TWI497063B - Three dimensional image processing method and three dimensional image processing device for fibrous fillers in composite materials - Google Patents

Description

Three-dimensional image processing method for fibrous filler in composite material and three-dimensional image processing device

The present invention relates to a three-dimensional image processing method and a three-dimensional image processing apparatus for a fibrous filler in a composite material.

In recent years, electronic components that are molded by resin injection have been miniaturized and slimmed for use in electronic devices that continue to be miniaturized. In order to increase the rigidity or adjust the linear expansion coefficient, the molded article which is formed by the resin is formed by mixing a composite material obtained by mixing a fibrous filler such as glass fiber into a resin base material. In order to ensure dimensional accuracy, the molded article is designed to be designed to predict warpage after molding in advance, and to perform molding under conditions in which warpage is less likely to occur. The warpage of the molded article is a distribution of the shape of the molded article, a distribution of heat history to the molded article, a distribution of thermomechanical properties such as a linear expansion coefficient or a Young's modulus in the molded article, and a heat load to the molded article. Produced by compounding. The warpage deformation occurs as a reaction of a molded article with a heat load or the like, and is determined by the distribution of thermal characteristics of the molded article and the distribution of mechanical properties. These characteristic distributions are determined by the distribution of the number of fibrous fillers at each point in the molded article and the distribution (orthogonal distribution) of the longitudinal direction of the fibrous filler. Therefore, the molding conditions or the forming mold can be optimized by grasping the distribution and the distribution of the fibrous filler in the molded article, so that the distribution can be optimized, thereby suppressing or controlling the warpage of the molded article. Deformation.

A method for evaluating the distribution of the fibrous filler in a molded article in a non-destructive manner In other words, X-ray computed tomography (CT) is used to obtain a three-dimensional image of a molded article, and the three-dimensional image is analyzed to determine the morphology or distribution of the fiber (for example, refer to Non-Patent Document 1). ). This evaluation method evaluates the tendency of the fiber alignment by using the tortuosity of the three-dimensional path formed by the fibers which are mutually connected to each other as an index. The tortuosity is to cut a layered rectangular area from the three-dimensional image, and to perform thinning processing of the image of the fiber in the rectangular area, and to find a path between the two ends of the path at both ends of the rectangular area. The ratio of the path length of the shortest path to the linear distance between the two ends, and is defined by the ratio.

Further, a method is known in which a three-dimensional coordinate of a blood vessel encapsulated by a resin in a semiconductor package and a blood vessel or a nervous system in a living body is directly extracted non-destructively (for example, refer to Patent Document 1). In this method, a plurality of tomographic images intersecting the predetermined direction of the subject are obtained, and two constituent points are respectively obtained from the two adjacent images as a cross section corresponding to the linear element in the subject, and are represented as The start and end of the vector of the line. Tomographic images are obtained by X-ray computed tomography or MRI. The constituent points in the tomographic image are determined by the center of gravity of the image representing the circle or ellipse of the cross section of the line, and the pair of the starting point and the end point are selected to be the shortest distance between adjacent images. The vector of the line elements is sequentially connected between the adjacent tomographic images, thereby extracting the three-dimensional coordinates of the line elements contained in the object.

[Previous Technical Literature]

[Non-patent literature]

[Non-patent Document 1] Nakano Ryo, "Vibration alignment and simulation using X-ray computed tomography", forming processing, Vol. 20, No. 4, pp237-241 (2008)

[Patent Literature]

[Patent Document 1] Japanese Patent Laid-Open Publication No. 2012-32293

However, in the evaluation method of the alignment distribution of the fibrous filler as shown in the above Non-Patent Document 1, since the image is thinned, it is possible to extract the noise portion in the image. Take as a filler. This is because the thinning process will lose the fiber thickness information and become indistinguishable from noise and fiber. Moreover, since it is not a method of independently extracting the respective fillers, it is only possible to obtain qualitative information on the tendency of the fibers to be aligned. Further, in the method of extracting the three-dimensional coordinates of the line as shown in the above Patent Document 1, the direction of the known line element is assumed to be the predetermined direction of the subject, and the case where the filler is complicatedly oriented in any direction cannot be applied. .

The present invention solves the above problems, and aims to provide a three-dimensional image processing method and a three-dimensional image processing device for a fibrous filler in a composite material, which can quantitatively extract each filler from a three-dimensional image of a composite material in which a fibrous filler is contained in a substrate. Directional information.

In order to achieve the above object, the three-dimensional image processing method of the fibrous filler in the composite material of the present invention extracts the alignment information of the filler from the three-dimensional image of the composite material containing the fibrous filler in the substrate, and is characterized in that the following steps are included. : a region selection step of selecting a candidate region estimated to be a pixel containing the representative filler from the three-dimensional image of the composite material by referring to a predetermined threshold; and extracting the step of randomly changing the shape and configuration of the shape model of the filler The Monte Carlo method of the parameter is adapted to the candidate region selected by the selection step of the region to extract the alignment information of the filler.

In the three-dimensional image processing method of the fibrous filler in the composite material, the extracting step removes the extracted region from the candidate region after extracting the alignment information of the filler from the candidate region, and removes the candidate region after the removal. The extraction using the Monte Carlo method is repeated.

The three-dimensional image processing method of the fibrous filler in the composite material may further include: a data input step of inputting the three-dimensional image as voxel data; and the voxel data is obtained by X-ray computed tomography of the composite material. Each voxel of the voxel data has a value based on the value of the X-ray intensity value.

In the three-dimensional image processing method of the fibrous filler in the composite material, the data input step includes: an interpolation step to form new voxel data, and the new voxel data has the data value between the voxels The value of the linear interpolation is used as the voxel of the data value.

In the three-dimensional image processing method of the fibrous filler in the composite material, the region selection step compares the data value of the voxel with a predetermined threshold value according to the X-ray intensity value, and has a larger than the threshold value. The set of voxels of the data values is defined as a candidate region.

In the three-dimensional image processing method of the fibrous filler in the composite material, a group formed by voxels adjacent to each other is generated from voxels belonging to the candidate region, and each of the generated groups is determined as New candidate area.

In the three-dimensional image processing method of the fibrous filler in the composite material, the extraction step adopts a virtual cylinder as the shape model, and the parameters representing the shape and configuration of the virtual cylinder are changed in the candidate region by using The degree of adaptation of the virtual cylinder to the candidate region is evaluated according to the cumulative value of the evaluation value of the data value of the voxel contained in the virtual cylinder.

In the three-dimensional image processing method of the fibrous filler in the composite material, the extracting step adopts a cloud curve defined by a plurality of control points as a central axis after performing the adaptation using the virtual cylinder. The virtual tube is used as the shape model, and the coordinates of the control point are randomly changed in the candidate area as a parameter, and the degree of adaptation of the virtual tube to the candidate area is evaluated, and the evaluation using the virtual tube is utilized. The evaluation of the virtual cylinder improves the situation above the predetermined ratio, and the adaptation using the virtual tube is adopted, otherwise the adaptation of the virtual cylinder is adopted to extract the alignment information of the filler.

