JP7601378B2 - 3D CAD model part search method - Google Patents

Description

Application of Article 30, Paragraph 2 of the Patent Act Published in the proceedings of the 12th Forum on Data Engineering and Information Management, posted on the website (URL: https://cms.deim-forum.org/deim2020/paper/?p=P333a3d8af1f389366c44b382b3e5314be57146ef2) on February 14, 2020

Application of Article 30, Paragraph 2 of the Patent Act Announced at the 12th Forum on Data Engineering and Information Management held online on March 4, 2020

The present invention relates to a method for searching parts of a three-dimensional CAD model.

In recent years, in the manufacturing industry and other fields, model design is being carried out using 3D computer design tools (CAD: Computer-Aided Design). In order to improve the efficiency of design work, it is essential to utilize accumulated 3D model data. For this reason, research and development is being conducted on various technologies such as searching, clustering, classification, and detection of 3D model data.

One of the common requirements for technologies that utilize 3D model data is the ability to calculate the similarity between models. Furthermore, much research has been conducted into techniques for searching between models as part of research into calculating the similarity between models. Although there has been active research into 3D model search techniques up to now, most of the research has been on search techniques that only consider information about the shape of the model, and no search techniques have been found that allow for rigorous comparison of the layout of parts.

Three-dimensional models created for machine design and other purposes are often assembly models made up of multiple parts. In particular, in the design of large, complex machines such as automobiles, parts are assembled across multiple layers, resulting in the creation of huge models containing a huge number of parts. Such models that combine multiple parts are called assembly models, and each grouping at each layer is called a subassembly model. With a part being the smallest unit of a 3D CAD model, a subassembly model is a collection of parts, and an assembly model is a collection of subassembly models. Due to their complexity, it has been difficult to apply partial search technology to assembly models.

As a search method for searching for such subassembly models, Patent Document 1 proposes a 3D CAD model search method that searches for each component that makes up the 3D CAD data and accesses the 3D CAD data of the finished product from the component.

JP 2018-136946 A

However, previously proposed search methods perform a 2D Radon transform on the projection image of 3D CAD data, and then use the properties of the resulting sinogram to perform a comparison to achieve partial searches. This creates a problem in that the amount of data for features calculated from the 3D data becomes large, making it still insufficient for rapid searches.

One of the objectives of the present invention is to provide a method for searching parts of a 3D CAD model that allows for fast and highly accurate searches.

A three-dimensional CAD model partial search method according to the present invention is a three-dimensional CAD model partial search method for comparing a search model with a comparison model stored in a database and evaluating the degree of difference, the search model and the comparison model are three-dimensional CAD data of an assembly model composed of a plurality of subassembly models, and the method includes a first calculation step in which a computer calculates a feature amount of the search model, a second calculation step in which the computer calculates a feature amount of the comparison model, and a comparison step in which the computer compares the feature amounts of the search model and the comparison model and calculates the degree of difference between the search model and the comparison model, and in the first calculation step and the second calculation step, for all subassembly models constituting the search model and the comparison model, the subassembly models and the assembly model excluding the subassembly models are each subjected to a three-dimensional Radon transform, and the feature amount is derived from the transformed value.

The present invention provides a method for searching parts of a 3D CAD model that allows for fast and highly accurate searches.

FIG. 1 is a perspective view of an assembly model shown as an example of a search model. FIG. 2 is a perspective view of an assembly model shown as an example of a comparison model. FIG. 3 is a perspective view of an assembly model shown as an example of a comparison model. FIG. 4 is a schematic diagram showing an example of a three-dimensional CAD model part search method according to this embodiment, and mainly shows a comparison step. FIG. 5 is a schematic diagram showing an example of a three-dimensional CAD model part search method according to this embodiment, and mainly shows the first direction fixing step of the comparison step.

The three-dimensional CAD data (hereinafter simply referred to as CAD data) targeted in the three-dimensional CAD model part search method (hereinafter simply referred to as the part search method) of this embodiment is an assembly model that includes multiple subassembly models.

The partial search method of this embodiment is a method of comparing a search model with a comparison model stored in a database and evaluating the degree of difference. Note that the partial search method of this embodiment can be used in both cases, whether the search model includes a part of the comparison model or the comparison model includes a part of the search model. In this embodiment, a case will be described in which one included in the search model is found from among multiple comparison models stored in a database.

The search model and the comparison model are three-dimensional CAD data of an assembly model composed of a plurality of subassembly models.
Moreover, the dissimilarity is an index that indicates the extent to which the comparison model is not included in the search model (how conceptually separated the search model and the comparison model are). The dissimilarity takes a positive value of 0 or greater. If the search model includes the comparison model, the dissimilarity becomes 0. Therefore, the smaller the dissimilarity value, the higher the extent to which the comparison model is included in the search model.

