focuses on piles that are parallel to one another in a straight line, learning the features of two adjacent piles. For example, it learns features such as the distance between two piles and whether the adjacent piles are corners.

[0049] For the third feature, as shown in the square frame in Figure 2(c), the area enclosed by the four foundations is regarded as a "surface," and the system focuses on the piles arranged in a rectangle, learning the features of the four piles arranged in a rectangle. For example, it learns features such as the spacing between the four piles, whether all four points are corners, or whether they form a T-shape.

[0050] In these learning processes, a human can select the range of learning to be performed. At this time, since there is an extremely high possibility that piles will be placed at the foundations of at least the four corners of the entire building, it is desirable to perform learning that includes these four corners. Note that while it is desirable to perform supervised learning in which a human specifies the learning range, it is also possible to perform classification and learning using unsupervised learning, or to set a reward and perform reinforcement learning.

[0051] In this way, we learn about each feature of the foundation where the piles are placed, and by using deep learning with a neural network, for example, we can obtain a learning dataset L that weights the relationship between the shape and position of the foundation and the probability that the piles should be placed.

[0052] "Processing procedure of this system" Next, a specific method for automatically arranging elements by the system 100 will be described with reference to Figures 3 to 8. In the following description, element E is assumed to be a pile, the layout plan D is assumed to be a foundation plan, and the placement target component P is assumed to be a foundation. First, in the flowchart of FIG. 3, in S1: Reading step, user U operates customer terminal C to send a foundation plan as a layout plan D to system 100. System 100 reads layout plan D and sets it as placement target data T. FIG. 4 shows the state in which the loaded placement target data T is displayed on customer terminal C. In FIG. 4, the shape and layout of the foundation as target component P, and each dimension, are displayed.

[0053] The layout plan D to be read is raster data obtained by scanning a paper drawing with a scanner S, but it may be slightly tilted due to the effects of scanning. In this case, the tilt can be corrected by using the straight line part of the drawing frame.

[0054] Next, in S2: Image Analysis Step, dimensional information such as the outer shape of the target component P may be detected using an OCR function. Specifically, as shown in the rectangular frame in Fig. 5, dimension areas A1 to A4 are detected using the OCR function, and the numerical values ​​aligned on the same line in the horizontal and vertical directions of the dimension areas A1 to A4 are read and added up. Of the added numerical values, the largest numerical value or a value that overlaps in the added values ​​of the upper, lower, left, and right dimension areas is determined to be highly likely to be the outermost dimension, and is recorded as the outer dimension.

[0055] When detecting exterior dimensions, if a balcony or the like (not shown) is shown on the foundation plan, the dimensions including the balcony or the like and the dimensions of only the foundation part excluding the balcony or the like may be recorded. In this case, it is possible to record both the maximum exterior dimensions including the balcony or the like and the maximum exterior dimensions of the part excluding the balcony or the like, and allow the user U to select the appropriate dimensions.

[0056] The minimum pitch width is also detected from the external dimension values ​​that have been read. By convention, structures such as building foundations are placed in intervals of 0.5 ken = 910 mm (0.91 m), with 1 ken = 1,820 mm (1.82 m) as the standard. When the external dimension that has been read is divided by an integer, the result that corresponds to these conventional standard dimensions is recorded as the standard pitch. This standard pitch can be used as an indicator for determining whether the spacing between piles is too far or too close when adjusting the final number of piles to be placed. Although the original value of 1 ken is 1818 mm, it is often rounded up to 1820 mm, and so this dimension is used in the present invention. In addition to these dimensions, any minimum pitch width can be selected, such as a meter module based on 1 m. Thus, if the read dimension is 1820 mm or 3620 mm, the minimum pitch width is 910 mm; if the read dimension is 1818 mm or 3636 mm, the minimum pitch width is 909 mm; and if the read dimension is 2000 mm or 3500 mm, the minimum pitch width is 1000 mm.

[0057] Next, in S3: calculation step, stakes are placed in appropriate positions for the loaded placement target data T. In the flowchart of FIG. 3, placement is performed through S31: probability calculation step and S32: placement determination step.

