Tool shape measuring device and measuring method in tool presetter

Abstract

To provide a device and method for measuring a tool shape on a tool presetter that can accurately measure a shape of a tool even when deposit adheres to a tool.SOLUTION: A device for measuring a tool shape on a tool presetter includes: a spindle 3 configured to detachably hold a tool while being rotatable; a column 4 configured to be capable of proceeding and retreating along an X-axis direction; a guide member 5 configured to be capable of ascending and descending along a Z-axis direction on the column 4; a camera 8 mounted to the guide member 5 and imaging a tool along a Y-axis direction; deposit removing means for removing deposit adhering to the tool from an image by the camera 8 on the basis of preliminarily learned AI model; image processing means for processing an image so as to visualize deposit of the tool which is removed by the deposit removing means and determining whether deposit of the tool is present or not; and shape measuring means for measuring a shape of the tool on the basis of a camera image or an image of the tool from which deposit is removed by the deposit removing means.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a tool shape measuring device and a measuring method in a tool presetter, and more particularly to a tool shape measuring device and a measuring method in a tool presetter that enable accurate measurement of the shape of a tool even when deposits are attached to the tool.

Background Art

[0002] In a machine tool, dust, grinding fluid, chips, etc. adhere to the tools (cutting tools) used (see, for example, Patent Document 1). Further, in a tool presetter used for measuring in advance the shape and dimensions of the cutting edge of a machine tool tool, when there are deposits on the tool, there is a problem that the shape and dimensions of the cutting edge of the tool cannot be accurately measured due to the influence of the deposits. Therefore, the operator needs to check in advance the presence or absence of deposits and, if there are deposits, remove them before performing the measurement work. Such confirmation work and removal work of deposits hinder the automation of the tool presetting work.

[0003] On the other hand, a master image of the tool (an image in a state without deposits) can be registered, and the presence or absence of deposits on the tool can be determined using pattern matching. However, when automatically measuring the shape and dimensions of the cutting edge of a tool by such a determination method, it is necessary to register the master images of all the tools to be used, and there is a problem that a lot of labor and time are required for the registration work.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] An object of the present invention is to provide a tool shape measuring device and a measuring method in a tool presetter that enables accurate measurement of the shape of a tool even when deposits are attached to the tool.

Means for Solving the Problems

[0006] The tool shape measuring device in the tool presetter of the present invention for achieving the above object includes a spindle configured to be rotatable while detachably holding a tool, a column configured to be movable forward and backward along the X-axis direction perpendicular to the rotation axis of the spindle, a guide member configured to be movable up and down along the Z-axis direction parallel to the rotation axis of the spindle on the column, a camera mounted on the guide member for photographing the tool along the Y-axis direction orthogonal to the X-axis direction and the Z-axis direction, deposits attached to the tool from the image of the camera based on a pre-learned AI model Output an image in a state where [it] has been removed Deposit removing means, and image processing means configured to perform image processing to visualize the deposits of the tool removed by the deposit removing means and to determine the presence or absence of deposits on the tool. When the image processing means determines that there is no deposit on the tool The image of the camera While measuring the shape of the tool based on [it], when the image processing means determines that there is a deposit on the tool It is characterized by comprising shape measuring means for measuring the shape of the tool based on an image in a state where the deposits on the tool have been removed by the deposit removing means.