In this specification, an assembly model is a three-dimensional model that is constructed by combining multiple subassembly models. Also, in this specification, a subassembly model is a three-dimensional model that constitutes a part of an assembly model. Therefore, an assembly model has multiple different or identical subassembly models. A subassembly model is a three-dimensional model that is constructed by combining multiple parts. Also, a subassembly model may be a three-dimensional model that has only a single part.

Illustrative Description
The partial search method of this embodiment will be described by way of example with reference to Figures 1 to 3. The description here is merely an example of the concept of the partial search method, and in the strict sense, the method is different from the partial search method of the embodiment described later.

Figures 1 to 3 are perspective views of the first to third assembly models 11, 12, and 13, respectively. In Figures 1 to 3, the subassembly models included in each assembly model are classified by pattern.

The first assembly model 11 shown in FIG. 1 has a base portion 10, a first sub-assembly model 21, a second sub-assembly model 22, a third sub-assembly model 23, and a fourth sub-assembly model 24.
The second assembly model 12 shown in FIG. 2 has a base portion 10 , a first sub-assembly model 21 , a second sub-assembly model 22 , and a fifth sub-assembly model 25 .
The third assembly model 13 shown in FIG. 3 has a base portion 10, a first sub-assembly model 21, a fifth sub-assembly model 25, and a sixth sub-assembly model 26.
The base portion 10 is also considered to be one of the subassembly models.

Here, we consider determining the degree of difference between the first assembly model 11 and the second and third assembly models 12, 13. That is, in this example, the first assembly model 11 corresponds to the search model, and the second and third assembly models 12, 13 correspond to the comparison models.

The second assembly model 12 has a fifth subassembly model 25 that the first assembly model 11 does not have. In addition, the third assembly model 13 has a fifth subassembly model 25 and a sixth subassembly model 26 that the first assembly model 11 does not have.

The fifth subassembly model 25 included in the second and third assembly models 12, 13 is part of the third subassembly model 23 included in the first assembly model 11. On the other hand, the sixth subassembly model 26 included in the third assembly model 13 is composed of the same parts as the subassembly model 22 included in the first assembly model 11, but has a different positional relationship with the other subassembly models than the first assembly model 11.

As a result of comparing the first assembly model 11 with the second and third assembly models 12 and 13, all of the parts included in the second assembly model 12 are included in the first assembly model 11 in the same arrangement. In contrast, all of the parts included in the third assembly model 13 are included in the first assembly model 11, but the arrangement of the sixth subassembly model 26 is different from that of the second subassembly model 22 included in the first assembly model 11. Therefore, in such an example, it is determined that the second assembly model 12 has a high similarity (i.e., a low degree of difference) with the first assembly model 11. In this way, the partial search method of this embodiment takes into account the arrangement of the subassembly models.

The first to third assembly models 11, 12, and 13 shown here have the same orientation. However, in the data of the assembly models that are actually created, the reference direction may differ for each model. For this reason, the partial search method needs to take into account the relative positional relationship between the assembly models when the relative orientation is changed.

<Embodiment>
The following describes the embodiments.
In this embodiment, the assembly model to be searched is called a search model, and the assembly model to be searched that is stored in the database is called a comparison model.

The partial search method of this embodiment includes a first calculation step, a second calculation step, a comparison step, and an output step. Note that the order of the first calculation step and the second calculation step may be reversed. In addition, the comparison step is performed after the first and second calculation steps because it uses the values (feature amounts) obtained in the first and second calculation steps.

The first calculation step and the second calculation step are performed mainly in the same manner, but the CAD data to be calculated is different. The first calculation step is a step of calculating the feature values from the search model to be searched. On the other hand, the second calculation step is a step of calculating the feature values of the comparison models stored in the database.

The algorithm for the partial search method of this embodiment is shown in Table 1 as the main algorithm.

We now define the main algorithm and the symbols used below.
M _comp represents the search model.
M _part represents the comparison model.
The search model M _comp and the comparison model M _part are collections of subassembly models sa.
V _comp and V _part are sets of directions for performing a three-dimensional Radon transform. The element v of V _comp and V _part is a two-dimensional vector that represents a direction in spatial polar coordinates. More specifically, the direction is specified by points uniformly distributed on a unit sphere.

In this embodiment, the number of directions in which the three-dimensional Radon transform is performed in the first calculation step (the number of elements included in V _comp ) is smaller than the number of directions in which the three-dimensional Radon transform is performed in the second calculation step (the number of elements included in V _part ). In general, the step of calculating the feature amount from the search model (the first calculation step) is performed online, and the step of calculating the feature amount from the comparison model (the second calculation step) is performed and stored in advance in the same server. By suppressing the number of directions in which the three-dimensional Radon transform is performed in the first calculation step, the calculation of the feature amount performed online can be performed at high speed, and by increasing the number of directions in which the three-dimensional Radon transform is performed in the second calculation step, the accuracy of the search results can be ensured.

