JP2018073111A - Retrieval system and retrieval program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform matching between a person who requests to manufacture an article that has a desired shape or the like and a manufacturer who actually performs manufacturing of the article with efficiency and high reliability.SOLUTION: A feature extraction unit 7b calculates a feature vector that characterizes at least a shape of a retrieval target on the basis of three-dimensional model data of an article to be the retrieval target. A manufacturing result database 7a stores a manufacturer who manufactures an article and a feature vector that characterizes a shape of an actual article manufactured in the past by the manufacturer in association with each other. A retrieval processing unit 7c specifies, by evaluating similarity with the feature vector of the actual article stored in the manufacturing result database 7a on the basis of the feature vector of the retrieval target, the manufacturer who manufactured the actual article that has similarity with the retrieval target.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a search system and a search program for searching for a manufacturer suitable for manufacturing an article having a desired shape or the like.

  Patent Document 1 discloses a cost estimation system that estimates a manufacturing cost of a processed article based on three-dimensional model data of the processed article in which a plurality of parts are incorporated. Specifically, first, based on the three-dimensional model data of the processed article, as the processed article information, the materials and processing steps necessary to manufacture each of the plurality of parts constituting the processed article, and the interval between these parts The processing steps required to join the are identified. Next, with reference to a reference price database in which reference prices of elements necessary for manufacturing the processed article are registered, a reference price of elements required for manufacturing the processed article is acquired as an estimation parameter. Then, the manufacturing cost of the processed article is calculated based on the processed article information and the estimated parameter.

JP 2008-158740 A

  By the way, as a result of technological advances such as 3D printers in recent years, an outsourcing service in which a manufacturer who owns manufacturing equipment acts on behalf of a manufacturer in response to a request from an individual who wants to manufacture an article having a desired shape, etc. Is spreading. The concern for the person who requests the manufacture of an article is to find a manufacturer who manufactures an article satisfying the required specifications under the conditions he desires. On the other hand, for the manufacturer of goods, it is how to obtain customer opportunities that make use of the manufacturing and processing technologies that are their strengths. Therefore, information sites have emerged that make the strengths and characteristics of many manufacturers who make things public on the network and help the client to make a selection. However, most of the conventional information sites merely provide information on the manufacturer to the viewer, and it is difficult to say that matching between the requesting side and the manufacturing side is performed appropriately.

  The present invention has been made in view of such circumstances, and an object thereof is to quickly and highly match a person who requests manufacture of an article having a desired shape and the like with a manufacturer who actually manufactures the article. It is done with reliability.

  In order to solve such a problem, the first invention provides a search system that includes a feature extraction unit, a manufacturing performance database, and a search processing unit, and searches for a manufacturer suitable for manufacturing an article. The feature extraction unit calculates a feature vector that characterizes at least the shape of the search target based on the three-dimensional model data of the article to be searched. The manufacturing performance database stores a manufacturer that manufactures an article and a feature vector that characterizes the shape of a past product manufactured by the manufacturer in association with each other. The search processing unit evaluates the similarity to the feature vector of the actual product stored in the manufacturing performance database based on the feature vector of the search target, thereby identifying the manufacturer who manufactured the actual product having similarity to the search target. Identify.

  Here, in the first invention, the manufacturing performance database is associated with the manufacturer and the feature vector, and is at least one of processing accuracy, size, material, manufacturing method, manufacturing apparatus, delivery date, and user evaluation of the article. It is preferable to store incidental information including In this case, when at least one of the processing accuracy, size, material, manufacturing method, manufacturing apparatus, delivery date, and user evaluation of the article is specified as the search condition, the search processing unit manufactures based on the search condition. The incidental information stored in the record database is searched to identify the manufacturer who manufactured the record product that satisfies the search conditions.

  In the first invention, the search processing unit preferably outputs a list of at least one manufacturer specified by searching the manufacturing performance database. In this case, when the search condition is changed by the searcher, the search processing unit may change the listed manufacturer according to the change of the search condition. In addition, when a plurality of manufacturers are specified by searching the manufacturing performance database, the search processing unit may rate the listed manufacturers according to the incidental information.

  In the first invention, when a new manufacturing record occurs for a specific manufacturer, an update processing unit for adding a feature vector of the actual product manufactured by the specific manufacturer to the manufacturing record database may be further provided. .

