JP2020157397A - Breakage prediction method of tool - Google Patents

Abstract

To provide a breakage prediction method of a tool which can improve breakage prediction accuracy of the tool.SOLUTION: A breakage prediction method of a tool includes the steps of: generating a model waveform by preliminarily mechanically learning a plurality of first waveforms indicating each normal relation between cutting time of a first workpiece finished in cutting and a cutting load; comparing a partial waveform appearing in the latest period in a second waveform indicating a relation until actual measurement of cutting time of a second workpiece to be cut after the first workpiece and a cutting load with a part of the model waveform corresponding to the partial waveform; and detecting a sign of the breakage of the tool for cutting the second workpiece when a cutting load actually measured next to the actual measurement time appears at a lower probability than a prescribed threshold which can specify a normal cutting load as a result of the comparison.SELECTED DRAWING: Figure 4

Description

The present invention relates to a method for predicting damage to a tool.

A technique for predicting drill breakage has been proposed. In particular, this technique determines that there is a risk of cutting of the drill when the spindle torque change function in the actual machining state fluctuates by a predetermined value or more with respect to the torque change model function in the reference state (for example, Patent Document). 1).

Japanese Unexamined Patent Publication No. 10-296589

However, when predicting breakage or breakage of a tool such as a drill (hereinafter referred to as breakage) using a predetermined value, it may not be possible to accurately predict the breakage of the tool depending on the predetermined value to be adopted.

Therefore, an object of the present invention is to improve the accuracy of predicting breakage of a tool.

The tool breakage prediction method according to the present invention generates a model waveform by preliminarily machine-learning a plurality of first waveforms representing the normal relationship between the cutting time of the first workpiece that has been cut and the cutting load. Then, the partial waveform appearing in the latest period in the second waveform representing the relationship between the cutting time of the second workpiece to be cut after the first workpiece and the cutting load up to the actual measurement corresponds to the partial waveform. As a result of comparing and comparing with a part of the model waveform, when the cutting load actually measured next to the actual measurement appears with a probability lower than a predetermined threshold value capable of specifying a normal cutting load, the second work is used. Detects signs of damage to cutting tools.

According to the present invention, it is possible to improve the accuracy of predicting breakage of a tool.

FIG. 1 is an example of a side view of a cutting machine to which a drill is attached. FIG. 2 is an example of the hardware configuration of the control system. FIG. 3 is an example of a block diagram of the terminal device. FIG. 4A is an example of a flowchart of the learning process. FIG. 4B is an example of a flowchart of the evaluation process. FIG. 5 is an example of the torque waveform of the spindle in the normal state. FIG. 6 is a diagram for explaining an example of the model waveform and the model waveform. FIG. 7 is a diagram for explaining a comparison between the torque waveform to be evaluated and the model waveform. FIG. 8 is an example of a probability graph.

Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings. In the present invention, a drill will be used as an example of a tool (specifically, a rotary cutting tool), but the tool is not limited to the drill, and may be, for example, a tap, an end mill, a milling cutter, a cutting tool, or the like. ..

FIG. 1 is an example of a side view of the cutting machine 100 to which the drill DL is attached. The column 11 is installed on the bed 10 which is the pedestal of the cutting machine 100. The column 11 can be moved along the X-axis guideway 12 provided on the bed 10 in the X-axis direction (not shown), which is the vertical direction of the paper surface. In other words, the column 11 can be moved toward the front side of the paper or the back side of the paper. Hereinafter, this movement is referred to as left-right movement.

A Y-axis guideway 13 extending in the vertical direction is provided on the side surface of the column 11 that moves left and right. A saddle 14 is suspended on the Y-axis guideway 13. The Y-axis guideway 13 guides the vertical movement of the saddle 14. As a result, the saddle 14 can be moved up and down along the Y-axis guideway 13 in the direction away from the bed 10 and in the direction approaching the bed 10. That is, the saddle 14 can move up and down in the Y-axis direction of the cutting machine 100.

A Z-axis guideway 15 extending in the horizontal direction is provided on the side surface of the saddle 14. A spindle head 16 is suspended on the Z-axis guideway 15. The spindle head 16 is connected to a support member 16A that is longer in the vertical direction than in the horizontal direction. The Z-axis guideway 15 guides the front-back movement (left-right direction in FIG. 1) of the spindle head 16. As a result, the spindle head 16 can move forward along the Z-axis guideway 15 in a direction approaching the work WK, or move backward in a direction away from the work WK. That is, the spindle head 16 can move forward and backward in the Z-axis direction of the cutting machine 100. The spindle head 16 rotatably supports the spindle 17. A drill DL is attached to the tip of the spindle 17. Instead of the drill DL, taps, end mills, milling cutters, cutting tools, etc. can be attached.

