JP7779779B2 - Performance prediction system and performance prediction method - Google Patents

Description

The present invention relates to a performance prediction system and method for predicting the amount of deflection when a load is applied to various metal panels, such as an automobile engine hood or door.

For example, an automobile engine hood needs to be rigid enough to prevent dents or deformation from minor impacts, and high dent resistance is required, but predicting engine hood performance is essential to satisfy these requirements.

Patent Document 1 below discloses a dent stiffness prediction method for predicting dent stiffness from design drawings of various metal panels, such as automobile engine hoods and doors. Specifically, the area of the stress-affected region defined around a measurement point when a load is applied to the measurement point is calculated as the deflection area. A predetermined relationship is then determined using multiple regression analysis or the like, using the curvature of the plate member at the measurement point, the plate thickness of the plate member at the measurement point, the plate material, and the deflection area as factors representing the plate member's stiffness. Then, by extracting the curvature of the plate member at the measurement point, the plate thickness of the plate member at the measurement point, the plate material, and the deflection area from the design drawings, the displacement of the measurement point in the load direction (dent stiffness) can be appropriately calculated at the design stage based on the predetermined relationship.

In this way, dent rigidity is determined early in the development process, preventing development delays due to delayed evaluation of dent rigidity.

JP 2007-33067 A (Patent No. 4568186)

The dent stiffness prediction method described in Patent Document 1 above has the problem that, when deriving the correlation equation (predetermined relational equation) and calculating the amount of deflection (dent stiffness), it is necessary to measure the curvature, plate thickness, material, and area of the stress-affected region defined around the measurement point each time a load is applied to the measurement point, which is a time-consuming process. Furthermore, these measurement results vary depending on the operator, and this variation affects the accuracy of the predicted displacement (dent stiffness) of the measurement point.

The present invention was made in consideration of the above-mentioned problems, and aims to provide a performance prediction system and method that significantly reduces the time required to calculate the amount of deflection when a load is applied to various automotive panels, and that can predict dent performance stably and with high accuracy.

The performance prediction system of the present invention predicts the performance of a metal panel that constitutes an automobile body. For example, the system includes a storage means for storing a dataset in which explanatory variables are data linking point cloud coordinate data of a point cloud around load points uniformly distributed on the panel with information on the material and thickness of each point that constitutes the point cloud, and the explanatory variables at each load point are linked to an objective variable that represents the amount of deflection when a certain load is applied to the load point. The model generation means uses the dataset read from the storage means to generate a performance prediction model through machine learning that predicts the amount of deflection when a certain load is applied to a performance prediction point at any location on the panel. The performance prediction means accepts input data linking point cloud coordinate data of a point cloud around the performance prediction point with information on the material and thickness of each point that constitutes the point cloud, and predicts the amount of deflection when a certain load is applied to the performance prediction point using the trained performance prediction model.

In the performance prediction system of the present invention, the storage means pre-stores 3D design information for the panel, information on the panel's material and thickness associated with the 3D design information, and the amount of deflection at each load point obtained using a known method. The point cloud coordinate data generation means then positions the panel in 3D coordinate space (X coordinate, Y coordinate, Z coordinate) based on the 3D design information for the panel, obtains multiple cross-sectional shapes of the panel centered on the load point or performance prediction point, divides the obtained cross-sectional shapes into squares of specified angles to generate a point cloud, which is a collection of their center points, and obtains the coordinates (X coordinate, Y coordinate, Z coordinate) of each point forming the point cloud to generate the point cloud coordinate data.

As a result, the dataset generation means can generate the dataset for model generation based on the point cloud coordinate data generated by the point cloud coordinate data generation means, 3D design information for the panel, information on the material and thickness of the panel associated with the 3D design information, and the amount of deflection at each load point acquired using the known method. Furthermore, the input data generation means can generate the input data for performance prediction based on the point cloud coordinate data generated by the point cloud coordinate data generation means, 3D design information for the panel, and information on the material and thickness of the panel associated with the 3D design information.

Therefore, with the performance prediction system of the present invention, the introduction of a performance prediction model can significantly reduce the time required to calculate the amount of deflection when a load is applied to any position (performance prediction point) on various automotive panels. Furthermore, the use of a performance prediction model eliminates variations that occur between workers, allowing for stable and highly accurate prediction of dent performance.

