JP2018032210A - Design prediction apparatus, design prediction program and design prediction method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a design prediction apparatus, a design prediction program and a design prediction method capable of accurately predicting a prediction object.SOLUTION: A classification unit 41 classifies plural pieces of design data 30A into clusters of design data 30A each of which has the same dominant feature relevant to a design item of prediction object based on the design information extracted from the plural pieces of design data 30A. A generation unit 42 generates a prediction model for predicting a design item of a prediction object for each of the classified clusters from the design information of the design data 30A included in the clusters.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a design prediction apparatus, a design prediction program, and a design prediction method.

  In designing a product, a design item may be predicted before starting the design. For example, in the mounting design of a printed circuit board, when the components and netlist to be mounted on the printed circuit board are determined, the number of signal layers, power supply layers, and ground layers of the printed circuit board is predicted before the mounting design. In this prediction, it is expected to predict an appropriate number of layers. If the predicted number of layers is less than the appropriate number of layers, the mounting design will be redone, resulting in lost man-hours and time. On the other hand, when the predicted number of layers is larger than the appropriate number of layers, the manufacturing cost of the printed circuit board increases.

  Therefore, there is a technique for performing prediction by regression analysis. For example, a linear prediction model that predicts a design item to be predicted is generated from past design assets using a linear kernel of SVR (Support Vector Regression). The SVR is also called SVM (Support Vector Machine). Then, a predicted value of the design item to be predicted is calculated using the generated prediction model.

International Publication No. 2010/029627 JP 2000-020504 A JP 2006-330988 A

  Since the linear prediction model is linear, it is easy to understand the basis for predicting the prediction value. However, since the linear prediction model is generated so that the total error is minimized with respect to the entire past design asset data, there may be a portion where the prediction accuracy of the prediction value is low.

  Therefore, it is conceivable to classify past design asset data by clustering and generate a linear prediction model for each cluster. For example, using a clustering method such as the k-means method or the mean-shift method, the past design asset data is classified into clusters for each similar data, and a linear prediction model is generated for each cluster. It is done. However, in this case, the prediction model becomes overfitting to the cluster data, and the prediction accuracy may be low.

  In addition, although the product design has been described as an example, such a problem is generally applicable to a case where a design item is predicted in another design such as a system design.

  In one aspect, an object is to provide a design prediction apparatus, a design prediction program, and a design prediction method that can accurately predict a prediction target.

  In the first plan, the design prediction apparatus includes a classification unit and a generation unit. The classification unit classifies the plurality of design data into clusters for each design data having the same dominant characteristics with respect to the design item to be predicted based on the design information extracted from the plurality of design data. For each cluster classified by the classification unit, the generation unit generates a prediction model for predicting a design item to be predicted from design information of design data included in the cluster.

  According to one embodiment of the present invention, there is an effect that a prediction target can be accurately predicted.

FIG. 1 is a diagram illustrating a schematic configuration of a design prediction apparatus. FIG. 2 is a diagram illustrating an example of a data configuration of the design information table. FIG. 3 is a diagram illustrating an example of a flow for extracting design information. FIG. 4 is a diagram showing an example of normalization of design information of each design item. FIG. 5 is a diagram functionally illustrating a classification flow by the classification unit according to the first embodiment. FIG. 6 is a diagram schematically showing a nonlinear prediction model. FIG. 7 is a diagram schematically showing a linear prediction model. FIG. 8 is a diagram for explaining the flow of classification. FIG. 9 is a diagram illustrating an example of a classification calculation flow according to the first embodiment. FIG. 10 is a diagram illustrating an example of a flow of generating a prediction model. FIG. 11 is a diagram illustrating an example of a flow of generating a cluster allocation model. FIG. 12 is a diagram illustrating an example of a flow for specifying a cluster. FIG. 13 is a diagram illustrating an example of a flow for calculating the number of signal layers. FIG. 14 is a diagram illustrating an example of a flow for calculating the contribution degree. FIG. 15 is a diagram illustrating an example of the contribution output. FIG. 16 is a flowchart illustrating an example of the procedure of the generation process according to the first embodiment. FIG. 17 is a flowchart illustrating an example of the procedure of the prediction process. FIG. 18 is a diagram functionally illustrating a classification flow by the classification unit according to the second embodiment. FIG. 19 is a diagram illustrating an example of a classification calculation flow according to the second embodiment. FIG. 20 is a flowchart illustrating an example of a generation processing procedure according to the second embodiment. FIG. 21 is a diagram illustrating an example of predicting other design items. FIG. 22 is a diagram illustrating an example of predicting other design items. FIG. 23 is a diagram illustrating a computer that executes a design prediction program.

  Hereinafter, a design prediction apparatus, a design prediction program, and a design prediction method according to the present invention will be described in detail based on the drawings. The disclosed technology is not limited by the present embodiment. Moreover, you may combine suitably the Example shown below in the range which does not cause contradiction.