[0007] Further, the tool shape measuring method in the tool presetter of the present invention includes a spindle configured to be rotatable while detachably holding a tool, a column configured to be movable forward and backward along the X-axis direction perpendicular to the rotation axis of the spindle, a guide member configured to be movable up and down along the Z-axis direction parallel to the rotation axis of the spindle on the column, a camera mounted on the guide member for photographing the tool along the Y-axis direction orthogonal to the X-axis direction and the Z-axis direction, deposits attached to the tool from the image of the camera based on a pre-learned AI model Output an image in a state where [it] has been removedAn adhesion removing means, an image processing means configured to perform image processing to visualize the adhesion of the tool removed by the adhesion removing means and to determine the presence or absence of the adhesion of the tool, When the image processing means determines that there is no deposit on the tool the image of the camera While measuring the shape of the tool based on [it], when the image processing means determines that there is a deposit on the tool using a measuring device comprising a shape measuring means for measuring the shape of the tool based on an image of the state in which the adhesion of the tool has been removed by the adhesion removing means, By the camera photographing the tool along the Y-axis direction, By the deposit removing means removing the adhesion adhering to the tool based on the image of the camera, and based on the image from which the adhesion has been removed By the image processing means determining the presence or absence of the adhesion of the tool, and when there is no adhesion on the tool, based on the image of the camera By the shape measuring means measuring the shape of the tool, and when there is adhesion on the tool, based on the image of the state in which the adhesion of the tool has been removed By the shape measuring means measuring the shape of the tool, which is characterized in that.

Effect of the Invention

[0008] In the present invention, a spindle configured to be rotatable while detachably holding a tool, a column configured to be movable forward and backward along an X-axis direction perpendicular to the rotation axis of the spindle, a guide member configured to be movable up and down along a Z-axis direction parallel to the rotation axis of the spindle on the column, a camera mounted on the guide member for photographing the tool along a Y-axis direction orthogonal to the X-axis direction and the Z-axis direction, an attachment removing means for removing an attachment attached to the tool from an image of the camera based on a pre-learned AI model, an image processing means for performing image processing to visualize the attachment of the tool removed by the attachment removing means and for determining the presence or absence of the attachment of the tool, and a shape measuring means for measuring the shape of the tool based on an image of the camera or an image in a state where the attachment of the tool has been removed by the attachment removing means. Therefore, even when an attachment is attached to the tool, there is no erroneous measurement due to the attachment, and the shape of the tool can be accurately measured. As a result, the confirmation work of the attachment by the operator (manual work) before measurement can be eliminated, and the automation of the preset work can be realized. Further, compared with the conventional method of discriminating attachments by pattern matching, it is not necessary to register master images of all tools, so the labor and time required for the registration work can be eliminated.

[0009] In the tool shape measuring device or measuring method in the tool presetter of the present invention, it is preferable that the AI model is constructed by deep learning based on teacher data in which pixels are pre-classified and labeled for an image of a tool with an attachment attached thereto and an image of a tool without an attachment.

[0010] When an image of the camera is input, the AI model preferably performs a process of classifying pixels into two categories: the tool and the background of the tool. Thereby, the pixels from which the attachment has been removed and the presence or absence of the attachment are clarified, erroneous measurement due to the attachment can be effectively suppressed, and the shape of the tool can be accurately measured.

Brief Description of the Drawings

[0011]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Embodiments for Carrying Out the Invention

[0012] Hereinafter, the configuration of the present invention will be described in detail with reference to the accompanying drawings. FIGS. 1 and 2 show a tool presetter equipped with a tool-shaped measuring device according to an embodiment of the present invention. In FIGS. 1 and 2, the direction perpendicular to the rotation axis E of the spindle is defined as the X-axis direction, the direction parallel to the rotation axis E of the spindle is defined as the Z-axis direction, and the direction orthogonal to the X-axis direction and the Z-axis direction is defined as the Y-axis direction.

[0013] As shown in FIGS. 1 and 2, on the machine base 1, there are installed a spindle 3 which is configured to be rotatable and detachably holds a tool T used in a machine tool, and a column 4 which is configured to be movable forward and backward along the X-axis direction with respect to the spindle 3. The tool T may be directly attached to the spindle 3, or may be attached to the spindle 3 via a taper adapter 2 as shown in FIG. 2. By using the taper adapter 2, it is advantageous because it can quickly respond to the shank sizes of various tools T.