Furthermore, it is desirable that the performance prediction system of the present invention further comprises a display means for displaying the predicted results of the amount of deflection when a certain load is applied to the performance prediction point, and for displaying a 3D image of the panel arranged in the 3D coordinate space.

The performance prediction method of the present invention predicts the performance of a metal panel that constitutes an automobile body, and is characterized by including, for example, a dataset storage step in which data relating point cloud coordinate data of a point cloud around load points uniformly provided on the panel to information on the material and thickness of each point that constitutes the point cloud is used as explanatory variables, and a dataset in which the explanatory variables at each load point are linked to an objective variable that is the amount of deflection when a certain load is applied to the load point is stored in memory; a model generation step in which, using the dataset read from the memory, a performance prediction model is generated by machine learning to predict the amount of deflection when a certain load is applied to a performance prediction point at any location on the panel; a performance prediction step in which input data relating point cloud coordinate data of a point cloud around the performance prediction point to information on the material and thickness of each point that constitutes the point cloud is received, and the learned performance prediction model is used to predict the amount of deflection when a certain load is applied to the performance prediction point; and a display step in which the results of the prediction are displayed on a display.

In addition, in the performance prediction method according to the present invention, the memory pre-stores 3D design information for the panel, information on the material and thickness of the panel associated with the 3D design information, and the amount of deflection at each load point obtained using a known method.

The preprocessing steps for generating the performance prediction model include a shape placement step for placing the panel in a 3D coordinate space (X coordinate, Y coordinate, Z coordinate) based on the 3D design information of the panel; a cross-sectional shape acquisition step for acquiring multiple cross-sectional shapes of the panel centered on the load point; a point cloud generation step for dividing the acquired cross-sectional shapes with squares of a predetermined angle to generate a point cloud, which is a collection of center points of the cross-sectional shapes; a point cloud coordinate data generation step for acquiring the coordinates (X coordinate, Y coordinate, Z coordinate) of each point forming the point cloud to generate point cloud coordinate data; and a dataset generation step for generating the dataset for model generation based on the point cloud coordinate data generated in the point cloud coordinate data generation step, the 3D design information of the panel, information on the material and thickness of the panel associated with the 3D design information, and the amount of deflection at each load point acquired using the known method.

Furthermore, the pre-processing for the performance prediction step includes the shape placement step, the cross-sectional shape acquisition step of acquiring multiple cross-sectional shapes of the panel centered on the performance prediction point, the point cloud generation step, the point cloud coordinate data generation step, and an input data generation step of generating the input data for performance prediction based on the point cloud coordinate data generated in the point cloud coordinate data generation step, 3D design information for the panel, and information on the material and thickness of the panel associated with the 3D design information.

Therefore, according to the performance prediction method of the present invention, by introducing a performance prediction model, it is possible to significantly reduce the time required to calculate the amount of deflection when a load is applied to any position (performance prediction point) on various automotive panels. Furthermore, the use of a performance prediction model eliminates variations that occur between workers, allowing for stable and highly accurate prediction of dent performance.

Furthermore, it is desirable that the performance prediction method according to the present invention further includes a display step of displaying the panel arranged in the 3D coordinate space as a 3D image on a display.

The performance prediction system and method of the present invention can significantly reduce the time required to calculate the amount of deflection when a load is applied to various automotive panels. It also makes it possible to predict dent performance stably and with high accuracy.

FIG. 1 is a diagram showing an example of the hardware configuration of a computer that operates as a performance prediction system according to the present invention. FIG. 2 is a functional block diagram showing the functions of the control unit that performs the learning model generation process. FIG. 3 is a flowchart illustrating an example of a learning model generation process. FIG. 4 is a diagram showing an example of a 3D image of an engine hood. FIG. 5 is a diagram showing the positions where the cross-sectional shape is acquired. FIG. 6 is a diagram showing an example of a cross-sectional shape of an engine hood and an image of the cross-sectional shape in a point cloud form. FIG. 7 is a diagram illustrating an example of point cloud coordinate data. FIG. 8 is a diagram illustrating an example of a data set for generating a performance prediction model. FIG. 9 is a diagram showing the evaluation results of the machine learning model. FIG. 10 is a diagram showing prediction results by a trained performance prediction model using test data. FIG. 11 is a functional block diagram showing the functions of the control unit that performs the performance prediction process. FIG. 12 is a flowchart illustrating an example of a performance prediction process.