[Configuration of design prediction device]
A design prediction apparatus 10 according to the first embodiment will be described. FIG. 1 is a diagram illustrating a schematic configuration of a design prediction apparatus. In designing a product, a design item may be predicted before starting the design. The design prediction device 10 is a device that predicts a certain design item before starting the design. In the present embodiment, a case where the number of signal layers of a printed circuit board is predicted before mounting design of the printed circuit board will be described as an example.

  The design prediction apparatus 10 is, for example, a computer such as a personal computer or a server computer. The design prediction apparatus 10 may be implemented as a single computer, or may be implemented as a cloud including a plurality of computers. In this embodiment, a case where the design prediction apparatus 10 is a single computer will be described as an example. The design prediction apparatus 10 may be a server computer that is communicably connected to a terminal device via a network, accepts various types of operation information from the terminal device, and transmits processing results corresponding to the operation information to the terminal device. Good. The design prediction apparatus 10 is, for example, a computer used by a developer who performs mounting design of a printed circuit board. The design prediction apparatus 10 may be a design apparatus that operates design software for designing hardware such as a CAD (Computer Aided Design) apparatus. As illustrated in FIG. 1, the design prediction apparatus 10 includes an input unit 20, a display unit 21, a storage unit 22, and a control unit 23. In addition, the design prediction apparatus 10 may have other apparatuses other than said apparatus.

  The input unit 20 is an input device that inputs various types of information. Examples of the input unit 20 include an input device that accepts an input of an operation such as a mouse or a keyboard. The input unit 20 receives input of various types of information. For example, the input unit 20 receives input of various operations on a virtual hardware model. The input unit 20 inputs operation information indicating the received operation content to the control unit 23.

  The display unit 21 is a display device that displays various types of information. Examples of the display unit 21 include display devices such as an LCD (Liquid Crystal Display) and a CRT (Cathode Ray Tube). The display unit 21 displays various information. For example, the display unit 21 displays various screens such as an operation screen.

  The storage unit 22 is a storage device such as a hard disk, an SSD (Solid State Drive), or an optical disk. The storage unit 22 may be a semiconductor memory that can rewrite data such as a random access memory (RAM), a flash memory, and a non-volatile static random access memory (NVSRAM).

  The storage unit 22 stores an OS (Operating System) executed by the control unit 23 and various programs. For example, the storage unit 22 stores a program that executes a generation process and a prediction process described later. Furthermore, the storage unit 22 stores various data used in programs executed by the control unit 23. For example, the storage unit 22 stores design asset data 30, a design information table 31, prediction model data 32, and cluster allocation model data 33.

  The design asset data 30 is data that stores various types of information related to design assets that can be used for design. For example, the design asset data 30 includes a plurality of designed design data 30A.

  The design data 30A is data of a printed circuit board that has been mounted and designed in the past. The design data 30A stores various design information related to a printed circuit board, a mounted component, and wiring that have been mounted and designed. For example, the design data 30A stores, as design information, the shape of the printed circuit board, the number of signal layers, position information of components, position information of wiring of each signal layer, and the like.

  The design information table 31 is data in which various design information extracted from the design asset data 30 is stored. For example, the design information table 31 stores design information of various design items extracted from each design data 30A.

  FIG. 2 is a diagram illustrating an example of a data configuration of the design information table. As shown in FIG. 2, the design information table 31 has items of data ID, design information A, design information B. The design information table 31 may store various information other than the above.

  The item of data ID is an area for storing identification information for identifying the design data 30A. A unique data ID is assigned to each of the design data 30A as identification information. In the data ID item, the data ID of the design data 30A from which the design information is extracted is stored. The items of design information A and design information B are areas for storing design information of design items respectively extracted from the design data 30A of the data ID.

  The prediction model data 32 is data that stores information related to a prediction model that predicts appropriate design information of a design item to be predicted. In this embodiment, the design item to be predicted is the number of signal layers of the printed circuit board. The prediction model data 32 stores information on a prediction model for predicting the number of signal layers of the printed circuit board. Details of the prediction model will be described later.

  The cluster allocation model data 33 is data that stores information on a cluster allocation model that predicts which cluster the design data to be predicted belongs to. Details of the cluster allocation model will be described later.

  Returning to FIG. 1, the control unit 23 is a device that controls the design prediction apparatus 10. As the control unit 23, an electronic circuit such as a CPU (Central Processing Unit) and an MPU (Micro Processing Unit), or an integrated circuit such as an ASIC (Application Specific Integrated Circuit) and an FPGA (Field Programmable Gate Array) can be employed. The control unit 23 has an internal memory for storing programs defining various processing procedures and control data, and executes various processes using these. The control unit 23 functions as various processing units by operating various programs. For example, the control unit 23 includes an extraction unit 40, a classification unit 41, a generation unit 42, a reception unit 43, a specification unit 44, a calculation unit 45, and an output unit 46.

  The extraction unit 40 extracts various information used for generation in the prediction model from the design asset data 30. For example, the extraction unit 40 extracts design information of a plurality of design items from the design data 30A. For example, the extraction unit 40 extracts design information of a design item to be predicted and design information of a design item that affects the design item to be predicted from the design data 30A. Design items to be extracted by the extraction unit 40 may be determined in advance or may be provided from the outside. The extraction unit 40 stores design information of each design item extracted from the design data 30 </ b> A in the design information table 31.