[0014] On the column 4, a guide member 5 is installed so as to be movable up and down along the Z-axis direction, and a pair of support members 6 and 7 extending in the horizontal direction are arranged on the guide member 5. A camera 8 (first camera) for photographing the tool T from the front is mounted on the support member 6. This camera 8 is configured to photograph the tool T in the Y-axis direction and measure the height and outer diameter of the tool T. On the other hand, a rotary drive device 9 having a rotation axis parallel to the Y-axis direction and a swing arm 10 that is rotationally driven by the rotary drive device 9 are arranged on the support member 7, and a camera 11 (second camera) for photographing the tool T from the side or above is mounted on the swing arm 10. This camera 11 is configured to be able to photograph the tool T from the X-axis direction or the Z-axis direction. In the embodiment of FIG. 1, the camera 11 is normally arranged in the X-axis direction, but when detecting the missing state of the tool T, the swing arm 10 swings around the axis and moves directly above the tool T in the Z-axis direction to perform photographing.

[0015] In the above-described tool presetter, an operation box 12 for operating the rotation of the spindle 3 and the movement of the column 4 or the guide member 5 is provided on the machine base 1. Further, on the mounting table 16 provided on the side surface of the machine base 1, a touch panel monitor 13 for displaying an image taken by the camera 8 or the camera 11, an operation screen, etc., a keyboard 14 for performing an input operation, and a printing machine 15 for printing measurement results and the like are installed. In FIGS. 1 and 2, an example in which all of the touch panel monitor 13, the keyboard 14, the printing machine 15, and the mounting table 16 are installed is shown, but these can be installed arbitrarily as needed.

[0016] A computer 20 is built into the machine base 1. As shown in FIG. 3, the computer 20 includes a control device 21, a storage device 22, an attachment removing means 23, an image processing means 24, and a shape measuring means 25. The control device 21 controls the moving amount of the column 4 in the X-axis direction, the moving amount of the guide member 5 in the Z-axis direction, the rotation amount of the spindle 3, the rotation amount of the rotation driving device 9, and the cameras 8 and 11 based on an instruction given by the operation box 12, the touch panel monitor 13, or the keyboard 14. The storage device 22 stores the measurement data and image data of the tool T. The attachment removing means 23 removes the attachment adhering to the tool T from the image of the camera 8 based on a pre-learned AI model described later. The image processing means 24 makes the pixels of the attachment removed by changing the color of the attachment of the tool T on the image of the camera 8 (for example, in red) to indicate the removed attachment and is configured to determine the presence or absence of the attachment. The shape measuring means 25 measures the shape of the tool T based on the image of the camera 8 (when there is no attachment on the tool T) or the image in a state where the attachment of the tool T has been removed by the attachment removing means 23 (when there is an attachment on the tool T). Such attachment removing means 23, image processing means 24, and shape measuring means 25 can be configured, for example, as a program capable of executing the above-described processing.

[0017] The deposit removal means 23 of the computer 20 is responsible for a pre-trained AI model. When an image of the tool T taken by the camera 8 is input into this AI model, regardless of the presence or absence of deposits on the tool T, it performs a process of classifying each pixel into a category (for example, the tool T and its background). Specifically, the AI model outputs a probability map indicating to which category, either the tool T or its background, each pixel of the image of the tool T belongs, and further converts this into a binary or 8-bit grayscale-coded image and outputs it. Note that the probability map represents, in pixel units, the probability of belonging to any of the predetermined categories.

[0018] Such an AI model is constructed by deep learning (deep neural network) based on teacher data using images of the tool T with deposits attached and images of the tool T without deposits. Specifically, the teacher data is data that has been pre-classified and labeled for each pixel into the categories of the tool T and its background for each of an image of the tool T without deposits as shown in Fig. 4(a) and an image of the tool T with at least one deposit g attached as shown in Fig. 4(b) for the same-shaped tool T. A large number of such teacher data can be prepared to construct the AI model. These multiple teacher data may be images of the same type of tool T with different deposit attachment states, or may include images of multiple types of tool T. Note that the AI model constructed by deep learning based on the above-described teacher data is evaluated using an image of the tool T for evaluation (an image of a tool T different from the teacher data), and if the evaluation result is good, it is installed in the computer 20.