  2nd invention provides the search program which searches the manufacturer suitable for manufacture of articles | goods. The search program includes a first step of calculating a feature vector that characterizes at least the shape of the search target based on the three-dimensional model data of the search target article, a manufacturer that manufactures the article, and a manufacturer By searching a manufacturing performance database that stores a feature vector that characterizes the shape of a past product manufactured in the past, and evaluating the similarity between the feature vector of the search target and the feature vector of the actual product Then, the computer is caused to execute a process including a second step of specifying a manufacturer who has manufactured the actual product having similarity to the search target.

  Here, in the second invention, the manufacturing result database is associated with the manufacturer and the feature vector, and is at least one of processing accuracy, size, material, manufacturing method, manufacturing apparatus, delivery date, and user evaluation of the article. It is preferable to store incidental information including In this case, in the case where at least one of the processing accuracy, size, material, manufacturing method, manufacturing apparatus, delivery date, and user evaluation of the article is specified as the search condition, the second step is based on the search condition. It is preferable to include the step of searching the incidental information stored in the manufacturing record database and identifying the manufacturer who manufactured the track record satisfying the search conditions.

  In the second invention, a third step of listing and outputting at least one manufacturer specified in the second step may be further provided. In this case, when the search condition is changed by the searcher, the third step may include a step of changing the listed manufacturer according to the change of the search condition. The third step may include a step of rating the listed manufacturers according to the incidental information when there are a plurality of manufacturers specified in the second step.

  In the second invention, when a new manufacturing result is generated for a specific manufacturer, a fourth step of adding a feature vector of the actual product manufactured by the specific manufacturer to the manufacturing result database may be further provided. Good.

  According to the present invention, a manufacturer of a track record product having shape similarity with an article to be searched by searching a manufacturing track database storing feature vectors of track records manufactured in the past by the manufacturer. Is identified. Since the similarity of the result product with respect to the search target is evaluated based on the feature vector converted into an abstract data structure called a vector, the matching between both can be performed quickly and with high reliability. As a result, the client can efficiently find a manufacturer that meets the conditions he / she desires, and the manufacturer can efficiently obtain customer opportunities that can make use of the manufacturing and processing technologies that are his strengths. Can do.

Overall view of manufacturing mediation network system Block diagram of manufacturing mediation server Configuration diagram of manufacturing results database Search process flowchart Illustration of SHOT Illustration of PFH Illustration of PPF

  FIG. 1 is an overall view of a manufacturing mediation network system according to this embodiment. The manufacturing intermediation network system 1 is a server client type mainly composed of a client 2 on the client side operated by the client, a client 3 on the manufacturer side operated by the manufacturer, and a manufacturing intermediary server 4 operated by the manufacturing intermediary. Network configuration. Here, the “requester” is a person who requests (orders) the manufacture of an article having a desired shape or the like from the outside. The “manufacturer” is a person who owns manufacturing equipment and undertakes (receives an order for) the manufacture (including processing) of an article. A manufacturer may be a single entity, but a plurality of entities may share the manufacturing process of an article, so a group consisting of a plurality of entities may be handled as one manufacturer. . Furthermore, the “manufacturing intermediary” is a person who mediates the transaction between the client and the manufacturer, and the ordering by the client and the order received by the manufacturer are performed via the manufacturing intermediary. The manufacturing intermediary server 4 provides a function of mediating and managing transactions between the client and the manufacturer, and receives a search request from the client, and is a manufacturer suitable for manufacturing an article desired by the client. It also provides a function to search.

  FIG. 2 is a block diagram of the production mediation server 4. This production mediation server 4 is mainly composed of an input unit 5, a transaction module 6, a search module 7, and an output unit 8. The input unit 5 receives information received from the requester side client 2 or the manufacturer side client 3 via a network such as the Internet, and transfers the input information to the transaction module 6 or the search module 7. The transaction module 6 includes a client database that manages various information about the client, a manufacturer database that manages various information about the manufacturer, a transaction database that manages the contents and progress of the transactions here, and the client. Manage the entire transaction process from ordering by the manufacturer to delivery by the manufacturer. The search module 7 searches for a specific product that the client wants to manufacture for a manufacturer that is suitable for manufacturing the product, and presents the search result to the client (searcher). The output unit 8 transmits the information output from the transaction module 6 or the search module 7 to the requester side client 2 or the manufacturer side client 3 via the network.