In the three-dimensional image processing method of the fibrous filler in the composite material, the extraction step adopts a virtual tube in which a cloud curve defined by a plurality of control points is defined as a central axis as the shape model, and is randomly selected in the candidate region. The coordinates of the control point are changed as parameters, and the degree of adaptation of the virtual tube to the candidate area is evaluated by the cumulative value of the evaluation value of the data value of the voxel contained in the virtual tube.

In the three-dimensional image processing method of the fibrous filler in the composite material, the extracting step changes the coordinates sequentially by one randomness for the majority of the control points of the cloud-shaped curve. Each time the one control point is changed, the cloud curve is determined and the degree of adaptation is evaluated, and all control points are reconfigured to be equally spaced on the determined cloud curve before changing the next control point.

In the three-dimensional image processing method of the fibrous filler in the composite material, the extracting step sets the evaluation point in the shape model at intervals below the size of the voxel, and according to the data value of the voxel The evaluation value is given to the evaluation point, and the degree of adaptation is evaluated by giving the cumulative value of the evaluation value of the evaluation point.

In the three-dimensional image processing method of the fibrous filler in the composite material, the extraction step is performed concentrically on the surface perpendicular to the central axis of the shape model.

In the three-dimensional image processing method of the fibrous filler in the composite material, the evaluation value given to the evaluation point is a value obtained by linearly interpolating using the data value of the voxel.

In the three-dimensional image processing method of the fibrous filler in the composite material, the extraction step adopts a value obtained by subtracting the threshold value from the evaluation value as a new evaluation value.

In the three-dimensional image processing method of the fibrous filler in the composite material, the extraction step may be a new evaluation value by multiplying the value by a predetermined positive value when the new evaluation value is a negative value.

In the three-dimensional image processing method of the fibrous filler in the composite material, the extraction step is performed by weighting the evaluation value of the position closer to the central axis of the shape model to calculate the integrated value.

The three-dimensional image processing device for the fibrous filler in the composite material of the present invention extracts the alignment information of the filler from a three-dimensional image of the composite material containing the fibrous filler in the substrate, and is characterized by: a region selection mechanism, by Selecting a candidate region that is presumed to contain pixels representing the filler from a three-dimensional image of the composite with reference to a predetermined threshold; and extracting mechanism using randomness The Monte Carlo method of "parameters for changing the shape and configuration of the shape model of the filler" is changed, and the shape model is adapted to the candidate region selected by the region selection mechanism to extract the alignment information of the filler.

According to the three-dimensional image processing method of the fibrous filler in the composite material of the present invention, since each of the fillers is extracted from the three-dimensional image according to the shape model of the filler having parameters which change the shape and configuration, and Monte Carlo method, it is possible to quantify Grounding information of the packing material.

1‧‧‧glass fiber reinforced resin molded article

2‧‧‧3D image processing device

10‧‧‧Control Department

11‧‧‧Regional Selection Department (Regional Election Agency)

12‧‧‧Model Setting Department (Model Setting Organization)

13‧‧‧Extracting the main body (extraction agency)

14‧‧‧Data Input Department

15‧‧‧Operation Department

16‧‧‧Display Department

A‧‧‧ arrow

b‧‧‧Evaluation point (sampling point)

B‧‧‧ disc

B0, Bx, bx‧‧‧ voxels

C‧‧‧Composite

C1~Cn, Ci‧‧‧ control points

F‧‧‧Filling

G1~G11‧‧‧3D image

L, L0‧‧‧ length

M‧‧‧Virtual Cylinder (Shape Model)

MP‧‧‧Virtual Tube (Shape Model)

P0‧‧‧ original location

P1, P2‧‧‧ endpoint

Pc‧‧‧ Center Point

S1~S4, S21~S25, S30~S32, S101~S110, S200~S210‧‧‧ steps

Sp‧‧‧Cloud Curve

#0~#2, #0a, #1a, #21~#23‧‧‧Steps

Δ1, Δ2, Δ3‧‧‧ interval

BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a flow chart showing a method of processing a three-dimensional image of a fibrous filler in a composite material according to an embodiment of the present invention.

Fig. 2 is a flow chart showing a modification of the image processing method.

Fig. 3(a) is a perspective view showing a composite material of an example of the object to be processed by the image processing method, and Fig. 3(b) is a perspective view showing a part of the composite material of Fig. 3(a).

Fig. 4 is a view showing a three-dimensional image of the composite material shown in Fig. 3(b) by X-ray computed tomography.

Fig. 5(a) shows a three-dimensional image of an example in which a three-dimensional image of a fibrous filler in a composite material is subjected to threshold processing, and Fig. 5(b) shows a white-black inverted image of the image of Fig. 5(a).

Fig. 6 is a flow chart showing another modification of the image processing method.

7(a) and 7(b) are plan views showing the voxels of the interpolation step in the modification.

Fig. 8 is a perspective view showing a voxel of another example of the interpolation step.

Fig. 9 is a flow chart showing another modification of the image processing method.

Fig. 10 is a view showing a three-dimensional image showing an example of a candidate region generated by the dividing step in the modification.

Figure 11 (a) is a perspective view of the virtual cylinder used in the image processing method, Figure 11 (b) is a perspective view of the virtual cylinder in the XYZ coordinate space, and Figure 11 (c) shows the virtual cylinder in the polar coordinate space Stereogram.

Figure 12 is a flow chart showing the steps of extracting the image processing method.

Figure 13 is a cross-sectional view showing the cumulative voxel and shape model of the evaluation values in the extraction step.

Fig. 14 (a), (b) and (c) show three-dimensional images of candidate regions observed from different viewpoints and virtual cylinders in the regions.

Fig. 15 is a flow chart showing a modification of the extraction step of the image processing method.

Fig. 16 is a flow chart showing another modification of the extraction step of the image processing method.

Fig. 17 is a flow chart showing another modification of the extraction step of the image processing method.

18(a) is a perspective view showing a pattern of changing the position of the end point of the virtual cylinder, and FIG. 18(b) is a perspective view of the virtual cylinder after changing the end point, in another modification of the extraction step of the image processing method. 18(c) is a plan view showing an evaluation point of the virtual cylinder set in Fig. 18(b).

Fig. 19 is a three-dimensional image showing a state in which an evaluation point is set to the virtual cylinder.

Figure 20 is a flow chart showing the steps of extracting the virtual cylinder and the evaluation point.

Figure 21 is a three dimensional image showing the condition of a composite fibrous material containing a curved fibrous filler.

Figure 22 is a perspective view schematically showing the appearance of adapting a virtual cylindrical model to a curved fibrous filler.

Fig. 23 is a flow chart showing another modification of the extraction step of the image processing method.