[0019] The AI model can be constructed, for example, by semantic segmentation using deep learning with a convolutional neural network. This convolutional neural network is suitable for learning targeted at image data. Note that deep learning is one of the machine learning methods represented by artificial intelligence (AI), and it uses a multi-layer structured neural network (deep neural network) modeled after the human brain's nerve circuits, where the artificial intelligence thinks and determines the setting and combination of feature quantities. That is, it is an evolution of conventional machine learning. In contrast to the need to specify the features (feature quantities) worthy of attention in machine learning, which discovers regularities and correlations from a large amount of data to make judgments and predictions, deep learning automatically learns without specifying features (feature quantities).

[0020] As semantic segmentation methods, examples include FCN (Fully Convolution Network), Seg-Net, U-Net, etc. When this semantic segmentation is used as the attachment removal means 23 (AI model), when an image of the tool T taken by the camera 8 is input, the feature amounts of categories (for example, the tool T and its background) are output for each pixel. By calculating this with a sigmoid function or the like, a probability map can be output. Semantic segmentation is composed of only an input layer, a convolutional layer, and a pooling layer. However, it is not limited to this structure and can be changed as appropriate. By stacking multiple convolutional layers and pooling layers, strong feature parts in the input image of the camera 8 can be identified, and even if the image of the tool T is complex, it can be accurately identified. In CNN (Convolution Neural Network), which is a typical algorithm for image recognition, the final output is obtained through a fully connected layer. However, in semantic segmentation, since the fully connected layer is replaced with a convolutional layer, not what the object is, but the contour and position of the object are output. For example, in U-Net, which is a typical method among semantic segmentation methods, the input image is repeatedly subjected to a convolution operation and pooling with a 1×1 kernel. It is advisable to use a ReLU function or the like as the activation function in the intermediate layer at this time. Next, the same number of Up-sampling (transpose convolution) is performed to return to the original number of pixels. At that time, after performing the convolution, the feature map is retained and added to the image for which Up-sampling is to be performed later, thereby compensating for the lost position information in the pooling. By this process, more accurate information on the position and contour of the object can be extracted.

[0021] Note that the AI model installed in the tool presetter of the present invention is not limited to the convolutional neural network described above, and those constructed by other deep learning other than the convolutional neural network or other machine learning based on teacher data may be used.

[0022] FIG. 5 shows the procedure of the tool shape measurement method according to the embodiment of the present invention. As shown in FIG. 5, the shape of the tool T is measured using a tool presetter equipped with the above-described tool shape measuring device. In step S1, after the tool T is attached to the taper adapter 2 attached to the spindle 3, the camera 8 takes a picture of the tool T in the Y-axis direction.

[0023] Next, proceed to step S2, and the deposit removing means 23 (AI model) removes the deposits attached to the tool T from the image taken by the camera 8. Specifically, regardless of the presence or absence of deposits on the tool T, the AI model outputs a probability map indicating to which category each pixel belongs, either the tool T or its background, and further converts this into a binary or 8-bit grayscale-encoded image and outputs it. The output image data (image data in a state where the deposits on the tool T are removed) is stored in the storage device 22. Note that the tool T shown in FIG. 6(c) shows a state where the deposits have been removed by the deposit removing means 23.

[0024] Also, in step S2, when removing the deposits attached to the tool T based on the image taken by the camera 8, a plurality of AI models can be combined in series for removal. For example, the type of the tool T is identified using an AI model such as a CNN effective for identifying the type of the tool T, and further, for the shape of the tool T, the deposits on the tool T are removed by an AI model using semantic segmentation, so that the removal accuracy of the deposits by the deposit removing means 23 can be improved for various tools T.

[0025] Next, proceed to step S3. The image processing means 24 compares the image from which deposits have been removed by the deposit removing means 23 (deposit-removed image) with the image captured by the camera 8 (original image) to determine the presence or absence of deposits on the tool T. Specifically, the image processing means 24 reads out the deposit-removed image, assigns a threshold value to each pixel of the deposit-removed image, and based on this threshold value, determines whether the pixel corresponding to the deposit-removed image in the original image exceeds the threshold value. For example, if the original image has at least one pixel exceeding the threshold value, the deposit-removed image is determined to have "deposits", while if the original image has no pixels exceeding the threshold value, the deposit-removed image is determined to have "no deposits".