The main algorithm performs the dissimilarity calculation between the search model and the comparison model in three steps:
(1) First calculation step (2) Second calculation step (3) Comparison step Each step will be described in more detail below.

<First and second calculation steps>
In the first calculation step, the feature quantities are calculated by performing a three-dimensional Radon transform on a search model from multiple directions. Similarly, in the second calculation step, the feature quantities are calculated by performing a three-dimensional Radon transform on a comparison model from multiple directions.

When a search model M _comp and a comparison model M _part are given, the correspondence of subassembly models between two assembly models (search model and comparison model) is generally unknown. In addition, the relationship between the positions and orientations of the two assembly models is also unknown. Therefore, in the partial search method of this embodiment, all combinations of subassembly models and all combinations of orientations are compared, and a combination that minimizes the degree of difference between the assembly models is selected.

The algorithm for the three-dimensional Radon transform is shown in Table 2 as the first sub-algorithm. The first sub-algorithm is part of the first and second calculation steps, and is an algorithm for deriving the three-dimensional Radon transform. The first sub-algorithm is the algorithm expressed as Radon3D in the main algorithm shown in Table 1.

In the first sub-algorithm, when performing a three-dimensional Radon transform along a direction vεV, first the model is rotated so that v is in the positive direction of the x-axis.
In the first sub-algorithm, in the function sa'=Rotate(sa, v) in the second line, v in the sub-assembly model sa before rotation is rotated so that it is in the positive direction of the x-axis of sa'.

Next, calculate the cross-sectional area of the model on a plane perpendicular to the x-axis along the positive x-axis direction. The function calculateCrossSectionArea(sa', x) on the fourth line of the first sub-algorithm is a function that calculates the cross-sectional area of model sa' where the value in the x-axis direction is x. Here, x changes at each specified sampling interval, and scans the section in which the model exists.

Actual 3D CAD models are stored in a variety of data formats. One of the standard shape description formats is the STL (Standard Triangulated Language) data format. The STL file format is a polygon format that represents the surface of a 3D solid as a collection of small triangular faces. The STL data format is a format that can be converted from all major 3D data formats. In this format, the cross-sectional area can be calculated from geometric calculations. The feature value obtained in this way, feature, is a quadratic real array of (number of directions) x (number of samples) with non-negative values.

In the first calculation step, the feature quantity F _part of the comparison model M _part is calculated by executing the first sub-algorithm. In the first calculation step, the feature quantity F _comp of the search model M _comp is calculated by executing the first sub-algorithm.

According to this embodiment, in the first calculation step and the second calculation step, the feature amount is calculated by performing a three-dimensional Radon transform on the three-dimensional CAD model. The three-dimensional Radon transform is a calculation method for calculating the cross-sectional area of the three-dimensional CAD model in the direction relative to the origin (coordinates on a spherical surface) and the distance from the origin. Therefore, the feature amount obtained by the three-dimensional Radon transform is a scalar value linked to the direction and distance relative to the origin, and the amount of data can be reduced compared to the search method disclosed in Patent Document 1. As a result, the calculation time of the feature amount in the first calculation step and the second calculation step, as well as the calculation time in the comparison step described below, can be significantly reduced. In addition, since the calculation time is reduced, it becomes easier to increase the number of samples, and the accuracy of the evaluation of the dissimilarity can be improved.

According to this embodiment, in the first calculation step and the second calculation step, the cross-sectional area is calculated from multiple directions. Therefore, even when comparing a search model and a comparison model that have different poses, a feature amount that can effectively evaluate the difference is calculated.

In the first and second calculation steps of this embodiment, it is required to calculate the relative positional relationship of the subassembly models included in the assembly model as a feature. Therefore, the inventors came up with the idea of combining values obtained by performing a three-dimensional Radon transform on the subassembly model to be compared and the assembly model excluding the subassembly model, and treating the combined values as a feature.

According to this embodiment, in the first calculation step and the second calculation step, for all constituent subassembly models, the subassembly model and the assembly model excluding the subassembly model are each subjected to a three-dimensional Radon transform, and the transformed values are regarded as feature quantities.
That is, the feature value f _comp derived in the first calculation step consists of the result of the three-dimensional Radon transform of the subassembly model sa (f _comp , _sa ) and the result of the three-dimensional Radon transform of the assembly model excluding the subassembly model sa (f _comp , _others ).
Similarly, the feature value f _part derived in the second calculation step consists of the result of the three-dimensional Radon transform of the subassembly model sa (f _part , _sa ) and the result of the three-dimensional Radon transform of the assembly model excluding the subassembly model sa (f _part , _others ).

