JP2023084018A - Tool determination device and tool determination method - Google Patents

Abstract

To provide a tool determination device and a tool determination method which can determine a processing surface of a tool in a short time and with high accuracy.SOLUTION: A tool determination device 20 determines a state of a processing surface 40 of a wire 11 that is a tool of a wire saw 10. A learning tool 51 learnt using teacher data 53 and a determination image 50 that are used in the determination are memorized in a memorizing part 34 of an electronic control device 30. Images constituting the teacher data 53 and the determination image 50 are constituted of a plurality of images extending in a moving direction of the wire 11 cut out from the photographed image of the processing surface 40 of the wire 11, which are arranged and interfaced in a direction crossing the moving direction. The electronic control device 30 outputs an index value of a state of the processing surface 40 corresponding to the determination image 50, from the learnt learning tool 51 memorized in the memorizing part 34, with the determination data 50 set as input data.SELECTED DRAWING: Figure 1

Description

The present invention relates to a tool determination device and a tool determination method for determining the state of a machined surface of a tool based on a captured image of the machined surface of the tool.

A wide variety of tools are used in processing apparatuses, such as rotating grindstones in grinding apparatuses and diamond wires in wire saws. During machining by a machining apparatus, the finish quality of the machined surface of the workpiece depends on the condition of the machined surface of the tool.

Conventionally, there has been proposed a tool determination device that determines the state of a machined surface of a tool (for example, Patent Document 1). In the apparatus described in Patent Document 1, the machined surface of the tool is imaged using a metallurgical microscope equipped with a camera, and the state of the machined surface is determined based on the imaged data.

In recent years, a tool determination device or the like has been proposed that uses a machine learning model to determine the state of a machined surface of a tool. In this apparatus, first, the machined surface of the tool is imaged by a camera, and the imaged image is used as teacher data to learn the learning device. When determining the machined surface of the tool, the machined surface of the tool to be judged is imaged, and output data is acquired from the learned learning device using the imaged image as input data. Based on this output data, the apparatus determines the state of the machined surface of the tool to be determined.

JP-A-2004-45078

Among the tools used in the processing apparatus, the grindstone and the diamond wire have an elongated working surface in the moving direction (specifically, the rotation direction of the grindstone or the axial direction of the diamond wire).

In the above-mentioned tool determination device, in order to determine the whole of such a long machined surface, the machined surface is imaged by a camera for each imaging range divided in the movement direction of the machined surface, and a large number of images are taken. I need to get an image. In this case, since the machined surface of the tool is determined based on a large number of captured images, the determination takes time.

By widening the imaging range of the camera, the number of captured images can be reduced. However, in this case, as the imaging range of the camera widens, the data of the captured image becomes coarser, and it becomes difficult to find "parts with a characteristic state on the processed surface" that appear in the same captured image. There are concerns about a decline in

A tool determination device for solving the above problems is applied to a tool that is used for processing by a processing device and has an elongated shape with a processing surface extending in a moving direction of the processing surface, In the tool determination device for determining the state of the machined surface, an image extending in the movement direction is cut from an image of the machined surface, and a plurality of the cut images are arranged in a direction intersecting the movement direction and connected. a storage unit for storing a learning device learned by teacher data composed of images in a combined state; An input unit that stores a plurality of images arranged and joined in a direction intersecting the moving direction as a judgment image, and from a learned learner stored in the storage unit using the judgment image as input data, a state output unit that outputs an index value of the state of the processing surface corresponding to the judgment image.

A tool determination method for solving the above-described problem is applied to a tool that is used for processing by a processing apparatus and has an elongated shape in which a processing surface extends in a moving direction of the processing surface. A tool determination method for determining the state of a machined surface, wherein an image of the machined surface is acquired, an image extending in the movement direction is cut from the acquired image, and the cut image is crossed with the movement direction. a storage step of generating a plurality of images arranged in a direction and joined together, causing a learning device to learn with teacher data composed of the generated images, and then storing the learning device in a storage unit of the tool determination device; An image obtained by imaging the surface to be processed is acquired, an image extending in the movement direction is cut from the acquired image, and an image is generated by arranging and connecting a plurality of the cut images in a direction intersecting the movement direction, and creating the same. an input step of storing an image as a judgment image in an input unit of the tool judgment device; and using the judgment image as input data, from a learned learning device stored in the storage unit, the machining surface corresponding to the judgment image. and a state output step of outputting an index value of the state of

According to the above-described tool determination device and tool determination method, an image obtained by collecting and connecting parts including "parts in a characteristic state" appearing on the machined surface from images captured by a camera of the machined surface of the tool is used as a teacher. It can be used as data or judgment image. As a result, it is possible to narrow the imaging range of the camera and acquire detailed captured images, and it is also possible to use images that include many of the above-mentioned "parts with characteristic states" as the above-mentioned teacher data and judgment images. Become. Therefore, it is possible to perform the processing of learning the learning device using the teacher data and the processing of determining the state of the machined surface using the learned learning device in a short time and with high accuracy.

According to the present invention, the machined surface of a tool can be determined in a short time with high accuracy.