[0058] S31: In the probability calculation step, the probability that a stake should be placed at each predetermined arbitrary position on the placement target data T is calculated based on the learning data set L obtained by machine learning. For example, the predetermined position can be determined by calculating the probability for each pixel in the image data of the entire placement target data T, and then using the integral value of the probability value in a 10-pixel square area. In this case, the upper left pixel of the placement target data T is used as the reference, and the 10-pixel square area is shifted by one pixel, and the integral value is calculated each time. When calculating the probability value for each pixel, for example, a neural network can be used to calculate the probability that a stake should be placed for each of the positions as a "point" corresponding to the first feature amount mentioned above, the position as a "line" corresponding to the second feature amount, and the position as a "surface" corresponding to the third feature amount.

[0059] In Figure 6(a), the probability that piles should be placed as "points" such as corners and T-shaped sections is calculated for each pixel and displayed as a heat map. In Figure 6(b), the probability that piles should be placed as "lines" such as on the straight parts of the foundation is calculated for each pixel and displayed as a heat map. In Figure 6(c), the probability that piles should be placed as "surfaces" surrounded by the foundation is calculated for each pixel and displayed as a heat map.

[0060] In Figure 6(d), the probabilities calculated using these three methods are multiplied for each position on the foundation (each pixel in the example of Figure 6(d)) and displayed as a heat map. In this way, the probabilities calculated from various perspectives are multiplied to comprehensively determine whether or not the piles should be placed, thereby preventing the pile placement from being biased towards a specific design policy.

[0061] To elaborate on the "specific design policy," the load acting on the rising part of the foundation differs depending on whether the foundation is a strip footing or a slab footing. In the case of a strip footing, the load acting on the rising part of the foundation tends to be greater, so the number of piles that need to be placed in a straight line on the rising part of the foundation tends to increase, and the spacing between the piles tends to be narrower. On the other hand, a mat foundation can support the load even on the surface area surrounded by the rising part of the foundation, and the load acting on the rising part can be distributed, so the number of piles that need to be placed on the rising part can be reduced and the piles can be spaced more widely.In addition, there are cases where it is not necessary to place piles at the tip of the peninsula part.

[0062] As such, whether or not piles should be placed depends on the type of foundation and the condition of the ground, and if the system requires input of such detailed information, the input work that user U must perform becomes cumbersome. On the other hand, if the pile placement is calculated based on the design principles of a slab foundation, when a strip foundation is actually used, the number of piles will be insufficient and the designed strength will not be ensured. Conversely, if the pile placement is calculated based on the design principles of a strip foundation, when a slab foundation is actually used, the number of piles will be redundant, resulting in increased costs. In this way, the system 100 uses a learning data set obtained by learning various framing plans, such as framing plans designed for slab foundations and framing plans designed for strip footings, and can therefore automatically calculate statistically whether piles should be placed to ensure a necessary, sufficient, and appropriate placement without the need to input various conditions, thereby reducing the workload of the user U.

[0063] Finally, in S32: placement determination step, it is determined whether or not to place a stake at a position higher than a preset threshold based on the probability calculated in S31: probability calculation step. Here, the placement of the stakes can be determined based on the probability of a given location, with the center of the area where the probability is relatively high. For example, using the probability value integrated over a 10-pixel range, areas showing higher values ​​than the surrounding areas can be extracted as candidate placement locations, and the stakes can be determined to be placed at the center of the range of probability values ​​(integrated values) higher than a threshold for each candidate placement location. In this case, the extracted positions may have slightly different dimensions, but if the deviation is within a specified range, the candidate placement locations can be aligned in a grid by aligning them to the average value of the deviation.

[0064] Next, each placement candidate position is ranked. When ranking, the rank is adjusted taking into consideration the probability of adjacent placement candidate positions and information such as the shape and probability of the foundation of the adjacent placement candidate positions. Specifically, even if the probability of placement in the middle part of a long foundation on a straight line is high, if the placement probability of four piles in a rectangular shape including the foundation is extremely high, it is highly likely that the foundation will be able to support the surface, so processing can be performed to intentionally lower the rank of the piles in the middle part of the foundation. Furthermore, even if the probability is low, if the position is a corner or a T-shaped part, it is highly likely that a pile would be better to place, so processing can be performed to intentionally raise the rank of the pile.