Here, the servomotor 21 is fixed to the bed 10. On the other hand, the column 11 is screw-engaged with the bed 10 by a feed screw (not shown) extending in the X-axis direction. Therefore, when the servomotor 21 is driven, the feed screw rotates and the column 11 moves in the X-axis direction. Similarly, the servomotor 22 is fixed to the upper surface of the column 11. On the other hand, the saddle 14 is screw-engaged with the column 11 by a feed screw 23 extending in the Y-axis direction. Therefore, when the servomotor 22 is driven, the feed screw 23 rotates and the saddle 14 moves in the Y-axis direction.

A servomotor 24 is fixed to the saddle 14. On the other hand, the support member 16A is screw-engaged with the saddle 14 by a feed screw 25 extending in the Z-axis direction. Therefore, when the servomotor 24 is driven, the feed screw 25 rotates and the support member 16A moves in the Z-axis direction. Since the support member 16A is connected to the spindle head 16, the spindle head 16 moves in the Z-axis direction in accordance with the movement of the support member 16A. The servomotor 26 is fixed to the spindle head 16 and rotationally drives the spindle 17. As a result, the drill DL attached to the tip of the spindle 17 rotates.

A turntable 28 is provided in front of the column 11 on the bed 10. The work WK to be cut is placed on the turntable 28. The drill hole W1 can be formed in the work WK by rotating the drill DL and moving the tip of the drill DL in contact with the work WK. As the work WK, for example, various vehicle parts can be adopted, but the type of the work WK is not particularly limited.

Next, the control of the cutting machine 100 will be described with reference to FIG.

FIG. 2 is an example of the hardware configuration of the control system ST. The control system ST includes an NC (Numerical Control) device 30 and a terminal device 40. The terminal device 40 may be a PC (Personal Computer) or a smart device including a tablet terminal. The NC device 30 is connected to the operation panel 33, the drive device 37, and the terminal device 40. The NC device 30 supplies a drive current to the drive device 37 based on commands and data input from the operation panel 33, and controls the operation of the drive device 37. As a result, a plurality of servomotors 21, 22, 24, 26 connected to the drive device 37 are driven.

The operation panel 33 includes switches 331, a numeric keypad 332, and a display unit 333. The switches 331 are various switches for inputting various commands to the NC device 30. The numeric keypad 332 is a key for inputting data to the NC device 30. The display unit 333 displays various information.

The drive device 37 includes an X-axis drive unit DU-X, a Y-axis drive unit DU-Y, a Z-axis drive unit DU-Z, and a C-axis drive unit DU-C. For example, the C-axis drive unit DU-C drives the corresponding servomotor 26 based on the drive current. That is, the C axis corresponds to the rotation axis of the main shaft 17. The X-axis drive unit DU-X, the Y-axis drive unit DU-Y, and the Z-axis drive unit DU-Z are basically the same as the C-axis drive unit DU-C, and thus description thereof will be omitted. ..

The terminal device 40 includes a CPU (Central Processing Unit) 40A, a RAM (Random Access Memory) 40B, a ROM (Read Only Memory) 40C, a display 40D, and a plurality of communication interfaces (denoted as I / F in FIG. 2) 40E and 40F. Includes. The CPU 40A, RAM 40B, ROM 40C, display 40D, and a plurality of communication interfaces 40E and 40F are connected to each other by an internal bus 40G. An auxiliary storage device such as an HDD (Hard Disk Drive) or SSD (Solid State Drive) may be connected to the internal bus 40G.

The communication interface 40E is provided with a connection terminal such as a USB (Universal Serial Bus) port, and is connected to the NC device 30 by a communication cable 50. On the other hand, the communication interface 40F includes, for example, a LAN (Local Area Network) port, and is connected to a communication network NW such as a LAN by a LAN cable 60. In addition, instead of the wired communication using the LAN cable 60, wireless communication such as Wi-Fi (registered trademark) may be used.

In the RAM 40B described above, the program stored in the ROM 40C is temporarily stored by the CPU 40A. When the CPU 40A executes the stored program, the CPU 40A realizes various functions and also executes various processes. The program may be adapted to the flowchart described later. Further, instead of the ROM 40C, the CPU 40A may realize various functions and execute various processes by executing the program stored in the auxiliary storage device described above by the CPU 40A.

Next, the functional configuration of the terminal device 40 will be described with reference to FIG.

FIG. 3 is an example of a block diagram of the terminal device 40. FIG. 3 shows a main part of the function of the terminal device 40. As shown in FIG. 3, the terminal device 40 includes a storage unit 41, a control unit 42, a first communication unit 43, a second communication unit 44, and a display unit 45. The storage unit 41 can be realized by at least one of a RAM 40B, a ROM 40C, and an auxiliary storage device. The control unit 42 can be realized by the CPU 40A. The first communication unit 43 can be realized by the communication interface 40E. The second communication unit 44 can be realized by the communication interface 40F. The display unit 45 can be realized by the display 40D. Therefore, the storage unit 41, the control unit 42, the first communication unit 43, the second communication unit 44, and the display unit 45 are connected to each other.