It is a schematic block diagram of the tool determination apparatus of one Embodiment. Fig. 2 is a plan view showing a processing roller and a wire of a wire saw to which the same tool determination device is applied; FIG. 10 is an explanatory diagram for explaining a procedure for generating teacher data and judgment images; FIG. 4 is an explanatory diagram for explaining a learning mode of a learning device; FIG. 4 is an explanatory diagram for explaining generation errors output from a learning device; FIG. 4 is a plan view of the wire; (a) to (e) are side views showing examples of wire states. 4 is a flowchart showing an execution procedure of operation control processing; (a) to (c) are side views showing an example of how abrasive grains are dispersed on the working surface of the wire. It is a schematic diagram showing a schematic configuration of a wire electric discharge machine to which a tool determination device of another embodiment is applied. It is a schematic diagram showing a schematic configuration of a grinding apparatus to which a tool determination device of another embodiment is applied. FIG. 10 is an explanatory diagram for explaining a procedure for generating teacher data and a determination image in the same embodiment;

An embodiment of a tool determination device and a tool determination method will be described below with reference to FIGS. 1 to 8. FIG.
As shown in FIGS. 1 and 2, the tool determination device 20 of the present embodiment is applied to a diamond wire (hereinafter referred to as wire 11) used for machining by the wire saw 10. As shown in FIG.

<Wire saw 10>
The wire saw 10 as a processing device will be described below.
A wire saw 10 has a pair of reels 12 and 13 around which the wire 11 is wound. These reels 12 and 13 are rotatably supported about axes extending in parallel. The wire saw 10 also has a pair of processing rollers 14 and 15 . The processing rollers 14 and 15 are rotatably supported about axes extending in parallel. A large number of annular grooves 141 and 151 are formed on the outer peripheral surfaces of the processing rollers 14 and 15 at a constant pitch. In the wire saw 10, a single wire 11 is wound in a suspended state between annular grooves 141 and 151 of processing rollers 14 and 15. As shown in FIG.

A rotation drive unit 16 is connected to the reel 12 . The rotation drive unit 16 includes a servomotor 16M that rotates the reel 12, an encoder 16E that detects the rotation phase and rotation speed of the servomotor 16M, and a motor control device 16C that controls the operation of the servomotor 16M. have. A rotation drive unit 17 is connected to the reel 13 . The rotary drive unit 17 includes a servomotor 17M that rotates the reel 13, an encoder 17E that detects the rotation phase and rotation speed of the servomotor 17M, and a motor control device 17C that controls the operation of the servomotor 17M. have.

When the work W1 is machined by the wire saw 10, the reels 12 and 13 are rotationally driven through operation control of the rotational drive units 16 and 17. As shown in FIG. As a result, the wire 11 wound around the reels 12 and 13 is guided onto the processing rollers 14 and 15 and travels around between the processing rollers 14 and 15 .

The wire saw 10 has a support portion 18 that detachably supports the work W1. The support portion 18 is arranged above the wire 11 between the working rollers 14 and 15 . The support portion 18 has a structure capable of moving in a direction toward the wire 11 (downward in FIG. 1) and a direction away from the wire 11 (upward in FIG. 1).

The wire saw 10 has an electronic controller 30 . Rotation drive units 16 and 17 are connected to the electronic control unit 30 . The electronic control unit 30 has a CPU 31, a ROM 32, a RAM 33, and a storage section 34 for storing various programs and data. The storage unit 34 is composed of a non-volatile memory such as a hard disk drive or a solid state drive that can be written and read at any time. The electronic control unit 30 is configured to execute various arithmetic processes based on various programs and data. The electronic control unit 30 controls the operation of the rotary drive units 16 and 17 based on the calculation results.

In the present embodiment, through the operation control of the rotary drive units 16 and 17, the wire saw 10 alternately repeats [forward movement] and [backward movement] described below (so-called traverse operation). there is

[Forward Movement] A wire 11 of a predetermined length LG is let out from the reel 12 and wound on the reel 13 .
[Return] A wire 11 of a predetermined length LB (where LG>LB) is wound on a reel 12 and unwound from a reel 13 .

In the operation control of the rotary drive units 16 and 17, the moving speed of the wire 11 is adjusted to be substantially constant except for the period in which the [forward motion] and the [backward motion] are switched. In this embodiment, an execution program that realizes the operation of the wire saw 10 is determined and stored in advance in the storage unit 34 of the electronic control unit 30 .

Machining of the work W1 by the wire saw 10 is performed as follows. First, the wire 11 is made to travel around between the processing rollers 14 and 15 through operation control of the rotary drive units 16 and 17 . After that, while the work W1 is supported on the lower part of the support part 18, the support part 18 is lowered as indicated by the white arrow in FIG. As a result, the work W1 is pressed against the wire 11, so that the wire 11 cuts the work W1. At this time, the wire 11 performs a traverse operation such that it is reeled out from the reel 12 by a predetermined length LG and then wound on the reel 12 by a predetermined length LB. Through this traverse operation, the cutting of the work W1 by the wire 11 proceeds.

<Tool determination device 20>
Next, the tool determination device 20 of this embodiment will be described.
The tool determination device 20 is a device for determining the state of the outer surface of the wire 11 (hereinafter referred to as the machined surface 40). The tool determination device 20 captures an image of the machined surface 40 of the wire 11 with the camera 21, and uses the captured image (more specifically, a judgment image 50 to be described later) as input data by using a learned learning device 51 to perform machining. Determine the condition of surface 40 .