  FIG. 3 is a diagram illustrating an example of a flow for extracting design information. In the example of FIG. 3, design items to be extracted are provided as a feature list. In the example of FIG. 3, as the design items that affect the number of signal layers of the printed circuit board, the board area, the total number of nets, the number of parts,..., The BGA (Ball Grid Array) pin pitch are described in the feature list. ing. The extraction unit 40 extracts design information of design items described in the feature list from the design data 30A. Then, the extraction unit 40 stores design information of each design item extracted from the design data 30 </ b> A in the design information table 31. In the example of FIG. 3, the board area, the total number of nets, the number of parts,..., The BGA pin pitch extracted from the design data 30 </ b> A with the data IDs “1” to “500” are stored in the design information table 31. ing.

  The classification unit 41 performs various classifications. For example, based on the design information of each design data 30A stored in the design information table 31, the classification unit 41 assigns each design data 30A to each design data 30A having the same dominant feature with respect to the design item to be predicted. Class.

  For example, first, the classification unit 41 normalizes the design information of each design item stored in the design information table 31 to a feature value within a predetermined range for each design item. FIG. 4 is a diagram showing an example of normalization of design information of each design item. For example, the classification unit 41 normalizes design information of each design item stored in the design information table 31 to a feature value in a range of 0 to 1 for each design item, and features of the design item for each design data 30A. A feature vector whose value is a component is obtained.

  Next, the classification unit 41 uses the feature vector of each design data 30A to classify into clusters for each design data 30A having the same dominant feature with respect to the design item to be predicted.

  FIG. 5 is a diagram functionally illustrating a classification flow by the classification unit according to the first embodiment. The classification unit 41 includes a nonlinear model generation unit 50, an inclination calculation unit 51, and a clustering unit 52.

  The nonlinear model generation unit 50 generates a nonlinear prediction model that predicts a design item to be predicted from the design information of each design data 30 </ b> A stored in the design information table 31. For example, the nonlinear model generation unit 50 generates a nonlinear prediction model that predicts a feature value indicating the number of signal layers of a printed circuit board from a feature vector obtained by normalizing design information of each design data 30A. For example, the non-linear model generation unit 50 generates a non-linear prediction model using an SVR prediction model of a radial basis function (RBF) kernel as the non-linear model. For example, the nonlinear model generation unit 50 receives a feature value of a design item that affects a design item to be predicted as an input, and generates a nonlinear prediction model that outputs the feature value of the design item to be predicted. In this embodiment, the nonlinear model generation unit 50 generates a nonlinear prediction model that predicts a feature value indicating the number of signal layers of a printed circuit board as a design value.

Here, the nonlinear prediction model and the linear prediction model will be described. First, the nonlinear prediction model will be described. FIG. 6 is a diagram schematically showing a nonlinear prediction model. The example of FIG. 6 shows a nonlinear prediction model curve L1 that predicts a design value indicating a design item to be predicted from a feature vector obtained by normalizing design information of each design data 30A. In the example of FIG. 6, the feature values of the design items that are input to the prediction model are simplified to two dimensions (f ₁ , f ₂ ) to simplify the vector space. Each feature value becomes an axis and becomes more multidimensional. In the example of FIG. 6, a point P corresponding to the feature vector of each design data 30 </ b> A is shown in the space centered on each component of the feature vector. The nonlinear prediction model is a nonlinear curve L1 that passes through each point P with less error. Such a non-linear prediction model has a small error in the prediction value, but it is not known why the prediction value is the value. For this reason, it is difficult for the designer to trust the prediction result obtained by the prediction model.

Next, a linear prediction model will be described. FIG. 7 is a diagram schematically showing a linear prediction model. In the example of FIG. 7, a straight line L2 of a linear prediction model for predicting a design value indicating a design item to be predicted from a feature vector obtained by normalizing design information of each design data 30A is shown. In the example of FIG. 7 as well, the feature values of the design items that are input to the prediction model are simplified to two dimensions (f ₁ , f ₂ ) to simplify the vector space. Each feature value becomes an axis and becomes more multidimensional. In the example of FIG. 7, a point P corresponding to the feature vector of each design data 30 </ b> A is shown in the space centered on each component of the feature vector. The linear prediction model is a straight line L2 that passes through each point P with less error. Since the linear prediction model is linear, it is easy to understand why the predicted value is the value. However, in the linear prediction model, there may be a portion where a large error occurs in the predicted value.

  For example, the nonlinear model generation unit 50 generates a nonlinear prediction model that fits each design data 30A from the feature vector obtained by normalizing the design information of each design data 30A stored in the design information table 31. FIG. 8 is a diagram for explaining the flow of classification. FIG. 8 shows a non-linear prediction model curve L1 for predicting the design value of the design item to be predicted and a point P corresponding to the feature vector of each design data 30A, as in FIG.