Here, the storage unit 41 includes a waveform storage unit 411 and a model storage unit 412 as components. The control unit 42 includes a model generation unit 421, a waveform comparison unit 422, and a damage detection unit 423 as components. Each component of the control unit 42 accesses at least one of the components of the storage unit 41 and executes various processes. For example, when the model generation unit 421 detects a model waveform generation instruction described later, it accesses the waveform storage unit 411 and acquires the torque waveform of the main shaft 17 in the normal state stored by the waveform storage unit 411. When the model generation unit 421 acquires the torque waveform, the model generation unit 421 generates a model waveform. The other components will be described in detail when the operation of the terminal device 40 is described.

Next, the operation of the terminal device 40 will be described with reference to FIGS. 4 to 8.

First, a learning process for generating a model waveform will be described with reference to FIGS. 4 (a), 5 and 6. As shown in FIG. 4A, the model generation unit 421 acquires a normal torque waveform (step S101). More specifically, the model generation unit 421 acquires the torque waveform of the main shaft 17 at the normal time from the waveform storage unit 411. The torque waveform of the main shaft 17 at the normal time may be stored in the waveform storage unit 411 in advance.

As a result, as shown in FIG. 5, the waveform storage unit 411 stores various torque waveforms of the main shaft 17 in the normal state. Each of these torque waveforms represents a normal relationship between the cutting time of the work WK that has been cut and the cutting torque. The cutting torque represents the ratio of the drive current applied to the rated torque of the servomotor 26. Therefore, the more drive current is supplied to the servomotor 26, the higher the cutting torque. As described above, the shapes of the torque waveforms are often generally similar under normal conditions. In the present embodiment, such a cutting torque is adopted as an example of the cutting load, but for example, the product of the distance from the rotation center of the spindle 17 and the force may be adopted as the cutting load.

Here, the torque waveform is a waveform in which the measured values of the cutting torque acquired at a predetermined sampling interval are connected by a polygonal line in time series. No processing such as processing has been performed on this waveform. The model generation unit 421 acquires a plurality of such torque waveforms. In the present embodiment, the millisecond unit is adopted as the predetermined sampling interval, but various unit times such as the microsecond unit may be adopted as the sampling interval.

When the process of step S101 is completed, the model generation unit 421 then generates a model waveform (step S102). More specifically, the model generation unit 421 learns the torque waveform acquired in the process of step S101 by a neural network which is one of machine learning, and generates an autoregressive model as a learning result as a normal model waveform. As shown in FIG. 6, since the model waveform G1 in the normal state is generated based on various torque waveforms in the normal state, it is similar to the torque waveform in the normal state.

Here, the model value of the cutting torque at the nth time point in the model waveform G1 is predicted from the model value at the (n-1) th time point immediately before the nth time point from the model value at the (n−M) th time point. To. Then, a probability distribution of the probability that the predicted model value appears is also generated. For example, the probability that the model value V1 appears at the nth time point immediately after the model value at the (n-1) th time point is 1%. The model values V2 to V7 can be described in the same manner as the model value V1. In this way, the model generation unit 421 generates the normal model waveform G1 of the drill DL that cuts the work WK together with the probability distribution. Both n and M are integers, and n is larger than M. In particular, M represents the number of samplings of the measured values of the cutting torque from the start of cutting one work WK to the end of cutting. M may be determined based on at least one of the performance of the terminal device 40, the accuracy of predicting damage to the drill DL, and the time until a sign of damage is detected.

When the process of step S102 is completed, the model generation unit 421 then saves the model waveform (step S103). That is, the model generation unit 421 stores the model waveform G1 in the model storage unit 412. As a result, the model storage unit 412 stores the model waveform G1 shown in FIG. 6 generated by the model generation unit 421.

Next, an evaluation process for evaluating the torque waveform to be evaluated will be described with reference to FIGS. 4 (b), 7 and 8. As shown in FIG. 4B, the waveform comparison unit 422 generates a target torque waveform (step S201). More specifically, the waveform comparison unit 422 acquires the measured value of the cutting torque of the spindle 17 from the NC device 30 via the first communication unit 43 after the NC device 30 starts cutting the work WK. The sampling interval for acquiring the measured value of the cutting torque is the same as the sampling interval described above. When the waveform comparison unit 422 acquires the measured value of the cutting torque, it generates the torque waveform up to the time of the actual measurement as the target torque waveform based on the acquired cutting torque. As a result, as shown in the upper part of FIG. 7, the torque waveform G2 of the spindle is generated.