Fig. 24(a) is a perspective view showing a virtual tube by a control point and a cloud-shaped curve, and Fig. 24(b) is a perspective view of the cloud-shaped curve in a case where one control point is changed.

Fig. 25(a) is a perspective view showing a disk for setting the evaluation point along a cloud-shaped curve of the virtual tube, and Fig. 25(b) is a plan view showing the evaluation point on the disk.

Fig. 26 is a three-dimensional image showing a state in which an evaluation point is set to a virtual tube.

Fig. 27 is a view showing a three-dimensional image in the middle of the adaptation using the virtual tube.

28(a), (b) and (c) are conceptual diagrams each showing an example of changing programs different from each other of the control points of the virtual tube.

Fig. 29 is a flow chart showing another modification of the image processing method.

30(a) to (h) are conceptual diagrams each illustrating a procedure for adapting the virtual tube and changing each control point.

Figure 31 is a block diagram showing a three-dimensional image processing apparatus for a fibrous filler in a composite material according to an embodiment.

32 is a three-dimensional image showing the image processing method and the fibrous filler extracted by the image processing device in a three-dimensional image.

33 is a view showing an image processing method and a fiber of a fibrous filler extracted by the same image processing device; The frequency distribution of the frequency ratio of the dimension length.

[Preferred form of implementing the invention]

Hereinafter, a three-dimensional image processing method and a three-dimensional image processing apparatus for a fibrous filler in a composite material according to an embodiment of the present invention will be described with reference to the drawings. Fig. 1 is a view showing a three-dimensional image processing method (hereinafter referred to as an image processing method) of a fibrous filler in a composite material according to an embodiment. The image processing method is as shown in FIG. 1 and includes: a region selection step (#1); and an extraction step (#2); and performing the steps of the three-dimensional image of the composite material containing the fibrous filler from the substrate, Remove the alignment information of the filler in the composite. The alignment information is information such as where the filler is placed and in what shape and direction. The region selection step (#1) obtains a three-dimensional image of the composite material, and by referring to a predetermined threshold, a candidate region estimated to contain pixels (stereopixels) representing the filler is selected from the three-dimensional image. The extraction step (#2) is to adapt the shape model to the candidate region selected by the region selection step by using the Monte Carlo method of changing the randomness of the shape and configuration parameters of the shape model of the predetermined filler. And the alignment information of the filler is taken. The alignment information of the filler is extracted from the shape and configuration of the shape model of the adaptation result. The composite material to be treated is, for example, a glass fiber reinforced resin, so that the substrate made of the liquid crystal polymer contains the glass fiber for reinforcement as a fibrous filler. In the case of the composite material, the image processing method extracts the alignment information of the glass fibers in the liquid crystal polymer resin.

Fig. 2 is a view showing a modification of the above image processing method. The image processing method in the embodiment of FIG. 1 includes: a data input step (#0) as a region selection step (#1) pre-step; and an extraction step (#2) includes: a model setting step (S1) ); extracting the main step (S2); and repeating the processing steps (S3, S4).

The data input step (#0) inputs a three-dimensional image of the fibrous filler in the composite material as voxel data. The voxel data is obtained by X-ray computed tomography of the composite material, and each voxel of the voxel data has a value based on the value of the X-ray intensity value. For example, as shown in FIGS. 3(a) and 3(b), the composite material C is a part of the glass fiber reinforced resin molded article 1 molded into a long box shape. In the resin molded article 1, when the resin is injected in the longitudinal direction indicated by the arrow a, the glass fiber is aligned substantially in the direction of the arrow a.

Fig. 4 is a view showing a three-dimensional image G1 of the composite material C shown in Fig. 3(b) by X-ray computed tomography. The three-dimensional image G1 causes X-rays to penetrate the composite material C from various directions, collects an intensity profile formed in accordance with the distribution of the X-ray absorption amount, and reconstructs a plurality of tomographic images of the composite material C from the intensity distributions. Three-dimensional display of black and white shading images. The three-dimensional image G1 containing the three-dimensional image information of the fibrous filler is obtained by the difference in the X-ray absorption amount of the substrate (for example, liquid crystal polymer) and the fibrous filler (for example, glass fiber). The data value of the voxel represents the presence of the filler. The data value of the voxel can be arbitrarily set depending on, for example, the use of the shade or the black and white to represent the fibrous filler, or can be arbitrarily set for the purpose of using the data value. Voxel, the term originally refers to the general three-dimensional pixel that constitutes the three-dimensional image space. In this image processing method, it is sometimes limited to use the stereo pixel selected as the image processing object by using a certain threshold. For example, in the performance of deleting voxels from candidate regions.

The region selection step (#1) selects a candidate region from the three-dimensional image of the composite material C, the candidate region including a region estimated to contain pixels representing the filler. The region selection step (#1) discriminates the voxel data value by the threshold value for the candidate region selection, for example, sets the voxel set having the data value larger than the threshold as the candidate region. At this time, the data value is an X-ray intensity value, and the threshold is set based on the distribution of the X-ray intensity values. The three-dimensional images G2 and G3 of Figs. 5(a) and (b) respectively show examples in which candidate regions are selected by such threshold processing. By the region selection step (#1), the region of the pixel representing the filler is cut out from the background region.

The model setting step (S1) of the extraction step (#2) sets the shape model to the candidate region set in the region selection step (#1) so as to constitute a shape having a parameter that changes shape and configuration and can reproduce the filler. The shape of the shape model is set in accordance with the shape of the filler contained in the composite material. For example, a cylinder, a long ellipse, a corner column, and a plurality of cylinders are defined as a shape model along a curved hollow tube extending along a curve. Each has an orientation with a defined long-side direction.

The shape of the shape model determines the shape and size of the shape model, such as the shape of a cylinder. The radius and length of the shape, the case of a long ellipse is the three-axis length of the three-dimensional ellipse, and the shape of the cross-section is the shape and size of the cross-section, and the length of the corner post, etc., and is set to be changed within a predetermined range. In the case of a curved hollow tube (round tube), the parameters of the shape are set such that the radius, the length of the curve, and the bending of the curve (for example, the approximation of the fold line, the length of each line segment, the direction and size of the mutual corners, etc.) are changed within a predetermined range. Just fine. Moreover, the parameter of the configuration of the shape model determines the position and posture of the shape model whose shape is determined for the candidate region in the space in which the coordinate axis of the three-dimensional space is set. For example, the position coordinates of the both ends of the shape model in the long side direction can be determined. Set as the parameter of the configuration. As long as the configuration parameters can uniquely determine the configuration of the shape model, the center position coordinates and azimuth of the shape model can also be used.