[0026] Furthermore, in step S3, when the image processing means 24 determines the presence or absence of deposits on the tool T, it performs image processing to visualize the deposits removed from the tool T by the deposit removing means 23. Specifically, if the image processing means 24 determines that there are deposits on the tool T, assuming that there is actually a deposit g attached to the tool T as shown in Fig. 6(a), then as shown in Fig. 6(b), the characteristic part detected by the image processing means 24 coincides with the location where the deposit g actually exists, and that location can be displayed. The characteristic part detected by this image processing means 24 can be configured to be viewable on the touch panel monitor 13. Also, the display method by the image processing means 24 is not particularly limited. For example, the location where the deposit g does not exist can be shown transparently, and the location where the deposit g exists can be shown in black and gray. Alternatively, it can be displayed as a color image. In that case, the location where the deposit g does not exist can be shown transparently, and the location where the deposit g exists can be shown in color. In particular, it is preferable to use a heat map that shows the presence of the deposit g by varying the type and shade of color at the location where the deposit g exists.

[0027] In step S4, the shape measurement means 25 measures the shape of the tool T. When the image processing means 24 determines in step S3 that there is no deposit on the tool T, the shape measurement means 25 measures the shape of the tool T based on the image (the original image shown in the figure) captured by the camera 8 in step S1. On the other hand, when the image processing means 24 determines in step S3 that there is a deposit on the tool T, the shape measurement means 25 reads out the image data (the deposit removal image shown in the figure) stored in the storage device 22 in step S2 from the storage device 22, and measures the shape of the tool T based on the image data.

[0028] Furthermore, in step S4, when measuring the shape and surface state of the cutting edge of the tool T, the camera 11 (second camera) is used. The tool T can be photographed from directly above in the Z-axis direction by the camera 11, and the defective state of the tool T can be detected.

[0029] In a tool presetter including the above-described tool-shaped measuring device, there are a spindle 3 configured to be rotatable while detachably holding a tool T, a column 4 configured to be movable back and forth along the X-axis direction, a guide member 5 configured to be movable up and down along the Z-axis direction on the column 4, a camera 8 mounted on the guide member 5 for photographing the tool T along the Y-axis direction, an attachment removing means 23 for removing attachments adhering to the tool T from the image of the camera 8 based on a pre-learned AI model, an image processing means 24 configured to perform image processing so as to visualize the attachments of the tool T removed by the attachment removing means 23 and to determine the presence or absence of the attachments of the tool T, and a shape measuring means 25 for measuring the shape of the tool T based on the image of the camera or the image in a state where the attachments of the tool T have been removed by the attachment removing means 23. Therefore, even when attachments are adhering to the tool T, there is no erroneous measurement due to the attachments, and the shape of the tool T can be accurately measured. As a result, the work of the operator (manual labor) to check for attachments before measurement can be eliminated, and thus automation of the preset work can be realized. In particular, by using the AI model to remove the attachments of the tool T, attachments that cannot be identified by the human eye can also be removed, contributing to the realization of full automation of the preset work. Also, compared with the conventional method of discriminating attachments by pattern matching, it is not necessary to register the master images of all the tools T, so the labor and time required for the registration work can be eliminated.

[0030] Furthermore, in the present invention, the attachments of the tool T are removed using the attachment removing means 23 (AI model), and image processing is performed so as to visualize the attachments of the tool T removed by the attachment removing means 23. Since the attachments are displayed by changing the color on the image, the image quality of the original image (the image photographed by the camera 8) is not deteriorated. On the other hand, as image processing for removing noise on the image, for example, spatial filtering, dilation / erosion method, etc. can be mentioned. However, in these cases, since the original image itself is processed in the process of image processing, the image quality of the original image deteriorates, so it is not suitable for precise measurement of the tool shape.