  The inclination calculation unit 51 calculates the inclination of the design information of each design data 30 </ b> A in the nonlinear prediction model generated by the nonlinear model generation unit 50. For example, the inclination calculation unit 51 calculates the inclination (gradient) with respect to the axis of the curve L1 at the position corresponding to each point P. For example, the inclination calculation unit 51 obtains the closest point to the curve L1 for each point P, and calculates the inclination of the curve L1 with respect to the axis at the closest point.

Based on the inclination calculated by the inclination calculation unit 51, the clustering unit 52 classifies each design data 30A into clusters for each design data 30A having the same dominant feature. For example, the clustering unit 52 specifies a design item having the largest inclination from the feature vector of each design data 30A. Then, the clustering unit 52 classifies the design data 30A into clusters so that the design data 30A having the same design item having the largest inclination becomes the same cluster. In the example of FIG. 8, the points P corresponding to the design data 30A is a largest cluster inclination of f _1, the slope of design items f ₂ are classified in the largest cluster.

An example of a specific calculation flow of classification by each processing unit of the classification unit 41 will be described. FIG. 9 is a diagram illustrating an example of a classification calculation flow according to the first embodiment. For example, the nonlinear model generation unit 50 models the feature vector obtained by normalizing the design information of each design data 30A with the SVR of the RBF kernel, and uses the prediction model to determine the number of signal layers of the printed circuit board (number of signal layers). Generate a function of the prediction model that predicts. The inclination calculation unit 51 calculates an inclination (gradient) with respect to the axis of the function of the prediction model at a position corresponding to the point P of each feature vector. The clustering unit 52 classifies each design data 30A into clusters for each design data 30A having the same dominant feature based on the gradient of each feature vector. For example, the clustering unit 52 specifies a design item having the largest inclination from each feature vector. For example, in the example of FIG. 9, for the feature vector x _1, components of the first design item inclination is "10.8" is identified as dominant features. Also, the feature vector x _2, components of the third design fields inclination is "14" is identified as dominant features. The clustering unit 52 classifies the design data 30A into clusters so that the design data 30A with the same design item having the largest inclination becomes the same cluster. For example, in the example of FIG. 9, the feature vector x ₁ , the feature vector x ₄ ,..., And the feature vector x ₅₀₀ are classified into the cluster a. Further, the feature vector x ₂ , the feature vector x ₃ ,... Are classified into the cluster b. In addition, although the clustering part 52 made the design item with the largest inclination the dominant feature, it is not limited to this. For example, the clustering unit 52 may classify the design data 30A having the same upper predetermined number of design items into clusters, with the upper predetermined number (for example, the upper third) design items as the dominant features in descending order of inclination. .

  Returning to FIG. 1, the generation unit 42 generates various prediction models. For example, the generation unit 42 generates, for each cluster classified by the classification unit 41, a prediction model that predicts a design item to be predicted from the design information of the design data 30A included in the cluster. For example, for each cluster, the generation unit 42 generates a linear prediction model that predicts a feature value indicating the number of signal layers of the printed circuit board as a design value from the feature vector of the design data 30A included in the cluster. The generation unit 42 stores the generated prediction model information in the prediction model data 32 for each cluster.

  FIG. 10 is a diagram illustrating an example of a flow of generating a prediction model. For each cluster, the generation unit 42 generates a linear prediction model that predicts a feature value indicating the number of signal layers of the printed circuit board as a design value from the feature vector of the design information of the design data 30A included in the cluster. In the example of FIG. 10, linear prediction models are generated for the clusters a and b, respectively, and prediction model information is stored in the prediction model data 32 for each cluster.

  Further, the generation unit 42 generates a cluster allocation model that assigns the design data 30A to the cluster based on the classification result by the classification unit 41. For example, the generation unit 42 generates a cluster assignment model that assigns the design data 30A to the clusters using a classification method such as SVC or Random Forest using the cluster classification result of each design data 30A as teacher data. The generation unit 42 stores the generated cluster allocation model information in the cluster allocation model data 33.

  FIG. 11 is a diagram illustrating an example of a flow of generating a cluster allocation model. In FIG. 11, any one of clusters a, b,... Is shown for 500 pieces of design data 30A. The design item data of each design data 30A and the cluster classification result of each design data 30A are shown. The generation unit 42 generates a cluster assignment model for estimating a cluster from data of each design item by a class classification method using the data of each design item of each design data 30A and the cluster classification result as teacher data. The generation unit 42 stores the generated cluster allocation model information in the cluster allocation model data 33.

  The reception unit 43 performs various types of reception. For example, the reception unit 43 receives design information of design data to be predicted. For example, when the mounting design of the printed circuit board is started, the reception unit 43 displays an operation screen for predicting the number of signal layers of the printed circuit board on the display unit 21. And the reception part 43 receives the input of the design information of each design item of the target printed circuit board which estimates the number of signal layers from the operation screen. The reception unit 43 displays an operation screen on a terminal device that is communicably connected via a network, and inputs design information of each design item of the target printed circuit board for predicting the number of signal layers from the terminal device. May be accepted.