The main step of extracting (S2) is to change the shape model parameter set by the model setting step (S1) to fit the shape model to the candidate region, and to define the shape and configuration information of the adapted shape model as the orientation of the filler. News. The main step of extracting (S2) is a process of randomly changing the parameters of the shape model according to the Monte Carlo method and adapting them. In the Monte Carlo method, for the voxel group with high probability of filler, the parameters are varied within a certain range by using random numbers, and the parameters of the shape model are optimized by evaluating the higher parameters. . In the Monte Carlo method, each time the parameter is changed, the degree of adaptation of the candidate region to the shape model is evaluated by the overall evaluation value obtained by the predetermined method. At the time when the entire evaluation value is updated to the direction of improvement, the shape model at that time is regarded as a filler, and the extraction processing of one filler is converged.

Repeating the processing step (S3), repeating the process of extracting the alignment information of one filler from the candidate region by extracting the main step (S2), and performing extraction from the entire region of the candidate region containing the majority of the filler Processing of alignment information for all fillers. The reverse processing step (S3) removes the final adapted region of the one shape model from the candidate region after extracting one filler, thereby efficiently performing the subsequent next extraction. After the area is removed, if there is still a remaining area of the candidate area (Yes in S4), the processing from the model setting step (S1) is repeated, and if not (No in S4), the image processing is ended.

According to the image processing method of the present embodiment, since each of the fillers is extracted from the three-dimensional image based on the shape model of the filler having the parameters that change the shape and the configuration and the Monte Carlo method, The quantitative information of the packing can be quantitatively extracted. Moreover, in the image processing method, the three-dimensional image of the fibrous filler in the composite material is obtained non-destructively by X-ray computed tomography or the like, so that the alignment information of the filler can be quantitatively extracted in a non-destructive manner.

Fig. 6 shows another modification of the above image processing method. This modification is based on the three-dimensional image processing method of the fibrous filler in the composite material, so that the data input step (#0) has an interpolation step (#0a) in which the data values of the voxels are linearly interpolated between the voxels. This linear interpolation can be performed using, for example, trilinear interpolation.

As shown in FIG. 7(a), the three-dimensional image data includes: a voxel Bx having a data value; and a voxel B0 which does not have an original data value due to influence of measurement conditions and the like. The interpolation step (#0a) is as shown in Fig. 7(b), and the data value is given to the voxel B0 based on the data value of the peripheral voxel Bx. Thereby, the accuracy of the adaptation can be improved. Further, as shown in FIG. 8, the interpolation step (#0a) can constitute a new voxel data having a new voxel bx obtained by linearly interpolating data values between voxels Bx. The value is set to the data value. That is, the data input step (#0) divides the voxel Bx to generate a small voxel bx, and sets the value of the linear interpolation between the voxels Bx as the data value of the voxel bx. The segmentation interpolation of such voxels can improve the accuracy of the adaptation when the size of the voxels is approximately equal to or larger than the size of the filler.

9 and 10 show another modification of the image processing method described above. This modification is, for example, as shown in FIG. 9, and the above-described area selection step (#1) has a division step (#1a) for dividing a large candidate area into smaller candidate areas. In the division step (#1a), groups (groups) composed of voxels adjacent to each other are generated from the voxels belonging to the candidate region, and each group is defined as a new candidate region. The voxels include adjacency, inter-edge abutment, and abutment between the vertices in the case where the two voxels are adjacent to each other. Therefore, for example, a set of voxels adjacent to each other and a set of voxels of a predetermined number or more is identified as a group, and the group is defined as a new individual candidate region. Even if the voxels adjacent to each other do not contain a predetermined number or more of voxels, they are also deleted from the candidate region.

The candidate area originally selected as a set of voxels having a larger data value than the threshold The domain is subdivided into a plurality of candidate regions via this segmentation step (#1a). Further, the following island-shaped voxels are excluded: they are not considered to constitute a filler, and are dispersed to a predetermined number or less. The predetermined number is set based on prior knowledge such as the shape of the filler or based on knowledge obtained by experimentally performing the extraction main step (S2). The three-dimensional image G4 of Fig. 10 shows an example of a candidate region set after the division step (#1a). Further, in the processing of the dividing step, the voxels adjacent to each other, the adjacent to each other, and the voxels adjacent to each other may be included in the group.

Since the image processing method can reduce the range of the Monte Carlo method by performing such a dividing step (#1a), the image processing can be performed efficiently, and the calculation load of the main step (S2) can be reduced. Further, since the method sets the candidate region by the adjacent voxels, the disadvantages are avoided when the grouping by the coordinates is performed: the filler between the groups is cut off, and two fillers are determined. The dividing step (#1a) can be applied to the material that has passed through the above interpolation step (#0a).

11 to 14 show still other modified examples of the above image processing method. This deformation is, for example, as shown in Fig. 11 (a), (b) and (c), using the virtual cylinder M as a shape model for reproducing the shape of the filler. Because the virtual cylinder M is shaped by its radius R and length L, and by the coordinates P1 (x1, y1, z1) and P2 (x2, y2, z2) of the two end points P1 and P2 on the central axis. To determine the spatial configuration in the XYZ orthogonal coordinate space, there are 8 parameters. Since the given length L is the distance between the two end points P1 and P2, the number of parameters can be made seven. When the diameter of the filler is known and can be fixed to a specific value, the radius R can be fixed and set to 6 parameters. The arrangement of the virtual cylinders M is not limited to being represented by the coordinates of the both end points P1 and P2, and the coordinates of the center point Pc may be used, and the polar coordinates (r, θ, Φ) may be used to indicate the tendency of the virtual cylinder.

Thus, in terms of the parameters of the virtual cylinder, the three-dimensional coordinates of the two ends, the three-dimensional coordinates of the center, the direction (azimuth, height), the length, the radius, and the like are included, wherein some of the independent variables are derived, and the rest are derived from other independent variables. The relationship between the independent variable and the dependent variable is complementary, and what is considered as an independent variable can be elastically changed in the soft body. For example, when the position of the endpoint changes, the center position, length, and direction are derived from the position of the endpoint. When the length changes, the end position is derived from the center position and length. When the diameter of the filler is known, as far as the parameters of the cylinder are concerned, only the three-dimensional coordinates (3 × 2 = 6 degrees of freedom) of the two end points can be optimized.

This image processing method is as shown in FIG. 12. In the model setting step (S1) of the extraction step (#2), the virtual cylinder is set as a shape model for reproducing the shape of the filler, and is used in the main step of extraction (S2). A virtual cylinder is used to extract the packing. In the main step of extracting (S2), the shape (radius, length) and the configuration (coordinates at both ends) of the virtual cylinder M are randomly changed within the candidate region, and the virtual cylinder having the changed shape and configuration is evaluated, and the candidate region is evaluated. The degree of adaptation while extracting the filler. The degree of adaptation is as shown in Fig. 13. The data value of the voxel Bx (shown by black dots) contained in the virtual cylinder M is used as an evaluation value within each voxel Bx, and is evaluated by the estimated value of the evaluation. . The evaluation is performed by, for example, using the integrated value, the number N of voxels Bx included in the virtual cylinder M, the value of the integrated value divided by the number N, and the like. For example, when the value of the normalization is fixed and the number N tends to increase, it is determined that the adaptation has not yet converged, and the virtual cylinder M can be further made. The shape and configuration of the virtual cylinder M with the largest cumulative value is extracted as the alignment information of the filler. The comprehensive evaluation value, for example, the three-dimensional image G5 of FIG. 14(a), (b), and (c) shows a three-dimensional image of the candidate region viewed from different perspectives and one virtual cylinder M adapted in the region. .