In the first calculation step and the second calculation step, the feature quantities of the subassembly model and the assembly model excluding the subassembly model are normalized, thereby making it possible to suppress the influence of the size of each component on the feature quantities.
More specifically, in the first calculation step and the second calculation step, the feature quantities of the subassembly model and the assembly model excluding the subassembly model are the sums of values obtained by individually performing a 3D Radon transform on all included parts and dividing the results by an approximation of the volume of the parts.

Figures 4 and 5 are schematic diagrams showing an example of a 3D CAD model part search method according to this embodiment. In each figure, subassembly models 31 and 32 of a comparison model 30 and subassembly models 41, 42, and 43 of a search model 40 are shown. In each figure, one cube represents one part. A collection of cubes represented by the same pattern represents one subassembly.

In this example, the comparative model 30 has a first subassembly model 31 and a second subassembly model 32. The first subassembly 31 of the comparative model 30 consists of two parts 31a and 31b. The second subassembly 32 of the comparative model 30 consists of one part 32a.

In this example, the search model 40 has a first subassembly model 41, a second subassembly model 42, and a third subassembly model 43. The first subassembly 41 of the search model 40 consists of one part 41a. The second subassembly 42 of the search model 40 consists of two parts 42a and 42b. The third subassembly 43 of the search model 40 consists of one part 43a.

Figure 4 shows combinations B1, B2, B3, B4, B5, and B6 of the subassembly models of the comparison model 30 and the search model 40 to be compared in the comparison step described below. In the comparison step, all combinations B1, B2, B3, B4, B5, and B6 in Figure 4 are compared in sequence.

Figure 5 is a diagram showing how the features of the first subassembly model 31 of the comparison model 30 and the first subassembly 41 of the search model 40 are handled when comparing them in the comparison step described below. Figure 5 is a diagram showing the comparison process for combination B1 in the table of Figure 4 during the comparison step, in which the first subassemblies 31, 41 of the comparison model 30 and the search model 40 are compared.

As shown in FIG. 5, the feature quantities of the comparison model 30 are calculated as the feature quantities of the first subassembly model 31 and the feature quantities of the assembly model excluding the first subassembly model 31 (i.e., the second subassembly model 32). Note that the step of calculating the feature quantities of the comparison model 30 is the first calculation step.

As described above, the features of the compared subassembly model and the assembly model excluding the compared subassembly model are the sum of the values obtained by individually performing a 3D Radon transform on all included parts and dividing the results by an approximation of the part's volume.

The first subassembly model 31 of the comparative model 30 includes two parts 31a and 31b. Therefore, the sum of the values obtained by performing a three-dimensional Radon transform on the two parts 31a and 31b individually and dividing the result by the approximate value of the volume of each part becomes the feature value of the first subassembly model 31 of the comparative model 30.

On the other hand, the assembly model excluding the first subassembly model 31 of the comparison model 30 (i.e., the second subassembly model 32) contains only one part 32a. Therefore, the value obtained by dividing the value obtained by performing the three-dimensional Radon transform on the part 32a by the approximate value of the volume of the part 32a becomes the feature value of the assembly model excluding the second subassembly model 31.

Furthermore, as the feature quantities of the search model 40, the feature quantities of the first subassembly model 41 and the feature quantities of the assembly model excluding the first subassembly model 41 (i.e., the second and third subassembly models 42 and 43) are calculated. Note that the step of calculating the feature quantities of the search model 40 is the second calculation step.

The first subassembly model 41 of the search model 40 contains only one part 41a. Therefore, the value obtained by dividing the three-dimensional Radon transform value of the part 41a by the approximate value of the volume of the part 41a becomes the feature value of the first subassembly model 41 of the search model 40.

On the other hand, the assembly model excluding the first subassembly model 41 of the search model 40 (i.e., the second and third subassembly models 42, 43) contains a total of three parts 42a, 42b, and 43a. Therefore, the three parts 42a, 42b, and 43a are individually subjected to a three-dimensional Radon transform, and the sum of the values obtained by dividing the results by the approximate values of the volumes of the respective parts becomes the feature quantity of the assembly model excluding the first subassembly model 41.

According to this embodiment, the subassembly model and the assembly model excluding the subassembly model are each individually subjected to a three-dimensional Radon transform to obtain feature quantities. By comparing these two feature quantities simultaneously, it is possible to calculate feature quantities that take into account the relative positional relationship of the subassembly model to the assembly model. As a result, it is possible to calculate the degree of difference between the search model and the comparison model, taking into account the geometric arrangement of the subassembly model.