Embodiments of a performance prediction system and a performance prediction method according to the present invention will be described in detail below with reference to the accompanying drawings. Note that the present invention is not limited to these embodiments. Furthermore, in the specification and drawings of this application, elements that can be similarly described may be designated by the same reference numerals, and duplicate explanations may be omitted.

<System Configuration>
1 is a diagram showing an example of the hardware configuration of a computer that operates as a performance prediction system according to the present invention. The performance prediction system 1 of this embodiment operates as a host computer that performs a process of predicting the amount of deflection (dent performance) when a load is applied to various metal panels for automobiles (hereinafter referred to as a performance prediction process) and a process of generating a learning model that executes the deflection amount prediction (hereinafter referred to as a learning model generation process).

In FIG. 1, performance prediction system 1 includes a control unit 11 composed of a CPU (Central Processing Unit) and FPGA (Field Programmable Gate Array), etc.; a storage unit 12 including various types of memory such as ROM (Read Only Memory) and RAM (Random Access Memory); an input unit 13 including a user interface such as a keyboard and mouse; an interface (I/F) unit 14 performing input/output processes such as printing and scanning; a display unit 15 serving as a display; and a communication unit 16 for communicating with the outside world via a predetermined network. While FIG. 1 shows an input unit 13 including a user interface such as a keyboard and mouse, the performance prediction system 1 of this embodiment is not limited to this. By providing touch panel functionality to the display unit 15, the input unit 13 may be omitted or used in combination with the input unit 13.

1, the control unit 11 executes, for example, a performance prediction program that predicts the amount of deflection when a load is applied to various automotive panels and a learning model generation program that generates a learning model for executing deflection prediction, in order to realize the performance prediction process and learning model generation process by the performance prediction system 1 of this embodiment. The memory unit 12 stores the programs (performance prediction program, learning model generation program) related to the performance prediction process and learning model generation process of this embodiment, various information (3D shape data indicating the shapes of various panels, material, plate thickness, and performance, datasets for machine learning, etc.), and various data (point cloud coordinate data, etc.) obtained during the process. The control unit 11 executes the performance prediction process and learning model generation process of this embodiment by reading the various programs stored in the memory unit 12. Note that the memory unit 12 is not limited to internal memory and may be an external storage medium such as a DVD (Digital Versatile Disc) or SD memory, or may be composed of both internal memory and external storage medium (such as a DVD or SD memory). Furthermore, for the sake of convenience, the hardware configuration of the performance prediction system 1 of this embodiment lists the configuration related to the performance prediction processing and learning model generation processing of this embodiment, and does not represent all of the functions of the computer that makes up the performance prediction system 1.

Furthermore, the performance prediction system 1 of this embodiment is intended for general-purpose PCs such as desktop personal computers and laptop computers, but is not limited to these and may also be, for example, mobile devices such as smartphones and tablet terminals.

<Learning model generation process>
Next, before describing the performance prediction process in the performance prediction system 1 of this embodiment, a learning model generation process, which is a prerequisite for the performance prediction process, will be described in detail. Fig. 2 is a functional block diagram showing the functions of the control unit 11 that performs the learning model generation process, and Fig. 3 is a flowchart showing an example of the learning model generation process.

In FIG. 2, the control unit 11 includes a point cloud coordinate data generation unit 21, a dataset generation unit 22, a machine learning unit 23 that generates a learning model using a machine learning algorithm, and a known CAD (computer-aided design) system 24 and CAE (computer-aided engineering) system 25. In this embodiment, a case will be described in which a learning model is generated using 3D design information for an automobile engine hood as an example of various metal panels. That is, in this embodiment, as an example of the learning model, a performance prediction model is generated that predicts the amount of deflection when a predetermined load (e.g., a constant load) is applied to an arbitrary position on the automobile engine hood. This performance prediction model is generated using a known machine learning algorithm, such as a neural network (CNN: Convolution Neural Network, RNN: Recurrent Neural Network, etc.).

The automobile engine hood is designed, for example, using the CAD system 24 shown in Figure 2, and 3D design information including 3D shape data is stored in advance in the storage unit 12. Information such as the material and thickness of each part that makes up the engine hood is also stored in advance in the storage unit 12 in association with the 3D design information (3D shape data).