The filler extraction process using the above-described virtual cylinder M will be described in more detail below. For the three-dimensional image of the filler, for example, the above-described three-dimensional image G4 is considered to contain a plurality of candidate regions, and each group contains images of a plurality of fillers. Therefore, in the model setting step (S1), for each of the groups, a plurality of virtual cylinders M are randomly generated, the fitness is evaluated for each virtual cylinder M, and the virtual cylinder M having the highest evaluation is selected, and the virtual cylinder M is selected. The parameter is used as the initial parameter. Secondly, in the main step of extracting (S2), starting from the initial parameter, the parameter is randomly changed within a predetermined range of variation, and the parameter is updated if the evaluation is improved. That is to say, the Monte Carlo method is used to optimize the basic algorithm, and the parameters are changed within a certain range by using random numbers, and the parameters with higher fitness are evaluated by fitness. This procedure is repeated for the selected virtual cylinder, and if the evaluation is not improved, the parameters are determined to converge, thereby extracting one filler.

Next, in the repeated processing step (S3), voxels relating to the extracted filler, that is, voxels belonging to the virtual cylinder M, are deleted from the candidate region, and the remaining candidate regions are in the same group. Repeat the procedure from the initial parameter to the later program. Whether the voxel belongs to the virtual cylinder M, for example, whether the point of the voxel is preset by the center or the vertex of the voxel It is included in the virtual cylinder M to determine. If the filler cannot be extracted in one group, the extraction of the filler is performed for the other groups. At the point in time when the filling of the packing for all the groups is ended, the extraction step (#2) is ended. Further, when one piece of the filler is extracted and the voxel is deleted from the candidate region, if a predetermined number of island-shaped voxel groups are generated, the island-shaped voxel is deleted at the same time, and the calculation load of the subsequent processing is reduced. Further, when the number of voxels to be processed is within a predetermined number or less in the group, it is determined that the filler cannot be extracted within the group, that is, the extraction is completed within the group.

Fig. 15 shows still more other modifications of the above image processing method. The extraction step (#2) of the image processing method is such that the extraction main step (S2) after the above-described model setting step (S1) includes an evaluation point setting step (S21). The evaluation point setting step (S21) sets a plurality of evaluation points (sampling points) in the virtual cylinder M at intervals below the voxel size, and gives the evaluation values to the evaluation points based on the data values of the voxels.

The evaluation point is, for example, a representative point of a smaller voxel obtained by subdividing the voxel Bx in the virtual cylinder M shown in FIG. 13 as shown in FIG. 8 above, and a representative of the original larger voxel. Just click. The evaluation value of each evaluation point is a data value of a larger voxel containing the original voxel generated by the subdivision. Such evaluation points and evaluation values are generated in the virtual cylinder M. Thereby, compared with the case where the data density is increased in the whole of the voxel data (for example, FIG. 8), there is an advantage that the calculation can be performed locally. The evaluation point is a point formed by the position and data value of the voxel and fixed in the voxel data space. The virtual cylinder M system changes position, expands and contracts within the voxel data space as its parameters change, and is evaluated by the obtained evaluation points.

The main step of extracting (S2) is to estimate the degree of adaptation by using the cumulative value of the evaluation values given to each evaluation point in place of the cumulative value of the data values of the voxels in the main step (S2) of the above-mentioned drawing. By using the evaluation point of the voxel subdivision and the evaluation value thereof, it is possible to obtain an integrated value more accurately with respect to the volume of the voxel contained in the virtual cylinder M, and to evaluate the degree of adaptation more accurately.

Fig. 16 shows still more other modifications of the above image processing method. The extraction step (#2) of this image processing method has an evaluation value interpolation step (S22) after the above-described evaluation point setting step (S21). That is, in terms of the evaluation value given to the evaluation point, the evaluation value interpolation step (S22) will be adopted. The value obtained by linearly interpolating the data value of the original ontology is determined as the evaluation value of the evaluation point. At this time, the evaluation point is interpolated between the representative point of the voxel and the representative point of the voxel, and the evaluation value of the evaluation point using the interpolation is obtained by linear interpolation using the data of the voxel. For linear interpolation methods, trilinear interpolation can be used. For example, when the evaluation point is set at an intermediate position between voxels, the evaluation value of the point can be evaluated by trilinear interpolation of the data value (evaluation value) of the 8 voxels surrounding the evaluation point. By having such an evaluation value interpolation step (S22), the degree of adaptation can be evaluated more accurately.

Fig. 17 shows still more other modifications of the above image processing method. The extraction step (#2) of the image processing method further includes an evaluation value subtraction step (S23), a penalty step (S24), and a weighting step (S25) after the evaluation value interpolation step (S22). The evaluation value subtraction step (S23) is to determine the value obtained by subtracting the predetermined threshold from the evaluation value as the evaluation value. The penalty step (S24) is a case where the evaluation value is negative, and the value obtained by multiplying the value by the predetermined positive number is determined as a new evaluation value. For example, when the evaluation value takes a negative value, the value obtained by multiplying the evaluation value by 10 times is set as the new evaluation value. Thereby, the virtual cylinder can be prevented from leaving the filler region, and the extraction efficiency of the filler can be improved. The weighting step (S25) is weighted in such a manner that the evaluation value of the position closer to the central axis of the virtual cylinder is larger, for example, in a manner proportional to the reciprocal of the distance from the central axis. With this weighting, the process can be accelerated so that the central axis of the virtual cylinder is closer to the portion where the evaluation value is higher.

18, 19, and 20 show still other modified examples of the image processing method. This image processing method uses an evaluation point that changes the position in accordance with the change in position, posture, and shape of the shape model (virtual cylinder M) to replace the evaluation point fixed in the voxel data space in the foregoing method. In the following, as shown in FIG. 18(a), the end point P1 of the virtual cylinder M of the length L0 is moved to the position Px by the variation of the parameter. The evaluation of the adaptation is performed as shown in Fig. 18(b) for the inclined virtual cylinder M having the length L of the end points P1, P2. End point P1 moves from the original position P0 to position Px. As shown in Fig. 18 (b) and (c), the evaluation point b is set between the end points P1 and P2 at a fixed interval Δ1 between the discs B perpendicular to the central axis of the virtual cylinder M (shape model). The arrangement of the discs B is set to be concentric. The evaluation points b on the disk B are arranged at a fixed interval Δ2 in the radius R direction at a fixed interval Δ3 in the circumferential direction. Fig. 19 is a view showing that the virtual cylinder M is finely divided for the height, the radius, and the circumferential direction, respectively, and the evaluation point b is set to each of the division points. kind. The gray portion around the virtual cylinder M in the image G6 represents a voxel having a threshold value or more and forming a candidate region.