When the process of step S202 is completed, the waveform comparison unit 422 then compares the torque waveforms (step S203). More specifically, as shown in the upper part of FIG. 7, the waveform comparison unit 422 identifies the partial waveform G21 that appears in the most recent period in the generated torque waveform G2. When the waveform comparison unit 422 specifies the partial waveform G21, the model waveform G1 is acquired from the model storage unit 412, and as shown in the lower part of FIG. 7, a part G11 of the model waveform G1 corresponding to the specified partial waveform G21 is specified. To do. When the waveform comparison unit 422 specifies a part G11 of the model waveform G1, the waveform comparison unit 422 compares the partial waveform G21 with the part G11 of the model waveform G1.

When the process of step S202 is completed, the damage detection unit 423 then generates a probability graph (step S203). More specifically, the damage detection unit 423 compares the partial waveform G21 with the partial G11 of the model waveform G1. Then, the damage detection unit 423 expresses which probability of the model value of each cutting torque constituting a part G11 of the model waveform G1 has passed with respect to each measured value of each cutting torque constituting the partial waveform G21. Generate a graph. As a result, as shown in FIG. 8, the probability graph G3 is generated. From the probability graph G3, it is possible to determine whether or not the actually measured value of each cutting torque constituting the partial waveform G21 appears with high probability.

When the process of step S203 is completed, the damage detection unit 423 then determines whether or not a low probability has appeared (step S204). For example, as shown in FIG. 8, when the measured value of the cutting load actually measured next to the actual measurement appears with a probability lower than a predetermined threshold value that can specify a normal cutting load, the damage detection unit 423 appears with a low probability. (Step S204: YES). In this case, the damage detection unit 423 detects a sign of damage to the drill DL that cuts the work WK (step S205), and ends the process. On the other hand, when the damage detection unit 423 determines that the low probability does not appear (step S204: NO), the process of step S205 is skipped and the process ends.

As described above, the terminal device 40 according to the present invention includes a control unit 42, and the control unit 42 includes a model generation unit 421, a waveform comparison unit 422, and a damage detection unit 423. The model generation unit 421 generates the model waveform G1 by machine learning in advance a plurality of torque waveforms representing the normal relationship between the cutting time of the work WK that has been cut and the cutting load. The waveform comparison unit 422 corresponds to the partial waveform G21 that appears in the latest period of the torque waveform G2, which represents the relationship between the cutting time of another work WK to be cut after the work WK and the cutting load until the actual measurement. It is compared with a part G11 of the model waveform G1. As a result of comparison, if the cutting load actually measured next to the actual measurement appears with a probability lower than a predetermined threshold value that can identify a normal cutting load, the damage detection unit 423 is a sign of damage to the drill DL that cuts another work WK. Is detected. Thereby, the damage prediction accuracy of the drill DL can be improved.

In particular, according to the present embodiment, even when the difference between the torque waveform to be evaluated and the model waveform in the normal state where signs of damage appear is small, a certain threshold value is used to predict the damage of the drill DL. Not detected. Therefore, false detection or over-detection that may occur when such a threshold value is used does not occur, and damage to the drill DL can be predicted with high accuracy.

A server device may be connected to the communication network NW so that the server device includes the model generation unit 421. When the performance of the server device is higher than the performance of the terminal device 40, the model waveform can be generated faster than the terminal device 40. On the other hand, if the performance of the CPU and various memories of the NC device 30 is high enough to execute the method of the present invention, the method of the present invention may be executed by the NC device 30. Furthermore, the entire probability graph until the work WK is completely cut is generated, and when a probability lower than a predetermined threshold that can identify a normal cutting load appears in the entire probability graph, a sign of damage to the drill DL is detected. You may.

Although the preferred embodiment of the present invention has been described in detail above, the present invention is not limited to the specific embodiment, and various modifications and modifications are made within the scope of the gist of the present invention described in the claims. It can be changed. For example, in the above-described embodiment, the cutting torque of the spindle 17 is adopted, but the cutting torque based on the drive current applied with respect to the rated torque of the servomotor 24 in the Z-axis direction may be adopted.

30 NC device 40 Terminal device 42 Control unit 421 Model generation unit 422 Waveform comparison unit 423 Damage detection unit 100 Cutting machine DL drill WK work

Claims (1)

A model waveform is generated by pre-machine learning a plurality of first waveforms representing the normal relationship between the cutting time of the first workpiece that has been cut and the cutting load.
The partial waveform appearing in the latest period in the second waveform representing the relationship between the cutting time of the second workpiece to be cut after the first workpiece and the cutting load until the actual measurement is the said corresponding to the partial waveform. Compare with part of the model waveform
As a result of comparison, when the cutting load actually measured next to the actual measurement appears with a probability lower than a predetermined threshold value that can specify a normal cutting load, a sign of damage to the tool for cutting the second workpiece is detected.
How to predict tool breakage.