<Camera 21>
The tool determination device 20 has a camera 21 for imaging the machined surface 40 of the wire 11 . The camera 21 is arranged between the reel 12 and the processing roller 14 so that its imaging portion faces the processing surface 40 of the wire 11 . A camera 21 is connected to the electronic control unit 30 . The electronic control unit 30 controls the operation of the camera 21 . Images captured by the camera 21 are stored in the storage unit 34 of the electronic control device 30 .

<Electronic control unit 30>
The storage unit 34 of the electronic control unit 30 has a storage area for storing various execution programs 52 and a storage area for storing teacher data 53 used for machine learning of the learning device 51 . The storage unit 34 has a storage area for storing a learned learning device 51 (more specifically, learning data created through execution of machine learning). The storage unit 34 also has a storage area as an input unit for storing an image of the machined surface 40 of the wire 11 to be judged (details, a judgment image 50 to be described later), and judges the state of the machined surface 40 of the wire 11. It has a storage area for storing determination result data 54 indicating the result of the determination.

The electronic control unit 30 executes various processes for determining the state of the machined surface 40 of the wire 11 based on various calculation results. In this embodiment, the control section having the CPU 31, the ROM 32, and the RAM 33 corresponds to the state output section.

The execution program 52 includes a program for executing machine learning by the learning device 51 and generating learning data, which is data indicating the execution result of the machine learning. The execution program 52 also includes an image generation program for generating the determination image 50 based on the image captured by the camera 21 . Further, the execution program 52 includes a determination program for executing processing (determination processing) for determining the state of the machined surface 40 of the wire 11 .

The teacher data 53 is data for performing machine learning of the learner 51 so as to acquire the ability to determine the state of the machined surface 40 of the wire 11 .
<Image generation>
In this embodiment, as the images constituting the teacher data 53 and the judgment image 50, the images captured by the camera 21 are not used as they are, but a single image obtained by processing a plurality of captured images is used. A procedure for generating the images constituting the training data 53 and the determination image 50 will be described below.

In this embodiment, under the condition that the wire saw 10 is operated, that is, under the condition that the wire 11 is moving (running), the image of the machined surface 40 of the wire 11 is captured by the camera 21 in a predetermined image capturing pattern. be done. Specifically, images are captured by the camera 21 at predetermined time intervals while the wire saw 10 is in operation. The captured image captured by the camera 21 is stored in the storage section 34 of the electronic control device 30 .

As shown in FIG. 3, when generating the teacher data 53 and the determination image 50, three captured images PA, PB, and PC are selected from captured images stored in the storage unit 34 of the electronic control unit 30. is obtained. Images PPA, PPB, and PPC extending in the axial direction of the wire 11 (horizontal direction in FIG. 3) are cut out from the captured images PA, PB, and PC by removing portions where the wire 11 is not shown. Specifically, the images PPA, PB, and PC are processed in such a manner that the portions of the images PA, PB, and PC where the wire 11 is shown (central portion in the vertical direction in FIG. 3) are left and the aspect ratio is [3:1]. PPB and PPC are clipped. Thereafter, these images PPA, PPB, and PPC are arranged in a direction (vertical direction in FIG. 3) that intersects (perpendicular to in this embodiment) the moving direction, and are joined together to generate a single rectangular image PT. be. In this embodiment, the image PT generated in this way is used as the image forming the teacher data 53 and the determination image 50 .

In the present embodiment, the generation of the images constituting the teacher data 53 is performed as follows. In this case, an unused wire 11 is prepared and attached to the wire saw 10 . Then, the wire saw 10 is operated without machining the work W1. In this state, the image of the machined surface 40 of the wire 11 is captured by the camera 21 and the captured image is stored in the storage unit 34 of the electronic control unit 30 . Three picked-up images PA, PB, and PC are obtained from the picked-up images stored in the storage unit 34 of the electronic control unit 30 in this manner, and a rectangular shape is obtained from the picked-up images PA, PB, and PC. is generated. This image PT is used as an image forming the teacher data 53 . It should be noted that when acquiring the image (the image PT described above) that constitutes the teaching data 53, in addition to using the wire saw 10 and the tool determination device 20 of the present embodiment, it is also possible to use a separate device. .

In this embodiment, when generating the judgment image 50, the generation is executed as follows. In this case, while the wire saw 10 is operated to machine the work W1, the camera 21 captures an image of the machined surface 40 of the wire 11, and the captured image is stored in the storage unit 34 of the electronic control unit 30. stored in Three picked-up images PA, PB, and PC are obtained from the picked-up images stored in the storage unit 34 of the electronic control unit 30 in this manner, and a rectangular shape is obtained from the picked-up images PA, PB, and PC. is generated. This image PT is stored in the storage section 34 of the electronic control device 30 as the determination image 50 . In this embodiment, the process of generating the determination image 50 and storing it in the storage unit 34 of the electronic control unit 30 corresponds to the input process.

<Learning device 51>
As shown in FIG. 1, in this embodiment, as the learner 51, a Generative Adversarial Network (GAN) is employed. In this embodiment, machine learning is performed by the learning device 51 based on the teacher data 53 stored in the storage unit 34 . More specifically, a learning process of extracting one of the plurality of images that make up the teacher data 53 and using the same image as input data to cause the learning device 51 to perform machine learning is repeatedly executed for each image that makes up the teacher data 53. . Learning data created through execution of machine learning is stored in the storage unit 34 . In this embodiment, an execution program 52 is constructed and stored in the storage unit 34 so that a series of processing related to machine learning by the learning device 51 is automatically executed. In this embodiment, the process of executing machine learning by the learner 51 in this way corresponds to the storage process.