In addition, in this embodiment, when a certain load is applied to predetermined positions (load points: for example, 143 locations) on the engine hood, the amount of deflection at each load point is obtained in advance, and these deflection amounts are associated with each load point and stored in advance in the memory unit 12. Specifically, for example, the CAE system 25 shown in FIG. 2 reads 3D design information for the engine hood from the memory unit 12, performs CAE analysis using the finite element method based on this 3D design information, and obtains the amount of deflection (performance information) at each load point as a result of the analysis. In CAAE analysis using the finite element method, the 3D shape of the engine hood is divided into a mesh, and by applying loads to the load points, deformation, etc. of each portion of the divided mesh is analyzed.

The 3D design using the CAD system 24 and the CAE analysis using the CAE system 25 are performed using well-known, commonly-used methods, and therefore detailed explanations are omitted. In this embodiment, CAE analysis using the finite element method is performed based on the 3D design information to obtain the amount of deflection at each load point in advance. However, this is not a limitation, and any known method may be used to obtain the amount of deflection at the load point. In this embodiment, as an example, 143 load points are used to apply the load, but this is not a limitation; it is desirable to have 143 or more load points depending on the required accuracy of performance prediction, etc. In other words, the load points to apply the load need only be uniformly distributed (evenly distributed) across the entire panel, and can be set appropriately depending on, for example, the shape, material, and thickness of the panel, the stress range when the load is applied, the required accuracy of performance prediction, etc.

In addition, in Figure 2, the CAD system 24 and CAE system 25 are implemented as functional blocks within the control unit 11 of the performance prediction system 1, but this configuration is not limited to this. For example, the CAD system 24 and CAE system 25 may be configured separately and connected to the performance prediction system 1 via a network.

In this embodiment, assuming that various types of information as described above are stored in advance in the memory unit 12, for example, as shown in FIG. 3, the control unit 11 executes a process (learning model generation process) to generate a performance prediction model using 3D design information for the automobile engine hood, information on the material and plate thickness of the engine hood, and the deflection amount at each load point acquired in advance.

In Figure 3, when an operator operates the input unit 13 to instruct the generation of a performance prediction model, the point cloud coordinate data generation unit 21 of the control unit 11 reads 3D design information for the engine hood from the memory unit 12 and places the engine hood (shape) in 3D coordinate space (X coordinate, Y coordinate, Z coordinate) based on the 3D shape data obtained from the 3D design information (step S1). At this time, a 3D image of the engine hood is displayed on the display unit 15. Figure 4 is a diagram showing an example of a 3D image of the engine hood.

Next, the point cloud coordinate data generator 21 acquires the cross-sectional shapes of the engine hood around the load points (143 locations) from the 3D design information (step S2). Specifically, the cross-sectional shapes of the engine hood, for example, 150 mm wide in the front-to-back and left-to-right directions, centered on the load points, are acquired. Figure 5 is a diagram showing the positions at which the cross-sectional shapes are acquired. In this embodiment, the cross-sectional shapes of the engine hood 150 mm wide in the left-to-right direction and centered on the load points (see line A-A' (a line parallel to the X-axis) centered on load point C in Figure 5) and the cross-sectional shapes of the engine hood 150 mm wide in the front-to-back direction and centered on the load points (see line B-B' (a line parallel to the Y-axis) centered on load point C in Figure 5) are acquired for each load point.

Figures 6(a) and (b) are diagrams showing an example of the cross-sectional shape of an engine hood. In detail, (a) shows the A-A' cross section at load point C, and (b) shows the B-B' cross section at load point C. In this embodiment, the width of the cross-sectional shape to be acquired is set to 150 mm as an example, but this is not limited to this. It is sufficient to acquire a uniform (evenly varying) cross-sectional shape from the entire panel. For example, this width can be set appropriately depending on the panel shape, material, and thickness, the stress range when a load is applied, the required accuracy of performance prediction, and other factors. In this embodiment, two orthogonal cross-sectional shapes are acquired, such as the A-A' cross section and the B-B' cross section shown in Figure 5. However, this is not limited to this. For example, to improve the accuracy of performance prediction, cross-sectional shapes in other directions centered on the load point may also be acquired.