  The identification unit 44 performs various types of identification. For example, the specifying unit 44 specifies to which cluster the design data to be predicted is classified based on the design information received by the receiving unit 43. For example, the specifying unit 44 uses the cluster allocation model stored in the cluster allocation model data 33 to determine which printed circuit board is from the design information of each design item of the target printed circuit board for which the number of signal layers is predicted. Identify whether it is classified as a cluster.

FIG. 12 is a diagram illustrating an example of a flow for specifying a cluster. FIG. 12 shows an example of data of each design item of the printed circuit board for which the number of signal layers is predicted as the data ID “501”. The specifying unit 44 normalizes the design information of each design item to a feature value within a predetermined range for each design item. FIG. 12 shows a normalized feature vector x ₅₀₁ of design information of each design item of a printed circuit board for which the number of signal layers is predicted. The specifying unit 44 uses the cluster assignment model stored in the cluster assignment model data 33 to specify to which cluster the feature vector x ₅₀₁ belongs. In the example of FIG. 12, the feature vector x ₅₀₁ is specified as belonging to the cluster b.

  The calculation unit 45 performs various calculations. For example, the calculation unit 45 calculates the predicted value of the design item to be predicted from the design information received by the receiving unit, using the prediction model corresponding to the cluster specified by the specifying unit 44. For example, the calculation unit 45 uses the prediction model corresponding to the identified cluster among the prediction models stored in the prediction model data 32, and outputs each design data of the target printed circuit board for predicting the number of signal layers. The number of signal layers is calculated from design item data.

FIG. 13 is a diagram illustrating an example of a flow for calculating the number of signal layers. In the example of FIG. 13, the feature vector x ₅₀₁ belongs to the cluster b. The calculation unit 45 calculates the number of signal layers from the feature vector x ₅₀₁ using the prediction model of the cluster b. In the example of FIG. 13, the number of signal layers is calculated as four.

  The calculation unit 45 also calculates a contribution indicating the degree to which the design information received by the reception unit 43 has an effect on the predicted value. For example, the calculation unit 45 calculates the contribution for each design item by multiplying the feature value by the weight value of the prediction model for each design item.

FIG. 14 is a diagram illustrating an example of a flow for calculating the contribution degree. In the example of FIG. 14, for each design item of the feature vector x ₅₀₁ , the feature value is multiplied by the weight of the design item to calculate the contribution for each design item. The design item having a large contribution is a design item having a large influence on the design item to be predicted.

  The output unit 46 performs various outputs. For example, the output unit 46 outputs the predicted value of the design item to be predicted calculated by the calculation unit 45. For example, the output unit 46 outputs the number of signal layers calculated by the calculation unit 45 to the display unit 21. The output unit 46 outputs the contribution for each design item calculated by the calculation unit 45. For example, the output unit 46 outputs the contribution for each design item calculated by the calculation unit 45 to the display unit 21. Note that the reception unit 43 may output the number of signal layers calculated on the screen of a terminal device that is communicably connected via a network and the degree of contribution for each design item.

  The designer can grasp the predicted value of the design item to be predicted from the output result. Further, the designer can grasp the design item having a large influence on the predicted value from the contribution degree, and can understand why the predicted value has become the value, and therefore can determine whether the basis for calculating the predicted value is appropriate. For this reason, the designer can trust the prediction result by the prediction model.

FIG. 15 is a diagram illustrating an example of the contribution output. In the example of FIG. 15, the contribution for each design item is output as a bar graph. In the example of FIG. 15, it can be understood that a large contribution of design items feature value f _3.

  Next, the flow of various processes executed by the design prediction apparatus 10 according to the present embodiment will be described. First, the flow of generation processing in which the design prediction apparatus 10 generates a prediction model will be described. FIG. 16 is a flowchart illustrating an example of the procedure of the generation process according to the first embodiment. This generation process is performed at a predetermined timing, for example, when a predetermined operation for instructing the generation start of a prediction model is performed from an administrator, or when a design item to be predicted is predicted, before a prediction process described later. It is executed at the timing.

  As illustrated in FIG. 16, the extraction unit 40 extracts, from each design data 30 </ b> A, design information on a design item to be predicted and design information on a design item that affects the design item to be predicted. The item design information is stored in the design information table 31 (S10).

  The nonlinear model generation unit 50 generates a nonlinear prediction model that predicts a design item to be predicted from the design information of each design data 30A stored in the design information table 31 (S11). The inclination calculation unit 51 calculates the inclination of the design information of each design data 30A in the generated nonlinear prediction model (S12). Based on the calculated slope, the clustering unit 52 classifies each design data 30A into clusters for each design data 30A having the same dominant feature (S13).

  For each cluster, the generation unit 42 generates a prediction model for predicting a design item to be predicted from the design information of the design data 30A included in the cluster, and stores the generated prediction model information in the prediction model data 32 ( S14). Further, the generation unit 42 generates a cluster allocation model that allocates the design data 30A to the cluster based on the classification result, stores the generated cluster allocation model information in the cluster allocation model data 33 (S15), and ends the processing. To do.