As conceptually shown in FIG. 4, in detail, the machine learning of the learning device 51 is performed by combining input data and generated image data generated by the learning device 51 based on the same input data with data representing the same image. (Image data A). A value corresponding to the difference between the input data and the generated image data (hereinafter referred to as generation error ΔP) is output from the learning device 51 that has already been trained. As a result, when the learning device 51 is properly learned, when the determination processing for determining the state of the wire 11 to be determined (more specifically, the machined surface 40 thereof) is executed, the learning device 51 outputs the following: A value (generated error ΔP) is output. In this embodiment, the generated error ΔP corresponds to the index value of the state of the processing surface 40 corresponding to the determination image 50, and the step of outputting the generated error ΔP corresponds to the state output step.

When the processing surface 40 of the wire 11 to be determined is in an unused state, the input data (determination image 50) and the generated image data generated by the learning device 51 are substantially the same. Therefore, in this case, the generated error ΔP output from the learning device 51 becomes a very small value (for example, "0").

On the other hand, when the wire 11 has been used to some extent and the machined surface 40 is in a deteriorated state, the input data (determination image 50) and the generated image data do not match. In this case, as shown in FIG. 5, the generated error ΔP output from the learning device 51 is a "positive value" corresponding to the difference between the input data and the generated image data. Moreover, in this case, the greater the difference between the actual state and the unused state, the greater the difference between the input data and the generated image data. become. Therefore, in the determination process using the learning device 51, it can be determined that the deterioration of the machined surface 40 of the wire 11 progresses as the value of the generation error ΔP increases.

<State of wire 11>
As shown in FIG. 6, the wire 11 has a structure in which a plurality of abrasive grains 41 are fixed by a binder 43 to the outer surface of a core material made of a metal wire. A machined surface 40 of the wire 11 has an elongated shape extending in the moving direction. As the wire 11 is used, the state of the machined surface 40 of the wire 11 changes as follows.

As shown in FIG. 7(a), in the unused wire 11, since the abrasive grains 41 are not abraded, the tips of the abrasive grains 41 are sharp. As shown in FIG. 7(b), when the processing surface 40 deteriorates with use of the wire 11, the tips of the abrasive grains 41 gradually flatten. As shown in FIG. 7(c), as the deterioration of the processing surface 40 of the wire 11 progresses, the protrusion amount of the abrasive grains 41 from the layer of the binder 43 of the wire 11 decreases. As the deterioration of the processed surface 40 of the wire 11 progresses further, the layer of the binder 43 on the wire 11 cracks as shown in FIG. Part of the layer 43 may be peeled off. In addition, as the wire 11 is used, shavings adhere to the surface of the binder 43 or the surface of the binder 43 is scraped, so that the outermost peripheral surface of the processing surface 40 of the wire 11 extends in the moving direction. Some parts may appear. In the present embodiment, the state of the machined surface 40 can be determined by capturing the "characteristic portion" that appears on the machined surface 40 of the wire 11 as the wire 11 is used.

<Statistical processing>
In the present embodiment, when determining the machined surface 40 of the wire 11, an index value for the state of the wire 11 is calculated by performing statistical processing on the "generated error ΔP". In this embodiment, as the index value, the median value of the "generation error ΔP" output from the learning device 51 and stored in the storage unit 34 of the electronic control unit 30 during a predetermined period T is calculated. . Then, the state of the machined surface 40 of the wire 11 is determined based on this index value. As described above, it can be determined that the deterioration of the machined surface 40 progresses as the "generation error ΔP" increases. can be determined to be

As the predetermined period T, one reciprocation period of the wire 11 consists of a [forward movement] period in which the wire 11 is paid out by a predetermined length LG and a [backward movement] period in which the wire 11 is wound by a predetermined length LB immediately after. It is defined. In this embodiment, the "generated error ΔP" stored in the storage unit 34 of the electronic control unit 30 is taken in during the predetermined period T, and statistical processing is performed on the "generated error ΔP" to An index value (the median value above) for the condition is calculated.

By performing such statistical processing, it is possible to grasp the output tendency of the "generation error ΔP" in the predetermined period T, so based on these "generation error ΔP", the state of the machined surface 40 of the wire 11 can be determined over a wide range. can do. As a result, it is possible to prevent the state of the entire machined surface 40 from being erroneously judged based on the judgment result of a part of the machined surface 40, as in the following [erroneous judgment 1] and [erroneous judgment 2]. can. [Incorrect judgment 1] Although the deterioration of the entire machined surface 40 has progressed, it is judged that the deterioration of the machined surface 40 has not progressed in a small part, so the deterioration of the machined surface 40 has not progressed. is determined. [Incorrect judgment 2] Although the deterioration of the entire machined surface 40 has not progressed, it is judged that the deterioration of the machined surface 40 has progressed in only a small part, so the deterioration of the machined surface 40 has progressed. is determined.

<Reflection of judgment results>
In the present embodiment, the determination result (specifically, the median value) of the determination process is reflected in the operation control of the rotation drive units 16 and 17 of the wire saw 10 (FIG. 1). Specifically, when the median value calculated in the determination process is equal to or greater than the determination value, a value (=LG+α) obtained by adding a predetermined value α to the “predetermined length LG” in the [forward movement] is used as a new predetermined value. defined as length LG. On the other hand, when the median value is less than the judgment value, a value (=LG-α) obtained by subtracting the predetermined value α from the "predetermined length LG" is determined as the new predetermined length LG.