  In the present embodiment, the search module 7 includes a manufacturing performance database 7a, a feature extraction unit 7b, a search processing unit 7c, and an update processing unit 7d. FIG. 3 is a configuration diagram of the manufacturing performance database 7a. This manufacturing result database 7a is used for managing products (result products) manufactured in the past by the manufacturer. One record provided for each transaction result includes “manufacturer” and “characteristics”. The “vector” and the “accompanying information” are stored in association with each other.

  Here, in “Manufacturer”, the manufacturer name of the record product is described. In the “feature vector”, the feature vector of the actual product is described. The feature vector at least characterizes the shape of the actual product, but may be characterized including the size and color of the actual product together with the shape. Here, the “three-dimensional model data” may be three-dimensional CAD data (file extensions such as IGES, STEP, PARASOLID, SAT, JT, VDA, etc.) or 3DCG (three-dimensional computer graphics). ) Data or point cloud data expressing a three-dimensional surface shape with a large number of points. The registration of the actual product in the manufacturing result database 7a is obtained by converting the 3D model data, not the actual 3D model data itself, so that the actual product can be searched using the shape or the like as a search key. Stored in the form of feature vectors. In this way, by converting the original 3D model data (unstructured data) into an abstract data structure called vectors, in other words, by abstracting so that machine learning and pattern recognition can be performed, It becomes possible to perform the search of the above with accuracy on a shape basis.

  In the “Attachment Information”, as supplementary information related to this actual product, “Processing Accuracy” of the actual product, “Size” indicating the size and thickness of the actual product, and “Material” used for manufacturing the actual product. "Manufacturing method" (for example, additive manufacturing, cutting, casting, etc.) used for manufacturing of the actual product, "Manufacturing apparatus" (for example, 3D printer, lathe, machining center, etc.) used for manufacturing the actual product, The “delivery date” of the actual product and the “evaluation” (user evaluation) of the requester for the actual product are described. Here, the “machining accuracy” refers to the accuracy that the entire actual product should satisfy. Therefore, for example, when high accuracy (for example, 0.05mm) is required only for a specific part in the actual product, and the required accuracy for other parts is lower (for example, 0.1mm), the minimum accuracy to be satisfied is 0.05mm. It becomes.

  The feature extraction unit 7b calculates a feature vector that characterizes the shape of the search target based on the three-dimensional model data of the search target article received from the client on the client side 2. The search processing unit 7c manufactures a performance product having similarity to the search target by evaluating the similarity with the “feature vector” of the performance product stored in the manufacturing performance database 7a based on the feature vector of the search target. The “manufacturer” is identified. At the same time, when at least one of the processing accuracy, size, material, manufacturing method, manufacturing apparatus, delivery date, and user evaluation of the article is designated by the client 2 as the search condition, the search processing unit 7c Based on the search condition, the “accompanying information” stored in the manufacturing performance database 7a is searched to identify the “manufacturer” who manufactured the actual product satisfying the search condition. Furthermore, in response to an update instruction from the transaction module 6, the update processing unit 7 d stores a record describing “feature vector” and “accompanying information” of the actual product manufactured by the “manufacturer” in the manufacturing result database 7 a. Add a new one. This update instruction is issued every time a new manufacturing result is produced for a certain manufacturer, for example, when delivery from the manufacturer to the client is completed.

  Next, the search process performed by the search module 7 will be described in detail with reference to FIG. This search process is executed by installing a computer program that causes the production mediation server 4 to execute the process of FIG.

  First, in step 1, information relating to a search target received from the client 2 on the requester side is input to the input unit 5. As input information for the search, 3D model data that identifies the shape of the article is indispensable, but as other optional search conditions, the processing accuracy of the article, size, material, manufacturing method, manufacturing equipment, delivery date, and It is also possible to specify at least one of the user evaluations.

  In step 2, the feature extraction unit 7b calculates a feature vector having a finite number of dimensions that characterizes the shape of the search target based on the three-dimensional model data of the article to be searched. As a method for characterizing an object in a three-dimensional shape (three-dimensional feature amount), for example, a key point-based method is known, and a method for describing information around a key point, two points, or three points It is classified into a method for describing the relationship between a plurality of points such as between.

  Examples of methods for describing information around key points include SHOT (Signature of Histograms of OrienTations) using coordinate data and normal vectors, and PFH (Point Feature Histograms). In the case of SHOT, as shown in FIG. 5, first, in the support sphere around the key point, the XY plane is divided into two parts, the inside of the sphere is divided into two parts at the central part and the peripheral part, and further, the Z axis is divided into eight parts ( 2 × 2 × 8 = 32 divisions). Next, the inner product of the normal r of the reference point and the normal vector nj of each of the divided spaces (32) is obtained, and an 11-bin histogram is created. As a result, a 352-dimensional feature amount of 32 × 11 is obtained.