[0065] One example of ranking is a method in which 1 is the highest rank and the rank is ascending. In this case, the larger the number, the lower the rank. When ranking is performed using this method based on the probability distribution in Figure 6, the result is shown in Figure 7. The round shapes are stake markers that indicate stakes, and the numbers displayed nearby are the ranks. If necessary, the stake number can also be displayed inside the stake marker.

[0066] Then, it is decided to place stakes at positions where the rank is equal to or greater than a preset threshold value. For example, if a stake is to be placed at a position where the rank number is less than 200, it will be placed at the position of the stake marker filled in black in Figure 7. In addition to automatically placed piles, the system 100 also allows forcible manual placement of piles at any desired position. For example, positions ranked in the 500s are forcibly added as lower-ranked candidate placement positions so that they can be added as needed, since the distance between adjacent positions is far compared to the standard pitch. Furthermore, positions ranked in the 600s are not far from adjacent candidate placement positions, but are forcibly added as lower-ranked candidate placement positions so that they can be added as needed for safety reasons (for example, when it is necessary to design a continuous footing).

[0067] The system 100 can also automatically create a projection diagram of the position of each placed pile based on the coordinate position of the placed pile, using the outermost dimension information and reference pitch information that have been read in advance. At this time, a pile marker that serves as the coordinate reference can be selected, and the vertical projection dimensions, horizontal projection dimensions, and diagonal dimensions of the outermost shape can be displayed. When creating an eviction drawing, if the coordinate values ​​of each stake marker are close to the standard pitch, the system may have a function to align them to the standard pitch. This allows the system to output an eviction drawing that is consistent with the standard pitch, even if the coordinate values ​​of stakes placed by the automatic element placement method of the system 100 are incomplete values. Note that if the value is clearly far from the standard pitch, the value can be considered to be the correct value, and the system can simply round the value without aligning it.

[0068] In the placement determination step S32 of the present system 100, the number of stakes can be adjusted to increase or decrease as needed. This adjustment is performed by varying the rank threshold value used to determine the placement of the stakes described above. For example, if you want to increase the number of stakes, you can lower the threshold value (increase the rank number), and the stake will be placed at the placement reference point with the next highest rank. In Figure 8, a stake would be added at the position indicated by the square. Also, if you want to reduce the number of stakes, you can raise the threshold value (decrease the rank number), and the stakes already placed will be deleted, starting with the lowest rank. In Figure 8, the stake at the position indicated by the triangle would be deleted.

[0069] As described above, in the present system 100, piles can be placed in necessary and sufficient positions relative to the foundation plan using a learning dataset that has been previously machine-learned to learn the pile plan.

[0070] "Variation 1" Next, an automatic element placement system 101 according to a modified example of the present invention will be described with reference to Fig. 9. In the following description, the same parts will be designated by the same reference numerals, and duplicated descriptions will be omitted.

[0071] This modified example differs from the configurations of Figures 1 to 8 in that element E is a rafter, the target component P on which the rafter should be placed is a beam, and the placement target plan D is a beam plan. In this modified example, a learning dataset is used that has been machine-learned using a large amount of roof floor plans with roof rafters arranged.

[0072] Then, following the same steps as in Figure 3, the reading means 1 reads the beam plan, and the image analysis means 2 reads the beam shape, position, dimensions, etc. Then, the calculation means 3 can automatically place the rafters in the appropriate positions on the beams.