Here, when the median value is large, it can be seen that the generation error ΔP tends to increase in the predetermined period T, so it can be determined that the deterioration of the wire 11 has progressed more than expected. In this case, since the work W1 is machined by the wire 11 in a state deteriorated more than expected, the machining accuracy of the work W1 may be lowered.

In this embodiment, in such a case, by increasing the predetermined length LG in [forward movement], the new feed amount of the wire 11, more specifically, from the predetermined length LG in [forward movement] to the predetermined length in [backward movement] The amount obtained by subtracting the length LB (=LG-LB) is increased. Therefore, the update time for updating the portion of the wire 11 that is used for machining the work W1 is shortened. Thereby, the machining accuracy of the workpiece W1 by the wire saw 10 can be maintained at a high level.

On the other hand, when the median value is small, it can be seen that the generation error ΔP tends to be small in the predetermined period T, so it can be determined that the deterioration of the wire 11 has not progressed as expected. In this case, although the machining accuracy of the work W1 can be kept high, there is a possibility that the wire 11 has been updated even though it is still usable. In this embodiment, in such a case, the new feeding amount of the wire 11 is reduced by reducing the predetermined length LG in the [forward movement]. As a result, the update time can be lengthened, so that the usage period of the wire 11 can be extended.

In the present embodiment, from the results of various experiments and simulations by the inventors, etc., the judgment value and the predetermined value α are set in advance so that the life of the wire 11 can be extended while maintaining the machining accuracy of the workpiece W1 at a high level. It has been demanded. These determination values and the predetermined value α are incorporated in the execution program for operating the wire saw 10 and stored in the storage section 34 of the electronic control unit 30 .

Various types of processing (operation control processing) relating to the operation of the wire saw 10 will be specifically described below. FIG. 8 shows the execution procedure of the operation control process. A series of processes shown in the flowchart of FIG. 8 are performed by the electronic control unit 30 as processes at predetermined intervals on the condition that the operation control of the rotation drive units 16 and 17 and the operation control of the camera 21 are being executed. executed.

As shown in FIG. 8, in this process, first, it is determined whether or not image generation conditions are satisfied (step S11). In the process of step S11, it is determined that the image generation condition is met when three captured images are newly acquired through the operation control of the camera 21. FIG.

If it is determined that the image generation condition is not satisfied (step S11: NO), this process is terminated without executing the following processes (processes of steps S12 to S17).

After that, this process is repeatedly executed, and if the image generation condition is satisfied (step S11: YES), the determination image 50 is generated based on the three newly acquired captured images (step S12). Then, using this determination image 50 as input data, the generated error ΔP is output from the learned learning device 51 (step S13), and this generated error ΔP is stored in the storage unit 34 of the electronic control unit 30 ( step S14).

Thereafter, it is determined whether or not the generation error ΔP has been stored for a predetermined period T (step S15). If it is determined that the period during which the generation error ΔP is stored has not reached the predetermined period T (step S15: NO), this process ends without executing the following processes (steps S16 and S17). be done.

After that, this process is repeatedly executed, and when it is determined that the generation error ΔP has been stored for a predetermined period of time T (step S15: YES), the index value (the median value) for the state of the wire 11 is calculated ( step S16). In the process of step S16, by performing statistical processing on the generated errors ΔP stored during the predetermined period T, the median value of the generated errors ΔP is calculated as the index value.

Then, based on the index value (the median value) of the state of the wire 11, the predetermined length LG in the [forward movement] is corrected (step S17). As a result, thereafter, the operation control of the rotary drive units 16 and 17 is executed based on the predetermined length LG corrected in the process of step S17. After the predetermined length LG is corrected in this manner, this process is terminated.

<Effect>
According to this embodiment, the effects described below can be obtained.
(1) The storage unit 34 of the electronic control unit 30 stores the learning device 51 trained by the teacher data 53 and the determination image 50 . The teacher data 53 and the judgment image 50 are obtained by cutting out images extending in the moving direction from the captured image of the machined surface 40 of the wire 11 captured by the camera 21, and arranging a plurality of the cut images in the direction orthogonal to the moving direction. It is an image in a state of being spliced. In the present embodiment, the judgment image 50 is used as input data, and the generated error ΔP, which is the index value of the state of the processing surface 40 corresponding to the judgment image 50 , is output from the learned learning device 51 .

According to the present embodiment, an image that constitutes the teacher data 53 is an image obtained by collecting and connecting the parts including the "part in a characteristic state (see FIG. 7)" appearing on the processing surface 40 from a plurality of captured images. and the determination image 50. As a result, an image that includes many of the above-described “parts in a characteristic state” can be used as the teacher data 53 and the determination image 50 . Moreover, since there is no need to widen the imaging range of the camera 21 in order to obtain an image containing many of the above-described "characteristic portions", it is possible to narrow the imaging range of the camera 21 and obtain a detailed captured image. . Therefore, the processing of learning the learning device 51 using the teacher data 53 and the processing of determining the state of the machined surface 40 using the learned learning device 51 can be executed in a short time and with high accuracy.