The extraction step (#2) of this image processing method will be described with reference to FIG. In the model setting step (S1), the positions of the end points P1, P2 of the virtual cylinder M and the initial values of the radius R are set. The extraction main step (S2) repeats the steps (S101 to S110) in order to extract the filler on a single straight line. The model setting step (S1) and the extraction main step (S2) are collectively referred to as a single straight line step (#21).

At the beginning of the extraction main step (S2), the end point P1 is set to the variable Pi (S101) so that the position of the end point P1 is changed (S102). The discs B are arranged at equal intervals between the end point P2 and the changed end point P1 (S103), and the evaluation point b (sampling point) is concentrically arranged on each of the discs B (S104). Disc B is also disposed at endpoints P1, P2. Since the evaluation point b is set to the virtual cylinder M, it is not fixed with respect to the voxel data space, but is arbitrarily configured. The evaluation value obtained by performing the triple linear interpolation based on the data value of the voxel is given to each evaluation point b (S105). The evaluation values of all the evaluation points b in the virtual cylinder M are accumulated, and the degree of adaptation is evaluated by the accumulated value (S106).

As a result of the evaluation of the variation of the end point P1, if the processing in the Monte Carlo method does not converge (S107, NO), the processing from the step (S102) is repeated. When the processing is converged (S107, YES), the endpoint P2 is set to the variable Pi (S109) via the step (S108), and the step (S102 to S107) is performed for the endpoint P2 as in the case of the endpoint P1. When the convergence of both end points P1 and P2 is confirmed (S108, YES), the process from step (S101) is forcibly repeated at least once in order to perform convergence determination. When the improvement of the evaluation of the fluctuation of the both end points P1 and P2 is judged to have been converged (S110, YES), the main extraction step (S2) is ended. Thereby, the single straight step (#21) is ended. The steps (S3, S4) are the same as described above.

By using the evaluation point b arranged on the disc B as a concentric shape, the information contained in the shape model, that is, the virtual cylinder M can be effectively utilized in the outer boundary portion of the virtual cylinder M without missing. The evaluation of the degree of adaptation in the above step (S106) may employ the method of subtracting the step (S23), the penalty step (S24), and the weighting step (S25) from the evaluation value in Fig. 17 described above.

21 to 28 show still other modified examples of the image processing method. This image processing method employs a shape model that can correspond to a curved filler. As shown in the image G7 of Fig. 21, a curved filler F is sometimes generated in the three-dimensional image of the fibrous filler in the composite material. This is because the strength of the composite material is increased to cause long fiberization of the fibrous filler, and as a result, the filler becomes more easily bent. As shown in the image G8 of Fig. 22, when the curved filler is adapted by the virtual cylinder M, the result of cutting into the short virtual cylinder M is obtained. Therefore, a virtual tube (hollow tube) having a fixed radius of the central axis as a central axis defined by a plurality of control points is used as a shape model that can correspond to the curved filler.

As shown in FIG. 23, the extraction step (#2) of the image processing method includes a single curve step (#22) combining the model setting step (S1) and the extraction main step (S2), and the subsequent step (S3, S4). In the model setting step (S1), n control points {Ci:i=1, . . . , n), C1=P1, and Cn=P2 are arranged at equal intervals between the end points P1 and P2, and a straight line is set. A cloud-shaped curve is used as an initial value. A radius R is attached to the cloud curve, and a virtual tube MP having a fixed radius R centered on the cloud curve is set. The cloud-shaped curve connects the curve between the control points with an arbitrary polynomial, so the curve must pass the curve or straight line of the control point.

In the main extraction step (S2), the full control points Ci are rearranged at equal intervals on the cloud curve (S200), and the variable i=1 is set (S201), and the position of the end point P1=C1 is changed (S202). . Fig. 24(a) shows a virtual pipe MP having a cloud curve Sp and a control point {Ci: i = 1, ..., n}. It also shows that the position of the control point C1 (end point P1) is moved. In addition, FIG. 24(b) shows a pattern in which the control points C1 are moved, and the full control points Ci are rearranged on the cloud curve Sp at equal intervals by the step (S200).

After the position of the control point C1 (the post-generalization control point Ci) performed in the above step (S202) is changed, as shown in FIG. 25(a), the disk B is placed along the cloud-shaped curve Sp, that is, the central axis of the virtual tube. They are arranged at equal intervals perpendicular to the central axis (S203). As shown in FIG. 25(b), the evaluation point b is arranged concentrically on each of the disks B (S204). The setting of the arrangement interval of the disk B or the arrangement of the evaluation points on the disk B can be performed by using the same values as Δ1, Δ2, and Δ3 shown in Fig. 18 (b) and (c) above. Because the evaluation point b is set in the virtual tube MP, it is relative to the voxel data space. Not fixed, but arbitrary configuration.

The evaluation value obtained by the 3-fold linear interpolation based on the data value of the voxel is given to each evaluation point b (S205). The evaluation value of all the evaluation points b in the virtual tube MP is accumulated, and the degree of adaptation is evaluated by the accumulated value (S206). As a result of the evaluation of the fluctuation of the single control point Ci, if the processing in the Monte Carlo method does not converge (S207, NO), the processing from the step (S202) is repeated. When the process is converged (S207, YES), the process proceeds to step (S208), the variable i is set to i=i+1 (S209), and the above-described steps (S202 to S207) are performed in the same manner for the next control point Ci.

When the convergence of all the control points C1 to Cn is confirmed (S108, YES), the process from the step (S201) is forcibly repeated at least once in order to perform the convergence determination. When the evaluation dimension has been converged for the evaluation of the change of the end points P1 and P2 (S210, YES), the main extraction step (S2) is ended. This ends the single curve step (#22). The steps (S3, S4) are the same as those described above. The image G9 of Fig. 26 shows a state in which the evaluation point b is set to the virtual tube MP. The image G10 of Fig. 27 shows the state in the middle of the adaptation using the virtual tube MP.

In the above, the number n of the control points Ci may be made variable. For example, as long as n=2 is initially determined, that is, it is set as the virtual cylinder M, and n is added to the timing of the steps (S200 to S208) of reconfiguring the control point Ci. Further, in the above, the control points {Ci:i=1, . . . , n} are as shown in FIG. 28(a), and the positional change is performed from the control point C1 in accordance with the arrangement order, and is also repeated when the whole is repeated. The position change is also performed from the control point C1. On the other hand, as shown in FIG. 28(b), the positional fluctuation may be performed from the control point C1, and the positional change may be reversed from the control point Cn when the entire repetition is performed. Further, as shown in FIG. 28(c), the positional change of the control point Ci may be alternately performed from both ends of the control point {Ci:i=1, . . . , n}. The order of the positional change of the control point Ci can be arbitrarily selected. For example, in order to make the fluctuation amount larger, the fluctuation is performed first, and the convergence is performed early.