The point cloud coordinate data generator 21 then divides the cross-sectional shape (150 mm wide) acquired for each load point into 5 mm intervals (5 mm x 5 mm squares), for example, and generates a point cloud of the cross-sectional shape (a collection of center points of 5 mm squares) for each load point (step S3). The spacing between each point constituting the point cloud was set to 5 mm based on the balance between the shape representation accuracy (fillet R, etc.) and the number of features when the cross-sectional shape was converted into a point cloud. The point cloud coordinate data generator 21 then acquires the coordinates (X coordinate, Y coordinate, Z coordinate) of each point constituting the point cloud based on the 3D shape data of the engine hood arranged in 3D coordinate space, generates point cloud coordinate data of the cross-sectional shape for each load point (step S4), and stores this in the memory unit 12. Figures 6(c) and (d) are schematic diagrams showing an image of the cross-sectional shape acquired at load point C in Figure 5 converted into a point cloud. In detail, (c) shows a point cloud image of the A-A' cross section at load point C, and (b) shows a point cloud image of the B-B' cross section at load point C. Figure 7 is a diagram showing an example of point cloud coordinate data. Note that in this embodiment, the spacing between points in the point cloud is set to 5 mm as an example, but this is not limited to this and can be set appropriately depending on, for example, the shape, material, and thickness of the panel, the stress range when a load is applied, the required accuracy of performance prediction, etc.

As described above, after the point cloud coordinate data generator 21 generates point cloud coordinate data for each load point, the dataset generator 22 reads from the storage unit 12 the point cloud coordinate data, 3D design information (including information on the material and thickness of each component constituting the engine hood), and the deflection amount (performance information) for each load point previously acquired through CAE analysis and stored. The dataset generator 22 then links the coordinates of each point forming the point cloud coordinate data, the material and thickness of each point, and the deflection amount for each load point to generate a dataset for machine learning (step S5). That is, the dataset used in the machine learning algorithm uses data linking the coordinates of each point forming the point cloud coordinate data corresponding to a single load point with the material and thickness of each point as explanatory variables (input data), the deflection amount for that load point previously acquired as the objective variable (training data), and the training data corresponding to each load point individually linked to the input data prepared for each load point. The dataset generation unit 22 then stores the generated dataset in the storage unit 12.

Figure 8 shows an example of a dataset for generating a performance prediction model (a dataset for 120 load points out of a dataset for 143 load points for generating a performance prediction model). The machine learning unit 23 of the control unit 11 reads the dataset for 120 load points shown in Figure 8 from the storage unit 12 and performs supervised learning using the dataset using a neural network, which is a machine learning algorithm, to generate a performance prediction model that predicts the amount of deflection when a constant load is applied to any location on the automobile engine hood (step S6). That is, in this embodiment, a machine learning algorithm is used to learn the correlation between the 3D shape data (e.g., point cloud coordinate data for each load point), material, and plate thickness of the automobile engine hood and the amount of deflection when a constant load is applied to each load point on the engine hood, thereby generating a performance prediction model that predicts the amount of deflection when a constant load is applied to any location on the automobile engine hood.

In this embodiment, as described above, 143 datasets (143 load points) were prepared for machine learning. Of these, 120 were used as datasets for generating the performance prediction model; that is, 80 were used as training data and 40 were used as evaluation data. The remaining 23 were used as test data. Here, the training data is a dataset used to train the performance prediction model, the evaluation data is a dataset used to verify the performance of the model, and the test data is a dataset used for final evaluation of the model by comparing it with previously acquired deflection amounts. Furthermore, a sufficient number of training data is prepared to achieve the desired prediction accuracy (e.g., a coefficient of determination R2 in the range of 0.8 to 0.9) without overfitting or underfitting. Furthermore, the performance of the performance prediction model in this embodiment is evaluated using evaluation indices such as mean squared error (MSE) and mean absolute error (MAE). In other words, the smaller these values, the less error there is in the model. Note that evaluation indices are not limited to these.

Figure 9 shows the evaluation results of a machine learning model, showing, for example, the trends in mean squared error and mean absolute error versus the degree of learning (epochs). As shown in Figure 9, the mean squared error and mean absolute error of the performance prediction model of this embodiment decrease as learning progresses, confirming that learning is progressing smoothly. Furthermore, since the mean squared error and mean absolute error become sufficiently small from around epoch number: 50 and change little thereafter, it can be said that in this embodiment, an epoch number of around 50 is suitable for training a performance prediction model.

Figure 10 shows the prediction results of the trained performance prediction model using 23 test data sets. Here, the predicted values (deflection amounts) output from the performance prediction model were plotted against the training data (actual measurements: deflection amounts acquired in advance), and the results were observed. In Figure 10, the dotted line indicates the range where the coefficient of determination for the line of actual measurements connecting the plotted data (circles) is R2 = 0.8 or greater. As a result, the approximation curve for the predicted values of the performance prediction model (deflection amounts at 23 locations indicated by crosses) satisfied the coefficient of determination R2 = 0.8 or greater, indicating that good prediction results were obtained.