  Next, a flow of prediction processing in which the design prediction apparatus 10 predicts a design item to be predicted will be described. FIG. 17 is a flowchart illustrating an example of the procedure of the prediction process. This prediction process is performed at a predetermined timing, for example, when a predetermined operation for instructing the start of processing is performed after design information of each design item of the target printed circuit board for which the number of signal layers is predicted is input on the operation screen. Is executed.

  As illustrated in FIG. 17, the specifying unit 44 uses the cluster allocation model stored in the cluster allocation model data 33 to determine which of the printed circuit boards from the design information of each design item of the printed circuit board input to the operation screen. To be classified as a cluster (S50).

  The calculation unit 45 uses the prediction model corresponding to the identified cluster among the prediction models stored in the prediction model data 32, and each design item of the design data of the target printed circuit board for predicting the number of signal layers. The number of signal layers is calculated from the data (S51). Further, the calculation unit 45 calculates a contribution indicating the degree to which the design information received by the reception unit 43 has an effect on the predicted value (S52).

  The output unit 46 outputs the calculated number of signal layers and the contribution for each design item to the display unit 21 (S53), and ends the process.

  As described above, the design prediction apparatus 10 determines, based on the design information extracted from the plurality of design data 30A, the plurality of design data 30A for each design data 30A having the same dominant characteristics with respect to the design item to be predicted. Class. For each classified cluster, the design prediction device 10 generates a prediction model that predicts a design item to be predicted from design information of design data 30A included in the cluster. Thereby, the design prediction apparatus 10 can predict the prediction target with high accuracy.

  In addition, the design prediction apparatus 10 generates a nonlinear prediction model that predicts a design item to be predicted from design information extracted from a plurality of design data 30A. The design prediction apparatus 10 calculates the inclination of the design information of each design data 30A in the generated nonlinear prediction model. Based on the calculated inclination, the design prediction apparatus 10 classifies the plurality of design data 30A into clusters for each design data 30A having the same dominant feature. Thereby, the design prediction apparatus 10 can classify the plurality of design data 30A with high accuracy into clusters for each design data 30A having the same dominant feature.

  Further, the design prediction apparatus 10 receives design information of design data to be predicted. The design prediction apparatus 10 identifies which cluster the design data to be predicted is classified based on the received design information. The design prediction apparatus 10 calculates a predicted value of a design item to be predicted from the received design information using a prediction model corresponding to the identified cluster. The design prediction apparatus 10 outputs the calculated predicted value of the design item to be predicted. Thereby, the design prediction apparatus 10 can present the predicted value of the design item to be predicted with high accuracy for the design data to be predicted, and can support efficient design.

  Further, the design prediction apparatus 10 further calculates a contribution indicating the degree to which the received design information has affected the predicted value. The design prediction apparatus 10 further outputs the calculated contribution. The designer can evaluate whether the predicted value is appropriate or reliable from the output contribution. That is, the design prediction apparatus 10 can output the basis for predicting the predicted value.

  Next, Example 2 will be described. The configuration of the design prediction apparatus 10 according to the second embodiment is the same as that of the first embodiment shown in FIG.

  The classification unit 41 uses the feature vector of each design data 30A to classify into clusters for each design data 30A having the same dominant feature with respect to the design item to be predicted.

  FIG. 18 is a diagram functionally illustrating a classification flow by the classification unit according to the second embodiment. The classification unit 41 includes a clustering unit 60, a model generation unit 61, and a combining unit 62.

  The clustering unit 60 classifies the plurality of design data 30A into clusters for each design data 30A having similar design information. For example, the clustering unit 60 uses a classification method such as a k-means method or a mean-shift method to classify each design data 30A into clusters of design data 30A having similar feature vectors. For each classified cluster, the model generation unit 61 generates a linear prediction model that predicts a design item to be predicted from the design information of the design data 30A included in the cluster. In the present embodiment, the model generation unit 61 generates a linear prediction model that predicts a feature value indicating the number of signal layers of a printed circuit board as a design value. The combining unit 62 combines similar clusters based on the weight value for the design information in the generated linear prediction model. For example, the combining unit 62 combines the clusters of prediction models having the same design item with the maximum weight value as similar clusters. Note that the combining unit 62 has the upper predetermined number (for example, the third highest) design item as the dominant feature in descending order of the weight value, and combines the clusters of the prediction models with the same upper predetermined number of design items as similar clusters. May be.

An example of a specific calculation flow of classification by each processing unit of the classification unit 41 will be described. FIG. 19 is a diagram illustrating an example of a classification calculation flow according to the second embodiment. For example, the clustering unit 60 classifies the design data 30A and the design data 30A having similar feature vector feature quantities into clusters using classification methods such as the k-means method and the mean-shift method. For example, in the example of FIG. 19, the feature vector x ₁ , the feature vector x ₄ ,... Further, the feature vector x ₂ , the feature vector x ₃ ,... Are classified into the cluster 2.