(2) An image extending in the axial direction of the wire 11 is cut out from the three captured images of the machined surface 40 of the wire 11 in such a manner as to remove the portion where the wire 11 is not captured. Then, by arranging and connecting the cut images in a direction perpendicular to the moving direction, one rectangular image is generated. The images generated in this manner are used as the images forming the teacher data 53 and the judgment image 50 .

According to this embodiment, it is possible to collect only the portions necessary for determining the state of the processed surface 40, ie, the portions where the processed surface 40 is shown, from a plurality of captured images into one image. In other words, it is possible to remove a portion that is useless for determining the state of the processed surface 40, that is, a portion that does not show the processed surface 40, from the captured image. Therefore, the number of images constituting the teacher data 53 and the number of determination images 50 can be reduced compared to the case where the captured images are used as they are for determination. As a result, the process of learning the learning device 51 using the teacher data 53 and the process of determining the state of the machined surface 40 using the learned learning device 51 can be executed in a short time. Moreover, compared to the case where the captured image is used as it is for the determination, it is possible to increase the above-mentioned "characteristic state portion" included in one image. As a result, it becomes easier to find the above-mentioned "characteristic portion" from the teacher data 53 and the judgment image 50, so that the state of the machined surface 40 can be judged with high accuracy.

<Modification>
It should be noted that the above embodiment can be implemented with the following modifications. The above embodiments and the following modifications can be combined with each other within a technically consistent range.

- The images constituting the teacher data 53 and the determination image 50 may be generated based on two captured images, or generated based on four or more captured images. In short, the images forming the teacher data 53 and the judgment image 50 should be generated as follows. First, a predetermined number (N) of captured images are acquired from captured images stored in the storage unit 34 of the electronic control unit 30 . Then, an image (partial image) is cut out from each captured image in such a manner that the aspect ratio becomes [N:1] while leaving the part where the wire 11 is shown. After that, by arranging and connecting the partial images in a direction orthogonal to the moving direction, one rectangular image is generated. The images generated in this manner are used as the images forming the teacher data 53 and the determination image 50 .

- The predetermined period T is not limited to one reciprocating period of the wire 11, and any period can be set. As the predetermined period T, for example, it is possible to set a multiple reciprocating period in which both the [forward motion] and the [backward motion] of the wire 11 are performed multiple times. In addition, as the predetermined period T, it is possible to set a period during which the wire 11 moves once [forward movement], or set a period during which the wire 11 moves once [backward movement].

- Timing for executing imaging of the processing surface 40 by the camera 21 can be set arbitrarily. For example, imaging by the camera 21 may be performed in a mode in which a “pause period” during which imaging by the camera 21 is suspended and an “execution period” during which imaging by the camera 21 is performed at predetermined time intervals are alternately set. . The imaging by the camera 21 may be performed only during one of the period in which the wire 11 is [forward] and the period in which the wire 11 is [backward]. In addition, imaging by the camera 21 may be performed only during part of the period during which the wire 11 is [forward], or imaging may be performed by the camera 21 only during part of the period during which the wire 11 is [backward]. is also possible. Note that the predetermined period T can be arbitrarily set in accordance with the execution mode of image pickup by the camera 21 .

・As the statistical processing to be performed on the generated error ΔP stored in the predetermined period T, in addition to adopting the processing of calculating the median value, the processing of calculating the average value or the processing of calculating the mode value is adopted. can be adopted. Any process can be employed as the statistical process as long as it is possible to calculate a value that remarkably shows the tendency of the generation error ΔP stored in the predetermined period T.

・An example of the reflection mode of reflecting the determination result (specifically, the median value) of the determination process in the operation control of the rotation drive units 16 and 17 is described in [reflection mode 1] and [reflection mode 2] below. can be changed as desired.

[Reflection mode 1] A new " Predetermined length LG" is calculated.

[Reflection mode 2] When the median value calculated in the determination process is equal to or greater than the determination value, the value obtained by subtracting the predetermined value β from the "predetermined length LB" in the above [Return movement] (= LB - β) is newly added. is defined as a predetermined length LB. On the other hand, when the median value is less than the judgment value, a value obtained by adding a predetermined value β to the "predetermined length LB" (=LB+β) is determined as a new predetermined length LB.

The point is that the [forward movement] is changed according to the determination result of the determination process so that the state of the machined surface 40 of the wire 11 can be extended while maintaining high machining accuracy. It suffices if the predetermined length LG and the predetermined length LB of the [backward movement] can be adjusted.

The process of reflecting the determination result (specifically, the median value) of the determination process in the operation control of the rotary drive units 16 and 17 (the process of step S17 in FIG. 8) may be omitted. With such a configuration as well, the state of the machined surface 40 of the wire 11 can be grasped based on the determination result of the determination process.

- A convolutional neural network (CNN) can be adopted as the learning device 51 . In this case, the machine learning of the learner 51 may be performed so as to enable classification of the state of the machined surface 40 .

As the class classification in this case, it is possible to adopt the class classification of the deterioration state of the machined surface 40 (see FIGS. 7A to 7E). According to this configuration, data indicating the deterioration state of the machined surface 40 can be output from the learned learning device 51 using the determination image 50 as input data.