<Example of a machine learning algorithm>
The machine learning algorithm used in this embodiment is, for example, a known neural network such as a convolutional neural network (CNN) or a recurrent neural network (RNN). The machine learning unit 23 of the control unit 11 calculates the mean square error between the predicted result (deflection amount), which is the output of the output layer of the neural network, and the training data (deflection amount), and repeatedly performs learning over a predetermined number of epochs so as to minimize this error.

In this embodiment, the above-mentioned performance prediction model is generated using neural networks such as CNN and RNN as an example of a machine learning algorithm, but the machine learning algorithm for generating the performance prediction model is not limited to this. For example, other machine learning algorithms such as random forests and boosted decision trees can also be used.

<Performance prediction processing>
Next, a detailed description will be given of the performance prediction process in the performance prediction system 1 of this embodiment. Fig. 11 is a functional block diagram showing the functions of the control unit 11 that performs the performance prediction process, and Fig. 12 is a flowchart showing an example of the performance prediction process.

In FIG. 11, the control unit 11 has the point cloud coordinate data generation unit 21 described above, the input data generation unit 26, and the performance prediction unit 27 that operates as a performance prediction model. In this embodiment, as above, 3D design information for an automobile engine hood is used as an example of various metal panels, and the learning model (performance prediction model) generated above is used to predict the amount of deflection at a performance prediction point on the engine hood (any point where the amount of deflection is desired). That is, in this embodiment, the learned performance prediction model is used to predict the amount of deflection when a certain load is applied to the performance prediction point on the automobile engine hood.

Specifically, when an operator operates the input unit 13 to instruct execution of the performance prediction process, the point cloud coordinate data generator 21 of the control unit 11 reads 3D design information for the engine hood from the memory unit 12 and places the engine hood (shape) in 3D coordinate space (X coordinate, Y coordinate, Z coordinate) based on the 3D shape data obtained from the 3D design information (step S11). At this time, a 3D image of the engine hood is displayed on the display unit 15 (see Figure 4).

Next, when the operator operates the input unit 13 to set a performance prediction point, which is an arbitrary position on the engine hood where the amount of deflection is desired, the point cloud coordinate data generator 21 acquires the cross-sectional shape of the engine hood around the performance prediction point from the 3D design information (step S12). Specifically, the cross-sectional shape of the engine hood, for example, 150 mm wide in the front-to-back and left-to-right directions, centered on the performance prediction point, is acquired. That is, in this embodiment, the cross-sectional shape of the engine hood, 150 mm wide in the left-to-right direction and centered on the performance prediction point, and the cross-sectional shape of the engine hood, 150 mm wide in the front-to-back direction and centered on the performance prediction point, are acquired (see Figures 5 and 6). Note that the width dimension of the cross-sectional shape acquired here is the same as the width dimension of the cross-sectional shape acquired when generating the performance prediction model (see step S2 in Figure 3).

Furthermore, the point cloud coordinate data generator 21 divides the cross-sectional shape (150 mm wide) in the longitudinal and lateral directions centered on the performance prediction point into, for example, 5 mm intervals (5 mm x 5 mm squares) and generates a point cloud (a collection of center points of 5 mm squares) of the cross-sectional shape at the performance prediction point (step S13). The point cloud coordinate data generator 21 then obtains the coordinates (X coordinate, Y coordinate, Z coordinate) of each point forming the point cloud based on the 3D shape data of the engine hood arranged in 3D coordinate space, generates point cloud coordinate data of the cross-sectional shape at the performance prediction point (step S14) (see Figures 6(c), (d), and 7), and stores this in the memory unit 12. The spacing between each point forming the point cloud of the cross-sectional shape at the performance prediction point is set to the same spacing as the spacing between each point forming the point cloud of the cross-sectional shape obtained for each load point when generating the performance prediction model (see step S3 in Figure 3).

After the point cloud coordinate data generation unit 21 generates the point cloud coordinate data for the performance prediction points as described above, the input data generation unit 26 then reads the point cloud coordinate data for the performance prediction points and 3D design information (including information on the material and thickness of each component that makes up the engine hood) from the storage unit 12, and links the coordinates of each point that form the point cloud coordinate data with the material and thickness of each point to generate input data that will be input to the performance prediction model (step S15). The input data generation unit 26 then inputs the generated input data to the performance prediction unit 27, which operates as a performance prediction model.