29 and 30 show still other modifications of the image processing method. This image processing method combines the following steps as shown in FIG. 29: the single straight line step (#21) shown in FIG. 20; the single curve step (#22) shown in FIG. 23; and the subsequent new determination step ( #twenty three). That is, this image processing method performs the adaptation of the virtual cylinder M to a single filler (#21), and the adoption will determine The two ends P1, P2 of the virtual cylinder M are defined as the virtual tube MP of the end point for adaptation to a single filler (#22). Thereafter, in the determination step (#23), the evaluation result by the single straight line step (#21) and the evaluation result by the single curve step (#22) are compared (S30). As a result of the comparison, if the predetermined ratio is improved by the evaluation of the virtual pipe MP, for example, 10% or more (S30, YES), the adaptation result using the virtual pipe MP (S31) is adopted, otherwise (S30, No), the virtual cylinder is used. The result of M (S32).

Specific examples of the above processing will be described below with reference to Figs. 30(a) to (h). As shown in Fig. 30 (a), when the adaptation is performed by the linear shape model, that is, the virtual cylinder M, one of the fillers is bent by the four virtual cylinders M. The short virtual cylinder M is extracted by the single straight step (#21) for the full length of the curved filler. As shown in FIG. 30(b), for the extracted virtual cylinder M, the linear cloud curve Sp is set by a single curve step (#22), and the position of the control point C1 at the end point P1 is changed. As shown in FIG. 30(c), the position of the control point C1 is changed every time, that is, the rearrangement of the control point Ci and the evaluation of the adaptation are performed, and the position of the control point C1 is changed. As shown in FIG. 30(d), when the position of the control point C1 reaches the end of the filler, the positional change of the control point C1 converges.

Thereafter, as shown in FIGS. 30(e) to (g), positional variation and evaluation are performed in the order of control points C2, C3, ..., Cn, and when the position of the control point Cn reaches the other end of the filler, The positional change of the control point Cn converges. Thereafter, the positional change and evaluation are performed from the control point C1 to the control point Cn, and the convergence of the extraction of one filler is confirmed based on the degree of change in the evaluation. As a result, as shown in FIG. 30(h), the dummy tube MP for one of the bent fillers is determined.

Fig. 31 is a view showing a three-dimensional image processing apparatus (hereinafter referred to as image processing apparatus 2) for fibrous filler in the composite material of the embodiment. The image processing device 2 includes a region selection unit 11 , a model setting unit 12 , an extraction main body unit 13 , a data input unit 14 , an operation unit 15 , a display unit 16 , and a control unit 10 that is controlled, and contains a fibrous material from a substrate. The alignment information of the filler is extracted from the three-dimensional image of the composite material of the filler. The three-dimensional image is input from the material input unit 14. The control unit 10 is constituted by a computer, and the area selecting unit 11, the model setting unit 12, and the extraction main unit 13 are configured by a program that operates on a computer. The data input unit 14, the operation unit 15, and the display unit 16 It consists of a USB 埠, DVD player, hard disk, keyboard, mouse, pointing device, flat panel display, etc., which are usually provided by computers.

The area selecting unit 11 selects and sets a candidate area including a region estimated to be a region representing the filler from the three-dimensional image of the composite material. The model setting unit 12 sets a shape model having parameters for changing the shape and arrangement to reproduce the shape of the filler, for the candidate regions set by the region selecting unit 11. The extracted body portion 13 adapts the shape model to the candidate region according to the Monte Carlo method by the parameters of the shape model set by the randomness model setting unit 12. The extraction body portion 13 extracts the shape and configuration of the adapted shape model as alignment information for the filler. The control unit 10 repeatedly performs the operations of the model setting unit 12 and the extraction main body unit 13 before the filling of the filler in the three-dimensional image is completed. The area selection unit 11, the model setting unit 12, the extraction main unit 13 and the data input unit 14 execute the area selection step (#1), the model setting step (S1), the extraction main step (S2), and the data input in Fig. 2 described above. Each process of step (#0). Further, the video processing device 2 is not limited to the processing described in the flowcharts of FIGS. 1 and 2, and the processing described in the other flowcharts may be performed.

(Example)

32 and 33 show an embodiment. Fig. 32 is a view showing a three-dimensional image G11 of the fibrous filler extracted by the image processing method and the image processing apparatus for the composite material C shown in Fig. 3(b). The composite material C (test material) is as shown in Fig. 3(a) above, and has an outer shape length of 17.8, a width of 1.83, a height of 0.4 mm, an opening length of 16.5 mm, a bottom portion thickness of 0.12 mm, and a side surface thickness of 0.163 mm. The central portion of the box-shaped slim injection molded article 1 is cut out. The composite material C was a type I liquid crystal polymer in which a glass fiber having an average fiber length of 88 μm was added as a superheat resistance of the filler, and the molding system was a premolded injection molding machine having a maximum clamping force of 196 kN.

The three-dimensional image using the X-ray computed tomography is reconstructed from the tomographic image data of the molded article taken by the micro-focus X-ray computed tomography. The reconstructed image is shown in FIG. 4 as the three-dimensional image G1, and the shape of the L-shaped molded article and the presence of the glass fiber inside can be visually confirmed. The size of the three-dimensional image data is 300×348×200 voxels, and the actual size of one side of one voxel is about 3 μm. In this embodiment, the virtual cylinder is used as the shape model for extraction. The glass fiber is known to have a diameter of 10 μm. So, in terms of virtual cylinder parameters, only the three-dimensional coordinates of the two ends (3 × 2 = 6 degrees of freedom) is optimized by Monte Carlo method. The evaluation of the suitability (degree of fit) is performed by using the evaluation point b (sampling point) shown in Figs. 18 and 19.

Fig. 33 is an example of the fiber length distribution of the filler extracted by the image processing method and the image processing apparatus, and the measured fiber length distribution of the filler is shown as a comparative example. The measured value of the comparative example was obtained by burning the resin portion of the test material and extracting only the glass fiber, and measuring the length thereof. The fiber length distribution of both the examples and the comparative examples showed good agreement. Further, for example, in the case of obtaining a maximum frequency of 60 to 80 μm, it is 22.2% in the embodiment, and in the comparative example, it is 27.0%, and the difference is limited to 4.8%. Moreover, by adopting the embodiment of the image processing method and the image processing apparatus, the alignment information of the filler can be extracted in a non-destructive manner without damaging the test material. Here, only the fiber length distribution of the filler obtained when extracting the alignment information of the filler is shown, but the distribution of the orientation or orientation of the filler in the composite material C (test material) can be directly from the alignment information of each filler. Calculated quantitatively.