According to this embodiment, the value obtained by three-dimensional Radon transform of the subassembly model and the value obtained by three-dimensional Radon transform of the assembly model excluding the subassembly model are treated as feature quantities separately. Therefore, compared to the case where the sum of these values is used (see Patent Document 1), for example, the information representing the relative positional relationship is not diluted, and it is easy to clearly determine the degree of difference by the feature quantity.
In addition, according to this embodiment, the data capacity of the feature amount can be made smaller relative to the amount of information by adopting the three-dimensional Radon transform as a method for calculating the feature amount, as compared with the search method disclosed in Patent Document 1. Therefore, even if the value obtained by three-dimensional Radon transform of the subassembly model and the value obtained by three-dimensional Radon transform of the assembly model excluding the subassembly model are each used as feature amounts individually, it is possible to prevent the overall data capacity from becoming too large.

<Comparison step>
The comparison step is a step of comparing the feature amounts of the search model and the comparison model and calculating the degree of difference between the search model and the comparison model.

As shown in the main algorithm of Table 1, the (3) comparison step includes a (3-1) direction fixing first step, a (3-2) direction fixing second step, and a (3-3) decision step.
(3-1) Fixing the Orientation The first step is to fix the orientation and then calculate the dissimilarity (specific combination dissimilarity) between the search model and the comparison model for all possible correspondences between the subassemblies based on the features of the search model and the comparison model.
(3-2)
The second direction fixing step is a step of selecting the smallest value from among the multiple specific combination dissimilarities derived in the first direction fixing step, and setting the minimum value as the dissimilarity in this direction (specific direction dissimilarity).
The (3-3) determination step is a step of selecting the smallest specific direction dissimilarity in each direction derived in the second direction fixing step, and deriving the dissimilarity between the comparison model and the search model.

(First step of fixing direction)
The first direction fixing step is a step of calculating the dissimilarity for all combinations of subassembly models of the search model and the comparison model for a specific direction. In the first direction fixing step, the dissimilarity between all subassembly models of the search model and all subassembly models of the comparison model is calculated in a brute force manner.

In the first direction fixing step, the search model and comparison model are compared for a specific fixed direction. As shown in the determination step described below, the specific fixed direction is changed to any direction, and finally the search model and comparison model are compared for all directions.

Here, the dissimilarity of each of the combinations of subassembly models calculated in the first direction fixing step is called the "specific combination dissimilarity." In the first direction fixing step, the same number of specific combination dissimilarity is calculated as the number of combinations of subassembly models calculated in the fixed direction by the brute force method.

In other words, the first direction fixing step is a step in which the direction is fixed and the sum of the feature differences for all combinations of all subassembly models of the search model and all subassembly models of the comparison model is calculated as a specific combination dissimilarity.

This point will be described with reference to the schematic diagram of FIG.
In this example, the comparison model 30 has two subassembly models (first and second subassembly models 31, 32). On the other hand, the search model 40 has three subassembly models (first, second and third subassembly models 41, 42, 43). In the first direction fixing step, the first, second and third subassembly models 41, 42, 43 of the search model 40 are compared with the first subassembly model 31 of the comparison model 30 in a fixed specific direction, and the first, second and third subassembly models 41, 42, 43 of the search model 40 are compared with the second subassembly model 32 of the comparison model 30. As a result, a number of specific combination dissimilarity degrees corresponding to the product (2×3 in FIG. 4 ) of the number of subassembly models of the comparison model 30 and the number of subassembly models of the search model 40 are calculated.

When the feature quantities f _comp and f _part to be compared of the search model M _comp and the comparison model M _part are determined in the first and second calculation steps, an algorithm for calculating a specific combination dissimilarity is shown as a second sub-algorithm in Table 3. The second sub-algorithm is an algorithm expressed as ComputePartialDissimilarity in the main algorithm shown in Table 1.

In general, the posture and position of an assembly model in a three-dimensional space differ depending on the data. Therefore, in order to derive a difference in feature quantities that is robust against rotation and translation, the feature quantities are translated to derive the difference. However, since relative position information of a subassembly model in an assembly is necessary, the feature quantities (f _comp , _others , f _part , _others ) of the entire assembly model excluding the subassembly model are translated at the same time as translating the subassembly model of interest, and the feature quantity difference is similarly derived for these as well.

That is, in the comparison step, the features to be compared are translated relative to each other, and the value that minimizes the sum of the differences between the features is taken as the dissimilarity.

According to this embodiment, the feature quantities of the search model and the comparison model are translated relatively to calculate the dissimilarity, so the relative positional relationship between the search model and the comparison model can be comprehensively checked to calculate the optimal dissimilarity, and the deviation in the relative positional relationship can be offset.