The performance prediction unit 27 accepts the input data generated by the input data generation unit 26 and uses the performance prediction model to predict the amount of deflection at the performance prediction point on the engine hood (any point where the amount of deflection is desired) (step S16). That is, in this embodiment, the trained performance prediction model is used to predict the amount of deflection (the amount of deflection at the performance prediction point) when a certain load is applied to the performance prediction point on the automobile engine hood. The amount of deflection at the performance prediction point is then displayed on the display unit 15 as a performance prediction value.

<Effects>
As described above, the performance prediction system 1 of this embodiment predicts the amount of deflection due to a load, which is one of the performance characteristics of a metal panel (e.g., an engine hood of an automobile) that constitutes the body of an automobile. Specifically, the system includes a storage unit 12 that stores a data set (see FIG. 8 ) in which, for each load point, explanatory variables are used, which associates point cloud coordinate data (see FIG. 7 ) of a point cloud around load points uniformly provided on the engine hood with information on the material and plate thickness of each point that constitutes the point cloud, and the explanatory variables (input data) at each load point are linked to a target variable (teacher data) that is the amount of deflection when a certain load is applied to the load point. The machine learning unit 23 uses the data set read from the storage unit 12 to generate, by machine learning, a performance prediction model that predicts the amount of deflection when a certain load is applied to a performance prediction point of the engine hood. The performance prediction unit 27 then receives input data that associates point cloud coordinate data of a point cloud around a performance prediction point, which is any location on the engine hood, with information on the material and plate thickness of each point that makes up the point cloud, and predicts the amount of deflection when a certain load is applied to the performance prediction point using the trained performance prediction model.

Furthermore, the memory unit 12 pre-stores 3D design information for the engine hood, information on the material and plate thickness of the engine hood associated with the 3D design information, and the amount of deflection at each load point previously obtained using a known method. The point cloud coordinate data generation unit 21 then positions the engine hood in 3D coordinate space (X coordinate, Y coordinate, Z coordinate) based on the 3D design information for the engine hood, obtains multiple cross-sectional shapes of the panel centered on the load point or performance prediction point, divides the obtained cross-sectional shapes into squares of specified angles to generate a point cloud, which is a collection of center points, and obtains the coordinates (X coordinate, Y coordinate, Z coordinate) of each point forming the point cloud to generate point cloud coordinate data.

As a result, the dataset generation unit 22 can generate a dataset for generating a learning model based on the point cloud coordinate data generated by the point cloud coordinate data generation unit 21, the 3D design information of the engine hood, the material and thickness information of the engine hood associated with the 3D design information, and the deflection amount at each load point acquired in advance. Furthermore, the input data generation unit 26 can generate input data for performance prediction based on the point cloud coordinate data generated by the point cloud coordinate data generation unit 21, the 3D design information of the engine hood, and the material and thickness information of the engine hood associated with the 3D design information.

In this way, with the performance prediction system 1 of this embodiment, by introducing a performance prediction model, it is possible to significantly reduce the time required to calculate the amount of deflection when a load is applied to any position (performance prediction point) on various automotive panels. Furthermore, the use of a performance prediction model eliminates variations that occur between workers, allowing for stable and highly accurate prediction of dent performance.

In this embodiment, an automobile engine hood is used as an example of various metal panels, but this is not limited to this. The learning model generation process and performance prediction process of this embodiment can be applied to all metal panels that make up an automobile body, such as engine hoods, front fenders, roof panels, doors, etc.

In addition, in this embodiment, a performance prediction model is generated that predicts the amount of deflection when a certain load is applied to an arbitrary position on a metal panel. However, it is also possible to generate a performance prediction model that predicts the amount of deflection corresponding to the magnitude and direction of the load by adding load information (such as the magnitude and direction of the load) as an explanatory variable.