  In the case of PFH, as shown in FIG. 6, first, a large number of two-point sets are selected from the neighboring points in the spherical region around the point of interest. Next, α, θ, φ, and pt-ps are calculated from the two points and converted into a histogram (125 dimensions). If a certain feature amount is larger than the average feature amount in the model, it is stored as a candidate. The same calculation is performed while changing the radius of the sphere, and candidate points supported from many radii are selected as the final unique feature points. Since SHOT and PFH have a high number of dimensions, there is an advantage that the feature expression power is high. On the other hand, PPF (Point Pair Feature) is an example of the latter. In the case of PPF, as shown in FIG. 7, first, a four-dimensional feature amount (F1 to F4) is specified from a pair of two points (PPH). This feature amount is calculated for each pair formed from all points on the object. These feature amounts are registered in a hash table. Next, referring to the hash table, PPHs having similar features are searched, and their geometric transformation parameters are calculated.

  Further, instead of the above-described method, at least one two-dimensional image (for example, a six-sided view) in which a product arranged in the virtual space is viewed from a predetermined viewpoint based on the three-dimensional model data of the object to be searched. , And a characterization method focusing on the density of the two-dimensional image may be used. In addition, it is also possible to use image labeling technology (deep auto encoder or the like) using deep learning.

  In step 3, the search processing unit 7c refers to the manufacturing result database 7a using the search target feature vector as a search key, and evaluates / calculates the similarity between the search target item and the actual product. A high degree of similarity between the two means that the feature point indicated by the feature vector to be searched is close to the feature point indicated by the feature vector of the actual product in the feature space. Therefore, if the distance between two feature points is calculated using a known method such as the Euclidean distance or the absolute sum of the coordinate component differences, the similarity between the two feature points can be evaluated.

  In step 4, the search processing unit 7c identifies a manufacturer who has manufactured a tracked product having similarity to the article to be searched from the description content of the manufacturing track record database 7a. As the actual product having this similarity, for example, the actual product whose similarity calculated in Step 3 is larger than a predetermined threshold value can be used. If the searcher (requester) can variably set this threshold value, the number of hits by search can be arbitrarily adjusted.

  In step 5, the search processing unit 7c creates a manufacturer list in which at least one manufacturer specified by searching the manufacturing performance database is listed. The manufacturer list created by the search processing unit 7 c is transmitted to the client client 2 via the output unit 8.

  The search processing unit 7c has the above-described shape when at least one of processing accuracy, size, material, manufacturing method, manufacturing apparatus, delivery date, and user evaluation of an article is specified as a search condition of the manufacturer. In addition to the similarity, the search condition may be satisfied as a hit requirement. In this case, the incidental information stored in the manufacturing record database 7a is also searched using the specified search condition as a search key. For example, when machining accuracy (for example, 0.1 mm or less) is specified as one of the search conditions, even a manufacturer who has produced a product with a geometric similarity satisfies the specified machining accuracy. Non-manufacturers will be filtered out. In addition, when the search condition is changed by the searcher, the search processing unit 7c changes the listed manufacturer according to the change of the search condition. In this way, it is possible to obtain the optimum search result for the searcher by repeatedly adding / deleting the search condition and filtering the manufacturer.

  Further, when a plurality of manufacturers are specified by searching the manufacturing performance database 7a, the search processing unit 7c may rank the manufacturers according to the incidental information, that is, perform rating. This rating may be performed based on, for example, user evaluation fed back from the client, or may be performed for each processing method.

  As described above, according to the present embodiment, by searching the manufacturing result database 7a in which the feature vector of the actual product manufactured in the past by the manufacturer is searched, the shape similarity with the article to be searched is obtained. The manufacturer of the track record product that he has is specified. The similarity of the actual product to the search target is not based on the 3D model itself, but based on the feature vector converted into an abstract data structure called a vector, so that both can be matched quickly and reliably. it can. As a result, the client can efficiently find a manufacturer that meets the conditions he / she desires, and the manufacturer can efficiently obtain customer opportunities that can make use of the manufacturing and processing technologies that are his strengths. Can do.