The feature quantity fpart, _sa of the subassembly model of the comparison model and the feature quantity _fcomp , _sa of the subassembly model of the search model include information on the shape and arrangement of the subassembly model under consideration, and the _difference ( _fpart , _sa - _fcomp , _sa ) is a value that represents the difference in shape and arrangement. When the comparison model is completely included in the search model, the elements of ( _fpart , _sa - _fcomp , _sa ) do not include positive values, and are only 0 or negative values.

In this case, as in lines 7 and 8 of the second sub-algorithm, by adopting the greater of the difference value and 0, if there is a perfect partial match, the value is set to 0, otherwise it represents the extent to which the non-zero difference is not included in the search model, and the value of the excess part becomes the difference between the features.

In addition, the feature amount f _comp , _others of the search model excluding the subassembly model and the feature amount f _part , _others of the comparison model excluding the subassembly model are also compared at the same time. As described with reference to FIG. 5, these are each calculated as the sum of the feature amounts of the parts that constitute the target subassembly model.

In lines 10 and 11 of the second sub-algorithm, the difference in the features of the sub-assembly model and the difference in the features of the assembly model excluding the sub-assembly model are each divided by the number of parts included in the sub-assembly model in the comparison model. The divided values are then used to compare the differences in the features in the comparison step. This makes it possible to prevent the number of parts included in the sub-assembly model from affecting the difference in the features.

As shown diagrammatically in FIG. 5, in the comparison step, it is preferable to use the sum of the three-dimensional Radon transform of the components constituting each of the subassembly models calculated in the first and second calculation steps and the assembly model excluding the subassembly models, divided by an approximate value of the volume of the components, as the feature quantities of the subassembly models calculated in the first and second calculation steps.

That is, in the first direction fixing step, for the search model and the comparison model, the specific combination dissimilarity is calculated based on the sum of the difference in feature amounts between the subassembly models being compared and the difference in feature amounts between the assembly models excluding the subassembly models being compared. The feature amount value obtained by the three-dimensional Radon transform is affected by the volume of the model. In this embodiment, the specific combination dissimilarity is calculated based on the sum of the results of three-dimensional Radon transforming the parts and dividing the results by an approximation of the volume of the parts, thereby reducing the effect of the part volume on the calculation of the specific combination dissimilarity.

The sum (integral value) of the cross-sectional areas calculated by the three-dimensional Radon transform can also be used as an approximation of the volume. This allows normalization to be performed without a separate volume calculation, thereby shortening the processing time.

Furthermore, in the first direction fixing step, it is preferable to calculate the specific combination dissimilarity by dividing the sum of the difference in feature amounts between the subassembly models to be compared in the search model and the comparison model and the difference in feature amounts between the assembly models excluding the subassembly model to be compared by the number of parts in the subassembly model to be compared in the comparison model. By dividing the above sum by the number of parts in the subassembly model in the comparison model, the effect of the number of parts on the calculation of the specific combination dissimilarity can be reduced.

This point will be described with reference to the schematic diagram of FIG.
The first direction fixing step is a step of fixing the direction and calculating the sum of the differences in the feature amounts for all combinations of subassembly models of the comparison model 30 and the search model 40 as a specific combination dissimilarity.

In Figure 5, the first subassembly model 31 of the comparison model 30 is compared with the first subassembly model 41 of the search model 40.

The feature amount of the first subassembly 31 of the comparison model 30 is the sum of two parts 31a and 31b individually subjected to three-dimensional Radon transform, divided by the approximate value of the volume of each part, and normalized. Meanwhile, the feature amount of the first subassembly 41 of the search model 40 is the part 41a subjected to three-dimensional Radon transform, divided by the approximate value of the volume of the part 41a, and normalized.
In this step, the difference (first difference D1) between the feature amount of the first subassembly 31 of the comparison model 30 and the feature amount of the first subassembly 41 of the search model 40 is calculated.

The feature amount of the assembly model excluding the first subassembly model 31 of the comparison model 30 (i.e., the second subassembly model 32) is obtained by performing a three-dimensional Radon transform on the part 32a, dividing the result by an approximate value of the volume of the part 32a, and normalizing the result. On the other hand, the feature amount of the assembly model excluding the first subassembly model 41 of the search model 40 (i.e., the second and third subassembly models 42, 43) is obtained by performing a three-dimensional Radon transform on the three parts 42a, 42b, and 43a individually, dividing the result by an approximate value of the volume of each part, and normalizing the result.
In this step, the difference (second difference D2) between the features of the assembly model excluding the first subassembly model 31 of the comparison model 30 and the assembly model excluding the first subassembly model 41 of the search model 40 is calculated.

In the direction fixing step, the sum of the first difference D1 and the second difference D2 is further divided by the number of parts in the first subassembly model 31 compared by the comparison model 30 to calculate the specific combination dissimilarity. In the example of FIG. 5, the number of parts in the first subassembly model 31 compared by the comparison model 30 is 2. Therefore, in this example, the sum of the first difference D1 and the second difference D2 is divided by 2.