Further, as the class classification, a class classification regarding the dispersed state of the abrasive grains 41 on the processing surface 40 of the wire 11 can be adopted. Here, the dispersion state of the abrasive grains 41 on the processing surface 40 may be a dense state in which the number of abrasive grains 41 arranged per unit area is relatively large as shown in FIG. As shown in , it becomes an average state in which the number of identical arrangements is an average number. Moreover, as shown in FIG. 9(c), the dispersed state of the abrasive grains 41 on the processing surface 40 may be a rough state in which the number of the abrasive grains 41 arranged per unit area is relatively small. According to the above configuration, the determination image 50 is used as input data, and output data indicating the dispersed state of the abrasive grains 41, examples of which are shown in FIGS. be able to. As a result, for example, in an inspection process in the manufacturing process of the wire 11, the output data can be used to inspect the dispersed state of the abrasive grains 41 on the processing surface 40 of the wire 11. FIG.

・The configuration for generating teacher data and judgment images from captured images of the camera, the configuration for storing the learner learned by the teacher data, and the configuration for outputting the index value of the state of the machined surface from the learned learner, It can also be applied to a wire electric discharge machine.

An example of such a wire electric discharge machine 60 is shown in FIG. As shown in FIG. 10, the wire electric discharge machine 60 has a wire 61, which is a tool used for machining the work W2, and a pair of reels 62, 63 around which the wire 61 is wound. A rotary drive unit (not shown) is connected to each of the reels 62 and 63 . When the work W2 is machined by the wire electric discharge machine 60, the reels 62 and 63 are rotationally driven such that the wire 11 is let out from one reel 62 and wound by the other reel 63. As shown in FIG. The wire electric discharge machine 60 has a power supply device 64 . During machining of the work W2, the positive terminal of the power supply 64 is connected to the wire 61 and the negative terminal of the power supply 64 is connected to the work W2. The wire electric discharge machine 60 has a moving device (not shown) for moving the work W2 relative to the wire 61 when machining the work W2. The wire electric discharge machine 60 has a camera 65 for imaging the machined surface 611 of the wire 61 . The imaging portion of the camera 65 faces the portion of the processing surface 611 of the wire 11 after passing the work W2. Images captured by the camera 65 are stored in a storage unit (not shown) of the electronic control unit.

The tool determination device applied to the wire electric discharge machining device 60 generates the images constituting the teaching data and the determination image as follows. First, an image extending in the axial direction of the wire 61 is cut out from a plurality of captured images of the machined surface 611 of the wire 61 captured by the camera 65 in such a manner as to remove portions where the wire 61 is not captured. Then, by arranging and connecting the cut images in a direction orthogonal to the moving direction of the wire 61, one rectangular image is generated. The images generated in this way are used as the images constituting the teacher data and the judgment images. By generating the images constituting the training data and the judgment image in this way, it is possible to obtain the effects according to the effects described in (2) above.

In the tool determination device applied to the wire electric discharge machine 60, the learner learned by the above teaching data is stored in the storage section of the electronic control device. Then, with the judgment image as input data, the learned learning device outputs an index value (generation error) of the state of the machined surface 611 .

Here, the wire 61 is wire-shaped and has a processed surface 611 on its outer surface. Specifically, a metal wire (for example, brass wire or galvanized brass wire) is used as the wire 61 . As the wire 61 used as a tool of the wire electric discharge machine 60 deteriorates with use, the machined surface 40 becomes uneven. According to the above configuration, it is possible to detect the shape change of the machined surface 611 of the wire 61 and appropriately determine the state of the machined surface 611 . Further, by adjusting the feeding speed of the wire 61 and the moving speed of the work W2 based on the determination result, it is possible to extend the life of the wire 61 while maintaining high machining accuracy of the work W2.

・The configuration for generating teacher data and judgment images from captured images of the camera, the configuration for storing the learner learned by the teacher data, and the configuration for outputting the index value of the state of the machined surface from the learned learner, It can also be applied to a grinding device.

FIG. 11 shows an example of such a grinding device 70. As shown in FIG. As shown in FIG. 11, the grinding device 70 has a substantially disk-shaped rotating grindstone 71 as a tool used for processing. The rotary grindstone 71 has a structure in which a large number of abrasive grains are dispersed on the outer peripheral surface of a base material and fixed with a binder. A processing surface 711 forming the outer peripheral surface of the emery wheel 71 has an elongated shape extending in the moving direction. The emery wheel 71 is rotatably supported. A rotary drive unit 72 having, for example, a servomotor is connected to the rotary grindstone 71 . The grinding device 70 is capable of relatively moving the rotary grindstone 71 and a work (not shown). In the grinding apparatus 70, the surface of the workpiece is ground by the machining surface 711 of the grindstone 71 by relatively moving the grindstone 71 and the work while rotating the grindstone 71 to bring the grindstone 71 into contact with the work. Grinded. The grinding device 70 has a camera 73 for imaging the machining surface 711 of the grindstone 71 . Images captured by the camera 73 are stored in a storage unit (not shown) of the electronic control unit.

The tool determination device applied to the grinding device 70 generates the images constituting the teacher data and the determination image as follows. As conceptually shown in FIG. 12, first, an image PE extending in the moving direction of the processing surface 711 (the vertical direction in FIG. 12) is cut out from the image PD captured by the camera 73 . Then, the image PE is divided in the movement direction into two images PE1 and PE2, and the images PB1 and PB2 are arranged in the direction perpendicular to the movement direction and joined together to form one rectangular image. A PTD is generated. The image PTD generated in this manner is used as an image forming teacher data and a judgment image. Similarly, images PF, PG, and PH extending in the moving direction of the processing surface 711 are cut from the captured image PD, and rectangular images PTF, PTG, and PTH are generated based on the images PF, PG, and PH. These images PTF, PTG, and PTH are also used as images constituting teacher data and judgment images.