[0073] In this modification, element E is a rafter, but it is also possible to place a column on the upper floor as element E in the beam plan. In this case, it is possible to choose whether to place a rafter or a column, or to place both. By configuring it as described above, it is possible to create a system that automatically places, as element E, various elements that are displayed on architectural floor plans in addition to piles. [Explanation of symbols]

[0074] 100,101 Element automatic placement system 1. Reading method 2. Image analysis methods 3 Calculation means 31 Probability calculation method 32 Placement determination means C. Customer terminal D Layout map The E element L training dataset P Target material U User S scanner T Data to be placed

Claims

1. An automatic element placement system for placing elements at appropriate positions in a building plan, a reading means for reading a layout plan showing target components on which the elements are to be placed, and setting the read plan as layout target data; A computing means that uses a learning dataset obtained by machine learning the features of target components whose elements are already placed in other framing plans; Equipped with The element automatic placement system is characterized in that the calculation means is a means for automatically placing the elements according to characteristics of target components in the placement target data by calculation based on the learning data set.

2. the target member is a combination of linear members, the calculation means includes a probability calculation means and an arrangement determination means; The probability calculation means is a means for using a learning data set obtained by machine learning the positional features of elements arranged next to the straight line portions of the target component in the other framing plan and the positional features of elements arranged at the intersection of the straight line portions, a means for determining, based on the learning data set, a probability that the element should be placed at a predetermined position in the target component of the placement target data; 2. The automatic element placement system according to claim 1, wherein the placement determination means is a means for placing the elements based on the probability calculated by the probability calculation means for each predetermined position in the target component of the placement target data.

3. The probability calculation means is a means that uses a learning data set obtained by machine learning a first feature in the other framing plan, which is a feature of the position of an element arranged at a corner of a target component, a second feature, which is a feature of the position of two elements arranged side by side on a straight line portion, and a third feature, which is a feature of the position of four elements arranged in a rectangular shape, a probability that the element should be placed at a predetermined position on the target component of the placement target data is calculated as a comprehensive probability by superimposing the probabilities calculated from the first feature, the second feature, and the third feature, The automatic element placement system according to claim 2, wherein the placement determination means calculates a rank based on the overall probability calculated by the probability calculation means for each predetermined position in the target component of the placement target data, and is a means for placing the element at a position that is equal to or higher than a predetermined rank.

4. 4. The automatic element placement system according to claim 2 or claim 3, wherein the placement determination means includes means for deleting an element at a predetermined position or adding an element at a predetermined position based on the relationship between the probability for each position calculated by the probability calculation means and the distance between adjacent elements, for the position where the element has been determined to be placed.

5. The layout plan is a foundation plan, the elements are piles, and the target members are foundations, The automatic element placement system according to claim 4, wherein the learning dataset is obtained by machine learning the characteristics of foundations where piles have already been placed in other pile plans.

6. An automatic element placement method for placing elements at appropriate positions in a building plan, comprising: a reading step of reading a layout plan showing target components on which the elements are to be placed, and setting the layout plan as placement target data; A calculation step using a learning data set obtained by machine learning the characteristics of the placement target component whose elements have already been placed relative to the target component in another framing plan; Equipped with The element automatic placement method, characterized in that the calculation step is a step of automatically placing the elements according to characteristics of target components in the placement target data by calculation based on the learning data set.

7. the target member is a combination of linear members, the calculation step includes a probability calculation step and an arrangement determination step, The probability calculation step is a step of using a learning data set obtained by machine learning a first feature in the other framing plan, which is a feature of the position of an element arranged at a corner of a target component, a second feature, which is a feature of the position of two elements arranged side by side on a straight line portion, and a third feature, which is a feature of the position of four elements arranged in a rectangular shape, determining a probability that the element should be placed at a predetermined position on the target component of the placement target data by superimposing the probabilities calculated from the first feature, the second feature, and the third feature, respectively; the placement determination step is a step of calculating a rank based on the overall probability calculated for each predetermined position in the target component of the placement target data by the probability calculation step, and placing the element at a position equal to or higher than a predetermined rank, 7. The method for automatically placing elements according to claim 6, further comprising a step of deleting an element from a predetermined position or adding an element to a predetermined position based on the relationship between the probability for each position calculated by the probability calculation step and the distance between adjacent elements, for the positions where the element has been decided to be placed.

8. The layout plan is a foundation plan, the elements are piles, and the target members are foundations, The automatic element placement method according to claim 7, wherein the learning dataset is obtained by machine learning the characteristics of foundations where piles have already been placed in other pile plans.

Images