  For each classified cluster, the model generation unit 61 generates a linear prediction model that predicts a design item to be predicted from the design information of the design data 30A included in the cluster. For each cluster, the model generation unit 61 generates a function of a linear prediction model that predicts the number of signal layers of the printed circuit board (the number of signal layers) from the feature vector of the design data 30A included in the cluster. For example, in the example of FIG. 19, a linear prediction model function is generated for each of the clusters 1 and 2.

The combining unit 62 combines similar clusters based on the weight value for the design information in the generated linear prediction model. For example, the combining unit 62 combines the clusters of prediction models having the same design item with the maximum weight value as similar clusters. By combining similar clusters in this manner, clusters having the same dominant feature are combined among the classified clusters. For example, in the example of FIG. 19, the feature vector x ₁ , the feature vector x ₄ ,..., The feature vector x ₅₀₀ are classified into the cluster a. Further, the feature vector x ₂ , the feature vector x ₃ ,... Are classified into the cluster b.

  Similarly to the generation unit 42 and the first embodiment, for each cluster classified by the classification unit 41, a prediction model for predicting a design item to be predicted is generated from the design information of the design data 30A included in the cluster.

  FIG. 20 is a flowchart illustrating an example of a generation processing procedure according to the second embodiment. The generation process according to the second embodiment is partly the same as the generation process according to the first embodiment illustrated in FIG. 16, and therefore, the same parts are denoted by the same reference numerals, the description thereof is omitted, and is mainly different. The part will be described.

  The clustering unit 60 uses a classification method such as the k-means method or the mean-shift method to classify the design data 30A into clusters of design data 30A having similar feature vector feature quantities (S20). The model generation unit 61 generates, for each classified cluster, a linear prediction model that predicts a design item to be predicted from design information of the design data 30A included in the cluster (S21). The combining unit 62 combines similar clusters based on the weight value for the design information in the generated linear prediction model (S22).

  As described above, the design prediction apparatus 10 classifies the plurality of design data 30A into clusters for each design data 30A having similar design information. For each classified cluster, the design prediction apparatus 10 generates a linear prediction model that predicts a design item to be predicted from design information of design data 30A included in the cluster. The design prediction apparatus 10 combines similar clusters based on the weight value for the design information in the generated linear prediction model. Thereby, the design prediction apparatus 10 can classify the plurality of design data 30A with high accuracy into clusters for each design data 30A having the same dominant feature.

  Although the embodiments related to the disclosed apparatus have been described so far, the disclosed technology may be implemented in various different forms other than the above-described embodiments. Therefore, another embodiment included in the present invention will be described below.

  For example, in the above embodiment, the case where the number of signal layers of the printed circuit board is predicted has been described as an example, but the disclosed apparatus is not limited thereto. For example, prediction of other design items may be used.

  An example of predicting other design items will be described. FIG. 21 is a diagram illustrating an example of predicting other design items. The example of FIG. 21 shows a case where the temperature of the printed circuit board is predicted with respect to the positions of the CPU and the heat sink. The extraction unit 40 extracts various information used for generation in the prediction model from the design asset data 30. For example, the extraction unit 40 extracts design information of a plurality of design items used for predicting the temperature of the printed circuit board, such as the substrate temperature, the CPU size, and the heat sink size, for each design data 30A. The classification unit 41 normalizes the design information of each design item to a feature value in the range of 0 to 1 for each design item, and obtains a feature vector having the feature value of the design item as a component for each design data 30A. By using this feature vector and generating a prediction model by classifying into clusters for each design data 30A having the same dominant feature, it is possible to accurately predict the temperature of the printed circuit board with respect to the positions of the CPU and the heat sink.

  For example, in the above-described embodiment, the case where the design item to be predicted is predicted in the mounting design of the printed board has been described as an example, but the disclosed apparatus is not limited thereto. For example, design items may be used for prediction in other designs such as system design.

  An example of predicting other design items will be described. FIG. 22 is a diagram illustrating an example of predicting other design items. The example of FIG. 22 shows a case where the disk capacity is predicted in the system design. The design asset data 30 stores various design information of a designed system as design data 30A. The extraction unit 40 extracts various information used for generation in the prediction model from the design asset data 30. For example, for each design data 30A, the extraction unit 40 uses a plurality of information used for prediction of disk capacity prediction such as information on the type of OS used, the number of users, and the type of RAID (Redundant Array of Inexpensive Disks). Extract design information of design items. The classification unit 41 normalizes the design information of each design item to a feature value in the range of 0 to 1 for each design item, and obtains a feature vector having the feature value of the design item as a component for each design data 30A. By using this feature vector and classifying into clusters for each design data 30A having the same dominant feature and generating a prediction model, the disk capacity of a newly designed system can be accurately predicted.