[0031] In the above tool presetter, when an image from the camera 8 is input, the AI model preferably performs a process of classifying each pixel into two categories: the tool T and the background of the tool T. By using such an AI model, the pixels from which the adherents have been removed and the presence or absence of adherents become clear, mismeasurement due to adherents can be effectively suppressed, and the shape of the tool T can be accurately measured.

Explanation of Signs

[0032] 1 Machine base 2 Taper adapter 3 Spindle 4 Column 5 Guide member 6,7 Support member 8 Camera (First camera) 9 Rotation drive device 10 Swivel arm 11 Camera (Second camera) 20 Computer 21 Control device 22 Storage device 23 Adherent removal means 24 Image processing means 25 Shape measurement means

Claims

1. A spindle configured to be rotatable while detachably holding a tool, a column configured to be movable forward and backward along the X-axis direction perpendicular to the rotation axis of the spindle, and a guide member configured to be movable up and down along the Z-axis direction parallel to the rotation axis of the spindle on the column, a camera mounted on the guide member for photographing the tool along the Y-axis direction orthogonal to the X-axis direction and the Z-axis direction, an attachment removing means for outputting an image in a state where an attachment attached to the tool is removed from the image of the camera based on a pre-learned AI model, an image processing means configured to perform image processing so as to visualize the attachment of the tool removed by the attachment removing means and to determine the presence or absence of the attachment of the tool, and a shape measuring means for measuring the shape of the tool based on the image of the camera when the image processing means determines that there is no attachment on the tool, and for measuring the shape of the tool based on the image in a state where the attachment of the tool has been removed by the attachment removing means when the image processing means determines that there is an attachment on the tool. A tool shape measuring device in a tool presetter, characterized by comprising the above components.

2. The tool shape measuring device in the tool presetter according to claim 1, characterized in that the AI model is constructed by deep learning based on teacher data in which categories are pre-classified and labeled for each pixel with respect to an image of the tool in a state where an attachment is attached and an image of the tool in a state where there is no attachment.

3. The tool shape measuring device in the tool presetter according to claim 1 or 2, characterized in that the AI model performs a process of classifying each pixel into two categories of the tool and the background of the tool when the image of the camera is input.

4. A spindle configured to be rotatable while detachably holding a tool, a column configured to be movable forward and backward along an X-axis direction perpendicular to the rotation axis of the spindle, a guide member configured to be movable up and down along a Z-axis direction parallel to the rotation axis of the spindle on the column, a camera mounted on the guide member for photographing the tool along a Y-axis direction orthogonal to the X-axis direction and the Z-axis direction, an attachment removing means for outputting an image in a state where an attachment attached to the tool has been removed from an image of the camera based on a pre-learned AI model, an image processing means for performing image processing to visualize the attachment of the tool removed by the attachment removing means and for determining the presence or absence of the attachment of the tool, and a shape measuring means for measuring the shape of the tool based on the image of the camera when the image processing means determines that there is no attachment on the tool, and for measuring the shape of the tool based on an image in a state where the attachment of the tool has been removed by the attachment removing means when the image processing means determines that there is an attachment on the tool. A method for measuring the shape of a tool in a tool presetter, comprising photographing the tool along the Y-axis direction with the camera, removing an attachment attached to the tool based on the image of the camera by the attachment removing means, determining the presence or absence of the attachment of the tool by the image processing means based on the image with the attachment removed, measuring the shape of the tool by the shape measuring means based on the image of the camera when there is no attachment on the tool, and measuring the shape of the tool by the shape measuring means based on an image in a state where the attachment of the tool has been removed when there is an attachment on the tool.

5. The method for measuring the shape of a tool in a tool presetter according to claim 4, wherein the AI model is constructed by deep learning based on teacher data in which categories are classified and labeled in advance for each pixel with respect to an image of the tool in a state where an attachment is attached and an image of the tool in a state where there is no attachment.

6. The method for measuring the shape of a tool in a tool presetter according to claim 4 or 5, wherein the AI model performs a process of classifying into two categories, namely the tool and the background of the tool, for each pixel when an image of the camera is input.

Images