According to this embodiment, in the first calculation step, the feature amount of the first subassembly model 31 of the comparison model 30 and the assembly model excluding the subassembly model 31 (i.e., the subassembly model 32) is the sum of values obtained by individually performing three-dimensional Radon transformation on all included parts and dividing the result by an approximate value of the volume of the part. Similarly, in the second calculation step of this embodiment, the feature amount of the first subassembly model 41 of the search model 40 and the assembly model excluding the subassembly model 41 (i.e., the subassembly models 42, 43) is the sum of values obtained by individually performing three-dimensional Radon transformation on all included parts and dividing the result by an approximate value of the volume of the part. Also, in the first direction fixing step of this embodiment, for the comparison model 30 and the search model 40, the specific combination dissimilarity is calculated based on the sum of the difference in the feature amount between the subassembly models (first subassembly models 31, 41) to be compared and the difference in the feature amount between the assembly models excluding the subassembly model to be compared. More specifically, this sum is further divided by the number of parts (two in this configuration) in the first subassembly model 31 compared with the comparison model 30 to calculate the specific combination dissimilarity. By performing these operations, all subassembly models are evaluated with equal weighting, regardless of the volume and number of parts included in the subassembly models, and the dissimilarity between assembly models can be accurately evaluated.

(Second step of fixing direction)
The second direction fixing step is to select the dissimilarity for a specific direction (hereinafter referred to as the "specific direction dissimilarity"). The specific direction dissimilarity derived in this step represents the dissimilarity between the search model and the comparison model when limited to a specific direction.

In the second direction fixing step, the smallest value is selected from the multiple specific combination dissimilarity values derived in the first direction fixing step. The selected smallest value is then defined as the dissimilarity in this direction (specific direction dissimilarity).

In other words, the second direction fixing step is a step in which a combination of subassembly models is derived that minimizes the specific combination dissimilarity calculated in the first direction fixing step, and the specific combination dissimilarity at this point is set as the specific direction dissimilarity for that direction.

This point will be described with reference to the schematic diagram of FIG.
In this example, the comparison model 30 has two subassembly models 31 and 32, and the search model 40 has three subassembly models 41, 42, and 43. Therefore, the comparison model 30 and the search model 40 compare the specific combination dissimilarity of the combinations B1, B2, B3, B4, B5, and B6 of the six subassembly models, and further select the smallest value among the specific combination dissimilarity of each of the six combinations B1, B2, B3, B4, B5, and B6 as the specific direction dissimilarity.

The second direction fixing step of the main algorithm shown in Table 1 is now explained in detail.
First, in line 16 of the main algorithm, for the features f _comp and f _part obtained in the processes up to this point, from among all combinations of sa _comp ∈ M _comp and sa _part ∈ M _part , a combination that minimizes the sum of the differences between the features is solved. In this embodiment, the number of subassemblies of the comparison model is less than the number of subassemblies of the search model. In this case, all combinations mean all injections (mappings) from the set of subassemblies of the comparison model to the set of subassemblies of the search model.

The sum of the differences between the features in the second direction fixing step is taken as the sum of the differences between the assemblies. The problem of finding this optimal combination is formulated and solved as a combinatorial optimization problem called an assignment problem.

(Decision Step)
The determination step is a step of determining the smallest dissimilarity for all possible directions. Through the determination step, a dissimilarity indicating the dissimilarity between the search model and the comparison model is calculated. In this embodiment, the number of directions (the number of elements included in V _comp ) for performing the three-dimensional Radon transform when calculating the feature amount of the search model is less than the number of directions (the number of elements included in V _part ) for performing the three-dimensional Radon transform when calculating the feature amount of the comparison model. The dissimilarity is obtained by finding an injection from V _comp to V _part that results in the smallest dissimilarity, and deriving the final dissimilarity between the assemblies from the dissimilarity at that time. The injection that results in the smallest dissimilarity is formulated and solved as an assignment problem.

In the determination step, first, the first direction fixing step and the second direction fixing step are performed for all directions. That is, the first direction fixing step and the second direction fixing step are repeated while changing the direction. In this way, the specific direction dissimilarity is obtained for all combinations of directions. Next, the smallest specific direction dissimilarity is selected from these. Furthermore, the smallest specific direction dissimilarity is divided by the number of subassemblies in the comparison model to derive the dissimilarity.

In other words, the determination step is a step in which a first direction fixing step and a second direction fixing step are performed for all directions to derive the direction in which the specific direction dissimilarity is minimal, and the dissimilarity between the search model and the comparison model is derived from the specific direction dissimilarity at this time.