Further, the present invention is not limited to the above configuration, and various modifications can be made. For example, the configuration of the above-described respective embodiments and modifications may be combined with each other, and the order of each step may be appropriately changed. For example, it is also possible to have a configuration in which the evaluation point setting step (S21) is after the area selection step (#1) and before the model setting step (S1). In this case, the evaluation point (sampling point) may be set to, for example, each division point in which the voxel is subdivided, and may be fixed to the evaluation point of the three-dimensional image space without depending on the shape model. Further, the evaluation value subtraction step (S23), the penalty step (S24), and the weighting step (S25) can also be applied to the case where the evaluation point is not used, that is, the case where the voxel data value is determined as the evaluation value.

Further, when the resolution of the three-dimensional image is high and the size of the voxel is sufficiently smaller than the size of the shape model (filler), it is not necessary to set the evaluation point (sampling point), but instead it can be configured as a step of integrating the voxels. For example, it is sufficient to set 4 individuals as one large voxel with the average value of their data values as the new data value. The data of the three-dimensional image is not limited to the data using the X-ray computed tomography, and the data of the acquisition mechanism using any three-dimensional image can be used. For example, MRI image data can be used, and in the case of a transparent resin, optical CT image data can be used. Moreover, these data forms are not limited to voxels, and stereoscopic pixels of arbitrary shapes may be employed.

#0‧‧‧Data input steps

#1‧‧‧Regional selection steps

#2‧‧‧ Extraction steps

S1‧‧‧Model setting steps

S2‧‧‧ extraction main steps

S3, S4‧‧‧Repeat processing steps

Claims (17)

A three-dimensional image processing method for a fibrous filler in a composite material, wherein the alignment information of the filler is extracted from a three-dimensional image of a composite material containing a fibrous filler, and the method comprises the following steps: a region selection step, by Selecting a candidate region estimated to be a pixel containing the representative filler from the three-dimensional image of the composite material with reference to a predetermined threshold; and extracting the Monte Carlo having a random change "parameter for changing the shape and configuration of the shape model of the filler" Luofa, the shape model is adapted to the candidate region selected in the selection step of the region to extract the alignment information of the filler. A three-dimensional image processing method for a fibrous filler in a composite material according to claim 1, wherein the extracting step removes the region related to the extraction after extracting the alignment information of the filler from the candidate region, and The candidate region after the removal is repeatedly subjected to extraction using the Monte Carlo method. A three-dimensional image processing method for a fibrous filler in a composite material according to claim 1, comprising: a data input step of inputting the three-dimensional image as voxel data; and the voxel data is by a composite material X Obtained by optical computed tomography, each voxel of the voxel data has a value based on the value of the X-ray intensity value. The method for processing a fibrous filler according to the third aspect of the patent application, wherein the data input step comprises: an interpolation step to form a new voxel data, the new voxel data having the data value The value of linear interpolation between voxels is used as a voxel of the data value. A three-dimensional image processing method for a fibrous filler in a composite material according to claim 3, wherein the region selection step compares the data value of the voxel with a predetermined threshold value according to the X-ray intensity value, and has a ratio The set of voxels of the larger threshold value is defined as the candidate region. A three-dimensional image processing method for a fibrous filler in a composite material according to claim 5, wherein a group formed of voxels adjacent to each other is generated from voxels belonging to the candidate region, and the resulting Each group is designated as a new candidate area. A three-dimensional image processing method for a fibrous filler in a composite material according to claim 3, wherein the extracting step adopts a virtual cylinder as the shape model, and changes the shape and configuration of the virtual cylinder in the candidate region. Parameters, and by containing according to the virtual cylinder The cumulative value of the evaluation value of the voxel data value is used to evaluate the degree of adaptation of the virtual cylinder to the candidate region. A three-dimensional image processing method for a fibrous filler in a composite material according to claim 7, wherein the extracting step adopts a cloud curve (spline) defined by a plurality of control points after performing adaptation using the virtual cylinder a virtual tube defined as a central axis as the shape model, in which the coordinates of the control point are randomly changed as parameters, and the degree of adaptation of the virtual tube to the candidate area is evaluated, and the virtual tube is utilized. The evaluation uses the adaptation of the virtual tube compared to the case where the evaluation of the virtual cylinder is used to improve the predetermined ratio. Otherwise, the adaptation of the virtual cylinder is used to extract the alignment information of the filler. A three-dimensional image processing method for a fibrous filler in a composite material according to claim 3, wherein the extracting step uses a virtual tube in which a cloud curve defined by a plurality of control points is defined as a central axis as the shape model. The coordinates of the control point are randomly changed in the candidate region as a parameter, and the adaptation of the virtual tube to the candidate region is evaluated by the cumulative value of the evaluation value of the data value of the voxel contained in the virtual tube. degree. The three-dimensional image processing method of the fibrous filler in the composite material of claim 8 or 9, wherein the extracting step sequentially changes the coordinates of the random majority of the control points of the cloud curve. Each time the one control point is changed, the cloud curve is determined and the degree of adaptation is evaluated, and all control points are reconfigured to be equally spaced on the determined cloud curve before changing the next control point. A three-dimensional image processing method for a fibrous filler in a composite material according to claim 7 or 9, wherein the extracting step sets an evaluation point within the shape model at intervals below the size of the voxel, and The evaluation point is given based on the evaluation value of the voxel data value, and the degree of adaptation is evaluated by giving the cumulative value of the evaluation value of the evaluation point. A three-dimensional image processing method for a fibrous filler in a composite material according to claim 11, wherein the extracting step concentrically arranges the evaluation point on a plane perpendicular to a central axis of the shape model. A three-dimensional image processing method for a fibrous filler in a composite material according to claim 11, wherein the evaluation value given to the evaluation point is a value obtained by linearly interpolating using the data value of the voxel. A three-dimensional image processing method for a fibrous filler in a composite material according to claim 7 or 9, wherein the extraction step uses a value obtained by subtracting a threshold from the evaluation value as a new evaluation value. A three-dimensional image processing method for a fibrous filler in a composite material according to claim 14, wherein the extraction step is performed by multiplying the value by a predetermined positive value when the new evaluation value is a negative value. New evaluation value. A three-dimensional image processing method for a fibrous filler in a composite material according to claim 7 or 9, wherein the extracting step is performed in such a manner that an evaluation value closer to a position of a central axis of the shape model is larger To calculate the accumulated value. A three-dimensional image processing device for a fibrous filler in a composite material, wherein the alignment information of the filler is extracted from a three-dimensional image of a composite material containing a fibrous filler in a substrate, and is characterized by: a region selection mechanism, which is defined by reference a candidate region selected as a pixel containing the representative filler from the three-dimensional image of the composite material; and an extraction mechanism for Monte Carlo method of changing the parameter of the shape and configuration of the shape model of the filler by randomness And adapting the shape model to the candidate region selected by the region selection mechanism to extract the alignment information of the filler.