In the tool determination device applied to the grinding device 70, the learner learned by the above teaching data is stored in the storage section of the electronic control device. Then, with the determination image as input data, the learned learning device outputs an index value (generation error) of the state of the machined surface 711 .

Here, in the emery wheel 71 in an unused state or in the emery wheel 71 just after being properly dressed, the outermost peripheral surface of the working surface 711 is composed only of the tip surfaces of the abrasive grains. When the grindstone 71 performs grinding to some extent, a portion PX other than the abrasive grains (a so-called bond tail mainly composed of a binder) appears on the outermost peripheral surface of the processing surface 711 . This portion PX tends to extend in the moving direction from the rear portion in the moving direction of the abrasive grains. Furthermore, if the total execution time of the grinding process by the grindstone 71 becomes longer, the state change of the processing surface 711 progresses, so that the portion PX extends in such a manner as to connect the abrasive grains arranged in the movement direction. Thus, the degree of state change of the processing surface 711 of the grindstone 71 appears in the portion extending in the moving direction on the outermost peripheral surface of the processing surface 711 .

According to the above-described configuration, when generating an image constituting teacher data and a judgment image, a captured image obtained by capturing the processing surface 711 of the rotary grindstone 71 has a long shape in the moving direction, that is, a “characteristic shape” appearing on the processing surface 711 . It is possible to cut out an image having a shape that is likely to include "parts in a bad state". In other words, it is possible to cut out an image that tends to include information on the above-mentioned "portion extending in the moving direction on the outermost peripheral surface of the machining surface 711" from the captured image of the machining surface 711 of the grindstone 71. Then, an image obtained by arranging and connecting a plurality of the cut-out images, in other words, an image including a large number of the above-described "characteristic portions" can be used as training data or a determination image. As a result, when determining the processing surface 711 of the emery wheel 71, it is possible to find the above-mentioned "portion in a characteristic state" with a high probability, so that the state of the processing surface 711 can be determined with high accuracy. become.

According to the above configuration, it is possible to use an image obtained by collecting and connecting parts including a "characteristic state" from the captured image captured by the camera 73 as an image constituting teacher data or a determination image. As a result, an image that includes many portions of the "characteristic state" can be used as training data or a determination image. Moreover, since there is no need to widen the imaging range of the camera 73 in order to obtain an image containing many "characteristic state" portions, it is possible to narrow the imaging range of the camera 73 and obtain a finely captured image. Therefore, it is possible to perform the processing of learning the learning device using the teacher data and the processing of determining the state of the machined surface 711 using the learned learning device in a short time and with high accuracy.

DESCRIPTION OF SYMBOLS 10... Wire saw 11, 61... Wire 20... Tool determination apparatus 21, 65, 73... Camera 30... Electronic control unit 31... CPU
32 ROM
33 RAM
34... Storage unit 40, 611, 711... Machining surface 50... Judgment image 51... Learning device 52... Execution program 53... Teacher data 60... Wire electric discharge machine 70... Grinding machine 71... Rotary whetstone

Claims (4)

A tool determination device that is applied to a tool that is used for processing by a processing device and that has a long shape with a processing surface extending in a moving direction of the processing surface to determine the state of the processing surface ,
An image extending in the direction of movement is cut out from an image of the processing surface, and a plurality of the cut out images are arranged in a direction intersecting the direction of movement and joined together. a storage unit that stores learning devices;
An image in which an image extending in the moving direction is cut from an image of the processed surface to be determined and a plurality of the cut images are arranged in a direction intersecting the moving direction and joined together is used as a determination image. an input unit for storing;
a state output unit for outputting an index value of the state of the machined surface corresponding to the judgment image from a learned learning device stored in the storage unit using the judgment image as input data;
A tool determination device comprising: The tool has a wire shape with an outer surface forming the processing surface,
The teacher data and the determination image are obtained by cutting out an image extending in the movement direction in such a manner that a portion where the processing surface is not captured is removed from an image of the processing surface, and the cut image intersects the movement direction. 2. The tool determination device according to claim 1, wherein a plurality of images are arranged in the direction of the tool and are joined together. The tool is a rotary grindstone having a structure in which a plurality of abrasive grains are fixed with a binder,
2. The tool determination apparatus according to claim 1, wherein said teacher data and said determination image comprise images containing information on a portion of said grinding surface extending in said moving direction. A tool determination method that is applied to a tool that is used for processing by a processing device and that has a long shape with a processing surface extending in a moving direction of the processing surface to determine the state of the processing surface. hand,
obtaining an image of the processed surface, cutting out an image extending in the moving direction from the obtained image, and generating an image by arranging and connecting a plurality of the cut images in a direction intersecting the moving direction; a storage step of storing the learning device in a storage unit of the tool determination device after making the learning device learn by the teacher data composed of the images obtained;
An image obtained by imaging the processing surface to be determined is obtained, an image extending in the movement direction is cut from the obtained image, and an image is generated by arranging and connecting a plurality of the cut images in a direction intersecting the movement direction. an input step of storing the created image as a determination image in the input unit of the tool determination device;
a state output step of outputting an index value of the state of the machined surface corresponding to the judgment image from a learned learning device stored in the storage unit using the judgment image as input data;
A tool determination method comprising:

Images