Furthermore, in this embodiment, as shown in line 19 of the main algorithm (see Table 1), the result obtained in the determination step (the degree of difference between the search model and the comparison model) is divided by the number of subassemblies (|M _part |) included in the comparison model. That is, according to this embodiment, the degree of difference is normalized by the number of subassemblies included in the comparison model. This makes it possible to suppress the number of subassemblies in the comparison model from affecting the degree of difference. When comparing the degree of difference between the search model and multiple comparison models, accurate search results can be calculated even if there is a difference in the number of subassemblies between the comparison models.

The search performed based on the dissimilarity obtained in this way is robust to the model's orientation and position, does not depend on the subassembly model's identifier, and is a partial search that takes into account both the model's shape and layout.

Here, we have described a case where one comparison model is prepared as the object to be searched. However, in an actual implementation, multiple comparison models are stored in the database as objects to be searched. It is preferable that the features of each comparison model are calculated in advance and stored as related information.

When searching a search model against multiple assembly models in a database, the search model is compared in order with multiple comparison models in the database to derive the comparison model with the smallest degree of difference. This makes it possible to provide the searcher with one of the multiple comparison models in the database that is most similar to the search model.

Although various embodiments of the present invention have been described above, each configuration and their combinations in each embodiment are merely examples, and additions, omissions, substitutions, and other modifications of configurations are possible without departing from the spirit of the present invention. Furthermore, the present invention is not limited to the embodiments.

10...base part, 11...first assembly model, 12...second assembly model, 13...third assembly model, 21...first sub-assembly model, 22...second sub-assembly model, 23...third sub-assembly model, 24...fourth sub-assembly model, 25...fifth sub-assembly model, 26...sixth sub-assembly model

Claims (9)

a method for retrieving a part of a 3D CAD model, the method comprising: comparing a retrieved model with a comparison model stored in a database and evaluating the degree of difference;
the search model and the comparison model are three-dimensional CAD data of an assembly model composed of a plurality of subassembly models;
A first calculation step in which a computer calculates a feature amount of the retrieval model;
a second calculation step in which the computer calculates a feature amount of the comparison model;
a comparison step in which the computer compares the feature amounts of the search model and the comparison model and calculates a degree of difference between the search model and the comparison model;
In the first calculation step and the second calculation step, for all subassembly models constituting the search model and the comparison model, the subassembly models and an assembly model excluding the subassembly models are each subjected to a three-dimensional Radon transform, and the feature quantity is derived from the transformed value.
A method for retrieving parts of a 3D CAD model. In the first calculation step and the second calculation step, the computer performs a three-dimensional Radon transform from a plurality of directions.
The method for searching a part of a three-dimensional CAD model according to claim 1. The comparison step includes:
a first direction fixing step in which the computer fixes the direction and calculates a sum of the differences in the feature amounts as a specific combination dissimilarity for all combinations of subassembly models of the search model and the comparison model;
a second direction fixing step in which the computer derives a combination of subassembly models that minimizes the specific combination dissimilarity calculated in the first direction fixing step, and sets the specific combination dissimilarity at this time as a specific direction dissimilarity for the direction;
The method further comprises a determination step in which the computer performs the first direction fixing step and the second direction fixing step for all directions, derives the direction in which the specific direction dissimilarity is the smallest, and derives the dissimilarity between the search model and the comparison model from the specific direction dissimilarity at this time.
The three-dimensional CAD model part searching method according to claim 2. In the determining step, the computer derives the direction in which the specific direction dissimilarity is minimized, and derives the dissimilarity by dividing the specific direction dissimilarity at this time by the number of subassemblies of the comparison model.
The three-dimensional CAD model part searching method according to claim 3. In the first calculation step and the second calculation step,
The feature quantities of the subassembly model and the assembly model excluding the subassembly model are sums of values obtained by performing three-dimensional Radon transform on all included parts individually and dividing the results by an approximate value of the volume of the parts.
The three-dimensional CAD model part searching method according to claim 4. In the first direction fixing step, the computer calculates the specific combination dissimilarity degree based on a sum of a difference in feature amounts between subassembly models to be compared and a difference in feature amounts between assembly models excluding the subassembly model to be compared, for the search model and the comparison model.
The three-dimensional CAD model part searching method according to claim 5. In the first direction fixing step, the computer further divides the sum by the number of parts of the subassembly model to be compared of the comparison model to calculate the specific combination dissimilarity.
The three-dimensional CAD model part searching method according to claim 6. In the comparing step, the computer relatively translates the feature amounts to be compared to obtain a value that minimizes a sum of differences between the feature amounts.
The three-dimensional CAD model part search method according to any one of claims 1 to 7. The database stores a plurality of comparison models to be searched;
The computer sequentially compares the search model with a plurality of the comparison models to derive the comparison model with the smallest degree of difference.
The three-dimensional CAD model part search method according to any one of claims 1 to 8.

Images