REFERENCE SIGNS LIST 1 Performance prediction system 11 Control unit 12 Memory unit 13 Input unit 14 Interface (I/F) unit 15 Display unit 16 Communication unit 21 Point cloud coordinate data generation unit 22 Data set generation unit 23 Machine learning unit 24 CAD system 25 CAE system 26 Input data generation unit 27 Performance prediction unit

Claims (5)

In a performance prediction system that predicts the performance of metal panels that make up an automobile body,
a storage means for storing a data set in which explanatory variables are data relating point cloud coordinate data of a point cloud around load points uniformly provided on the panel with information on the material and plate thickness of each point constituting the point cloud, and the explanatory variables at each load point are linked to a target variable, which is the amount of deflection when a certain load is applied to the load point;
a model generation means for generating a performance prediction model by machine learning, using the data set read from the storage means, to predict the amount of deflection when a certain load is applied to a performance prediction point, which is an arbitrary location of the panel; and
a performance prediction means for receiving input data that associates point cloud coordinate data of a point cloud around a performance prediction point with information on the material and plate thickness of each point that constitutes the point cloud, and predicting the amount of deflection when a certain load is applied to the performance prediction point using a trained performance prediction model;
Equipped with
A performance prediction system comprising: The storage means stores in advance 3D design information of the panel, information on the material and thickness of the panel associated with the 3D design information, and the amount of deflection at each load point obtained by a known method;
moreover,
a point cloud coordinate data generation means for arranging the panel in a 3D coordinate space (X coordinate, Y coordinate, Z coordinate) based on 3D design information of the panel, acquiring a plurality of cross-sectional shapes of the panel centered on a load point or a performance prediction point, dividing the acquired cross-sectional shapes into squares of a predetermined angle to generate a point cloud which is a collection of center points, acquiring the coordinates (X coordinate, Y coordinate, Z coordinate) of each point forming the point cloud to generate the point cloud coordinate data;
a dataset generation means for generating the dataset for model generation based on the point cloud coordinate data generated by the point cloud coordinate data generation means, 3D design information of the panel, information on the material and thickness of the panel associated with the 3D design information, and the amount of deflection at each load point obtained by the known method;
an input data generation means for generating the input data for performance prediction based on the point cloud coordinate data generated by the point cloud coordinate data generation means, 3D design information of the panel, and information on the material and thickness of the panel associated with the 3D design information;
Equipped with
The performance prediction system according to claim 1 . A performance prediction method for predicting the performance of a metal panel that constitutes a body of an automobile, comprising:
a data set storage step of storing in memory a data set in which, for each load point, the explanatory variables are linked to a response variable, which is the amount of deflection when a certain load is applied to the load point; and the data set is stored in memory, the explanatory variables being data that associates point cloud coordinate data of a point cloud around a load point uniformly provided on the panel with information on the material and plate thickness of each point constituting the point cloud.
a model generation step of generating, by machine learning, a performance prediction model that predicts the amount of deflection when a certain load is applied to a performance prediction point, which is an arbitrary location of the panel, using the data set read from the memory; and
a performance prediction step of receiving input data that associates point cloud coordinate data of a point cloud around a performance prediction point with information on the material and plate thickness of each point that constitutes the point cloud, and predicting the amount of deflection when a certain load is applied to the performance prediction point using a trained performance prediction model;
a display step of displaying the result of the prediction on a display;
Including,
A performance prediction method comprising: The memory stores in advance 3D design information of the panel, information on the material and thickness of the panel associated with the 3D design information, and a deflection amount at each load point obtained by a known method;
Furthermore, as a preprocessing for generating the performance prediction model,
a shape placement step of placing the panel in a 3D coordinate space (X coordinate, Y coordinate, Z coordinate) based on 3D design information of the panel;
a cross-sectional shape acquisition step of acquiring a plurality of cross-sectional shapes of the panel centered on a load point;
a point cloud generation step of dividing the obtained cross-sectional shape into squares of a predetermined angle to generate a point cloud which is a collection of center points of the squares;
a point cloud coordinate data generation step of acquiring coordinates (X coordinate, Y coordinate, Z coordinate) of each point forming the point cloud and generating point cloud coordinate data;
a dataset generation step of generating the dataset for model generation based on the point cloud coordinate data generated in the point cloud coordinate data generation step, 3D design information of the panel, information on the material and thickness of the panel associated with the 3D design information, and the deflection amount at each load point acquired by the known method;
Including,
4. The performance prediction method according to claim 3. Furthermore, as a preprocessing step of the performance prediction step,
the shape placement step;
a cross-sectional shape acquisition step of acquiring a plurality of cross-sectional shapes of the panel centered on a performance prediction point;
the point cloud generation step;
the point cloud coordinate data generation step;
an input data generation step of generating the input data for performance prediction based on the point cloud coordinate data generated in the point cloud coordinate data generation step, 3D design information of the panel, and information on the material and plate thickness of the panel associated with the 3D design information;
Including,
5. The performance prediction method according to claim 4.