  Further, each component of each illustrated apparatus is functionally conceptual, and does not necessarily need to be physically configured as illustrated. In other words, the specific state of distribution / integration of each device is not limited to the one shown in the figure, and all or a part thereof may be functionally or physically distributed or arbitrarily distributed in arbitrary units according to various loads or usage conditions. Can be integrated and configured. For example, the processing units of the extraction unit 40, the classification unit 41, the generation unit 42, the reception unit 43, the specification unit 44, the calculation unit 45, and the output unit 46 may be appropriately integrated or divided. Each processing function performed by each processing unit may be realized in whole or in part by a CPU and a program that is analyzed and executed by the CPU, or may be realized by hardware using wired logic. .

[Design Prediction Program]
The various processes described in the above embodiments can also be realized by executing a program prepared in advance on a computer system such as a personal computer or a workstation. Therefore, in the following, an example of a computer system that executes a program having the same function as in the above embodiment will be described. FIG. 23 is a diagram illustrating a computer that executes a design prediction program.

  As shown in FIG. 23, the computer 300 includes a CPU 310, an HDD (Hard Disk Drive) 320, and a RAM (Random Access Memory) 340. These units 310 to 340 are connected via a bus 400.

  The HDD 320 stores in advance a design prediction program 320A that performs the same function as each processing unit of the above embodiment. For example, the design prediction program 320A that exhibits the same functions as those of the extraction unit 40, the classification unit 41, the generation unit 42, the reception unit 43, the specification unit 44, the calculation unit 45, and the output unit 46 of the above embodiment is stored. Note that the design prediction program 320A may be separated as appropriate.

  The HDD 320 stores various data. For example, the HDD 320 stores the OS and various data.

  Then, the CPU 310 reads out and executes the design prediction program 320A from the HDD 320, thereby executing the same operation as each processing unit of the embodiment. That is, the design prediction program 320A performs the same operations as the extraction unit 40, the classification unit 41, the generation unit 42, the reception unit 43, the specification unit 44, the calculation unit 45, and the output unit 46 of the embodiment.

  Note that the above-described design prediction program 320A does not necessarily need to be stored in the HDD 320 from the beginning. The design prediction program 320A may be stored in the HDD 320 from an external device.

  For example, a program is stored in a “portable physical medium” such as a flexible disk (FD), a compact disk read only memory (CD-ROM), a digital versatile disk (DVD), a magneto-optical disk, or an IC card inserted into the computer 300. Remember. Then, the computer 300 may read and execute the program from these.

  Furthermore, the program is stored in “another computer (or server)” connected to the computer 300 via a public line, the Internet, a LAN, a WAN, or the like. Then, the computer 300 may read and execute the program from these.

DESCRIPTION OF SYMBOLS 10 Design prediction apparatus 20 Input part 21 Display part 22 Storage part 23 Control part 30 Design asset data 30A Design data 31 Design information table 32 Prediction model data 33 Cluster allocation model data 40 Extraction part 41 Classification part 42 Generation part 43 Reception part 44 Identification Unit 45 calculation unit 46 output unit 50 nonlinear model generation unit 51 slope calculation unit 52 clustering unit 60 clustering unit 61 model generation unit 62 combination unit

Claims (7)

Based on design information extracted from a plurality of design data, a classification unit for classifying the plurality of design data into clusters for each design data having the same dominant characteristics with respect to the design item to be predicted;
For each cluster classified by the classification unit, a generation unit that generates a prediction model for predicting a design item to be predicted from design information of design data included in the cluster,
A design prediction apparatus characterized by comprising: The classification unit generates a non-linear prediction model for predicting the design item to be predicted from design information extracted from the plurality of design data, and the inclination of the design information of each design data in the generated non-linear prediction model The design prediction apparatus according to claim 1, wherein the plurality of design data is classified into clusters for each design data having the same dominant feature based on the calculated slope. The classification unit classifies the plurality of design data into clusters for each design data having similar design information, and predicts a design item to be predicted from the design information of the design data included in the cluster for each classified cluster. The design prediction apparatus according to claim 1, further comprising: combining similar clusters based on a weight value for design information in the generated linear prediction model. A reception unit for receiving design information of design data to be predicted;
Based on the design information received by the receiving unit, a specifying unit that identifies which cluster the design data of the prediction target is classified;
A calculation unit that calculates a predicted value of a design item to be predicted from design information received by the receiving unit using a prediction model corresponding to the cluster specified by the specifying unit;
An output unit that outputs a predicted value of the design item to be predicted calculated by the calculation unit;
The design prediction apparatus according to claim 1, further comprising: The calculation unit further calculates a contribution indicating the degree to which the design information received by the reception unit affects the predicted value, and the output unit further outputs the calculated contribution. 4. The design prediction apparatus according to 4. Based on design information extracted from a plurality of design data, classify the plurality of design data into clusters for each design data having the same dominant characteristics with respect to the design item to be predicted,
A design prediction program that causes a computer to execute a process for generating a prediction model for predicting a design item to be predicted from design information of design data included in the cluster for each classified cluster. Based on design information extracted from a plurality of design data, classify the plurality of design data into clusters for each design data having the same dominant characteristics with respect to the design item to be predicted,
A design prediction method, wherein a computer executes a process for generating a prediction model for predicting a design item to be predicted from design information of design data included in the cluster for each classified cluster.

Images