  In addition, according to the present embodiment, not only the shape similarity of the actual product with respect to the search target, but also the search of incidental conditions using the search condition as a search key is performed, thereby satisfying the request specifications of the client. Can be searched with higher accuracy.

  In the above-described embodiment, the processing accuracy, size, material, manufacturing method, manufacturing apparatus, delivery date, and user evaluation of the article are used as incidental information stored in the manufacturing performance database 7a. However, the present invention is not limited to this, and only some of these items may be used, or other items may be added as supplementary information. In this case, the search condition can be appropriately specified within a range of items provided as incidental items.

DESCRIPTION OF SYMBOLS 1 Manufacturing mediation network system 2 Client of client side 3 Client of manufacturing side 4 Manufacturing mediation server 5 Input part 6 Transaction module 7 Search module 7a Manufacturing performance database 7b Feature extraction part 7c Search processing part 7d Update processing part 8 Output part

Claims (12)

In a search system for searching for a manufacturer suitable for manufacturing an article,
A feature extraction unit that calculates a feature vector that characterizes at least the shape of the search target based on the three-dimensional model data of the search target article;
A manufacturing record database that stores a manufacturer that manufactures an article and a feature vector that characterizes the shape of a track record manufactured in the past by the manufacturer;
Based on the feature vector of the search target, by evaluating the similarity with the feature vector of the actual product stored in the manufacturing result database, the manufacturer who manufactured the actual product having similarity to the search target A search system comprising a search processing unit for specifying. The manufacturing performance database stores incidental information including at least one of processing accuracy, size, material, manufacturing method, manufacturing apparatus, delivery date, and user evaluation of an article in association with the manufacturer and the feature vector. And
When at least one of processing accuracy, size, material, manufacturing method, manufacturing apparatus, delivery date, and user evaluation of an article is specified as a search condition, the search processing unit is configured to use the manufacturing performance database based on the search condition. 2. The search system according to claim 1, wherein the accessory information stored in the search is searched to identify a manufacturer who manufactured the actual product satisfying the search condition.   The search system according to claim 1 or 2, wherein the search processing unit lists and outputs at least one manufacturer specified by searching the manufacturing performance database.   4. The search according to claim 3, wherein when the search condition is changed by a searcher, the search processing unit changes the listed manufacturer according to the change of the search condition. system.   4. The search processor according to claim 3, wherein when a plurality of manufacturers are specified by searching the manufacturing performance database, the listed manufacturers are rated according to the incidental information. The described search system.   The apparatus further comprises an update processing unit for adding a feature vector of a performance product manufactured by the specific manufacturer to the manufacturing performance database when a new manufacturing performance occurs for the specific manufacturer. The search system described in any one of 1 to 5. In a search program for searching for a manufacturer suitable for manufacturing an article,
A first step of calculating a feature vector characterizing at least the shape of the search target based on the three-dimensional model data of the article to be searched;
A manufacturing performance database that stores a manufacturer that manufactures an article and a feature vector that characterizes the shape of a product manufactured in the past by the manufacturer in association with each other is searched, and the search target feature vector, By causing the computer to execute a process having a second step of identifying a manufacturer who has produced the actual product having similarity to the search target by evaluating the similarity with the feature vector of the actual product. Search program. The manufacturing performance database stores incidental information including at least one of processing accuracy, size, material, manufacturing method, manufacturing apparatus, delivery date, and user evaluation of an article in association with the manufacturer and the feature vector. And
In the second step, when at least one of processing accuracy, size, material, manufacturing method, manufacturing apparatus, delivery date, and user evaluation of an article is specified as a search condition, the manufacturing performance is based on the search condition. The search program according to claim 7, further comprising a step of searching the incidental information stored in a database and identifying a manufacturer who manufactured the actual product satisfying the search condition.   9. The search program according to claim 7, further comprising a third step of listing and outputting at least one manufacturer specified in the second step. The third step includes
The search program according to claim 9, further comprising a step of changing the listed manufacturer according to the change of the search condition when the search condition is changed by the searcher. The third step includes
10. The search according to claim 9, further comprising a step of rating the listed manufacturers according to the incidental information when there are a plurality of manufacturers specified in the second step. program. The method further comprises a fourth step of adding a feature vector of a performance product manufactured by the specific manufacturer to the manufacturing performance database when a new manufacturing performance occurs for the specific manufacturer. Item 12. The search program according to any one of Items 7 to 11.